]>
Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
4 | * |
5 | * This copyrighted material is made available to anyone wishing to use, | |
6 | * modify, copy, or redistribute it subject to the terms and conditions | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
10 | /* | |
61e085a8 SW |
11 | * Implements Extendible Hashing as described in: |
12 | * "Extendible Hashing" by Fagin, et al in | |
13 | * __ACM Trans. on Database Systems__, Sept 1979. | |
14 | * | |
15 | * | |
16 | * Here's the layout of dirents which is essentially the same as that of ext2 | |
17 | * within a single block. The field de_name_len is the number of bytes | |
18 | * actually required for the name (no null terminator). The field de_rec_len | |
19 | * is the number of bytes allocated to the dirent. The offset of the next | |
20 | * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is | |
21 | * deleted, the preceding dirent inherits its allocated space, ie | |
22 | * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained | |
23 | * by adding de_rec_len to the current dirent, this essentially causes the | |
24 | * deleted dirent to get jumped over when iterating through all the dirents. | |
25 | * | |
26 | * When deleting the first dirent in a block, there is no previous dirent so | |
27 | * the field de_ino is set to zero to designate it as deleted. When allocating | |
28 | * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the | |
29 | * first dirent has (de_ino == 0) and de_rec_len is large enough, this first | |
30 | * dirent is allocated. Otherwise it must go through all the 'used' dirents | |
31 | * searching for one in which the amount of total space minus the amount of | |
32 | * used space will provide enough space for the new dirent. | |
33 | * | |
34 | * There are two types of blocks in which dirents reside. In a stuffed dinode, | |
35 | * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of | |
36 | * the block. In leaves, they begin at offset sizeof(struct gfs2_leaf) from the | |
37 | * beginning of the leaf block. The dirents reside in leaves when | |
38 | * | |
39 | * dip->i_di.di_flags & GFS2_DIF_EXHASH is true | |
40 | * | |
41 | * Otherwise, the dirents are "linear", within a single stuffed dinode block. | |
42 | * | |
43 | * When the dirents are in leaves, the actual contents of the directory file are | |
44 | * used as an array of 64-bit block pointers pointing to the leaf blocks. The | |
45 | * dirents are NOT in the directory file itself. There can be more than one | |
46 | * block pointer in the array that points to the same leaf. In fact, when a | |
47 | * directory is first converted from linear to exhash, all of the pointers | |
48 | * point to the same leaf. | |
49 | * | |
50 | * When a leaf is completely full, the size of the hash table can be | |
51 | * doubled unless it is already at the maximum size which is hard coded into | |
52 | * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list, | |
53 | * but never before the maximum hash table size has been reached. | |
54 | */ | |
b3b94faa DT |
55 | |
56 | #include <linux/sched.h> | |
57 | #include <linux/slab.h> | |
58 | #include <linux/spinlock.h> | |
b3b94faa DT |
59 | #include <linux/buffer_head.h> |
60 | #include <linux/sort.h> | |
5c676f6d | 61 | #include <linux/gfs2_ondisk.h> |
71b86f56 | 62 | #include <linux/crc32.h> |
fe1bdedc | 63 | #include <linux/vmalloc.h> |
7d308590 | 64 | #include <linux/lm_interface.h> |
b3b94faa DT |
65 | |
66 | #include "gfs2.h" | |
5c676f6d | 67 | #include "incore.h" |
b3b94faa DT |
68 | #include "dir.h" |
69 | #include "glock.h" | |
70 | #include "inode.h" | |
b3b94faa DT |
71 | #include "meta_io.h" |
72 | #include "quota.h" | |
73 | #include "rgrp.h" | |
74 | #include "trans.h" | |
e13940ba | 75 | #include "bmap.h" |
5c676f6d | 76 | #include "util.h" |
b3b94faa DT |
77 | |
78 | #define IS_LEAF 1 /* Hashed (leaf) directory */ | |
79 | #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */ | |
80 | ||
cd915493 SW |
81 | #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1) |
82 | #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1)) | |
b3b94faa | 83 | |
907b9bce | 84 | typedef int (*leaf_call_t) (struct gfs2_inode *dip, u32 index, u32 len, |
2bdbc5d7 SW |
85 | u64 leaf_no, void *data); |
86 | typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent, | |
87 | const struct qstr *name, void *opaque); | |
b3b94faa | 88 | |
61e085a8 | 89 | |
cd915493 | 90 | int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block, |
61e085a8 | 91 | struct buffer_head **bhp) |
e13940ba SW |
92 | { |
93 | struct buffer_head *bh; | |
e13940ba | 94 | |
61e085a8 SW |
95 | bh = gfs2_meta_new(ip->i_gl, block); |
96 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
97 | gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD); | |
98 | gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header)); | |
e13940ba SW |
99 | *bhp = bh; |
100 | return 0; | |
101 | } | |
102 | ||
cd915493 | 103 | static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block, |
61e085a8 SW |
104 | struct buffer_head **bhp) |
105 | { | |
106 | struct buffer_head *bh; | |
107 | int error; | |
e13940ba | 108 | |
7276b3b0 | 109 | error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, &bh); |
61e085a8 SW |
110 | if (error) |
111 | return error; | |
feaa7bba | 112 | if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) { |
61e085a8 SW |
113 | brelse(bh); |
114 | return -EIO; | |
115 | } | |
116 | *bhp = bh; | |
117 | return 0; | |
118 | } | |
e13940ba SW |
119 | |
120 | static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf, | |
121 | unsigned int offset, unsigned int size) | |
e13940ba SW |
122 | { |
123 | struct buffer_head *dibh; | |
124 | int error; | |
125 | ||
126 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
127 | if (error) | |
128 | return error; | |
129 | ||
130 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
c752666c | 131 | memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size); |
e13940ba SW |
132 | if (ip->i_di.di_size < offset + size) |
133 | ip->i_di.di_size = offset + size; | |
134 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
135 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | |
136 | ||
137 | brelse(dibh); | |
138 | ||
139 | return size; | |
140 | } | |
141 | ||
142 | ||
143 | ||
144 | /** | |
145 | * gfs2_dir_write_data - Write directory information to the inode | |
146 | * @ip: The GFS2 inode | |
147 | * @buf: The buffer containing information to be written | |
148 | * @offset: The file offset to start writing at | |
149 | * @size: The amount of data to write | |
150 | * | |
151 | * Returns: The number of bytes correctly written or error code | |
152 | */ | |
153 | static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf, | |
cd915493 | 154 | u64 offset, unsigned int size) |
e13940ba | 155 | { |
feaa7bba | 156 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
e13940ba | 157 | struct buffer_head *dibh; |
cd915493 SW |
158 | u64 lblock, dblock; |
159 | u32 extlen = 0; | |
e13940ba SW |
160 | unsigned int o; |
161 | int copied = 0; | |
162 | int error = 0; | |
163 | ||
164 | if (!size) | |
165 | return 0; | |
166 | ||
167 | if (gfs2_is_stuffed(ip) && | |
168 | offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) | |
568f4c96 SW |
169 | return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset, |
170 | size); | |
e13940ba SW |
171 | |
172 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
173 | return -EINVAL; | |
174 | ||
175 | if (gfs2_is_stuffed(ip)) { | |
f25ef0c1 | 176 | error = gfs2_unstuff_dinode(ip, NULL); |
e13940ba | 177 | if (error) |
c752666c | 178 | return error; |
e13940ba SW |
179 | } |
180 | ||
181 | lblock = offset; | |
182 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); | |
183 | ||
184 | while (copied < size) { | |
185 | unsigned int amount; | |
186 | struct buffer_head *bh; | |
187 | int new; | |
188 | ||
189 | amount = size - copied; | |
190 | if (amount > sdp->sd_sb.sb_bsize - o) | |
191 | amount = sdp->sd_sb.sb_bsize - o; | |
192 | ||
193 | if (!extlen) { | |
194 | new = 1; | |
feaa7bba | 195 | error = gfs2_extent_map(&ip->i_inode, lblock, &new, |
fd88de56 | 196 | &dblock, &extlen); |
e13940ba SW |
197 | if (error) |
198 | goto fail; | |
199 | error = -EIO; | |
200 | if (gfs2_assert_withdraw(sdp, dblock)) | |
201 | goto fail; | |
202 | } | |
203 | ||
61e085a8 SW |
204 | if (amount == sdp->sd_jbsize || new) |
205 | error = gfs2_dir_get_new_buffer(ip, dblock, &bh); | |
206 | else | |
207 | error = gfs2_dir_get_existing_buffer(ip, dblock, &bh); | |
208 | ||
e13940ba SW |
209 | if (error) |
210 | goto fail; | |
211 | ||
212 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
213 | memcpy(bh->b_data + o, buf, amount); | |
214 | brelse(bh); | |
e13940ba | 215 | |
899bb264 | 216 | buf += amount; |
e13940ba SW |
217 | copied += amount; |
218 | lblock++; | |
219 | dblock++; | |
220 | extlen--; | |
221 | ||
222 | o = sizeof(struct gfs2_meta_header); | |
223 | } | |
224 | ||
225 | out: | |
226 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
227 | if (error) | |
228 | return error; | |
229 | ||
230 | if (ip->i_di.di_size < offset + copied) | |
231 | ip->i_di.di_size = offset + copied; | |
232 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
233 | ||
234 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
235 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | |
236 | brelse(dibh); | |
237 | ||
238 | return copied; | |
239 | fail: | |
240 | if (copied) | |
241 | goto out; | |
242 | return error; | |
243 | } | |
244 | ||
245 | static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, char *buf, | |
7276b3b0 | 246 | u64 offset, unsigned int size) |
e13940ba SW |
247 | { |
248 | struct buffer_head *dibh; | |
249 | int error; | |
250 | ||
251 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
252 | if (!error) { | |
253 | offset += sizeof(struct gfs2_dinode); | |
254 | memcpy(buf, dibh->b_data + offset, size); | |
255 | brelse(dibh); | |
256 | } | |
257 | ||
258 | return (error) ? error : size; | |
259 | } | |
260 | ||
261 | ||
262 | /** | |
263 | * gfs2_dir_read_data - Read a data from a directory inode | |
264 | * @ip: The GFS2 Inode | |
265 | * @buf: The buffer to place result into | |
266 | * @offset: File offset to begin jdata_readng from | |
267 | * @size: Amount of data to transfer | |
268 | * | |
269 | * Returns: The amount of data actually copied or the error | |
270 | */ | |
7276b3b0 SW |
271 | static int gfs2_dir_read_data(struct gfs2_inode *ip, char *buf, u64 offset, |
272 | unsigned int size, unsigned ra) | |
e13940ba | 273 | { |
feaa7bba | 274 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
cd915493 SW |
275 | u64 lblock, dblock; |
276 | u32 extlen = 0; | |
e13940ba SW |
277 | unsigned int o; |
278 | int copied = 0; | |
279 | int error = 0; | |
280 | ||
281 | if (offset >= ip->i_di.di_size) | |
282 | return 0; | |
283 | ||
c5392124 | 284 | if (offset + size > ip->i_di.di_size) |
e13940ba SW |
285 | size = ip->i_di.di_size - offset; |
286 | ||
287 | if (!size) | |
288 | return 0; | |
289 | ||
290 | if (gfs2_is_stuffed(ip)) | |
7276b3b0 | 291 | return gfs2_dir_read_stuffed(ip, buf, offset, size); |
e13940ba SW |
292 | |
293 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
294 | return -EINVAL; | |
295 | ||
296 | lblock = offset; | |
297 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); | |
298 | ||
299 | while (copied < size) { | |
300 | unsigned int amount; | |
301 | struct buffer_head *bh; | |
302 | int new; | |
303 | ||
304 | amount = size - copied; | |
305 | if (amount > sdp->sd_sb.sb_bsize - o) | |
306 | amount = sdp->sd_sb.sb_bsize - o; | |
307 | ||
308 | if (!extlen) { | |
309 | new = 0; | |
feaa7bba | 310 | error = gfs2_extent_map(&ip->i_inode, lblock, &new, |
fd88de56 | 311 | &dblock, &extlen); |
7276b3b0 | 312 | if (error || !dblock) |
e13940ba | 313 | goto fail; |
7276b3b0 SW |
314 | BUG_ON(extlen < 1); |
315 | if (!ra) | |
316 | extlen = 1; | |
317 | bh = gfs2_meta_ra(ip->i_gl, dblock, extlen); | |
e13940ba | 318 | } |
7276b3b0 SW |
319 | if (!bh) { |
320 | error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, &bh); | |
e13940ba SW |
321 | if (error) |
322 | goto fail; | |
7276b3b0 SW |
323 | } |
324 | error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD); | |
325 | if (error) { | |
326 | brelse(bh); | |
327 | goto fail; | |
328 | } | |
329 | dblock++; | |
330 | extlen--; | |
e13940ba SW |
331 | memcpy(buf, bh->b_data + o, amount); |
332 | brelse(bh); | |
7276b3b0 | 333 | bh = NULL; |
899bb264 | 334 | buf += amount; |
e13940ba SW |
335 | copied += amount; |
336 | lblock++; | |
e13940ba SW |
337 | o = sizeof(struct gfs2_meta_header); |
338 | } | |
339 | ||
340 | return copied; | |
341 | fail: | |
342 | return (copied) ? copied : error; | |
343 | } | |
344 | ||
c752666c SW |
345 | static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent, |
346 | const struct qstr *name, int ret) | |
347 | { | |
348 | if (dent->de_inum.no_addr != 0 && | |
349 | be32_to_cpu(dent->de_hash) == name->hash && | |
350 | be16_to_cpu(dent->de_name_len) == name->len && | |
2bdbc5d7 | 351 | memcmp(dent+1, name->name, name->len) == 0) |
c752666c SW |
352 | return ret; |
353 | return 0; | |
354 | } | |
355 | ||
356 | static int gfs2_dirent_find(const struct gfs2_dirent *dent, | |
71b86f56 SW |
357 | const struct qstr *name, |
358 | void *opaque) | |
c752666c SW |
359 | { |
360 | return __gfs2_dirent_find(dent, name, 1); | |
361 | } | |
362 | ||
363 | static int gfs2_dirent_prev(const struct gfs2_dirent *dent, | |
71b86f56 SW |
364 | const struct qstr *name, |
365 | void *opaque) | |
c752666c SW |
366 | { |
367 | return __gfs2_dirent_find(dent, name, 2); | |
368 | } | |
369 | ||
370 | /* | |
371 | * name->name holds ptr to start of block. | |
372 | * name->len holds size of block. | |
b3b94faa | 373 | */ |
c752666c | 374 | static int gfs2_dirent_last(const struct gfs2_dirent *dent, |
71b86f56 SW |
375 | const struct qstr *name, |
376 | void *opaque) | |
c752666c SW |
377 | { |
378 | const char *start = name->name; | |
379 | const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len); | |
380 | if (name->len == (end - start)) | |
381 | return 1; | |
382 | return 0; | |
383 | } | |
b3b94faa | 384 | |
c752666c | 385 | static int gfs2_dirent_find_space(const struct gfs2_dirent *dent, |
71b86f56 SW |
386 | const struct qstr *name, |
387 | void *opaque) | |
b3b94faa | 388 | { |
c752666c SW |
389 | unsigned required = GFS2_DIRENT_SIZE(name->len); |
390 | unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); | |
391 | unsigned totlen = be16_to_cpu(dent->de_rec_len); | |
392 | ||
71b86f56 SW |
393 | if (!dent->de_inum.no_addr) |
394 | actual = GFS2_DIRENT_SIZE(0); | |
c5392124 | 395 | if (totlen - actual >= required) |
c752666c SW |
396 | return 1; |
397 | return 0; | |
398 | } | |
399 | ||
71b86f56 SW |
400 | struct dirent_gather { |
401 | const struct gfs2_dirent **pdent; | |
402 | unsigned offset; | |
403 | }; | |
404 | ||
405 | static int gfs2_dirent_gather(const struct gfs2_dirent *dent, | |
406 | const struct qstr *name, | |
407 | void *opaque) | |
408 | { | |
409 | struct dirent_gather *g = opaque; | |
410 | if (dent->de_inum.no_addr) { | |
411 | g->pdent[g->offset++] = dent; | |
412 | } | |
413 | return 0; | |
414 | } | |
415 | ||
c752666c SW |
416 | /* |
417 | * Other possible things to check: | |
418 | * - Inode located within filesystem size (and on valid block) | |
419 | * - Valid directory entry type | |
420 | * Not sure how heavy-weight we want to make this... could also check | |
421 | * hash is correct for example, but that would take a lot of extra time. | |
422 | * For now the most important thing is to check that the various sizes | |
423 | * are correct. | |
424 | */ | |
425 | static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset, | |
426 | unsigned int size, unsigned int len, int first) | |
427 | { | |
428 | const char *msg = "gfs2_dirent too small"; | |
429 | if (unlikely(size < sizeof(struct gfs2_dirent))) | |
430 | goto error; | |
431 | msg = "gfs2_dirent misaligned"; | |
432 | if (unlikely(offset & 0x7)) | |
433 | goto error; | |
434 | msg = "gfs2_dirent points beyond end of block"; | |
435 | if (unlikely(offset + size > len)) | |
436 | goto error; | |
437 | msg = "zero inode number"; | |
438 | if (unlikely(!first && !dent->de_inum.no_addr)) | |
439 | goto error; | |
440 | msg = "name length is greater than space in dirent"; | |
441 | if (dent->de_inum.no_addr && | |
442 | unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) > | |
443 | size)) | |
444 | goto error; | |
445 | return 0; | |
446 | error: | |
447 | printk(KERN_WARNING "gfs2_check_dirent: %s (%s)\n", msg, | |
448 | first ? "first in block" : "not first in block"); | |
449 | return -EIO; | |
b3b94faa DT |
450 | } |
451 | ||
71b86f56 | 452 | static int gfs2_dirent_offset(const void *buf) |
c752666c | 453 | { |
71b86f56 SW |
454 | const struct gfs2_meta_header *h = buf; |
455 | int offset; | |
c752666c SW |
456 | |
457 | BUG_ON(buf == NULL); | |
c752666c | 458 | |
e3167ded | 459 | switch(be32_to_cpu(h->mh_type)) { |
c752666c SW |
460 | case GFS2_METATYPE_LF: |
461 | offset = sizeof(struct gfs2_leaf); | |
462 | break; | |
463 | case GFS2_METATYPE_DI: | |
464 | offset = sizeof(struct gfs2_dinode); | |
465 | break; | |
466 | default: | |
467 | goto wrong_type; | |
468 | } | |
71b86f56 SW |
469 | return offset; |
470 | wrong_type: | |
471 | printk(KERN_WARNING "gfs2_scan_dirent: wrong block type %u\n", | |
e3167ded | 472 | be32_to_cpu(h->mh_type)); |
71b86f56 SW |
473 | return -1; |
474 | } | |
475 | ||
2bdbc5d7 | 476 | static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf, |
71b86f56 SW |
477 | unsigned int len, gfs2_dscan_t scan, |
478 | const struct qstr *name, | |
479 | void *opaque) | |
480 | { | |
481 | struct gfs2_dirent *dent, *prev; | |
482 | unsigned offset; | |
483 | unsigned size; | |
484 | int ret = 0; | |
c752666c | 485 | |
71b86f56 SW |
486 | ret = gfs2_dirent_offset(buf); |
487 | if (ret < 0) | |
488 | goto consist_inode; | |
489 | ||
490 | offset = ret; | |
c752666c | 491 | prev = NULL; |
2bdbc5d7 | 492 | dent = buf + offset; |
c752666c SW |
493 | size = be16_to_cpu(dent->de_rec_len); |
494 | if (gfs2_check_dirent(dent, offset, size, len, 1)) | |
495 | goto consist_inode; | |
496 | do { | |
71b86f56 | 497 | ret = scan(dent, name, opaque); |
c752666c SW |
498 | if (ret) |
499 | break; | |
500 | offset += size; | |
501 | if (offset == len) | |
502 | break; | |
503 | prev = dent; | |
2bdbc5d7 | 504 | dent = buf + offset; |
c752666c SW |
505 | size = be16_to_cpu(dent->de_rec_len); |
506 | if (gfs2_check_dirent(dent, offset, size, len, 0)) | |
507 | goto consist_inode; | |
508 | } while(1); | |
509 | ||
510 | switch(ret) { | |
511 | case 0: | |
512 | return NULL; | |
513 | case 1: | |
514 | return dent; | |
515 | case 2: | |
516 | return prev ? prev : dent; | |
517 | default: | |
518 | BUG_ON(ret > 0); | |
519 | return ERR_PTR(ret); | |
520 | } | |
521 | ||
c752666c | 522 | consist_inode: |
feaa7bba | 523 | gfs2_consist_inode(GFS2_I(inode)); |
c752666c SW |
524 | return ERR_PTR(-EIO); |
525 | } | |
526 | ||
527 | ||
b3b94faa DT |
528 | /** |
529 | * dirent_first - Return the first dirent | |
530 | * @dip: the directory | |
531 | * @bh: The buffer | |
532 | * @dent: Pointer to list of dirents | |
533 | * | |
534 | * return first dirent whether bh points to leaf or stuffed dinode | |
535 | * | |
536 | * Returns: IS_LEAF, IS_DINODE, or -errno | |
537 | */ | |
538 | ||
539 | static int dirent_first(struct gfs2_inode *dip, struct buffer_head *bh, | |
540 | struct gfs2_dirent **dent) | |
541 | { | |
542 | struct gfs2_meta_header *h = (struct gfs2_meta_header *)bh->b_data; | |
543 | ||
e3167ded | 544 | if (be32_to_cpu(h->mh_type) == GFS2_METATYPE_LF) { |
feaa7bba | 545 | if (gfs2_meta_check(GFS2_SB(&dip->i_inode), bh)) |
b3b94faa DT |
546 | return -EIO; |
547 | *dent = (struct gfs2_dirent *)(bh->b_data + | |
548 | sizeof(struct gfs2_leaf)); | |
549 | return IS_LEAF; | |
550 | } else { | |
feaa7bba | 551 | if (gfs2_metatype_check(GFS2_SB(&dip->i_inode), bh, GFS2_METATYPE_DI)) |
b3b94faa DT |
552 | return -EIO; |
553 | *dent = (struct gfs2_dirent *)(bh->b_data + | |
554 | sizeof(struct gfs2_dinode)); | |
555 | return IS_DINODE; | |
556 | } | |
557 | } | |
558 | ||
2bdbc5d7 SW |
559 | static int dirent_check_reclen(struct gfs2_inode *dip, |
560 | const struct gfs2_dirent *d, const void *end_p) | |
561 | { | |
562 | const void *ptr = d; | |
563 | u16 rec_len = be16_to_cpu(d->de_rec_len); | |
564 | ||
565 | if (unlikely(rec_len < sizeof(struct gfs2_dirent))) | |
566 | goto broken; | |
567 | ptr += rec_len; | |
568 | if (ptr < end_p) | |
569 | return rec_len; | |
570 | if (ptr == end_p) | |
571 | return -ENOENT; | |
572 | broken: | |
573 | gfs2_consist_inode(dip); | |
574 | return -EIO; | |
575 | } | |
576 | ||
b3b94faa DT |
577 | /** |
578 | * dirent_next - Next dirent | |
579 | * @dip: the directory | |
580 | * @bh: The buffer | |
581 | * @dent: Pointer to list of dirents | |
582 | * | |
583 | * Returns: 0 on success, error code otherwise | |
584 | */ | |
585 | ||
586 | static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh, | |
587 | struct gfs2_dirent **dent) | |
588 | { | |
2bdbc5d7 SW |
589 | struct gfs2_dirent *cur = *dent, *tmp; |
590 | char *bh_end = bh->b_data + bh->b_size; | |
591 | int ret; | |
b3b94faa | 592 | |
2bdbc5d7 SW |
593 | ret = dirent_check_reclen(dip, cur, bh_end); |
594 | if (ret < 0) | |
595 | return ret; | |
4dd651ad | 596 | |
2bdbc5d7 SW |
597 | tmp = (void *)cur + ret; |
598 | ret = dirent_check_reclen(dip, tmp, bh_end); | |
599 | if (ret == -EIO) | |
600 | return ret; | |
4dd651ad | 601 | |
b3b94faa DT |
602 | /* Only the first dent could ever have de_inum.no_addr == 0 */ |
603 | if (!tmp->de_inum.no_addr) { | |
604 | gfs2_consist_inode(dip); | |
605 | return -EIO; | |
606 | } | |
607 | ||
608 | *dent = tmp; | |
b3b94faa DT |
609 | return 0; |
610 | } | |
611 | ||
612 | /** | |
613 | * dirent_del - Delete a dirent | |
614 | * @dip: The GFS2 inode | |
615 | * @bh: The buffer | |
616 | * @prev: The previous dirent | |
617 | * @cur: The current dirent | |
618 | * | |
619 | */ | |
620 | ||
621 | static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh, | |
622 | struct gfs2_dirent *prev, struct gfs2_dirent *cur) | |
623 | { | |
cd915493 | 624 | u16 cur_rec_len, prev_rec_len; |
b3b94faa DT |
625 | |
626 | if (!cur->de_inum.no_addr) { | |
627 | gfs2_consist_inode(dip); | |
628 | return; | |
629 | } | |
630 | ||
d4e9c4c3 | 631 | gfs2_trans_add_bh(dip->i_gl, bh, 1); |
b3b94faa DT |
632 | |
633 | /* If there is no prev entry, this is the first entry in the block. | |
634 | The de_rec_len is already as big as it needs to be. Just zero | |
635 | out the inode number and return. */ | |
636 | ||
637 | if (!prev) { | |
638 | cur->de_inum.no_addr = 0; /* No endianess worries */ | |
639 | return; | |
640 | } | |
641 | ||
642 | /* Combine this dentry with the previous one. */ | |
643 | ||
fc69d0d3 SW |
644 | prev_rec_len = be16_to_cpu(prev->de_rec_len); |
645 | cur_rec_len = be16_to_cpu(cur->de_rec_len); | |
b3b94faa DT |
646 | |
647 | if ((char *)prev + prev_rec_len != (char *)cur) | |
648 | gfs2_consist_inode(dip); | |
649 | if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size) | |
650 | gfs2_consist_inode(dip); | |
651 | ||
652 | prev_rec_len += cur_rec_len; | |
fc69d0d3 | 653 | prev->de_rec_len = cpu_to_be16(prev_rec_len); |
b3b94faa DT |
654 | } |
655 | ||
c752666c SW |
656 | /* |
657 | * Takes a dent from which to grab space as an argument. Returns the | |
658 | * newly created dent. | |
b3b94faa | 659 | */ |
08bc2dbc AB |
660 | static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode, |
661 | struct gfs2_dirent *dent, | |
662 | const struct qstr *name, | |
663 | struct buffer_head *bh) | |
b3b94faa | 664 | { |
feaa7bba | 665 | struct gfs2_inode *ip = GFS2_I(inode); |
c752666c SW |
666 | struct gfs2_dirent *ndent; |
667 | unsigned offset = 0, totlen; | |
668 | ||
669 | if (dent->de_inum.no_addr) | |
670 | offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); | |
671 | totlen = be16_to_cpu(dent->de_rec_len); | |
672 | BUG_ON(offset + name->len > totlen); | |
673 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
674 | ndent = (struct gfs2_dirent *)((char *)dent + offset); | |
675 | dent->de_rec_len = cpu_to_be16(offset); | |
676 | gfs2_qstr2dirent(name, totlen - offset, ndent); | |
677 | return ndent; | |
b3b94faa DT |
678 | } |
679 | ||
c752666c SW |
680 | static struct gfs2_dirent *gfs2_dirent_alloc(struct inode *inode, |
681 | struct buffer_head *bh, | |
682 | const struct qstr *name) | |
b3b94faa DT |
683 | { |
684 | struct gfs2_dirent *dent; | |
907b9bce | 685 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, |
71b86f56 | 686 | gfs2_dirent_find_space, name, NULL); |
c752666c SW |
687 | if (!dent || IS_ERR(dent)) |
688 | return dent; | |
689 | return gfs2_init_dirent(inode, dent, name, bh); | |
b3b94faa DT |
690 | } |
691 | ||
cd915493 | 692 | static int get_leaf(struct gfs2_inode *dip, u64 leaf_no, |
b3b94faa DT |
693 | struct buffer_head **bhp) |
694 | { | |
695 | int error; | |
696 | ||
7276b3b0 | 697 | error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, bhp); |
feaa7bba SW |
698 | if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) { |
699 | /* printk(KERN_INFO "block num=%llu\n", leaf_no); */ | |
b3b94faa | 700 | error = -EIO; |
feaa7bba | 701 | } |
b3b94faa DT |
702 | |
703 | return error; | |
704 | } | |
705 | ||
706 | /** | |
707 | * get_leaf_nr - Get a leaf number associated with the index | |
708 | * @dip: The GFS2 inode | |
709 | * @index: | |
710 | * @leaf_out: | |
711 | * | |
712 | * Returns: 0 on success, error code otherwise | |
713 | */ | |
714 | ||
cd915493 SW |
715 | static int get_leaf_nr(struct gfs2_inode *dip, u32 index, |
716 | u64 *leaf_out) | |
b3b94faa | 717 | { |
cd915493 | 718 | u64 leaf_no; |
b3b94faa DT |
719 | int error; |
720 | ||
e13940ba | 721 | error = gfs2_dir_read_data(dip, (char *)&leaf_no, |
cd915493 | 722 | index * sizeof(u64), |
7276b3b0 | 723 | sizeof(u64), 0); |
cd915493 | 724 | if (error != sizeof(u64)) |
b3b94faa DT |
725 | return (error < 0) ? error : -EIO; |
726 | ||
727 | *leaf_out = be64_to_cpu(leaf_no); | |
728 | ||
729 | return 0; | |
730 | } | |
731 | ||
cd915493 | 732 | static int get_first_leaf(struct gfs2_inode *dip, u32 index, |
b3b94faa DT |
733 | struct buffer_head **bh_out) |
734 | { | |
cd915493 | 735 | u64 leaf_no; |
b3b94faa DT |
736 | int error; |
737 | ||
738 | error = get_leaf_nr(dip, index, &leaf_no); | |
739 | if (!error) | |
740 | error = get_leaf(dip, leaf_no, bh_out); | |
741 | ||
742 | return error; | |
743 | } | |
744 | ||
c752666c SW |
745 | static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode, |
746 | const struct qstr *name, | |
747 | gfs2_dscan_t scan, | |
748 | struct buffer_head **pbh) | |
b3b94faa | 749 | { |
c752666c SW |
750 | struct buffer_head *bh; |
751 | struct gfs2_dirent *dent; | |
feaa7bba | 752 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
753 | int error; |
754 | ||
c752666c SW |
755 | if (ip->i_di.di_flags & GFS2_DIF_EXHASH) { |
756 | struct gfs2_leaf *leaf; | |
757 | unsigned hsize = 1 << ip->i_di.di_depth; | |
758 | unsigned index; | |
759 | u64 ln; | |
760 | if (hsize * sizeof(u64) != ip->i_di.di_size) { | |
761 | gfs2_consist_inode(ip); | |
762 | return ERR_PTR(-EIO); | |
763 | } | |
907b9bce | 764 | |
c752666c SW |
765 | index = name->hash >> (32 - ip->i_di.di_depth); |
766 | error = get_first_leaf(ip, index, &bh); | |
767 | if (error) | |
768 | return ERR_PTR(error); | |
769 | do { | |
770 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, | |
71b86f56 | 771 | scan, name, NULL); |
c752666c SW |
772 | if (dent) |
773 | goto got_dent; | |
774 | leaf = (struct gfs2_leaf *)bh->b_data; | |
775 | ln = be64_to_cpu(leaf->lf_next); | |
f4154ea0 | 776 | brelse(bh); |
c752666c SW |
777 | if (!ln) |
778 | break; | |
907b9bce | 779 | |
c752666c SW |
780 | error = get_leaf(ip, ln, &bh); |
781 | } while(!error); | |
b3b94faa | 782 | |
c752666c | 783 | return error ? ERR_PTR(error) : NULL; |
b3b94faa | 784 | } |
b3b94faa | 785 | |
907b9bce | 786 | |
c752666c SW |
787 | error = gfs2_meta_inode_buffer(ip, &bh); |
788 | if (error) | |
789 | return ERR_PTR(error); | |
71b86f56 | 790 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL); |
c752666c | 791 | got_dent: |
f4154ea0 | 792 | if (unlikely(dent == NULL || IS_ERR(dent))) { |
ed386507 SW |
793 | brelse(bh); |
794 | bh = NULL; | |
795 | } | |
c752666c SW |
796 | *pbh = bh; |
797 | return dent; | |
798 | } | |
b3b94faa | 799 | |
c752666c SW |
800 | static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth) |
801 | { | |
feaa7bba | 802 | struct gfs2_inode *ip = GFS2_I(inode); |
c752666c SW |
803 | u64 bn = gfs2_alloc_meta(ip); |
804 | struct buffer_head *bh = gfs2_meta_new(ip->i_gl, bn); | |
805 | struct gfs2_leaf *leaf; | |
806 | struct gfs2_dirent *dent; | |
71b86f56 | 807 | struct qstr name = { .name = "", .len = 0, .hash = 0 }; |
c752666c SW |
808 | if (!bh) |
809 | return NULL; | |
907b9bce | 810 | |
c752666c SW |
811 | gfs2_trans_add_bh(ip->i_gl, bh, 1); |
812 | gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF); | |
813 | leaf = (struct gfs2_leaf *)bh->b_data; | |
814 | leaf->lf_depth = cpu_to_be16(depth); | |
2bdbc5d7 | 815 | leaf->lf_entries = 0; |
a2d7d021 | 816 | leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE); |
2bdbc5d7 | 817 | leaf->lf_next = 0; |
c752666c SW |
818 | memset(leaf->lf_reserved, 0, sizeof(leaf->lf_reserved)); |
819 | dent = (struct gfs2_dirent *)(leaf+1); | |
71b86f56 | 820 | gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent); |
c752666c SW |
821 | *pbh = bh; |
822 | return leaf; | |
b3b94faa DT |
823 | } |
824 | ||
825 | /** | |
826 | * dir_make_exhash - Convert a stuffed directory into an ExHash directory | |
827 | * @dip: The GFS2 inode | |
828 | * | |
829 | * Returns: 0 on success, error code otherwise | |
830 | */ | |
831 | ||
c752666c | 832 | static int dir_make_exhash(struct inode *inode) |
b3b94faa | 833 | { |
feaa7bba SW |
834 | struct gfs2_inode *dip = GFS2_I(inode); |
835 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
b3b94faa | 836 | struct gfs2_dirent *dent; |
c752666c | 837 | struct qstr args; |
b3b94faa DT |
838 | struct buffer_head *bh, *dibh; |
839 | struct gfs2_leaf *leaf; | |
840 | int y; | |
cd915493 SW |
841 | u32 x; |
842 | u64 *lp, bn; | |
b3b94faa DT |
843 | int error; |
844 | ||
845 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
846 | if (error) | |
847 | return error; | |
848 | ||
b3b94faa DT |
849 | /* Turn over a new leaf */ |
850 | ||
c752666c SW |
851 | leaf = new_leaf(inode, &bh, 0); |
852 | if (!leaf) | |
853 | return -ENOSPC; | |
854 | bn = bh->b_blocknr; | |
b3b94faa DT |
855 | |
856 | gfs2_assert(sdp, dip->i_di.di_entries < (1 << 16)); | |
b3b94faa DT |
857 | leaf->lf_entries = cpu_to_be16(dip->i_di.di_entries); |
858 | ||
859 | /* Copy dirents */ | |
860 | ||
861 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh, | |
862 | sizeof(struct gfs2_dinode)); | |
863 | ||
864 | /* Find last entry */ | |
865 | ||
866 | x = 0; | |
c752666c SW |
867 | args.len = bh->b_size - sizeof(struct gfs2_dinode) + |
868 | sizeof(struct gfs2_leaf); | |
869 | args.name = bh->b_data; | |
feaa7bba | 870 | dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size, |
71b86f56 | 871 | gfs2_dirent_last, &args, NULL); |
c752666c SW |
872 | if (!dent) { |
873 | brelse(bh); | |
874 | brelse(dibh); | |
875 | return -EIO; | |
876 | } | |
877 | if (IS_ERR(dent)) { | |
878 | brelse(bh); | |
879 | brelse(dibh); | |
880 | return PTR_ERR(dent); | |
b3b94faa | 881 | } |
b3b94faa DT |
882 | |
883 | /* Adjust the last dirent's record length | |
884 | (Remember that dent still points to the last entry.) */ | |
885 | ||
4dd651ad | 886 | dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) + |
b3b94faa | 887 | sizeof(struct gfs2_dinode) - |
4dd651ad | 888 | sizeof(struct gfs2_leaf)); |
b3b94faa DT |
889 | |
890 | brelse(bh); | |
891 | ||
892 | /* We're done with the new leaf block, now setup the new | |
893 | hash table. */ | |
894 | ||
d4e9c4c3 | 895 | gfs2_trans_add_bh(dip->i_gl, dibh, 1); |
b3b94faa DT |
896 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
897 | ||
cd915493 | 898 | lp = (u64 *)(dibh->b_data + sizeof(struct gfs2_dinode)); |
b3b94faa DT |
899 | |
900 | for (x = sdp->sd_hash_ptrs; x--; lp++) | |
901 | *lp = cpu_to_be64(bn); | |
902 | ||
903 | dip->i_di.di_size = sdp->sd_sb.sb_bsize / 2; | |
904 | dip->i_di.di_blocks++; | |
905 | dip->i_di.di_flags |= GFS2_DIF_EXHASH; | |
906 | dip->i_di.di_payload_format = 0; | |
907 | ||
908 | for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ; | |
909 | dip->i_di.di_depth = y; | |
910 | ||
911 | gfs2_dinode_out(&dip->i_di, dibh->b_data); | |
912 | ||
913 | brelse(dibh); | |
914 | ||
915 | return 0; | |
916 | } | |
917 | ||
918 | /** | |
919 | * dir_split_leaf - Split a leaf block into two | |
920 | * @dip: The GFS2 inode | |
921 | * @index: | |
922 | * @leaf_no: | |
923 | * | |
924 | * Returns: 0 on success, error code on failure | |
925 | */ | |
926 | ||
c752666c | 927 | static int dir_split_leaf(struct inode *inode, const struct qstr *name) |
b3b94faa | 928 | { |
feaa7bba | 929 | struct gfs2_inode *dip = GFS2_I(inode); |
b3b94faa DT |
930 | struct buffer_head *nbh, *obh, *dibh; |
931 | struct gfs2_leaf *nleaf, *oleaf; | |
4da3c646 | 932 | struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new; |
cd915493 SW |
933 | u32 start, len, half_len, divider; |
934 | u64 bn, *lp, leaf_no; | |
935 | u32 index; | |
b3b94faa DT |
936 | int x, moved = 0; |
937 | int error; | |
938 | ||
c752666c SW |
939 | index = name->hash >> (32 - dip->i_di.di_depth); |
940 | error = get_leaf_nr(dip, index, &leaf_no); | |
941 | if (error) | |
942 | return error; | |
b3b94faa DT |
943 | |
944 | /* Get the old leaf block */ | |
b3b94faa DT |
945 | error = get_leaf(dip, leaf_no, &obh); |
946 | if (error) | |
e90deff5 | 947 | return error; |
b3b94faa | 948 | |
b3b94faa | 949 | oleaf = (struct gfs2_leaf *)obh->b_data; |
e90deff5 SW |
950 | if (dip->i_di.di_depth == be16_to_cpu(oleaf->lf_depth)) { |
951 | brelse(obh); | |
952 | return 1; /* can't split */ | |
953 | } | |
954 | ||
955 | gfs2_trans_add_bh(dip->i_gl, obh, 1); | |
b3b94faa | 956 | |
c752666c SW |
957 | nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1); |
958 | if (!nleaf) { | |
959 | brelse(obh); | |
960 | return -ENOSPC; | |
961 | } | |
962 | bn = nbh->b_blocknr; | |
b3b94faa | 963 | |
c752666c | 964 | /* Compute the start and len of leaf pointers in the hash table. */ |
b3b94faa DT |
965 | len = 1 << (dip->i_di.di_depth - be16_to_cpu(oleaf->lf_depth)); |
966 | half_len = len >> 1; | |
967 | if (!half_len) { | |
e90deff5 | 968 | printk(KERN_WARNING "di_depth %u lf_depth %u index %u\n", dip->i_di.di_depth, be16_to_cpu(oleaf->lf_depth), index); |
b3b94faa DT |
969 | gfs2_consist_inode(dip); |
970 | error = -EIO; | |
971 | goto fail_brelse; | |
972 | } | |
973 | ||
974 | start = (index & ~(len - 1)); | |
975 | ||
976 | /* Change the pointers. | |
977 | Don't bother distinguishing stuffed from non-stuffed. | |
978 | This code is complicated enough already. */ | |
cd915493 | 979 | lp = kmalloc(half_len * sizeof(u64), GFP_NOFS | __GFP_NOFAIL); |
b3b94faa | 980 | /* Change the pointers */ |
b3b94faa DT |
981 | for (x = 0; x < half_len; x++) |
982 | lp[x] = cpu_to_be64(bn); | |
983 | ||
cd915493 SW |
984 | error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64), |
985 | half_len * sizeof(u64)); | |
986 | if (error != half_len * sizeof(u64)) { | |
b3b94faa DT |
987 | if (error >= 0) |
988 | error = -EIO; | |
989 | goto fail_lpfree; | |
990 | } | |
991 | ||
992 | kfree(lp); | |
993 | ||
994 | /* Compute the divider */ | |
b3b94faa DT |
995 | divider = (start + half_len) << (32 - dip->i_di.di_depth); |
996 | ||
997 | /* Copy the entries */ | |
b3b94faa DT |
998 | dirent_first(dip, obh, &dent); |
999 | ||
1000 | do { | |
1001 | next = dent; | |
1002 | if (dirent_next(dip, obh, &next)) | |
1003 | next = NULL; | |
1004 | ||
1005 | if (dent->de_inum.no_addr && | |
1006 | be32_to_cpu(dent->de_hash) < divider) { | |
c752666c SW |
1007 | struct qstr str; |
1008 | str.name = (char*)(dent+1); | |
1009 | str.len = be16_to_cpu(dent->de_name_len); | |
1010 | str.hash = be32_to_cpu(dent->de_hash); | |
71b86f56 | 1011 | new = gfs2_dirent_alloc(inode, nbh, &str); |
c752666c SW |
1012 | if (IS_ERR(new)) { |
1013 | error = PTR_ERR(new); | |
1014 | break; | |
1015 | } | |
b3b94faa DT |
1016 | |
1017 | new->de_inum = dent->de_inum; /* No endian worries */ | |
b3b94faa | 1018 | new->de_type = dent->de_type; /* No endian worries */ |
c752666c | 1019 | nleaf->lf_entries = cpu_to_be16(be16_to_cpu(nleaf->lf_entries)+1); |
b3b94faa DT |
1020 | |
1021 | dirent_del(dip, obh, prev, dent); | |
1022 | ||
1023 | if (!oleaf->lf_entries) | |
1024 | gfs2_consist_inode(dip); | |
c752666c | 1025 | oleaf->lf_entries = cpu_to_be16(be16_to_cpu(oleaf->lf_entries)-1); |
b3b94faa DT |
1026 | |
1027 | if (!prev) | |
1028 | prev = dent; | |
1029 | ||
1030 | moved = 1; | |
c752666c | 1031 | } else { |
b3b94faa | 1032 | prev = dent; |
c752666c | 1033 | } |
b3b94faa | 1034 | dent = next; |
c752666c | 1035 | } while (dent); |
b3b94faa | 1036 | |
c752666c | 1037 | oleaf->lf_depth = nleaf->lf_depth; |
b3b94faa DT |
1038 | |
1039 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
feaa7bba | 1040 | if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) { |
b3b94faa DT |
1041 | dip->i_di.di_blocks++; |
1042 | gfs2_dinode_out(&dip->i_di, dibh->b_data); | |
1043 | brelse(dibh); | |
1044 | } | |
1045 | ||
1046 | brelse(obh); | |
1047 | brelse(nbh); | |
1048 | ||
1049 | return error; | |
1050 | ||
e90deff5 | 1051 | fail_lpfree: |
b3b94faa DT |
1052 | kfree(lp); |
1053 | ||
e90deff5 | 1054 | fail_brelse: |
b3b94faa | 1055 | brelse(obh); |
b3b94faa DT |
1056 | brelse(nbh); |
1057 | return error; | |
1058 | } | |
1059 | ||
1060 | /** | |
1061 | * dir_double_exhash - Double size of ExHash table | |
1062 | * @dip: The GFS2 dinode | |
1063 | * | |
1064 | * Returns: 0 on success, error code on failure | |
1065 | */ | |
1066 | ||
1067 | static int dir_double_exhash(struct gfs2_inode *dip) | |
1068 | { | |
feaa7bba | 1069 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); |
b3b94faa | 1070 | struct buffer_head *dibh; |
cd915493 SW |
1071 | u32 hsize; |
1072 | u64 *buf; | |
1073 | u64 *from, *to; | |
1074 | u64 block; | |
b3b94faa DT |
1075 | int x; |
1076 | int error = 0; | |
1077 | ||
1078 | hsize = 1 << dip->i_di.di_depth; | |
cd915493 | 1079 | if (hsize * sizeof(u64) != dip->i_di.di_size) { |
b3b94faa DT |
1080 | gfs2_consist_inode(dip); |
1081 | return -EIO; | |
1082 | } | |
1083 | ||
1084 | /* Allocate both the "from" and "to" buffers in one big chunk */ | |
1085 | ||
1086 | buf = kcalloc(3, sdp->sd_hash_bsize, GFP_KERNEL | __GFP_NOFAIL); | |
1087 | ||
1088 | for (block = dip->i_di.di_size >> sdp->sd_hash_bsize_shift; block--;) { | |
e13940ba | 1089 | error = gfs2_dir_read_data(dip, (char *)buf, |
b3b94faa | 1090 | block * sdp->sd_hash_bsize, |
7276b3b0 | 1091 | sdp->sd_hash_bsize, 1); |
b3b94faa DT |
1092 | if (error != sdp->sd_hash_bsize) { |
1093 | if (error >= 0) | |
1094 | error = -EIO; | |
1095 | goto fail; | |
1096 | } | |
1097 | ||
1098 | from = buf; | |
cd915493 | 1099 | to = (u64 *)((char *)buf + sdp->sd_hash_bsize); |
b3b94faa DT |
1100 | |
1101 | for (x = sdp->sd_hash_ptrs; x--; from++) { | |
1102 | *to++ = *from; /* No endianess worries */ | |
1103 | *to++ = *from; | |
1104 | } | |
1105 | ||
e13940ba | 1106 | error = gfs2_dir_write_data(dip, |
b3b94faa DT |
1107 | (char *)buf + sdp->sd_hash_bsize, |
1108 | block * sdp->sd_sb.sb_bsize, | |
1109 | sdp->sd_sb.sb_bsize); | |
1110 | if (error != sdp->sd_sb.sb_bsize) { | |
1111 | if (error >= 0) | |
1112 | error = -EIO; | |
1113 | goto fail; | |
1114 | } | |
1115 | } | |
1116 | ||
1117 | kfree(buf); | |
1118 | ||
1119 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
1120 | if (!gfs2_assert_withdraw(sdp, !error)) { | |
1121 | dip->i_di.di_depth++; | |
1122 | gfs2_dinode_out(&dip->i_di, dibh->b_data); | |
1123 | brelse(dibh); | |
1124 | } | |
1125 | ||
1126 | return error; | |
1127 | ||
a91ea69f | 1128 | fail: |
b3b94faa | 1129 | kfree(buf); |
b3b94faa DT |
1130 | return error; |
1131 | } | |
1132 | ||
1133 | /** | |
1134 | * compare_dents - compare directory entries by hash value | |
1135 | * @a: first dent | |
1136 | * @b: second dent | |
1137 | * | |
1138 | * When comparing the hash entries of @a to @b: | |
1139 | * gt: returns 1 | |
1140 | * lt: returns -1 | |
1141 | * eq: returns 0 | |
1142 | */ | |
1143 | ||
1144 | static int compare_dents(const void *a, const void *b) | |
1145 | { | |
2bdbc5d7 | 1146 | const struct gfs2_dirent *dent_a, *dent_b; |
cd915493 | 1147 | u32 hash_a, hash_b; |
b3b94faa DT |
1148 | int ret = 0; |
1149 | ||
2bdbc5d7 | 1150 | dent_a = *(const struct gfs2_dirent **)a; |
c752666c | 1151 | hash_a = be32_to_cpu(dent_a->de_hash); |
b3b94faa | 1152 | |
2bdbc5d7 | 1153 | dent_b = *(const struct gfs2_dirent **)b; |
c752666c | 1154 | hash_b = be32_to_cpu(dent_b->de_hash); |
b3b94faa DT |
1155 | |
1156 | if (hash_a > hash_b) | |
1157 | ret = 1; | |
1158 | else if (hash_a < hash_b) | |
1159 | ret = -1; | |
1160 | else { | |
4dd651ad SW |
1161 | unsigned int len_a = be16_to_cpu(dent_a->de_name_len); |
1162 | unsigned int len_b = be16_to_cpu(dent_b->de_name_len); | |
b3b94faa DT |
1163 | |
1164 | if (len_a > len_b) | |
1165 | ret = 1; | |
1166 | else if (len_a < len_b) | |
1167 | ret = -1; | |
1168 | else | |
2bdbc5d7 | 1169 | ret = memcmp(dent_a + 1, dent_b + 1, len_a); |
b3b94faa DT |
1170 | } |
1171 | ||
1172 | return ret; | |
1173 | } | |
1174 | ||
1175 | /** | |
1176 | * do_filldir_main - read out directory entries | |
1177 | * @dip: The GFS2 inode | |
1178 | * @offset: The offset in the file to read from | |
1179 | * @opaque: opaque data to pass to filldir | |
1180 | * @filldir: The function to pass entries to | |
1181 | * @darr: an array of struct gfs2_dirent pointers to read | |
1182 | * @entries: the number of entries in darr | |
1183 | * @copied: pointer to int that's non-zero if a entry has been copied out | |
1184 | * | |
1185 | * Jump through some hoops to make sure that if there are hash collsions, | |
1186 | * they are read out at the beginning of a buffer. We want to minimize | |
1187 | * the possibility that they will fall into different readdir buffers or | |
1188 | * that someone will want to seek to that location. | |
1189 | * | |
1190 | * Returns: errno, >0 on exception from filldir | |
1191 | */ | |
1192 | ||
cd915493 | 1193 | static int do_filldir_main(struct gfs2_inode *dip, u64 *offset, |
b3b94faa | 1194 | void *opaque, gfs2_filldir_t filldir, |
cd915493 | 1195 | const struct gfs2_dirent **darr, u32 entries, |
b3b94faa DT |
1196 | int *copied) |
1197 | { | |
71b86f56 | 1198 | const struct gfs2_dirent *dent, *dent_next; |
b3b94faa | 1199 | struct gfs2_inum inum; |
cd915493 | 1200 | u64 off, off_next; |
b3b94faa DT |
1201 | unsigned int x, y; |
1202 | int run = 0; | |
1203 | int error = 0; | |
1204 | ||
1205 | sort(darr, entries, sizeof(struct gfs2_dirent *), compare_dents, NULL); | |
1206 | ||
1207 | dent_next = darr[0]; | |
1208 | off_next = be32_to_cpu(dent_next->de_hash); | |
1209 | off_next = gfs2_disk_hash2offset(off_next); | |
1210 | ||
1211 | for (x = 0, y = 1; x < entries; x++, y++) { | |
1212 | dent = dent_next; | |
1213 | off = off_next; | |
1214 | ||
1215 | if (y < entries) { | |
1216 | dent_next = darr[y]; | |
1217 | off_next = be32_to_cpu(dent_next->de_hash); | |
1218 | off_next = gfs2_disk_hash2offset(off_next); | |
1219 | ||
1220 | if (off < *offset) | |
1221 | continue; | |
1222 | *offset = off; | |
1223 | ||
1224 | if (off_next == off) { | |
1225 | if (*copied && !run) | |
1226 | return 1; | |
1227 | run = 1; | |
1228 | } else | |
1229 | run = 0; | |
1230 | } else { | |
1231 | if (off < *offset) | |
1232 | continue; | |
1233 | *offset = off; | |
1234 | } | |
1235 | ||
1236 | gfs2_inum_in(&inum, (char *)&dent->de_inum); | |
1237 | ||
2bdbc5d7 | 1238 | error = filldir(opaque, (const char *)(dent + 1), |
4dd651ad | 1239 | be16_to_cpu(dent->de_name_len), |
b3b94faa | 1240 | off, &inum, |
4dd651ad | 1241 | be16_to_cpu(dent->de_type)); |
b3b94faa DT |
1242 | if (error) |
1243 | return 1; | |
1244 | ||
1245 | *copied = 1; | |
1246 | } | |
1247 | ||
1248 | /* Increment the *offset by one, so the next time we come into the | |
1249 | do_filldir fxn, we get the next entry instead of the last one in the | |
1250 | current leaf */ | |
1251 | ||
1252 | (*offset)++; | |
1253 | ||
1254 | return 0; | |
1255 | } | |
1256 | ||
71b86f56 SW |
1257 | static int gfs2_dir_read_leaf(struct inode *inode, u64 *offset, void *opaque, |
1258 | gfs2_filldir_t filldir, int *copied, | |
1259 | unsigned *depth, u64 leaf_no) | |
b3b94faa | 1260 | { |
feaa7bba | 1261 | struct gfs2_inode *ip = GFS2_I(inode); |
71b86f56 SW |
1262 | struct buffer_head *bh; |
1263 | struct gfs2_leaf *lf; | |
1264 | unsigned entries = 0; | |
1265 | unsigned leaves = 0; | |
1266 | const struct gfs2_dirent **darr, *dent; | |
1267 | struct dirent_gather g; | |
1268 | struct buffer_head **larr; | |
1269 | int leaf = 0; | |
1270 | int error, i; | |
1271 | u64 lfn = leaf_no; | |
b3b94faa | 1272 | |
b3b94faa | 1273 | do { |
71b86f56 | 1274 | error = get_leaf(ip, lfn, &bh); |
b3b94faa | 1275 | if (error) |
71b86f56 SW |
1276 | goto out; |
1277 | lf = (struct gfs2_leaf *)bh->b_data; | |
1278 | if (leaves == 0) | |
1279 | *depth = be16_to_cpu(lf->lf_depth); | |
1280 | entries += be16_to_cpu(lf->lf_entries); | |
1281 | leaves++; | |
1282 | lfn = be64_to_cpu(lf->lf_next); | |
1283 | brelse(bh); | |
1284 | } while(lfn); | |
b3b94faa DT |
1285 | |
1286 | if (!entries) | |
1287 | return 0; | |
1288 | ||
71b86f56 | 1289 | error = -ENOMEM; |
2bdbc5d7 | 1290 | larr = vmalloc((leaves + entries) * sizeof(void *)); |
71b86f56 SW |
1291 | if (!larr) |
1292 | goto out; | |
1293 | darr = (const struct gfs2_dirent **)(larr + leaves); | |
1294 | g.pdent = darr; | |
1295 | g.offset = 0; | |
1296 | lfn = leaf_no; | |
b3b94faa | 1297 | |
71b86f56 SW |
1298 | do { |
1299 | error = get_leaf(ip, lfn, &bh); | |
b3b94faa | 1300 | if (error) |
71b86f56 SW |
1301 | goto out_kfree; |
1302 | lf = (struct gfs2_leaf *)bh->b_data; | |
1303 | lfn = be64_to_cpu(lf->lf_next); | |
1304 | if (lf->lf_entries) { | |
1305 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, | |
1306 | gfs2_dirent_gather, NULL, &g); | |
1307 | error = PTR_ERR(dent); | |
1308 | if (IS_ERR(dent)) { | |
1309 | goto out_kfree; | |
1310 | } | |
1311 | error = 0; | |
1312 | larr[leaf++] = bh; | |
b3b94faa | 1313 | } else { |
71b86f56 | 1314 | brelse(bh); |
b3b94faa | 1315 | } |
71b86f56 | 1316 | } while(lfn); |
b3b94faa | 1317 | |
71b86f56 | 1318 | error = do_filldir_main(ip, offset, opaque, filldir, darr, |
b3b94faa | 1319 | entries, copied); |
71b86f56 SW |
1320 | out_kfree: |
1321 | for(i = 0; i < leaf; i++) | |
1322 | brelse(larr[i]); | |
fe1bdedc | 1323 | vfree(larr); |
71b86f56 | 1324 | out: |
b3b94faa DT |
1325 | return error; |
1326 | } | |
1327 | ||
1328 | /** | |
c752666c SW |
1329 | * dir_e_read - Reads the entries from a directory into a filldir buffer |
1330 | * @dip: dinode pointer | |
1331 | * @offset: the hash of the last entry read shifted to the right once | |
1332 | * @opaque: buffer for the filldir function to fill | |
1333 | * @filldir: points to the filldir function to use | |
b3b94faa | 1334 | * |
c752666c | 1335 | * Returns: errno |
b3b94faa DT |
1336 | */ |
1337 | ||
cd915493 | 1338 | static int dir_e_read(struct inode *inode, u64 *offset, void *opaque, |
c752666c | 1339 | gfs2_filldir_t filldir) |
b3b94faa | 1340 | { |
feaa7bba SW |
1341 | struct gfs2_inode *dip = GFS2_I(inode); |
1342 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
cd915493 SW |
1343 | u32 hsize, len = 0; |
1344 | u32 ht_offset, lp_offset, ht_offset_cur = -1; | |
1345 | u32 hash, index; | |
1346 | u64 *lp; | |
c752666c SW |
1347 | int copied = 0; |
1348 | int error = 0; | |
4da3c646 | 1349 | unsigned depth = 0; |
b3b94faa DT |
1350 | |
1351 | hsize = 1 << dip->i_di.di_depth; | |
cd915493 | 1352 | if (hsize * sizeof(u64) != dip->i_di.di_size) { |
b3b94faa DT |
1353 | gfs2_consist_inode(dip); |
1354 | return -EIO; | |
1355 | } | |
1356 | ||
1357 | hash = gfs2_dir_offset2hash(*offset); | |
1358 | index = hash >> (32 - dip->i_di.di_depth); | |
1359 | ||
1360 | lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL); | |
1361 | if (!lp) | |
1362 | return -ENOMEM; | |
1363 | ||
1364 | while (index < hsize) { | |
1365 | lp_offset = index & (sdp->sd_hash_ptrs - 1); | |
1366 | ht_offset = index - lp_offset; | |
1367 | ||
1368 | if (ht_offset_cur != ht_offset) { | |
e13940ba | 1369 | error = gfs2_dir_read_data(dip, (char *)lp, |
cd915493 | 1370 | ht_offset * sizeof(u64), |
7276b3b0 | 1371 | sdp->sd_hash_bsize, 1); |
b3b94faa DT |
1372 | if (error != sdp->sd_hash_bsize) { |
1373 | if (error >= 0) | |
1374 | error = -EIO; | |
1375 | goto out; | |
1376 | } | |
1377 | ht_offset_cur = ht_offset; | |
1378 | } | |
1379 | ||
71b86f56 SW |
1380 | error = gfs2_dir_read_leaf(inode, offset, opaque, filldir, |
1381 | &copied, &depth, | |
1382 | be64_to_cpu(lp[lp_offset])); | |
b3b94faa | 1383 | if (error) |
71b86f56 | 1384 | break; |
b3b94faa | 1385 | |
71b86f56 | 1386 | len = 1 << (dip->i_di.di_depth - depth); |
b3b94faa DT |
1387 | index = (index & ~(len - 1)) + len; |
1388 | } | |
1389 | ||
71b86f56 | 1390 | out: |
b3b94faa | 1391 | kfree(lp); |
71b86f56 SW |
1392 | if (error > 0) |
1393 | error = 0; | |
b3b94faa DT |
1394 | return error; |
1395 | } | |
1396 | ||
cd915493 | 1397 | int gfs2_dir_read(struct inode *inode, u64 *offset, void *opaque, |
71b86f56 | 1398 | gfs2_filldir_t filldir) |
b3b94faa | 1399 | { |
feaa7bba | 1400 | struct gfs2_inode *dip = GFS2_I(inode); |
71b86f56 SW |
1401 | struct dirent_gather g; |
1402 | const struct gfs2_dirent **darr, *dent; | |
b3b94faa DT |
1403 | struct buffer_head *dibh; |
1404 | int copied = 0; | |
1405 | int error; | |
1406 | ||
71b86f56 SW |
1407 | if (!dip->i_di.di_entries) |
1408 | return 0; | |
1409 | ||
1410 | if (dip->i_di.di_flags & GFS2_DIF_EXHASH) | |
1411 | return dir_e_read(inode, offset, opaque, filldir); | |
1412 | ||
b3b94faa DT |
1413 | if (!gfs2_is_stuffed(dip)) { |
1414 | gfs2_consist_inode(dip); | |
1415 | return -EIO; | |
1416 | } | |
1417 | ||
b3b94faa DT |
1418 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1419 | if (error) | |
1420 | return error; | |
1421 | ||
71b86f56 SW |
1422 | error = -ENOMEM; |
1423 | darr = kmalloc(dip->i_di.di_entries * sizeof(struct gfs2_dirent *), | |
1424 | GFP_KERNEL); | |
1425 | if (darr) { | |
1426 | g.pdent = darr; | |
1427 | g.offset = 0; | |
1428 | dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size, | |
1429 | gfs2_dirent_gather, NULL, &g); | |
1430 | if (IS_ERR(dent)) { | |
1431 | error = PTR_ERR(dent); | |
1432 | goto out; | |
1433 | } | |
1434 | error = do_filldir_main(dip, offset, opaque, filldir, darr, | |
1435 | dip->i_di.di_entries, &copied); | |
1436 | out: | |
1437 | kfree(darr); | |
1438 | } | |
1439 | ||
b3b94faa DT |
1440 | if (error > 0) |
1441 | error = 0; | |
1442 | ||
1443 | brelse(dibh); | |
1444 | ||
1445 | return error; | |
1446 | } | |
1447 | ||
b3b94faa DT |
1448 | /** |
1449 | * gfs2_dir_search - Search a directory | |
1450 | * @dip: The GFS2 inode | |
1451 | * @filename: | |
1452 | * @inode: | |
1453 | * | |
1454 | * This routine searches a directory for a file or another directory. | |
1455 | * Assumes a glock is held on dip. | |
1456 | * | |
1457 | * Returns: errno | |
1458 | */ | |
1459 | ||
c752666c | 1460 | int gfs2_dir_search(struct inode *dir, const struct qstr *name, |
b3b94faa DT |
1461 | struct gfs2_inum *inum, unsigned int *type) |
1462 | { | |
c752666c SW |
1463 | struct buffer_head *bh; |
1464 | struct gfs2_dirent *dent; | |
1465 | ||
1466 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1467 | if (dent) { | |
1468 | if (IS_ERR(dent)) | |
1469 | return PTR_ERR(dent); | |
1470 | if (inum) | |
1471 | gfs2_inum_in(inum, (char *)&dent->de_inum); | |
1472 | if (type) | |
1473 | *type = be16_to_cpu(dent->de_type); | |
1474 | brelse(bh); | |
1475 | return 0; | |
1476 | } | |
1477 | return -ENOENT; | |
1478 | } | |
1479 | ||
1480 | static int dir_new_leaf(struct inode *inode, const struct qstr *name) | |
1481 | { | |
1482 | struct buffer_head *bh, *obh; | |
feaa7bba | 1483 | struct gfs2_inode *ip = GFS2_I(inode); |
c752666c | 1484 | struct gfs2_leaf *leaf, *oleaf; |
b3b94faa | 1485 | int error; |
c752666c SW |
1486 | u32 index; |
1487 | u64 bn; | |
b3b94faa | 1488 | |
c752666c SW |
1489 | index = name->hash >> (32 - ip->i_di.di_depth); |
1490 | error = get_first_leaf(ip, index, &obh); | |
1491 | if (error) | |
1492 | return error; | |
1493 | do { | |
1494 | oleaf = (struct gfs2_leaf *)obh->b_data; | |
1495 | bn = be64_to_cpu(oleaf->lf_next); | |
1496 | if (!bn) | |
1497 | break; | |
1498 | brelse(obh); | |
1499 | error = get_leaf(ip, bn, &obh); | |
1500 | if (error) | |
1501 | return error; | |
1502 | } while(1); | |
b3b94faa | 1503 | |
c752666c SW |
1504 | gfs2_trans_add_bh(ip->i_gl, obh, 1); |
1505 | ||
1506 | leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth)); | |
1507 | if (!leaf) { | |
1508 | brelse(obh); | |
1509 | return -ENOSPC; | |
1510 | } | |
4d8012b6 | 1511 | oleaf->lf_next = cpu_to_be64(bh->b_blocknr); |
c752666c SW |
1512 | brelse(bh); |
1513 | brelse(obh); | |
1514 | ||
1515 | error = gfs2_meta_inode_buffer(ip, &bh); | |
1516 | if (error) | |
1517 | return error; | |
1518 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
1519 | ip->i_di.di_blocks++; | |
1520 | gfs2_dinode_out(&ip->i_di, bh->b_data); | |
1521 | brelse(bh); | |
1522 | return 0; | |
b3b94faa DT |
1523 | } |
1524 | ||
1525 | /** | |
1526 | * gfs2_dir_add - Add new filename into directory | |
1527 | * @dip: The GFS2 inode | |
1528 | * @filename: The new name | |
1529 | * @inode: The inode number of the entry | |
1530 | * @type: The type of the entry | |
1531 | * | |
1532 | * Returns: 0 on success, error code on failure | |
1533 | */ | |
1534 | ||
c752666c SW |
1535 | int gfs2_dir_add(struct inode *inode, const struct qstr *name, |
1536 | const struct gfs2_inum *inum, unsigned type) | |
b3b94faa | 1537 | { |
feaa7bba | 1538 | struct gfs2_inode *ip = GFS2_I(inode); |
c752666c SW |
1539 | struct buffer_head *bh; |
1540 | struct gfs2_dirent *dent; | |
1541 | struct gfs2_leaf *leaf; | |
b3b94faa DT |
1542 | int error; |
1543 | ||
c752666c SW |
1544 | while(1) { |
1545 | dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, | |
1546 | &bh); | |
1547 | if (dent) { | |
1548 | if (IS_ERR(dent)) | |
1549 | return PTR_ERR(dent); | |
1550 | dent = gfs2_init_dirent(inode, dent, name, bh); | |
1551 | gfs2_inum_out(inum, (char *)&dent->de_inum); | |
1552 | dent->de_type = cpu_to_be16(type); | |
1553 | if (ip->i_di.di_flags & GFS2_DIF_EXHASH) { | |
1554 | leaf = (struct gfs2_leaf *)bh->b_data; | |
1555 | leaf->lf_entries = cpu_to_be16(be16_to_cpu(leaf->lf_entries) + 1); | |
1556 | } | |
1557 | brelse(bh); | |
1558 | error = gfs2_meta_inode_buffer(ip, &bh); | |
1559 | if (error) | |
1560 | break; | |
1561 | gfs2_trans_add_bh(ip->i_gl, bh, 1); | |
1562 | ip->i_di.di_entries++; | |
1563 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | |
1564 | gfs2_dinode_out(&ip->i_di, bh->b_data); | |
1565 | brelse(bh); | |
1566 | error = 0; | |
1567 | break; | |
1568 | } | |
1569 | if (!(ip->i_di.di_flags & GFS2_DIF_EXHASH)) { | |
1570 | error = dir_make_exhash(inode); | |
1571 | if (error) | |
1572 | break; | |
1573 | continue; | |
1574 | } | |
1575 | error = dir_split_leaf(inode, name); | |
1576 | if (error == 0) | |
1577 | continue; | |
e90deff5 | 1578 | if (error < 0) |
c752666c SW |
1579 | break; |
1580 | if (ip->i_di.di_depth < GFS2_DIR_MAX_DEPTH) { | |
1581 | error = dir_double_exhash(ip); | |
1582 | if (error) | |
1583 | break; | |
1584 | error = dir_split_leaf(inode, name); | |
e90deff5 | 1585 | if (error < 0) |
c752666c | 1586 | break; |
e90deff5 SW |
1587 | if (error == 0) |
1588 | continue; | |
c752666c SW |
1589 | } |
1590 | error = dir_new_leaf(inode, name); | |
1591 | if (!error) | |
1592 | continue; | |
1593 | error = -ENOSPC; | |
1594 | break; | |
1595 | } | |
b3b94faa DT |
1596 | return error; |
1597 | } | |
1598 | ||
c752666c | 1599 | |
b3b94faa DT |
1600 | /** |
1601 | * gfs2_dir_del - Delete a directory entry | |
1602 | * @dip: The GFS2 inode | |
1603 | * @filename: The filename | |
1604 | * | |
1605 | * Returns: 0 on success, error code on failure | |
1606 | */ | |
1607 | ||
c752666c | 1608 | int gfs2_dir_del(struct gfs2_inode *dip, const struct qstr *name) |
b3b94faa | 1609 | { |
c752666c SW |
1610 | struct gfs2_dirent *dent, *prev = NULL; |
1611 | struct buffer_head *bh; | |
b3b94faa DT |
1612 | int error; |
1613 | ||
c752666c SW |
1614 | /* Returns _either_ the entry (if its first in block) or the |
1615 | previous entry otherwise */ | |
feaa7bba | 1616 | dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh); |
c752666c SW |
1617 | if (!dent) { |
1618 | gfs2_consist_inode(dip); | |
1619 | return -EIO; | |
1620 | } | |
1621 | if (IS_ERR(dent)) { | |
1622 | gfs2_consist_inode(dip); | |
1623 | return PTR_ERR(dent); | |
1624 | } | |
1625 | /* If not first in block, adjust pointers accordingly */ | |
71b86f56 | 1626 | if (gfs2_dirent_find(dent, name, NULL) == 0) { |
c752666c SW |
1627 | prev = dent; |
1628 | dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len)); | |
1629 | } | |
1630 | ||
1631 | dirent_del(dip, bh, prev, dent); | |
1632 | if (dip->i_di.di_flags & GFS2_DIF_EXHASH) { | |
1633 | struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data; | |
1634 | u16 entries = be16_to_cpu(leaf->lf_entries); | |
1635 | if (!entries) | |
1636 | gfs2_consist_inode(dip); | |
1637 | leaf->lf_entries = cpu_to_be16(--entries); | |
c752666c | 1638 | } |
ed386507 | 1639 | brelse(bh); |
c752666c SW |
1640 | |
1641 | error = gfs2_meta_inode_buffer(dip, &bh); | |
1642 | if (error) | |
1643 | return error; | |
1644 | ||
1645 | if (!dip->i_di.di_entries) | |
1646 | gfs2_consist_inode(dip); | |
1647 | gfs2_trans_add_bh(dip->i_gl, bh, 1); | |
1648 | dip->i_di.di_entries--; | |
1649 | dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds(); | |
1650 | gfs2_dinode_out(&dip->i_di, bh->b_data); | |
1651 | brelse(bh); | |
feaa7bba | 1652 | mark_inode_dirty(&dip->i_inode); |
b3b94faa DT |
1653 | |
1654 | return error; | |
1655 | } | |
1656 | ||
b3b94faa DT |
1657 | /** |
1658 | * gfs2_dir_mvino - Change inode number of directory entry | |
1659 | * @dip: The GFS2 inode | |
1660 | * @filename: | |
1661 | * @new_inode: | |
1662 | * | |
1663 | * This routine changes the inode number of a directory entry. It's used | |
1664 | * by rename to change ".." when a directory is moved. | |
1665 | * Assumes a glock is held on dvp. | |
1666 | * | |
1667 | * Returns: errno | |
1668 | */ | |
1669 | ||
c752666c | 1670 | int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename, |
b3b94faa DT |
1671 | struct gfs2_inum *inum, unsigned int new_type) |
1672 | { | |
c752666c SW |
1673 | struct buffer_head *bh; |
1674 | struct gfs2_dirent *dent; | |
b3b94faa DT |
1675 | int error; |
1676 | ||
feaa7bba | 1677 | dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh); |
c752666c SW |
1678 | if (!dent) { |
1679 | gfs2_consist_inode(dip); | |
1680 | return -EIO; | |
1681 | } | |
1682 | if (IS_ERR(dent)) | |
1683 | return PTR_ERR(dent); | |
b3b94faa | 1684 | |
c752666c SW |
1685 | gfs2_trans_add_bh(dip->i_gl, bh, 1); |
1686 | gfs2_inum_out(inum, (char *)&dent->de_inum); | |
1687 | dent->de_type = cpu_to_be16(new_type); | |
1688 | ||
1689 | if (dip->i_di.di_flags & GFS2_DIF_EXHASH) { | |
1690 | brelse(bh); | |
1691 | error = gfs2_meta_inode_buffer(dip, &bh); | |
1692 | if (error) | |
1693 | return error; | |
1694 | gfs2_trans_add_bh(dip->i_gl, bh, 1); | |
1695 | } | |
1696 | ||
1697 | dip->i_di.di_mtime = dip->i_di.di_ctime = get_seconds(); | |
1698 | gfs2_dinode_out(&dip->i_di, bh->b_data); | |
1699 | brelse(bh); | |
1700 | return 0; | |
b3b94faa DT |
1701 | } |
1702 | ||
1703 | /** | |
1704 | * foreach_leaf - call a function for each leaf in a directory | |
1705 | * @dip: the directory | |
1706 | * @lc: the function to call for each each | |
1707 | * @data: private data to pass to it | |
1708 | * | |
1709 | * Returns: errno | |
1710 | */ | |
1711 | ||
1712 | static int foreach_leaf(struct gfs2_inode *dip, leaf_call_t lc, void *data) | |
1713 | { | |
feaa7bba | 1714 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); |
b3b94faa | 1715 | struct buffer_head *bh; |
c752666c | 1716 | struct gfs2_leaf *leaf; |
cd915493 SW |
1717 | u32 hsize, len; |
1718 | u32 ht_offset, lp_offset, ht_offset_cur = -1; | |
1719 | u32 index = 0; | |
1720 | u64 *lp; | |
1721 | u64 leaf_no; | |
b3b94faa DT |
1722 | int error = 0; |
1723 | ||
1724 | hsize = 1 << dip->i_di.di_depth; | |
cd915493 | 1725 | if (hsize * sizeof(u64) != dip->i_di.di_size) { |
b3b94faa DT |
1726 | gfs2_consist_inode(dip); |
1727 | return -EIO; | |
1728 | } | |
1729 | ||
1730 | lp = kmalloc(sdp->sd_hash_bsize, GFP_KERNEL); | |
1731 | if (!lp) | |
1732 | return -ENOMEM; | |
1733 | ||
1734 | while (index < hsize) { | |
1735 | lp_offset = index & (sdp->sd_hash_ptrs - 1); | |
1736 | ht_offset = index - lp_offset; | |
1737 | ||
1738 | if (ht_offset_cur != ht_offset) { | |
e13940ba | 1739 | error = gfs2_dir_read_data(dip, (char *)lp, |
cd915493 | 1740 | ht_offset * sizeof(u64), |
7276b3b0 | 1741 | sdp->sd_hash_bsize, 1); |
b3b94faa DT |
1742 | if (error != sdp->sd_hash_bsize) { |
1743 | if (error >= 0) | |
1744 | error = -EIO; | |
1745 | goto out; | |
1746 | } | |
1747 | ht_offset_cur = ht_offset; | |
1748 | } | |
1749 | ||
1750 | leaf_no = be64_to_cpu(lp[lp_offset]); | |
1751 | if (leaf_no) { | |
1752 | error = get_leaf(dip, leaf_no, &bh); | |
1753 | if (error) | |
1754 | goto out; | |
c752666c | 1755 | leaf = (struct gfs2_leaf *)bh->b_data; |
c752666c | 1756 | len = 1 << (dip->i_di.di_depth - be16_to_cpu(leaf->lf_depth)); |
634ee0b9 | 1757 | brelse(bh); |
b3b94faa DT |
1758 | |
1759 | error = lc(dip, index, len, leaf_no, data); | |
1760 | if (error) | |
1761 | goto out; | |
1762 | ||
1763 | index = (index & ~(len - 1)) + len; | |
1764 | } else | |
1765 | index++; | |
1766 | } | |
1767 | ||
1768 | if (index != hsize) { | |
1769 | gfs2_consist_inode(dip); | |
1770 | error = -EIO; | |
1771 | } | |
1772 | ||
634ee0b9 | 1773 | out: |
b3b94faa DT |
1774 | kfree(lp); |
1775 | ||
1776 | return error; | |
1777 | } | |
1778 | ||
1779 | /** | |
1780 | * leaf_dealloc - Deallocate a directory leaf | |
1781 | * @dip: the directory | |
1782 | * @index: the hash table offset in the directory | |
1783 | * @len: the number of pointers to this leaf | |
1784 | * @leaf_no: the leaf number | |
1785 | * @data: not used | |
1786 | * | |
1787 | * Returns: errno | |
1788 | */ | |
1789 | ||
cd915493 SW |
1790 | static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len, |
1791 | u64 leaf_no, void *data) | |
b3b94faa | 1792 | { |
feaa7bba | 1793 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); |
c752666c | 1794 | struct gfs2_leaf *tmp_leaf; |
b3b94faa DT |
1795 | struct gfs2_rgrp_list rlist; |
1796 | struct buffer_head *bh, *dibh; | |
cd915493 | 1797 | u64 blk, nblk; |
b3b94faa DT |
1798 | unsigned int rg_blocks = 0, l_blocks = 0; |
1799 | char *ht; | |
cd915493 | 1800 | unsigned int x, size = len * sizeof(u64); |
b3b94faa DT |
1801 | int error; |
1802 | ||
1803 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); | |
1804 | ||
1805 | ht = kzalloc(size, GFP_KERNEL); | |
1806 | if (!ht) | |
1807 | return -ENOMEM; | |
1808 | ||
1809 | gfs2_alloc_get(dip); | |
1810 | ||
1811 | error = gfs2_quota_hold(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
1812 | if (error) | |
1813 | goto out; | |
1814 | ||
1815 | error = gfs2_rindex_hold(sdp, &dip->i_alloc.al_ri_gh); | |
1816 | if (error) | |
1817 | goto out_qs; | |
1818 | ||
1819 | /* Count the number of leaves */ | |
1820 | ||
c752666c | 1821 | for (blk = leaf_no; blk; blk = nblk) { |
b3b94faa DT |
1822 | error = get_leaf(dip, blk, &bh); |
1823 | if (error) | |
1824 | goto out_rlist; | |
c752666c SW |
1825 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
1826 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
b3b94faa DT |
1827 | brelse(bh); |
1828 | ||
1829 | gfs2_rlist_add(sdp, &rlist, blk); | |
1830 | l_blocks++; | |
1831 | } | |
1832 | ||
1833 | gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, 0); | |
1834 | ||
1835 | for (x = 0; x < rlist.rl_rgrps; x++) { | |
1836 | struct gfs2_rgrpd *rgd; | |
5c676f6d | 1837 | rgd = rlist.rl_ghs[x].gh_gl->gl_object; |
b3b94faa DT |
1838 | rg_blocks += rgd->rd_ri.ri_length; |
1839 | } | |
1840 | ||
1841 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
1842 | if (error) | |
1843 | goto out_rlist; | |
1844 | ||
1845 | error = gfs2_trans_begin(sdp, | |
5c676f6d | 1846 | rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) + |
b3b94faa DT |
1847 | RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks); |
1848 | if (error) | |
1849 | goto out_rg_gunlock; | |
1850 | ||
c752666c | 1851 | for (blk = leaf_no; blk; blk = nblk) { |
b3b94faa DT |
1852 | error = get_leaf(dip, blk, &bh); |
1853 | if (error) | |
1854 | goto out_end_trans; | |
c752666c SW |
1855 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
1856 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
b3b94faa DT |
1857 | brelse(bh); |
1858 | ||
1859 | gfs2_free_meta(dip, blk, 1); | |
1860 | ||
1861 | if (!dip->i_di.di_blocks) | |
1862 | gfs2_consist_inode(dip); | |
1863 | dip->i_di.di_blocks--; | |
1864 | } | |
1865 | ||
cd915493 | 1866 | error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size); |
b3b94faa DT |
1867 | if (error != size) { |
1868 | if (error >= 0) | |
1869 | error = -EIO; | |
1870 | goto out_end_trans; | |
1871 | } | |
1872 | ||
1873 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
1874 | if (error) | |
1875 | goto out_end_trans; | |
1876 | ||
d4e9c4c3 | 1877 | gfs2_trans_add_bh(dip->i_gl, dibh, 1); |
b3b94faa DT |
1878 | gfs2_dinode_out(&dip->i_di, dibh->b_data); |
1879 | brelse(dibh); | |
1880 | ||
a91ea69f | 1881 | out_end_trans: |
b3b94faa | 1882 | gfs2_trans_end(sdp); |
a91ea69f | 1883 | out_rg_gunlock: |
b3b94faa | 1884 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); |
a91ea69f | 1885 | out_rlist: |
b3b94faa DT |
1886 | gfs2_rlist_free(&rlist); |
1887 | gfs2_glock_dq_uninit(&dip->i_alloc.al_ri_gh); | |
a91ea69f | 1888 | out_qs: |
b3b94faa | 1889 | gfs2_quota_unhold(dip); |
a91ea69f | 1890 | out: |
b3b94faa DT |
1891 | gfs2_alloc_put(dip); |
1892 | kfree(ht); | |
b3b94faa DT |
1893 | return error; |
1894 | } | |
1895 | ||
1896 | /** | |
1897 | * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory | |
1898 | * @dip: the directory | |
1899 | * | |
1900 | * Dealloc all on-disk directory leaves to FREEMETA state | |
1901 | * Change on-disk inode type to "regular file" | |
1902 | * | |
1903 | * Returns: errno | |
1904 | */ | |
1905 | ||
1906 | int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip) | |
1907 | { | |
feaa7bba | 1908 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); |
b3b94faa DT |
1909 | struct buffer_head *bh; |
1910 | int error; | |
1911 | ||
1912 | /* Dealloc on-disk leaves to FREEMETA state */ | |
1913 | error = foreach_leaf(dip, leaf_dealloc, NULL); | |
1914 | if (error) | |
1915 | return error; | |
1916 | ||
1917 | /* Make this a regular file in case we crash. | |
1918 | (We don't want to free these blocks a second time.) */ | |
1919 | ||
1920 | error = gfs2_trans_begin(sdp, RES_DINODE, 0); | |
1921 | if (error) | |
1922 | return error; | |
1923 | ||
1924 | error = gfs2_meta_inode_buffer(dip, &bh); | |
1925 | if (!error) { | |
d4e9c4c3 | 1926 | gfs2_trans_add_bh(dip->i_gl, bh, 1); |
568f4c96 SW |
1927 | ((struct gfs2_dinode *)bh->b_data)->di_mode = |
1928 | cpu_to_be32(S_IFREG); | |
b3b94faa DT |
1929 | brelse(bh); |
1930 | } | |
1931 | ||
1932 | gfs2_trans_end(sdp); | |
1933 | ||
1934 | return error; | |
1935 | } | |
1936 | ||
1937 | /** | |
1938 | * gfs2_diradd_alloc_required - find if adding entry will require an allocation | |
1939 | * @ip: the file being written to | |
1940 | * @filname: the filename that's going to be added | |
b3b94faa | 1941 | * |
c752666c | 1942 | * Returns: 1 if alloc required, 0 if not, -ve on error |
b3b94faa DT |
1943 | */ |
1944 | ||
4d8012b6 | 1945 | int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name) |
b3b94faa | 1946 | { |
c752666c SW |
1947 | struct gfs2_dirent *dent; |
1948 | struct buffer_head *bh; | |
b3b94faa | 1949 | |
c752666c | 1950 | dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh); |
ed386507 | 1951 | if (!dent) { |
c752666c | 1952 | return 1; |
ed386507 | 1953 | } |
c752666c SW |
1954 | if (IS_ERR(dent)) |
1955 | return PTR_ERR(dent); | |
1956 | brelse(bh); | |
1957 | return 0; | |
b3b94faa DT |
1958 | } |
1959 |