]>
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 | * | |
383f01fb | 39 | * dip->i_diskflags & GFS2_DIF_EXHASH is true |
61e085a8 SW |
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 | 55 | |
d77d1b58 JP |
56 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
57 | ||
b3b94faa DT |
58 | #include <linux/slab.h> |
59 | #include <linux/spinlock.h> | |
b3b94faa DT |
60 | #include <linux/buffer_head.h> |
61 | #include <linux/sort.h> | |
5c676f6d | 62 | #include <linux/gfs2_ondisk.h> |
71b86f56 | 63 | #include <linux/crc32.h> |
fe1bdedc | 64 | #include <linux/vmalloc.h> |
2f8b5444 | 65 | #include <linux/bio.h> |
b3b94faa DT |
66 | |
67 | #include "gfs2.h" | |
5c676f6d | 68 | #include "incore.h" |
b3b94faa DT |
69 | #include "dir.h" |
70 | #include "glock.h" | |
71 | #include "inode.h" | |
b3b94faa DT |
72 | #include "meta_io.h" |
73 | #include "quota.h" | |
74 | #include "rgrp.h" | |
75 | #include "trans.h" | |
e13940ba | 76 | #include "bmap.h" |
5c676f6d | 77 | #include "util.h" |
b3b94faa DT |
78 | |
79 | #define IS_LEAF 1 /* Hashed (leaf) directory */ | |
80 | #define IS_DINODE 2 /* Linear (stuffed dinode block) directory */ | |
81 | ||
dfe4d34b BP |
82 | #define MAX_RA_BLOCKS 32 /* max read-ahead blocks */ |
83 | ||
cd915493 SW |
84 | #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1) |
85 | #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1)) | |
471f3db2 BM |
86 | #define GFS2_HASH_INDEX_MASK 0xffffc000 |
87 | #define GFS2_USE_HASH_FLAG 0x2000 | |
b3b94faa | 88 | |
8d123585 SW |
89 | struct qstr gfs2_qdot __read_mostly; |
90 | struct qstr gfs2_qdotdot __read_mostly; | |
91 | ||
2bdbc5d7 SW |
92 | typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent, |
93 | const struct qstr *name, void *opaque); | |
b3b94faa | 94 | |
cd915493 | 95 | int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block, |
61e085a8 | 96 | struct buffer_head **bhp) |
e13940ba SW |
97 | { |
98 | struct buffer_head *bh; | |
e13940ba | 99 | |
61e085a8 | 100 | bh = gfs2_meta_new(ip->i_gl, block); |
350a9b0a | 101 | gfs2_trans_add_meta(ip->i_gl, bh); |
61e085a8 SW |
102 | gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD); |
103 | gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header)); | |
e13940ba SW |
104 | *bhp = bh; |
105 | return 0; | |
106 | } | |
107 | ||
cd915493 | 108 | static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block, |
61e085a8 SW |
109 | struct buffer_head **bhp) |
110 | { | |
111 | struct buffer_head *bh; | |
112 | int error; | |
e13940ba | 113 | |
c8d57703 | 114 | error = gfs2_meta_read(ip->i_gl, block, DIO_WAIT, 0, &bh); |
61e085a8 SW |
115 | if (error) |
116 | return error; | |
feaa7bba | 117 | if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) { |
61e085a8 SW |
118 | brelse(bh); |
119 | return -EIO; | |
120 | } | |
121 | *bhp = bh; | |
122 | return 0; | |
123 | } | |
e13940ba SW |
124 | |
125 | static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf, | |
126 | unsigned int offset, unsigned int size) | |
e13940ba SW |
127 | { |
128 | struct buffer_head *dibh; | |
129 | int error; | |
130 | ||
131 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
132 | if (error) | |
133 | return error; | |
134 | ||
350a9b0a | 135 | gfs2_trans_add_meta(ip->i_gl, dibh); |
c752666c | 136 | memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size); |
a2e0f799 SW |
137 | if (ip->i_inode.i_size < offset + size) |
138 | i_size_write(&ip->i_inode, offset + size); | |
078cd827 | 139 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
539e5d6b | 140 | gfs2_dinode_out(ip, dibh->b_data); |
e13940ba SW |
141 | |
142 | brelse(dibh); | |
143 | ||
144 | return size; | |
145 | } | |
146 | ||
147 | ||
148 | ||
149 | /** | |
150 | * gfs2_dir_write_data - Write directory information to the inode | |
151 | * @ip: The GFS2 inode | |
152 | * @buf: The buffer containing information to be written | |
153 | * @offset: The file offset to start writing at | |
154 | * @size: The amount of data to write | |
155 | * | |
156 | * Returns: The number of bytes correctly written or error code | |
157 | */ | |
158 | static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf, | |
cd915493 | 159 | u64 offset, unsigned int size) |
e13940ba | 160 | { |
feaa7bba | 161 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
e13940ba | 162 | struct buffer_head *dibh; |
cd915493 SW |
163 | u64 lblock, dblock; |
164 | u32 extlen = 0; | |
e13940ba SW |
165 | unsigned int o; |
166 | int copied = 0; | |
167 | int error = 0; | |
9b8c81d1 | 168 | int new = 0; |
e13940ba SW |
169 | |
170 | if (!size) | |
171 | return 0; | |
172 | ||
173 | if (gfs2_is_stuffed(ip) && | |
174 | offset + size <= sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) | |
568f4c96 SW |
175 | return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset, |
176 | size); | |
e13940ba SW |
177 | |
178 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
179 | return -EINVAL; | |
180 | ||
181 | if (gfs2_is_stuffed(ip)) { | |
f25ef0c1 | 182 | error = gfs2_unstuff_dinode(ip, NULL); |
e13940ba | 183 | if (error) |
c752666c | 184 | return error; |
e13940ba SW |
185 | } |
186 | ||
187 | lblock = offset; | |
188 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); | |
189 | ||
190 | while (copied < size) { | |
191 | unsigned int amount; | |
192 | struct buffer_head *bh; | |
e13940ba SW |
193 | |
194 | amount = size - copied; | |
195 | if (amount > sdp->sd_sb.sb_bsize - o) | |
196 | amount = sdp->sd_sb.sb_bsize - o; | |
197 | ||
198 | if (!extlen) { | |
199 | new = 1; | |
feaa7bba | 200 | error = gfs2_extent_map(&ip->i_inode, lblock, &new, |
fd88de56 | 201 | &dblock, &extlen); |
e13940ba SW |
202 | if (error) |
203 | goto fail; | |
204 | error = -EIO; | |
205 | if (gfs2_assert_withdraw(sdp, dblock)) | |
206 | goto fail; | |
207 | } | |
208 | ||
61e085a8 SW |
209 | if (amount == sdp->sd_jbsize || new) |
210 | error = gfs2_dir_get_new_buffer(ip, dblock, &bh); | |
211 | else | |
212 | error = gfs2_dir_get_existing_buffer(ip, dblock, &bh); | |
213 | ||
e13940ba SW |
214 | if (error) |
215 | goto fail; | |
216 | ||
350a9b0a | 217 | gfs2_trans_add_meta(ip->i_gl, bh); |
e13940ba SW |
218 | memcpy(bh->b_data + o, buf, amount); |
219 | brelse(bh); | |
e13940ba | 220 | |
899bb264 | 221 | buf += amount; |
e13940ba SW |
222 | copied += amount; |
223 | lblock++; | |
224 | dblock++; | |
225 | extlen--; | |
226 | ||
227 | o = sizeof(struct gfs2_meta_header); | |
228 | } | |
229 | ||
230 | out: | |
231 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
232 | if (error) | |
233 | return error; | |
234 | ||
a2e0f799 SW |
235 | if (ip->i_inode.i_size < offset + copied) |
236 | i_size_write(&ip->i_inode, offset + copied); | |
078cd827 | 237 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = current_time(&ip->i_inode); |
e13940ba | 238 | |
350a9b0a | 239 | gfs2_trans_add_meta(ip->i_gl, dibh); |
539e5d6b | 240 | gfs2_dinode_out(ip, dibh->b_data); |
e13940ba SW |
241 | brelse(dibh); |
242 | ||
243 | return copied; | |
244 | fail: | |
245 | if (copied) | |
246 | goto out; | |
247 | return error; | |
248 | } | |
249 | ||
4c28d338 SW |
250 | static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf, |
251 | unsigned int size) | |
e13940ba SW |
252 | { |
253 | struct buffer_head *dibh; | |
254 | int error; | |
255 | ||
256 | error = gfs2_meta_inode_buffer(ip, &dibh); | |
257 | if (!error) { | |
4c28d338 | 258 | memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size); |
e13940ba SW |
259 | brelse(dibh); |
260 | } | |
261 | ||
262 | return (error) ? error : size; | |
263 | } | |
264 | ||
265 | ||
266 | /** | |
267 | * gfs2_dir_read_data - Read a data from a directory inode | |
268 | * @ip: The GFS2 Inode | |
269 | * @buf: The buffer to place result into | |
e13940ba SW |
270 | * @size: Amount of data to transfer |
271 | * | |
272 | * Returns: The amount of data actually copied or the error | |
273 | */ | |
4c28d338 SW |
274 | static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf, |
275 | unsigned int size) | |
e13940ba | 276 | { |
feaa7bba | 277 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode); |
cd915493 SW |
278 | u64 lblock, dblock; |
279 | u32 extlen = 0; | |
e13940ba SW |
280 | unsigned int o; |
281 | int copied = 0; | |
282 | int error = 0; | |
e13940ba SW |
283 | |
284 | if (gfs2_is_stuffed(ip)) | |
4c28d338 | 285 | return gfs2_dir_read_stuffed(ip, buf, size); |
e13940ba SW |
286 | |
287 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip))) | |
288 | return -EINVAL; | |
289 | ||
4c28d338 | 290 | lblock = 0; |
e13940ba SW |
291 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header); |
292 | ||
293 | while (copied < size) { | |
294 | unsigned int amount; | |
295 | struct buffer_head *bh; | |
296 | int new; | |
297 | ||
298 | amount = size - copied; | |
299 | if (amount > sdp->sd_sb.sb_bsize - o) | |
300 | amount = sdp->sd_sb.sb_bsize - o; | |
301 | ||
302 | if (!extlen) { | |
303 | new = 0; | |
feaa7bba | 304 | error = gfs2_extent_map(&ip->i_inode, lblock, &new, |
fd88de56 | 305 | &dblock, &extlen); |
7276b3b0 | 306 | if (error || !dblock) |
e13940ba | 307 | goto fail; |
7276b3b0 | 308 | BUG_ON(extlen < 1); |
7276b3b0 | 309 | bh = gfs2_meta_ra(ip->i_gl, dblock, extlen); |
b7d8ac3e | 310 | } else { |
c8d57703 | 311 | error = gfs2_meta_read(ip->i_gl, dblock, DIO_WAIT, 0, &bh); |
e13940ba SW |
312 | if (error) |
313 | goto fail; | |
7276b3b0 SW |
314 | } |
315 | error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD); | |
316 | if (error) { | |
317 | brelse(bh); | |
318 | goto fail; | |
319 | } | |
320 | dblock++; | |
321 | extlen--; | |
e13940ba SW |
322 | memcpy(buf, bh->b_data + o, amount); |
323 | brelse(bh); | |
4c28d338 | 324 | buf += (amount/sizeof(__be64)); |
e13940ba SW |
325 | copied += amount; |
326 | lblock++; | |
e13940ba SW |
327 | o = sizeof(struct gfs2_meta_header); |
328 | } | |
329 | ||
330 | return copied; | |
331 | fail: | |
332 | return (copied) ? copied : error; | |
333 | } | |
334 | ||
17d539f0 SW |
335 | /** |
336 | * gfs2_dir_get_hash_table - Get pointer to the dir hash table | |
337 | * @ip: The inode in question | |
338 | * | |
339 | * Returns: The hash table or an error | |
340 | */ | |
341 | ||
342 | static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip) | |
343 | { | |
344 | struct inode *inode = &ip->i_inode; | |
345 | int ret; | |
346 | u32 hsize; | |
347 | __be64 *hc; | |
348 | ||
349 | BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH)); | |
350 | ||
351 | hc = ip->i_hash_cache; | |
352 | if (hc) | |
353 | return hc; | |
354 | ||
47a9a527 | 355 | hsize = BIT(ip->i_depth); |
17d539f0 SW |
356 | hsize *= sizeof(__be64); |
357 | if (hsize != i_size_read(&ip->i_inode)) { | |
358 | gfs2_consist_inode(ip); | |
359 | return ERR_PTR(-EIO); | |
360 | } | |
361 | ||
e8830d88 BP |
362 | hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN); |
363 | if (hc == NULL) | |
364 | hc = __vmalloc(hsize, GFP_NOFS, PAGE_KERNEL); | |
365 | ||
17d539f0 SW |
366 | if (hc == NULL) |
367 | return ERR_PTR(-ENOMEM); | |
368 | ||
4c28d338 | 369 | ret = gfs2_dir_read_data(ip, hc, hsize); |
17d539f0 | 370 | if (ret < 0) { |
3cdcf63e | 371 | kvfree(hc); |
17d539f0 SW |
372 | return ERR_PTR(ret); |
373 | } | |
374 | ||
375 | spin_lock(&inode->i_lock); | |
9265f1d0 | 376 | if (likely(!ip->i_hash_cache)) { |
17d539f0 | 377 | ip->i_hash_cache = hc; |
9265f1d0 AV |
378 | hc = NULL; |
379 | } | |
17d539f0 | 380 | spin_unlock(&inode->i_lock); |
9265f1d0 | 381 | kvfree(hc); |
17d539f0 SW |
382 | |
383 | return ip->i_hash_cache; | |
384 | } | |
385 | ||
386 | /** | |
387 | * gfs2_dir_hash_inval - Invalidate dir hash | |
388 | * @ip: The directory inode | |
389 | * | |
390 | * Must be called with an exclusive glock, or during glock invalidation. | |
391 | */ | |
392 | void gfs2_dir_hash_inval(struct gfs2_inode *ip) | |
393 | { | |
c36b97e9 BP |
394 | __be64 *hc; |
395 | ||
396 | spin_lock(&ip->i_inode.i_lock); | |
397 | hc = ip->i_hash_cache; | |
17d539f0 | 398 | ip->i_hash_cache = NULL; |
c36b97e9 BP |
399 | spin_unlock(&ip->i_inode.i_lock); |
400 | ||
3cdcf63e | 401 | kvfree(hc); |
17d539f0 SW |
402 | } |
403 | ||
5e7d65cd SW |
404 | static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent) |
405 | { | |
406 | return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0; | |
407 | } | |
408 | ||
c752666c SW |
409 | static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent, |
410 | const struct qstr *name, int ret) | |
411 | { | |
5e7d65cd | 412 | if (!gfs2_dirent_sentinel(dent) && |
c752666c SW |
413 | be32_to_cpu(dent->de_hash) == name->hash && |
414 | be16_to_cpu(dent->de_name_len) == name->len && | |
2bdbc5d7 | 415 | memcmp(dent+1, name->name, name->len) == 0) |
c752666c SW |
416 | return ret; |
417 | return 0; | |
418 | } | |
419 | ||
420 | static int gfs2_dirent_find(const struct gfs2_dirent *dent, | |
71b86f56 SW |
421 | const struct qstr *name, |
422 | void *opaque) | |
c752666c SW |
423 | { |
424 | return __gfs2_dirent_find(dent, name, 1); | |
425 | } | |
426 | ||
427 | static int gfs2_dirent_prev(const struct gfs2_dirent *dent, | |
71b86f56 SW |
428 | const struct qstr *name, |
429 | void *opaque) | |
c752666c SW |
430 | { |
431 | return __gfs2_dirent_find(dent, name, 2); | |
432 | } | |
433 | ||
434 | /* | |
435 | * name->name holds ptr to start of block. | |
436 | * name->len holds size of block. | |
b3b94faa | 437 | */ |
c752666c | 438 | static int gfs2_dirent_last(const struct gfs2_dirent *dent, |
71b86f56 SW |
439 | const struct qstr *name, |
440 | void *opaque) | |
c752666c SW |
441 | { |
442 | const char *start = name->name; | |
443 | const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len); | |
444 | if (name->len == (end - start)) | |
445 | return 1; | |
446 | return 0; | |
447 | } | |
b3b94faa | 448 | |
34017472 BM |
449 | /* Look for the dirent that contains the offset specified in data. Once we |
450 | * find that dirent, there must be space available there for the new dirent */ | |
451 | static int gfs2_dirent_find_offset(const struct gfs2_dirent *dent, | |
452 | const struct qstr *name, | |
453 | void *ptr) | |
454 | { | |
455 | unsigned required = GFS2_DIRENT_SIZE(name->len); | |
456 | unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); | |
457 | unsigned totlen = be16_to_cpu(dent->de_rec_len); | |
458 | ||
459 | if (ptr < (void *)dent || ptr >= (void *)dent + totlen) | |
460 | return 0; | |
461 | if (gfs2_dirent_sentinel(dent)) | |
462 | actual = 0; | |
463 | if (ptr < (void *)dent + actual) | |
464 | return -1; | |
465 | if ((void *)dent + totlen >= ptr + required) | |
466 | return 1; | |
467 | return -1; | |
468 | } | |
469 | ||
c752666c | 470 | static int gfs2_dirent_find_space(const struct gfs2_dirent *dent, |
71b86f56 SW |
471 | const struct qstr *name, |
472 | void *opaque) | |
b3b94faa | 473 | { |
c752666c SW |
474 | unsigned required = GFS2_DIRENT_SIZE(name->len); |
475 | unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); | |
476 | unsigned totlen = be16_to_cpu(dent->de_rec_len); | |
477 | ||
5e7d65cd | 478 | if (gfs2_dirent_sentinel(dent)) |
728a756b | 479 | actual = 0; |
c5392124 | 480 | if (totlen - actual >= required) |
c752666c SW |
481 | return 1; |
482 | return 0; | |
483 | } | |
484 | ||
71b86f56 SW |
485 | struct dirent_gather { |
486 | const struct gfs2_dirent **pdent; | |
487 | unsigned offset; | |
488 | }; | |
489 | ||
490 | static int gfs2_dirent_gather(const struct gfs2_dirent *dent, | |
491 | const struct qstr *name, | |
492 | void *opaque) | |
493 | { | |
494 | struct dirent_gather *g = opaque; | |
5e7d65cd | 495 | if (!gfs2_dirent_sentinel(dent)) { |
71b86f56 SW |
496 | g->pdent[g->offset++] = dent; |
497 | } | |
498 | return 0; | |
499 | } | |
500 | ||
c752666c SW |
501 | /* |
502 | * Other possible things to check: | |
503 | * - Inode located within filesystem size (and on valid block) | |
504 | * - Valid directory entry type | |
505 | * Not sure how heavy-weight we want to make this... could also check | |
506 | * hash is correct for example, but that would take a lot of extra time. | |
507 | * For now the most important thing is to check that the various sizes | |
508 | * are correct. | |
509 | */ | |
510 | static int gfs2_check_dirent(struct gfs2_dirent *dent, unsigned int offset, | |
511 | unsigned int size, unsigned int len, int first) | |
512 | { | |
513 | const char *msg = "gfs2_dirent too small"; | |
514 | if (unlikely(size < sizeof(struct gfs2_dirent))) | |
515 | goto error; | |
516 | msg = "gfs2_dirent misaligned"; | |
517 | if (unlikely(offset & 0x7)) | |
518 | goto error; | |
519 | msg = "gfs2_dirent points beyond end of block"; | |
520 | if (unlikely(offset + size > len)) | |
521 | goto error; | |
522 | msg = "zero inode number"; | |
5e7d65cd | 523 | if (unlikely(!first && gfs2_dirent_sentinel(dent))) |
c752666c SW |
524 | goto error; |
525 | msg = "name length is greater than space in dirent"; | |
5e7d65cd | 526 | if (!gfs2_dirent_sentinel(dent) && |
c752666c SW |
527 | unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) > |
528 | size)) | |
529 | goto error; | |
530 | return 0; | |
531 | error: | |
d77d1b58 JP |
532 | pr_warn("%s: %s (%s)\n", |
533 | __func__, msg, first ? "first in block" : "not first in block"); | |
c752666c | 534 | return -EIO; |
b3b94faa DT |
535 | } |
536 | ||
71b86f56 | 537 | static int gfs2_dirent_offset(const void *buf) |
c752666c | 538 | { |
71b86f56 SW |
539 | const struct gfs2_meta_header *h = buf; |
540 | int offset; | |
c752666c SW |
541 | |
542 | BUG_ON(buf == NULL); | |
c752666c | 543 | |
e3167ded | 544 | switch(be32_to_cpu(h->mh_type)) { |
c752666c SW |
545 | case GFS2_METATYPE_LF: |
546 | offset = sizeof(struct gfs2_leaf); | |
547 | break; | |
548 | case GFS2_METATYPE_DI: | |
549 | offset = sizeof(struct gfs2_dinode); | |
550 | break; | |
551 | default: | |
552 | goto wrong_type; | |
553 | } | |
71b86f56 SW |
554 | return offset; |
555 | wrong_type: | |
d77d1b58 | 556 | pr_warn("%s: wrong block type %u\n", __func__, be32_to_cpu(h->mh_type)); |
71b86f56 SW |
557 | return -1; |
558 | } | |
559 | ||
2bdbc5d7 | 560 | static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf, |
71b86f56 SW |
561 | unsigned int len, gfs2_dscan_t scan, |
562 | const struct qstr *name, | |
563 | void *opaque) | |
564 | { | |
565 | struct gfs2_dirent *dent, *prev; | |
566 | unsigned offset; | |
567 | unsigned size; | |
568 | int ret = 0; | |
c752666c | 569 | |
71b86f56 SW |
570 | ret = gfs2_dirent_offset(buf); |
571 | if (ret < 0) | |
572 | goto consist_inode; | |
573 | ||
574 | offset = ret; | |
c752666c | 575 | prev = NULL; |
2bdbc5d7 | 576 | dent = buf + offset; |
c752666c SW |
577 | size = be16_to_cpu(dent->de_rec_len); |
578 | if (gfs2_check_dirent(dent, offset, size, len, 1)) | |
579 | goto consist_inode; | |
580 | do { | |
71b86f56 | 581 | ret = scan(dent, name, opaque); |
c752666c SW |
582 | if (ret) |
583 | break; | |
584 | offset += size; | |
585 | if (offset == len) | |
586 | break; | |
587 | prev = dent; | |
2bdbc5d7 | 588 | dent = buf + offset; |
c752666c SW |
589 | size = be16_to_cpu(dent->de_rec_len); |
590 | if (gfs2_check_dirent(dent, offset, size, len, 0)) | |
591 | goto consist_inode; | |
592 | } while(1); | |
593 | ||
594 | switch(ret) { | |
595 | case 0: | |
596 | return NULL; | |
597 | case 1: | |
598 | return dent; | |
599 | case 2: | |
600 | return prev ? prev : dent; | |
601 | default: | |
602 | BUG_ON(ret > 0); | |
603 | return ERR_PTR(ret); | |
604 | } | |
605 | ||
c752666c | 606 | consist_inode: |
feaa7bba | 607 | gfs2_consist_inode(GFS2_I(inode)); |
c752666c SW |
608 | return ERR_PTR(-EIO); |
609 | } | |
610 | ||
2bdbc5d7 SW |
611 | static int dirent_check_reclen(struct gfs2_inode *dip, |
612 | const struct gfs2_dirent *d, const void *end_p) | |
613 | { | |
614 | const void *ptr = d; | |
615 | u16 rec_len = be16_to_cpu(d->de_rec_len); | |
616 | ||
617 | if (unlikely(rec_len < sizeof(struct gfs2_dirent))) | |
618 | goto broken; | |
619 | ptr += rec_len; | |
620 | if (ptr < end_p) | |
621 | return rec_len; | |
622 | if (ptr == end_p) | |
623 | return -ENOENT; | |
624 | broken: | |
625 | gfs2_consist_inode(dip); | |
626 | return -EIO; | |
627 | } | |
628 | ||
b3b94faa DT |
629 | /** |
630 | * dirent_next - Next dirent | |
631 | * @dip: the directory | |
632 | * @bh: The buffer | |
633 | * @dent: Pointer to list of dirents | |
634 | * | |
635 | * Returns: 0 on success, error code otherwise | |
636 | */ | |
637 | ||
638 | static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh, | |
639 | struct gfs2_dirent **dent) | |
640 | { | |
2bdbc5d7 SW |
641 | struct gfs2_dirent *cur = *dent, *tmp; |
642 | char *bh_end = bh->b_data + bh->b_size; | |
643 | int ret; | |
b3b94faa | 644 | |
2bdbc5d7 SW |
645 | ret = dirent_check_reclen(dip, cur, bh_end); |
646 | if (ret < 0) | |
647 | return ret; | |
4dd651ad | 648 | |
2bdbc5d7 SW |
649 | tmp = (void *)cur + ret; |
650 | ret = dirent_check_reclen(dip, tmp, bh_end); | |
651 | if (ret == -EIO) | |
652 | return ret; | |
4dd651ad | 653 | |
b3b94faa | 654 | /* Only the first dent could ever have de_inum.no_addr == 0 */ |
5e7d65cd | 655 | if (gfs2_dirent_sentinel(tmp)) { |
b3b94faa DT |
656 | gfs2_consist_inode(dip); |
657 | return -EIO; | |
658 | } | |
659 | ||
660 | *dent = tmp; | |
b3b94faa DT |
661 | return 0; |
662 | } | |
663 | ||
664 | /** | |
665 | * dirent_del - Delete a dirent | |
666 | * @dip: The GFS2 inode | |
667 | * @bh: The buffer | |
668 | * @prev: The previous dirent | |
669 | * @cur: The current dirent | |
670 | * | |
671 | */ | |
672 | ||
673 | static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh, | |
674 | struct gfs2_dirent *prev, struct gfs2_dirent *cur) | |
675 | { | |
cd915493 | 676 | u16 cur_rec_len, prev_rec_len; |
b3b94faa | 677 | |
5e7d65cd | 678 | if (gfs2_dirent_sentinel(cur)) { |
b3b94faa DT |
679 | gfs2_consist_inode(dip); |
680 | return; | |
681 | } | |
682 | ||
350a9b0a | 683 | gfs2_trans_add_meta(dip->i_gl, bh); |
b3b94faa DT |
684 | |
685 | /* If there is no prev entry, this is the first entry in the block. | |
686 | The de_rec_len is already as big as it needs to be. Just zero | |
687 | out the inode number and return. */ | |
688 | ||
689 | if (!prev) { | |
5e7d65cd SW |
690 | cur->de_inum.no_addr = 0; |
691 | cur->de_inum.no_formal_ino = 0; | |
b3b94faa DT |
692 | return; |
693 | } | |
694 | ||
695 | /* Combine this dentry with the previous one. */ | |
696 | ||
fc69d0d3 SW |
697 | prev_rec_len = be16_to_cpu(prev->de_rec_len); |
698 | cur_rec_len = be16_to_cpu(cur->de_rec_len); | |
b3b94faa DT |
699 | |
700 | if ((char *)prev + prev_rec_len != (char *)cur) | |
701 | gfs2_consist_inode(dip); | |
702 | if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size) | |
703 | gfs2_consist_inode(dip); | |
704 | ||
705 | prev_rec_len += cur_rec_len; | |
fc69d0d3 | 706 | prev->de_rec_len = cpu_to_be16(prev_rec_len); |
b3b94faa DT |
707 | } |
708 | ||
34017472 BM |
709 | |
710 | static struct gfs2_dirent *do_init_dirent(struct inode *inode, | |
711 | struct gfs2_dirent *dent, | |
712 | const struct qstr *name, | |
713 | struct buffer_head *bh, | |
714 | unsigned offset) | |
715 | { | |
716 | struct gfs2_inode *ip = GFS2_I(inode); | |
717 | struct gfs2_dirent *ndent; | |
718 | unsigned totlen; | |
719 | ||
720 | totlen = be16_to_cpu(dent->de_rec_len); | |
721 | BUG_ON(offset + name->len > totlen); | |
722 | gfs2_trans_add_meta(ip->i_gl, bh); | |
723 | ndent = (struct gfs2_dirent *)((char *)dent + offset); | |
724 | dent->de_rec_len = cpu_to_be16(offset); | |
725 | gfs2_qstr2dirent(name, totlen - offset, ndent); | |
726 | return ndent; | |
727 | } | |
728 | ||
729 | ||
c752666c SW |
730 | /* |
731 | * Takes a dent from which to grab space as an argument. Returns the | |
732 | * newly created dent. | |
b3b94faa | 733 | */ |
08bc2dbc AB |
734 | static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode, |
735 | struct gfs2_dirent *dent, | |
736 | const struct qstr *name, | |
737 | struct buffer_head *bh) | |
b3b94faa | 738 | { |
34017472 | 739 | unsigned offset = 0; |
c752666c | 740 | |
5e7d65cd | 741 | if (!gfs2_dirent_sentinel(dent)) |
c752666c | 742 | offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len)); |
34017472 | 743 | return do_init_dirent(inode, dent, name, bh, offset); |
b3b94faa DT |
744 | } |
745 | ||
34017472 BM |
746 | static struct gfs2_dirent *gfs2_dirent_split_alloc(struct inode *inode, |
747 | struct buffer_head *bh, | |
748 | const struct qstr *name, | |
749 | void *ptr) | |
b3b94faa DT |
750 | { |
751 | struct gfs2_dirent *dent; | |
907b9bce | 752 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, |
34017472 | 753 | gfs2_dirent_find_offset, name, ptr); |
c752666c SW |
754 | if (!dent || IS_ERR(dent)) |
755 | return dent; | |
34017472 BM |
756 | return do_init_dirent(inode, dent, name, bh, |
757 | (unsigned)(ptr - (void *)dent)); | |
b3b94faa DT |
758 | } |
759 | ||
cd915493 | 760 | static int get_leaf(struct gfs2_inode *dip, u64 leaf_no, |
b3b94faa DT |
761 | struct buffer_head **bhp) |
762 | { | |
763 | int error; | |
764 | ||
c8d57703 | 765 | error = gfs2_meta_read(dip->i_gl, leaf_no, DIO_WAIT, 0, bhp); |
feaa7bba | 766 | if (!error && gfs2_metatype_check(GFS2_SB(&dip->i_inode), *bhp, GFS2_METATYPE_LF)) { |
d77d1b58 | 767 | /* pr_info("block num=%llu\n", leaf_no); */ |
b3b94faa | 768 | error = -EIO; |
feaa7bba | 769 | } |
b3b94faa DT |
770 | |
771 | return error; | |
772 | } | |
773 | ||
774 | /** | |
775 | * get_leaf_nr - Get a leaf number associated with the index | |
776 | * @dip: The GFS2 inode | |
777 | * @index: | |
778 | * @leaf_out: | |
779 | * | |
780 | * Returns: 0 on success, error code otherwise | |
781 | */ | |
782 | ||
cd915493 SW |
783 | static int get_leaf_nr(struct gfs2_inode *dip, u32 index, |
784 | u64 *leaf_out) | |
b3b94faa | 785 | { |
17d539f0 | 786 | __be64 *hash; |
287980e4 | 787 | int error; |
b3b94faa | 788 | |
17d539f0 | 789 | hash = gfs2_dir_get_hash_table(dip); |
287980e4 AB |
790 | error = PTR_ERR_OR_ZERO(hash); |
791 | ||
792 | if (!error) | |
793 | *leaf_out = be64_to_cpu(*(hash + index)); | |
794 | ||
795 | return error; | |
b3b94faa DT |
796 | } |
797 | ||
cd915493 | 798 | static int get_first_leaf(struct gfs2_inode *dip, u32 index, |
b3b94faa DT |
799 | struct buffer_head **bh_out) |
800 | { | |
cd915493 | 801 | u64 leaf_no; |
b3b94faa DT |
802 | int error; |
803 | ||
804 | error = get_leaf_nr(dip, index, &leaf_no); | |
287980e4 | 805 | if (!error) |
b3b94faa DT |
806 | error = get_leaf(dip, leaf_no, bh_out); |
807 | ||
808 | return error; | |
809 | } | |
810 | ||
c752666c SW |
811 | static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode, |
812 | const struct qstr *name, | |
813 | gfs2_dscan_t scan, | |
814 | struct buffer_head **pbh) | |
b3b94faa | 815 | { |
c752666c SW |
816 | struct buffer_head *bh; |
817 | struct gfs2_dirent *dent; | |
feaa7bba | 818 | struct gfs2_inode *ip = GFS2_I(inode); |
b3b94faa DT |
819 | int error; |
820 | ||
383f01fb | 821 | if (ip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c | 822 | struct gfs2_leaf *leaf; |
47a9a527 FF |
823 | unsigned int hsize = BIT(ip->i_depth); |
824 | unsigned int index; | |
c752666c | 825 | u64 ln; |
a2e0f799 | 826 | if (hsize * sizeof(u64) != i_size_read(inode)) { |
c752666c SW |
827 | gfs2_consist_inode(ip); |
828 | return ERR_PTR(-EIO); | |
829 | } | |
907b9bce | 830 | |
9a004508 | 831 | index = name->hash >> (32 - ip->i_depth); |
c752666c SW |
832 | error = get_first_leaf(ip, index, &bh); |
833 | if (error) | |
834 | return ERR_PTR(error); | |
835 | do { | |
836 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, | |
71b86f56 | 837 | scan, name, NULL); |
c752666c SW |
838 | if (dent) |
839 | goto got_dent; | |
840 | leaf = (struct gfs2_leaf *)bh->b_data; | |
841 | ln = be64_to_cpu(leaf->lf_next); | |
f4154ea0 | 842 | brelse(bh); |
c752666c SW |
843 | if (!ln) |
844 | break; | |
907b9bce | 845 | |
c752666c SW |
846 | error = get_leaf(ip, ln, &bh); |
847 | } while(!error); | |
b3b94faa | 848 | |
c752666c | 849 | return error ? ERR_PTR(error) : NULL; |
b3b94faa | 850 | } |
b3b94faa | 851 | |
907b9bce | 852 | |
c752666c SW |
853 | error = gfs2_meta_inode_buffer(ip, &bh); |
854 | if (error) | |
855 | return ERR_PTR(error); | |
71b86f56 | 856 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL); |
c752666c | 857 | got_dent: |
f4154ea0 | 858 | if (unlikely(dent == NULL || IS_ERR(dent))) { |
ed386507 SW |
859 | brelse(bh); |
860 | bh = NULL; | |
861 | } | |
c752666c SW |
862 | *pbh = bh; |
863 | return dent; | |
864 | } | |
b3b94faa | 865 | |
c752666c SW |
866 | static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth) |
867 | { | |
feaa7bba | 868 | struct gfs2_inode *ip = GFS2_I(inode); |
b45e41d7 | 869 | unsigned int n = 1; |
09010978 SW |
870 | u64 bn; |
871 | int error; | |
872 | struct buffer_head *bh; | |
c752666c SW |
873 | struct gfs2_leaf *leaf; |
874 | struct gfs2_dirent *dent; | |
26fe5750 | 875 | struct qstr name = { .name = "" }; |
078cd827 | 876 | struct timespec tv = current_time(inode); |
09010978 | 877 | |
6e87ed0f | 878 | error = gfs2_alloc_blocks(ip, &bn, &n, 0, NULL); |
09010978 SW |
879 | if (error) |
880 | return NULL; | |
881 | bh = gfs2_meta_new(ip->i_gl, bn); | |
c752666c SW |
882 | if (!bh) |
883 | return NULL; | |
09010978 | 884 | |
5731be53 | 885 | gfs2_trans_add_unrevoke(GFS2_SB(inode), bn, 1); |
350a9b0a | 886 | gfs2_trans_add_meta(ip->i_gl, bh); |
c752666c SW |
887 | gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF); |
888 | leaf = (struct gfs2_leaf *)bh->b_data; | |
889 | leaf->lf_depth = cpu_to_be16(depth); | |
2bdbc5d7 | 890 | leaf->lf_entries = 0; |
a2d7d021 | 891 | leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE); |
2bdbc5d7 | 892 | leaf->lf_next = 0; |
01bcb0de SW |
893 | leaf->lf_inode = cpu_to_be64(ip->i_no_addr); |
894 | leaf->lf_dist = cpu_to_be32(1); | |
895 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); | |
896 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
897 | memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2)); | |
c752666c | 898 | dent = (struct gfs2_dirent *)(leaf+1); |
71b86f56 | 899 | gfs2_qstr2dirent(&name, bh->b_size - sizeof(struct gfs2_leaf), dent); |
c752666c SW |
900 | *pbh = bh; |
901 | return leaf; | |
b3b94faa DT |
902 | } |
903 | ||
904 | /** | |
905 | * dir_make_exhash - Convert a stuffed directory into an ExHash directory | |
906 | * @dip: The GFS2 inode | |
907 | * | |
908 | * Returns: 0 on success, error code otherwise | |
909 | */ | |
910 | ||
c752666c | 911 | static int dir_make_exhash(struct inode *inode) |
b3b94faa | 912 | { |
feaa7bba SW |
913 | struct gfs2_inode *dip = GFS2_I(inode); |
914 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
b3b94faa | 915 | struct gfs2_dirent *dent; |
c752666c | 916 | struct qstr args; |
b3b94faa DT |
917 | struct buffer_head *bh, *dibh; |
918 | struct gfs2_leaf *leaf; | |
919 | int y; | |
cd915493 | 920 | u32 x; |
b44b84d7 AV |
921 | __be64 *lp; |
922 | u64 bn; | |
b3b94faa DT |
923 | int error; |
924 | ||
925 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
926 | if (error) | |
927 | return error; | |
928 | ||
b3b94faa DT |
929 | /* Turn over a new leaf */ |
930 | ||
c752666c SW |
931 | leaf = new_leaf(inode, &bh, 0); |
932 | if (!leaf) | |
933 | return -ENOSPC; | |
934 | bn = bh->b_blocknr; | |
b3b94faa | 935 | |
47a9a527 | 936 | gfs2_assert(sdp, dip->i_entries < BIT(16)); |
ad6203f2 | 937 | leaf->lf_entries = cpu_to_be16(dip->i_entries); |
b3b94faa DT |
938 | |
939 | /* Copy dirents */ | |
940 | ||
941 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh, | |
942 | sizeof(struct gfs2_dinode)); | |
943 | ||
944 | /* Find last entry */ | |
945 | ||
946 | x = 0; | |
c752666c SW |
947 | args.len = bh->b_size - sizeof(struct gfs2_dinode) + |
948 | sizeof(struct gfs2_leaf); | |
949 | args.name = bh->b_data; | |
feaa7bba | 950 | dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size, |
71b86f56 | 951 | gfs2_dirent_last, &args, NULL); |
c752666c SW |
952 | if (!dent) { |
953 | brelse(bh); | |
954 | brelse(dibh); | |
955 | return -EIO; | |
956 | } | |
957 | if (IS_ERR(dent)) { | |
958 | brelse(bh); | |
959 | brelse(dibh); | |
960 | return PTR_ERR(dent); | |
b3b94faa | 961 | } |
b3b94faa DT |
962 | |
963 | /* Adjust the last dirent's record length | |
964 | (Remember that dent still points to the last entry.) */ | |
965 | ||
4dd651ad | 966 | dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) + |
b3b94faa | 967 | sizeof(struct gfs2_dinode) - |
4dd651ad | 968 | sizeof(struct gfs2_leaf)); |
b3b94faa DT |
969 | |
970 | brelse(bh); | |
971 | ||
972 | /* We're done with the new leaf block, now setup the new | |
973 | hash table. */ | |
974 | ||
350a9b0a | 975 | gfs2_trans_add_meta(dip->i_gl, dibh); |
b3b94faa DT |
976 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode)); |
977 | ||
b44b84d7 | 978 | lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode)); |
b3b94faa DT |
979 | |
980 | for (x = sdp->sd_hash_ptrs; x--; lp++) | |
981 | *lp = cpu_to_be64(bn); | |
982 | ||
a2e0f799 | 983 | i_size_write(inode, sdp->sd_sb.sb_bsize / 2); |
77658aad | 984 | gfs2_add_inode_blocks(&dip->i_inode, 1); |
383f01fb | 985 | dip->i_diskflags |= GFS2_DIF_EXHASH; |
b3b94faa DT |
986 | |
987 | for (x = sdp->sd_hash_ptrs, y = -1; x; x >>= 1, y++) ; | |
9a004508 | 988 | dip->i_depth = y; |
b3b94faa | 989 | |
539e5d6b | 990 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
991 | |
992 | brelse(dibh); | |
993 | ||
994 | return 0; | |
995 | } | |
996 | ||
997 | /** | |
998 | * dir_split_leaf - Split a leaf block into two | |
999 | * @dip: The GFS2 inode | |
1000 | * @index: | |
1001 | * @leaf_no: | |
1002 | * | |
1003 | * Returns: 0 on success, error code on failure | |
1004 | */ | |
1005 | ||
c752666c | 1006 | static int dir_split_leaf(struct inode *inode, const struct qstr *name) |
b3b94faa | 1007 | { |
feaa7bba | 1008 | struct gfs2_inode *dip = GFS2_I(inode); |
b3b94faa DT |
1009 | struct buffer_head *nbh, *obh, *dibh; |
1010 | struct gfs2_leaf *nleaf, *oleaf; | |
4da3c646 | 1011 | struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new; |
cd915493 | 1012 | u32 start, len, half_len, divider; |
b44b84d7 AV |
1013 | u64 bn, leaf_no; |
1014 | __be64 *lp; | |
cd915493 | 1015 | u32 index; |
b3b94faa DT |
1016 | int x, moved = 0; |
1017 | int error; | |
1018 | ||
9a004508 | 1019 | index = name->hash >> (32 - dip->i_depth); |
c752666c | 1020 | error = get_leaf_nr(dip, index, &leaf_no); |
287980e4 | 1021 | if (error) |
c752666c | 1022 | return error; |
b3b94faa DT |
1023 | |
1024 | /* Get the old leaf block */ | |
b3b94faa DT |
1025 | error = get_leaf(dip, leaf_no, &obh); |
1026 | if (error) | |
e90deff5 | 1027 | return error; |
b3b94faa | 1028 | |
b3b94faa | 1029 | oleaf = (struct gfs2_leaf *)obh->b_data; |
9a004508 | 1030 | if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) { |
e90deff5 SW |
1031 | brelse(obh); |
1032 | return 1; /* can't split */ | |
1033 | } | |
1034 | ||
350a9b0a | 1035 | gfs2_trans_add_meta(dip->i_gl, obh); |
b3b94faa | 1036 | |
c752666c SW |
1037 | nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1); |
1038 | if (!nleaf) { | |
1039 | brelse(obh); | |
1040 | return -ENOSPC; | |
1041 | } | |
1042 | bn = nbh->b_blocknr; | |
b3b94faa | 1043 | |
c752666c | 1044 | /* Compute the start and len of leaf pointers in the hash table. */ |
47a9a527 | 1045 | len = BIT(dip->i_depth - be16_to_cpu(oleaf->lf_depth)); |
b3b94faa DT |
1046 | half_len = len >> 1; |
1047 | if (!half_len) { | |
d77d1b58 JP |
1048 | pr_warn("i_depth %u lf_depth %u index %u\n", |
1049 | dip->i_depth, be16_to_cpu(oleaf->lf_depth), index); | |
b3b94faa DT |
1050 | gfs2_consist_inode(dip); |
1051 | error = -EIO; | |
1052 | goto fail_brelse; | |
1053 | } | |
1054 | ||
1055 | start = (index & ~(len - 1)); | |
1056 | ||
1057 | /* Change the pointers. | |
1058 | Don't bother distinguishing stuffed from non-stuffed. | |
1059 | This code is complicated enough already. */ | |
4244b52e DR |
1060 | lp = kmalloc(half_len * sizeof(__be64), GFP_NOFS); |
1061 | if (!lp) { | |
1062 | error = -ENOMEM; | |
1063 | goto fail_brelse; | |
1064 | } | |
1065 | ||
b3b94faa | 1066 | /* Change the pointers */ |
b3b94faa DT |
1067 | for (x = 0; x < half_len; x++) |
1068 | lp[x] = cpu_to_be64(bn); | |
1069 | ||
17d539f0 SW |
1070 | gfs2_dir_hash_inval(dip); |
1071 | ||
cd915493 SW |
1072 | error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64), |
1073 | half_len * sizeof(u64)); | |
1074 | if (error != half_len * sizeof(u64)) { | |
b3b94faa DT |
1075 | if (error >= 0) |
1076 | error = -EIO; | |
1077 | goto fail_lpfree; | |
1078 | } | |
1079 | ||
1080 | kfree(lp); | |
1081 | ||
1082 | /* Compute the divider */ | |
9a004508 | 1083 | divider = (start + half_len) << (32 - dip->i_depth); |
b3b94faa DT |
1084 | |
1085 | /* Copy the entries */ | |
1579343a | 1086 | dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf)); |
b3b94faa DT |
1087 | |
1088 | do { | |
1089 | next = dent; | |
1090 | if (dirent_next(dip, obh, &next)) | |
1091 | next = NULL; | |
1092 | ||
5e7d65cd | 1093 | if (!gfs2_dirent_sentinel(dent) && |
b3b94faa | 1094 | be32_to_cpu(dent->de_hash) < divider) { |
c752666c | 1095 | struct qstr str; |
34017472 | 1096 | void *ptr = ((char *)dent - obh->b_data) + nbh->b_data; |
c752666c SW |
1097 | str.name = (char*)(dent+1); |
1098 | str.len = be16_to_cpu(dent->de_name_len); | |
1099 | str.hash = be32_to_cpu(dent->de_hash); | |
34017472 | 1100 | new = gfs2_dirent_split_alloc(inode, nbh, &str, ptr); |
c752666c SW |
1101 | if (IS_ERR(new)) { |
1102 | error = PTR_ERR(new); | |
1103 | break; | |
1104 | } | |
b3b94faa DT |
1105 | |
1106 | new->de_inum = dent->de_inum; /* No endian worries */ | |
b3b94faa | 1107 | new->de_type = dent->de_type; /* No endian worries */ |
bb16b342 | 1108 | be16_add_cpu(&nleaf->lf_entries, 1); |
b3b94faa DT |
1109 | |
1110 | dirent_del(dip, obh, prev, dent); | |
1111 | ||
1112 | if (!oleaf->lf_entries) | |
1113 | gfs2_consist_inode(dip); | |
bb16b342 | 1114 | be16_add_cpu(&oleaf->lf_entries, -1); |
b3b94faa DT |
1115 | |
1116 | if (!prev) | |
1117 | prev = dent; | |
1118 | ||
1119 | moved = 1; | |
c752666c | 1120 | } else { |
b3b94faa | 1121 | prev = dent; |
c752666c | 1122 | } |
b3b94faa | 1123 | dent = next; |
c752666c | 1124 | } while (dent); |
b3b94faa | 1125 | |
c752666c | 1126 | oleaf->lf_depth = nleaf->lf_depth; |
b3b94faa DT |
1127 | |
1128 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
feaa7bba | 1129 | if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) { |
350a9b0a | 1130 | gfs2_trans_add_meta(dip->i_gl, dibh); |
77658aad | 1131 | gfs2_add_inode_blocks(&dip->i_inode, 1); |
539e5d6b | 1132 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
1133 | brelse(dibh); |
1134 | } | |
1135 | ||
1136 | brelse(obh); | |
1137 | brelse(nbh); | |
1138 | ||
1139 | return error; | |
1140 | ||
e90deff5 | 1141 | fail_lpfree: |
b3b94faa DT |
1142 | kfree(lp); |
1143 | ||
e90deff5 | 1144 | fail_brelse: |
b3b94faa | 1145 | brelse(obh); |
b3b94faa DT |
1146 | brelse(nbh); |
1147 | return error; | |
1148 | } | |
1149 | ||
1150 | /** | |
1151 | * dir_double_exhash - Double size of ExHash table | |
1152 | * @dip: The GFS2 dinode | |
1153 | * | |
1154 | * Returns: 0 on success, error code on failure | |
1155 | */ | |
1156 | ||
1157 | static int dir_double_exhash(struct gfs2_inode *dip) | |
1158 | { | |
b3b94faa | 1159 | struct buffer_head *dibh; |
cd915493 | 1160 | u32 hsize; |
17d539f0 SW |
1161 | u32 hsize_bytes; |
1162 | __be64 *hc; | |
1163 | __be64 *hc2, *h; | |
b3b94faa DT |
1164 | int x; |
1165 | int error = 0; | |
1166 | ||
47a9a527 | 1167 | hsize = BIT(dip->i_depth); |
17d539f0 | 1168 | hsize_bytes = hsize * sizeof(__be64); |
b3b94faa | 1169 | |
17d539f0 SW |
1170 | hc = gfs2_dir_get_hash_table(dip); |
1171 | if (IS_ERR(hc)) | |
1172 | return PTR_ERR(hc); | |
b3b94faa | 1173 | |
512cbf02 | 1174 | hc2 = kmalloc(hsize_bytes * 2, GFP_NOFS | __GFP_NOWARN); |
e8830d88 BP |
1175 | if (hc2 == NULL) |
1176 | hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS, PAGE_KERNEL); | |
1177 | ||
17d539f0 | 1178 | if (!hc2) |
4244b52e | 1179 | return -ENOMEM; |
b3b94faa | 1180 | |
512cbf02 | 1181 | h = hc2; |
17d539f0 SW |
1182 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1183 | if (error) | |
1184 | goto out_kfree; | |
b3b94faa | 1185 | |
17d539f0 SW |
1186 | for (x = 0; x < hsize; x++) { |
1187 | *h++ = *hc; | |
1188 | *h++ = *hc; | |
1189 | hc++; | |
b3b94faa DT |
1190 | } |
1191 | ||
17d539f0 SW |
1192 | error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2); |
1193 | if (error != (hsize_bytes * 2)) | |
1194 | goto fail; | |
b3b94faa | 1195 | |
17d539f0 SW |
1196 | gfs2_dir_hash_inval(dip); |
1197 | dip->i_hash_cache = hc2; | |
1198 | dip->i_depth++; | |
1199 | gfs2_dinode_out(dip, dibh->b_data); | |
1200 | brelse(dibh); | |
1201 | return 0; | |
b3b94faa | 1202 | |
a91ea69f | 1203 | fail: |
17d539f0 SW |
1204 | /* Replace original hash table & size */ |
1205 | gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes); | |
1206 | i_size_write(&dip->i_inode, hsize_bytes); | |
1207 | gfs2_dinode_out(dip, dibh->b_data); | |
1208 | brelse(dibh); | |
1209 | out_kfree: | |
3cdcf63e | 1210 | kvfree(hc2); |
b3b94faa DT |
1211 | return error; |
1212 | } | |
1213 | ||
1214 | /** | |
1215 | * compare_dents - compare directory entries by hash value | |
1216 | * @a: first dent | |
1217 | * @b: second dent | |
1218 | * | |
1219 | * When comparing the hash entries of @a to @b: | |
1220 | * gt: returns 1 | |
1221 | * lt: returns -1 | |
1222 | * eq: returns 0 | |
1223 | */ | |
1224 | ||
1225 | static int compare_dents(const void *a, const void *b) | |
1226 | { | |
2bdbc5d7 | 1227 | const struct gfs2_dirent *dent_a, *dent_b; |
cd915493 | 1228 | u32 hash_a, hash_b; |
b3b94faa DT |
1229 | int ret = 0; |
1230 | ||
2bdbc5d7 | 1231 | dent_a = *(const struct gfs2_dirent **)a; |
471f3db2 | 1232 | hash_a = dent_a->de_cookie; |
b3b94faa | 1233 | |
2bdbc5d7 | 1234 | dent_b = *(const struct gfs2_dirent **)b; |
471f3db2 | 1235 | hash_b = dent_b->de_cookie; |
b3b94faa DT |
1236 | |
1237 | if (hash_a > hash_b) | |
1238 | ret = 1; | |
1239 | else if (hash_a < hash_b) | |
1240 | ret = -1; | |
1241 | else { | |
4dd651ad SW |
1242 | unsigned int len_a = be16_to_cpu(dent_a->de_name_len); |
1243 | unsigned int len_b = be16_to_cpu(dent_b->de_name_len); | |
b3b94faa DT |
1244 | |
1245 | if (len_a > len_b) | |
1246 | ret = 1; | |
1247 | else if (len_a < len_b) | |
1248 | ret = -1; | |
1249 | else | |
2bdbc5d7 | 1250 | ret = memcmp(dent_a + 1, dent_b + 1, len_a); |
b3b94faa DT |
1251 | } |
1252 | ||
1253 | return ret; | |
1254 | } | |
1255 | ||
1256 | /** | |
1257 | * do_filldir_main - read out directory entries | |
1258 | * @dip: The GFS2 inode | |
d81a8ef5 | 1259 | * @ctx: what to feed the entries to |
b3b94faa DT |
1260 | * @darr: an array of struct gfs2_dirent pointers to read |
1261 | * @entries: the number of entries in darr | |
1262 | * @copied: pointer to int that's non-zero if a entry has been copied out | |
1263 | * | |
1264 | * Jump through some hoops to make sure that if there are hash collsions, | |
1265 | * they are read out at the beginning of a buffer. We want to minimize | |
1266 | * the possibility that they will fall into different readdir buffers or | |
1267 | * that someone will want to seek to that location. | |
1268 | * | |
d81a8ef5 | 1269 | * Returns: errno, >0 if the actor tells you to stop |
b3b94faa DT |
1270 | */ |
1271 | ||
d81a8ef5 | 1272 | static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx, |
471f3db2 BM |
1273 | struct gfs2_dirent **darr, u32 entries, |
1274 | u32 sort_start, int *copied) | |
b3b94faa | 1275 | { |
71b86f56 | 1276 | const struct gfs2_dirent *dent, *dent_next; |
cd915493 | 1277 | u64 off, off_next; |
b3b94faa DT |
1278 | unsigned int x, y; |
1279 | int run = 0; | |
b3b94faa | 1280 | |
471f3db2 BM |
1281 | if (sort_start < entries) |
1282 | sort(&darr[sort_start], entries - sort_start, | |
1283 | sizeof(struct gfs2_dirent *), compare_dents, NULL); | |
b3b94faa DT |
1284 | |
1285 | dent_next = darr[0]; | |
471f3db2 | 1286 | off_next = dent_next->de_cookie; |
b3b94faa DT |
1287 | |
1288 | for (x = 0, y = 1; x < entries; x++, y++) { | |
1289 | dent = dent_next; | |
1290 | off = off_next; | |
1291 | ||
1292 | if (y < entries) { | |
1293 | dent_next = darr[y]; | |
471f3db2 | 1294 | off_next = dent_next->de_cookie; |
b3b94faa | 1295 | |
d81a8ef5 | 1296 | if (off < ctx->pos) |
b3b94faa | 1297 | continue; |
d81a8ef5 | 1298 | ctx->pos = off; |
b3b94faa DT |
1299 | |
1300 | if (off_next == off) { | |
1301 | if (*copied && !run) | |
1302 | return 1; | |
1303 | run = 1; | |
1304 | } else | |
1305 | run = 0; | |
1306 | } else { | |
d81a8ef5 | 1307 | if (off < ctx->pos) |
b3b94faa | 1308 | continue; |
d81a8ef5 | 1309 | ctx->pos = off; |
b3b94faa DT |
1310 | } |
1311 | ||
d81a8ef5 | 1312 | if (!dir_emit(ctx, (const char *)(dent + 1), |
4dd651ad | 1313 | be16_to_cpu(dent->de_name_len), |
d81a8ef5 AV |
1314 | be64_to_cpu(dent->de_inum.no_addr), |
1315 | be16_to_cpu(dent->de_type))) | |
b3b94faa DT |
1316 | return 1; |
1317 | ||
1318 | *copied = 1; | |
1319 | } | |
1320 | ||
d81a8ef5 | 1321 | /* Increment the ctx->pos by one, so the next time we come into the |
b3b94faa DT |
1322 | do_filldir fxn, we get the next entry instead of the last one in the |
1323 | current leaf */ | |
1324 | ||
d81a8ef5 | 1325 | ctx->pos++; |
b3b94faa DT |
1326 | |
1327 | return 0; | |
1328 | } | |
1329 | ||
d2a97a4e SW |
1330 | static void *gfs2_alloc_sort_buffer(unsigned size) |
1331 | { | |
1332 | void *ptr = NULL; | |
1333 | ||
1334 | if (size < KMALLOC_MAX_SIZE) | |
1335 | ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN); | |
1336 | if (!ptr) | |
1337 | ptr = __vmalloc(size, GFP_NOFS, PAGE_KERNEL); | |
1338 | return ptr; | |
1339 | } | |
1340 | ||
471f3db2 BM |
1341 | |
1342 | static int gfs2_set_cookies(struct gfs2_sbd *sdp, struct buffer_head *bh, | |
1343 | unsigned leaf_nr, struct gfs2_dirent **darr, | |
1344 | unsigned entries) | |
1345 | { | |
1346 | int sort_id = -1; | |
1347 | int i; | |
1348 | ||
1349 | for (i = 0; i < entries; i++) { | |
1350 | unsigned offset; | |
1351 | ||
1352 | darr[i]->de_cookie = be32_to_cpu(darr[i]->de_hash); | |
1353 | darr[i]->de_cookie = gfs2_disk_hash2offset(darr[i]->de_cookie); | |
1354 | ||
1355 | if (!sdp->sd_args.ar_loccookie) | |
1356 | continue; | |
1357 | offset = (char *)(darr[i]) - | |
1358 | (bh->b_data + gfs2_dirent_offset(bh->b_data)); | |
1359 | offset /= GFS2_MIN_DIRENT_SIZE; | |
1360 | offset += leaf_nr * sdp->sd_max_dents_per_leaf; | |
1361 | if (offset >= GFS2_USE_HASH_FLAG || | |
1362 | leaf_nr >= GFS2_USE_HASH_FLAG) { | |
1363 | darr[i]->de_cookie |= GFS2_USE_HASH_FLAG; | |
1364 | if (sort_id < 0) | |
1365 | sort_id = i; | |
1366 | continue; | |
1367 | } | |
1368 | darr[i]->de_cookie &= GFS2_HASH_INDEX_MASK; | |
1369 | darr[i]->de_cookie |= offset; | |
1370 | } | |
1371 | return sort_id; | |
1372 | } | |
1373 | ||
1374 | ||
d81a8ef5 AV |
1375 | static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx, |
1376 | int *copied, unsigned *depth, | |
3699e3a4 | 1377 | u64 leaf_no) |
b3b94faa | 1378 | { |
feaa7bba | 1379 | struct gfs2_inode *ip = GFS2_I(inode); |
bdd19a22 | 1380 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
71b86f56 SW |
1381 | struct buffer_head *bh; |
1382 | struct gfs2_leaf *lf; | |
bdd19a22 | 1383 | unsigned entries = 0, entries2 = 0; |
471f3db2 BM |
1384 | unsigned leaves = 0, leaf = 0, offset, sort_offset; |
1385 | struct gfs2_dirent **darr, *dent; | |
71b86f56 SW |
1386 | struct dirent_gather g; |
1387 | struct buffer_head **larr; | |
471f3db2 | 1388 | int error, i, need_sort = 0, sort_id; |
71b86f56 | 1389 | u64 lfn = leaf_no; |
b3b94faa | 1390 | |
b3b94faa | 1391 | do { |
71b86f56 | 1392 | error = get_leaf(ip, lfn, &bh); |
b3b94faa | 1393 | if (error) |
71b86f56 SW |
1394 | goto out; |
1395 | lf = (struct gfs2_leaf *)bh->b_data; | |
1396 | if (leaves == 0) | |
1397 | *depth = be16_to_cpu(lf->lf_depth); | |
1398 | entries += be16_to_cpu(lf->lf_entries); | |
1399 | leaves++; | |
1400 | lfn = be64_to_cpu(lf->lf_next); | |
1401 | brelse(bh); | |
1402 | } while(lfn); | |
b3b94faa | 1403 | |
471f3db2 BM |
1404 | if (*depth < GFS2_DIR_MAX_DEPTH || !sdp->sd_args.ar_loccookie) { |
1405 | need_sort = 1; | |
1406 | sort_offset = 0; | |
1407 | } | |
1408 | ||
b3b94faa DT |
1409 | if (!entries) |
1410 | return 0; | |
1411 | ||
71b86f56 | 1412 | error = -ENOMEM; |
bdd19a22 SW |
1413 | /* |
1414 | * The extra 99 entries are not normally used, but are a buffer | |
1415 | * zone in case the number of entries in the leaf is corrupt. | |
1416 | * 99 is the maximum number of entries that can fit in a single | |
1417 | * leaf block. | |
1418 | */ | |
d2a97a4e | 1419 | larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *)); |
71b86f56 SW |
1420 | if (!larr) |
1421 | goto out; | |
471f3db2 BM |
1422 | darr = (struct gfs2_dirent **)(larr + leaves); |
1423 | g.pdent = (const struct gfs2_dirent **)darr; | |
71b86f56 SW |
1424 | g.offset = 0; |
1425 | lfn = leaf_no; | |
b3b94faa | 1426 | |
71b86f56 SW |
1427 | do { |
1428 | error = get_leaf(ip, lfn, &bh); | |
b3b94faa | 1429 | if (error) |
d2a97a4e | 1430 | goto out_free; |
71b86f56 SW |
1431 | lf = (struct gfs2_leaf *)bh->b_data; |
1432 | lfn = be64_to_cpu(lf->lf_next); | |
1433 | if (lf->lf_entries) { | |
471f3db2 | 1434 | offset = g.offset; |
bdd19a22 | 1435 | entries2 += be16_to_cpu(lf->lf_entries); |
71b86f56 SW |
1436 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, |
1437 | gfs2_dirent_gather, NULL, &g); | |
1438 | error = PTR_ERR(dent); | |
bdd19a22 | 1439 | if (IS_ERR(dent)) |
d2a97a4e | 1440 | goto out_free; |
bdd19a22 | 1441 | if (entries2 != g.offset) { |
f391a4ea AM |
1442 | fs_warn(sdp, "Number of entries corrupt in dir " |
1443 | "leaf %llu, entries2 (%u) != " | |
1444 | "g.offset (%u)\n", | |
1445 | (unsigned long long)bh->b_blocknr, | |
1446 | entries2, g.offset); | |
d87d62b7 | 1447 | gfs2_consist_inode(ip); |
bdd19a22 | 1448 | error = -EIO; |
d2a97a4e | 1449 | goto out_free; |
71b86f56 SW |
1450 | } |
1451 | error = 0; | |
471f3db2 BM |
1452 | sort_id = gfs2_set_cookies(sdp, bh, leaf, &darr[offset], |
1453 | be16_to_cpu(lf->lf_entries)); | |
1454 | if (!need_sort && sort_id >= 0) { | |
1455 | need_sort = 1; | |
1456 | sort_offset = offset + sort_id; | |
1457 | } | |
71b86f56 | 1458 | larr[leaf++] = bh; |
b3b94faa | 1459 | } else { |
471f3db2 | 1460 | larr[leaf++] = NULL; |
71b86f56 | 1461 | brelse(bh); |
b3b94faa | 1462 | } |
71b86f56 | 1463 | } while(lfn); |
b3b94faa | 1464 | |
bdd19a22 | 1465 | BUG_ON(entries2 != entries); |
471f3db2 BM |
1466 | error = do_filldir_main(ip, ctx, darr, entries, need_sort ? |
1467 | sort_offset : entries, copied); | |
d2a97a4e | 1468 | out_free: |
71b86f56 | 1469 | for(i = 0; i < leaf; i++) |
471f3db2 BM |
1470 | if (larr[i]) |
1471 | brelse(larr[i]); | |
3cdcf63e | 1472 | kvfree(larr); |
71b86f56 | 1473 | out: |
b3b94faa DT |
1474 | return error; |
1475 | } | |
1476 | ||
79c4c379 SW |
1477 | /** |
1478 | * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks. | |
dfe4d34b BP |
1479 | * |
1480 | * Note: we can't calculate each index like dir_e_read can because we don't | |
1481 | * have the leaf, and therefore we don't have the depth, and therefore we | |
1482 | * don't have the length. So we have to just read enough ahead to make up | |
79c4c379 SW |
1483 | * for the loss of information. |
1484 | */ | |
dfe4d34b BP |
1485 | static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index, |
1486 | struct file_ra_state *f_ra) | |
1487 | { | |
1488 | struct gfs2_inode *ip = GFS2_I(inode); | |
1489 | struct gfs2_glock *gl = ip->i_gl; | |
1490 | struct buffer_head *bh; | |
1491 | u64 blocknr = 0, last; | |
1492 | unsigned count; | |
1493 | ||
1494 | /* First check if we've already read-ahead for the whole range. */ | |
79c4c379 | 1495 | if (index + MAX_RA_BLOCKS < f_ra->start) |
dfe4d34b BP |
1496 | return; |
1497 | ||
1498 | f_ra->start = max((pgoff_t)index, f_ra->start); | |
1499 | for (count = 0; count < MAX_RA_BLOCKS; count++) { | |
1500 | if (f_ra->start >= hsize) /* if exceeded the hash table */ | |
1501 | break; | |
1502 | ||
1503 | last = blocknr; | |
1504 | blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]); | |
1505 | f_ra->start++; | |
1506 | if (blocknr == last) | |
1507 | continue; | |
1508 | ||
1509 | bh = gfs2_getbuf(gl, blocknr, 1); | |
1510 | if (trylock_buffer(bh)) { | |
1511 | if (buffer_uptodate(bh)) { | |
1512 | unlock_buffer(bh); | |
1513 | brelse(bh); | |
1514 | continue; | |
1515 | } | |
1516 | bh->b_end_io = end_buffer_read_sync; | |
70246286 | 1517 | submit_bh(REQ_OP_READ, REQ_RAHEAD | REQ_META, bh); |
dfe4d34b BP |
1518 | continue; |
1519 | } | |
1520 | brelse(bh); | |
1521 | } | |
1522 | } | |
17d539f0 | 1523 | |
b3b94faa | 1524 | /** |
c752666c SW |
1525 | * dir_e_read - Reads the entries from a directory into a filldir buffer |
1526 | * @dip: dinode pointer | |
d81a8ef5 | 1527 | * @ctx: actor to feed the entries to |
b3b94faa | 1528 | * |
c752666c | 1529 | * Returns: errno |
b3b94faa DT |
1530 | */ |
1531 | ||
d81a8ef5 AV |
1532 | static int dir_e_read(struct inode *inode, struct dir_context *ctx, |
1533 | struct file_ra_state *f_ra) | |
b3b94faa | 1534 | { |
feaa7bba | 1535 | struct gfs2_inode *dip = GFS2_I(inode); |
cd915493 | 1536 | u32 hsize, len = 0; |
cd915493 | 1537 | u32 hash, index; |
b44b84d7 | 1538 | __be64 *lp; |
c752666c SW |
1539 | int copied = 0; |
1540 | int error = 0; | |
4da3c646 | 1541 | unsigned depth = 0; |
b3b94faa | 1542 | |
47a9a527 | 1543 | hsize = BIT(dip->i_depth); |
d81a8ef5 | 1544 | hash = gfs2_dir_offset2hash(ctx->pos); |
9a004508 | 1545 | index = hash >> (32 - dip->i_depth); |
b3b94faa | 1546 | |
79c4c379 | 1547 | if (dip->i_hash_cache == NULL) |
dfe4d34b | 1548 | f_ra->start = 0; |
17d539f0 SW |
1549 | lp = gfs2_dir_get_hash_table(dip); |
1550 | if (IS_ERR(lp)) | |
1551 | return PTR_ERR(lp); | |
b3b94faa | 1552 | |
dfe4d34b BP |
1553 | gfs2_dir_readahead(inode, hsize, index, f_ra); |
1554 | ||
b3b94faa | 1555 | while (index < hsize) { |
d81a8ef5 | 1556 | error = gfs2_dir_read_leaf(inode, ctx, |
71b86f56 | 1557 | &copied, &depth, |
17d539f0 | 1558 | be64_to_cpu(lp[index])); |
b3b94faa | 1559 | if (error) |
71b86f56 | 1560 | break; |
b3b94faa | 1561 | |
47a9a527 | 1562 | len = BIT(dip->i_depth - depth); |
b3b94faa DT |
1563 | index = (index & ~(len - 1)) + len; |
1564 | } | |
1565 | ||
71b86f56 SW |
1566 | if (error > 0) |
1567 | error = 0; | |
b3b94faa DT |
1568 | return error; |
1569 | } | |
1570 | ||
d81a8ef5 AV |
1571 | int gfs2_dir_read(struct inode *inode, struct dir_context *ctx, |
1572 | struct file_ra_state *f_ra) | |
b3b94faa | 1573 | { |
feaa7bba | 1574 | struct gfs2_inode *dip = GFS2_I(inode); |
bdd19a22 | 1575 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
71b86f56 | 1576 | struct dirent_gather g; |
471f3db2 | 1577 | struct gfs2_dirent **darr, *dent; |
b3b94faa DT |
1578 | struct buffer_head *dibh; |
1579 | int copied = 0; | |
1580 | int error; | |
1581 | ||
ad6203f2 | 1582 | if (!dip->i_entries) |
71b86f56 SW |
1583 | return 0; |
1584 | ||
383f01fb | 1585 | if (dip->i_diskflags & GFS2_DIF_EXHASH) |
d81a8ef5 | 1586 | return dir_e_read(inode, ctx, f_ra); |
71b86f56 | 1587 | |
b3b94faa DT |
1588 | if (!gfs2_is_stuffed(dip)) { |
1589 | gfs2_consist_inode(dip); | |
1590 | return -EIO; | |
1591 | } | |
1592 | ||
b3b94faa DT |
1593 | error = gfs2_meta_inode_buffer(dip, &dibh); |
1594 | if (error) | |
1595 | return error; | |
1596 | ||
71b86f56 | 1597 | error = -ENOMEM; |
bdd19a22 | 1598 | /* 96 is max number of dirents which can be stuffed into an inode */ |
16c5f06f | 1599 | darr = kmalloc(96 * sizeof(struct gfs2_dirent *), GFP_NOFS); |
71b86f56 | 1600 | if (darr) { |
471f3db2 | 1601 | g.pdent = (const struct gfs2_dirent **)darr; |
71b86f56 SW |
1602 | g.offset = 0; |
1603 | dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size, | |
1604 | gfs2_dirent_gather, NULL, &g); | |
1605 | if (IS_ERR(dent)) { | |
1606 | error = PTR_ERR(dent); | |
1607 | goto out; | |
1608 | } | |
ad6203f2 | 1609 | if (dip->i_entries != g.offset) { |
bdd19a22 | 1610 | fs_warn(sdp, "Number of entries corrupt in dir %llu, " |
ad6203f2 | 1611 | "ip->i_entries (%u) != g.offset (%u)\n", |
dbb7cae2 | 1612 | (unsigned long long)dip->i_no_addr, |
ad6203f2 | 1613 | dip->i_entries, |
bdd19a22 | 1614 | g.offset); |
d87d62b7 | 1615 | gfs2_consist_inode(dip); |
bdd19a22 SW |
1616 | error = -EIO; |
1617 | goto out; | |
1618 | } | |
471f3db2 | 1619 | gfs2_set_cookies(sdp, dibh, 0, darr, dip->i_entries); |
d81a8ef5 | 1620 | error = do_filldir_main(dip, ctx, darr, |
471f3db2 | 1621 | dip->i_entries, 0, &copied); |
71b86f56 SW |
1622 | out: |
1623 | kfree(darr); | |
1624 | } | |
1625 | ||
b3b94faa DT |
1626 | if (error > 0) |
1627 | error = 0; | |
1628 | ||
1629 | brelse(dibh); | |
1630 | ||
1631 | return error; | |
1632 | } | |
1633 | ||
b3b94faa DT |
1634 | /** |
1635 | * gfs2_dir_search - Search a directory | |
5a00f3cc SW |
1636 | * @dip: The GFS2 dir inode |
1637 | * @name: The name we are looking up | |
1638 | * @fail_on_exist: Fail if the name exists rather than looking it up | |
b3b94faa DT |
1639 | * |
1640 | * This routine searches a directory for a file or another directory. | |
1641 | * Assumes a glock is held on dip. | |
1642 | * | |
1643 | * Returns: errno | |
1644 | */ | |
1645 | ||
5a00f3cc SW |
1646 | struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name, |
1647 | bool fail_on_exist) | |
b3b94faa | 1648 | { |
c752666c SW |
1649 | struct buffer_head *bh; |
1650 | struct gfs2_dirent *dent; | |
5a00f3cc SW |
1651 | u64 addr, formal_ino; |
1652 | u16 dtype; | |
dbb7cae2 SW |
1653 | |
1654 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1655 | if (dent) { | |
c8d57703 AG |
1656 | struct inode *inode; |
1657 | u16 rahead; | |
1658 | ||
dbb7cae2 | 1659 | if (IS_ERR(dent)) |
e231c2ee | 1660 | return ERR_CAST(dent); |
5a00f3cc | 1661 | dtype = be16_to_cpu(dent->de_type); |
c8d57703 | 1662 | rahead = be16_to_cpu(dent->de_rahead); |
5a00f3cc SW |
1663 | addr = be64_to_cpu(dent->de_inum.no_addr); |
1664 | formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino); | |
dbb7cae2 | 1665 | brelse(bh); |
5a00f3cc SW |
1666 | if (fail_on_exist) |
1667 | return ERR_PTR(-EEXIST); | |
3ce37b2c AG |
1668 | inode = gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino, |
1669 | GFS2_BLKST_FREE /* ignore */); | |
c8d57703 AG |
1670 | if (!IS_ERR(inode)) |
1671 | GFS2_I(inode)->i_rahead = rahead; | |
1672 | return inode; | |
dbb7cae2 SW |
1673 | } |
1674 | return ERR_PTR(-ENOENT); | |
1675 | } | |
1676 | ||
1677 | int gfs2_dir_check(struct inode *dir, const struct qstr *name, | |
1678 | const struct gfs2_inode *ip) | |
1679 | { | |
1680 | struct buffer_head *bh; | |
1681 | struct gfs2_dirent *dent; | |
1682 | int ret = -ENOENT; | |
c752666c SW |
1683 | |
1684 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh); | |
1685 | if (dent) { | |
1686 | if (IS_ERR(dent)) | |
1687 | return PTR_ERR(dent); | |
dbb7cae2 SW |
1688 | if (ip) { |
1689 | if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr) | |
1690 | goto out; | |
1691 | if (be64_to_cpu(dent->de_inum.no_formal_ino) != | |
1692 | ip->i_no_formal_ino) | |
1693 | goto out; | |
1694 | if (unlikely(IF2DT(ip->i_inode.i_mode) != | |
1695 | be16_to_cpu(dent->de_type))) { | |
1696 | gfs2_consist_inode(GFS2_I(dir)); | |
1697 | ret = -EIO; | |
1698 | goto out; | |
1699 | } | |
1700 | } | |
1701 | ret = 0; | |
1702 | out: | |
c752666c | 1703 | brelse(bh); |
c752666c | 1704 | } |
dbb7cae2 | 1705 | return ret; |
c752666c SW |
1706 | } |
1707 | ||
01bcb0de SW |
1708 | /** |
1709 | * dir_new_leaf - Add a new leaf onto hash chain | |
1710 | * @inode: The directory | |
1711 | * @name: The name we are adding | |
1712 | * | |
1713 | * This adds a new dir leaf onto an existing leaf when there is not | |
1714 | * enough space to add a new dir entry. This is a last resort after | |
1715 | * we've expanded the hash table to max size and also split existing | |
1716 | * leaf blocks, so it will only occur for very large directories. | |
1717 | * | |
1718 | * The dist parameter is set to 1 for leaf blocks directly attached | |
1719 | * to the hash table, 2 for one layer of indirection, 3 for two layers | |
1720 | * etc. We are thus able to tell the difference between an old leaf | |
1721 | * with dist set to zero (i.e. "don't know") and a new one where we | |
1722 | * set this information for debug/fsck purposes. | |
1723 | * | |
1724 | * Returns: 0 on success, or -ve on error | |
1725 | */ | |
1726 | ||
c752666c SW |
1727 | static int dir_new_leaf(struct inode *inode, const struct qstr *name) |
1728 | { | |
1729 | struct buffer_head *bh, *obh; | |
feaa7bba | 1730 | struct gfs2_inode *ip = GFS2_I(inode); |
c752666c | 1731 | struct gfs2_leaf *leaf, *oleaf; |
01bcb0de | 1732 | u32 dist = 1; |
b3b94faa | 1733 | int error; |
c752666c SW |
1734 | u32 index; |
1735 | u64 bn; | |
b3b94faa | 1736 | |
9a004508 | 1737 | index = name->hash >> (32 - ip->i_depth); |
c752666c SW |
1738 | error = get_first_leaf(ip, index, &obh); |
1739 | if (error) | |
1740 | return error; | |
1741 | do { | |
01bcb0de | 1742 | dist++; |
c752666c SW |
1743 | oleaf = (struct gfs2_leaf *)obh->b_data; |
1744 | bn = be64_to_cpu(oleaf->lf_next); | |
1745 | if (!bn) | |
1746 | break; | |
1747 | brelse(obh); | |
1748 | error = get_leaf(ip, bn, &obh); | |
1749 | if (error) | |
1750 | return error; | |
1751 | } while(1); | |
b3b94faa | 1752 | |
350a9b0a | 1753 | gfs2_trans_add_meta(ip->i_gl, obh); |
c752666c SW |
1754 | |
1755 | leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth)); | |
1756 | if (!leaf) { | |
1757 | brelse(obh); | |
1758 | return -ENOSPC; | |
1759 | } | |
01bcb0de | 1760 | leaf->lf_dist = cpu_to_be32(dist); |
4d8012b6 | 1761 | oleaf->lf_next = cpu_to_be64(bh->b_blocknr); |
c752666c SW |
1762 | brelse(bh); |
1763 | brelse(obh); | |
1764 | ||
1765 | error = gfs2_meta_inode_buffer(ip, &bh); | |
1766 | if (error) | |
1767 | return error; | |
350a9b0a | 1768 | gfs2_trans_add_meta(ip->i_gl, bh); |
77658aad | 1769 | gfs2_add_inode_blocks(&ip->i_inode, 1); |
539e5d6b | 1770 | gfs2_dinode_out(ip, bh->b_data); |
c752666c SW |
1771 | brelse(bh); |
1772 | return 0; | |
b3b94faa DT |
1773 | } |
1774 | ||
44aaada9 SW |
1775 | static u16 gfs2_inode_ra_len(const struct gfs2_inode *ip) |
1776 | { | |
1777 | u64 where = ip->i_no_addr + 1; | |
1778 | if (ip->i_eattr == where) | |
1779 | return 1; | |
1780 | return 0; | |
1781 | } | |
1782 | ||
b3b94faa DT |
1783 | /** |
1784 | * gfs2_dir_add - Add new filename into directory | |
2b47dad8 SW |
1785 | * @inode: The directory inode |
1786 | * @name: The new name | |
1787 | * @nip: The GFS2 inode to be linked in to the directory | |
1788 | * @da: The directory addition info | |
1789 | * | |
1790 | * If the call to gfs2_diradd_alloc_required resulted in there being | |
1791 | * no need to allocate any new directory blocks, then it will contain | |
1792 | * a pointer to the directory entry and the bh in which it resides. We | |
1793 | * can use that without having to repeat the search. If there was no | |
1794 | * free space, then we must now create more space. | |
b3b94faa DT |
1795 | * |
1796 | * Returns: 0 on success, error code on failure | |
1797 | */ | |
1798 | ||
c752666c | 1799 | int gfs2_dir_add(struct inode *inode, const struct qstr *name, |
2b47dad8 | 1800 | const struct gfs2_inode *nip, struct gfs2_diradd *da) |
b3b94faa | 1801 | { |
feaa7bba | 1802 | struct gfs2_inode *ip = GFS2_I(inode); |
2b47dad8 SW |
1803 | struct buffer_head *bh = da->bh; |
1804 | struct gfs2_dirent *dent = da->dent; | |
01bcb0de | 1805 | struct timespec tv; |
c752666c | 1806 | struct gfs2_leaf *leaf; |
b3b94faa DT |
1807 | int error; |
1808 | ||
c752666c | 1809 | while(1) { |
2b47dad8 SW |
1810 | if (da->bh == NULL) { |
1811 | dent = gfs2_dirent_search(inode, name, | |
1812 | gfs2_dirent_find_space, &bh); | |
1813 | } | |
c752666c SW |
1814 | if (dent) { |
1815 | if (IS_ERR(dent)) | |
1816 | return PTR_ERR(dent); | |
1817 | dent = gfs2_init_dirent(inode, dent, name, bh); | |
dbb7cae2 | 1818 | gfs2_inum_out(nip, dent); |
3d6ecb7d | 1819 | dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode)); |
44aaada9 | 1820 | dent->de_rahead = cpu_to_be16(gfs2_inode_ra_len(nip)); |
078cd827 | 1821 | tv = current_time(&ip->i_inode); |
383f01fb | 1822 | if (ip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c | 1823 | leaf = (struct gfs2_leaf *)bh->b_data; |
bb16b342 | 1824 | be16_add_cpu(&leaf->lf_entries, 1); |
01bcb0de SW |
1825 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); |
1826 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
c752666c | 1827 | } |
2b47dad8 SW |
1828 | da->dent = NULL; |
1829 | da->bh = NULL; | |
c752666c | 1830 | brelse(bh); |
ad6203f2 | 1831 | ip->i_entries++; |
01bcb0de | 1832 | ip->i_inode.i_mtime = ip->i_inode.i_ctime = tv; |
3d6ecb7d SW |
1833 | if (S_ISDIR(nip->i_inode.i_mode)) |
1834 | inc_nlink(&ip->i_inode); | |
343cd8f0 | 1835 | mark_inode_dirty(inode); |
c752666c SW |
1836 | error = 0; |
1837 | break; | |
1838 | } | |
383f01fb | 1839 | if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) { |
c752666c SW |
1840 | error = dir_make_exhash(inode); |
1841 | if (error) | |
1842 | break; | |
1843 | continue; | |
1844 | } | |
1845 | error = dir_split_leaf(inode, name); | |
1846 | if (error == 0) | |
1847 | continue; | |
e90deff5 | 1848 | if (error < 0) |
c752666c | 1849 | break; |
9a004508 | 1850 | if (ip->i_depth < GFS2_DIR_MAX_DEPTH) { |
c752666c SW |
1851 | error = dir_double_exhash(ip); |
1852 | if (error) | |
1853 | break; | |
1854 | error = dir_split_leaf(inode, name); | |
e90deff5 | 1855 | if (error < 0) |
c752666c | 1856 | break; |
e90deff5 SW |
1857 | if (error == 0) |
1858 | continue; | |
c752666c SW |
1859 | } |
1860 | error = dir_new_leaf(inode, name); | |
1861 | if (!error) | |
1862 | continue; | |
1863 | error = -ENOSPC; | |
1864 | break; | |
1865 | } | |
b3b94faa DT |
1866 | return error; |
1867 | } | |
1868 | ||
c752666c | 1869 | |
b3b94faa DT |
1870 | /** |
1871 | * gfs2_dir_del - Delete a directory entry | |
1872 | * @dip: The GFS2 inode | |
1873 | * @filename: The filename | |
1874 | * | |
1875 | * Returns: 0 on success, error code on failure | |
1876 | */ | |
1877 | ||
855d23ce | 1878 | int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry) |
b3b94faa | 1879 | { |
855d23ce | 1880 | const struct qstr *name = &dentry->d_name; |
c752666c SW |
1881 | struct gfs2_dirent *dent, *prev = NULL; |
1882 | struct buffer_head *bh; | |
078cd827 | 1883 | struct timespec tv = current_time(&dip->i_inode); |
b3b94faa | 1884 | |
c752666c SW |
1885 | /* Returns _either_ the entry (if its first in block) or the |
1886 | previous entry otherwise */ | |
feaa7bba | 1887 | dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh); |
c752666c SW |
1888 | if (!dent) { |
1889 | gfs2_consist_inode(dip); | |
1890 | return -EIO; | |
1891 | } | |
1892 | if (IS_ERR(dent)) { | |
1893 | gfs2_consist_inode(dip); | |
1894 | return PTR_ERR(dent); | |
1895 | } | |
1896 | /* If not first in block, adjust pointers accordingly */ | |
71b86f56 | 1897 | if (gfs2_dirent_find(dent, name, NULL) == 0) { |
c752666c SW |
1898 | prev = dent; |
1899 | dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len)); | |
1900 | } | |
1901 | ||
1902 | dirent_del(dip, bh, prev, dent); | |
383f01fb | 1903 | if (dip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c SW |
1904 | struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data; |
1905 | u16 entries = be16_to_cpu(leaf->lf_entries); | |
1906 | if (!entries) | |
1907 | gfs2_consist_inode(dip); | |
1908 | leaf->lf_entries = cpu_to_be16(--entries); | |
01bcb0de SW |
1909 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec); |
1910 | leaf->lf_sec = cpu_to_be64(tv.tv_sec); | |
c752666c | 1911 | } |
ed386507 | 1912 | brelse(bh); |
c752666c | 1913 | |
ad6203f2 | 1914 | if (!dip->i_entries) |
c752666c | 1915 | gfs2_consist_inode(dip); |
ad6203f2 | 1916 | dip->i_entries--; |
01bcb0de | 1917 | dip->i_inode.i_mtime = dip->i_inode.i_ctime = tv; |
e36cb0b8 | 1918 | if (d_is_dir(dentry)) |
855d23ce | 1919 | drop_nlink(&dip->i_inode); |
feaa7bba | 1920 | mark_inode_dirty(&dip->i_inode); |
b3b94faa | 1921 | |
ab9bbda0 | 1922 | return 0; |
b3b94faa DT |
1923 | } |
1924 | ||
b3b94faa DT |
1925 | /** |
1926 | * gfs2_dir_mvino - Change inode number of directory entry | |
1927 | * @dip: The GFS2 inode | |
1928 | * @filename: | |
1929 | * @new_inode: | |
1930 | * | |
1931 | * This routine changes the inode number of a directory entry. It's used | |
1932 | * by rename to change ".." when a directory is moved. | |
1933 | * Assumes a glock is held on dvp. | |
1934 | * | |
1935 | * Returns: errno | |
1936 | */ | |
1937 | ||
c752666c | 1938 | int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename, |
dbb7cae2 | 1939 | const struct gfs2_inode *nip, unsigned int new_type) |
b3b94faa | 1940 | { |
c752666c SW |
1941 | struct buffer_head *bh; |
1942 | struct gfs2_dirent *dent; | |
b3b94faa DT |
1943 | int error; |
1944 | ||
feaa7bba | 1945 | dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh); |
c752666c SW |
1946 | if (!dent) { |
1947 | gfs2_consist_inode(dip); | |
1948 | return -EIO; | |
1949 | } | |
1950 | if (IS_ERR(dent)) | |
1951 | return PTR_ERR(dent); | |
b3b94faa | 1952 | |
350a9b0a | 1953 | gfs2_trans_add_meta(dip->i_gl, bh); |
dbb7cae2 | 1954 | gfs2_inum_out(nip, dent); |
c752666c SW |
1955 | dent->de_type = cpu_to_be16(new_type); |
1956 | ||
383f01fb | 1957 | if (dip->i_diskflags & GFS2_DIF_EXHASH) { |
c752666c SW |
1958 | brelse(bh); |
1959 | error = gfs2_meta_inode_buffer(dip, &bh); | |
1960 | if (error) | |
1961 | return error; | |
350a9b0a | 1962 | gfs2_trans_add_meta(dip->i_gl, bh); |
c752666c SW |
1963 | } |
1964 | ||
078cd827 | 1965 | dip->i_inode.i_mtime = dip->i_inode.i_ctime = current_time(&dip->i_inode); |
539e5d6b | 1966 | gfs2_dinode_out(dip, bh->b_data); |
c752666c SW |
1967 | brelse(bh); |
1968 | return 0; | |
b3b94faa DT |
1969 | } |
1970 | ||
b3b94faa DT |
1971 | /** |
1972 | * leaf_dealloc - Deallocate a directory leaf | |
1973 | * @dip: the directory | |
1974 | * @index: the hash table offset in the directory | |
1975 | * @len: the number of pointers to this leaf | |
1976 | * @leaf_no: the leaf number | |
ec038c82 | 1977 | * @leaf_bh: buffer_head for the starting leaf |
d24a7a43 | 1978 | * last_dealloc: 1 if this is the final dealloc for the leaf, else 0 |
b3b94faa DT |
1979 | * |
1980 | * Returns: errno | |
1981 | */ | |
1982 | ||
cd915493 | 1983 | static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len, |
ec038c82 BP |
1984 | u64 leaf_no, struct buffer_head *leaf_bh, |
1985 | int last_dealloc) | |
b3b94faa | 1986 | { |
feaa7bba | 1987 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode); |
c752666c | 1988 | struct gfs2_leaf *tmp_leaf; |
b3b94faa DT |
1989 | struct gfs2_rgrp_list rlist; |
1990 | struct buffer_head *bh, *dibh; | |
cd915493 | 1991 | u64 blk, nblk; |
b3b94faa DT |
1992 | unsigned int rg_blocks = 0, l_blocks = 0; |
1993 | char *ht; | |
cd915493 | 1994 | unsigned int x, size = len * sizeof(u64); |
b3b94faa DT |
1995 | int error; |
1996 | ||
5e2f7d61 BP |
1997 | error = gfs2_rindex_update(sdp); |
1998 | if (error) | |
1999 | return error; | |
2000 | ||
b3b94faa DT |
2001 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list)); |
2002 | ||
8be04b93 | 2003 | ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN); |
e8830d88 | 2004 | if (ht == NULL) |
7456a37d OD |
2005 | ht = __vmalloc(size, GFP_NOFS | __GFP_NOWARN | __GFP_ZERO, |
2006 | PAGE_KERNEL); | |
b3b94faa DT |
2007 | if (!ht) |
2008 | return -ENOMEM; | |
2009 | ||
f4108a60 | 2010 | error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE); |
b3b94faa | 2011 | if (error) |
5407e242 | 2012 | goto out; |
b3b94faa | 2013 | |
b3b94faa | 2014 | /* Count the number of leaves */ |
ec038c82 | 2015 | bh = leaf_bh; |
b3b94faa | 2016 | |
c752666c | 2017 | for (blk = leaf_no; blk; blk = nblk) { |
ec038c82 BP |
2018 | if (blk != leaf_no) { |
2019 | error = get_leaf(dip, blk, &bh); | |
2020 | if (error) | |
2021 | goto out_rlist; | |
2022 | } | |
c752666c SW |
2023 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
2024 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
ec038c82 BP |
2025 | if (blk != leaf_no) |
2026 | brelse(bh); | |
b3b94faa | 2027 | |
70b0c365 | 2028 | gfs2_rlist_add(dip, &rlist, blk); |
b3b94faa DT |
2029 | l_blocks++; |
2030 | } | |
2031 | ||
fe6c991c | 2032 | gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE); |
b3b94faa DT |
2033 | |
2034 | for (x = 0; x < rlist.rl_rgrps; x++) { | |
6f6597ba AG |
2035 | struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(rlist.rl_ghs[x].gh_gl); |
2036 | ||
bb8d8a6f | 2037 | rg_blocks += rgd->rd_length; |
b3b94faa DT |
2038 | } |
2039 | ||
2040 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs); | |
2041 | if (error) | |
2042 | goto out_rlist; | |
2043 | ||
2044 | error = gfs2_trans_begin(sdp, | |
5c676f6d | 2045 | rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) + |
b3b94faa DT |
2046 | RES_DINODE + RES_STATFS + RES_QUOTA, l_blocks); |
2047 | if (error) | |
2048 | goto out_rg_gunlock; | |
2049 | ||
ec038c82 BP |
2050 | bh = leaf_bh; |
2051 | ||
c752666c | 2052 | for (blk = leaf_no; blk; blk = nblk) { |
ec038c82 BP |
2053 | if (blk != leaf_no) { |
2054 | error = get_leaf(dip, blk, &bh); | |
2055 | if (error) | |
2056 | goto out_end_trans; | |
2057 | } | |
c752666c SW |
2058 | tmp_leaf = (struct gfs2_leaf *)bh->b_data; |
2059 | nblk = be64_to_cpu(tmp_leaf->lf_next); | |
ec038c82 BP |
2060 | if (blk != leaf_no) |
2061 | brelse(bh); | |
b3b94faa DT |
2062 | |
2063 | gfs2_free_meta(dip, blk, 1); | |
77658aad | 2064 | gfs2_add_inode_blocks(&dip->i_inode, -1); |
b3b94faa DT |
2065 | } |
2066 | ||
cd915493 | 2067 | error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size); |
b3b94faa DT |
2068 | if (error != size) { |
2069 | if (error >= 0) | |
2070 | error = -EIO; | |
2071 | goto out_end_trans; | |
2072 | } | |
2073 | ||
2074 | error = gfs2_meta_inode_buffer(dip, &dibh); | |
2075 | if (error) | |
2076 | goto out_end_trans; | |
2077 | ||
350a9b0a | 2078 | gfs2_trans_add_meta(dip->i_gl, dibh); |
d24a7a43 BP |
2079 | /* On the last dealloc, make this a regular file in case we crash. |
2080 | (We don't want to free these blocks a second time.) */ | |
2081 | if (last_dealloc) | |
2082 | dip->i_inode.i_mode = S_IFREG; | |
539e5d6b | 2083 | gfs2_dinode_out(dip, dibh->b_data); |
b3b94faa DT |
2084 | brelse(dibh); |
2085 | ||
a91ea69f | 2086 | out_end_trans: |
b3b94faa | 2087 | gfs2_trans_end(sdp); |
a91ea69f | 2088 | out_rg_gunlock: |
b3b94faa | 2089 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs); |
a91ea69f | 2090 | out_rlist: |
b3b94faa | 2091 | gfs2_rlist_free(&rlist); |
b3b94faa | 2092 | gfs2_quota_unhold(dip); |
182fe5ab | 2093 | out: |
3cdcf63e | 2094 | kvfree(ht); |
b3b94faa DT |
2095 | return error; |
2096 | } | |
2097 | ||
2098 | /** | |
2099 | * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory | |
2100 | * @dip: the directory | |
2101 | * | |
2102 | * Dealloc all on-disk directory leaves to FREEMETA state | |
2103 | * Change on-disk inode type to "regular file" | |
2104 | * | |
2105 | * Returns: errno | |
2106 | */ | |
2107 | ||
2108 | int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip) | |
2109 | { | |
556bb179 BP |
2110 | struct buffer_head *bh; |
2111 | struct gfs2_leaf *leaf; | |
2112 | u32 hsize, len; | |
556bb179 BP |
2113 | u32 index = 0, next_index; |
2114 | __be64 *lp; | |
2115 | u64 leaf_no; | |
2116 | int error = 0, last; | |
2117 | ||
47a9a527 | 2118 | hsize = BIT(dip->i_depth); |
556bb179 | 2119 | |
17d539f0 SW |
2120 | lp = gfs2_dir_get_hash_table(dip); |
2121 | if (IS_ERR(lp)) | |
2122 | return PTR_ERR(lp); | |
556bb179 BP |
2123 | |
2124 | while (index < hsize) { | |
17d539f0 | 2125 | leaf_no = be64_to_cpu(lp[index]); |
556bb179 BP |
2126 | if (leaf_no) { |
2127 | error = get_leaf(dip, leaf_no, &bh); | |
2128 | if (error) | |
2129 | goto out; | |
2130 | leaf = (struct gfs2_leaf *)bh->b_data; | |
47a9a527 | 2131 | len = BIT(dip->i_depth - be16_to_cpu(leaf->lf_depth)); |
556bb179 BP |
2132 | |
2133 | next_index = (index & ~(len - 1)) + len; | |
2134 | last = ((next_index >= hsize) ? 1 : 0); | |
2135 | error = leaf_dealloc(dip, index, len, leaf_no, bh, | |
2136 | last); | |
2137 | brelse(bh); | |
2138 | if (error) | |
2139 | goto out; | |
2140 | index = next_index; | |
2141 | } else | |
2142 | index++; | |
2143 | } | |
2144 | ||
2145 | if (index != hsize) { | |
2146 | gfs2_consist_inode(dip); | |
2147 | error = -EIO; | |
2148 | } | |
2149 | ||
2150 | out: | |
556bb179 BP |
2151 | |
2152 | return error; | |
b3b94faa DT |
2153 | } |
2154 | ||
2155 | /** | |
2156 | * gfs2_diradd_alloc_required - find if adding entry will require an allocation | |
2157 | * @ip: the file being written to | |
2158 | * @filname: the filename that's going to be added | |
3c1c0ae1 | 2159 | * @da: The structure to return dir alloc info |
b3b94faa | 2160 | * |
3c1c0ae1 | 2161 | * Returns: 0 if ok, -ve on error |
b3b94faa DT |
2162 | */ |
2163 | ||
3c1c0ae1 SW |
2164 | int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name, |
2165 | struct gfs2_diradd *da) | |
b3b94faa | 2166 | { |
22b5a6c0 | 2167 | struct gfs2_inode *ip = GFS2_I(inode); |
3c1c0ae1 | 2168 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
22b5a6c0 | 2169 | const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf); |
c752666c SW |
2170 | struct gfs2_dirent *dent; |
2171 | struct buffer_head *bh; | |
b3b94faa | 2172 | |
3c1c0ae1 | 2173 | da->nr_blocks = 0; |
2b47dad8 SW |
2174 | da->bh = NULL; |
2175 | da->dent = NULL; | |
3c1c0ae1 | 2176 | |
c752666c | 2177 | dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh); |
ed386507 | 2178 | if (!dent) { |
3c1c0ae1 | 2179 | da->nr_blocks = sdp->sd_max_dirres; |
22b5a6c0 SW |
2180 | if (!(ip->i_diskflags & GFS2_DIF_EXHASH) && |
2181 | (GFS2_DIRENT_SIZE(name->len) < extra)) | |
2182 | da->nr_blocks = 1; | |
3c1c0ae1 | 2183 | return 0; |
ed386507 | 2184 | } |
c752666c SW |
2185 | if (IS_ERR(dent)) |
2186 | return PTR_ERR(dent); | |
19aeb5a6 BP |
2187 | |
2188 | if (da->save_loc) { | |
2189 | da->bh = bh; | |
2190 | da->dent = dent; | |
2191 | } else { | |
2192 | brelse(bh); | |
2193 | } | |
c752666c | 2194 | return 0; |
b3b94faa DT |
2195 | } |
2196 |