| 1 | |
| 2 | |
| 3 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
|
| 4 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
|
| 5 |
|
| 6 |
|
| 7 | |
| 8 | * Implements Extendible Hashing as described in:
|
| 9 | * "Extendible Hashing" by Fagin, et al in
|
| 10 | * __ACM Trans. on Database Systems__, Sept 1979.
|
| 11 | *
|
| 12 | *
|
| 13 | * Here's the layout of dirents which is essentially the same as that of ext2
|
| 14 | * within a single block. The field de_name_len is the number of bytes
|
| 15 | * actually required for the name (no null terminator). The field de_rec_len
|
| 16 | * is the number of bytes allocated to the dirent. The offset of the next
|
| 17 | * dirent in the block is (dirent + dirent->de_rec_len). When a dirent is
|
| 18 | * deleted, the preceding dirent inherits its allocated space, ie
|
| 19 | * prev->de_rec_len += deleted->de_rec_len. Since the next dirent is obtained
|
| 20 | * by adding de_rec_len to the current dirent, this essentially causes the
|
| 21 | * deleted dirent to get jumped over when iterating through all the dirents.
|
| 22 | *
|
| 23 | * When deleting the first dirent in a block, there is no previous dirent so
|
| 24 | * the field de_ino is set to zero to designate it as deleted. When allocating
|
| 25 | * a dirent, gfs2_dirent_alloc iterates through the dirents in a block. If the
|
| 26 | * first dirent has (de_ino == 0) and de_rec_len is large enough, this first
|
| 27 | * dirent is allocated. Otherwise it must go through all the 'used' dirents
|
| 28 | * searching for one in which the amount of total space minus the amount of
|
| 29 | * used space will provide enough space for the new dirent.
|
| 30 | *
|
| 31 | * There are two types of blocks in which dirents reside. In a stuffed dinode,
|
| 32 | * the dirents begin at offset sizeof(struct gfs2_dinode) from the beginning of
|
| 33 | * the block. In leaves, they begin at offset sizeof(struct gfs2_leaf) from the
|
| 34 | * beginning of the leaf block. The dirents reside in leaves when
|
| 35 | *
|
| 36 | * dip->i_diskflags & GFS2_DIF_EXHASH is true
|
| 37 | *
|
| 38 | * Otherwise, the dirents are "linear", within a single stuffed dinode block.
|
| 39 | *
|
| 40 | * When the dirents are in leaves, the actual contents of the directory file are
|
| 41 | * used as an array of 64-bit block pointers pointing to the leaf blocks. The
|
| 42 | * dirents are NOT in the directory file itself. There can be more than one
|
| 43 | * block pointer in the array that points to the same leaf. In fact, when a
|
| 44 | * directory is first converted from linear to exhash, all of the pointers
|
| 45 | * point to the same leaf.
|
| 46 | *
|
| 47 | * When a leaf is completely full, the size of the hash table can be
|
| 48 | * doubled unless it is already at the maximum size which is hard coded into
|
| 49 | * GFS2_DIR_MAX_DEPTH. After that, leaves are chained together in a linked list,
|
| 50 | * but never before the maximum hash table size has been reached.
|
| 51 |
|
| 52 |
|
| 53 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
|
| 54 |
|
| 55 | #include <linux/slab.h>
|
| 56 | #include <linux/spinlock.h>
|
| 57 | #include <linux/buffer_head.h>
|
| 58 | #include <linux/sort.h>
|
| 59 | #include <linux/gfs2_ondisk.h>
|
| 60 | #include <linux/crc32.h>
|
| 61 | #include <linux/vmalloc.h>
|
| 62 | #include <linux/bio.h>
|
| 63 | #include <linux/log2.h>
|
| 64 |
|
| 65 | #include "gfs2.h"
|
| 66 | #include "incore.h"
|
| 67 | #include "dir.h"
|
| 68 | #include "glock.h"
|
| 69 | #include "inode.h"
|
| 70 | #include "meta_io.h"
|
| 71 | #include "quota.h"
|
| 72 | #include "rgrp.h"
|
| 73 | #include "trans.h"
|
| 74 | #include "bmap.h"
|
| 75 | #include "util.h"
|
| 76 |
|
| 77 | #define MAX_RA_BLOCKS 32
|
| 78 |
|
| 79 | #define gfs2_disk_hash2offset(h) (((u64)(h)) >> 1)
|
| 80 | #define gfs2_dir_offset2hash(p) ((u32)(((u64)(p)) << 1))
|
| 81 | #define GFS2_HASH_INDEX_MASK 0xffffc000
|
| 82 | #define GFS2_USE_HASH_FLAG 0x2000
|
| 83 |
|
| 84 | struct qstr gfs2_qdot __read_mostly;
|
| 85 | struct qstr gfs2_qdotdot __read_mostly;
|
| 86 |
|
| 87 | typedef int (*gfs2_dscan_t)(const struct gfs2_dirent *dent,
|
| 88 | const struct qstr *name, void *opaque);
|
| 89 |
|
| 90 | int gfs2_dir_get_new_buffer(struct gfs2_inode *ip, u64 block,
|
| 91 | struct buffer_head **bhp)
|
| 92 | {
|
| 93 | struct gfs2_glock *gl = gfs2_inode_glock(&ip->i_inode);
|
| 94 | struct buffer_head *bh;
|
| 95 |
|
| 96 | bh = gfs2_meta_new(gl, block);
|
| 97 | gfs2_trans_add_meta(gl, bh);
|
| 98 | gfs2_metatype_set(bh, GFS2_METATYPE_JD, GFS2_FORMAT_JD);
|
| 99 | gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
|
| 100 | *bhp = bh;
|
| 101 | return 0;
|
| 102 | }
|
| 103 |
|
| 104 | static int gfs2_dir_get_existing_buffer(struct gfs2_inode *ip, u64 block,
|
| 105 | struct buffer_head **bhp)
|
| 106 | {
|
| 107 | struct gfs2_glock *gl = gfs2_inode_glock(&ip->i_inode);
|
| 108 | struct buffer_head *bh;
|
| 109 | int error;
|
| 110 |
|
| 111 | error = gfs2_meta_read(gl, block, DIO_WAIT, 0, &bh);
|
| 112 | if (error)
|
| 113 | return error;
|
| 114 | if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh, GFS2_METATYPE_JD)) {
|
| 115 | brelse(bh);
|
| 116 | return -EIO;
|
| 117 | }
|
| 118 | *bhp = bh;
|
| 119 | return 0;
|
| 120 | }
|
| 121 |
|
| 122 | static int gfs2_dir_write_stuffed(struct gfs2_inode *ip, const char *buf,
|
| 123 | unsigned int offset, unsigned int size)
|
| 124 | {
|
| 125 | struct gfs2_glock *gl = gfs2_inode_glock(&ip->i_inode);
|
| 126 | struct buffer_head *dibh;
|
| 127 | int error;
|
| 128 |
|
| 129 | error = gfs2_meta_inode_buffer(ip, &dibh);
|
| 130 | if (error)
|
| 131 | return error;
|
| 132 |
|
| 133 | gfs2_trans_add_meta(gl, dibh);
|
| 134 | memcpy(dibh->b_data + offset + sizeof(struct gfs2_dinode), buf, size);
|
| 135 | if (ip->i_inode.i_size < offset + size)
|
| 136 | i_size_write(&ip->i_inode, offset + size);
|
| 137 | inode_set_mtime_to_ts(&ip->i_inode, inode_set_ctime_current(&ip->i_inode));
|
| 138 | gfs2_dinode_out(ip, dibh->b_data);
|
| 139 |
|
| 140 | brelse(dibh);
|
| 141 |
|
| 142 | return size;
|
| 143 | }
|
| 144 |
|
| 145 |
|
| 146 |
|
| 147 | |
| 148 | * gfs2_dir_write_data - Write directory information to the inode
|
| 149 | * @ip: The GFS2 inode
|
| 150 | * @buf: The buffer containing information to be written
|
| 151 | * @offset: The file offset to start writing at
|
| 152 | * @size: The amount of data to write
|
| 153 | *
|
| 154 | * Returns: The number of bytes correctly written or error code
|
| 155 |
|
| 156 | static int gfs2_dir_write_data(struct gfs2_inode *ip, const char *buf,
|
| 157 | u64 offset, unsigned int size)
|
| 158 | {
|
| 159 | struct gfs2_glock *gl = gfs2_inode_glock(&ip->i_inode);
|
| 160 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
|
| 161 | struct buffer_head *dibh;
|
| 162 | u64 lblock, dblock;
|
| 163 | u32 extlen = 0;
|
| 164 | unsigned int o;
|
| 165 | int copied = 0;
|
| 166 | int error = 0;
|
| 167 | bool new = false;
|
| 168 |
|
| 169 | if (!size)
|
| 170 | return 0;
|
| 171 |
|
| 172 | if (gfs2_is_stuffed(ip) && offset + size <= gfs2_max_stuffed_size(ip))
|
| 173 | return gfs2_dir_write_stuffed(ip, buf, (unsigned int)offset,
|
| 174 | size);
|
| 175 |
|
| 176 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
|
| 177 | return -EINVAL;
|
| 178 |
|
| 179 | if (gfs2_is_stuffed(ip)) {
|
| 180 | error = gfs2_unstuff_dinode(ip);
|
| 181 | if (error)
|
| 182 | return error;
|
| 183 | }
|
| 184 |
|
| 185 | lblock = offset;
|
| 186 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
|
| 187 |
|
| 188 | while (copied < size) {
|
| 189 | unsigned int amount;
|
| 190 | struct buffer_head *bh;
|
| 191 |
|
| 192 | amount = size - copied;
|
| 193 | if (amount > sdp->sd_sb.sb_bsize - o)
|
| 194 | amount = sdp->sd_sb.sb_bsize - o;
|
| 195 |
|
| 196 | if (!extlen) {
|
| 197 | extlen = 1;
|
| 198 | error = gfs2_alloc_extent(&ip->i_inode, lblock, &dblock,
|
| 199 | &extlen, &new);
|
| 200 | if (error)
|
| 201 | goto fail;
|
| 202 | error = -EIO;
|
| 203 | if (gfs2_assert_withdraw(sdp, dblock))
|
| 204 | goto fail;
|
| 205 | }
|
| 206 |
|
| 207 | if (amount == sdp->sd_jbsize || new)
|
| 208 | error = gfs2_dir_get_new_buffer(ip, dblock, &bh);
|
| 209 | else
|
| 210 | error = gfs2_dir_get_existing_buffer(ip, dblock, &bh);
|
| 211 |
|
| 212 | if (error)
|
| 213 | goto fail;
|
| 214 |
|
| 215 | gfs2_trans_add_meta(gl, bh);
|
| 216 | memcpy(bh->b_data + o, buf, amount);
|
| 217 | brelse(bh);
|
| 218 |
|
| 219 | buf += amount;
|
| 220 | copied += amount;
|
| 221 | lblock++;
|
| 222 | dblock++;
|
| 223 | extlen--;
|
| 224 |
|
| 225 | o = sizeof(struct gfs2_meta_header);
|
| 226 | }
|
| 227 |
|
| 228 | out:
|
| 229 | error = gfs2_meta_inode_buffer(ip, &dibh);
|
| 230 | if (error)
|
| 231 | return error;
|
| 232 |
|
| 233 | if (ip->i_inode.i_size < offset + copied)
|
| 234 | i_size_write(&ip->i_inode, offset + copied);
|
| 235 | inode_set_mtime_to_ts(&ip->i_inode, inode_set_ctime_current(&ip->i_inode));
|
| 236 |
|
| 237 | gfs2_trans_add_meta(gl, dibh);
|
| 238 | gfs2_dinode_out(ip, dibh->b_data);
|
| 239 | brelse(dibh);
|
| 240 |
|
| 241 | return copied;
|
| 242 | fail:
|
| 243 | if (copied)
|
| 244 | goto out;
|
| 245 | return error;
|
| 246 | }
|
| 247 |
|
| 248 | static int gfs2_dir_read_stuffed(struct gfs2_inode *ip, __be64 *buf,
|
| 249 | unsigned int size)
|
| 250 | {
|
| 251 | struct buffer_head *dibh;
|
| 252 | int error;
|
| 253 |
|
| 254 | error = gfs2_meta_inode_buffer(ip, &dibh);
|
| 255 | if (!error) {
|
| 256 | memcpy(buf, dibh->b_data + sizeof(struct gfs2_dinode), size);
|
| 257 | brelse(dibh);
|
| 258 | }
|
| 259 |
|
| 260 | return (error) ? error : size;
|
| 261 | }
|
| 262 |
|
| 263 |
|
| 264 | |
| 265 | * gfs2_dir_read_data - Read a data from a directory inode
|
| 266 | * @ip: The GFS2 Inode
|
| 267 | * @buf: The buffer to place result into
|
| 268 | * @size: Amount of data to transfer
|
| 269 | *
|
| 270 | * Returns: The amount of data actually copied or the error
|
| 271 |
|
| 272 | static int gfs2_dir_read_data(struct gfs2_inode *ip, __be64 *buf,
|
| 273 | unsigned int size)
|
| 274 | {
|
| 275 | struct gfs2_glock *gl = gfs2_inode_glock(&ip->i_inode);
|
| 276 | struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
|
| 277 | u64 lblock, dblock;
|
| 278 | u32 extlen = 0;
|
| 279 | unsigned int o;
|
| 280 | int copied = 0;
|
| 281 | int error = 0;
|
| 282 |
|
| 283 | if (gfs2_is_stuffed(ip))
|
| 284 | return gfs2_dir_read_stuffed(ip, buf, size);
|
| 285 |
|
| 286 | if (gfs2_assert_warn(sdp, gfs2_is_jdata(ip)))
|
| 287 | return -EINVAL;
|
| 288 |
|
| 289 | lblock = 0;
|
| 290 | o = do_div(lblock, sdp->sd_jbsize) + sizeof(struct gfs2_meta_header);
|
| 291 |
|
| 292 | while (copied < size) {
|
| 293 | unsigned int amount;
|
| 294 | struct buffer_head *bh;
|
| 295 |
|
| 296 | amount = size - copied;
|
| 297 | if (amount > sdp->sd_sb.sb_bsize - o)
|
| 298 | amount = sdp->sd_sb.sb_bsize - o;
|
| 299 |
|
| 300 | if (!extlen) {
|
| 301 | extlen = 32;
|
| 302 | error = gfs2_get_extent(&ip->i_inode, lblock,
|
| 303 | &dblock, &extlen);
|
| 304 | if (error || !dblock)
|
| 305 | goto fail;
|
| 306 | BUG_ON(extlen < 1);
|
| 307 | bh = gfs2_meta_ra(gl, dblock, extlen);
|
| 308 | } else {
|
| 309 | error = gfs2_meta_read(gl, dblock, DIO_WAIT, 0, &bh);
|
| 310 | if (error)
|
| 311 | goto fail;
|
| 312 | }
|
| 313 | error = gfs2_metatype_check(sdp, bh, GFS2_METATYPE_JD);
|
| 314 | if (error) {
|
| 315 | brelse(bh);
|
| 316 | goto fail;
|
| 317 | }
|
| 318 | dblock++;
|
| 319 | extlen--;
|
| 320 | memcpy(buf, bh->b_data + o, amount);
|
| 321 | brelse(bh);
|
| 322 | buf += (amount/sizeof(__be64));
|
| 323 | copied += amount;
|
| 324 | lblock++;
|
| 325 | o = sizeof(struct gfs2_meta_header);
|
| 326 | }
|
| 327 |
|
| 328 | return copied;
|
| 329 | fail:
|
| 330 | return (copied) ? copied : error;
|
| 331 | }
|
| 332 |
|
| 333 | |
| 334 | * gfs2_dir_get_hash_table - Get pointer to the dir hash table
|
| 335 | * @ip: The inode in question
|
| 336 | *
|
| 337 | * Returns: The hash table or an error
|
| 338 |
|
| 339 |
|
| 340 | static __be64 *gfs2_dir_get_hash_table(struct gfs2_inode *ip)
|
| 341 | {
|
| 342 | struct inode *inode = &ip->i_inode;
|
| 343 | int ret;
|
| 344 | u32 hsize;
|
| 345 | __be64 *hc;
|
| 346 |
|
| 347 | BUG_ON(!(ip->i_diskflags & GFS2_DIF_EXHASH));
|
| 348 |
|
| 349 | hc = ip->i_hash_cache;
|
| 350 | if (hc)
|
| 351 | return hc;
|
| 352 |
|
| 353 | hsize = BIT(ip->i_depth);
|
| 354 | hsize *= sizeof(__be64);
|
| 355 | if (hsize != i_size_read(&ip->i_inode)) {
|
| 356 | gfs2_consist_inode(ip);
|
| 357 | return ERR_PTR(-EIO);
|
| 358 | }
|
| 359 |
|
| 360 | hc = kmalloc(hsize, GFP_NOFS | __GFP_NOWARN);
|
| 361 | if (hc == NULL)
|
| 362 | hc = __vmalloc(hsize, GFP_NOFS);
|
| 363 |
|
| 364 | if (hc == NULL)
|
| 365 | return ERR_PTR(-ENOMEM);
|
| 366 |
|
| 367 | ret = gfs2_dir_read_data(ip, hc, hsize);
|
| 368 | if (ret < 0) {
|
| 369 | kvfree(hc);
|
| 370 | return ERR_PTR(ret);
|
| 371 | }
|
| 372 |
|
| 373 | spin_lock(&inode->i_lock);
|
| 374 | if (likely(!ip->i_hash_cache)) {
|
| 375 | ip->i_hash_cache = hc;
|
| 376 | hc = NULL;
|
| 377 | }
|
| 378 | spin_unlock(&inode->i_lock);
|
| 379 | kvfree(hc);
|
| 380 |
|
| 381 | return ip->i_hash_cache;
|
| 382 | }
|
| 383 |
|
| 384 | |
| 385 | * gfs2_dir_hash_inval - Invalidate dir hash
|
| 386 | * @ip: The directory inode
|
| 387 | *
|
| 388 | * Must be called with an exclusive glock, or during glock invalidation.
|
| 389 |
|
| 390 | void gfs2_dir_hash_inval(struct gfs2_inode *ip)
|
| 391 | {
|
| 392 | __be64 *hc;
|
| 393 |
|
| 394 | spin_lock(&ip->i_inode.i_lock);
|
| 395 | hc = ip->i_hash_cache;
|
| 396 | ip->i_hash_cache = NULL;
|
| 397 | spin_unlock(&ip->i_inode.i_lock);
|
| 398 |
|
| 399 | kvfree(hc);
|
| 400 | }
|
| 401 |
|
| 402 | static inline int gfs2_dirent_sentinel(const struct gfs2_dirent *dent)
|
| 403 | {
|
| 404 | return dent->de_inum.no_addr == 0 || dent->de_inum.no_formal_ino == 0;
|
| 405 | }
|
| 406 |
|
| 407 | static inline int __gfs2_dirent_find(const struct gfs2_dirent *dent,
|
| 408 | const struct qstr *name, int ret)
|
| 409 | {
|
| 410 | if (!gfs2_dirent_sentinel(dent) &&
|
| 411 | be32_to_cpu(dent->de_hash) == name->hash &&
|
| 412 | be16_to_cpu(dent->de_name_len) == name->len &&
|
| 413 | memcmp(dent+1, name->name, name->len) == 0)
|
| 414 | return ret;
|
| 415 | return 0;
|
| 416 | }
|
| 417 |
|
| 418 | static int gfs2_dirent_find(const struct gfs2_dirent *dent,
|
| 419 | const struct qstr *name,
|
| 420 | void *opaque)
|
| 421 | {
|
| 422 | return __gfs2_dirent_find(dent, name, 1);
|
| 423 | }
|
| 424 |
|
| 425 | static int gfs2_dirent_prev(const struct gfs2_dirent *dent,
|
| 426 | const struct qstr *name,
|
| 427 | void *opaque)
|
| 428 | {
|
| 429 | return __gfs2_dirent_find(dent, name, 2);
|
| 430 | }
|
| 431 |
|
| 432 | |
| 433 | * name->name holds ptr to start of block.
|
| 434 | * name->len holds size of block.
|
| 435 |
|
| 436 | static int gfs2_dirent_last(const struct gfs2_dirent *dent,
|
| 437 | const struct qstr *name,
|
| 438 | void *opaque)
|
| 439 | {
|
| 440 | const char *start = name->name;
|
| 441 | const char *end = (const char *)dent + be16_to_cpu(dent->de_rec_len);
|
| 442 | if (name->len == (end - start))
|
| 443 | return 1;
|
| 444 | return 0;
|
| 445 | }
|
| 446 |
|
| 447 | |
| 448 | * find that dirent, there must be space available there for the new dirent
|
| 449 | static int gfs2_dirent_find_offset(const struct gfs2_dirent *dent,
|
| 450 | const struct qstr *name,
|
| 451 | void *ptr)
|
| 452 | {
|
| 453 | unsigned required = GFS2_DIRENT_SIZE(name->len);
|
| 454 | unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
|
| 455 | unsigned totlen = be16_to_cpu(dent->de_rec_len);
|
| 456 |
|
| 457 | if (ptr < (void *)dent || ptr >= (void *)dent + totlen)
|
| 458 | return 0;
|
| 459 | if (gfs2_dirent_sentinel(dent))
|
| 460 | actual = 0;
|
| 461 | if (ptr < (void *)dent + actual)
|
| 462 | return -1;
|
| 463 | if ((void *)dent + totlen >= ptr + required)
|
| 464 | return 1;
|
| 465 | return -1;
|
| 466 | }
|
| 467 |
|
| 468 | static int gfs2_dirent_find_space(const struct gfs2_dirent *dent,
|
| 469 | const struct qstr *name,
|
| 470 | void *opaque)
|
| 471 | {
|
| 472 | unsigned required = GFS2_DIRENT_SIZE(name->len);
|
| 473 | unsigned actual = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
|
| 474 | unsigned totlen = be16_to_cpu(dent->de_rec_len);
|
| 475 |
|
| 476 | if (gfs2_dirent_sentinel(dent))
|
| 477 | actual = 0;
|
| 478 | if (totlen - actual >= required)
|
| 479 | return 1;
|
| 480 | return 0;
|
| 481 | }
|
| 482 |
|
| 483 | struct dirent_gather {
|
| 484 | const struct gfs2_dirent **pdent;
|
| 485 | unsigned offset;
|
| 486 | };
|
| 487 |
|
| 488 | static int gfs2_dirent_gather(const struct gfs2_dirent *dent,
|
| 489 | const struct qstr *name,
|
| 490 | void *opaque)
|
| 491 | {
|
| 492 | struct dirent_gather *g = opaque;
|
| 493 | if (!gfs2_dirent_sentinel(dent)) {
|
| 494 | g->pdent[g->offset++] = dent;
|
| 495 | }
|
| 496 | return 0;
|
| 497 | }
|
| 498 |
|
| 499 | |
| 500 | * Other possible things to check:
|
| 501 | * - Inode located within filesystem size (and on valid block)
|
| 502 | * - Valid directory entry type
|
| 503 | * Not sure how heavy-weight we want to make this... could also check
|
| 504 | * hash is correct for example, but that would take a lot of extra time.
|
| 505 | * For now the most important thing is to check that the various sizes
|
| 506 | * are correct.
|
| 507 |
|
| 508 | static int gfs2_check_dirent(struct gfs2_sbd *sdp,
|
| 509 | struct gfs2_dirent *dent, unsigned int offset,
|
| 510 | unsigned int size, unsigned int len, int first)
|
| 511 | {
|
| 512 | const char *msg = "gfs2_dirent too small";
|
| 513 | if (unlikely(size < sizeof(struct gfs2_dirent)))
|
| 514 | goto error;
|
| 515 | msg = "gfs2_dirent misaligned";
|
| 516 | if (unlikely(offset & 0x7))
|
| 517 | goto error;
|
| 518 | msg = "gfs2_dirent points beyond end of block";
|
| 519 | if (unlikely(offset + size > len))
|
| 520 | goto error;
|
| 521 | msg = "zero inode number";
|
| 522 | if (unlikely(!first && gfs2_dirent_sentinel(dent)))
|
| 523 | goto error;
|
| 524 | msg = "name length is greater than space in dirent";
|
| 525 | if (!gfs2_dirent_sentinel(dent) &&
|
| 526 | unlikely(sizeof(struct gfs2_dirent)+be16_to_cpu(dent->de_name_len) >
|
| 527 | size))
|
| 528 | goto error;
|
| 529 | return 0;
|
| 530 | error:
|
| 531 | fs_warn(sdp, "%s: %s (%s)\n",
|
| 532 | __func__, msg, first ? "first in block" : "not first in block");
|
| 533 | return -EIO;
|
| 534 | }
|
| 535 |
|
| 536 | static int gfs2_dirent_offset(struct gfs2_sbd *sdp, const void *buf)
|
| 537 | {
|
| 538 | const struct gfs2_meta_header *h = buf;
|
| 539 | int offset;
|
| 540 |
|
| 541 | BUG_ON(buf == NULL);
|
| 542 |
|
| 543 | switch(be32_to_cpu(h->mh_type)) {
|
| 544 | case GFS2_METATYPE_LF:
|
| 545 | offset = sizeof(struct gfs2_leaf);
|
| 546 | break;
|
| 547 | case GFS2_METATYPE_DI:
|
| 548 | offset = sizeof(struct gfs2_dinode);
|
| 549 | break;
|
| 550 | default:
|
| 551 | goto wrong_type;
|
| 552 | }
|
| 553 | return offset;
|
| 554 | wrong_type:
|
| 555 | fs_warn(sdp, "%s: wrong block type %u\n", __func__,
|
| 556 | be32_to_cpu(h->mh_type));
|
| 557 | return -1;
|
| 558 | }
|
| 559 |
|
| 560 | static struct gfs2_dirent *gfs2_dirent_scan(struct inode *inode, void *buf,
|
| 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;
|
| 569 |
|
| 570 | ret = gfs2_dirent_offset(GFS2_SB(inode), buf);
|
| 571 | if (ret < 0) {
|
| 572 | gfs2_consist_inode(GFS2_I(inode));
|
| 573 | return ERR_PTR(-EIO);
|
| 574 | }
|
| 575 | offset = ret;
|
| 576 | prev = NULL;
|
| 577 | dent = buf + offset;
|
| 578 | size = be16_to_cpu(dent->de_rec_len);
|
| 579 | if (gfs2_check_dirent(GFS2_SB(inode), dent, offset, size, len, 1)) {
|
| 580 | gfs2_consist_inode(GFS2_I(inode));
|
| 581 | return ERR_PTR(-EIO);
|
| 582 | }
|
| 583 | do {
|
| 584 | ret = scan(dent, name, opaque);
|
| 585 | if (ret)
|
| 586 | break;
|
| 587 | offset += size;
|
| 588 | if (offset == len)
|
| 589 | break;
|
| 590 | prev = dent;
|
| 591 | dent = buf + offset;
|
| 592 | size = be16_to_cpu(dent->de_rec_len);
|
| 593 | if (gfs2_check_dirent(GFS2_SB(inode), dent, offset, size,
|
| 594 | len, 0)) {
|
| 595 | gfs2_consist_inode(GFS2_I(inode));
|
| 596 | return ERR_PTR(-EIO);
|
| 597 | }
|
| 598 | } while(1);
|
| 599 |
|
| 600 | switch(ret) {
|
| 601 | case 0:
|
| 602 | return NULL;
|
| 603 | case 1:
|
| 604 | return dent;
|
| 605 | case 2:
|
| 606 | return prev ? prev : dent;
|
| 607 | default:
|
| 608 | BUG_ON(ret > 0);
|
| 609 | return ERR_PTR(ret);
|
| 610 | }
|
| 611 | }
|
| 612 |
|
| 613 | static int dirent_check_reclen(struct gfs2_inode *dip,
|
| 614 | const struct gfs2_dirent *d, const void *end_p)
|
| 615 | {
|
| 616 | const void *ptr = d;
|
| 617 | u16 rec_len = be16_to_cpu(d->de_rec_len);
|
| 618 |
|
| 619 | if (unlikely(rec_len < sizeof(struct gfs2_dirent))) {
|
| 620 | gfs2_consist_inode(dip);
|
| 621 | return -EIO;
|
| 622 | }
|
| 623 | ptr += rec_len;
|
| 624 | if (ptr < end_p)
|
| 625 | return rec_len;
|
| 626 | if (ptr == end_p)
|
| 627 | return -ENOENT;
|
| 628 |
|
| 629 | gfs2_consist_inode(dip);
|
| 630 | return -EIO;
|
| 631 | }
|
| 632 |
|
| 633 | |
| 634 | * dirent_next - Next dirent
|
| 635 | * @dip: the directory
|
| 636 | * @bh: The buffer
|
| 637 | * @dent: Pointer to list of dirents
|
| 638 | *
|
| 639 | * Returns: 0 on success, error code otherwise
|
| 640 |
|
| 641 |
|
| 642 | static int dirent_next(struct gfs2_inode *dip, struct buffer_head *bh,
|
| 643 | struct gfs2_dirent **dent)
|
| 644 | {
|
| 645 | struct gfs2_dirent *cur = *dent, *tmp;
|
| 646 | char *bh_end = bh->b_data + bh->b_size;
|
| 647 | int ret;
|
| 648 |
|
| 649 | ret = dirent_check_reclen(dip, cur, bh_end);
|
| 650 | if (ret < 0)
|
| 651 | return ret;
|
| 652 |
|
| 653 | tmp = (void *)cur + ret;
|
| 654 | ret = dirent_check_reclen(dip, tmp, bh_end);
|
| 655 | if (ret == -EIO)
|
| 656 | return ret;
|
| 657 |
|
| 658 |
|
| 659 | if (gfs2_dirent_sentinel(tmp)) {
|
| 660 | gfs2_consist_inode(dip);
|
| 661 | return -EIO;
|
| 662 | }
|
| 663 |
|
| 664 | *dent = tmp;
|
| 665 | return 0;
|
| 666 | }
|
| 667 |
|
| 668 | |
| 669 | * dirent_del - Delete a dirent
|
| 670 | * @dip: The GFS2 inode
|
| 671 | * @bh: The buffer
|
| 672 | * @prev: The previous dirent
|
| 673 | * @cur: The current dirent
|
| 674 | *
|
| 675 |
|
| 676 |
|
| 677 | static void dirent_del(struct gfs2_inode *dip, struct buffer_head *bh,
|
| 678 | struct gfs2_dirent *prev, struct gfs2_dirent *cur)
|
| 679 | {
|
| 680 | struct gfs2_glock *gl = gfs2_inode_glock(&dip->i_inode);
|
| 681 | u16 cur_rec_len, prev_rec_len;
|
| 682 |
|
| 683 | if (gfs2_dirent_sentinel(cur)) {
|
| 684 | gfs2_consist_inode(dip);
|
| 685 | return;
|
| 686 | }
|
| 687 |
|
| 688 | gfs2_trans_add_meta(gl, bh);
|
| 689 |
|
| 690 | |
| 691 | The de_rec_len is already as big as it needs to be. Just zero
|
| 692 | out the inode number and return.
|
| 693 |
|
| 694 | if (!prev) {
|
| 695 | cur->de_inum.no_addr = 0;
|
| 696 | cur->de_inum.no_formal_ino = 0;
|
| 697 | return;
|
| 698 | }
|
| 699 |
|
| 700 |
|
| 701 |
|
| 702 | prev_rec_len = be16_to_cpu(prev->de_rec_len);
|
| 703 | cur_rec_len = be16_to_cpu(cur->de_rec_len);
|
| 704 |
|
| 705 | if ((char *)prev + prev_rec_len != (char *)cur)
|
| 706 | gfs2_consist_inode(dip);
|
| 707 | if ((char *)cur + cur_rec_len > bh->b_data + bh->b_size)
|
| 708 | gfs2_consist_inode(dip);
|
| 709 |
|
| 710 | prev_rec_len += cur_rec_len;
|
| 711 | prev->de_rec_len = cpu_to_be16(prev_rec_len);
|
| 712 | }
|
| 713 |
|
| 714 |
|
| 715 | static struct gfs2_dirent *do_init_dirent(struct inode *inode,
|
| 716 | struct gfs2_dirent *dent,
|
| 717 | const struct qstr *name,
|
| 718 | struct buffer_head *bh,
|
| 719 | unsigned offset)
|
| 720 | {
|
| 721 | struct gfs2_glock *gl = gfs2_inode_glock(inode);
|
| 722 | struct gfs2_dirent *ndent;
|
| 723 | unsigned totlen;
|
| 724 |
|
| 725 | totlen = be16_to_cpu(dent->de_rec_len);
|
| 726 | BUG_ON(offset + name->len > totlen);
|
| 727 | gfs2_trans_add_meta(gl, bh);
|
| 728 | ndent = (struct gfs2_dirent *)((char *)dent + offset);
|
| 729 | dent->de_rec_len = cpu_to_be16(offset);
|
| 730 | gfs2_qstr2dirent(name, totlen - offset, ndent);
|
| 731 | return ndent;
|
| 732 | }
|
| 733 |
|
| 734 |
|
| 735 | |
| 736 | * Takes a dent from which to grab space as an argument. Returns the
|
| 737 | * newly created dent.
|
| 738 |
|
| 739 | static struct gfs2_dirent *gfs2_init_dirent(struct inode *inode,
|
| 740 | struct gfs2_dirent *dent,
|
| 741 | const struct qstr *name,
|
| 742 | struct buffer_head *bh)
|
| 743 | {
|
| 744 | unsigned offset = 0;
|
| 745 |
|
| 746 | if (!gfs2_dirent_sentinel(dent))
|
| 747 | offset = GFS2_DIRENT_SIZE(be16_to_cpu(dent->de_name_len));
|
| 748 | return do_init_dirent(inode, dent, name, bh, offset);
|
| 749 | }
|
| 750 |
|
| 751 | static struct gfs2_dirent *gfs2_dirent_split_alloc(struct inode *inode,
|
| 752 | struct buffer_head *bh,
|
| 753 | const struct qstr *name,
|
| 754 | void *ptr)
|
| 755 | {
|
| 756 | struct gfs2_dirent *dent;
|
| 757 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
|
| 758 | gfs2_dirent_find_offset, name, ptr);
|
| 759 | if (IS_ERR_OR_NULL(dent))
|
| 760 | return dent;
|
| 761 | return do_init_dirent(inode, dent, name, bh,
|
| 762 | (unsigned)(ptr - (void *)dent));
|
| 763 | }
|
| 764 |
|
| 765 | static int get_leaf(struct gfs2_inode *dip, u64 leaf_no,
|
| 766 | struct buffer_head **bhp)
|
| 767 | {
|
| 768 | struct inode *inode = &dip->i_inode;
|
| 769 | struct gfs2_glock *gl = gfs2_inode_glock(inode);
|
| 770 | struct gfs2_sbd *sdp = GFS2_SB(inode);
|
| 771 | int error;
|
| 772 |
|
| 773 | error = gfs2_meta_read(gl, leaf_no, DIO_WAIT, 0, bhp);
|
| 774 | if (!error && gfs2_metatype_check(sdp, *bhp, GFS2_METATYPE_LF)) {
|
| 775 |
|
| 776 | error = -EIO;
|
| 777 | }
|
| 778 |
|
| 779 | return error;
|
| 780 | }
|
| 781 |
|
| 782 | |
| 783 | * get_leaf_nr - Get a leaf number associated with the index
|
| 784 | * @dip: The GFS2 inode
|
| 785 | * @index: hash table index of the targeted leaf
|
| 786 | * @leaf_out: Resulting leaf block number
|
| 787 | *
|
| 788 | * Returns: 0 on success, error code otherwise
|
| 789 |
|
| 790 |
|
| 791 | static int get_leaf_nr(struct gfs2_inode *dip, u32 index, u64 *leaf_out)
|
| 792 | {
|
| 793 | __be64 *hash;
|
| 794 | int error;
|
| 795 |
|
| 796 | hash = gfs2_dir_get_hash_table(dip);
|
| 797 | error = PTR_ERR_OR_ZERO(hash);
|
| 798 |
|
| 799 | if (!error)
|
| 800 | *leaf_out = be64_to_cpu(*(hash + index));
|
| 801 |
|
| 802 | return error;
|
| 803 | }
|
| 804 |
|
| 805 | static int get_first_leaf(struct gfs2_inode *dip, u32 index,
|
| 806 | struct buffer_head **bh_out)
|
| 807 | {
|
| 808 | u64 leaf_no;
|
| 809 | int error;
|
| 810 |
|
| 811 | error = get_leaf_nr(dip, index, &leaf_no);
|
| 812 | if (!error)
|
| 813 | error = get_leaf(dip, leaf_no, bh_out);
|
| 814 |
|
| 815 | return error;
|
| 816 | }
|
| 817 |
|
| 818 | static struct gfs2_dirent *gfs2_dirent_search(struct inode *inode,
|
| 819 | const struct qstr *name,
|
| 820 | gfs2_dscan_t scan,
|
| 821 | struct buffer_head **pbh)
|
| 822 | {
|
| 823 | struct buffer_head *bh;
|
| 824 | struct gfs2_dirent *dent;
|
| 825 | struct gfs2_inode *ip = GFS2_I(inode);
|
| 826 | int error;
|
| 827 |
|
| 828 | if (ip->i_diskflags & GFS2_DIF_EXHASH) {
|
| 829 | struct gfs2_leaf *leaf;
|
| 830 | unsigned int hsize = BIT(ip->i_depth);
|
| 831 | unsigned int index;
|
| 832 | u64 ln;
|
| 833 | if (hsize * sizeof(u64) != i_size_read(inode)) {
|
| 834 | gfs2_consist_inode(ip);
|
| 835 | return ERR_PTR(-EIO);
|
| 836 | }
|
| 837 |
|
| 838 | index = name->hash >> (32 - ip->i_depth);
|
| 839 | error = get_first_leaf(ip, index, &bh);
|
| 840 | if (error)
|
| 841 | return ERR_PTR(error);
|
| 842 | do {
|
| 843 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
|
| 844 | scan, name, NULL);
|
| 845 | if (dent)
|
| 846 | goto got_dent;
|
| 847 | leaf = (struct gfs2_leaf *)bh->b_data;
|
| 848 | ln = be64_to_cpu(leaf->lf_next);
|
| 849 | brelse(bh);
|
| 850 | if (!ln)
|
| 851 | break;
|
| 852 |
|
| 853 | error = get_leaf(ip, ln, &bh);
|
| 854 | } while(!error);
|
| 855 |
|
| 856 | return error ? ERR_PTR(error) : NULL;
|
| 857 | }
|
| 858 |
|
| 859 |
|
| 860 | error = gfs2_meta_inode_buffer(ip, &bh);
|
| 861 | if (error)
|
| 862 | return ERR_PTR(error);
|
| 863 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size, scan, name, NULL);
|
| 864 | got_dent:
|
| 865 | if (IS_ERR_OR_NULL(dent)) {
|
| 866 | brelse(bh);
|
| 867 | bh = NULL;
|
| 868 | }
|
| 869 | *pbh = bh;
|
| 870 | return dent;
|
| 871 | }
|
| 872 |
|
| 873 | static struct gfs2_leaf *new_leaf(struct inode *inode, struct buffer_head **pbh, u16 depth)
|
| 874 | {
|
| 875 | struct gfs2_glock *gl = gfs2_inode_glock(inode);
|
| 876 | struct gfs2_inode *ip = GFS2_I(inode);
|
| 877 | unsigned int n = 1;
|
| 878 | u64 bn;
|
| 879 | int error;
|
| 880 | struct buffer_head *bh;
|
| 881 | struct gfs2_leaf *leaf;
|
| 882 | struct gfs2_dirent *dent;
|
| 883 | struct timespec64 tv = current_time(inode);
|
| 884 |
|
| 885 | error = gfs2_alloc_blocks(ip, &bn, &n, 0);
|
| 886 | if (error)
|
| 887 | return NULL;
|
| 888 | bh = gfs2_meta_new(gl, bn);
|
| 889 | if (!bh)
|
| 890 | return NULL;
|
| 891 |
|
| 892 | gfs2_trans_remove_revoke(GFS2_SB(inode), bn, 1);
|
| 893 | gfs2_trans_add_meta(gl, bh);
|
| 894 | gfs2_metatype_set(bh, GFS2_METATYPE_LF, GFS2_FORMAT_LF);
|
| 895 | leaf = (struct gfs2_leaf *)bh->b_data;
|
| 896 | leaf->lf_depth = cpu_to_be16(depth);
|
| 897 | leaf->lf_entries = 0;
|
| 898 | leaf->lf_dirent_format = cpu_to_be32(GFS2_FORMAT_DE);
|
| 899 | leaf->lf_next = 0;
|
| 900 | leaf->lf_inode = cpu_to_be64(ip->i_no_addr);
|
| 901 | leaf->lf_dist = cpu_to_be32(1);
|
| 902 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
|
| 903 | leaf->lf_sec = cpu_to_be64(tv.tv_sec);
|
| 904 | memset(leaf->lf_reserved2, 0, sizeof(leaf->lf_reserved2));
|
| 905 | dent = (struct gfs2_dirent *)(leaf+1);
|
| 906 | gfs2_qstr2dirent(&empty_name, bh->b_size - sizeof(struct gfs2_leaf), dent);
|
| 907 | *pbh = bh;
|
| 908 | return leaf;
|
| 909 | }
|
| 910 |
|
| 911 | |
| 912 | * dir_make_exhash - Convert a stuffed directory into an ExHash directory
|
| 913 | * @inode: The directory inode to be converted to exhash
|
| 914 | *
|
| 915 | * Returns: 0 on success, error code otherwise
|
| 916 |
|
| 917 |
|
| 918 | static int dir_make_exhash(struct inode *inode)
|
| 919 | {
|
| 920 | struct gfs2_glock *gl = gfs2_inode_glock(inode);
|
| 921 | struct gfs2_inode *dip = GFS2_I(inode);
|
| 922 | struct gfs2_sbd *sdp = GFS2_SB(inode);
|
| 923 | struct gfs2_dirent *dent;
|
| 924 | struct qstr args;
|
| 925 | struct buffer_head *bh, *dibh;
|
| 926 | struct gfs2_leaf *leaf;
|
| 927 | u32 x;
|
| 928 | __be64 *lp;
|
| 929 | u64 bn;
|
| 930 | int error;
|
| 931 |
|
| 932 | error = gfs2_meta_inode_buffer(dip, &dibh);
|
| 933 | if (error)
|
| 934 | return error;
|
| 935 |
|
| 936 |
|
| 937 |
|
| 938 | leaf = new_leaf(inode, &bh, 0);
|
| 939 | if (!leaf)
|
| 940 | return -ENOSPC;
|
| 941 | bn = bh->b_blocknr;
|
| 942 |
|
| 943 | gfs2_assert(sdp, dip->i_entries < BIT(16));
|
| 944 | leaf->lf_entries = cpu_to_be16(dip->i_entries);
|
| 945 |
|
| 946 |
|
| 947 |
|
| 948 | gfs2_buffer_copy_tail(bh, sizeof(struct gfs2_leaf), dibh,
|
| 949 | sizeof(struct gfs2_dinode));
|
| 950 |
|
| 951 |
|
| 952 |
|
| 953 | x = 0;
|
| 954 | args.len = bh->b_size - sizeof(struct gfs2_dinode) +
|
| 955 | sizeof(struct gfs2_leaf);
|
| 956 | args.name = bh->b_data;
|
| 957 | dent = gfs2_dirent_scan(&dip->i_inode, bh->b_data, bh->b_size,
|
| 958 | gfs2_dirent_last, &args, NULL);
|
| 959 | if (!dent) {
|
| 960 | brelse(bh);
|
| 961 | brelse(dibh);
|
| 962 | return -EIO;
|
| 963 | }
|
| 964 | if (IS_ERR(dent)) {
|
| 965 | brelse(bh);
|
| 966 | brelse(dibh);
|
| 967 | return PTR_ERR(dent);
|
| 968 | }
|
| 969 |
|
| 970 | |
| 971 | (Remember that dent still points to the last entry.)
|
| 972 |
|
| 973 | dent->de_rec_len = cpu_to_be16(be16_to_cpu(dent->de_rec_len) +
|
| 974 | sizeof(struct gfs2_dinode) -
|
| 975 | sizeof(struct gfs2_leaf));
|
| 976 |
|
| 977 | brelse(bh);
|
| 978 |
|
| 979 | |
| 980 | hash table.
|
| 981 |
|
| 982 | gfs2_trans_add_meta(gl, dibh);
|
| 983 | gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
|
| 984 |
|
| 985 | lp = (__be64 *)(dibh->b_data + sizeof(struct gfs2_dinode));
|
| 986 |
|
| 987 | for (x = sdp->sd_hash_ptrs; x--; lp++)
|
| 988 | *lp = cpu_to_be64(bn);
|
| 989 |
|
| 990 | i_size_write(inode, sdp->sd_sb.sb_bsize / 2);
|
| 991 | gfs2_add_inode_blocks(&dip->i_inode, 1);
|
| 992 | dip->i_diskflags |= GFS2_DIF_EXHASH;
|
| 993 | dip->i_depth = ilog2(sdp->sd_hash_ptrs);
|
| 994 |
|
| 995 | gfs2_dinode_out(dip, dibh->b_data);
|
| 996 |
|
| 997 | brelse(dibh);
|
| 998 |
|
| 999 | return 0;
|
| 1000 | }
|
| 1001 |
|
| 1002 | |
| 1003 | * dir_split_leaf - Split a leaf block into two
|
| 1004 | * @inode: The directory inode to be split
|
| 1005 | * @name: name of the dirent we're trying to insert
|
| 1006 | *
|
| 1007 | * Returns: 0 on success, error code on failure
|
| 1008 |
|
| 1009 |
|
| 1010 | static int dir_split_leaf(struct inode *inode, const struct qstr *name)
|
| 1011 | {
|
| 1012 | struct gfs2_glock *gl = gfs2_inode_glock(inode);
|
| 1013 | struct gfs2_inode *dip = GFS2_I(inode);
|
| 1014 | struct buffer_head *nbh, *obh, *dibh;
|
| 1015 | struct gfs2_leaf *nleaf, *oleaf;
|
| 1016 | struct gfs2_dirent *dent = NULL, *prev = NULL, *next = NULL, *new;
|
| 1017 | u32 start, len, half_len, divider;
|
| 1018 | u64 bn, leaf_no;
|
| 1019 | __be64 *lp;
|
| 1020 | u32 index;
|
| 1021 | int x;
|
| 1022 | int error;
|
| 1023 |
|
| 1024 | index = name->hash >> (32 - dip->i_depth);
|
| 1025 | error = get_leaf_nr(dip, index, &leaf_no);
|
| 1026 | if (error)
|
| 1027 | return error;
|
| 1028 |
|
| 1029 |
|
| 1030 | error = get_leaf(dip, leaf_no, &obh);
|
| 1031 | if (error)
|
| 1032 | return error;
|
| 1033 |
|
| 1034 | oleaf = (struct gfs2_leaf *)obh->b_data;
|
| 1035 | if (dip->i_depth == be16_to_cpu(oleaf->lf_depth)) {
|
| 1036 | brelse(obh);
|
| 1037 | return 1;
|
| 1038 | }
|
| 1039 |
|
| 1040 | gfs2_trans_add_meta(gl, obh);
|
| 1041 |
|
| 1042 | nleaf = new_leaf(inode, &nbh, be16_to_cpu(oleaf->lf_depth) + 1);
|
| 1043 | if (!nleaf) {
|
| 1044 | brelse(obh);
|
| 1045 | return -ENOSPC;
|
| 1046 | }
|
| 1047 | bn = nbh->b_blocknr;
|
| 1048 |
|
| 1049 |
|
| 1050 | len = BIT(dip->i_depth - be16_to_cpu(oleaf->lf_depth));
|
| 1051 | half_len = len >> 1;
|
| 1052 | if (!half_len) {
|
| 1053 | fs_warn(GFS2_SB(inode), "i_depth %u lf_depth %u index %u\n",
|
| 1054 | dip->i_depth, be16_to_cpu(oleaf->lf_depth), index);
|
| 1055 | gfs2_consist_inode(dip);
|
| 1056 | error = -EIO;
|
| 1057 | goto fail_brelse;
|
| 1058 | }
|
| 1059 |
|
| 1060 | start = (index & ~(len - 1));
|
| 1061 |
|
| 1062 | |
| 1063 | Don't bother distinguishing stuffed from non-stuffed.
|
| 1064 | This code is complicated enough already.
|
| 1065 | lp = kmalloc_array(half_len, sizeof(__be64), GFP_NOFS);
|
| 1066 | if (!lp) {
|
| 1067 | error = -ENOMEM;
|
| 1068 | goto fail_brelse;
|
| 1069 | }
|
| 1070 |
|
| 1071 |
|
| 1072 | for (x = 0; x < half_len; x++)
|
| 1073 | lp[x] = cpu_to_be64(bn);
|
| 1074 |
|
| 1075 | gfs2_dir_hash_inval(dip);
|
| 1076 |
|
| 1077 | error = gfs2_dir_write_data(dip, (char *)lp, start * sizeof(u64),
|
| 1078 | half_len * sizeof(u64));
|
| 1079 | if (error != half_len * sizeof(u64)) {
|
| 1080 | if (error >= 0)
|
| 1081 | error = -EIO;
|
| 1082 | goto fail_lpfree;
|
| 1083 | }
|
| 1084 |
|
| 1085 | kfree(lp);
|
| 1086 |
|
| 1087 |
|
| 1088 | divider = (start + half_len) << (32 - dip->i_depth);
|
| 1089 |
|
| 1090 |
|
| 1091 | dent = (struct gfs2_dirent *)(obh->b_data + sizeof(struct gfs2_leaf));
|
| 1092 |
|
| 1093 | do {
|
| 1094 | next = dent;
|
| 1095 | if (dirent_next(dip, obh, &next))
|
| 1096 | next = NULL;
|
| 1097 |
|
| 1098 | if (!gfs2_dirent_sentinel(dent) &&
|
| 1099 | be32_to_cpu(dent->de_hash) < divider) {
|
| 1100 | struct qstr str;
|
| 1101 | void *ptr = ((char *)dent - obh->b_data) + nbh->b_data;
|
| 1102 | str.name = (char*)(dent+1);
|
| 1103 | str.len = be16_to_cpu(dent->de_name_len);
|
| 1104 | str.hash = be32_to_cpu(dent->de_hash);
|
| 1105 | new = gfs2_dirent_split_alloc(inode, nbh, &str, ptr);
|
| 1106 | if (IS_ERR(new)) {
|
| 1107 | error = PTR_ERR(new);
|
| 1108 | break;
|
| 1109 | }
|
| 1110 |
|
| 1111 | new->de_inum = dent->de_inum;
|
| 1112 | new->de_type = dent->de_type;
|
| 1113 | be16_add_cpu(&nleaf->lf_entries, 1);
|
| 1114 |
|
| 1115 | dirent_del(dip, obh, prev, dent);
|
| 1116 |
|
| 1117 | if (!oleaf->lf_entries)
|
| 1118 | gfs2_consist_inode(dip);
|
| 1119 | be16_add_cpu(&oleaf->lf_entries, -1);
|
| 1120 |
|
| 1121 | if (!prev)
|
| 1122 | prev = dent;
|
| 1123 | } else {
|
| 1124 | prev = dent;
|
| 1125 | }
|
| 1126 | dent = next;
|
| 1127 | } while (dent);
|
| 1128 |
|
| 1129 | oleaf->lf_depth = nleaf->lf_depth;
|
| 1130 |
|
| 1131 | error = gfs2_meta_inode_buffer(dip, &dibh);
|
| 1132 | if (!gfs2_assert_withdraw(GFS2_SB(&dip->i_inode), !error)) {
|
| 1133 | gfs2_trans_add_meta(gl, dibh);
|
| 1134 | gfs2_add_inode_blocks(&dip->i_inode, 1);
|
| 1135 | gfs2_dinode_out(dip, dibh->b_data);
|
| 1136 | brelse(dibh);
|
| 1137 | }
|
| 1138 |
|
| 1139 | brelse(obh);
|
| 1140 | brelse(nbh);
|
| 1141 |
|
| 1142 | return error;
|
| 1143 |
|
| 1144 | fail_lpfree:
|
| 1145 | kfree(lp);
|
| 1146 |
|
| 1147 | fail_brelse:
|
| 1148 | brelse(obh);
|
| 1149 | brelse(nbh);
|
| 1150 | return error;
|
| 1151 | }
|
| 1152 |
|
| 1153 | |
| 1154 | * dir_double_exhash - Double size of ExHash table
|
| 1155 | * @dip: The GFS2 dinode
|
| 1156 | *
|
| 1157 | * Returns: 0 on success, error code on failure
|
| 1158 |
|
| 1159 |
|
| 1160 | static int dir_double_exhash(struct gfs2_inode *dip)
|
| 1161 | {
|
| 1162 | struct buffer_head *dibh;
|
| 1163 | u32 hsize;
|
| 1164 | u32 hsize_bytes;
|
| 1165 | __be64 *hc;
|
| 1166 | __be64 *hc2, *h;
|
| 1167 | int x;
|
| 1168 | int error = 0;
|
| 1169 |
|
| 1170 | hsize = BIT(dip->i_depth);
|
| 1171 | hsize_bytes = hsize * sizeof(__be64);
|
| 1172 |
|
| 1173 | hc = gfs2_dir_get_hash_table(dip);
|
| 1174 | if (IS_ERR(hc))
|
| 1175 | return PTR_ERR(hc);
|
| 1176 |
|
| 1177 | hc2 = kmalloc_array(hsize_bytes, 2, GFP_NOFS | __GFP_NOWARN);
|
| 1178 | if (hc2 == NULL)
|
| 1179 | hc2 = __vmalloc(hsize_bytes * 2, GFP_NOFS);
|
| 1180 |
|
| 1181 | if (!hc2)
|
| 1182 | return -ENOMEM;
|
| 1183 |
|
| 1184 | h = hc2;
|
| 1185 | error = gfs2_meta_inode_buffer(dip, &dibh);
|
| 1186 | if (error)
|
| 1187 | goto out_kfree;
|
| 1188 |
|
| 1189 | for (x = 0; x < hsize; x++) {
|
| 1190 | *h++ = *hc;
|
| 1191 | *h++ = *hc;
|
| 1192 | hc++;
|
| 1193 | }
|
| 1194 |
|
| 1195 | error = gfs2_dir_write_data(dip, (char *)hc2, 0, hsize_bytes * 2);
|
| 1196 | if (error != (hsize_bytes * 2))
|
| 1197 | goto fail;
|
| 1198 |
|
| 1199 | gfs2_dir_hash_inval(dip);
|
| 1200 | dip->i_hash_cache = hc2;
|
| 1201 | dip->i_depth++;
|
| 1202 | gfs2_dinode_out(dip, dibh->b_data);
|
| 1203 | brelse(dibh);
|
| 1204 | return 0;
|
| 1205 |
|
| 1206 | fail:
|
| 1207 |
|
| 1208 | gfs2_dir_write_data(dip, (char *)hc, 0, hsize_bytes);
|
| 1209 | i_size_write(&dip->i_inode, hsize_bytes);
|
| 1210 | gfs2_dinode_out(dip, dibh->b_data);
|
| 1211 | brelse(dibh);
|
| 1212 | out_kfree:
|
| 1213 | kvfree(hc2);
|
| 1214 | return error;
|
| 1215 | }
|
| 1216 |
|
| 1217 | |
| 1218 | * compare_dents - compare directory entries by hash value
|
| 1219 | * @a: first dent
|
| 1220 | * @b: second dent
|
| 1221 | *
|
| 1222 | * When comparing the hash entries of @a to @b:
|
| 1223 | * gt: returns 1
|
| 1224 | * lt: returns -1
|
| 1225 | * eq: returns 0
|
| 1226 |
|
| 1227 |
|
| 1228 | static int compare_dents(const void *a, const void *b)
|
| 1229 | {
|
| 1230 | const struct gfs2_dirent *dent_a, *dent_b;
|
| 1231 | u32 hash_a, hash_b;
|
| 1232 | int ret = 0;
|
| 1233 |
|
| 1234 | dent_a = *(const struct gfs2_dirent **)a;
|
| 1235 | hash_a = dent_a->de_cookie;
|
| 1236 |
|
| 1237 | dent_b = *(const struct gfs2_dirent **)b;
|
| 1238 | hash_b = dent_b->de_cookie;
|
| 1239 |
|
| 1240 | if (hash_a > hash_b)
|
| 1241 | ret = 1;
|
| 1242 | else if (hash_a < hash_b)
|
| 1243 | ret = -1;
|
| 1244 | else {
|
| 1245 | unsigned int len_a = be16_to_cpu(dent_a->de_name_len);
|
| 1246 | unsigned int len_b = be16_to_cpu(dent_b->de_name_len);
|
| 1247 |
|
| 1248 | if (len_a > len_b)
|
| 1249 | ret = 1;
|
| 1250 | else if (len_a < len_b)
|
| 1251 | ret = -1;
|
| 1252 | else
|
| 1253 | ret = memcmp(dent_a + 1, dent_b + 1, len_a);
|
| 1254 | }
|
| 1255 |
|
| 1256 | return ret;
|
| 1257 | }
|
| 1258 |
|
| 1259 | |
| 1260 | * do_filldir_main - read out directory entries
|
| 1261 | * @dip: The GFS2 inode
|
| 1262 | * @ctx: what to feed the entries to
|
| 1263 | * @darr: an array of struct gfs2_dirent pointers to read
|
| 1264 | * @entries: the number of entries in darr
|
| 1265 | * @sort_start: index of the directory array to start our sort
|
| 1266 | * @copied: pointer to int that's non-zero if a entry has been copied out
|
| 1267 | *
|
| 1268 | * Jump through some hoops to make sure that if there are hash collsions,
|
| 1269 | * they are read out at the beginning of a buffer. We want to minimize
|
| 1270 | * the possibility that they will fall into different readdir buffers or
|
| 1271 | * that someone will want to seek to that location.
|
| 1272 | *
|
| 1273 | * Returns: errno, >0 if the actor tells you to stop
|
| 1274 |
|
| 1275 |
|
| 1276 | static int do_filldir_main(struct gfs2_inode *dip, struct dir_context *ctx,
|
| 1277 | struct gfs2_dirent **darr, u32 entries,
|
| 1278 | u32 sort_start, int *copied)
|
| 1279 | {
|
| 1280 | const struct gfs2_dirent *dent, *dent_next;
|
| 1281 | u64 off, off_next;
|
| 1282 | unsigned int x, y;
|
| 1283 | int run = 0;
|
| 1284 |
|
| 1285 | if (sort_start < entries)
|
| 1286 | sort(&darr[sort_start], entries - sort_start,
|
| 1287 | sizeof(struct gfs2_dirent *), compare_dents, NULL);
|
| 1288 |
|
| 1289 | dent_next = darr[0];
|
| 1290 | off_next = dent_next->de_cookie;
|
| 1291 |
|
| 1292 | for (x = 0, y = 1; x < entries; x++, y++) {
|
| 1293 | dent = dent_next;
|
| 1294 | off = off_next;
|
| 1295 |
|
| 1296 | if (y < entries) {
|
| 1297 | dent_next = darr[y];
|
| 1298 | off_next = dent_next->de_cookie;
|
| 1299 |
|
| 1300 | if (off < ctx->pos)
|
| 1301 | continue;
|
| 1302 | ctx->pos = off;
|
| 1303 |
|
| 1304 | if (off_next == off) {
|
| 1305 | if (*copied && !run)
|
| 1306 | return 1;
|
| 1307 | run = 1;
|
| 1308 | } else
|
| 1309 | run = 0;
|
| 1310 | } else {
|
| 1311 | if (off < ctx->pos)
|
| 1312 | continue;
|
| 1313 | ctx->pos = off;
|
| 1314 | }
|
| 1315 |
|
| 1316 | if (!dir_emit(ctx, (const char *)(dent + 1),
|
| 1317 | be16_to_cpu(dent->de_name_len),
|
| 1318 | be64_to_cpu(dent->de_inum.no_addr),
|
| 1319 | be16_to_cpu(dent->de_type)))
|
| 1320 | return 1;
|
| 1321 |
|
| 1322 | *copied = 1;
|
| 1323 | }
|
| 1324 |
|
| 1325 | |
| 1326 | do_filldir fxn, we get the next entry instead of the last one in the
|
| 1327 | current leaf
|
| 1328 |
|
| 1329 | ctx->pos++;
|
| 1330 |
|
| 1331 | return 0;
|
| 1332 | }
|
| 1333 |
|
| 1334 | static void *gfs2_alloc_sort_buffer(unsigned size)
|
| 1335 | {
|
| 1336 | void *ptr = NULL;
|
| 1337 |
|
| 1338 | if (size < KMALLOC_MAX_SIZE)
|
| 1339 | ptr = kmalloc(size, GFP_NOFS | __GFP_NOWARN);
|
| 1340 | if (!ptr)
|
| 1341 | ptr = __vmalloc(size, GFP_NOFS);
|
| 1342 | return ptr;
|
| 1343 | }
|
| 1344 |
|
| 1345 |
|
| 1346 | static int gfs2_set_cookies(struct gfs2_sbd *sdp, struct buffer_head *bh,
|
| 1347 | unsigned leaf_nr, struct gfs2_dirent **darr,
|
| 1348 | unsigned entries)
|
| 1349 | {
|
| 1350 | int sort_id = -1;
|
| 1351 | int i;
|
| 1352 |
|
| 1353 | for (i = 0; i < entries; i++) {
|
| 1354 | unsigned offset;
|
| 1355 |
|
| 1356 | darr[i]->de_cookie = be32_to_cpu(darr[i]->de_hash);
|
| 1357 | darr[i]->de_cookie = gfs2_disk_hash2offset(darr[i]->de_cookie);
|
| 1358 |
|
| 1359 | if (!sdp->sd_args.ar_loccookie)
|
| 1360 | continue;
|
| 1361 | offset = (char *)(darr[i]) -
|
| 1362 | (bh->b_data + gfs2_dirent_offset(sdp, bh->b_data));
|
| 1363 | offset /= GFS2_MIN_DIRENT_SIZE;
|
| 1364 | offset += leaf_nr * sdp->sd_max_dents_per_leaf;
|
| 1365 | if (offset >= GFS2_USE_HASH_FLAG ||
|
| 1366 | leaf_nr >= GFS2_USE_HASH_FLAG) {
|
| 1367 | darr[i]->de_cookie |= GFS2_USE_HASH_FLAG;
|
| 1368 | if (sort_id < 0)
|
| 1369 | sort_id = i;
|
| 1370 | continue;
|
| 1371 | }
|
| 1372 | darr[i]->de_cookie &= GFS2_HASH_INDEX_MASK;
|
| 1373 | darr[i]->de_cookie |= offset;
|
| 1374 | }
|
| 1375 | return sort_id;
|
| 1376 | }
|
| 1377 |
|
| 1378 |
|
| 1379 | static int gfs2_dir_read_leaf(struct inode *inode, struct dir_context *ctx,
|
| 1380 | int *copied, unsigned *depth,
|
| 1381 | u64 leaf_no)
|
| 1382 | {
|
| 1383 | struct gfs2_inode *ip = GFS2_I(inode);
|
| 1384 | struct gfs2_sbd *sdp = GFS2_SB(inode);
|
| 1385 | struct buffer_head *bh;
|
| 1386 | struct gfs2_leaf *lf;
|
| 1387 | unsigned entries = 0, entries2 = 0;
|
| 1388 | unsigned leaves = 0, leaf = 0, offset, sort_offset;
|
| 1389 | struct gfs2_dirent **darr, *dent;
|
| 1390 | struct dirent_gather g;
|
| 1391 | struct buffer_head **larr;
|
| 1392 | int error, i, need_sort = 0, sort_id;
|
| 1393 | u64 lfn = leaf_no;
|
| 1394 |
|
| 1395 | do {
|
| 1396 | error = get_leaf(ip, lfn, &bh);
|
| 1397 | if (error)
|
| 1398 | goto out;
|
| 1399 | lf = (struct gfs2_leaf *)bh->b_data;
|
| 1400 | if (leaves == 0)
|
| 1401 | *depth = be16_to_cpu(lf->lf_depth);
|
| 1402 | entries += be16_to_cpu(lf->lf_entries);
|
| 1403 | leaves++;
|
| 1404 | lfn = be64_to_cpu(lf->lf_next);
|
| 1405 | brelse(bh);
|
| 1406 | } while(lfn);
|
| 1407 |
|
| 1408 | if (*depth < GFS2_DIR_MAX_DEPTH || !sdp->sd_args.ar_loccookie) {
|
| 1409 | need_sort = 1;
|
| 1410 | sort_offset = 0;
|
| 1411 | }
|
| 1412 |
|
| 1413 | if (!entries)
|
| 1414 | return 0;
|
| 1415 |
|
| 1416 | error = -ENOMEM;
|
| 1417 | |
| 1418 | * The extra 99 entries are not normally used, but are a buffer
|
| 1419 | * zone in case the number of entries in the leaf is corrupt.
|
| 1420 | * 99 is the maximum number of entries that can fit in a single
|
| 1421 | * leaf block.
|
| 1422 |
|
| 1423 | larr = gfs2_alloc_sort_buffer((leaves + entries + 99) * sizeof(void *));
|
| 1424 | if (!larr)
|
| 1425 | goto out;
|
| 1426 | darr = (struct gfs2_dirent **)(larr + leaves);
|
| 1427 | g.pdent = (const struct gfs2_dirent **)darr;
|
| 1428 | g.offset = 0;
|
| 1429 | lfn = leaf_no;
|
| 1430 |
|
| 1431 | do {
|
| 1432 | error = get_leaf(ip, lfn, &bh);
|
| 1433 | if (error)
|
| 1434 | goto out_free;
|
| 1435 | lf = (struct gfs2_leaf *)bh->b_data;
|
| 1436 | lfn = be64_to_cpu(lf->lf_next);
|
| 1437 | if (lf->lf_entries) {
|
| 1438 | offset = g.offset;
|
| 1439 | entries2 += be16_to_cpu(lf->lf_entries);
|
| 1440 | dent = gfs2_dirent_scan(inode, bh->b_data, bh->b_size,
|
| 1441 | gfs2_dirent_gather, NULL, &g);
|
| 1442 | error = PTR_ERR(dent);
|
| 1443 | if (IS_ERR(dent))
|
| 1444 | goto out_free;
|
| 1445 | if (entries2 != g.offset) {
|
| 1446 | fs_warn(sdp, "Number of entries corrupt in dir "
|
| 1447 | "leaf %llu, entries2 (%u) != "
|
| 1448 | "g.offset (%u)\n",
|
| 1449 | (unsigned long long)bh->b_blocknr,
|
| 1450 | entries2, g.offset);
|
| 1451 | gfs2_consist_inode(ip);
|
| 1452 | error = -EIO;
|
| 1453 | goto out_free;
|
| 1454 | }
|
| 1455 | error = 0;
|
| 1456 | sort_id = gfs2_set_cookies(sdp, bh, leaf, &darr[offset],
|
| 1457 | be16_to_cpu(lf->lf_entries));
|
| 1458 | if (!need_sort && sort_id >= 0) {
|
| 1459 | need_sort = 1;
|
| 1460 | sort_offset = offset + sort_id;
|
| 1461 | }
|
| 1462 | larr[leaf++] = bh;
|
| 1463 | } else {
|
| 1464 | larr[leaf++] = NULL;
|
| 1465 | brelse(bh);
|
| 1466 | }
|
| 1467 | } while(lfn);
|
| 1468 |
|
| 1469 | BUG_ON(entries2 != entries);
|
| 1470 | error = do_filldir_main(ip, ctx, darr, entries, need_sort ?
|
| 1471 | sort_offset : entries, copied);
|
| 1472 | out_free:
|
| 1473 | for(i = 0; i < leaf; i++)
|
| 1474 | brelse(larr[i]);
|
| 1475 | kvfree(larr);
|
| 1476 | out:
|
| 1477 | return error;
|
| 1478 | }
|
| 1479 |
|
| 1480 | |
| 1481 | * gfs2_dir_readahead - Issue read-ahead requests for leaf blocks.
|
| 1482 | * @inode: the directory inode
|
| 1483 | * @hsize: hash table size
|
| 1484 | * @index: index into the hash table
|
| 1485 | * @f_ra: read-ahead parameters
|
| 1486 | *
|
| 1487 | * Note: we can't calculate each index like dir_e_read can because we don't
|
| 1488 | * have the leaf, and therefore we don't have the depth, and therefore we
|
| 1489 | * don't have the length. So we have to just read enough ahead to make up
|
| 1490 | * for the loss of information.
|
| 1491 |
|
| 1492 | static void gfs2_dir_readahead(struct inode *inode, unsigned hsize, u32 index,
|
| 1493 | struct file_ra_state *f_ra)
|
| 1494 | {
|
| 1495 | struct gfs2_glock *gl = gfs2_inode_glock(inode);
|
| 1496 | struct gfs2_inode *ip = GFS2_I(inode);
|
| 1497 | struct buffer_head *bh;
|
| 1498 | u64 blocknr = 0, last;
|
| 1499 | unsigned count;
|
| 1500 |
|
| 1501 |
|
| 1502 | if (index + MAX_RA_BLOCKS < f_ra->start)
|
| 1503 | return;
|
| 1504 |
|
| 1505 | f_ra->start = max((pgoff_t)index, f_ra->start);
|
| 1506 | for (count = 0; count < MAX_RA_BLOCKS; count++) {
|
| 1507 | if (f_ra->start >= hsize)
|
| 1508 | break;
|
| 1509 |
|
| 1510 | last = blocknr;
|
| 1511 | blocknr = be64_to_cpu(ip->i_hash_cache[f_ra->start]);
|
| 1512 | f_ra->start++;
|
| 1513 | if (blocknr == last)
|
| 1514 | continue;
|
| 1515 |
|
| 1516 | bh = gfs2_getbuf(gl, blocknr, 1);
|
| 1517 | if (trylock_buffer(bh)) {
|
| 1518 | if (buffer_uptodate(bh)) {
|
| 1519 | unlock_buffer(bh);
|
| 1520 | } else {
|
| 1521 | bh_submit(bh, REQ_OP_READ | REQ_RAHEAD |
|
| 1522 | REQ_META | REQ_PRIO,
|
| 1523 | bh_end_read);
|
| 1524 | }
|
| 1525 | }
|
| 1526 | put_bh(bh);
|
| 1527 | }
|
| 1528 | }
|
| 1529 |
|
| 1530 | |
| 1531 | * dir_e_read - Reads the entries from a directory into a filldir buffer
|
| 1532 | * @inode: the directory inode
|
| 1533 | * @ctx: actor to feed the entries to
|
| 1534 | * @f_ra: read-ahead parameters
|
| 1535 | *
|
| 1536 | * Returns: errno
|
| 1537 |
|
| 1538 |
|
| 1539 | static int dir_e_read(struct inode *inode, struct dir_context *ctx,
|
| 1540 | struct file_ra_state *f_ra)
|
| 1541 | {
|
| 1542 | struct gfs2_inode *dip = GFS2_I(inode);
|
| 1543 | u32 hsize, len = 0;
|
| 1544 | u32 hash, index;
|
| 1545 | __be64 *lp;
|
| 1546 | int copied = 0;
|
| 1547 | int error = 0;
|
| 1548 | unsigned depth = 0;
|
| 1549 |
|
| 1550 | hsize = BIT(dip->i_depth);
|
| 1551 | hash = gfs2_dir_offset2hash(ctx->pos);
|
| 1552 | index = hash >> (32 - dip->i_depth);
|
| 1553 |
|
| 1554 | if (dip->i_hash_cache == NULL)
|
| 1555 | f_ra->start = 0;
|
| 1556 | lp = gfs2_dir_get_hash_table(dip);
|
| 1557 | if (IS_ERR(lp))
|
| 1558 | return PTR_ERR(lp);
|
| 1559 |
|
| 1560 | gfs2_dir_readahead(inode, hsize, index, f_ra);
|
| 1561 |
|
| 1562 | while (index < hsize) {
|
| 1563 | error = gfs2_dir_read_leaf(inode, ctx,
|
| 1564 | &copied, &depth,
|
| 1565 | be64_to_cpu(lp[index]));
|
| 1566 | if (error)
|
| 1567 | break;
|
| 1568 |
|
| 1569 | len = BIT(dip->i_depth - depth);
|
| 1570 | index = (index & ~(len - 1)) + len;
|
| 1571 | }
|
| 1572 |
|
| 1573 | if (error > 0)
|
| 1574 | error = 0;
|
| 1575 | return error;
|
| 1576 | }
|
| 1577 |
|
| 1578 | int gfs2_dir_read(struct inode *inode, struct dir_context *ctx,
|
| 1579 | struct file_ra_state *f_ra)
|
| 1580 | {
|
| 1581 | struct gfs2_inode *dip = GFS2_I(inode);
|
| 1582 | struct gfs2_sbd *sdp = GFS2_SB(inode);
|
| 1583 | struct dirent_gather g;
|
| 1584 | struct gfs2_dirent **darr, *dent;
|
| 1585 | struct buffer_head *dibh;
|
| 1586 | int copied = 0;
|
| 1587 | int error;
|
| 1588 |
|
| 1589 | if (!dip->i_entries)
|
| 1590 | return 0;
|
| 1591 |
|
| 1592 | if (dip->i_diskflags & GFS2_DIF_EXHASH)
|
| 1593 | return dir_e_read(inode, ctx, f_ra);
|
| 1594 |
|
| 1595 | if (!gfs2_is_stuffed(dip)) {
|
| 1596 | gfs2_consist_inode(dip);
|
| 1597 | return -EIO;
|
| 1598 | }
|
| 1599 |
|
| 1600 | error = gfs2_meta_inode_buffer(dip, &dibh);
|
| 1601 | if (error)
|
| 1602 | return error;
|
| 1603 |
|
| 1604 | error = -ENOMEM;
|
| 1605 |
|
| 1606 | darr = kmalloc_objs(struct gfs2_dirent *, 96, GFP_NOFS);
|
| 1607 | if (darr) {
|
| 1608 | g.pdent = (const struct gfs2_dirent **)darr;
|
| 1609 | g.offset = 0;
|
| 1610 | dent = gfs2_dirent_scan(inode, dibh->b_data, dibh->b_size,
|
| 1611 | gfs2_dirent_gather, NULL, &g);
|
| 1612 | if (IS_ERR(dent)) {
|
| 1613 | error = PTR_ERR(dent);
|
| 1614 | goto out;
|
| 1615 | }
|
| 1616 | if (dip->i_entries != g.offset) {
|
| 1617 | fs_warn(sdp, "Number of entries corrupt in dir %llu, "
|
| 1618 | "ip->i_entries (%u) != g.offset (%u)\n",
|
| 1619 | (unsigned long long)dip->i_no_addr,
|
| 1620 | dip->i_entries,
|
| 1621 | g.offset);
|
| 1622 | gfs2_consist_inode(dip);
|
| 1623 | error = -EIO;
|
| 1624 | goto out;
|
| 1625 | }
|
| 1626 | gfs2_set_cookies(sdp, dibh, 0, darr, dip->i_entries);
|
| 1627 | error = do_filldir_main(dip, ctx, darr,
|
| 1628 | dip->i_entries, 0, &copied);
|
| 1629 | out:
|
| 1630 | kfree(darr);
|
| 1631 | }
|
| 1632 |
|
| 1633 | if (error > 0)
|
| 1634 | error = 0;
|
| 1635 |
|
| 1636 | brelse(dibh);
|
| 1637 |
|
| 1638 | return error;
|
| 1639 | }
|
| 1640 |
|
| 1641 | |
| 1642 | * gfs2_dir_search - Search a directory
|
| 1643 | * @dir: The GFS2 directory inode
|
| 1644 | * @name: The name we are looking up
|
| 1645 | * @fail_on_exist: Fail if the name exists rather than looking it up
|
| 1646 | *
|
| 1647 | * This routine searches a directory for a file or another directory.
|
| 1648 | * Assumes a glock is held on dip.
|
| 1649 | *
|
| 1650 | * Returns: errno
|
| 1651 |
|
| 1652 |
|
| 1653 | struct inode *gfs2_dir_search(struct inode *dir, const struct qstr *name,
|
| 1654 | bool fail_on_exist)
|
| 1655 | {
|
| 1656 | struct buffer_head *bh;
|
| 1657 | struct gfs2_dirent *dent;
|
| 1658 | u64 addr, formal_ino;
|
| 1659 | u16 dtype;
|
| 1660 |
|
| 1661 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
|
| 1662 | if (dent) {
|
| 1663 | struct inode *inode;
|
| 1664 | u16 rahead;
|
| 1665 |
|
| 1666 | if (IS_ERR(dent))
|
| 1667 | return ERR_CAST(dent);
|
| 1668 | dtype = be16_to_cpu(dent->de_type);
|
| 1669 | rahead = be16_to_cpu(dent->de_rahead);
|
| 1670 | addr = be64_to_cpu(dent->de_inum.no_addr);
|
| 1671 | formal_ino = be64_to_cpu(dent->de_inum.no_formal_ino);
|
| 1672 | brelse(bh);
|
| 1673 | if (fail_on_exist)
|
| 1674 | return ERR_PTR(-EEXIST);
|
| 1675 | inode = gfs2_inode_lookup(dir->i_sb, dtype, addr, formal_ino,
|
| 1676 | GFS2_BLKST_FREE );
|
| 1677 | if (!IS_ERR(inode))
|
| 1678 | GFS2_I(inode)->i_rahead = rahead;
|
| 1679 | return inode;
|
| 1680 | }
|
| 1681 | return ERR_PTR(-ENOENT);
|
| 1682 | }
|
| 1683 |
|
| 1684 | int gfs2_dir_check(struct inode *dir, const struct qstr *name,
|
| 1685 | const struct gfs2_inode *ip)
|
| 1686 | {
|
| 1687 | struct buffer_head *bh;
|
| 1688 | struct gfs2_dirent *dent;
|
| 1689 | int ret = -ENOENT;
|
| 1690 |
|
| 1691 | dent = gfs2_dirent_search(dir, name, gfs2_dirent_find, &bh);
|
| 1692 | if (dent) {
|
| 1693 | if (IS_ERR(dent))
|
| 1694 | return PTR_ERR(dent);
|
| 1695 | if (ip) {
|
| 1696 | if (be64_to_cpu(dent->de_inum.no_addr) != ip->i_no_addr)
|
| 1697 | goto out;
|
| 1698 | if (be64_to_cpu(dent->de_inum.no_formal_ino) !=
|
| 1699 | ip->i_no_formal_ino)
|
| 1700 | goto out;
|
| 1701 | if (unlikely(IF2DT(ip->i_inode.i_mode) !=
|
| 1702 | be16_to_cpu(dent->de_type))) {
|
| 1703 | gfs2_consist_inode(GFS2_I(dir));
|
| 1704 | ret = -EIO;
|
| 1705 | goto out;
|
| 1706 | }
|
| 1707 | }
|
| 1708 | ret = 0;
|
| 1709 | out:
|
| 1710 | brelse(bh);
|
| 1711 | }
|
| 1712 | return ret;
|
| 1713 | }
|
| 1714 |
|
| 1715 | |
| 1716 | * dir_new_leaf - Add a new leaf onto hash chain
|
| 1717 | * @inode: The directory
|
| 1718 | * @name: The name we are adding
|
| 1719 | *
|
| 1720 | * This adds a new dir leaf onto an existing leaf when there is not
|
| 1721 | * enough space to add a new dir entry. This is a last resort after
|
| 1722 | * we've expanded the hash table to max size and also split existing
|
| 1723 | * leaf blocks, so it will only occur for very large directories.
|
| 1724 | *
|
| 1725 | * The dist parameter is set to 1 for leaf blocks directly attached
|
| 1726 | * to the hash table, 2 for one layer of indirection, 3 for two layers
|
| 1727 | * etc. We are thus able to tell the difference between an old leaf
|
| 1728 | * with dist set to zero (i.e. "don't know") and a new one where we
|
| 1729 | * set this information for debug/fsck purposes.
|
| 1730 | *
|
| 1731 | * Returns: 0 on success, or -ve on error
|
| 1732 |
|
| 1733 |
|
| 1734 | static int dir_new_leaf(struct inode *inode, const struct qstr *name)
|
| 1735 | {
|
| 1736 | struct buffer_head *bh, *obh;
|
| 1737 | struct gfs2_glock *gl = gfs2_inode_glock(inode);
|
| 1738 | struct gfs2_inode *ip = GFS2_I(inode);
|
| 1739 | struct gfs2_leaf *leaf, *oleaf;
|
| 1740 | u32 dist = 1;
|
| 1741 | int error;
|
| 1742 | u32 index;
|
| 1743 | u64 bn;
|
| 1744 |
|
| 1745 | index = name->hash >> (32 - ip->i_depth);
|
| 1746 | error = get_first_leaf(ip, index, &obh);
|
| 1747 | if (error)
|
| 1748 | return error;
|
| 1749 | do {
|
| 1750 | dist++;
|
| 1751 | oleaf = (struct gfs2_leaf *)obh->b_data;
|
| 1752 | bn = be64_to_cpu(oleaf->lf_next);
|
| 1753 | if (!bn)
|
| 1754 | break;
|
| 1755 | brelse(obh);
|
| 1756 | error = get_leaf(ip, bn, &obh);
|
| 1757 | if (error)
|
| 1758 | return error;
|
| 1759 | } while(1);
|
| 1760 |
|
| 1761 | gfs2_trans_add_meta(gl, obh);
|
| 1762 |
|
| 1763 | leaf = new_leaf(inode, &bh, be16_to_cpu(oleaf->lf_depth));
|
| 1764 | if (!leaf) {
|
| 1765 | brelse(obh);
|
| 1766 | return -ENOSPC;
|
| 1767 | }
|
| 1768 | leaf->lf_dist = cpu_to_be32(dist);
|
| 1769 | oleaf->lf_next = cpu_to_be64(bh->b_blocknr);
|
| 1770 | brelse(bh);
|
| 1771 | brelse(obh);
|
| 1772 |
|
| 1773 | error = gfs2_meta_inode_buffer(ip, &bh);
|
| 1774 | if (error)
|
| 1775 | return error;
|
| 1776 | gfs2_trans_add_meta(gl, bh);
|
| 1777 | gfs2_add_inode_blocks(&ip->i_inode, 1);
|
| 1778 | gfs2_dinode_out(ip, bh->b_data);
|
| 1779 | brelse(bh);
|
| 1780 | return 0;
|
| 1781 | }
|
| 1782 |
|
| 1783 | static u16 gfs2_inode_ra_len(const struct gfs2_inode *ip)
|
| 1784 | {
|
| 1785 | u64 where = ip->i_no_addr + 1;
|
| 1786 | if (ip->i_eattr == where)
|
| 1787 | return 1;
|
| 1788 | return 0;
|
| 1789 | }
|
| 1790 |
|
| 1791 | |
| 1792 | * gfs2_dir_add - Add new filename into directory
|
| 1793 | * @inode: The directory inode
|
| 1794 | * @name: The new name
|
| 1795 | * @nip: The GFS2 inode to be linked in to the directory
|
| 1796 | * @da: The directory addition info
|
| 1797 | *
|
| 1798 | * If the call to gfs2_diradd_alloc_required resulted in there being
|
| 1799 | * no need to allocate any new directory blocks, then it will contain
|
| 1800 | * a pointer to the directory entry and the bh in which it resides. We
|
| 1801 | * can use that without having to repeat the search. If there was no
|
| 1802 | * free space, then we must now create more space.
|
| 1803 | *
|
| 1804 | * Returns: 0 on success, error code on failure
|
| 1805 |
|
| 1806 |
|
| 1807 | int gfs2_dir_add(struct inode *inode, const struct qstr *name,
|
| 1808 | const struct gfs2_inode *nip, struct gfs2_diradd *da)
|
| 1809 | {
|
| 1810 | struct gfs2_inode *ip = GFS2_I(inode);
|
| 1811 | struct buffer_head *bh = da->bh;
|
| 1812 | struct gfs2_dirent *dent = da->dent;
|
| 1813 | struct timespec64 tv;
|
| 1814 | struct gfs2_leaf *leaf;
|
| 1815 | int error;
|
| 1816 |
|
| 1817 | while(1) {
|
| 1818 | if (da->bh == NULL) {
|
| 1819 | dent = gfs2_dirent_search(inode, name,
|
| 1820 | gfs2_dirent_find_space, &bh);
|
| 1821 | }
|
| 1822 | if (dent) {
|
| 1823 | if (IS_ERR(dent))
|
| 1824 | return PTR_ERR(dent);
|
| 1825 | dent = gfs2_init_dirent(inode, dent, name, bh);
|
| 1826 | gfs2_inum_out(nip, dent);
|
| 1827 | dent->de_type = cpu_to_be16(IF2DT(nip->i_inode.i_mode));
|
| 1828 | dent->de_rahead = cpu_to_be16(gfs2_inode_ra_len(nip));
|
| 1829 | tv = inode_set_ctime_current(&ip->i_inode);
|
| 1830 | if (ip->i_diskflags & GFS2_DIF_EXHASH) {
|
| 1831 | leaf = (struct gfs2_leaf *)bh->b_data;
|
| 1832 | be16_add_cpu(&leaf->lf_entries, 1);
|
| 1833 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
|
| 1834 | leaf->lf_sec = cpu_to_be64(tv.tv_sec);
|
| 1835 | }
|
| 1836 | da->dent = NULL;
|
| 1837 | da->bh = NULL;
|
| 1838 | brelse(bh);
|
| 1839 | ip->i_entries++;
|
| 1840 | inode_set_mtime_to_ts(&ip->i_inode, tv);
|
| 1841 | if (S_ISDIR(nip->i_inode.i_mode))
|
| 1842 | inc_nlink(&ip->i_inode);
|
| 1843 | mark_inode_dirty(inode);
|
| 1844 | error = 0;
|
| 1845 | break;
|
| 1846 | }
|
| 1847 | if (!(ip->i_diskflags & GFS2_DIF_EXHASH)) {
|
| 1848 | error = dir_make_exhash(inode);
|
| 1849 | if (error)
|
| 1850 | break;
|
| 1851 | continue;
|
| 1852 | }
|
| 1853 | error = dir_split_leaf(inode, name);
|
| 1854 | if (error == 0)
|
| 1855 | continue;
|
| 1856 | if (error < 0)
|
| 1857 | break;
|
| 1858 | if (ip->i_depth < GFS2_DIR_MAX_DEPTH) {
|
| 1859 | error = dir_double_exhash(ip);
|
| 1860 | if (error)
|
| 1861 | break;
|
| 1862 | error = dir_split_leaf(inode, name);
|
| 1863 | if (error < 0)
|
| 1864 | break;
|
| 1865 | if (error == 0)
|
| 1866 | continue;
|
| 1867 | }
|
| 1868 | error = dir_new_leaf(inode, name);
|
| 1869 | if (!error)
|
| 1870 | continue;
|
| 1871 | error = -ENOSPC;
|
| 1872 | break;
|
| 1873 | }
|
| 1874 | return error;
|
| 1875 | }
|
| 1876 |
|
| 1877 |
|
| 1878 | |
| 1879 | * gfs2_dir_del - Delete a directory entry
|
| 1880 | * @dip: The GFS2 inode
|
| 1881 | * @dentry: The directory entry we want to delete
|
| 1882 | *
|
| 1883 | * Returns: 0 on success, error code on failure
|
| 1884 |
|
| 1885 |
|
| 1886 | int gfs2_dir_del(struct gfs2_inode *dip, const struct dentry *dentry)
|
| 1887 | {
|
| 1888 | const struct qstr *name = &dentry->d_name;
|
| 1889 | struct gfs2_dirent *dent, *prev = NULL;
|
| 1890 | struct buffer_head *bh;
|
| 1891 | struct timespec64 tv;
|
| 1892 |
|
| 1893 | |
| 1894 | previous entry otherwise
|
| 1895 | dent = gfs2_dirent_search(&dip->i_inode, name, gfs2_dirent_prev, &bh);
|
| 1896 | if (!dent) {
|
| 1897 | gfs2_consist_inode(dip);
|
| 1898 | return -EIO;
|
| 1899 | }
|
| 1900 | if (IS_ERR(dent)) {
|
| 1901 | gfs2_consist_inode(dip);
|
| 1902 | return PTR_ERR(dent);
|
| 1903 | }
|
| 1904 |
|
| 1905 | if (gfs2_dirent_find(dent, name, NULL) == 0) {
|
| 1906 | prev = dent;
|
| 1907 | dent = (struct gfs2_dirent *)((char *)dent + be16_to_cpu(prev->de_rec_len));
|
| 1908 | }
|
| 1909 |
|
| 1910 | dirent_del(dip, bh, prev, dent);
|
| 1911 | tv = inode_set_ctime_current(&dip->i_inode);
|
| 1912 | if (dip->i_diskflags & GFS2_DIF_EXHASH) {
|
| 1913 | struct gfs2_leaf *leaf = (struct gfs2_leaf *)bh->b_data;
|
| 1914 | u16 entries = be16_to_cpu(leaf->lf_entries);
|
| 1915 | if (!entries)
|
| 1916 | gfs2_consist_inode(dip);
|
| 1917 | leaf->lf_entries = cpu_to_be16(--entries);
|
| 1918 | leaf->lf_nsec = cpu_to_be32(tv.tv_nsec);
|
| 1919 | leaf->lf_sec = cpu_to_be64(tv.tv_sec);
|
| 1920 | }
|
| 1921 | brelse(bh);
|
| 1922 |
|
| 1923 | if (!dip->i_entries)
|
| 1924 | gfs2_consist_inode(dip);
|
| 1925 | dip->i_entries--;
|
| 1926 | inode_set_mtime_to_ts(&dip->i_inode, tv);
|
| 1927 | if (d_is_dir(dentry))
|
| 1928 | drop_nlink(&dip->i_inode);
|
| 1929 | mark_inode_dirty(&dip->i_inode);
|
| 1930 |
|
| 1931 | return 0;
|
| 1932 | }
|
| 1933 |
|
| 1934 | |
| 1935 | * gfs2_dir_mvino - Change inode number of directory entry
|
| 1936 | * @dip: The GFS2 directory inode
|
| 1937 | * @filename: the filename to be moved
|
| 1938 | * @nip: the new GFS2 inode
|
| 1939 | * @new_type: the de_type of the new dirent
|
| 1940 | *
|
| 1941 | * This routine changes the inode number of a directory entry. It's used
|
| 1942 | * by rename to change ".." when a directory is moved.
|
| 1943 | * Assumes a glock is held on dvp.
|
| 1944 | *
|
| 1945 | * Returns: errno
|
| 1946 |
|
| 1947 |
|
| 1948 | int gfs2_dir_mvino(struct gfs2_inode *dip, const struct qstr *filename,
|
| 1949 | const struct gfs2_inode *nip, unsigned int new_type)
|
| 1950 | {
|
| 1951 | struct gfs2_glock *gl = gfs2_inode_glock(&dip->i_inode);
|
| 1952 | struct buffer_head *bh;
|
| 1953 | struct gfs2_dirent *dent;
|
| 1954 |
|
| 1955 | dent = gfs2_dirent_search(&dip->i_inode, filename, gfs2_dirent_find, &bh);
|
| 1956 | if (!dent) {
|
| 1957 | gfs2_consist_inode(dip);
|
| 1958 | return -EIO;
|
| 1959 | }
|
| 1960 | if (IS_ERR(dent))
|
| 1961 | return PTR_ERR(dent);
|
| 1962 |
|
| 1963 | gfs2_trans_add_meta(gl, bh);
|
| 1964 | gfs2_inum_out(nip, dent);
|
| 1965 | dent->de_type = cpu_to_be16(new_type);
|
| 1966 | brelse(bh);
|
| 1967 |
|
| 1968 | inode_set_mtime_to_ts(&dip->i_inode, inode_set_ctime_current(&dip->i_inode));
|
| 1969 | mark_inode_dirty_sync(&dip->i_inode);
|
| 1970 | return 0;
|
| 1971 | }
|
| 1972 |
|
| 1973 | |
| 1974 | * leaf_dealloc - Deallocate a directory leaf
|
| 1975 | * @dip: the directory
|
| 1976 | * @index: the hash table offset in the directory
|
| 1977 | * @len: the number of pointers to this leaf
|
| 1978 | * @leaf_no: the leaf number
|
| 1979 | * @leaf_bh: buffer_head for the starting leaf
|
| 1980 | * @last_dealloc: 1 if this is the final dealloc for the leaf, else 0
|
| 1981 | *
|
| 1982 | * Returns: errno
|
| 1983 |
|
| 1984 |
|
| 1985 | static int leaf_dealloc(struct gfs2_inode *dip, u32 index, u32 len,
|
| 1986 | u64 leaf_no, struct buffer_head *leaf_bh,
|
| 1987 | int last_dealloc)
|
| 1988 | {
|
| 1989 | struct gfs2_glock *gl = gfs2_inode_glock(&dip->i_inode);
|
| 1990 | struct gfs2_sbd *sdp = GFS2_SB(&dip->i_inode);
|
| 1991 | struct gfs2_leaf *tmp_leaf;
|
| 1992 | struct gfs2_rgrp_list rlist;
|
| 1993 | struct buffer_head *bh, *dibh;
|
| 1994 | u64 blk, nblk;
|
| 1995 | unsigned int rg_blocks = 0, l_blocks = 0;
|
| 1996 | char *ht;
|
| 1997 | unsigned int x, size = len * sizeof(u64);
|
| 1998 | int error;
|
| 1999 |
|
| 2000 | error = gfs2_rindex_update(sdp);
|
| 2001 | if (error)
|
| 2002 | return error;
|
| 2003 |
|
| 2004 | memset(&rlist, 0, sizeof(struct gfs2_rgrp_list));
|
| 2005 |
|
| 2006 | ht = kzalloc(size, GFP_NOFS | __GFP_NOWARN);
|
| 2007 | if (ht == NULL)
|
| 2008 | ht = __vmalloc(size, GFP_NOFS | __GFP_NOWARN | __GFP_ZERO);
|
| 2009 | if (!ht)
|
| 2010 | return -ENOMEM;
|
| 2011 |
|
| 2012 | error = gfs2_quota_hold(dip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
|
| 2013 | if (error)
|
| 2014 | goto out;
|
| 2015 |
|
| 2016 |
|
| 2017 | bh = leaf_bh;
|
| 2018 |
|
| 2019 | for (blk = leaf_no; blk; blk = nblk) {
|
| 2020 | if (blk != leaf_no) {
|
| 2021 | error = get_leaf(dip, blk, &bh);
|
| 2022 | if (error)
|
| 2023 | goto out_rlist;
|
| 2024 | }
|
| 2025 | tmp_leaf = (struct gfs2_leaf *)bh->b_data;
|
| 2026 | nblk = be64_to_cpu(tmp_leaf->lf_next);
|
| 2027 | if (blk != leaf_no)
|
| 2028 | brelse(bh);
|
| 2029 |
|
| 2030 | gfs2_rlist_add(dip, &rlist, blk);
|
| 2031 | l_blocks++;
|
| 2032 | }
|
| 2033 |
|
| 2034 | gfs2_rlist_alloc(&rlist, LM_ST_EXCLUSIVE, LM_FLAG_NODE_SCOPE);
|
| 2035 |
|
| 2036 | for (x = 0; x < rlist.rl_rgrps; x++) {
|
| 2037 | struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(rlist.rl_ghs[x].gh_gl);
|
| 2038 |
|
| 2039 | rg_blocks += rgd->rd_length;
|
| 2040 | }
|
| 2041 |
|
| 2042 | error = gfs2_glock_nq_m(rlist.rl_rgrps, rlist.rl_ghs);
|
| 2043 | if (error)
|
| 2044 | goto out_rlist;
|
| 2045 |
|
| 2046 | error = gfs2_trans_begin(sdp,
|
| 2047 | rg_blocks + (DIV_ROUND_UP(size, sdp->sd_jbsize) + 1) +
|
| 2048 | RES_DINODE + RES_STATFS + RES_QUOTA, RES_DINODE +
|
| 2049 | l_blocks);
|
| 2050 | if (error)
|
| 2051 | goto out_rg_gunlock;
|
| 2052 |
|
| 2053 | bh = leaf_bh;
|
| 2054 |
|
| 2055 | for (blk = leaf_no; blk; blk = nblk) {
|
| 2056 | struct gfs2_rgrpd *rgd;
|
| 2057 |
|
| 2058 | if (blk != leaf_no) {
|
| 2059 | error = get_leaf(dip, blk, &bh);
|
| 2060 | if (error)
|
| 2061 | goto out_end_trans;
|
| 2062 | }
|
| 2063 | tmp_leaf = (struct gfs2_leaf *)bh->b_data;
|
| 2064 | nblk = be64_to_cpu(tmp_leaf->lf_next);
|
| 2065 | if (blk != leaf_no)
|
| 2066 | brelse(bh);
|
| 2067 |
|
| 2068 | rgd = gfs2_blk2rgrpd(sdp, blk, true);
|
| 2069 | gfs2_free_meta(dip, rgd, blk, 1);
|
| 2070 | gfs2_add_inode_blocks(&dip->i_inode, -1);
|
| 2071 | }
|
| 2072 |
|
| 2073 | error = gfs2_dir_write_data(dip, ht, index * sizeof(u64), size);
|
| 2074 | if (error != size) {
|
| 2075 | if (error >= 0)
|
| 2076 | error = -EIO;
|
| 2077 | goto out_end_trans;
|
| 2078 | }
|
| 2079 |
|
| 2080 | error = gfs2_meta_inode_buffer(dip, &dibh);
|
| 2081 | if (error)
|
| 2082 | goto out_end_trans;
|
| 2083 |
|
| 2084 | gfs2_trans_add_meta(gl, dibh);
|
| 2085 | |
| 2086 | (We don't want to free these blocks a second time.)
|
| 2087 | if (last_dealloc)
|
| 2088 | dip->i_inode.i_mode = S_IFREG;
|
| 2089 | gfs2_dinode_out(dip, dibh->b_data);
|
| 2090 | brelse(dibh);
|
| 2091 |
|
| 2092 | out_end_trans:
|
| 2093 | gfs2_trans_end(sdp);
|
| 2094 | out_rg_gunlock:
|
| 2095 | gfs2_glock_dq_m(rlist.rl_rgrps, rlist.rl_ghs);
|
| 2096 | out_rlist:
|
| 2097 | gfs2_rlist_free(&rlist);
|
| 2098 | gfs2_quota_unhold(dip);
|
| 2099 | out:
|
| 2100 | kvfree(ht);
|
| 2101 | return error;
|
| 2102 | }
|
| 2103 |
|
| 2104 | |
| 2105 | * gfs2_dir_exhash_dealloc - free all the leaf blocks in a directory
|
| 2106 | * @dip: the directory
|
| 2107 | *
|
| 2108 | * Dealloc all on-disk directory leaves to FREEMETA state
|
| 2109 | * Change on-disk inode type to "regular file"
|
| 2110 | *
|
| 2111 | * Returns: errno
|
| 2112 |
|
| 2113 |
|
| 2114 | int gfs2_dir_exhash_dealloc(struct gfs2_inode *dip)
|
| 2115 | {
|
| 2116 | struct buffer_head *bh;
|
| 2117 | struct gfs2_leaf *leaf;
|
| 2118 | u32 hsize, len;
|
| 2119 | u32 index = 0, next_index;
|
| 2120 | __be64 *lp;
|
| 2121 | u64 leaf_no;
|
| 2122 | int error = 0, last;
|
| 2123 |
|
| 2124 | hsize = BIT(dip->i_depth);
|
| 2125 |
|
| 2126 | lp = gfs2_dir_get_hash_table(dip);
|
| 2127 | if (IS_ERR(lp))
|
| 2128 | return PTR_ERR(lp);
|
| 2129 |
|
| 2130 | while (index < hsize) {
|
| 2131 | leaf_no = be64_to_cpu(lp[index]);
|
| 2132 | if (leaf_no) {
|
| 2133 | error = get_leaf(dip, leaf_no, &bh);
|
| 2134 | if (error)
|
| 2135 | goto out;
|
| 2136 | leaf = (struct gfs2_leaf *)bh->b_data;
|
| 2137 | len = BIT(dip->i_depth - be16_to_cpu(leaf->lf_depth));
|
| 2138 |
|
| 2139 | next_index = (index & ~(len - 1)) + len;
|
| 2140 | last = ((next_index >= hsize) ? 1 : 0);
|
| 2141 | error = leaf_dealloc(dip, index, len, leaf_no, bh,
|
| 2142 | last);
|
| 2143 | brelse(bh);
|
| 2144 | if (error)
|
| 2145 | goto out;
|
| 2146 | index = next_index;
|
| 2147 | } else
|
| 2148 | index++;
|
| 2149 | }
|
| 2150 |
|
| 2151 | if (index != hsize) {
|
| 2152 | gfs2_consist_inode(dip);
|
| 2153 | error = -EIO;
|
| 2154 | }
|
| 2155 |
|
| 2156 | out:
|
| 2157 |
|
| 2158 | return error;
|
| 2159 | }
|
| 2160 |
|
| 2161 | |
| 2162 | * gfs2_diradd_alloc_required - find if adding entry will require an allocation
|
| 2163 | * @inode: the directory inode being written to
|
| 2164 | * @name: the filename that's going to be added
|
| 2165 | * @da: The structure to return dir alloc info
|
| 2166 | *
|
| 2167 | * Returns: 0 if ok, -ve on error
|
| 2168 |
|
| 2169 |
|
| 2170 | int gfs2_diradd_alloc_required(struct inode *inode, const struct qstr *name,
|
| 2171 | struct gfs2_diradd *da)
|
| 2172 | {
|
| 2173 | struct gfs2_inode *ip = GFS2_I(inode);
|
| 2174 | struct gfs2_sbd *sdp = GFS2_SB(inode);
|
| 2175 | const unsigned int extra = sizeof(struct gfs2_dinode) - sizeof(struct gfs2_leaf);
|
| 2176 | struct gfs2_dirent *dent;
|
| 2177 | struct buffer_head *bh;
|
| 2178 |
|
| 2179 | da->nr_blocks = 0;
|
| 2180 | da->bh = NULL;
|
| 2181 | da->dent = NULL;
|
| 2182 |
|
| 2183 | dent = gfs2_dirent_search(inode, name, gfs2_dirent_find_space, &bh);
|
| 2184 | if (!dent) {
|
| 2185 | da->nr_blocks = sdp->sd_max_dirres;
|
| 2186 | if (!(ip->i_diskflags & GFS2_DIF_EXHASH) &&
|
| 2187 | (GFS2_DIRENT_SIZE(name->len) < extra))
|
| 2188 | da->nr_blocks = 1;
|
| 2189 | return 0;
|
| 2190 | }
|
| 2191 | if (IS_ERR(dent))
|
| 2192 | return PTR_ERR(dent);
|
| 2193 |
|
| 2194 | if (da->save_loc) {
|
| 2195 | da->bh = bh;
|
| 2196 | da->dent = dent;
|
| 2197 | } else {
|
| 2198 | brelse(bh);
|
| 2199 | }
|
| 2200 | return 0;
|
| 2201 | }
|
| 2202 |
|