blob: 42c4c0c892ed9b442617207b00acbdc7be99249f [file] [log] [blame]
Alex Tomasa86c6182006-10-11 01:21:03 -07001/*
2 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
3 * Written by Alex Tomas <alex@clusterfs.com>
4 *
5 * Architecture independence:
6 * Copyright (c) 2005, Bull S.A.
7 * Written by Pierre Peiffer <pierre.peiffer@bull.net>
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public Licens
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
21 */
22
23/*
24 * Extents support for EXT4
25 *
26 * TODO:
27 * - ext4*_error() should be used in some situations
28 * - analyze all BUG()/BUG_ON(), use -EIO where appropriate
29 * - smart tree reduction
30 */
31
32#include <linux/module.h>
33#include <linux/fs.h>
34#include <linux/time.h>
Mingming Caocd02ff02007-10-16 18:38:25 -040035#include <linux/jbd2.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070036#include <linux/highuid.h>
37#include <linux/pagemap.h>
38#include <linux/quotaops.h>
39#include <linux/string.h>
40#include <linux/slab.h>
Amit Aroraa2df2a62007-07-17 21:42:41 -040041#include <linux/falloc.h>
Alex Tomasa86c6182006-10-11 01:21:03 -070042#include <asm/uaccess.h>
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040043#include "ext4_jbd2.h"
44#include "ext4_extents.h"
Alex Tomasa86c6182006-10-11 01:21:03 -070045
46
Randy Dunlapd0d856e2006-10-11 01:21:07 -070047/*
48 * ext_pblock:
49 * combine low and high parts of physical block number into ext4_fsblk_t
50 */
Avantika Mathur09b88252006-12-06 20:41:36 -080051static ext4_fsblk_t ext_pblock(struct ext4_extent *ex)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070052{
53 ext4_fsblk_t block;
54
Aneesh Kumar K.Vb3776112007-10-16 18:38:25 -040055 block = le32_to_cpu(ex->ee_start_lo);
Mingming Cao9b8f1f02006-10-11 01:21:13 -070056 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -070057 return block;
58}
59
Randy Dunlapd0d856e2006-10-11 01:21:07 -070060/*
61 * idx_pblock:
62 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
63 */
Aneesh Kumar K.Vc14c6fd2008-01-28 23:58:26 -050064ext4_fsblk_t idx_pblock(struct ext4_extent_idx *ix)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070065{
66 ext4_fsblk_t block;
67
Aneesh Kumar K.Vd8dd0b42007-10-16 18:38:25 -040068 block = le32_to_cpu(ix->ei_leaf_lo);
Mingming Cao9b8f1f02006-10-11 01:21:13 -070069 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -070070 return block;
71}
72
Randy Dunlapd0d856e2006-10-11 01:21:07 -070073/*
74 * ext4_ext_store_pblock:
75 * stores a large physical block number into an extent struct,
76 * breaking it into parts
77 */
Aneesh Kumar K.Vc14c6fd2008-01-28 23:58:26 -050078void ext4_ext_store_pblock(struct ext4_extent *ex, ext4_fsblk_t pb)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070079{
Aneesh Kumar K.Vb3776112007-10-16 18:38:25 -040080 ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
Mingming Cao9b8f1f02006-10-11 01:21:13 -070081 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
Alex Tomasf65e6fb2006-10-11 01:21:05 -070082}
83
Randy Dunlapd0d856e2006-10-11 01:21:07 -070084/*
85 * ext4_idx_store_pblock:
86 * stores a large physical block number into an index struct,
87 * breaking it into parts
88 */
Avantika Mathur09b88252006-12-06 20:41:36 -080089static void ext4_idx_store_pblock(struct ext4_extent_idx *ix, ext4_fsblk_t pb)
Alex Tomasf65e6fb2006-10-11 01:21:05 -070090{
Aneesh Kumar K.Vd8dd0b42007-10-16 18:38:25 -040091 ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
Mingming Cao9b8f1f02006-10-11 01:21:13 -070092 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) & 0xffff);
Alex Tomasf65e6fb2006-10-11 01:21:05 -070093}
94
Shen Feng9102e4f2008-07-11 19:27:31 -040095static int ext4_ext_journal_restart(handle_t *handle, int needed)
Alex Tomasa86c6182006-10-11 01:21:03 -070096{
97 int err;
98
99 if (handle->h_buffer_credits > needed)
Shen Feng9102e4f2008-07-11 19:27:31 -0400100 return 0;
101 err = ext4_journal_extend(handle, needed);
102 if (err)
103 return err;
104 return ext4_journal_restart(handle, needed);
Alex Tomasa86c6182006-10-11 01:21:03 -0700105}
106
107/*
108 * could return:
109 * - EROFS
110 * - ENOMEM
111 */
112static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
113 struct ext4_ext_path *path)
114{
115 if (path->p_bh) {
116 /* path points to block */
117 return ext4_journal_get_write_access(handle, path->p_bh);
118 }
119 /* path points to leaf/index in inode body */
120 /* we use in-core data, no need to protect them */
121 return 0;
122}
123
124/*
125 * could return:
126 * - EROFS
127 * - ENOMEM
128 * - EIO
129 */
130static int ext4_ext_dirty(handle_t *handle, struct inode *inode,
131 struct ext4_ext_path *path)
132{
133 int err;
134 if (path->p_bh) {
135 /* path points to block */
136 err = ext4_journal_dirty_metadata(handle, path->p_bh);
137 } else {
138 /* path points to leaf/index in inode body */
139 err = ext4_mark_inode_dirty(handle, inode);
140 }
141 return err;
142}
143
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700144static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700145 struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500146 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700147{
148 struct ext4_inode_info *ei = EXT4_I(inode);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700149 ext4_fsblk_t bg_start;
Valerie Clement74d34872008-02-15 13:43:07 -0500150 ext4_fsblk_t last_block;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700151 ext4_grpblk_t colour;
Alex Tomasa86c6182006-10-11 01:21:03 -0700152 int depth;
153
154 if (path) {
155 struct ext4_extent *ex;
156 depth = path->p_depth;
157
158 /* try to predict block placement */
Avantika Mathur7e028972006-12-06 20:41:33 -0800159 ex = path[depth].p_ext;
160 if (ex)
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700161 return ext_pblock(ex)+(block-le32_to_cpu(ex->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700162
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700163 /* it looks like index is empty;
164 * try to find starting block from index itself */
Alex Tomasa86c6182006-10-11 01:21:03 -0700165 if (path[depth].p_bh)
166 return path[depth].p_bh->b_blocknr;
167 }
168
169 /* OK. use inode's group */
170 bg_start = (ei->i_block_group * EXT4_BLOCKS_PER_GROUP(inode->i_sb)) +
171 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_first_data_block);
Valerie Clement74d34872008-02-15 13:43:07 -0500172 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
173
174 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
175 colour = (current->pid % 16) *
Alex Tomasa86c6182006-10-11 01:21:03 -0700176 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
Valerie Clement74d34872008-02-15 13:43:07 -0500177 else
178 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
Alex Tomasa86c6182006-10-11 01:21:03 -0700179 return bg_start + colour + block;
180}
181
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400182/*
183 * Allocation for a meta data block
184 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700185static ext4_fsblk_t
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400186ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -0700187 struct ext4_ext_path *path,
188 struct ext4_extent *ex, int *err)
189{
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700190 ext4_fsblk_t goal, newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700191
192 goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
Aneesh Kumar K.V7061eba2008-07-11 19:27:31 -0400193 newblock = ext4_new_meta_block(handle, inode, goal, err);
Alex Tomasa86c6182006-10-11 01:21:03 -0700194 return newblock;
195}
196
Avantika Mathur09b88252006-12-06 20:41:36 -0800197static int ext4_ext_space_block(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700198{
199 int size;
200
201 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
202 / sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100203#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700204 if (size > 6)
205 size = 6;
206#endif
207 return size;
208}
209
Avantika Mathur09b88252006-12-06 20:41:36 -0800210static int ext4_ext_space_block_idx(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700211{
212 int size;
213
214 size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
215 / sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100216#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700217 if (size > 5)
218 size = 5;
219#endif
220 return size;
221}
222
Avantika Mathur09b88252006-12-06 20:41:36 -0800223static int ext4_ext_space_root(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700224{
225 int size;
226
227 size = sizeof(EXT4_I(inode)->i_data);
228 size -= sizeof(struct ext4_extent_header);
229 size /= sizeof(struct ext4_extent);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100230#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700231 if (size > 3)
232 size = 3;
233#endif
234 return size;
235}
236
Avantika Mathur09b88252006-12-06 20:41:36 -0800237static int ext4_ext_space_root_idx(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -0700238{
239 int size;
240
241 size = sizeof(EXT4_I(inode)->i_data);
242 size -= sizeof(struct ext4_extent_header);
243 size /= sizeof(struct ext4_extent_idx);
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +0100244#ifdef AGGRESSIVE_TEST
Alex Tomasa86c6182006-10-11 01:21:03 -0700245 if (size > 4)
246 size = 4;
247#endif
248 return size;
249}
250
Mingming Caod2a17632008-07-14 17:52:37 -0400251/*
252 * Calculate the number of metadata blocks needed
253 * to allocate @blocks
254 * Worse case is one block per extent
255 */
256int ext4_ext_calc_metadata_amount(struct inode *inode, int blocks)
257{
258 int lcap, icap, rcap, leafs, idxs, num;
259 int newextents = blocks;
260
261 rcap = ext4_ext_space_root_idx(inode);
262 lcap = ext4_ext_space_block(inode);
263 icap = ext4_ext_space_block_idx(inode);
264
265 /* number of new leaf blocks needed */
266 num = leafs = (newextents + lcap - 1) / lcap;
267
268 /*
269 * Worse case, we need separate index block(s)
270 * to link all new leaf blocks
271 */
272 idxs = (leafs + icap - 1) / icap;
273 do {
274 num += idxs;
275 idxs = (idxs + icap - 1) / icap;
276 } while (idxs > rcap);
277
278 return num;
279}
280
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400281static int
282ext4_ext_max_entries(struct inode *inode, int depth)
283{
284 int max;
285
286 if (depth == ext_depth(inode)) {
287 if (depth == 0)
288 max = ext4_ext_space_root(inode);
289 else
290 max = ext4_ext_space_root_idx(inode);
291 } else {
292 if (depth == 0)
293 max = ext4_ext_space_block(inode);
294 else
295 max = ext4_ext_space_block_idx(inode);
296 }
297
298 return max;
299}
300
301static int __ext4_ext_check_header(const char *function, struct inode *inode,
302 struct ext4_extent_header *eh,
303 int depth)
304{
305 const char *error_msg;
306 int max = 0;
307
308 if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
309 error_msg = "invalid magic";
310 goto corrupted;
311 }
312 if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
313 error_msg = "unexpected eh_depth";
314 goto corrupted;
315 }
316 if (unlikely(eh->eh_max == 0)) {
317 error_msg = "invalid eh_max";
318 goto corrupted;
319 }
320 max = ext4_ext_max_entries(inode, depth);
321 if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
322 error_msg = "too large eh_max";
323 goto corrupted;
324 }
325 if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
326 error_msg = "invalid eh_entries";
327 goto corrupted;
328 }
329 return 0;
330
331corrupted:
332 ext4_error(inode->i_sb, function,
333 "bad header in inode #%lu: %s - magic %x, "
334 "entries %u, max %u(%u), depth %u(%u)",
335 inode->i_ino, error_msg, le16_to_cpu(eh->eh_magic),
336 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
337 max, le16_to_cpu(eh->eh_depth), depth);
338
339 return -EIO;
340}
341
342#define ext4_ext_check_header(inode, eh, depth) \
Harvey Harrison46e665e2008-04-17 10:38:59 -0400343 __ext4_ext_check_header(__func__, inode, eh, depth)
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400344
Alex Tomasa86c6182006-10-11 01:21:03 -0700345#ifdef EXT_DEBUG
346static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
347{
348 int k, l = path->p_depth;
349
350 ext_debug("path:");
351 for (k = 0; k <= l; k++, path++) {
352 if (path->p_idx) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700353 ext_debug(" %d->%llu", le32_to_cpu(path->p_idx->ei_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700354 idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700355 } else if (path->p_ext) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700356 ext_debug(" %d:%d:%llu ",
Alex Tomasa86c6182006-10-11 01:21:03 -0700357 le32_to_cpu(path->p_ext->ee_block),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400358 ext4_ext_get_actual_len(path->p_ext),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700359 ext_pblock(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700360 } else
361 ext_debug(" []");
362 }
363 ext_debug("\n");
364}
365
366static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
367{
368 int depth = ext_depth(inode);
369 struct ext4_extent_header *eh;
370 struct ext4_extent *ex;
371 int i;
372
373 if (!path)
374 return;
375
376 eh = path[depth].p_hdr;
377 ex = EXT_FIRST_EXTENT(eh);
378
379 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700380 ext_debug("%d:%d:%llu ", le32_to_cpu(ex->ee_block),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400381 ext4_ext_get_actual_len(ex), ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -0700382 }
383 ext_debug("\n");
384}
385#else
386#define ext4_ext_show_path(inode,path)
387#define ext4_ext_show_leaf(inode,path)
388#endif
389
Aneesh Kumar K.Vb35905c2008-02-25 16:54:37 -0500390void ext4_ext_drop_refs(struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700391{
392 int depth = path->p_depth;
393 int i;
394
395 for (i = 0; i <= depth; i++, path++)
396 if (path->p_bh) {
397 brelse(path->p_bh);
398 path->p_bh = NULL;
399 }
400}
401
402/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700403 * ext4_ext_binsearch_idx:
404 * binary search for the closest index of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400405 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700406 */
407static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500408ext4_ext_binsearch_idx(struct inode *inode,
409 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700410{
411 struct ext4_extent_header *eh = path->p_hdr;
412 struct ext4_extent_idx *r, *l, *m;
413
Alex Tomasa86c6182006-10-11 01:21:03 -0700414
Eric Sandeenbba90742008-01-28 23:58:27 -0500415 ext_debug("binsearch for %u(idx): ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700416
417 l = EXT_FIRST_INDEX(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400418 r = EXT_LAST_INDEX(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700419 while (l <= r) {
420 m = l + (r - l) / 2;
421 if (block < le32_to_cpu(m->ei_block))
422 r = m - 1;
423 else
424 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400425 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
426 m, le32_to_cpu(m->ei_block),
427 r, le32_to_cpu(r->ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700428 }
429
430 path->p_idx = l - 1;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700431 ext_debug(" -> %d->%lld ", le32_to_cpu(path->p_idx->ei_block),
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400432 idx_pblock(path->p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700433
434#ifdef CHECK_BINSEARCH
435 {
436 struct ext4_extent_idx *chix, *ix;
437 int k;
438
439 chix = ix = EXT_FIRST_INDEX(eh);
440 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
441 if (k != 0 &&
442 le32_to_cpu(ix->ei_block) <= le32_to_cpu(ix[-1].ei_block)) {
443 printk("k=%d, ix=0x%p, first=0x%p\n", k,
444 ix, EXT_FIRST_INDEX(eh));
445 printk("%u <= %u\n",
446 le32_to_cpu(ix->ei_block),
447 le32_to_cpu(ix[-1].ei_block));
448 }
449 BUG_ON(k && le32_to_cpu(ix->ei_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400450 <= le32_to_cpu(ix[-1].ei_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700451 if (block < le32_to_cpu(ix->ei_block))
452 break;
453 chix = ix;
454 }
455 BUG_ON(chix != path->p_idx);
456 }
457#endif
458
459}
460
461/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700462 * ext4_ext_binsearch:
463 * binary search for closest extent of the given block
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400464 * the header must be checked before calling this
Alex Tomasa86c6182006-10-11 01:21:03 -0700465 */
466static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500467ext4_ext_binsearch(struct inode *inode,
468 struct ext4_ext_path *path, ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -0700469{
470 struct ext4_extent_header *eh = path->p_hdr;
471 struct ext4_extent *r, *l, *m;
472
Alex Tomasa86c6182006-10-11 01:21:03 -0700473 if (eh->eh_entries == 0) {
474 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700475 * this leaf is empty:
476 * we get such a leaf in split/add case
Alex Tomasa86c6182006-10-11 01:21:03 -0700477 */
478 return;
479 }
480
Eric Sandeenbba90742008-01-28 23:58:27 -0500481 ext_debug("binsearch for %u: ", block);
Alex Tomasa86c6182006-10-11 01:21:03 -0700482
483 l = EXT_FIRST_EXTENT(eh) + 1;
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400484 r = EXT_LAST_EXTENT(eh);
Alex Tomasa86c6182006-10-11 01:21:03 -0700485
486 while (l <= r) {
487 m = l + (r - l) / 2;
488 if (block < le32_to_cpu(m->ee_block))
489 r = m - 1;
490 else
491 l = m + 1;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400492 ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
493 m, le32_to_cpu(m->ee_block),
494 r, le32_to_cpu(r->ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700495 }
496
497 path->p_ext = l - 1;
Mingming Cao2ae02102006-10-11 01:21:11 -0700498 ext_debug(" -> %d:%llu:%d ",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400499 le32_to_cpu(path->p_ext->ee_block),
500 ext_pblock(path->p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400501 ext4_ext_get_actual_len(path->p_ext));
Alex Tomasa86c6182006-10-11 01:21:03 -0700502
503#ifdef CHECK_BINSEARCH
504 {
505 struct ext4_extent *chex, *ex;
506 int k;
507
508 chex = ex = EXT_FIRST_EXTENT(eh);
509 for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
510 BUG_ON(k && le32_to_cpu(ex->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400511 <= le32_to_cpu(ex[-1].ee_block));
Alex Tomasa86c6182006-10-11 01:21:03 -0700512 if (block < le32_to_cpu(ex->ee_block))
513 break;
514 chex = ex;
515 }
516 BUG_ON(chex != path->p_ext);
517 }
518#endif
519
520}
521
522int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
523{
524 struct ext4_extent_header *eh;
525
526 eh = ext_inode_hdr(inode);
527 eh->eh_depth = 0;
528 eh->eh_entries = 0;
529 eh->eh_magic = EXT4_EXT_MAGIC;
530 eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode));
531 ext4_mark_inode_dirty(handle, inode);
532 ext4_ext_invalidate_cache(inode);
533 return 0;
534}
535
536struct ext4_ext_path *
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500537ext4_ext_find_extent(struct inode *inode, ext4_lblk_t block,
538 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -0700539{
540 struct ext4_extent_header *eh;
541 struct buffer_head *bh;
542 short int depth, i, ppos = 0, alloc = 0;
543
544 eh = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400545 depth = ext_depth(inode);
546 if (ext4_ext_check_header(inode, eh, depth))
Alex Tomasa86c6182006-10-11 01:21:03 -0700547 return ERR_PTR(-EIO);
548
Alex Tomasa86c6182006-10-11 01:21:03 -0700549
550 /* account possible depth increase */
551 if (!path) {
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800552 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 2),
Alex Tomasa86c6182006-10-11 01:21:03 -0700553 GFP_NOFS);
554 if (!path)
555 return ERR_PTR(-ENOMEM);
556 alloc = 1;
557 }
Alex Tomasa86c6182006-10-11 01:21:03 -0700558 path[0].p_hdr = eh;
Shen Feng1973adc2008-07-11 19:27:31 -0400559 path[0].p_bh = NULL;
Alex Tomasa86c6182006-10-11 01:21:03 -0700560
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400561 i = depth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700562 /* walk through the tree */
563 while (i) {
564 ext_debug("depth %d: num %d, max %d\n",
565 ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400566
Alex Tomasa86c6182006-10-11 01:21:03 -0700567 ext4_ext_binsearch_idx(inode, path + ppos, block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700568 path[ppos].p_block = idx_pblock(path[ppos].p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -0700569 path[ppos].p_depth = i;
570 path[ppos].p_ext = NULL;
571
572 bh = sb_bread(inode->i_sb, path[ppos].p_block);
573 if (!bh)
574 goto err;
575
576 eh = ext_block_hdr(bh);
577 ppos++;
578 BUG_ON(ppos > depth);
579 path[ppos].p_bh = bh;
580 path[ppos].p_hdr = eh;
581 i--;
582
Alex Tomasc29c0ae2007-07-18 09:19:09 -0400583 if (ext4_ext_check_header(inode, eh, i))
Alex Tomasa86c6182006-10-11 01:21:03 -0700584 goto err;
585 }
586
587 path[ppos].p_depth = i;
Alex Tomasa86c6182006-10-11 01:21:03 -0700588 path[ppos].p_ext = NULL;
589 path[ppos].p_idx = NULL;
590
Alex Tomasa86c6182006-10-11 01:21:03 -0700591 /* find extent */
592 ext4_ext_binsearch(inode, path + ppos, block);
Shen Feng1973adc2008-07-11 19:27:31 -0400593 /* if not an empty leaf */
594 if (path[ppos].p_ext)
595 path[ppos].p_block = ext_pblock(path[ppos].p_ext);
Alex Tomasa86c6182006-10-11 01:21:03 -0700596
597 ext4_ext_show_path(inode, path);
598
599 return path;
600
601err:
602 ext4_ext_drop_refs(path);
603 if (alloc)
604 kfree(path);
605 return ERR_PTR(-EIO);
606}
607
608/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700609 * ext4_ext_insert_index:
610 * insert new index [@logical;@ptr] into the block at @curp;
611 * check where to insert: before @curp or after @curp
Alex Tomasa86c6182006-10-11 01:21:03 -0700612 */
613static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
614 struct ext4_ext_path *curp,
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700615 int logical, ext4_fsblk_t ptr)
Alex Tomasa86c6182006-10-11 01:21:03 -0700616{
617 struct ext4_extent_idx *ix;
618 int len, err;
619
Avantika Mathur7e028972006-12-06 20:41:33 -0800620 err = ext4_ext_get_access(handle, inode, curp);
621 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700622 return err;
623
624 BUG_ON(logical == le32_to_cpu(curp->p_idx->ei_block));
625 len = EXT_MAX_INDEX(curp->p_hdr) - curp->p_idx;
626 if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
627 /* insert after */
628 if (curp->p_idx != EXT_LAST_INDEX(curp->p_hdr)) {
629 len = (len - 1) * sizeof(struct ext4_extent_idx);
630 len = len < 0 ? 0 : len;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400631 ext_debug("insert new index %d after: %llu. "
Alex Tomasa86c6182006-10-11 01:21:03 -0700632 "move %d from 0x%p to 0x%p\n",
633 logical, ptr, len,
634 (curp->p_idx + 1), (curp->p_idx + 2));
635 memmove(curp->p_idx + 2, curp->p_idx + 1, len);
636 }
637 ix = curp->p_idx + 1;
638 } else {
639 /* insert before */
640 len = len * sizeof(struct ext4_extent_idx);
641 len = len < 0 ? 0 : len;
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400642 ext_debug("insert new index %d before: %llu. "
Alex Tomasa86c6182006-10-11 01:21:03 -0700643 "move %d from 0x%p to 0x%p\n",
644 logical, ptr, len,
645 curp->p_idx, (curp->p_idx + 1));
646 memmove(curp->p_idx + 1, curp->p_idx, len);
647 ix = curp->p_idx;
648 }
649
650 ix->ei_block = cpu_to_le32(logical);
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700651 ext4_idx_store_pblock(ix, ptr);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400652 le16_add_cpu(&curp->p_hdr->eh_entries, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -0700653
654 BUG_ON(le16_to_cpu(curp->p_hdr->eh_entries)
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400655 > le16_to_cpu(curp->p_hdr->eh_max));
Alex Tomasa86c6182006-10-11 01:21:03 -0700656 BUG_ON(ix > EXT_LAST_INDEX(curp->p_hdr));
657
658 err = ext4_ext_dirty(handle, inode, curp);
659 ext4_std_error(inode->i_sb, err);
660
661 return err;
662}
663
664/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700665 * ext4_ext_split:
666 * inserts new subtree into the path, using free index entry
667 * at depth @at:
668 * - allocates all needed blocks (new leaf and all intermediate index blocks)
669 * - makes decision where to split
670 * - moves remaining extents and index entries (right to the split point)
671 * into the newly allocated blocks
672 * - initializes subtree
Alex Tomasa86c6182006-10-11 01:21:03 -0700673 */
674static int ext4_ext_split(handle_t *handle, struct inode *inode,
675 struct ext4_ext_path *path,
676 struct ext4_extent *newext, int at)
677{
678 struct buffer_head *bh = NULL;
679 int depth = ext_depth(inode);
680 struct ext4_extent_header *neh;
681 struct ext4_extent_idx *fidx;
682 struct ext4_extent *ex;
683 int i = at, k, m, a;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700684 ext4_fsblk_t newblock, oldblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700685 __le32 border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700686 ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
Alex Tomasa86c6182006-10-11 01:21:03 -0700687 int err = 0;
688
689 /* make decision: where to split? */
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700690 /* FIXME: now decision is simplest: at current extent */
Alex Tomasa86c6182006-10-11 01:21:03 -0700691
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700692 /* if current leaf will be split, then we should use
Alex Tomasa86c6182006-10-11 01:21:03 -0700693 * border from split point */
694 BUG_ON(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr));
695 if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
696 border = path[depth].p_ext[1].ee_block;
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700697 ext_debug("leaf will be split."
Alex Tomasa86c6182006-10-11 01:21:03 -0700698 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400699 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700700 } else {
701 border = newext->ee_block;
702 ext_debug("leaf will be added."
703 " next leaf starts at %d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400704 le32_to_cpu(border));
Alex Tomasa86c6182006-10-11 01:21:03 -0700705 }
706
707 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700708 * If error occurs, then we break processing
709 * and mark filesystem read-only. index won't
Alex Tomasa86c6182006-10-11 01:21:03 -0700710 * be inserted and tree will be in consistent
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700711 * state. Next mount will repair buffers too.
Alex Tomasa86c6182006-10-11 01:21:03 -0700712 */
713
714 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700715 * Get array to track all allocated blocks.
716 * We need this to handle errors and free blocks
717 * upon them.
Alex Tomasa86c6182006-10-11 01:21:03 -0700718 */
Avantika Mathur5d4958f2006-12-06 20:41:35 -0800719 ablocks = kzalloc(sizeof(ext4_fsblk_t) * depth, GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -0700720 if (!ablocks)
721 return -ENOMEM;
Alex Tomasa86c6182006-10-11 01:21:03 -0700722
723 /* allocate all needed blocks */
724 ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
725 for (a = 0; a < depth - at; a++) {
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400726 newblock = ext4_ext_new_meta_block(handle, inode, path,
727 newext, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -0700728 if (newblock == 0)
729 goto cleanup;
730 ablocks[a] = newblock;
731 }
732
733 /* initialize new leaf */
734 newblock = ablocks[--a];
735 BUG_ON(newblock == 0);
736 bh = sb_getblk(inode->i_sb, newblock);
737 if (!bh) {
738 err = -EIO;
739 goto cleanup;
740 }
741 lock_buffer(bh);
742
Avantika Mathur7e028972006-12-06 20:41:33 -0800743 err = ext4_journal_get_create_access(handle, bh);
744 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700745 goto cleanup;
746
747 neh = ext_block_hdr(bh);
748 neh->eh_entries = 0;
749 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
750 neh->eh_magic = EXT4_EXT_MAGIC;
751 neh->eh_depth = 0;
752 ex = EXT_FIRST_EXTENT(neh);
753
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700754 /* move remainder of path[depth] to the new leaf */
Alex Tomasa86c6182006-10-11 01:21:03 -0700755 BUG_ON(path[depth].p_hdr->eh_entries != path[depth].p_hdr->eh_max);
756 /* start copy from next extent */
757 /* TODO: we could do it by single memmove */
758 m = 0;
759 path[depth].p_ext++;
760 while (path[depth].p_ext <=
761 EXT_MAX_EXTENT(path[depth].p_hdr)) {
Mingming Cao2ae02102006-10-11 01:21:11 -0700762 ext_debug("move %d:%llu:%d in new leaf %llu\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400763 le32_to_cpu(path[depth].p_ext->ee_block),
764 ext_pblock(path[depth].p_ext),
Amit Aroraa2df2a62007-07-17 21:42:41 -0400765 ext4_ext_get_actual_len(path[depth].p_ext),
Alex Tomasa86c6182006-10-11 01:21:03 -0700766 newblock);
767 /*memmove(ex++, path[depth].p_ext++,
768 sizeof(struct ext4_extent));
769 neh->eh_entries++;*/
770 path[depth].p_ext++;
771 m++;
772 }
773 if (m) {
774 memmove(ex, path[depth].p_ext-m, sizeof(struct ext4_extent)*m);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400775 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -0700776 }
777
778 set_buffer_uptodate(bh);
779 unlock_buffer(bh);
780
Avantika Mathur7e028972006-12-06 20:41:33 -0800781 err = ext4_journal_dirty_metadata(handle, bh);
782 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700783 goto cleanup;
784 brelse(bh);
785 bh = NULL;
786
787 /* correct old leaf */
788 if (m) {
Avantika Mathur7e028972006-12-06 20:41:33 -0800789 err = ext4_ext_get_access(handle, inode, path + depth);
790 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700791 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -0400792 le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
Avantika Mathur7e028972006-12-06 20:41:33 -0800793 err = ext4_ext_dirty(handle, inode, path + depth);
794 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700795 goto cleanup;
796
797 }
798
799 /* create intermediate indexes */
800 k = depth - at - 1;
801 BUG_ON(k < 0);
802 if (k)
803 ext_debug("create %d intermediate indices\n", k);
804 /* insert new index into current index block */
805 /* current depth stored in i var */
806 i = depth - 1;
807 while (k--) {
808 oldblock = newblock;
809 newblock = ablocks[--a];
Eric Sandeenbba90742008-01-28 23:58:27 -0500810 bh = sb_getblk(inode->i_sb, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700811 if (!bh) {
812 err = -EIO;
813 goto cleanup;
814 }
815 lock_buffer(bh);
816
Avantika Mathur7e028972006-12-06 20:41:33 -0800817 err = ext4_journal_get_create_access(handle, bh);
818 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700819 goto cleanup;
820
821 neh = ext_block_hdr(bh);
822 neh->eh_entries = cpu_to_le16(1);
823 neh->eh_magic = EXT4_EXT_MAGIC;
824 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
825 neh->eh_depth = cpu_to_le16(depth - i);
826 fidx = EXT_FIRST_INDEX(neh);
827 fidx->ei_block = border;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700828 ext4_idx_store_pblock(fidx, oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700829
Eric Sandeenbba90742008-01-28 23:58:27 -0500830 ext_debug("int.index at %d (block %llu): %u -> %llu\n",
831 i, newblock, le32_to_cpu(border), oldblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700832 /* copy indexes */
833 m = 0;
834 path[i].p_idx++;
835
836 ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
837 EXT_MAX_INDEX(path[i].p_hdr));
838 BUG_ON(EXT_MAX_INDEX(path[i].p_hdr) !=
839 EXT_LAST_INDEX(path[i].p_hdr));
840 while (path[i].p_idx <= EXT_MAX_INDEX(path[i].p_hdr)) {
Dmitry Monakhov26d535e2007-07-18 08:33:37 -0400841 ext_debug("%d: move %d:%llu in new index %llu\n", i,
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400842 le32_to_cpu(path[i].p_idx->ei_block),
843 idx_pblock(path[i].p_idx),
844 newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700845 /*memmove(++fidx, path[i].p_idx++,
846 sizeof(struct ext4_extent_idx));
847 neh->eh_entries++;
848 BUG_ON(neh->eh_entries > neh->eh_max);*/
849 path[i].p_idx++;
850 m++;
851 }
852 if (m) {
853 memmove(++fidx, path[i].p_idx - m,
854 sizeof(struct ext4_extent_idx) * m);
Marcin Slusarze8546d02008-04-17 10:38:59 -0400855 le16_add_cpu(&neh->eh_entries, m);
Alex Tomasa86c6182006-10-11 01:21:03 -0700856 }
857 set_buffer_uptodate(bh);
858 unlock_buffer(bh);
859
Avantika Mathur7e028972006-12-06 20:41:33 -0800860 err = ext4_journal_dirty_metadata(handle, bh);
861 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700862 goto cleanup;
863 brelse(bh);
864 bh = NULL;
865
866 /* correct old index */
867 if (m) {
868 err = ext4_ext_get_access(handle, inode, path + i);
869 if (err)
870 goto cleanup;
Marcin Slusarze8546d02008-04-17 10:38:59 -0400871 le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
Alex Tomasa86c6182006-10-11 01:21:03 -0700872 err = ext4_ext_dirty(handle, inode, path + i);
873 if (err)
874 goto cleanup;
875 }
876
877 i--;
878 }
879
880 /* insert new index */
Alex Tomasa86c6182006-10-11 01:21:03 -0700881 err = ext4_ext_insert_index(handle, inode, path + at,
882 le32_to_cpu(border), newblock);
883
884cleanup:
885 if (bh) {
886 if (buffer_locked(bh))
887 unlock_buffer(bh);
888 brelse(bh);
889 }
890
891 if (err) {
892 /* free all allocated blocks in error case */
893 for (i = 0; i < depth; i++) {
894 if (!ablocks[i])
895 continue;
Alex Tomasc9de5602008-01-29 00:19:52 -0500896 ext4_free_blocks(handle, inode, ablocks[i], 1, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -0700897 }
898 }
899 kfree(ablocks);
900
901 return err;
902}
903
904/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700905 * ext4_ext_grow_indepth:
906 * implements tree growing procedure:
907 * - allocates new block
908 * - moves top-level data (index block or leaf) into the new block
909 * - initializes new top-level, creating index that points to the
910 * just created block
Alex Tomasa86c6182006-10-11 01:21:03 -0700911 */
912static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
913 struct ext4_ext_path *path,
914 struct ext4_extent *newext)
915{
916 struct ext4_ext_path *curp = path;
917 struct ext4_extent_header *neh;
918 struct ext4_extent_idx *fidx;
919 struct buffer_head *bh;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700920 ext4_fsblk_t newblock;
Alex Tomasa86c6182006-10-11 01:21:03 -0700921 int err = 0;
922
Aneesh Kumar K.V654b4902008-07-11 19:27:31 -0400923 newblock = ext4_ext_new_meta_block(handle, inode, path, newext, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -0700924 if (newblock == 0)
925 return err;
926
927 bh = sb_getblk(inode->i_sb, newblock);
928 if (!bh) {
929 err = -EIO;
930 ext4_std_error(inode->i_sb, err);
931 return err;
932 }
933 lock_buffer(bh);
934
Avantika Mathur7e028972006-12-06 20:41:33 -0800935 err = ext4_journal_get_create_access(handle, bh);
936 if (err) {
Alex Tomasa86c6182006-10-11 01:21:03 -0700937 unlock_buffer(bh);
938 goto out;
939 }
940
941 /* move top-level index/leaf into new block */
942 memmove(bh->b_data, curp->p_hdr, sizeof(EXT4_I(inode)->i_data));
943
944 /* set size of new block */
945 neh = ext_block_hdr(bh);
946 /* old root could have indexes or leaves
947 * so calculate e_max right way */
948 if (ext_depth(inode))
949 neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode));
950 else
951 neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode));
952 neh->eh_magic = EXT4_EXT_MAGIC;
953 set_buffer_uptodate(bh);
954 unlock_buffer(bh);
955
Avantika Mathur7e028972006-12-06 20:41:33 -0800956 err = ext4_journal_dirty_metadata(handle, bh);
957 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700958 goto out;
959
960 /* create index in new top-level index: num,max,pointer */
Avantika Mathur7e028972006-12-06 20:41:33 -0800961 err = ext4_ext_get_access(handle, inode, curp);
962 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -0700963 goto out;
964
965 curp->p_hdr->eh_magic = EXT4_EXT_MAGIC;
966 curp->p_hdr->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode));
967 curp->p_hdr->eh_entries = cpu_to_le16(1);
968 curp->p_idx = EXT_FIRST_INDEX(curp->p_hdr);
Dmitry Monakhove9f410b2007-07-18 09:09:15 -0400969
970 if (path[0].p_hdr->eh_depth)
971 curp->p_idx->ei_block =
972 EXT_FIRST_INDEX(path[0].p_hdr)->ei_block;
973 else
974 curp->p_idx->ei_block =
975 EXT_FIRST_EXTENT(path[0].p_hdr)->ee_block;
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700976 ext4_idx_store_pblock(curp->p_idx, newblock);
Alex Tomasa86c6182006-10-11 01:21:03 -0700977
978 neh = ext_inode_hdr(inode);
979 fidx = EXT_FIRST_INDEX(neh);
Mingming Cao2ae02102006-10-11 01:21:11 -0700980 ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -0700981 le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700982 le32_to_cpu(fidx->ei_block), idx_pblock(fidx));
Alex Tomasa86c6182006-10-11 01:21:03 -0700983
984 neh->eh_depth = cpu_to_le16(path->p_depth + 1);
985 err = ext4_ext_dirty(handle, inode, curp);
986out:
987 brelse(bh);
988
989 return err;
990}
991
992/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700993 * ext4_ext_create_new_leaf:
994 * finds empty index and adds new leaf.
995 * if no free index is found, then it requests in-depth growing.
Alex Tomasa86c6182006-10-11 01:21:03 -0700996 */
997static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
998 struct ext4_ext_path *path,
999 struct ext4_extent *newext)
1000{
1001 struct ext4_ext_path *curp;
1002 int depth, i, err = 0;
1003
1004repeat:
1005 i = depth = ext_depth(inode);
1006
1007 /* walk up to the tree and look for free index entry */
1008 curp = path + depth;
1009 while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1010 i--;
1011 curp--;
1012 }
1013
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001014 /* we use already allocated block for index block,
1015 * so subsequent data blocks should be contiguous */
Alex Tomasa86c6182006-10-11 01:21:03 -07001016 if (EXT_HAS_FREE_INDEX(curp)) {
1017 /* if we found index with free entry, then use that
1018 * entry: create all needed subtree and add new leaf */
1019 err = ext4_ext_split(handle, inode, path, newext, i);
Shen Feng787e0982008-07-11 19:27:31 -04001020 if (err)
1021 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07001022
1023 /* refill path */
1024 ext4_ext_drop_refs(path);
1025 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001026 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1027 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001028 if (IS_ERR(path))
1029 err = PTR_ERR(path);
1030 } else {
1031 /* tree is full, time to grow in depth */
1032 err = ext4_ext_grow_indepth(handle, inode, path, newext);
1033 if (err)
1034 goto out;
1035
1036 /* refill path */
1037 ext4_ext_drop_refs(path);
1038 path = ext4_ext_find_extent(inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001039 (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1040 path);
Alex Tomasa86c6182006-10-11 01:21:03 -07001041 if (IS_ERR(path)) {
1042 err = PTR_ERR(path);
1043 goto out;
1044 }
1045
1046 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001047 * only first (depth 0 -> 1) produces free space;
1048 * in all other cases we have to split the grown tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001049 */
1050 depth = ext_depth(inode);
1051 if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001052 /* now we need to split */
Alex Tomasa86c6182006-10-11 01:21:03 -07001053 goto repeat;
1054 }
1055 }
1056
1057out:
1058 return err;
1059}
1060
1061/*
Alex Tomas1988b512008-01-28 23:58:27 -05001062 * search the closest allocated block to the left for *logical
1063 * and returns it at @logical + it's physical address at @phys
1064 * if *logical is the smallest allocated block, the function
1065 * returns 0 at @phys
1066 * return value contains 0 (success) or error code
1067 */
1068int
1069ext4_ext_search_left(struct inode *inode, struct ext4_ext_path *path,
1070 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1071{
1072 struct ext4_extent_idx *ix;
1073 struct ext4_extent *ex;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001074 int depth, ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001075
1076 BUG_ON(path == NULL);
1077 depth = path->p_depth;
1078 *phys = 0;
1079
1080 if (depth == 0 && path->p_ext == NULL)
1081 return 0;
1082
1083 /* usually extent in the path covers blocks smaller
1084 * then *logical, but it can be that extent is the
1085 * first one in the file */
1086
1087 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001088 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001089 if (*logical < le32_to_cpu(ex->ee_block)) {
1090 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1091 while (--depth >= 0) {
1092 ix = path[depth].p_idx;
1093 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1094 }
1095 return 0;
1096 }
1097
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001098 BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
Alex Tomas1988b512008-01-28 23:58:27 -05001099
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001100 *logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1101 *phys = ext_pblock(ex) + ee_len - 1;
Alex Tomas1988b512008-01-28 23:58:27 -05001102 return 0;
1103}
1104
1105/*
1106 * search the closest allocated block to the right for *logical
1107 * and returns it at @logical + it's physical address at @phys
1108 * if *logical is the smallest allocated block, the function
1109 * returns 0 at @phys
1110 * return value contains 0 (success) or error code
1111 */
1112int
1113ext4_ext_search_right(struct inode *inode, struct ext4_ext_path *path,
1114 ext4_lblk_t *logical, ext4_fsblk_t *phys)
1115{
1116 struct buffer_head *bh = NULL;
1117 struct ext4_extent_header *eh;
1118 struct ext4_extent_idx *ix;
1119 struct ext4_extent *ex;
1120 ext4_fsblk_t block;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001121 int depth, ee_len;
Alex Tomas1988b512008-01-28 23:58:27 -05001122
1123 BUG_ON(path == NULL);
1124 depth = path->p_depth;
1125 *phys = 0;
1126
1127 if (depth == 0 && path->p_ext == NULL)
1128 return 0;
1129
1130 /* usually extent in the path covers blocks smaller
1131 * then *logical, but it can be that extent is the
1132 * first one in the file */
1133
1134 ex = path[depth].p_ext;
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001135 ee_len = ext4_ext_get_actual_len(ex);
Alex Tomas1988b512008-01-28 23:58:27 -05001136 if (*logical < le32_to_cpu(ex->ee_block)) {
1137 BUG_ON(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex);
1138 while (--depth >= 0) {
1139 ix = path[depth].p_idx;
1140 BUG_ON(ix != EXT_FIRST_INDEX(path[depth].p_hdr));
1141 }
1142 *logical = le32_to_cpu(ex->ee_block);
1143 *phys = ext_pblock(ex);
1144 return 0;
1145 }
1146
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001147 BUG_ON(*logical < (le32_to_cpu(ex->ee_block) + ee_len));
Alex Tomas1988b512008-01-28 23:58:27 -05001148
1149 if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1150 /* next allocated block in this leaf */
1151 ex++;
1152 *logical = le32_to_cpu(ex->ee_block);
1153 *phys = ext_pblock(ex);
1154 return 0;
1155 }
1156
1157 /* go up and search for index to the right */
1158 while (--depth >= 0) {
1159 ix = path[depth].p_idx;
1160 if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1161 break;
1162 }
1163
1164 if (depth < 0) {
1165 /* we've gone up to the root and
1166 * found no index to the right */
1167 return 0;
1168 }
1169
1170 /* we've found index to the right, let's
1171 * follow it and find the closest allocated
1172 * block to the right */
1173 ix++;
1174 block = idx_pblock(ix);
1175 while (++depth < path->p_depth) {
1176 bh = sb_bread(inode->i_sb, block);
1177 if (bh == NULL)
1178 return -EIO;
1179 eh = ext_block_hdr(bh);
1180 if (ext4_ext_check_header(inode, eh, depth)) {
1181 put_bh(bh);
1182 return -EIO;
1183 }
1184 ix = EXT_FIRST_INDEX(eh);
1185 block = idx_pblock(ix);
1186 put_bh(bh);
1187 }
1188
1189 bh = sb_bread(inode->i_sb, block);
1190 if (bh == NULL)
1191 return -EIO;
1192 eh = ext_block_hdr(bh);
1193 if (ext4_ext_check_header(inode, eh, path->p_depth - depth)) {
1194 put_bh(bh);
1195 return -EIO;
1196 }
1197 ex = EXT_FIRST_EXTENT(eh);
1198 *logical = le32_to_cpu(ex->ee_block);
1199 *phys = ext_pblock(ex);
1200 put_bh(bh);
1201 return 0;
1202
1203}
1204
1205/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001206 * ext4_ext_next_allocated_block:
1207 * returns allocated block in subsequent extent or EXT_MAX_BLOCK.
1208 * NOTE: it considers block number from index entry as
1209 * allocated block. Thus, index entries have to be consistent
1210 * with leaves.
Alex Tomasa86c6182006-10-11 01:21:03 -07001211 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001212static ext4_lblk_t
Alex Tomasa86c6182006-10-11 01:21:03 -07001213ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1214{
1215 int depth;
1216
1217 BUG_ON(path == NULL);
1218 depth = path->p_depth;
1219
1220 if (depth == 0 && path->p_ext == NULL)
1221 return EXT_MAX_BLOCK;
1222
1223 while (depth >= 0) {
1224 if (depth == path->p_depth) {
1225 /* leaf */
1226 if (path[depth].p_ext !=
1227 EXT_LAST_EXTENT(path[depth].p_hdr))
1228 return le32_to_cpu(path[depth].p_ext[1].ee_block);
1229 } else {
1230 /* index */
1231 if (path[depth].p_idx !=
1232 EXT_LAST_INDEX(path[depth].p_hdr))
1233 return le32_to_cpu(path[depth].p_idx[1].ei_block);
1234 }
1235 depth--;
1236 }
1237
1238 return EXT_MAX_BLOCK;
1239}
1240
1241/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001242 * ext4_ext_next_leaf_block:
Alex Tomasa86c6182006-10-11 01:21:03 -07001243 * returns first allocated block from next leaf or EXT_MAX_BLOCK
1244 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001245static ext4_lblk_t ext4_ext_next_leaf_block(struct inode *inode,
Andrew Morton63f57932006-10-11 01:21:24 -07001246 struct ext4_ext_path *path)
Alex Tomasa86c6182006-10-11 01:21:03 -07001247{
1248 int depth;
1249
1250 BUG_ON(path == NULL);
1251 depth = path->p_depth;
1252
1253 /* zero-tree has no leaf blocks at all */
1254 if (depth == 0)
1255 return EXT_MAX_BLOCK;
1256
1257 /* go to index block */
1258 depth--;
1259
1260 while (depth >= 0) {
1261 if (path[depth].p_idx !=
1262 EXT_LAST_INDEX(path[depth].p_hdr))
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001263 return (ext4_lblk_t)
1264 le32_to_cpu(path[depth].p_idx[1].ei_block);
Alex Tomasa86c6182006-10-11 01:21:03 -07001265 depth--;
1266 }
1267
1268 return EXT_MAX_BLOCK;
1269}
1270
1271/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001272 * ext4_ext_correct_indexes:
1273 * if leaf gets modified and modified extent is first in the leaf,
1274 * then we have to correct all indexes above.
Alex Tomasa86c6182006-10-11 01:21:03 -07001275 * TODO: do we need to correct tree in all cases?
1276 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001277static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001278 struct ext4_ext_path *path)
1279{
1280 struct ext4_extent_header *eh;
1281 int depth = ext_depth(inode);
1282 struct ext4_extent *ex;
1283 __le32 border;
1284 int k, err = 0;
1285
1286 eh = path[depth].p_hdr;
1287 ex = path[depth].p_ext;
1288 BUG_ON(ex == NULL);
1289 BUG_ON(eh == NULL);
1290
1291 if (depth == 0) {
1292 /* there is no tree at all */
1293 return 0;
1294 }
1295
1296 if (ex != EXT_FIRST_EXTENT(eh)) {
1297 /* we correct tree if first leaf got modified only */
1298 return 0;
1299 }
1300
1301 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001302 * TODO: we need correction if border is smaller than current one
Alex Tomasa86c6182006-10-11 01:21:03 -07001303 */
1304 k = depth - 1;
1305 border = path[depth].p_ext->ee_block;
Avantika Mathur7e028972006-12-06 20:41:33 -08001306 err = ext4_ext_get_access(handle, inode, path + k);
1307 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001308 return err;
1309 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001310 err = ext4_ext_dirty(handle, inode, path + k);
1311 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001312 return err;
1313
1314 while (k--) {
1315 /* change all left-side indexes */
1316 if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1317 break;
Avantika Mathur7e028972006-12-06 20:41:33 -08001318 err = ext4_ext_get_access(handle, inode, path + k);
1319 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001320 break;
1321 path[k].p_idx->ei_block = border;
Avantika Mathur7e028972006-12-06 20:41:33 -08001322 err = ext4_ext_dirty(handle, inode, path + k);
1323 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001324 break;
1325 }
1326
1327 return err;
1328}
1329
Avantika Mathur09b88252006-12-06 20:41:36 -08001330static int
Alex Tomasa86c6182006-10-11 01:21:03 -07001331ext4_can_extents_be_merged(struct inode *inode, struct ext4_extent *ex1,
1332 struct ext4_extent *ex2)
1333{
Amit Arora749269f2007-07-18 09:02:56 -04001334 unsigned short ext1_ee_len, ext2_ee_len, max_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001335
1336 /*
1337 * Make sure that either both extents are uninitialized, or
1338 * both are _not_.
1339 */
1340 if (ext4_ext_is_uninitialized(ex1) ^ ext4_ext_is_uninitialized(ex2))
1341 return 0;
1342
Amit Arora749269f2007-07-18 09:02:56 -04001343 if (ext4_ext_is_uninitialized(ex1))
1344 max_len = EXT_UNINIT_MAX_LEN;
1345 else
1346 max_len = EXT_INIT_MAX_LEN;
1347
Amit Aroraa2df2a62007-07-17 21:42:41 -04001348 ext1_ee_len = ext4_ext_get_actual_len(ex1);
1349 ext2_ee_len = ext4_ext_get_actual_len(ex2);
1350
1351 if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
Andrew Morton63f57932006-10-11 01:21:24 -07001352 le32_to_cpu(ex2->ee_block))
Alex Tomasa86c6182006-10-11 01:21:03 -07001353 return 0;
1354
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001355 /*
1356 * To allow future support for preallocated extents to be added
1357 * as an RO_COMPAT feature, refuse to merge to extents if
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001358 * this can result in the top bit of ee_len being set.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001359 */
Amit Arora749269f2007-07-18 09:02:56 -04001360 if (ext1_ee_len + ext2_ee_len > max_len)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07001361 return 0;
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01001362#ifdef AGGRESSIVE_TEST
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05001363 if (ext1_ee_len >= 4)
Alex Tomasa86c6182006-10-11 01:21:03 -07001364 return 0;
1365#endif
1366
Amit Aroraa2df2a62007-07-17 21:42:41 -04001367 if (ext_pblock(ex1) + ext1_ee_len == ext_pblock(ex2))
Alex Tomasa86c6182006-10-11 01:21:03 -07001368 return 1;
1369 return 0;
1370}
1371
1372/*
Amit Arora56055d32007-07-17 21:42:38 -04001373 * This function tries to merge the "ex" extent to the next extent in the tree.
1374 * It always tries to merge towards right. If you want to merge towards
1375 * left, pass "ex - 1" as argument instead of "ex".
1376 * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1377 * 1 if they got merged.
1378 */
1379int ext4_ext_try_to_merge(struct inode *inode,
1380 struct ext4_ext_path *path,
1381 struct ext4_extent *ex)
1382{
1383 struct ext4_extent_header *eh;
1384 unsigned int depth, len;
1385 int merge_done = 0;
1386 int uninitialized = 0;
1387
1388 depth = ext_depth(inode);
1389 BUG_ON(path[depth].p_hdr == NULL);
1390 eh = path[depth].p_hdr;
1391
1392 while (ex < EXT_LAST_EXTENT(eh)) {
1393 if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1394 break;
1395 /* merge with next extent! */
1396 if (ext4_ext_is_uninitialized(ex))
1397 uninitialized = 1;
1398 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1399 + ext4_ext_get_actual_len(ex + 1));
1400 if (uninitialized)
1401 ext4_ext_mark_uninitialized(ex);
1402
1403 if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1404 len = (EXT_LAST_EXTENT(eh) - ex - 1)
1405 * sizeof(struct ext4_extent);
1406 memmove(ex + 1, ex + 2, len);
1407 }
Marcin Slusarze8546d02008-04-17 10:38:59 -04001408 le16_add_cpu(&eh->eh_entries, -1);
Amit Arora56055d32007-07-17 21:42:38 -04001409 merge_done = 1;
1410 WARN_ON(eh->eh_entries == 0);
1411 if (!eh->eh_entries)
1412 ext4_error(inode->i_sb, "ext4_ext_try_to_merge",
1413 "inode#%lu, eh->eh_entries = 0!", inode->i_ino);
1414 }
1415
1416 return merge_done;
1417}
1418
1419/*
Amit Arora25d14f92007-05-24 13:04:13 -04001420 * check if a portion of the "newext" extent overlaps with an
1421 * existing extent.
1422 *
1423 * If there is an overlap discovered, it updates the length of the newext
1424 * such that there will be no overlap, and then returns 1.
1425 * If there is no overlap found, it returns 0.
1426 */
1427unsigned int ext4_ext_check_overlap(struct inode *inode,
1428 struct ext4_extent *newext,
1429 struct ext4_ext_path *path)
1430{
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001431 ext4_lblk_t b1, b2;
Amit Arora25d14f92007-05-24 13:04:13 -04001432 unsigned int depth, len1;
1433 unsigned int ret = 0;
1434
1435 b1 = le32_to_cpu(newext->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001436 len1 = ext4_ext_get_actual_len(newext);
Amit Arora25d14f92007-05-24 13:04:13 -04001437 depth = ext_depth(inode);
1438 if (!path[depth].p_ext)
1439 goto out;
1440 b2 = le32_to_cpu(path[depth].p_ext->ee_block);
1441
1442 /*
1443 * get the next allocated block if the extent in the path
1444 * is before the requested block(s)
1445 */
1446 if (b2 < b1) {
1447 b2 = ext4_ext_next_allocated_block(path);
1448 if (b2 == EXT_MAX_BLOCK)
1449 goto out;
1450 }
1451
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001452 /* check for wrap through zero on extent logical start block*/
Amit Arora25d14f92007-05-24 13:04:13 -04001453 if (b1 + len1 < b1) {
1454 len1 = EXT_MAX_BLOCK - b1;
1455 newext->ee_len = cpu_to_le16(len1);
1456 ret = 1;
1457 }
1458
1459 /* check for overlap */
1460 if (b1 + len1 > b2) {
1461 newext->ee_len = cpu_to_le16(b2 - b1);
1462 ret = 1;
1463 }
1464out:
1465 return ret;
1466}
1467
1468/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001469 * ext4_ext_insert_extent:
1470 * tries to merge requsted extent into the existing extent or
1471 * inserts requested extent as new one into the tree,
1472 * creating new leaf in the no-space case.
Alex Tomasa86c6182006-10-11 01:21:03 -07001473 */
1474int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1475 struct ext4_ext_path *path,
1476 struct ext4_extent *newext)
1477{
1478 struct ext4_extent_header * eh;
1479 struct ext4_extent *ex, *fex;
1480 struct ext4_extent *nearex; /* nearest extent */
1481 struct ext4_ext_path *npath = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001482 int depth, len, err;
1483 ext4_lblk_t next;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001484 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001485
Amit Aroraa2df2a62007-07-17 21:42:41 -04001486 BUG_ON(ext4_ext_get_actual_len(newext) == 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07001487 depth = ext_depth(inode);
1488 ex = path[depth].p_ext;
1489 BUG_ON(path[depth].p_hdr == NULL);
1490
1491 /* try to insert block into found extent and return */
1492 if (ex && ext4_can_extents_be_merged(inode, ex, newext)) {
Mingming Cao2ae02102006-10-11 01:21:11 -07001493 ext_debug("append %d block to %d:%d (from %llu)\n",
Amit Aroraa2df2a62007-07-17 21:42:41 -04001494 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001495 le32_to_cpu(ex->ee_block),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001496 ext4_ext_get_actual_len(ex), ext_pblock(ex));
Avantika Mathur7e028972006-12-06 20:41:33 -08001497 err = ext4_ext_get_access(handle, inode, path + depth);
1498 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001499 return err;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001500
1501 /*
1502 * ext4_can_extents_be_merged should have checked that either
1503 * both extents are uninitialized, or both aren't. Thus we
1504 * need to check only one of them here.
1505 */
1506 if (ext4_ext_is_uninitialized(ex))
1507 uninitialized = 1;
1508 ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1509 + ext4_ext_get_actual_len(newext));
1510 if (uninitialized)
1511 ext4_ext_mark_uninitialized(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001512 eh = path[depth].p_hdr;
1513 nearex = ex;
1514 goto merge;
1515 }
1516
1517repeat:
1518 depth = ext_depth(inode);
1519 eh = path[depth].p_hdr;
1520 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
1521 goto has_space;
1522
1523 /* probably next leaf has space for us? */
1524 fex = EXT_LAST_EXTENT(eh);
1525 next = ext4_ext_next_leaf_block(inode, path);
1526 if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block)
1527 && next != EXT_MAX_BLOCK) {
1528 ext_debug("next leaf block - %d\n", next);
1529 BUG_ON(npath != NULL);
1530 npath = ext4_ext_find_extent(inode, next, NULL);
1531 if (IS_ERR(npath))
1532 return PTR_ERR(npath);
1533 BUG_ON(npath->p_depth != path->p_depth);
1534 eh = npath[depth].p_hdr;
1535 if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
1536 ext_debug("next leaf isnt full(%d)\n",
1537 le16_to_cpu(eh->eh_entries));
1538 path = npath;
1539 goto repeat;
1540 }
1541 ext_debug("next leaf has no free space(%d,%d)\n",
1542 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
1543 }
1544
1545 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001546 * There is no free space in the found leaf.
1547 * We're gonna add a new leaf in the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -07001548 */
1549 err = ext4_ext_create_new_leaf(handle, inode, path, newext);
1550 if (err)
1551 goto cleanup;
1552 depth = ext_depth(inode);
1553 eh = path[depth].p_hdr;
1554
1555has_space:
1556 nearex = path[depth].p_ext;
1557
Avantika Mathur7e028972006-12-06 20:41:33 -08001558 err = ext4_ext_get_access(handle, inode, path + depth);
1559 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001560 goto cleanup;
1561
1562 if (!nearex) {
1563 /* there is no extent in this leaf, create first one */
Mingming Cao2ae02102006-10-11 01:21:11 -07001564 ext_debug("first extent in the leaf: %d:%llu:%d\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001565 le32_to_cpu(newext->ee_block),
1566 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001567 ext4_ext_get_actual_len(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07001568 path[depth].p_ext = EXT_FIRST_EXTENT(eh);
1569 } else if (le32_to_cpu(newext->ee_block)
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001570 > le32_to_cpu(nearex->ee_block)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001571/* BUG_ON(newext->ee_block == nearex->ee_block); */
1572 if (nearex != EXT_LAST_EXTENT(eh)) {
1573 len = EXT_MAX_EXTENT(eh) - nearex;
1574 len = (len - 1) * sizeof(struct ext4_extent);
1575 len = len < 0 ? 0 : len;
Mingming Cao2ae02102006-10-11 01:21:11 -07001576 ext_debug("insert %d:%llu:%d after: nearest 0x%p, "
Alex Tomasa86c6182006-10-11 01:21:03 -07001577 "move %d from 0x%p to 0x%p\n",
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001578 le32_to_cpu(newext->ee_block),
1579 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001580 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001581 nearex, len, nearex + 1, nearex + 2);
1582 memmove(nearex + 2, nearex + 1, len);
1583 }
1584 path[depth].p_ext = nearex + 1;
1585 } else {
1586 BUG_ON(newext->ee_block == nearex->ee_block);
1587 len = (EXT_MAX_EXTENT(eh) - nearex) * sizeof(struct ext4_extent);
1588 len = len < 0 ? 0 : len;
Mingming Cao2ae02102006-10-11 01:21:11 -07001589 ext_debug("insert %d:%llu:%d before: nearest 0x%p, "
Alex Tomasa86c6182006-10-11 01:21:03 -07001590 "move %d from 0x%p to 0x%p\n",
1591 le32_to_cpu(newext->ee_block),
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001592 ext_pblock(newext),
Amit Aroraa2df2a62007-07-17 21:42:41 -04001593 ext4_ext_get_actual_len(newext),
Alex Tomasa86c6182006-10-11 01:21:03 -07001594 nearex, len, nearex + 1, nearex + 2);
1595 memmove(nearex + 1, nearex, len);
1596 path[depth].p_ext = nearex;
1597 }
1598
Marcin Slusarze8546d02008-04-17 10:38:59 -04001599 le16_add_cpu(&eh->eh_entries, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07001600 nearex = path[depth].p_ext;
1601 nearex->ee_block = newext->ee_block;
Aneesh Kumar K.Vb3776112007-10-16 18:38:25 -04001602 ext4_ext_store_pblock(nearex, ext_pblock(newext));
Alex Tomasa86c6182006-10-11 01:21:03 -07001603 nearex->ee_len = newext->ee_len;
Alex Tomasa86c6182006-10-11 01:21:03 -07001604
1605merge:
1606 /* try to merge extents to the right */
Amit Arora56055d32007-07-17 21:42:38 -04001607 ext4_ext_try_to_merge(inode, path, nearex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001608
1609 /* try to merge extents to the left */
1610
1611 /* time to correct all indexes above */
1612 err = ext4_ext_correct_indexes(handle, inode, path);
1613 if (err)
1614 goto cleanup;
1615
1616 err = ext4_ext_dirty(handle, inode, path + depth);
1617
1618cleanup:
1619 if (npath) {
1620 ext4_ext_drop_refs(npath);
1621 kfree(npath);
1622 }
1623 ext4_ext_tree_changed(inode);
1624 ext4_ext_invalidate_cache(inode);
1625 return err;
1626}
1627
Avantika Mathur09b88252006-12-06 20:41:36 -08001628static void
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001629ext4_ext_put_in_cache(struct inode *inode, ext4_lblk_t block,
Mingming Caodd545672007-07-31 00:37:46 -07001630 __u32 len, ext4_fsblk_t start, int type)
Alex Tomasa86c6182006-10-11 01:21:03 -07001631{
1632 struct ext4_ext_cache *cex;
1633 BUG_ON(len == 0);
1634 cex = &EXT4_I(inode)->i_cached_extent;
1635 cex->ec_type = type;
1636 cex->ec_block = block;
1637 cex->ec_len = len;
1638 cex->ec_start = start;
1639}
1640
1641/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001642 * ext4_ext_put_gap_in_cache:
1643 * calculate boundaries of the gap that the requested block fits into
Alex Tomasa86c6182006-10-11 01:21:03 -07001644 * and cache this gap
1645 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001646static void
Alex Tomasa86c6182006-10-11 01:21:03 -07001647ext4_ext_put_gap_in_cache(struct inode *inode, struct ext4_ext_path *path,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001648 ext4_lblk_t block)
Alex Tomasa86c6182006-10-11 01:21:03 -07001649{
1650 int depth = ext_depth(inode);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001651 unsigned long len;
1652 ext4_lblk_t lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001653 struct ext4_extent *ex;
1654
1655 ex = path[depth].p_ext;
1656 if (ex == NULL) {
1657 /* there is no extent yet, so gap is [0;-] */
1658 lblock = 0;
1659 len = EXT_MAX_BLOCK;
1660 ext_debug("cache gap(whole file):");
1661 } else if (block < le32_to_cpu(ex->ee_block)) {
1662 lblock = block;
1663 len = le32_to_cpu(ex->ee_block) - block;
Eric Sandeenbba90742008-01-28 23:58:27 -05001664 ext_debug("cache gap(before): %u [%u:%u]",
1665 block,
1666 le32_to_cpu(ex->ee_block),
1667 ext4_ext_get_actual_len(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07001668 } else if (block >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001669 + ext4_ext_get_actual_len(ex)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001670 ext4_lblk_t next;
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001671 lblock = le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001672 + ext4_ext_get_actual_len(ex);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001673
1674 next = ext4_ext_next_allocated_block(path);
Eric Sandeenbba90742008-01-28 23:58:27 -05001675 ext_debug("cache gap(after): [%u:%u] %u",
1676 le32_to_cpu(ex->ee_block),
1677 ext4_ext_get_actual_len(ex),
1678 block);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001679 BUG_ON(next == lblock);
1680 len = next - lblock;
Alex Tomasa86c6182006-10-11 01:21:03 -07001681 } else {
1682 lblock = len = 0;
1683 BUG();
1684 }
1685
Eric Sandeenbba90742008-01-28 23:58:27 -05001686 ext_debug(" -> %u:%lu\n", lblock, len);
Alex Tomasa86c6182006-10-11 01:21:03 -07001687 ext4_ext_put_in_cache(inode, lblock, len, 0, EXT4_EXT_CACHE_GAP);
1688}
1689
Avantika Mathur09b88252006-12-06 20:41:36 -08001690static int
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001691ext4_ext_in_cache(struct inode *inode, ext4_lblk_t block,
Alex Tomasa86c6182006-10-11 01:21:03 -07001692 struct ext4_extent *ex)
1693{
1694 struct ext4_ext_cache *cex;
1695
1696 cex = &EXT4_I(inode)->i_cached_extent;
1697
1698 /* has cache valid data? */
1699 if (cex->ec_type == EXT4_EXT_CACHE_NO)
1700 return EXT4_EXT_CACHE_NO;
1701
1702 BUG_ON(cex->ec_type != EXT4_EXT_CACHE_GAP &&
1703 cex->ec_type != EXT4_EXT_CACHE_EXTENT);
1704 if (block >= cex->ec_block && block < cex->ec_block + cex->ec_len) {
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001705 ex->ee_block = cpu_to_le32(cex->ec_block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001706 ext4_ext_store_pblock(ex, cex->ec_start);
Dave Kleikamp8c55e202007-05-24 13:04:54 -04001707 ex->ee_len = cpu_to_le16(cex->ec_len);
Eric Sandeenbba90742008-01-28 23:58:27 -05001708 ext_debug("%u cached by %u:%u:%llu\n",
1709 block,
1710 cex->ec_block, cex->ec_len, cex->ec_start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001711 return cex->ec_type;
1712 }
1713
1714 /* not in cache */
1715 return EXT4_EXT_CACHE_NO;
1716}
1717
1718/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001719 * ext4_ext_rm_idx:
1720 * removes index from the index block.
1721 * It's used in truncate case only, thus all requests are for
1722 * last index in the block only.
Alex Tomasa86c6182006-10-11 01:21:03 -07001723 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001724static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001725 struct ext4_ext_path *path)
1726{
1727 struct buffer_head *bh;
1728 int err;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001729 ext4_fsblk_t leaf;
Alex Tomasa86c6182006-10-11 01:21:03 -07001730
1731 /* free index block */
1732 path--;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001733 leaf = idx_pblock(path->p_idx);
Alex Tomasa86c6182006-10-11 01:21:03 -07001734 BUG_ON(path->p_hdr->eh_entries == 0);
Avantika Mathur7e028972006-12-06 20:41:33 -08001735 err = ext4_ext_get_access(handle, inode, path);
1736 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001737 return err;
Marcin Slusarze8546d02008-04-17 10:38:59 -04001738 le16_add_cpu(&path->p_hdr->eh_entries, -1);
Avantika Mathur7e028972006-12-06 20:41:33 -08001739 err = ext4_ext_dirty(handle, inode, path);
1740 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001741 return err;
Mingming Cao2ae02102006-10-11 01:21:11 -07001742 ext_debug("index is empty, remove it, free block %llu\n", leaf);
Alex Tomasa86c6182006-10-11 01:21:03 -07001743 bh = sb_find_get_block(inode->i_sb, leaf);
1744 ext4_forget(handle, 1, inode, bh, leaf);
Alex Tomasc9de5602008-01-29 00:19:52 -05001745 ext4_free_blocks(handle, inode, leaf, 1, 1);
Alex Tomasa86c6182006-10-11 01:21:03 -07001746 return err;
1747}
1748
1749/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001750 * ext4_ext_calc_credits_for_insert:
1751 * This routine returns max. credits that the extent tree can consume.
Alex Tomasa86c6182006-10-11 01:21:03 -07001752 * It should be OK for low-performance paths like ->writepage()
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001753 * To allow many writing processes to fit into a single transaction,
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05001754 * the caller should calculate credits under i_data_sem and
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001755 * pass the actual path.
Alex Tomasa86c6182006-10-11 01:21:03 -07001756 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001757int ext4_ext_calc_credits_for_insert(struct inode *inode,
Alex Tomasa86c6182006-10-11 01:21:03 -07001758 struct ext4_ext_path *path)
1759{
1760 int depth, needed;
1761
1762 if (path) {
1763 /* probably there is space in leaf? */
1764 depth = ext_depth(inode);
1765 if (le16_to_cpu(path[depth].p_hdr->eh_entries)
1766 < le16_to_cpu(path[depth].p_hdr->eh_max))
1767 return 1;
1768 }
1769
1770 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001771 * given 32-bit logical block (4294967296 blocks), max. tree
Alex Tomasa86c6182006-10-11 01:21:03 -07001772 * can be 4 levels in depth -- 4 * 340^4 == 53453440000.
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001773 * Let's also add one more level for imbalance.
Alex Tomasa86c6182006-10-11 01:21:03 -07001774 */
1775 depth = 5;
1776
1777 /* allocation of new data block(s) */
1778 needed = 2;
1779
1780 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001781 * tree can be full, so it would need to grow in depth:
Johann Lombardifeb18922006-12-06 20:40:46 -08001782 * we need one credit to modify old root, credits for
1783 * new root will be added in split accounting
Alex Tomasa86c6182006-10-11 01:21:03 -07001784 */
Johann Lombardifeb18922006-12-06 20:40:46 -08001785 needed += 1;
Alex Tomasa86c6182006-10-11 01:21:03 -07001786
1787 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001788 * Index split can happen, we would need:
Alex Tomasa86c6182006-10-11 01:21:03 -07001789 * allocate intermediate indexes (bitmap + group)
1790 * + change two blocks at each level, but root (already included)
1791 */
Johann Lombardifeb18922006-12-06 20:40:46 -08001792 needed += (depth * 2) + (depth * 2);
Alex Tomasa86c6182006-10-11 01:21:03 -07001793
1794 /* any allocation modifies superblock */
1795 needed += 1;
1796
1797 return needed;
1798}
1799
1800static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
1801 struct ext4_extent *ex,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001802 ext4_lblk_t from, ext4_lblk_t to)
Alex Tomasa86c6182006-10-11 01:21:03 -07001803{
1804 struct buffer_head *bh;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001805 unsigned short ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasc9de5602008-01-29 00:19:52 -05001806 int i, metadata = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001807
Alex Tomasc9de5602008-01-29 00:19:52 -05001808 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1809 metadata = 1;
Alex Tomasa86c6182006-10-11 01:21:03 -07001810#ifdef EXTENTS_STATS
1811 {
1812 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Alex Tomasa86c6182006-10-11 01:21:03 -07001813 spin_lock(&sbi->s_ext_stats_lock);
1814 sbi->s_ext_blocks += ee_len;
1815 sbi->s_ext_extents++;
1816 if (ee_len < sbi->s_ext_min)
1817 sbi->s_ext_min = ee_len;
1818 if (ee_len > sbi->s_ext_max)
1819 sbi->s_ext_max = ee_len;
1820 if (ext_depth(inode) > sbi->s_depth_max)
1821 sbi->s_depth_max = ext_depth(inode);
1822 spin_unlock(&sbi->s_ext_stats_lock);
1823 }
1824#endif
1825 if (from >= le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001826 && to == le32_to_cpu(ex->ee_block) + ee_len - 1) {
Alex Tomasa86c6182006-10-11 01:21:03 -07001827 /* tail removal */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001828 ext4_lblk_t num;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001829 ext4_fsblk_t start;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001830
Amit Aroraa2df2a62007-07-17 21:42:41 -04001831 num = le32_to_cpu(ex->ee_block) + ee_len - from;
1832 start = ext_pblock(ex) + ee_len - num;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001833 ext_debug("free last %u blocks starting %llu\n", num, start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001834 for (i = 0; i < num; i++) {
1835 bh = sb_find_get_block(inode->i_sb, start + i);
1836 ext4_forget(handle, 0, inode, bh, start + i);
1837 }
Alex Tomasc9de5602008-01-29 00:19:52 -05001838 ext4_free_blocks(handle, inode, start, num, metadata);
Alex Tomasa86c6182006-10-11 01:21:03 -07001839 } else if (from == le32_to_cpu(ex->ee_block)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001840 && to <= le32_to_cpu(ex->ee_block) + ee_len - 1) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001841 printk(KERN_INFO "strange request: removal %u-%u from %u:%u\n",
Amit Aroraa2df2a62007-07-17 21:42:41 -04001842 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07001843 } else {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001844 printk(KERN_INFO "strange request: removal(2) "
1845 "%u-%u from %u:%u\n",
1846 from, to, le32_to_cpu(ex->ee_block), ee_len);
Alex Tomasa86c6182006-10-11 01:21:03 -07001847 }
1848 return 0;
1849}
1850
1851static int
1852ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001853 struct ext4_ext_path *path, ext4_lblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07001854{
1855 int err = 0, correct_index = 0;
1856 int depth = ext_depth(inode), credits;
1857 struct ext4_extent_header *eh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001858 ext4_lblk_t a, b, block;
1859 unsigned num;
1860 ext4_lblk_t ex_ee_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07001861 unsigned short ex_ee_len;
Amit Aroraa2df2a62007-07-17 21:42:41 -04001862 unsigned uninitialized = 0;
Alex Tomasa86c6182006-10-11 01:21:03 -07001863 struct ext4_extent *ex;
1864
Alex Tomasc29c0ae2007-07-18 09:19:09 -04001865 /* the header must be checked already in ext4_ext_remove_space() */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001866 ext_debug("truncate since %u in leaf\n", start);
Alex Tomasa86c6182006-10-11 01:21:03 -07001867 if (!path[depth].p_hdr)
1868 path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
1869 eh = path[depth].p_hdr;
1870 BUG_ON(eh == NULL);
Alex Tomasa86c6182006-10-11 01:21:03 -07001871
1872 /* find where to start removing */
1873 ex = EXT_LAST_EXTENT(eh);
1874
1875 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001876 if (ext4_ext_is_uninitialized(ex))
1877 uninitialized = 1;
1878 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001879
1880 while (ex >= EXT_FIRST_EXTENT(eh) &&
1881 ex_ee_block + ex_ee_len > start) {
1882 ext_debug("remove ext %lu:%u\n", ex_ee_block, ex_ee_len);
1883 path[depth].p_ext = ex;
1884
1885 a = ex_ee_block > start ? ex_ee_block : start;
1886 b = ex_ee_block + ex_ee_len - 1 < EXT_MAX_BLOCK ?
1887 ex_ee_block + ex_ee_len - 1 : EXT_MAX_BLOCK;
1888
1889 ext_debug(" border %u:%u\n", a, b);
1890
1891 if (a != ex_ee_block && b != ex_ee_block + ex_ee_len - 1) {
1892 block = 0;
1893 num = 0;
1894 BUG();
1895 } else if (a != ex_ee_block) {
1896 /* remove tail of the extent */
1897 block = ex_ee_block;
1898 num = a - block;
1899 } else if (b != ex_ee_block + ex_ee_len - 1) {
1900 /* remove head of the extent */
1901 block = a;
1902 num = b - a;
1903 /* there is no "make a hole" API yet */
1904 BUG();
1905 } else {
1906 /* remove whole extent: excellent! */
1907 block = ex_ee_block;
1908 num = 0;
1909 BUG_ON(a != ex_ee_block);
1910 BUG_ON(b != ex_ee_block + ex_ee_len - 1);
1911 }
1912
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001913 /* at present, extent can't cross block group: */
Alex Tomasa86c6182006-10-11 01:21:03 -07001914 /* leaf + bitmap + group desc + sb + inode */
1915 credits = 5;
1916 if (ex == EXT_FIRST_EXTENT(eh)) {
1917 correct_index = 1;
1918 credits += (ext_depth(inode)) + 1;
1919 }
1920#ifdef CONFIG_QUOTA
1921 credits += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
1922#endif
1923
Shen Feng9102e4f2008-07-11 19:27:31 -04001924 err = ext4_ext_journal_restart(handle, credits);
1925 if (err)
Alex Tomasa86c6182006-10-11 01:21:03 -07001926 goto out;
Alex Tomasa86c6182006-10-11 01:21:03 -07001927
1928 err = ext4_ext_get_access(handle, inode, path + depth);
1929 if (err)
1930 goto out;
1931
1932 err = ext4_remove_blocks(handle, inode, ex, a, b);
1933 if (err)
1934 goto out;
1935
1936 if (num == 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001937 /* this extent is removed; mark slot entirely unused */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001938 ext4_ext_store_pblock(ex, 0);
Marcin Slusarze8546d02008-04-17 10:38:59 -04001939 le16_add_cpu(&eh->eh_entries, -1);
Alex Tomasa86c6182006-10-11 01:21:03 -07001940 }
1941
1942 ex->ee_block = cpu_to_le32(block);
1943 ex->ee_len = cpu_to_le16(num);
Amit Arora749269f2007-07-18 09:02:56 -04001944 /*
1945 * Do not mark uninitialized if all the blocks in the
1946 * extent have been removed.
1947 */
1948 if (uninitialized && num)
Amit Aroraa2df2a62007-07-17 21:42:41 -04001949 ext4_ext_mark_uninitialized(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001950
1951 err = ext4_ext_dirty(handle, inode, path + depth);
1952 if (err)
1953 goto out;
1954
Mingming Cao2ae02102006-10-11 01:21:11 -07001955 ext_debug("new extent: %u:%u:%llu\n", block, num,
Alex Tomasf65e6fb2006-10-11 01:21:05 -07001956 ext_pblock(ex));
Alex Tomasa86c6182006-10-11 01:21:03 -07001957 ex--;
1958 ex_ee_block = le32_to_cpu(ex->ee_block);
Amit Aroraa2df2a62007-07-17 21:42:41 -04001959 ex_ee_len = ext4_ext_get_actual_len(ex);
Alex Tomasa86c6182006-10-11 01:21:03 -07001960 }
1961
1962 if (correct_index && eh->eh_entries)
1963 err = ext4_ext_correct_indexes(handle, inode, path);
1964
1965 /* if this leaf is free, then we should
1966 * remove it from index block above */
1967 if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
1968 err = ext4_ext_rm_idx(handle, inode, path + depth);
1969
1970out:
1971 return err;
1972}
1973
1974/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001975 * ext4_ext_more_to_rm:
1976 * returns 1 if current index has to be freed (even partial)
Alex Tomasa86c6182006-10-11 01:21:03 -07001977 */
Avantika Mathur09b88252006-12-06 20:41:36 -08001978static int
Alex Tomasa86c6182006-10-11 01:21:03 -07001979ext4_ext_more_to_rm(struct ext4_ext_path *path)
1980{
1981 BUG_ON(path->p_idx == NULL);
1982
1983 if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
1984 return 0;
1985
1986 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07001987 * if truncate on deeper level happened, it wasn't partial,
Alex Tomasa86c6182006-10-11 01:21:03 -07001988 * so we have to consider current index for truncation
1989 */
1990 if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
1991 return 0;
1992 return 1;
1993}
1994
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05001995static int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start)
Alex Tomasa86c6182006-10-11 01:21:03 -07001996{
1997 struct super_block *sb = inode->i_sb;
1998 int depth = ext_depth(inode);
1999 struct ext4_ext_path *path;
2000 handle_t *handle;
2001 int i = 0, err = 0;
2002
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002003 ext_debug("truncate since %u\n", start);
Alex Tomasa86c6182006-10-11 01:21:03 -07002004
2005 /* probably first extent we're gonna free will be last in block */
2006 handle = ext4_journal_start(inode, depth + 1);
2007 if (IS_ERR(handle))
2008 return PTR_ERR(handle);
2009
2010 ext4_ext_invalidate_cache(inode);
2011
2012 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002013 * We start scanning from right side, freeing all the blocks
2014 * after i_size and walking into the tree depth-wise.
Alex Tomasa86c6182006-10-11 01:21:03 -07002015 */
Josef Bacik216553c2008-04-29 22:02:02 -04002016 path = kzalloc(sizeof(struct ext4_ext_path) * (depth + 1), GFP_NOFS);
Alex Tomasa86c6182006-10-11 01:21:03 -07002017 if (path == NULL) {
2018 ext4_journal_stop(handle);
2019 return -ENOMEM;
2020 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002021 path[0].p_hdr = ext_inode_hdr(inode);
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002022 if (ext4_ext_check_header(inode, path[0].p_hdr, depth)) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002023 err = -EIO;
2024 goto out;
2025 }
2026 path[0].p_depth = depth;
2027
2028 while (i >= 0 && err == 0) {
2029 if (i == depth) {
2030 /* this is leaf block */
2031 err = ext4_ext_rm_leaf(handle, inode, path, start);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002032 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002033 brelse(path[i].p_bh);
2034 path[i].p_bh = NULL;
2035 i--;
2036 continue;
2037 }
2038
2039 /* this is index block */
2040 if (!path[i].p_hdr) {
2041 ext_debug("initialize header\n");
2042 path[i].p_hdr = ext_block_hdr(path[i].p_bh);
Alex Tomasa86c6182006-10-11 01:21:03 -07002043 }
2044
Alex Tomasa86c6182006-10-11 01:21:03 -07002045 if (!path[i].p_idx) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002046 /* this level hasn't been touched yet */
Alex Tomasa86c6182006-10-11 01:21:03 -07002047 path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2048 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2049 ext_debug("init index ptr: hdr 0x%p, num %d\n",
2050 path[i].p_hdr,
2051 le16_to_cpu(path[i].p_hdr->eh_entries));
2052 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002053 /* we were already here, see at next index */
Alex Tomasa86c6182006-10-11 01:21:03 -07002054 path[i].p_idx--;
2055 }
2056
2057 ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
2058 i, EXT_FIRST_INDEX(path[i].p_hdr),
2059 path[i].p_idx);
2060 if (ext4_ext_more_to_rm(path + i)) {
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002061 struct buffer_head *bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002062 /* go to the next level */
Mingming Cao2ae02102006-10-11 01:21:11 -07002063 ext_debug("move to level %d (block %llu)\n",
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002064 i + 1, idx_pblock(path[i].p_idx));
Alex Tomasa86c6182006-10-11 01:21:03 -07002065 memset(path + i + 1, 0, sizeof(*path));
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002066 bh = sb_bread(sb, idx_pblock(path[i].p_idx));
2067 if (!bh) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002068 /* should we reset i_size? */
2069 err = -EIO;
2070 break;
2071 }
Alex Tomasc29c0ae2007-07-18 09:19:09 -04002072 if (WARN_ON(i + 1 > depth)) {
2073 err = -EIO;
2074 break;
2075 }
2076 if (ext4_ext_check_header(inode, ext_block_hdr(bh),
2077 depth - i - 1)) {
2078 err = -EIO;
2079 break;
2080 }
2081 path[i + 1].p_bh = bh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002082
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002083 /* save actual number of indexes since this
2084 * number is changed at the next iteration */
Alex Tomasa86c6182006-10-11 01:21:03 -07002085 path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
2086 i++;
2087 } else {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002088 /* we finished processing this index, go up */
Alex Tomasa86c6182006-10-11 01:21:03 -07002089 if (path[i].p_hdr->eh_entries == 0 && i > 0) {
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002090 /* index is empty, remove it;
Alex Tomasa86c6182006-10-11 01:21:03 -07002091 * handle must be already prepared by the
2092 * truncatei_leaf() */
2093 err = ext4_ext_rm_idx(handle, inode, path + i);
2094 }
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002095 /* root level has p_bh == NULL, brelse() eats this */
Alex Tomasa86c6182006-10-11 01:21:03 -07002096 brelse(path[i].p_bh);
2097 path[i].p_bh = NULL;
2098 i--;
2099 ext_debug("return to level %d\n", i);
2100 }
2101 }
2102
2103 /* TODO: flexible tree reduction should be here */
2104 if (path->p_hdr->eh_entries == 0) {
2105 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002106 * truncate to zero freed all the tree,
2107 * so we need to correct eh_depth
Alex Tomasa86c6182006-10-11 01:21:03 -07002108 */
2109 err = ext4_ext_get_access(handle, inode, path);
2110 if (err == 0) {
2111 ext_inode_hdr(inode)->eh_depth = 0;
2112 ext_inode_hdr(inode)->eh_max =
2113 cpu_to_le16(ext4_ext_space_root(inode));
2114 err = ext4_ext_dirty(handle, inode, path);
2115 }
2116 }
2117out:
2118 ext4_ext_tree_changed(inode);
2119 ext4_ext_drop_refs(path);
2120 kfree(path);
2121 ext4_journal_stop(handle);
2122
2123 return err;
2124}
2125
2126/*
2127 * called at mount time
2128 */
2129void ext4_ext_init(struct super_block *sb)
2130{
2131 /*
2132 * possible initialization would be here
2133 */
2134
2135 if (test_opt(sb, EXTENTS)) {
2136 printk("EXT4-fs: file extents enabled");
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +01002137#ifdef AGGRESSIVE_TEST
2138 printk(", aggressive tests");
Alex Tomasa86c6182006-10-11 01:21:03 -07002139#endif
2140#ifdef CHECK_BINSEARCH
2141 printk(", check binsearch");
2142#endif
2143#ifdef EXTENTS_STATS
2144 printk(", stats");
2145#endif
2146 printk("\n");
2147#ifdef EXTENTS_STATS
2148 spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
2149 EXT4_SB(sb)->s_ext_min = 1 << 30;
2150 EXT4_SB(sb)->s_ext_max = 0;
2151#endif
2152 }
2153}
2154
2155/*
2156 * called at umount time
2157 */
2158void ext4_ext_release(struct super_block *sb)
2159{
2160 if (!test_opt(sb, EXTENTS))
2161 return;
2162
2163#ifdef EXTENTS_STATS
2164 if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
2165 struct ext4_sb_info *sbi = EXT4_SB(sb);
2166 printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
2167 sbi->s_ext_blocks, sbi->s_ext_extents,
2168 sbi->s_ext_blocks / sbi->s_ext_extents);
2169 printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
2170 sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
2171 }
2172#endif
2173}
2174
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002175static void bi_complete(struct bio *bio, int error)
2176{
2177 complete((struct completion *)bio->bi_private);
2178}
2179
2180/* FIXME!! we need to try to merge to left or right after zero-out */
2181static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
2182{
2183 int ret = -EIO;
2184 struct bio *bio;
2185 int blkbits, blocksize;
2186 sector_t ee_pblock;
2187 struct completion event;
2188 unsigned int ee_len, len, done, offset;
2189
2190
2191 blkbits = inode->i_blkbits;
2192 blocksize = inode->i_sb->s_blocksize;
2193 ee_len = ext4_ext_get_actual_len(ex);
2194 ee_pblock = ext_pblock(ex);
2195
2196 /* convert ee_pblock to 512 byte sectors */
2197 ee_pblock = ee_pblock << (blkbits - 9);
2198
2199 while (ee_len > 0) {
2200
2201 if (ee_len > BIO_MAX_PAGES)
2202 len = BIO_MAX_PAGES;
2203 else
2204 len = ee_len;
2205
2206 bio = bio_alloc(GFP_NOIO, len);
2207 if (!bio)
2208 return -ENOMEM;
2209 bio->bi_sector = ee_pblock;
2210 bio->bi_bdev = inode->i_sb->s_bdev;
2211
2212 done = 0;
2213 offset = 0;
2214 while (done < len) {
2215 ret = bio_add_page(bio, ZERO_PAGE(0),
2216 blocksize, offset);
2217 if (ret != blocksize) {
2218 /*
2219 * We can't add any more pages because of
2220 * hardware limitations. Start a new bio.
2221 */
2222 break;
2223 }
2224 done++;
2225 offset += blocksize;
2226 if (offset >= PAGE_CACHE_SIZE)
2227 offset = 0;
2228 }
2229
2230 init_completion(&event);
2231 bio->bi_private = &event;
2232 bio->bi_end_io = bi_complete;
2233 submit_bio(WRITE, bio);
2234 wait_for_completion(&event);
2235
2236 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
2237 ret = 0;
2238 else {
2239 ret = -EIO;
2240 break;
2241 }
2242 bio_put(bio);
2243 ee_len -= done;
2244 ee_pblock += done << (blkbits - 9);
2245 }
2246 return ret;
2247}
2248
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002249#define EXT4_EXT_ZERO_LEN 7
2250
Amit Arora56055d32007-07-17 21:42:38 -04002251/*
2252 * This function is called by ext4_ext_get_blocks() if someone tries to write
2253 * to an uninitialized extent. It may result in splitting the uninitialized
2254 * extent into multiple extents (upto three - one initialized and two
2255 * uninitialized).
2256 * There are three possibilities:
2257 * a> There is no split required: Entire extent should be initialized
2258 * b> Splits in two extents: Write is happening at either end of the extent
2259 * c> Splits in three extents: Somone is writing in middle of the extent
2260 */
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002261static int ext4_ext_convert_to_initialized(handle_t *handle,
2262 struct inode *inode,
2263 struct ext4_ext_path *path,
2264 ext4_lblk_t iblock,
2265 unsigned long max_blocks)
Amit Arora56055d32007-07-17 21:42:38 -04002266{
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002267 struct ext4_extent *ex, newex, orig_ex;
Amit Arora56055d32007-07-17 21:42:38 -04002268 struct ext4_extent *ex1 = NULL;
2269 struct ext4_extent *ex2 = NULL;
2270 struct ext4_extent *ex3 = NULL;
2271 struct ext4_extent_header *eh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002272 ext4_lblk_t ee_block;
2273 unsigned int allocated, ee_len, depth;
Amit Arora56055d32007-07-17 21:42:38 -04002274 ext4_fsblk_t newblock;
2275 int err = 0;
2276 int ret = 0;
2277
2278 depth = ext_depth(inode);
2279 eh = path[depth].p_hdr;
2280 ex = path[depth].p_ext;
2281 ee_block = le32_to_cpu(ex->ee_block);
2282 ee_len = ext4_ext_get_actual_len(ex);
2283 allocated = ee_len - (iblock - ee_block);
2284 newblock = iblock - ee_block + ext_pblock(ex);
2285 ex2 = ex;
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002286 orig_ex.ee_block = ex->ee_block;
2287 orig_ex.ee_len = cpu_to_le16(ee_len);
2288 ext4_ext_store_pblock(&orig_ex, ext_pblock(ex));
Amit Arora56055d32007-07-17 21:42:38 -04002289
Aneesh Kumar K.V9df56432008-02-22 06:17:31 -05002290 err = ext4_ext_get_access(handle, inode, path + depth);
2291 if (err)
2292 goto out;
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002293 /* If extent has less than 2*EXT4_EXT_ZERO_LEN zerout directly */
2294 if (ee_len <= 2*EXT4_EXT_ZERO_LEN) {
2295 err = ext4_ext_zeroout(inode, &orig_ex);
2296 if (err)
2297 goto fix_extent_len;
2298 /* update the extent length and mark as initialized */
2299 ex->ee_block = orig_ex.ee_block;
2300 ex->ee_len = orig_ex.ee_len;
2301 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2302 ext4_ext_dirty(handle, inode, path + depth);
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002303 /* zeroed the full extent */
2304 return allocated;
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002305 }
Aneesh Kumar K.V9df56432008-02-22 06:17:31 -05002306
Amit Arora56055d32007-07-17 21:42:38 -04002307 /* ex1: ee_block to iblock - 1 : uninitialized */
2308 if (iblock > ee_block) {
2309 ex1 = ex;
2310 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2311 ext4_ext_mark_uninitialized(ex1);
2312 ex2 = &newex;
2313 }
2314 /*
2315 * for sanity, update the length of the ex2 extent before
2316 * we insert ex3, if ex1 is NULL. This is to avoid temporary
2317 * overlap of blocks.
2318 */
2319 if (!ex1 && allocated > max_blocks)
2320 ex2->ee_len = cpu_to_le16(max_blocks);
2321 /* ex3: to ee_block + ee_len : uninitialised */
2322 if (allocated > max_blocks) {
2323 unsigned int newdepth;
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002324 /* If extent has less than EXT4_EXT_ZERO_LEN zerout directly */
2325 if (allocated <= EXT4_EXT_ZERO_LEN) {
2326 /* Mark first half uninitialized.
2327 * Mark second half initialized and zero out the
2328 * initialized extent
2329 */
2330 ex->ee_block = orig_ex.ee_block;
2331 ex->ee_len = cpu_to_le16(ee_len - allocated);
2332 ext4_ext_mark_uninitialized(ex);
2333 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2334 ext4_ext_dirty(handle, inode, path + depth);
2335
2336 ex3 = &newex;
2337 ex3->ee_block = cpu_to_le32(iblock);
2338 ext4_ext_store_pblock(ex3, newblock);
2339 ex3->ee_len = cpu_to_le16(allocated);
2340 err = ext4_ext_insert_extent(handle, inode, path, ex3);
2341 if (err == -ENOSPC) {
2342 err = ext4_ext_zeroout(inode, &orig_ex);
2343 if (err)
2344 goto fix_extent_len;
2345 ex->ee_block = orig_ex.ee_block;
2346 ex->ee_len = orig_ex.ee_len;
2347 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2348 ext4_ext_dirty(handle, inode, path + depth);
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002349 /* zeroed the full extent */
2350 return allocated;
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002351
2352 } else if (err)
2353 goto fix_extent_len;
2354
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002355 /*
2356 * We need to zero out the second half because
2357 * an fallocate request can update file size and
2358 * converting the second half to initialized extent
2359 * implies that we can leak some junk data to user
2360 * space.
2361 */
2362 err = ext4_ext_zeroout(inode, ex3);
2363 if (err) {
2364 /*
2365 * We should actually mark the
2366 * second half as uninit and return error
2367 * Insert would have changed the extent
2368 */
2369 depth = ext_depth(inode);
2370 ext4_ext_drop_refs(path);
2371 path = ext4_ext_find_extent(inode,
2372 iblock, path);
2373 if (IS_ERR(path)) {
2374 err = PTR_ERR(path);
2375 return err;
2376 }
2377 ex = path[depth].p_ext;
2378 err = ext4_ext_get_access(handle, inode,
2379 path + depth);
2380 if (err)
2381 return err;
2382 ext4_ext_mark_uninitialized(ex);
2383 ext4_ext_dirty(handle, inode, path + depth);
2384 return err;
2385 }
2386
2387 /* zeroed the second half */
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002388 return allocated;
2389 }
Amit Arora56055d32007-07-17 21:42:38 -04002390 ex3 = &newex;
2391 ex3->ee_block = cpu_to_le32(iblock + max_blocks);
2392 ext4_ext_store_pblock(ex3, newblock + max_blocks);
2393 ex3->ee_len = cpu_to_le16(allocated - max_blocks);
2394 ext4_ext_mark_uninitialized(ex3);
2395 err = ext4_ext_insert_extent(handle, inode, path, ex3);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002396 if (err == -ENOSPC) {
2397 err = ext4_ext_zeroout(inode, &orig_ex);
2398 if (err)
2399 goto fix_extent_len;
2400 /* update the extent length and mark as initialized */
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002401 ex->ee_block = orig_ex.ee_block;
2402 ex->ee_len = orig_ex.ee_len;
2403 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002404 ext4_ext_dirty(handle, inode, path + depth);
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002405 /* zeroed the full extent */
2406 return allocated;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002407
2408 } else if (err)
2409 goto fix_extent_len;
Amit Arora56055d32007-07-17 21:42:38 -04002410 /*
2411 * The depth, and hence eh & ex might change
2412 * as part of the insert above.
2413 */
2414 newdepth = ext_depth(inode);
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002415 /*
2416 * update the extent length after successfull insert of the
2417 * split extent
2418 */
2419 orig_ex.ee_len = cpu_to_le16(ee_len -
2420 ext4_ext_get_actual_len(ex3));
Amit Arora56055d32007-07-17 21:42:38 -04002421 if (newdepth != depth) {
2422 depth = newdepth;
Aneesh Kumar K.Vb35905c2008-02-25 16:54:37 -05002423 ext4_ext_drop_refs(path);
2424 path = ext4_ext_find_extent(inode, iblock, path);
Amit Arora56055d32007-07-17 21:42:38 -04002425 if (IS_ERR(path)) {
2426 err = PTR_ERR(path);
Amit Arora56055d32007-07-17 21:42:38 -04002427 goto out;
2428 }
2429 eh = path[depth].p_hdr;
2430 ex = path[depth].p_ext;
2431 if (ex2 != &newex)
2432 ex2 = ex;
Aneesh Kumar K.V9df56432008-02-22 06:17:31 -05002433
2434 err = ext4_ext_get_access(handle, inode, path + depth);
2435 if (err)
2436 goto out;
Amit Arora56055d32007-07-17 21:42:38 -04002437 }
2438 allocated = max_blocks;
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002439
2440 /* If extent has less than EXT4_EXT_ZERO_LEN and we are trying
2441 * to insert a extent in the middle zerout directly
2442 * otherwise give the extent a chance to merge to left
2443 */
2444 if (le16_to_cpu(orig_ex.ee_len) <= EXT4_EXT_ZERO_LEN &&
2445 iblock != ee_block) {
2446 err = ext4_ext_zeroout(inode, &orig_ex);
2447 if (err)
2448 goto fix_extent_len;
2449 /* update the extent length and mark as initialized */
2450 ex->ee_block = orig_ex.ee_block;
2451 ex->ee_len = orig_ex.ee_len;
2452 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2453 ext4_ext_dirty(handle, inode, path + depth);
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002454 /* zero out the first half */
2455 return allocated;
Aneesh Kumar K.V3977c962008-04-17 10:38:59 -04002456 }
Amit Arora56055d32007-07-17 21:42:38 -04002457 }
2458 /*
2459 * If there was a change of depth as part of the
2460 * insertion of ex3 above, we need to update the length
2461 * of the ex1 extent again here
2462 */
2463 if (ex1 && ex1 != ex) {
2464 ex1 = ex;
2465 ex1->ee_len = cpu_to_le16(iblock - ee_block);
2466 ext4_ext_mark_uninitialized(ex1);
2467 ex2 = &newex;
2468 }
2469 /* ex2: iblock to iblock + maxblocks-1 : initialised */
2470 ex2->ee_block = cpu_to_le32(iblock);
Amit Arora56055d32007-07-17 21:42:38 -04002471 ext4_ext_store_pblock(ex2, newblock);
2472 ex2->ee_len = cpu_to_le16(allocated);
2473 if (ex2 != ex)
2474 goto insert;
Amit Arora56055d32007-07-17 21:42:38 -04002475 /*
2476 * New (initialized) extent starts from the first block
2477 * in the current extent. i.e., ex2 == ex
2478 * We have to see if it can be merged with the extent
2479 * on the left.
2480 */
2481 if (ex2 > EXT_FIRST_EXTENT(eh)) {
2482 /*
2483 * To merge left, pass "ex2 - 1" to try_to_merge(),
2484 * since it merges towards right _only_.
2485 */
2486 ret = ext4_ext_try_to_merge(inode, path, ex2 - 1);
2487 if (ret) {
2488 err = ext4_ext_correct_indexes(handle, inode, path);
2489 if (err)
2490 goto out;
2491 depth = ext_depth(inode);
2492 ex2--;
2493 }
2494 }
2495 /*
2496 * Try to Merge towards right. This might be required
2497 * only when the whole extent is being written to.
2498 * i.e. ex2 == ex and ex3 == NULL.
2499 */
2500 if (!ex3) {
2501 ret = ext4_ext_try_to_merge(inode, path, ex2);
2502 if (ret) {
2503 err = ext4_ext_correct_indexes(handle, inode, path);
2504 if (err)
2505 goto out;
2506 }
2507 }
2508 /* Mark modified extent as dirty */
2509 err = ext4_ext_dirty(handle, inode, path + depth);
2510 goto out;
2511insert:
2512 err = ext4_ext_insert_extent(handle, inode, path, &newex);
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002513 if (err == -ENOSPC) {
2514 err = ext4_ext_zeroout(inode, &orig_ex);
2515 if (err)
2516 goto fix_extent_len;
2517 /* update the extent length and mark as initialized */
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002518 ex->ee_block = orig_ex.ee_block;
2519 ex->ee_len = orig_ex.ee_len;
2520 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
Aneesh Kumar K.V95c38892008-04-17 10:38:59 -04002521 ext4_ext_dirty(handle, inode, path + depth);
Aneesh Kumar K.V161e7b72008-04-29 22:03:59 -04002522 /* zero out the first half */
2523 return allocated;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002524 } else if (err)
2525 goto fix_extent_len;
Amit Arora56055d32007-07-17 21:42:38 -04002526out:
2527 return err ? err : allocated;
Aneesh Kumar K.V093a0882008-04-29 08:11:12 -04002528
2529fix_extent_len:
2530 ex->ee_block = orig_ex.ee_block;
2531 ex->ee_len = orig_ex.ee_len;
2532 ext4_ext_store_pblock(ex, ext_pblock(&orig_ex));
2533 ext4_ext_mark_uninitialized(ex);
2534 ext4_ext_dirty(handle, inode, path + depth);
2535 return err;
Amit Arora56055d32007-07-17 21:42:38 -04002536}
2537
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05002538/*
Mingming Caof5ab0d12008-02-25 15:29:55 -05002539 * Block allocation/map/preallocation routine for extents based files
2540 *
2541 *
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05002542 * Need to be called with
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05002543 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
2544 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
Mingming Caof5ab0d12008-02-25 15:29:55 -05002545 *
2546 * return > 0, number of of blocks already mapped/allocated
2547 * if create == 0 and these are pre-allocated blocks
2548 * buffer head is unmapped
2549 * otherwise blocks are mapped
2550 *
2551 * return = 0, if plain look up failed (blocks have not been allocated)
2552 * buffer head is unmapped
2553 *
2554 * return < 0, error case.
Aneesh Kumar K.Vc278bfe2008-01-28 23:58:27 -05002555 */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002556int ext4_ext_get_blocks(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002557 ext4_lblk_t iblock,
Alex Tomasa86c6182006-10-11 01:21:03 -07002558 unsigned long max_blocks, struct buffer_head *bh_result,
2559 int create, int extend_disksize)
2560{
2561 struct ext4_ext_path *path = NULL;
Amit Arora56055d32007-07-17 21:42:38 -04002562 struct ext4_extent_header *eh;
Alex Tomasa86c6182006-10-11 01:21:03 -07002563 struct ext4_extent newex, *ex;
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002564 ext4_fsblk_t goal, newblock;
Amit Arora56055d32007-07-17 21:42:38 -04002565 int err = 0, depth, ret;
Alex Tomasa86c6182006-10-11 01:21:03 -07002566 unsigned long allocated = 0;
Alex Tomasc9de5602008-01-29 00:19:52 -05002567 struct ext4_allocation_request ar;
Mingming Cao61628a32008-07-11 19:27:31 -04002568 loff_t disksize;
Alex Tomasa86c6182006-10-11 01:21:03 -07002569
2570 __clear_bit(BH_New, &bh_result->b_state);
Eric Sandeenbba90742008-01-28 23:58:27 -05002571 ext_debug("blocks %u/%lu requested for inode %u\n",
2572 iblock, max_blocks, inode->i_ino);
Alex Tomasa86c6182006-10-11 01:21:03 -07002573
2574 /* check in cache */
Avantika Mathur7e028972006-12-06 20:41:33 -08002575 goal = ext4_ext_in_cache(inode, iblock, &newex);
2576 if (goal) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002577 if (goal == EXT4_EXT_CACHE_GAP) {
2578 if (!create) {
Amit Arora56055d32007-07-17 21:42:38 -04002579 /*
2580 * block isn't allocated yet and
2581 * user doesn't want to allocate it
2582 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002583 goto out2;
2584 }
2585 /* we should allocate requested block */
2586 } else if (goal == EXT4_EXT_CACHE_EXTENT) {
2587 /* block is already allocated */
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002588 newblock = iblock
2589 - le32_to_cpu(newex.ee_block)
2590 + ext_pblock(&newex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002591 /* number of remaining blocks in the extent */
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05002592 allocated = ext4_ext_get_actual_len(&newex) -
Alex Tomasa86c6182006-10-11 01:21:03 -07002593 (iblock - le32_to_cpu(newex.ee_block));
2594 goto out;
2595 } else {
2596 BUG();
2597 }
2598 }
2599
2600 /* find extent for this block */
2601 path = ext4_ext_find_extent(inode, iblock, NULL);
2602 if (IS_ERR(path)) {
2603 err = PTR_ERR(path);
2604 path = NULL;
2605 goto out2;
2606 }
2607
2608 depth = ext_depth(inode);
2609
2610 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002611 * consistent leaf must not be empty;
2612 * this situation is possible, though, _during_ tree modification;
Alex Tomasa86c6182006-10-11 01:21:03 -07002613 * this is why assert can't be put in ext4_ext_find_extent()
2614 */
2615 BUG_ON(path[depth].p_ext == NULL && depth != 0);
Amit Arora56055d32007-07-17 21:42:38 -04002616 eh = path[depth].p_hdr;
Alex Tomasa86c6182006-10-11 01:21:03 -07002617
Avantika Mathur7e028972006-12-06 20:41:33 -08002618 ex = path[depth].p_ext;
2619 if (ex) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002620 ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002621 ext4_fsblk_t ee_start = ext_pblock(ex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002622 unsigned short ee_len;
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002623
2624 /*
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002625 * Uninitialized extents are treated as holes, except that
Amit Arora56055d32007-07-17 21:42:38 -04002626 * we split out initialized portions during a write.
Suparna Bhattacharya471d4012006-10-11 01:21:06 -07002627 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04002628 ee_len = ext4_ext_get_actual_len(ex);
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002629 /* if found extent covers block, simply return it */
Dave Kleikamp8c55e202007-05-24 13:04:54 -04002630 if (iblock >= ee_block && iblock < ee_block + ee_len) {
Alex Tomasa86c6182006-10-11 01:21:03 -07002631 newblock = iblock - ee_block + ee_start;
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002632 /* number of remaining blocks in the extent */
Alex Tomasa86c6182006-10-11 01:21:03 -07002633 allocated = ee_len - (iblock - ee_block);
Eric Sandeenbba90742008-01-28 23:58:27 -05002634 ext_debug("%u fit into %lu:%d -> %llu\n", iblock,
Alex Tomasa86c6182006-10-11 01:21:03 -07002635 ee_block, ee_len, newblock);
Amit Arora56055d32007-07-17 21:42:38 -04002636
Amit Aroraa2df2a62007-07-17 21:42:41 -04002637 /* Do not put uninitialized extent in the cache */
Amit Arora56055d32007-07-17 21:42:38 -04002638 if (!ext4_ext_is_uninitialized(ex)) {
Amit Aroraa2df2a62007-07-17 21:42:41 -04002639 ext4_ext_put_in_cache(inode, ee_block,
2640 ee_len, ee_start,
2641 EXT4_EXT_CACHE_EXTENT);
Amit Arora56055d32007-07-17 21:42:38 -04002642 goto out;
2643 }
2644 if (create == EXT4_CREATE_UNINITIALIZED_EXT)
2645 goto out;
Aneesh Kumar K.Ve067ba02008-04-29 08:11:12 -04002646 if (!create) {
2647 /*
2648 * We have blocks reserved already. We
2649 * return allocated blocks so that delalloc
2650 * won't do block reservation for us. But
2651 * the buffer head will be unmapped so that
2652 * a read from the block returns 0s.
2653 */
2654 if (allocated > max_blocks)
2655 allocated = max_blocks;
Eric Sandeen953e6222008-07-11 19:27:31 -04002656 set_buffer_unwritten(bh_result);
Amit Arora56055d32007-07-17 21:42:38 -04002657 goto out2;
Aneesh Kumar K.Ve067ba02008-04-29 08:11:12 -04002658 }
Amit Arora56055d32007-07-17 21:42:38 -04002659
2660 ret = ext4_ext_convert_to_initialized(handle, inode,
2661 path, iblock,
2662 max_blocks);
Dmitry Monakhovdbf9d7d2008-01-28 23:58:27 -05002663 if (ret <= 0) {
2664 err = ret;
Amit Arora56055d32007-07-17 21:42:38 -04002665 goto out2;
Dmitry Monakhovdbf9d7d2008-01-28 23:58:27 -05002666 } else
Amit Arora56055d32007-07-17 21:42:38 -04002667 allocated = ret;
2668 goto outnew;
Alex Tomasa86c6182006-10-11 01:21:03 -07002669 }
2670 }
2671
2672 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002673 * requested block isn't allocated yet;
Alex Tomasa86c6182006-10-11 01:21:03 -07002674 * we couldn't try to create block if create flag is zero
2675 */
2676 if (!create) {
Amit Arora56055d32007-07-17 21:42:38 -04002677 /*
2678 * put just found gap into cache to speed up
2679 * subsequent requests
2680 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002681 ext4_ext_put_gap_in_cache(inode, path, iblock);
2682 goto out2;
2683 }
2684 /*
Andrew Morton63f57932006-10-11 01:21:24 -07002685 * Okay, we need to do block allocation. Lazily initialize the block
2686 * allocation info here if necessary.
2687 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002688 if (S_ISREG(inode->i_mode) && (!EXT4_I(inode)->i_block_alloc_info))
2689 ext4_init_block_alloc_info(inode);
2690
Alex Tomasc9de5602008-01-29 00:19:52 -05002691 /* find neighbour allocated blocks */
2692 ar.lleft = iblock;
2693 err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
2694 if (err)
2695 goto out2;
2696 ar.lright = iblock;
2697 err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright);
2698 if (err)
2699 goto out2;
Amit Arora25d14f92007-05-24 13:04:13 -04002700
Amit Arora749269f2007-07-18 09:02:56 -04002701 /*
2702 * See if request is beyond maximum number of blocks we can have in
2703 * a single extent. For an initialized extent this limit is
2704 * EXT_INIT_MAX_LEN and for an uninitialized extent this limit is
2705 * EXT_UNINIT_MAX_LEN.
2706 */
2707 if (max_blocks > EXT_INIT_MAX_LEN &&
2708 create != EXT4_CREATE_UNINITIALIZED_EXT)
2709 max_blocks = EXT_INIT_MAX_LEN;
2710 else if (max_blocks > EXT_UNINIT_MAX_LEN &&
2711 create == EXT4_CREATE_UNINITIALIZED_EXT)
2712 max_blocks = EXT_UNINIT_MAX_LEN;
2713
Amit Arora25d14f92007-05-24 13:04:13 -04002714 /* Check if we can really insert (iblock)::(iblock+max_blocks) extent */
2715 newex.ee_block = cpu_to_le32(iblock);
2716 newex.ee_len = cpu_to_le16(max_blocks);
2717 err = ext4_ext_check_overlap(inode, &newex, path);
2718 if (err)
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05002719 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora25d14f92007-05-24 13:04:13 -04002720 else
2721 allocated = max_blocks;
Alex Tomasc9de5602008-01-29 00:19:52 -05002722
2723 /* allocate new block */
2724 ar.inode = inode;
2725 ar.goal = ext4_ext_find_goal(inode, path, iblock);
2726 ar.logical = iblock;
2727 ar.len = allocated;
2728 if (S_ISREG(inode->i_mode))
2729 ar.flags = EXT4_MB_HINT_DATA;
2730 else
2731 /* disable in-core preallocation for non-regular files */
2732 ar.flags = 0;
2733 newblock = ext4_mb_new_blocks(handle, &ar, &err);
Alex Tomasa86c6182006-10-11 01:21:03 -07002734 if (!newblock)
2735 goto out2;
Mingming Cao2ae02102006-10-11 01:21:11 -07002736 ext_debug("allocate new block: goal %llu, found %llu/%lu\n",
Alex Tomasa86c6182006-10-11 01:21:03 -07002737 goal, newblock, allocated);
2738
2739 /* try to insert new extent into found leaf and return */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002740 ext4_ext_store_pblock(&newex, newblock);
Alex Tomasc9de5602008-01-29 00:19:52 -05002741 newex.ee_len = cpu_to_le16(ar.len);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002742 if (create == EXT4_CREATE_UNINITIALIZED_EXT) /* Mark uninitialized */
2743 ext4_ext_mark_uninitialized(&newex);
Alex Tomasa86c6182006-10-11 01:21:03 -07002744 err = ext4_ext_insert_extent(handle, inode, path, &newex);
Alex Tomas315054f2007-05-24 13:04:25 -04002745 if (err) {
2746 /* free data blocks we just allocated */
Alex Tomasc9de5602008-01-29 00:19:52 -05002747 /* not a good idea to call discard here directly,
2748 * but otherwise we'd need to call it every free() */
2749 ext4_mb_discard_inode_preallocations(inode);
Alex Tomas315054f2007-05-24 13:04:25 -04002750 ext4_free_blocks(handle, inode, ext_pblock(&newex),
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05002751 ext4_ext_get_actual_len(&newex), 0);
Alex Tomasa86c6182006-10-11 01:21:03 -07002752 goto out2;
Alex Tomas315054f2007-05-24 13:04:25 -04002753 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002754
Alex Tomasa86c6182006-10-11 01:21:03 -07002755 /* previous routine could use block we allocated */
Alex Tomasf65e6fb2006-10-11 01:21:05 -07002756 newblock = ext_pblock(&newex);
Aneesh Kumar K.Vb939e372008-01-28 23:58:27 -05002757 allocated = ext4_ext_get_actual_len(&newex);
Amit Arora56055d32007-07-17 21:42:38 -04002758outnew:
Mingming Cao61628a32008-07-11 19:27:31 -04002759 if (extend_disksize) {
2760 disksize = ((loff_t) iblock + ar.len) << inode->i_blkbits;
2761 if (disksize > i_size_read(inode))
2762 disksize = i_size_read(inode);
2763 if (disksize > EXT4_I(inode)->i_disksize)
2764 EXT4_I(inode)->i_disksize = disksize;
2765 }
Aneesh Kumar K.Va379cd12008-07-11 19:27:31 -04002766
Eric Sandeen953e6222008-07-11 19:27:31 -04002767 set_buffer_new(bh_result);
Alex Tomasa86c6182006-10-11 01:21:03 -07002768
Amit Aroraa2df2a62007-07-17 21:42:41 -04002769 /* Cache only when it is _not_ an uninitialized extent */
2770 if (create != EXT4_CREATE_UNINITIALIZED_EXT)
2771 ext4_ext_put_in_cache(inode, iblock, allocated, newblock,
2772 EXT4_EXT_CACHE_EXTENT);
Alex Tomasa86c6182006-10-11 01:21:03 -07002773out:
2774 if (allocated > max_blocks)
2775 allocated = max_blocks;
2776 ext4_ext_show_leaf(inode, path);
Eric Sandeen953e6222008-07-11 19:27:31 -04002777 set_buffer_mapped(bh_result);
Alex Tomasa86c6182006-10-11 01:21:03 -07002778 bh_result->b_bdev = inode->i_sb->s_bdev;
2779 bh_result->b_blocknr = newblock;
2780out2:
2781 if (path) {
2782 ext4_ext_drop_refs(path);
2783 kfree(path);
2784 }
Alex Tomasa86c6182006-10-11 01:21:03 -07002785 return err ? err : allocated;
2786}
2787
Jan Karacf108bc2008-07-11 19:27:31 -04002788void ext4_ext_truncate(struct inode *inode)
Alex Tomasa86c6182006-10-11 01:21:03 -07002789{
2790 struct address_space *mapping = inode->i_mapping;
2791 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002792 ext4_lblk_t last_block;
Alex Tomasa86c6182006-10-11 01:21:03 -07002793 handle_t *handle;
2794 int err = 0;
2795
2796 /*
2797 * probably first extent we're gonna free will be last in block
2798 */
2799 err = ext4_writepage_trans_blocks(inode) + 3;
2800 handle = ext4_journal_start(inode, err);
Jan Karacf108bc2008-07-11 19:27:31 -04002801 if (IS_ERR(handle))
Alex Tomasa86c6182006-10-11 01:21:03 -07002802 return;
Alex Tomasa86c6182006-10-11 01:21:03 -07002803
Jan Karacf108bc2008-07-11 19:27:31 -04002804 if (inode->i_size & (sb->s_blocksize - 1))
2805 ext4_block_truncate_page(handle, mapping, inode->i_size);
Alex Tomasa86c6182006-10-11 01:21:03 -07002806
Jan Kara9ddfc3d2008-07-11 19:27:31 -04002807 if (ext4_orphan_add(handle, inode))
2808 goto out_stop;
2809
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -05002810 down_write(&EXT4_I(inode)->i_data_sem);
Alex Tomasa86c6182006-10-11 01:21:03 -07002811 ext4_ext_invalidate_cache(inode);
2812
Alex Tomasc9de5602008-01-29 00:19:52 -05002813 ext4_mb_discard_inode_preallocations(inode);
2814
Alex Tomasa86c6182006-10-11 01:21:03 -07002815 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002816 * TODO: optimization is possible here.
2817 * Probably we need not scan at all,
2818 * because page truncation is enough.
Alex Tomasa86c6182006-10-11 01:21:03 -07002819 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002820
2821 /* we have to know where to truncate from in crash case */
2822 EXT4_I(inode)->i_disksize = inode->i_size;
2823 ext4_mark_inode_dirty(handle, inode);
2824
2825 last_block = (inode->i_size + sb->s_blocksize - 1)
2826 >> EXT4_BLOCK_SIZE_BITS(sb);
2827 err = ext4_ext_remove_space(inode, last_block);
2828
2829 /* In a multi-transaction truncate, we only make the final
Amit Arora56055d32007-07-17 21:42:38 -04002830 * transaction synchronous.
2831 */
Alex Tomasa86c6182006-10-11 01:21:03 -07002832 if (IS_SYNC(inode))
2833 handle->h_sync = 1;
2834
2835out_stop:
Jan Kara9ddfc3d2008-07-11 19:27:31 -04002836 up_write(&EXT4_I(inode)->i_data_sem);
Alex Tomasa86c6182006-10-11 01:21:03 -07002837 /*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002838 * If this was a simple ftruncate() and the file will remain alive,
Alex Tomasa86c6182006-10-11 01:21:03 -07002839 * then we need to clear up the orphan record which we created above.
2840 * However, if this was a real unlink then we were called by
2841 * ext4_delete_inode(), and we allow that function to clean up the
2842 * orphan info for us.
2843 */
2844 if (inode->i_nlink)
2845 ext4_orphan_del(handle, inode);
2846
Solofo Ramangalahyef737722008-04-29 22:00:41 -04002847 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
2848 ext4_mark_inode_dirty(handle, inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07002849 ext4_journal_stop(handle);
2850}
2851
2852/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002853 * ext4_ext_writepage_trans_blocks:
2854 * calculate max number of blocks we could modify
Alex Tomasa86c6182006-10-11 01:21:03 -07002855 * in order to allocate new block for an inode
2856 */
2857int ext4_ext_writepage_trans_blocks(struct inode *inode, int num)
2858{
2859 int needed;
2860
2861 needed = ext4_ext_calc_credits_for_insert(inode, NULL);
2862
Randy Dunlapd0d856e2006-10-11 01:21:07 -07002863 /* caller wants to allocate num blocks, but note it includes sb */
Alex Tomasa86c6182006-10-11 01:21:03 -07002864 needed = needed * num - (num - 1);
2865
2866#ifdef CONFIG_QUOTA
2867 needed += 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
2868#endif
2869
2870 return needed;
2871}
Amit Aroraa2df2a62007-07-17 21:42:41 -04002872
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04002873static void ext4_falloc_update_inode(struct inode *inode,
2874 int mode, loff_t new_size, int update_ctime)
2875{
2876 struct timespec now;
2877
2878 if (update_ctime) {
2879 now = current_fs_time(inode->i_sb);
2880 if (!timespec_equal(&inode->i_ctime, &now))
2881 inode->i_ctime = now;
2882 }
2883 /*
2884 * Update only when preallocation was requested beyond
2885 * the file size.
2886 */
2887 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
2888 new_size > i_size_read(inode)) {
2889 i_size_write(inode, new_size);
2890 EXT4_I(inode)->i_disksize = new_size;
2891 }
2892
2893}
2894
Amit Aroraa2df2a62007-07-17 21:42:41 -04002895/*
2896 * preallocate space for a file. This implements ext4's fallocate inode
2897 * operation, which gets called from sys_fallocate system call.
2898 * For block-mapped files, posix_fallocate should fall back to the method
2899 * of writing zeroes to the required new blocks (the same behavior which is
2900 * expected for file systems which do not support fallocate() system call).
2901 */
2902long ext4_fallocate(struct inode *inode, int mode, loff_t offset, loff_t len)
2903{
2904 handle_t *handle;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002905 ext4_lblk_t block;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04002906 loff_t new_size;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05002907 unsigned long max_blocks;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002908 int ret = 0;
2909 int ret2 = 0;
2910 int retries = 0;
2911 struct buffer_head map_bh;
2912 unsigned int credits, blkbits = inode->i_blkbits;
2913
2914 /*
2915 * currently supporting (pre)allocate mode for extent-based
2916 * files _only_
2917 */
2918 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
2919 return -EOPNOTSUPP;
2920
2921 /* preallocation to directories is currently not supported */
2922 if (S_ISDIR(inode->i_mode))
2923 return -ENODEV;
2924
2925 block = offset >> blkbits;
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04002926 /*
2927 * We can't just convert len to max_blocks because
2928 * If blocksize = 4096 offset = 3072 and len = 2048
2929 */
Amit Aroraa2df2a62007-07-17 21:42:41 -04002930 max_blocks = (EXT4_BLOCK_ALIGN(len + offset, blkbits) >> blkbits)
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04002931 - block;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002932 /*
2933 * credits to insert 1 extent into extent tree + buffers to be able to
2934 * modify 1 super block, 1 block bitmap and 1 group descriptor.
2935 */
2936 credits = EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + 3;
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05002937 mutex_lock(&inode->i_mutex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002938retry:
2939 while (ret >= 0 && ret < max_blocks) {
2940 block = block + ret;
2941 max_blocks = max_blocks - ret;
2942 handle = ext4_journal_start(inode, credits);
2943 if (IS_ERR(handle)) {
2944 ret = PTR_ERR(handle);
2945 break;
2946 }
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05002947 ret = ext4_get_blocks_wrap(handle, inode, block,
Amit Aroraa2df2a62007-07-17 21:42:41 -04002948 max_blocks, &map_bh,
Mingming Caod2a17632008-07-14 17:52:37 -04002949 EXT4_CREATE_UNINITIALIZED_EXT, 0, 0);
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05002950 if (ret <= 0) {
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05002951#ifdef EXT4FS_DEBUG
2952 WARN_ON(ret <= 0);
2953 printk(KERN_ERR "%s: ext4_ext_get_blocks "
2954 "returned error inode#%lu, block=%u, "
2955 "max_blocks=%lu", __func__,
Aneesh Kumar K.V221879c2008-01-28 23:58:27 -05002956 inode->i_ino, block, max_blocks);
Aneesh Kumar K.V2c986152008-02-25 15:41:35 -05002957#endif
Amit Aroraa2df2a62007-07-17 21:42:41 -04002958 ext4_mark_inode_dirty(handle, inode);
2959 ret2 = ext4_journal_stop(handle);
2960 break;
2961 }
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04002962 if ((block + ret) >= (EXT4_BLOCK_ALIGN(offset + len,
2963 blkbits) >> blkbits))
2964 new_size = offset + len;
2965 else
2966 new_size = (block + ret) << blkbits;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002967
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04002968 ext4_falloc_update_inode(inode, mode, new_size,
2969 buffer_new(&map_bh));
Amit Aroraa2df2a62007-07-17 21:42:41 -04002970 ext4_mark_inode_dirty(handle, inode);
2971 ret2 = ext4_journal_stop(handle);
2972 if (ret2)
2973 break;
2974 }
Aneesh Kumar K.Vfd287842008-04-29 08:11:12 -04002975 if (ret == -ENOSPC &&
2976 ext4_should_retry_alloc(inode->i_sb, &retries)) {
2977 ret = 0;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002978 goto retry;
Amit Aroraa2df2a62007-07-17 21:42:41 -04002979 }
Aneesh Kumar K.V55bd7252008-02-15 12:47:21 -05002980 mutex_unlock(&inode->i_mutex);
Amit Aroraa2df2a62007-07-17 21:42:41 -04002981 return ret > 0 ? ret2 : ret;
2982}