| 1 | |
| 2 | |
| 3 | * suballoc.c
|
| 4 | *
|
| 5 | * metadata alloc and free
|
| 6 | * Inspired by ext3 block groups.
|
| 7 | *
|
| 8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved.
|
| 9 |
|
| 10 |
|
| 11 | #include <linux/fs.h>
|
| 12 | #include <linux/types.h>
|
| 13 | #include <linux/slab.h>
|
| 14 | #include <linux/string.h>
|
| 15 | #include <linux/highmem.h>
|
| 16 |
|
| 17 | #include <cluster/masklog.h>
|
| 18 |
|
| 19 | #include "ocfs2.h"
|
| 20 |
|
| 21 | #include "alloc.h"
|
| 22 | #include "blockcheck.h"
|
| 23 | #include "dlmglue.h"
|
| 24 | #include "inode.h"
|
| 25 | #include "journal.h"
|
| 26 | #include "localalloc.h"
|
| 27 | #include "suballoc.h"
|
| 28 | #include "super.h"
|
| 29 | #include "sysfile.h"
|
| 30 | #include "uptodate.h"
|
| 31 | #include "ocfs2_trace.h"
|
| 32 |
|
| 33 | #include "buffer_head_io.h"
|
| 34 |
|
| 35 | #define NOT_ALLOC_NEW_GROUP 0
|
| 36 | #define ALLOC_NEW_GROUP 0x1
|
| 37 | #define ALLOC_GROUPS_FROM_GLOBAL 0x2
|
| 38 |
|
| 39 | #define OCFS2_MAX_TO_STEAL 1024
|
| 40 |
|
| 41 | struct ocfs2_suballoc_result {
|
| 42 | u64 sr_bg_blkno; |
| 43 | to 0 when a block group is
|
| 44 | contiguous.
|
| 45 | u64 sr_bg_stable_blkno; |
| 46 | * Doesn't change, always
|
| 47 | * set to target block
|
| 48 | * group descriptor
|
| 49 | * block.
|
| 50 |
|
| 51 | u64 sr_blkno;
|
| 52 | unsigned int sr_bit_offset;
|
| 53 | unsigned int sr_bits;
|
| 54 | unsigned int sr_max_contig_bits; |
| 55 | * free bits, only available
|
| 56 | * for cluster group
|
| 57 |
|
| 58 | };
|
| 59 |
|
| 60 | static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
|
| 61 | {
|
| 62 | if (res->sr_blkno == 0)
|
| 63 | return 0;
|
| 64 |
|
| 65 | if (res->sr_bg_blkno)
|
| 66 | return res->sr_bg_blkno;
|
| 67 |
|
| 68 | return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
|
| 69 | }
|
| 70 |
|
| 71 | static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
|
| 72 | static int ocfs2_block_group_fill(handle_t *handle,
|
| 73 | struct inode *alloc_inode,
|
| 74 | struct buffer_head *bg_bh,
|
| 75 | u64 group_blkno,
|
| 76 | unsigned int group_clusters,
|
| 77 | u16 my_chain,
|
| 78 | struct ocfs2_chain_list *cl);
|
| 79 | static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
|
| 80 | struct inode *alloc_inode,
|
| 81 | struct buffer_head *bh,
|
| 82 | u64 max_block,
|
| 83 | u64 *last_alloc_group,
|
| 84 | int flags);
|
| 85 |
|
| 86 | static int ocfs2_cluster_group_search(struct inode *inode,
|
| 87 | struct buffer_head *group_bh,
|
| 88 | u32 bits_wanted, u32 min_bits,
|
| 89 | u64 max_block,
|
| 90 | struct ocfs2_suballoc_result *res);
|
| 91 | static int ocfs2_block_group_search(struct inode *inode,
|
| 92 | struct buffer_head *group_bh,
|
| 93 | u32 bits_wanted, u32 min_bits,
|
| 94 | u64 max_block,
|
| 95 | struct ocfs2_suballoc_result *res);
|
| 96 | static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
|
| 97 | handle_t *handle,
|
| 98 | u32 bits_wanted,
|
| 99 | u32 min_bits,
|
| 100 | struct ocfs2_suballoc_result *res);
|
| 101 | static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
|
| 102 | int nr);
|
| 103 | static int ocfs2_relink_block_group(handle_t *handle,
|
| 104 | struct inode *alloc_inode,
|
| 105 | struct buffer_head *fe_bh,
|
| 106 | struct buffer_head *bg_bh,
|
| 107 | struct buffer_head *prev_bg_bh,
|
| 108 | u16 chain);
|
| 109 | static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
|
| 110 | u32 wanted);
|
| 111 | static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
|
| 112 | u64 bg_blkno,
|
| 113 | u16 bg_bit_off);
|
| 114 | static inline void ocfs2_block_to_cluster_group(struct inode *inode,
|
| 115 | u64 data_blkno,
|
| 116 | u64 *bg_blkno,
|
| 117 | u16 *bg_bit_off);
|
| 118 | static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
|
| 119 | u32 bits_wanted, u64 max_block,
|
| 120 | int flags,
|
| 121 | struct ocfs2_alloc_context **ac);
|
| 122 |
|
| 123 | void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
|
| 124 | {
|
| 125 | struct inode *inode = ac->ac_inode;
|
| 126 |
|
| 127 | if (inode) {
|
| 128 | if (ac->ac_which != OCFS2_AC_USE_LOCAL)
|
| 129 | ocfs2_inode_unlock(inode, 1);
|
| 130 |
|
| 131 | inode_unlock(inode);
|
| 132 |
|
| 133 | iput(inode);
|
| 134 | ac->ac_inode = NULL;
|
| 135 | }
|
| 136 | brelse(ac->ac_bh);
|
| 137 | ac->ac_bh = NULL;
|
| 138 | ac->ac_resv = NULL;
|
| 139 | kfree(ac->ac_find_loc_priv);
|
| 140 | ac->ac_find_loc_priv = NULL;
|
| 141 | }
|
| 142 |
|
| 143 | void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
|
| 144 | {
|
| 145 | ocfs2_free_ac_resource(ac);
|
| 146 | kfree(ac);
|
| 147 | }
|
| 148 |
|
| 149 | static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
|
| 150 | {
|
| 151 | return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
|
| 152 | }
|
| 153 |
|
| 154 | #define do_error(fmt, ...) \
|
| 155 | do { \
|
| 156 | if (resize) \
|
| 157 | mlog(ML_ERROR, fmt, ##__VA_ARGS__); \
|
| 158 | else \
|
| 159 | return ocfs2_error(sb, fmt, ##__VA_ARGS__); \
|
| 160 | } while (0)
|
| 161 |
|
| 162 | static int ocfs2_validate_gd_self(struct super_block *sb,
|
| 163 | struct buffer_head *bh,
|
| 164 | int resize)
|
| 165 | {
|
| 166 | struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
|
| 167 |
|
| 168 | if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
|
| 169 | do_error("Group descriptor #%llu has bad signature %.*s\n",
|
| 170 | (unsigned long long)bh->b_blocknr, 7,
|
| 171 | gd->bg_signature);
|
| 172 | }
|
| 173 |
|
| 174 | if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
|
| 175 | do_error("Group descriptor #%llu has an invalid bg_blkno of %llu\n",
|
| 176 | (unsigned long long)bh->b_blocknr,
|
| 177 | (unsigned long long)le64_to_cpu(gd->bg_blkno));
|
| 178 | }
|
| 179 |
|
| 180 | if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
|
| 181 | do_error("Group descriptor #%llu has an invalid fs_generation of #%u\n",
|
| 182 | (unsigned long long)bh->b_blocknr,
|
| 183 | le32_to_cpu(gd->bg_generation));
|
| 184 | }
|
| 185 |
|
| 186 | if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
|
| 187 | do_error("Group descriptor #%llu has bit count %u but claims that %u are free\n",
|
| 188 | (unsigned long long)bh->b_blocknr,
|
| 189 | le16_to_cpu(gd->bg_bits),
|
| 190 | le16_to_cpu(gd->bg_free_bits_count));
|
| 191 | }
|
| 192 |
|
| 193 | if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
|
| 194 | do_error("Group descriptor #%llu has bit count %u but max bitmap bits of %u\n",
|
| 195 | (unsigned long long)bh->b_blocknr,
|
| 196 | le16_to_cpu(gd->bg_bits),
|
| 197 | 8 * le16_to_cpu(gd->bg_size));
|
| 198 | }
|
| 199 |
|
| 200 | |
| 201 | * For discontiguous block groups, validate the on-disk extent list
|
| 202 | * against the maximum number of extent records that can physically
|
| 203 | * fit in a single block.
|
| 204 |
|
| 205 | if (ocfs2_gd_is_discontig(gd)) {
|
| 206 | u16 max_recs = ocfs2_extent_recs_per_gd(sb);
|
| 207 | u16 l_count = le16_to_cpu(gd->bg_list.l_count);
|
| 208 | u16 l_next_free_rec = le16_to_cpu(gd->bg_list.l_next_free_rec);
|
| 209 |
|
| 210 | if (l_count != max_recs) {
|
| 211 | do_error("Group descriptor #%llu bad discontig l_count %u expected %u\n",
|
| 212 | (unsigned long long)bh->b_blocknr,
|
| 213 | l_count,
|
| 214 | max_recs);
|
| 215 | }
|
| 216 |
|
| 217 | if (l_next_free_rec > l_count) {
|
| 218 | do_error("Group descriptor #%llu bad discontig l_next_free_rec %u max %u\n",
|
| 219 | (unsigned long long)bh->b_blocknr,
|
| 220 | l_next_free_rec,
|
| 221 | l_count);
|
| 222 | }
|
| 223 | }
|
| 224 |
|
| 225 | return 0;
|
| 226 | }
|
| 227 |
|
| 228 | static int ocfs2_validate_gd_parent(struct super_block *sb,
|
| 229 | struct ocfs2_dinode *di,
|
| 230 | struct buffer_head *bh,
|
| 231 | int resize)
|
| 232 | {
|
| 233 | unsigned int max_bits;
|
| 234 | unsigned int max_bitmap_bits;
|
| 235 | unsigned int max_bitmap_size;
|
| 236 | int suballocator;
|
| 237 | struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
|
| 238 |
|
| 239 | suballocator = le64_to_cpu(di->i_blkno) != OCFS2_SB(sb)->bitmap_blkno;
|
| 240 | max_bitmap_size = ocfs2_group_bitmap_size(sb, suballocator,
|
| 241 | OCFS2_SB(sb)->s_feature_incompat);
|
| 242 | max_bitmap_bits = max_bitmap_size * 8;
|
| 243 |
|
| 244 | if (di->i_blkno != gd->bg_parent_dinode) {
|
| 245 | do_error("Group descriptor #%llu has bad parent pointer (%llu, expected %llu)\n",
|
| 246 | (unsigned long long)bh->b_blocknr,
|
| 247 | (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
|
| 248 | (unsigned long long)le64_to_cpu(di->i_blkno));
|
| 249 | }
|
| 250 |
|
| 251 | if (le16_to_cpu(gd->bg_size) > max_bitmap_size) {
|
| 252 | do_error("Group descriptor #%llu has bitmap size %u but physical max of %u\n",
|
| 253 | (unsigned long long)bh->b_blocknr,
|
| 254 | le16_to_cpu(gd->bg_size),
|
| 255 | max_bitmap_size);
|
| 256 | }
|
| 257 |
|
| 258 | if (le16_to_cpu(gd->bg_bits) > max_bitmap_bits) {
|
| 259 | do_error("Group descriptor #%llu has bit count %u but physical max of %u\n",
|
| 260 | (unsigned long long)bh->b_blocknr,
|
| 261 | le16_to_cpu(gd->bg_bits),
|
| 262 | max_bitmap_bits);
|
| 263 | }
|
| 264 |
|
| 265 | max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
|
| 266 | if (le16_to_cpu(gd->bg_bits) > max_bits) {
|
| 267 | do_error("Group descriptor #%llu has bit count of %u\n",
|
| 268 | (unsigned long long)bh->b_blocknr,
|
| 269 | le16_to_cpu(gd->bg_bits));
|
| 270 | }
|
| 271 |
|
| 272 |
|
| 273 | if ((le16_to_cpu(gd->bg_chain) >
|
| 274 | le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
|
| 275 | ((le16_to_cpu(gd->bg_chain) ==
|
| 276 | le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
|
| 277 | do_error("Group descriptor #%llu has bad chain %u\n",
|
| 278 | (unsigned long long)bh->b_blocknr,
|
| 279 | le16_to_cpu(gd->bg_chain));
|
| 280 | }
|
| 281 |
|
| 282 | return 0;
|
| 283 | }
|
| 284 |
|
| 285 | #undef do_error
|
| 286 |
|
| 287 | |
| 288 | * This version only prints errors. It does not fail the filesystem, and
|
| 289 | * exists only for resize.
|
| 290 |
|
| 291 | int ocfs2_check_group_descriptor(struct super_block *sb,
|
| 292 | struct ocfs2_dinode *di,
|
| 293 | struct buffer_head *bh)
|
| 294 | {
|
| 295 | int rc;
|
| 296 | struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
|
| 297 |
|
| 298 | BUG_ON(!buffer_uptodate(bh));
|
| 299 |
|
| 300 | |
| 301 | * If the ecc fails, we return the error but otherwise
|
| 302 | * leave the filesystem running. We know any error is
|
| 303 | * local to this block.
|
| 304 |
|
| 305 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
|
| 306 | if (rc) {
|
| 307 | mlog(ML_ERROR,
|
| 308 | "Checksum failed for group descriptor %llu\n",
|
| 309 | (unsigned long long)bh->b_blocknr);
|
| 310 | } else
|
| 311 | rc = ocfs2_validate_gd_self(sb, bh, 1);
|
| 312 | if (!rc)
|
| 313 | rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
|
| 314 |
|
| 315 | return rc;
|
| 316 | }
|
| 317 |
|
| 318 | static int ocfs2_validate_group_descriptor(struct super_block *sb,
|
| 319 | struct buffer_head *bh)
|
| 320 | {
|
| 321 | int rc;
|
| 322 | struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
|
| 323 |
|
| 324 | trace_ocfs2_validate_group_descriptor(
|
| 325 | (unsigned long long)bh->b_blocknr);
|
| 326 |
|
| 327 | BUG_ON(!buffer_uptodate(bh));
|
| 328 |
|
| 329 | |
| 330 | * If the ecc fails, we return the error but otherwise
|
| 331 | * leave the filesystem running. We know any error is
|
| 332 | * local to this block.
|
| 333 |
|
| 334 | rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
|
| 335 | if (rc)
|
| 336 | return rc;
|
| 337 |
|
| 338 | |
| 339 | * Errors after here are fatal.
|
| 340 |
|
| 341 |
|
| 342 | return ocfs2_validate_gd_self(sb, bh, 0);
|
| 343 | }
|
| 344 |
|
| 345 | |
| 346 | * The hint group descriptor (gd) may already have been released
|
| 347 | * in _ocfs2_free_suballoc_bits(). We first check the gd signature,
|
| 348 | * then perform the standard ocfs2_read_group_descriptor() jobs.
|
| 349 | *
|
| 350 | * If the gd signature is invalid, we return 'rc=0' and set
|
| 351 | * '*released=1'. The caller is expected to handle this specific case.
|
| 352 | * Otherwise, we return the actual error code.
|
| 353 | *
|
| 354 | * We treat gd signature corruption case as a release case. The
|
| 355 | * caller ocfs2_claim_suballoc_bits() will use ocfs2_search_chain()
|
| 356 | * to search each gd block. The code will eventually find this
|
| 357 | * corrupted gd block - Late, but not missed.
|
| 358 | *
|
| 359 | * Note:
|
| 360 | * The caller is responsible for initializing the '*released' status.
|
| 361 |
|
| 362 | static int ocfs2_read_hint_group_descriptor(struct inode *inode,
|
| 363 | struct ocfs2_dinode *di, u64 gd_blkno,
|
| 364 | struct buffer_head **bh, int *released)
|
| 365 | {
|
| 366 | int rc;
|
| 367 | struct buffer_head *tmp = *bh;
|
| 368 | struct ocfs2_group_desc *gd;
|
| 369 |
|
| 370 | rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp, NULL);
|
| 371 | if (rc)
|
| 372 | goto out;
|
| 373 |
|
| 374 | gd = (struct ocfs2_group_desc *) tmp->b_data;
|
| 375 | if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
|
| 376 | |
| 377 | * Invalid gd cache was set in ocfs2_read_block(),
|
| 378 | * which will affect block_group allocation.
|
| 379 | * Path:
|
| 380 | * ocfs2_reserve_suballoc_bits
|
| 381 | * ocfs2_block_group_alloc
|
| 382 | * ocfs2_block_group_alloc_contig
|
| 383 | * ocfs2_set_new_buffer_uptodate
|
| 384 |
|
| 385 | ocfs2_remove_from_cache(INODE_CACHE(inode), tmp);
|
| 386 | *released = 1;
|
| 387 | goto free_bh;
|
| 388 | }
|
| 389 |
|
| 390 |
|
| 391 | if (!buffer_jbd(tmp)) {
|
| 392 | rc = ocfs2_validate_group_descriptor(inode->i_sb, tmp);
|
| 393 | if (rc)
|
| 394 | goto free_bh;
|
| 395 | }
|
| 396 |
|
| 397 | rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
|
| 398 | if (rc)
|
| 399 | goto free_bh;
|
| 400 |
|
| 401 |
|
| 402 | if (!*bh)
|
| 403 | *bh = tmp;
|
| 404 |
|
| 405 | return rc;
|
| 406 |
|
| 407 | free_bh:
|
| 408 | brelse(tmp);
|
| 409 | out:
|
| 410 | return rc;
|
| 411 | }
|
| 412 |
|
| 413 | int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
|
| 414 | u64 gd_blkno, struct buffer_head **bh)
|
| 415 | {
|
| 416 | int rc;
|
| 417 | struct buffer_head *tmp = *bh;
|
| 418 |
|
| 419 | rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
|
| 420 | ocfs2_validate_group_descriptor);
|
| 421 | if (rc)
|
| 422 | goto out;
|
| 423 |
|
| 424 | rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
|
| 425 | if (rc) {
|
| 426 | brelse(tmp);
|
| 427 | goto out;
|
| 428 | }
|
| 429 |
|
| 430 |
|
| 431 | if (!*bh)
|
| 432 | *bh = tmp;
|
| 433 |
|
| 434 | out:
|
| 435 | return rc;
|
| 436 | }
|
| 437 |
|
| 438 | static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
|
| 439 | struct ocfs2_group_desc *bg,
|
| 440 | struct ocfs2_chain_list *cl,
|
| 441 | u64 p_blkno, unsigned int clusters)
|
| 442 | {
|
| 443 | struct ocfs2_extent_list *el = &bg->bg_list;
|
| 444 | struct ocfs2_extent_rec *rec;
|
| 445 |
|
| 446 | BUG_ON(!ocfs2_supports_discontig_bg(osb));
|
| 447 | if (!el->l_next_free_rec)
|
| 448 | el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
|
| 449 | rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
|
| 450 | rec->e_blkno = cpu_to_le64(p_blkno);
|
| 451 | rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
|
| 452 | le16_to_cpu(cl->cl_bpc));
|
| 453 | rec->e_leaf_clusters = cpu_to_le16(clusters);
|
| 454 | le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
|
| 455 | le16_add_cpu(&bg->bg_free_bits_count,
|
| 456 | clusters * le16_to_cpu(cl->cl_bpc));
|
| 457 | le16_add_cpu(&el->l_next_free_rec, 1);
|
| 458 | }
|
| 459 |
|
| 460 | static int ocfs2_block_group_fill(handle_t *handle,
|
| 461 | struct inode *alloc_inode,
|
| 462 | struct buffer_head *bg_bh,
|
| 463 | u64 group_blkno,
|
| 464 | unsigned int group_clusters,
|
| 465 | u16 my_chain,
|
| 466 | struct ocfs2_chain_list *cl)
|
| 467 | {
|
| 468 | int status = 0;
|
| 469 | struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
|
| 470 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
|
| 471 | struct super_block * sb = alloc_inode->i_sb;
|
| 472 |
|
| 473 | if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
|
| 474 | status = ocfs2_error(alloc_inode->i_sb,
|
| 475 | "group block (%llu) != b_blocknr (%llu)\n",
|
| 476 | (unsigned long long)group_blkno,
|
| 477 | (unsigned long long) bg_bh->b_blocknr);
|
| 478 | goto bail;
|
| 479 | }
|
| 480 |
|
| 481 | status = ocfs2_journal_access_gd(handle,
|
| 482 | INODE_CACHE(alloc_inode),
|
| 483 | bg_bh,
|
| 484 | OCFS2_JOURNAL_ACCESS_CREATE);
|
| 485 | if (status < 0) {
|
| 486 | mlog_errno(status);
|
| 487 | goto bail;
|
| 488 | }
|
| 489 |
|
| 490 | memset(bg, 0, sb->s_blocksize);
|
| 491 | strscpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
|
| 492 | bg->bg_generation = cpu_to_le32(osb->fs_generation);
|
| 493 | bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
|
| 494 | osb->s_feature_incompat));
|
| 495 | bg->bg_chain = cpu_to_le16(my_chain);
|
| 496 | bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
|
| 497 | bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
|
| 498 | bg->bg_blkno = cpu_to_le64(group_blkno);
|
| 499 | if (group_clusters == le16_to_cpu(cl->cl_cpg))
|
| 500 | bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
|
| 501 | else
|
| 502 | ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
|
| 503 | group_clusters);
|
| 504 |
|
| 505 |
|
| 506 | ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
|
| 507 | bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
|
| 508 |
|
| 509 | ocfs2_journal_dirty(handle, bg_bh);
|
| 510 |
|
| 511 | |
| 512 | * other blocks in a group - All valid FS metadata in a block
|
| 513 | * group stores the superblock fs_generation value at
|
| 514 | * allocation time.
|
| 515 |
|
| 516 | bail:
|
| 517 | if (status)
|
| 518 | mlog_errno(status);
|
| 519 | return status;
|
| 520 | }
|
| 521 |
|
| 522 | static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
|
| 523 | {
|
| 524 | u16 curr, best;
|
| 525 |
|
| 526 | best = curr = 0;
|
| 527 | while (curr < le16_to_cpu(cl->cl_count)) {
|
| 528 | if (le32_to_cpu(cl->cl_recs[best].c_total) >
|
| 529 | le32_to_cpu(cl->cl_recs[curr].c_total))
|
| 530 | best = curr;
|
| 531 | curr++;
|
| 532 | }
|
| 533 | return best;
|
| 534 | }
|
| 535 |
|
| 536 | static struct buffer_head *
|
| 537 | ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
|
| 538 | struct inode *alloc_inode,
|
| 539 | struct ocfs2_alloc_context *ac,
|
| 540 | struct ocfs2_chain_list *cl)
|
| 541 | {
|
| 542 | int status;
|
| 543 | u32 bit_off, num_bits;
|
| 544 | u64 bg_blkno;
|
| 545 | struct buffer_head *bg_bh;
|
| 546 | unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
|
| 547 |
|
| 548 | status = ocfs2_claim_clusters(handle, ac,
|
| 549 | le16_to_cpu(cl->cl_cpg), &bit_off,
|
| 550 | &num_bits);
|
| 551 | if (status < 0) {
|
| 552 | if (status != -ENOSPC)
|
| 553 | mlog_errno(status);
|
| 554 | goto bail;
|
| 555 | }
|
| 556 |
|
| 557 |
|
| 558 | bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
|
| 559 | trace_ocfs2_block_group_alloc_contig(
|
| 560 | (unsigned long long)bg_blkno, alloc_rec);
|
| 561 |
|
| 562 | bg_bh = sb_getblk(osb->sb, bg_blkno);
|
| 563 | if (!bg_bh) {
|
| 564 | status = -ENOMEM;
|
| 565 | mlog_errno(status);
|
| 566 | goto bail;
|
| 567 | }
|
| 568 | ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
|
| 569 |
|
| 570 | status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
|
| 571 | bg_blkno, num_bits, alloc_rec, cl);
|
| 572 | if (status < 0) {
|
| 573 | brelse(bg_bh);
|
| 574 | mlog_errno(status);
|
| 575 | }
|
| 576 |
|
| 577 | bail:
|
| 578 | return status ? ERR_PTR(status) : bg_bh;
|
| 579 | }
|
| 580 |
|
| 581 | static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
|
| 582 | handle_t *handle,
|
| 583 | struct ocfs2_alloc_context *ac,
|
| 584 | unsigned int min_bits,
|
| 585 | u32 *bit_off, u32 *num_bits)
|
| 586 | {
|
| 587 | int status = 0;
|
| 588 |
|
| 589 | while (min_bits) {
|
| 590 | status = ocfs2_claim_clusters(handle, ac, min_bits,
|
| 591 | bit_off, num_bits);
|
| 592 | if (status != -ENOSPC)
|
| 593 | break;
|
| 594 |
|
| 595 | min_bits >>= 1;
|
| 596 | }
|
| 597 |
|
| 598 | return status;
|
| 599 | }
|
| 600 |
|
| 601 | static int ocfs2_block_group_grow_discontig(handle_t *handle,
|
| 602 | struct inode *alloc_inode,
|
| 603 | struct buffer_head *bg_bh,
|
| 604 | struct ocfs2_alloc_context *ac,
|
| 605 | struct ocfs2_chain_list *cl,
|
| 606 | unsigned int min_bits)
|
| 607 | {
|
| 608 | int status;
|
| 609 | struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
|
| 610 | struct ocfs2_group_desc *bg =
|
| 611 | (struct ocfs2_group_desc *)bg_bh->b_data;
|
| 612 | unsigned int needed = le16_to_cpu(cl->cl_cpg) -
|
| 613 | le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
|
| 614 | u32 p_cpos, clusters;
|
| 615 | u64 p_blkno;
|
| 616 | struct ocfs2_extent_list *el = &bg->bg_list;
|
| 617 |
|
| 618 | status = ocfs2_journal_access_gd(handle,
|
| 619 | INODE_CACHE(alloc_inode),
|
| 620 | bg_bh,
|
| 621 | OCFS2_JOURNAL_ACCESS_CREATE);
|
| 622 | if (status < 0) {
|
| 623 | mlog_errno(status);
|
| 624 | goto bail;
|
| 625 | }
|
| 626 |
|
| 627 | while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
|
| 628 | le16_to_cpu(el->l_count))) {
|
| 629 | if (min_bits > needed)
|
| 630 | min_bits = needed;
|
| 631 | status = ocfs2_block_group_claim_bits(osb, handle, ac,
|
| 632 | min_bits, &p_cpos,
|
| 633 | &clusters);
|
| 634 | if (status < 0) {
|
| 635 | if (status != -ENOSPC)
|
| 636 | mlog_errno(status);
|
| 637 | goto bail;
|
| 638 | }
|
| 639 | p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
|
| 640 | ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
|
| 641 | clusters);
|
| 642 |
|
| 643 | min_bits = clusters;
|
| 644 | needed = le16_to_cpu(cl->cl_cpg) -
|
| 645 | le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
|
| 646 | }
|
| 647 |
|
| 648 | if (needed > 0) {
|
| 649 | |
| 650 | * We have used up all the extent rec but can't fill up
|
| 651 | * the cpg. So bail out.
|
| 652 |
|
| 653 | status = -ENOSPC;
|
| 654 | goto bail;
|
| 655 | }
|
| 656 |
|
| 657 | ocfs2_journal_dirty(handle, bg_bh);
|
| 658 |
|
| 659 | bail:
|
| 660 | return status;
|
| 661 | }
|
| 662 |
|
| 663 | static void ocfs2_bg_alloc_cleanup(handle_t *handle,
|
| 664 | struct ocfs2_alloc_context *cluster_ac,
|
| 665 | struct inode *alloc_inode,
|
| 666 | struct buffer_head *bg_bh)
|
| 667 | {
|
| 668 | int i, ret;
|
| 669 | struct ocfs2_group_desc *bg;
|
| 670 | struct ocfs2_extent_list *el;
|
| 671 | struct ocfs2_extent_rec *rec;
|
| 672 |
|
| 673 | if (!bg_bh)
|
| 674 | return;
|
| 675 |
|
| 676 | bg = (struct ocfs2_group_desc *)bg_bh->b_data;
|
| 677 | el = &bg->bg_list;
|
| 678 | for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
|
| 679 | rec = &el->l_recs[i];
|
| 680 | ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
|
| 681 | cluster_ac->ac_bh,
|
| 682 | le64_to_cpu(rec->e_blkno),
|
| 683 | le16_to_cpu(rec->e_leaf_clusters));
|
| 684 | if (ret)
|
| 685 | mlog_errno(ret);
|
| 686 |
|
| 687 | }
|
| 688 |
|
| 689 | ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
|
| 690 | brelse(bg_bh);
|
| 691 | }
|
| 692 |
|
| 693 | static struct buffer_head *
|
| 694 | ocfs2_block_group_alloc_discontig(handle_t *handle,
|
| 695 | struct inode *alloc_inode,
|
| 696 | struct ocfs2_alloc_context *ac,
|
| 697 | struct ocfs2_chain_list *cl)
|
| 698 | {
|
| 699 | int status;
|
| 700 | u32 bit_off, num_bits;
|
| 701 | u64 bg_blkno;
|
| 702 | unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
|
| 703 | struct buffer_head *bg_bh = NULL;
|
| 704 | unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
|
| 705 | struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
|
| 706 |
|
| 707 | if (!ocfs2_supports_discontig_bg(osb)) {
|
| 708 | status = -ENOSPC;
|
| 709 | goto bail;
|
| 710 | }
|
| 711 |
|
| 712 | status = ocfs2_extend_trans(handle,
|
| 713 | ocfs2_calc_bg_discontig_credits(osb->sb));
|
| 714 | if (status) {
|
| 715 | mlog_errno(status);
|
| 716 | goto bail;
|
| 717 | }
|
| 718 |
|
| 719 | |
| 720 | * We're going to be grabbing from multiple cluster groups.
|
| 721 | * We don't have enough credits to relink them all, and the
|
| 722 | * cluster groups will be staying in cache for the duration of
|
| 723 | * this operation.
|
| 724 |
|
| 725 | ac->ac_disable_chain_relink = 1;
|
| 726 |
|
| 727 |
|
| 728 | status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
|
| 729 | &bit_off, &num_bits);
|
| 730 | if (status < 0) {
|
| 731 | if (status != -ENOSPC)
|
| 732 | mlog_errno(status);
|
| 733 | goto bail;
|
| 734 | }
|
| 735 | min_bits = num_bits;
|
| 736 |
|
| 737 |
|
| 738 | bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
|
| 739 | trace_ocfs2_block_group_alloc_discontig(
|
| 740 | (unsigned long long)bg_blkno, alloc_rec);
|
| 741 |
|
| 742 | bg_bh = sb_getblk(osb->sb, bg_blkno);
|
| 743 | if (!bg_bh) {
|
| 744 | status = -ENOMEM;
|
| 745 | mlog_errno(status);
|
| 746 | goto bail;
|
| 747 | }
|
| 748 | ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
|
| 749 |
|
| 750 | status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
|
| 751 | bg_blkno, num_bits, alloc_rec, cl);
|
| 752 | if (status < 0) {
|
| 753 | mlog_errno(status);
|
| 754 | goto bail;
|
| 755 | }
|
| 756 |
|
| 757 | status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
|
| 758 | bg_bh, ac, cl, min_bits);
|
| 759 | if (status)
|
| 760 | mlog_errno(status);
|
| 761 |
|
| 762 | bail:
|
| 763 | if (status)
|
| 764 | ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
|
| 765 | return status ? ERR_PTR(status) : bg_bh;
|
| 766 | }
|
| 767 |
|
| 768 | |
| 769 | * We expect the block group allocator to already be locked.
|
| 770 |
|
| 771 | static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
|
| 772 | struct inode *alloc_inode,
|
| 773 | struct buffer_head *bh,
|
| 774 | u64 max_block,
|
| 775 | u64 *last_alloc_group,
|
| 776 | int flags)
|
| 777 | {
|
| 778 | int status, credits;
|
| 779 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
|
| 780 | struct ocfs2_chain_list *cl;
|
| 781 | struct ocfs2_alloc_context *ac = NULL;
|
| 782 | handle_t *handle = NULL;
|
| 783 | u16 alloc_rec;
|
| 784 | struct buffer_head *bg_bh = NULL;
|
| 785 | struct ocfs2_group_desc *bg;
|
| 786 |
|
| 787 | BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
|
| 788 |
|
| 789 | cl = &fe->id2.i_chain;
|
| 790 | status = ocfs2_reserve_clusters_with_limit(osb,
|
| 791 | le16_to_cpu(cl->cl_cpg),
|
| 792 | max_block, flags, &ac);
|
| 793 | if (status < 0) {
|
| 794 | if (status != -ENOSPC)
|
| 795 | mlog_errno(status);
|
| 796 | goto bail;
|
| 797 | }
|
| 798 |
|
| 799 | credits = ocfs2_calc_group_alloc_credits(osb->sb,
|
| 800 | le16_to_cpu(cl->cl_cpg));
|
| 801 | handle = ocfs2_start_trans(osb, credits);
|
| 802 | if (IS_ERR(handle)) {
|
| 803 | status = PTR_ERR(handle);
|
| 804 | handle = NULL;
|
| 805 | mlog_errno(status);
|
| 806 | goto bail;
|
| 807 | }
|
| 808 |
|
| 809 | if (last_alloc_group && *last_alloc_group != 0) {
|
| 810 | trace_ocfs2_block_group_alloc(
|
| 811 | (unsigned long long)*last_alloc_group);
|
| 812 | ac->ac_last_group = *last_alloc_group;
|
| 813 | }
|
| 814 |
|
| 815 | bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
|
| 816 | ac, cl);
|
| 817 | if (PTR_ERR(bg_bh) == -ENOSPC) {
|
| 818 | ac->ac_which = OCFS2_AC_USE_MAIN_DISCONTIG;
|
| 819 | bg_bh = ocfs2_block_group_alloc_discontig(handle,
|
| 820 | alloc_inode,
|
| 821 | ac, cl);
|
| 822 | }
|
| 823 | if (IS_ERR(bg_bh)) {
|
| 824 | status = PTR_ERR(bg_bh);
|
| 825 | bg_bh = NULL;
|
| 826 | if (status != -ENOSPC)
|
| 827 | mlog_errno(status);
|
| 828 | goto bail;
|
| 829 | }
|
| 830 | bg = (struct ocfs2_group_desc *) bg_bh->b_data;
|
| 831 |
|
| 832 | status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
|
| 833 | bh, OCFS2_JOURNAL_ACCESS_WRITE);
|
| 834 | if (status < 0) {
|
| 835 | mlog_errno(status);
|
| 836 | goto bail;
|
| 837 | }
|
| 838 |
|
| 839 | alloc_rec = le16_to_cpu(bg->bg_chain);
|
| 840 | le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
|
| 841 | le16_to_cpu(bg->bg_free_bits_count));
|
| 842 | le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
|
| 843 | le16_to_cpu(bg->bg_bits));
|
| 844 | cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
|
| 845 | if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
|
| 846 | le16_add_cpu(&cl->cl_next_free_rec, 1);
|
| 847 |
|
| 848 | le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
|
| 849 | le16_to_cpu(bg->bg_free_bits_count));
|
| 850 | le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
|
| 851 | le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
|
| 852 |
|
| 853 | ocfs2_journal_dirty(handle, bh);
|
| 854 |
|
| 855 | spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
|
| 856 | OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
|
| 857 | fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
|
| 858 | le32_to_cpu(fe->i_clusters)));
|
| 859 | spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
|
| 860 | i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
|
| 861 | alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
|
| 862 | ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
|
| 863 |
|
| 864 | status = 0;
|
| 865 |
|
| 866 |
|
| 867 | if (last_alloc_group)
|
| 868 | *last_alloc_group = ac->ac_last_group;
|
| 869 |
|
| 870 | bail:
|
| 871 | if (handle)
|
| 872 | ocfs2_commit_trans(osb, handle);
|
| 873 |
|
| 874 | if (ac)
|
| 875 | ocfs2_free_alloc_context(ac);
|
| 876 |
|
| 877 | brelse(bg_bh);
|
| 878 |
|
| 879 | if (status)
|
| 880 | mlog_errno(status);
|
| 881 | return status;
|
| 882 | }
|
| 883 |
|
| 884 | static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
|
| 885 | struct ocfs2_alloc_context *ac,
|
| 886 | int type,
|
| 887 | u32 slot,
|
| 888 | u64 *last_alloc_group,
|
| 889 | int flags)
|
| 890 | {
|
| 891 | int status;
|
| 892 | u32 bits_wanted = ac->ac_bits_wanted;
|
| 893 | struct inode *alloc_inode;
|
| 894 | struct buffer_head *bh = NULL;
|
| 895 | struct ocfs2_dinode *fe;
|
| 896 | u32 free_bits;
|
| 897 |
|
| 898 | alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
|
| 899 | if (!alloc_inode) {
|
| 900 | mlog_errno(-EINVAL);
|
| 901 | return -EINVAL;
|
| 902 | }
|
| 903 |
|
| 904 | inode_lock(alloc_inode);
|
| 905 |
|
| 906 | status = ocfs2_inode_lock(alloc_inode, &bh, 1);
|
| 907 | if (status < 0) {
|
| 908 | inode_unlock(alloc_inode);
|
| 909 | iput(alloc_inode);
|
| 910 |
|
| 911 | mlog_errno(status);
|
| 912 | return status;
|
| 913 | }
|
| 914 |
|
| 915 | ac->ac_inode = alloc_inode;
|
| 916 | ac->ac_alloc_slot = slot;
|
| 917 |
|
| 918 | fe = (struct ocfs2_dinode *) bh->b_data;
|
| 919 |
|
| 920 | |
| 921 | * ocfs2_inode_lock(). Any corruption is a code bug.
|
| 922 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
|
| 923 |
|
| 924 | if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
|
| 925 | status = ocfs2_error(alloc_inode->i_sb,
|
| 926 | "Invalid chain allocator %llu\n",
|
| 927 | (unsigned long long)le64_to_cpu(fe->i_blkno));
|
| 928 | goto bail;
|
| 929 | }
|
| 930 |
|
| 931 | free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
|
| 932 | le32_to_cpu(fe->id1.bitmap1.i_used);
|
| 933 |
|
| 934 | if (bits_wanted > free_bits) {
|
| 935 |
|
| 936 | if (ocfs2_is_cluster_bitmap(alloc_inode)) {
|
| 937 | trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
|
| 938 | free_bits);
|
| 939 | status = -ENOSPC;
|
| 940 | goto bail;
|
| 941 | }
|
| 942 |
|
| 943 | if (!(flags & ALLOC_NEW_GROUP)) {
|
| 944 | trace_ocfs2_reserve_suballoc_bits_no_new_group(
|
| 945 | slot, bits_wanted, free_bits);
|
| 946 | status = -ENOSPC;
|
| 947 | goto bail;
|
| 948 | }
|
| 949 |
|
| 950 | status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
|
| 951 | ac->ac_max_block,
|
| 952 | last_alloc_group, flags);
|
| 953 | if (status < 0) {
|
| 954 | if (status != -ENOSPC)
|
| 955 | mlog_errno(status);
|
| 956 | goto bail;
|
| 957 | }
|
| 958 | atomic_inc(&osb->alloc_stats.bg_extends);
|
| 959 |
|
| 960 |
|
| 961 | BUG_ON(bits_wanted >
|
| 962 | (le32_to_cpu(fe->id1.bitmap1.i_total)
|
| 963 | - le32_to_cpu(fe->id1.bitmap1.i_used)));
|
| 964 | }
|
| 965 |
|
| 966 | get_bh(bh);
|
| 967 | ac->ac_bh = bh;
|
| 968 | bail:
|
| 969 | brelse(bh);
|
| 970 |
|
| 971 | if (status)
|
| 972 | mlog_errno(status);
|
| 973 | return status;
|
| 974 | }
|
| 975 |
|
| 976 | static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
|
| 977 | {
|
| 978 | spin_lock(&osb->osb_lock);
|
| 979 | osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
|
| 980 | spin_unlock(&osb->osb_lock);
|
| 981 | atomic_set(&osb->s_num_inodes_stolen, 0);
|
| 982 | }
|
| 983 |
|
| 984 | static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
|
| 985 | {
|
| 986 | spin_lock(&osb->osb_lock);
|
| 987 | osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
|
| 988 | spin_unlock(&osb->osb_lock);
|
| 989 | atomic_set(&osb->s_num_meta_stolen, 0);
|
| 990 | }
|
| 991 |
|
| 992 | void ocfs2_init_steal_slots(struct ocfs2_super *osb)
|
| 993 | {
|
| 994 | ocfs2_init_inode_steal_slot(osb);
|
| 995 | ocfs2_init_meta_steal_slot(osb);
|
| 996 | }
|
| 997 |
|
| 998 | static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
|
| 999 | {
|
| 1000 | spin_lock(&osb->osb_lock);
|
| 1001 | if (type == INODE_ALLOC_SYSTEM_INODE)
|
| 1002 | osb->s_inode_steal_slot = (u16)slot;
|
| 1003 | else if (type == EXTENT_ALLOC_SYSTEM_INODE)
|
| 1004 | osb->s_meta_steal_slot = (u16)slot;
|
| 1005 | spin_unlock(&osb->osb_lock);
|
| 1006 | }
|
| 1007 |
|
| 1008 | static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
|
| 1009 | {
|
| 1010 | int slot = OCFS2_INVALID_SLOT;
|
| 1011 |
|
| 1012 | spin_lock(&osb->osb_lock);
|
| 1013 | if (type == INODE_ALLOC_SYSTEM_INODE)
|
| 1014 | slot = osb->s_inode_steal_slot;
|
| 1015 | else if (type == EXTENT_ALLOC_SYSTEM_INODE)
|
| 1016 | slot = osb->s_meta_steal_slot;
|
| 1017 | spin_unlock(&osb->osb_lock);
|
| 1018 |
|
| 1019 | return slot;
|
| 1020 | }
|
| 1021 |
|
| 1022 | static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
|
| 1023 | {
|
| 1024 | return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
|
| 1025 | }
|
| 1026 |
|
| 1027 | static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
|
| 1028 | {
|
| 1029 | return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
|
| 1030 | }
|
| 1031 |
|
| 1032 | static int ocfs2_steal_resource(struct ocfs2_super *osb,
|
| 1033 | struct ocfs2_alloc_context *ac,
|
| 1034 | int type)
|
| 1035 | {
|
| 1036 | int i, status = -ENOSPC;
|
| 1037 | int slot = __ocfs2_get_steal_slot(osb, type);
|
| 1038 |
|
| 1039 |
|
| 1040 | if (slot == OCFS2_INVALID_SLOT)
|
| 1041 | slot = osb->slot_num + 1;
|
| 1042 |
|
| 1043 | for (i = 0; i < osb->max_slots; i++, slot++) {
|
| 1044 | if (slot == osb->max_slots)
|
| 1045 | slot = 0;
|
| 1046 |
|
| 1047 | if (slot == osb->slot_num)
|
| 1048 | continue;
|
| 1049 |
|
| 1050 | status = ocfs2_reserve_suballoc_bits(osb, ac,
|
| 1051 | type,
|
| 1052 | (u32)slot, NULL,
|
| 1053 | NOT_ALLOC_NEW_GROUP);
|
| 1054 | if (status >= 0) {
|
| 1055 | __ocfs2_set_steal_slot(osb, slot, type);
|
| 1056 | break;
|
| 1057 | }
|
| 1058 |
|
| 1059 | ocfs2_free_ac_resource(ac);
|
| 1060 | }
|
| 1061 |
|
| 1062 | return status;
|
| 1063 | }
|
| 1064 |
|
| 1065 | static int ocfs2_steal_inode(struct ocfs2_super *osb,
|
| 1066 | struct ocfs2_alloc_context *ac)
|
| 1067 | {
|
| 1068 | return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
|
| 1069 | }
|
| 1070 |
|
| 1071 | static int ocfs2_steal_meta(struct ocfs2_super *osb,
|
| 1072 | struct ocfs2_alloc_context *ac)
|
| 1073 | {
|
| 1074 | return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
|
| 1075 | }
|
| 1076 |
|
| 1077 | int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
|
| 1078 | int blocks,
|
| 1079 | struct ocfs2_alloc_context **ac)
|
| 1080 | {
|
| 1081 | int status;
|
| 1082 | int slot = ocfs2_get_meta_steal_slot(osb);
|
| 1083 |
|
| 1084 | *ac = kzalloc_obj(struct ocfs2_alloc_context);
|
| 1085 | if (!(*ac)) {
|
| 1086 | status = -ENOMEM;
|
| 1087 | mlog_errno(status);
|
| 1088 | goto bail;
|
| 1089 | }
|
| 1090 |
|
| 1091 | (*ac)->ac_bits_wanted = blocks;
|
| 1092 | (*ac)->ac_which = OCFS2_AC_USE_META;
|
| 1093 | (*ac)->ac_group_search = ocfs2_block_group_search;
|
| 1094 |
|
| 1095 | if (slot != OCFS2_INVALID_SLOT &&
|
| 1096 | atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
|
| 1097 | goto extent_steal;
|
| 1098 |
|
| 1099 | atomic_set(&osb->s_num_meta_stolen, 0);
|
| 1100 | status = ocfs2_reserve_suballoc_bits(osb, (*ac),
|
| 1101 | EXTENT_ALLOC_SYSTEM_INODE,
|
| 1102 | (u32)osb->slot_num, NULL,
|
| 1103 | ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
|
| 1104 |
|
| 1105 |
|
| 1106 | if (status >= 0) {
|
| 1107 | status = 0;
|
| 1108 | if (slot != OCFS2_INVALID_SLOT)
|
| 1109 | ocfs2_init_meta_steal_slot(osb);
|
| 1110 | goto bail;
|
| 1111 | } else if (status < 0 && status != -ENOSPC) {
|
| 1112 | mlog_errno(status);
|
| 1113 | goto bail;
|
| 1114 | }
|
| 1115 |
|
| 1116 | ocfs2_free_ac_resource(*ac);
|
| 1117 |
|
| 1118 | extent_steal:
|
| 1119 | status = ocfs2_steal_meta(osb, *ac);
|
| 1120 | atomic_inc(&osb->s_num_meta_stolen);
|
| 1121 | if (status < 0) {
|
| 1122 | if (status != -ENOSPC)
|
| 1123 | mlog_errno(status);
|
| 1124 | goto bail;
|
| 1125 | }
|
| 1126 |
|
| 1127 | status = 0;
|
| 1128 | bail:
|
| 1129 | if ((status < 0) && *ac) {
|
| 1130 | ocfs2_free_alloc_context(*ac);
|
| 1131 | *ac = NULL;
|
| 1132 | }
|
| 1133 |
|
| 1134 | if (status)
|
| 1135 | mlog_errno(status);
|
| 1136 | return status;
|
| 1137 | }
|
| 1138 |
|
| 1139 | int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
|
| 1140 | struct ocfs2_extent_list *root_el,
|
| 1141 | struct ocfs2_alloc_context **ac)
|
| 1142 | {
|
| 1143 | return ocfs2_reserve_new_metadata_blocks(osb,
|
| 1144 | ocfs2_extend_meta_needed(root_el),
|
| 1145 | ac);
|
| 1146 | }
|
| 1147 |
|
| 1148 | int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
|
| 1149 | struct ocfs2_alloc_context **ac)
|
| 1150 | {
|
| 1151 | int status;
|
| 1152 | int slot = ocfs2_get_inode_steal_slot(osb);
|
| 1153 | u64 alloc_group;
|
| 1154 |
|
| 1155 | *ac = kzalloc_obj(struct ocfs2_alloc_context);
|
| 1156 | if (!(*ac)) {
|
| 1157 | status = -ENOMEM;
|
| 1158 | mlog_errno(status);
|
| 1159 | goto bail;
|
| 1160 | }
|
| 1161 |
|
| 1162 | (*ac)->ac_bits_wanted = 1;
|
| 1163 | (*ac)->ac_which = OCFS2_AC_USE_INODE;
|
| 1164 |
|
| 1165 | (*ac)->ac_group_search = ocfs2_block_group_search;
|
| 1166 |
|
| 1167 | |
| 1168 | * stat(2) can't handle i_ino > 32bits, so we tell the
|
| 1169 | * lower levels not to allocate us a block group past that
|
| 1170 | * limit. The 'inode64' mount option avoids this behavior.
|
| 1171 |
|
| 1172 | if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
|
| 1173 | (*ac)->ac_max_block = (u32)~0U;
|
| 1174 |
|
| 1175 | |
| 1176 | * slot is set when we successfully steal inode from other nodes.
|
| 1177 | * It is reset in 3 places:
|
| 1178 | * 1. when we flush the truncate log
|
| 1179 | * 2. when we complete local alloc recovery.
|
| 1180 | * 3. when we successfully allocate from our own slot.
|
| 1181 | * After it is set, we will go on stealing inodes until we find the
|
| 1182 | * need to check our slots to see whether there is some space for us.
|
| 1183 |
|
| 1184 | if (slot != OCFS2_INVALID_SLOT &&
|
| 1185 | atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
|
| 1186 | goto inode_steal;
|
| 1187 |
|
| 1188 | atomic_set(&osb->s_num_inodes_stolen, 0);
|
| 1189 | alloc_group = osb->osb_inode_alloc_group;
|
| 1190 | status = ocfs2_reserve_suballoc_bits(osb, *ac,
|
| 1191 | INODE_ALLOC_SYSTEM_INODE,
|
| 1192 | (u32)osb->slot_num,
|
| 1193 | &alloc_group,
|
| 1194 | ALLOC_NEW_GROUP |
|
| 1195 | ALLOC_GROUPS_FROM_GLOBAL);
|
| 1196 | if (status >= 0) {
|
| 1197 | status = 0;
|
| 1198 |
|
| 1199 | spin_lock(&osb->osb_lock);
|
| 1200 | osb->osb_inode_alloc_group = alloc_group;
|
| 1201 | spin_unlock(&osb->osb_lock);
|
| 1202 | trace_ocfs2_reserve_new_inode_new_group(
|
| 1203 | (unsigned long long)alloc_group);
|
| 1204 |
|
| 1205 | |
| 1206 | * Some inodes must be freed by us, so try to allocate
|
| 1207 | * from our own next time.
|
| 1208 |
|
| 1209 | if (slot != OCFS2_INVALID_SLOT)
|
| 1210 | ocfs2_init_inode_steal_slot(osb);
|
| 1211 | goto bail;
|
| 1212 | } else if (status < 0 && status != -ENOSPC) {
|
| 1213 | mlog_errno(status);
|
| 1214 | goto bail;
|
| 1215 | }
|
| 1216 |
|
| 1217 | ocfs2_free_ac_resource(*ac);
|
| 1218 |
|
| 1219 | inode_steal:
|
| 1220 | status = ocfs2_steal_inode(osb, *ac);
|
| 1221 | atomic_inc(&osb->s_num_inodes_stolen);
|
| 1222 | if (status < 0) {
|
| 1223 | if (status != -ENOSPC)
|
| 1224 | mlog_errno(status);
|
| 1225 | goto bail;
|
| 1226 | }
|
| 1227 |
|
| 1228 | status = 0;
|
| 1229 | bail:
|
| 1230 | if ((status < 0) && *ac) {
|
| 1231 | ocfs2_free_alloc_context(*ac);
|
| 1232 | *ac = NULL;
|
| 1233 | }
|
| 1234 |
|
| 1235 | if (status)
|
| 1236 | mlog_errno(status);
|
| 1237 | return status;
|
| 1238 | }
|
| 1239 |
|
| 1240 | |
| 1241 | * twice..
|
| 1242 | int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
|
| 1243 | struct ocfs2_alloc_context *ac)
|
| 1244 | {
|
| 1245 | int status;
|
| 1246 |
|
| 1247 | ac->ac_which = OCFS2_AC_USE_MAIN;
|
| 1248 | ac->ac_group_search = ocfs2_cluster_group_search;
|
| 1249 |
|
| 1250 | status = ocfs2_reserve_suballoc_bits(osb, ac,
|
| 1251 | GLOBAL_BITMAP_SYSTEM_INODE,
|
| 1252 | OCFS2_INVALID_SLOT, NULL,
|
| 1253 | ALLOC_NEW_GROUP);
|
| 1254 | if (status < 0 && status != -ENOSPC)
|
| 1255 | mlog_errno(status);
|
| 1256 |
|
| 1257 | return status;
|
| 1258 | }
|
| 1259 |
|
| 1260 | |
| 1261 | * use so we figure it out for them, but unfortunately this clutters
|
| 1262 | * things a bit.
|
| 1263 | static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
|
| 1264 | u32 bits_wanted, u64 max_block,
|
| 1265 | int flags,
|
| 1266 | struct ocfs2_alloc_context **ac)
|
| 1267 | {
|
| 1268 | int status, ret = 0;
|
| 1269 | int retried = 0;
|
| 1270 |
|
| 1271 | *ac = kzalloc_obj(struct ocfs2_alloc_context);
|
| 1272 | if (!(*ac)) {
|
| 1273 | status = -ENOMEM;
|
| 1274 | mlog_errno(status);
|
| 1275 | goto bail;
|
| 1276 | }
|
| 1277 |
|
| 1278 | (*ac)->ac_bits_wanted = bits_wanted;
|
| 1279 | (*ac)->ac_max_block = max_block;
|
| 1280 |
|
| 1281 | status = -ENOSPC;
|
| 1282 | if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
|
| 1283 | ocfs2_alloc_should_use_local(osb, bits_wanted)) {
|
| 1284 | status = ocfs2_reserve_local_alloc_bits(osb,
|
| 1285 | bits_wanted,
|
| 1286 | *ac);
|
| 1287 | if ((status < 0) && (status != -ENOSPC)) {
|
| 1288 | mlog_errno(status);
|
| 1289 | goto bail;
|
| 1290 | }
|
| 1291 | }
|
| 1292 |
|
| 1293 | if (status == -ENOSPC) {
|
| 1294 | retry:
|
| 1295 | status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
|
| 1296 |
|
| 1297 | if (status == -ENOSPC && !retried) {
|
| 1298 | retried = 1;
|
| 1299 | ocfs2_inode_unlock((*ac)->ac_inode, 1);
|
| 1300 | inode_unlock((*ac)->ac_inode);
|
| 1301 |
|
| 1302 | ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
|
| 1303 | if (ret == 1) {
|
| 1304 | iput((*ac)->ac_inode);
|
| 1305 | (*ac)->ac_inode = NULL;
|
| 1306 | goto retry;
|
| 1307 | }
|
| 1308 |
|
| 1309 | if (ret < 0)
|
| 1310 | mlog_errno(ret);
|
| 1311 |
|
| 1312 | inode_lock((*ac)->ac_inode);
|
| 1313 | ret = ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
|
| 1314 | if (ret < 0) {
|
| 1315 | mlog_errno(ret);
|
| 1316 | inode_unlock((*ac)->ac_inode);
|
| 1317 | iput((*ac)->ac_inode);
|
| 1318 | (*ac)->ac_inode = NULL;
|
| 1319 | goto bail;
|
| 1320 | }
|
| 1321 | }
|
| 1322 | if (status < 0) {
|
| 1323 | if (status != -ENOSPC)
|
| 1324 | mlog_errno(status);
|
| 1325 | goto bail;
|
| 1326 | }
|
| 1327 | }
|
| 1328 |
|
| 1329 | status = 0;
|
| 1330 | bail:
|
| 1331 | if ((status < 0) && *ac) {
|
| 1332 | ocfs2_free_alloc_context(*ac);
|
| 1333 | *ac = NULL;
|
| 1334 | }
|
| 1335 |
|
| 1336 | if (status)
|
| 1337 | mlog_errno(status);
|
| 1338 | return status;
|
| 1339 | }
|
| 1340 |
|
| 1341 | int ocfs2_reserve_clusters(struct ocfs2_super *osb,
|
| 1342 | u32 bits_wanted,
|
| 1343 | struct ocfs2_alloc_context **ac)
|
| 1344 | {
|
| 1345 | return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
|
| 1346 | ALLOC_NEW_GROUP, ac);
|
| 1347 | }
|
| 1348 |
|
| 1349 | |
| 1350 | * More or less lifted from ext3. I'll leave their description below:
|
| 1351 | *
|
| 1352 | * "For ext3 allocations, we must not reuse any blocks which are
|
| 1353 | * allocated in the bitmap buffer's "last committed data" copy. This
|
| 1354 | * prevents deletes from freeing up the page for reuse until we have
|
| 1355 | * committed the delete transaction.
|
| 1356 | *
|
| 1357 | * If we didn't do this, then deleting something and reallocating it as
|
| 1358 | * data would allow the old block to be overwritten before the
|
| 1359 | * transaction committed (because we force data to disk before commit).
|
| 1360 | * This would lead to corruption if we crashed between overwriting the
|
| 1361 | * data and committing the delete.
|
| 1362 | *
|
| 1363 | * @@@ We may want to make this allocation behaviour conditional on
|
| 1364 | * data-writes at some point, and disable it for metadata allocations or
|
| 1365 | * sync-data inodes."
|
| 1366 | *
|
| 1367 | * Note: OCFS2 already does this differently for metadata vs data
|
| 1368 | * allocations, as those bitmaps are separate and undo access is never
|
| 1369 | * called on a metadata group descriptor.
|
| 1370 |
|
| 1371 | static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
|
| 1372 | int nr)
|
| 1373 | {
|
| 1374 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
|
| 1375 | struct journal_head *jh;
|
| 1376 | int ret;
|
| 1377 |
|
| 1378 | if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
|
| 1379 | return 0;
|
| 1380 |
|
| 1381 | jh = jbd2_journal_grab_journal_head(bg_bh);
|
| 1382 | if (!jh)
|
| 1383 | return 1;
|
| 1384 |
|
| 1385 | spin_lock(&jh->b_state_lock);
|
| 1386 | bg = (struct ocfs2_group_desc *) jh->b_committed_data;
|
| 1387 | if (bg)
|
| 1388 | ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
|
| 1389 | else
|
| 1390 | ret = 1;
|
| 1391 | spin_unlock(&jh->b_state_lock);
|
| 1392 | jbd2_journal_put_journal_head(jh);
|
| 1393 |
|
| 1394 | return ret;
|
| 1395 | }
|
| 1396 |
|
| 1397 | u16 ocfs2_find_max_contig_free_bits(void *bitmap,
|
| 1398 | u16 total_bits, u16 start)
|
| 1399 | {
|
| 1400 | u16 offset, free_bits;
|
| 1401 | u16 contig_bits = 0;
|
| 1402 |
|
| 1403 | while (start < total_bits) {
|
| 1404 | offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start);
|
| 1405 | if (offset == total_bits)
|
| 1406 | break;
|
| 1407 |
|
| 1408 | start = ocfs2_find_next_bit(bitmap, total_bits, offset);
|
| 1409 | free_bits = start - offset;
|
| 1410 | if (contig_bits < free_bits)
|
| 1411 | contig_bits = free_bits;
|
| 1412 | }
|
| 1413 |
|
| 1414 | return contig_bits;
|
| 1415 | }
|
| 1416 |
|
| 1417 | static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
|
| 1418 | struct buffer_head *bg_bh,
|
| 1419 | unsigned int bits_wanted,
|
| 1420 | unsigned int total_bits,
|
| 1421 | struct ocfs2_suballoc_result *res)
|
| 1422 | {
|
| 1423 | void *bitmap;
|
| 1424 | u16 best_offset, best_size;
|
| 1425 | u16 prev_best_size = 0;
|
| 1426 | int offset, start, found, status = 0;
|
| 1427 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
|
| 1428 |
|
| 1429 | |
| 1430 | * ocfs2_read_group_descriptor(). Any corruption is a code bug.
|
| 1431 | BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
|
| 1432 |
|
| 1433 | found = start = best_offset = best_size = 0;
|
| 1434 | bitmap = bg->bg_bitmap;
|
| 1435 |
|
| 1436 | while ((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) <
|
| 1437 | total_bits) {
|
| 1438 | if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
|
| 1439 | |
| 1440 | * hasn't been put to disk yet!
|
| 1441 | found = 0;
|
| 1442 | start = offset + 1;
|
| 1443 | } else if (offset == start) {
|
| 1444 |
|
| 1445 | found++;
|
| 1446 |
|
| 1447 | start++;
|
| 1448 | } else {
|
| 1449 |
|
| 1450 | found = 1;
|
| 1451 | start = offset + 1;
|
| 1452 | prev_best_size = best_size;
|
| 1453 | }
|
| 1454 | if (found > best_size) {
|
| 1455 | best_size = found;
|
| 1456 | best_offset = start - found;
|
| 1457 | }
|
| 1458 |
|
| 1459 | if (found == bits_wanted) {
|
| 1460 |
|
| 1461 | break;
|
| 1462 | }
|
| 1463 | }
|
| 1464 |
|
| 1465 |
|
| 1466 | res->sr_max_contig_bits = prev_best_size;
|
| 1467 | if (best_size) {
|
| 1468 | res->sr_bit_offset = best_offset;
|
| 1469 | res->sr_bits = best_size;
|
| 1470 | } else {
|
| 1471 | status = -ENOSPC;
|
| 1472 | |
| 1473 | * ocfs2_test_bg_bit_allocatable
|
| 1474 | }
|
| 1475 |
|
| 1476 | return status;
|
| 1477 | }
|
| 1478 |
|
| 1479 | int ocfs2_block_group_set_bits(handle_t *handle,
|
| 1480 | struct inode *alloc_inode,
|
| 1481 | struct ocfs2_group_desc *bg,
|
| 1482 | struct buffer_head *group_bh,
|
| 1483 | unsigned int bit_off,
|
| 1484 | unsigned int num_bits,
|
| 1485 | unsigned int max_contig_bits,
|
| 1486 | int fastpath)
|
| 1487 | {
|
| 1488 | int status;
|
| 1489 | void *bitmap = bg->bg_bitmap;
|
| 1490 | int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
|
| 1491 | unsigned int start = bit_off + num_bits;
|
| 1492 | u16 contig_bits;
|
| 1493 | struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
|
| 1494 |
|
| 1495 | |
| 1496 | * ocfs2_read_group_descriptor(). Any corruption is a code bug.
|
| 1497 | BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
|
| 1498 | BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
|
| 1499 |
|
| 1500 | trace_ocfs2_block_group_set_bits(bit_off, num_bits);
|
| 1501 |
|
| 1502 | if (ocfs2_is_cluster_bitmap(alloc_inode))
|
| 1503 | journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
|
| 1504 |
|
| 1505 | status = ocfs2_journal_access_gd(handle,
|
| 1506 | INODE_CACHE(alloc_inode),
|
| 1507 | group_bh,
|
| 1508 | journal_type);
|
| 1509 | if (status < 0) {
|
| 1510 | mlog_errno(status);
|
| 1511 | goto bail;
|
| 1512 | }
|
| 1513 |
|
| 1514 | le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
|
| 1515 | if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
|
| 1516 | return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
|
| 1517 | (unsigned long long)le64_to_cpu(bg->bg_blkno),
|
| 1518 | le16_to_cpu(bg->bg_bits),
|
| 1519 | le16_to_cpu(bg->bg_free_bits_count),
|
| 1520 | num_bits);
|
| 1521 | }
|
| 1522 | while(num_bits--)
|
| 1523 | ocfs2_set_bit(bit_off++, bitmap);
|
| 1524 |
|
| 1525 | |
| 1526 | * this is optimize path, caller set old contig value
|
| 1527 | * in max_contig_bits to bypass finding action.
|
| 1528 |
|
| 1529 | if (fastpath) {
|
| 1530 | bg->bg_contig_free_bits = cpu_to_le16(max_contig_bits);
|
| 1531 | } else if (ocfs2_is_cluster_bitmap(alloc_inode)) {
|
| 1532 | |
| 1533 | * Usually, the block group bitmap allocates only 1 bit
|
| 1534 | * at a time, while the cluster group allocates n bits
|
| 1535 | * each time. Therefore, we only save the contig bits for
|
| 1536 | * the cluster group.
|
| 1537 |
|
| 1538 | contig_bits = ocfs2_find_max_contig_free_bits(bitmap,
|
| 1539 | le16_to_cpu(bg->bg_bits), start);
|
| 1540 | if (contig_bits > max_contig_bits)
|
| 1541 | max_contig_bits = contig_bits;
|
| 1542 | bg->bg_contig_free_bits = cpu_to_le16(max_contig_bits);
|
| 1543 | ocfs2_local_alloc_seen_free_bits(osb, max_contig_bits);
|
| 1544 | } else {
|
| 1545 | bg->bg_contig_free_bits = 0;
|
| 1546 | }
|
| 1547 |
|
| 1548 | ocfs2_journal_dirty(handle, group_bh);
|
| 1549 |
|
| 1550 | bail:
|
| 1551 | return status;
|
| 1552 | }
|
| 1553 |
|
| 1554 |
|
| 1555 | static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
|
| 1556 | {
|
| 1557 | u16 curr, best;
|
| 1558 |
|
| 1559 | BUG_ON(!cl->cl_next_free_rec);
|
| 1560 |
|
| 1561 | best = curr = 0;
|
| 1562 | while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
|
| 1563 | if (le32_to_cpu(cl->cl_recs[curr].c_free) >
|
| 1564 | le32_to_cpu(cl->cl_recs[best].c_free))
|
| 1565 | best = curr;
|
| 1566 | curr++;
|
| 1567 | }
|
| 1568 |
|
| 1569 | BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
|
| 1570 | return best;
|
| 1571 | }
|
| 1572 |
|
| 1573 | static int ocfs2_relink_block_group(handle_t *handle,
|
| 1574 | struct inode *alloc_inode,
|
| 1575 | struct buffer_head *fe_bh,
|
| 1576 | struct buffer_head *bg_bh,
|
| 1577 | struct buffer_head *prev_bg_bh,
|
| 1578 | u16 chain)
|
| 1579 | {
|
| 1580 | int status;
|
| 1581 | |
| 1582 | * but we wouldn't want inconsistent blocks in *any* case.
|
| 1583 | u64 bg_ptr, prev_bg_ptr;
|
| 1584 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
|
| 1585 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
|
| 1586 | struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
|
| 1587 |
|
| 1588 | |
| 1589 | * ocfs2_read_group_descriptor(). Any corruption is a code bug.
|
| 1590 | BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
|
| 1591 | BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
|
| 1592 |
|
| 1593 | trace_ocfs2_relink_block_group(
|
| 1594 | (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
|
| 1595 | (unsigned long long)le64_to_cpu(bg->bg_blkno),
|
| 1596 | (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
|
| 1597 |
|
| 1598 | bg_ptr = le64_to_cpu(bg->bg_next_group);
|
| 1599 | prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
|
| 1600 |
|
| 1601 | status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
|
| 1602 | prev_bg_bh,
|
| 1603 | OCFS2_JOURNAL_ACCESS_WRITE);
|
| 1604 | if (status < 0)
|
| 1605 | goto out;
|
| 1606 |
|
| 1607 | prev_bg->bg_next_group = bg->bg_next_group;
|
| 1608 | ocfs2_journal_dirty(handle, prev_bg_bh);
|
| 1609 |
|
| 1610 | status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
|
| 1611 | bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
|
| 1612 | if (status < 0)
|
| 1613 | goto out_rollback_prev_bg;
|
| 1614 |
|
| 1615 | bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
|
| 1616 | ocfs2_journal_dirty(handle, bg_bh);
|
| 1617 |
|
| 1618 | status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
|
| 1619 | fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
|
| 1620 | if (status < 0)
|
| 1621 | goto out_rollback_bg;
|
| 1622 |
|
| 1623 | fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
|
| 1624 | ocfs2_journal_dirty(handle, fe_bh);
|
| 1625 |
|
| 1626 | out:
|
| 1627 | if (status < 0)
|
| 1628 | mlog_errno(status);
|
| 1629 | return status;
|
| 1630 |
|
| 1631 | out_rollback_bg:
|
| 1632 | bg->bg_next_group = cpu_to_le64(bg_ptr);
|
| 1633 | out_rollback_prev_bg:
|
| 1634 | prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
|
| 1635 | goto out;
|
| 1636 | }
|
| 1637 |
|
| 1638 | static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
|
| 1639 | u32 wanted)
|
| 1640 | {
|
| 1641 | return le16_to_cpu(bg->bg_free_bits_count) > wanted;
|
| 1642 | }
|
| 1643 |
|
| 1644 | |
| 1645 | * value on error.
|
| 1646 | static int ocfs2_cluster_group_search(struct inode *inode,
|
| 1647 | struct buffer_head *group_bh,
|
| 1648 | u32 bits_wanted, u32 min_bits,
|
| 1649 | u64 max_block,
|
| 1650 | struct ocfs2_suballoc_result *res)
|
| 1651 | {
|
| 1652 | int search = -ENOSPC;
|
| 1653 | int ret;
|
| 1654 | u64 blkoff;
|
| 1655 | struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
|
| 1656 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
|
| 1657 | unsigned int max_bits, gd_cluster_off;
|
| 1658 |
|
| 1659 | BUG_ON(!ocfs2_is_cluster_bitmap(inode));
|
| 1660 |
|
| 1661 | if (le16_to_cpu(gd->bg_contig_free_bits) &&
|
| 1662 | le16_to_cpu(gd->bg_contig_free_bits) < bits_wanted)
|
| 1663 | return -ENOSPC;
|
| 1664 |
|
| 1665 |
|
| 1666 | if (le16_to_cpu(gd->bg_free_bits_count) >= bits_wanted) {
|
| 1667 | max_bits = le16_to_cpu(gd->bg_bits);
|
| 1668 |
|
| 1669 | |
| 1670 | * aligned are prone to partial extension by a failed
|
| 1671 | * fs resize. If the file system resize never got to
|
| 1672 | * update the dinode cluster count, then we don't want
|
| 1673 | * to trust any clusters past it, regardless of what
|
| 1674 | * the group descriptor says.
|
| 1675 | gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
|
| 1676 | le64_to_cpu(gd->bg_blkno));
|
| 1677 | if ((gd_cluster_off + max_bits) >
|
| 1678 | OCFS2_I(inode)->ip_clusters) {
|
| 1679 | max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
|
| 1680 | trace_ocfs2_cluster_group_search_wrong_max_bits(
|
| 1681 | (unsigned long long)le64_to_cpu(gd->bg_blkno),
|
| 1682 | le16_to_cpu(gd->bg_bits),
|
| 1683 | OCFS2_I(inode)->ip_clusters, max_bits);
|
| 1684 | }
|
| 1685 |
|
| 1686 | ret = ocfs2_block_group_find_clear_bits(osb,
|
| 1687 | group_bh, bits_wanted,
|
| 1688 | max_bits, res);
|
| 1689 | if (ret)
|
| 1690 | return ret;
|
| 1691 |
|
| 1692 | if (max_block) {
|
| 1693 | blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
|
| 1694 | gd_cluster_off +
|
| 1695 | res->sr_bit_offset +
|
| 1696 | res->sr_bits);
|
| 1697 | trace_ocfs2_cluster_group_search_max_block(
|
| 1698 | (unsigned long long)blkoff,
|
| 1699 | (unsigned long long)max_block);
|
| 1700 | if (blkoff > max_block)
|
| 1701 | return -ENOSPC;
|
| 1702 | }
|
| 1703 |
|
| 1704 | |
| 1705 | * return success, but we still want to return
|
| 1706 | * -ENOSPC unless it found the minimum number
|
| 1707 | * of bits.
|
| 1708 | if (min_bits <= res->sr_bits)
|
| 1709 | search = 0;
|
| 1710 | }
|
| 1711 |
|
| 1712 | return search;
|
| 1713 | }
|
| 1714 |
|
| 1715 | static int ocfs2_block_group_search(struct inode *inode,
|
| 1716 | struct buffer_head *group_bh,
|
| 1717 | u32 bits_wanted, u32 min_bits,
|
| 1718 | u64 max_block,
|
| 1719 | struct ocfs2_suballoc_result *res)
|
| 1720 | {
|
| 1721 | int ret = -ENOSPC;
|
| 1722 | u64 blkoff;
|
| 1723 | struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
|
| 1724 |
|
| 1725 | BUG_ON(min_bits != 1);
|
| 1726 | BUG_ON(ocfs2_is_cluster_bitmap(inode));
|
| 1727 |
|
| 1728 | if (le16_to_cpu(bg->bg_free_bits_count) >= bits_wanted) {
|
| 1729 | ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
|
| 1730 | group_bh, bits_wanted,
|
| 1731 | le16_to_cpu(bg->bg_bits),
|
| 1732 | res);
|
| 1733 | if (!ret && max_block) {
|
| 1734 | blkoff = le64_to_cpu(bg->bg_blkno) +
|
| 1735 | res->sr_bit_offset + res->sr_bits;
|
| 1736 | trace_ocfs2_block_group_search_max_block(
|
| 1737 | (unsigned long long)blkoff,
|
| 1738 | (unsigned long long)max_block);
|
| 1739 | if (blkoff > max_block)
|
| 1740 | ret = -ENOSPC;
|
| 1741 | }
|
| 1742 | }
|
| 1743 |
|
| 1744 | return ret;
|
| 1745 | }
|
| 1746 |
|
| 1747 | int ocfs2_alloc_dinode_update_counts(struct inode *inode,
|
| 1748 | handle_t *handle,
|
| 1749 | struct buffer_head *di_bh,
|
| 1750 | u32 num_bits,
|
| 1751 | u16 chain)
|
| 1752 | {
|
| 1753 | int ret;
|
| 1754 | u32 tmp_used;
|
| 1755 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
|
| 1756 | struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
|
| 1757 |
|
| 1758 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
|
| 1759 | OCFS2_JOURNAL_ACCESS_WRITE);
|
| 1760 | if (ret < 0) {
|
| 1761 | mlog_errno(ret);
|
| 1762 | goto out;
|
| 1763 | }
|
| 1764 |
|
| 1765 | tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
|
| 1766 | di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
|
| 1767 | le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
|
| 1768 | ocfs2_journal_dirty(handle, di_bh);
|
| 1769 |
|
| 1770 | out:
|
| 1771 | return ret;
|
| 1772 | }
|
| 1773 |
|
| 1774 | void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
|
| 1775 | struct buffer_head *di_bh,
|
| 1776 | u32 num_bits,
|
| 1777 | u16 chain)
|
| 1778 | {
|
| 1779 | u32 tmp_used;
|
| 1780 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
|
| 1781 | struct ocfs2_chain_list *cl;
|
| 1782 |
|
| 1783 | cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
|
| 1784 | tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
|
| 1785 | di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
|
| 1786 | le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
|
| 1787 | }
|
| 1788 |
|
| 1789 | static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
|
| 1790 | struct ocfs2_extent_rec *rec,
|
| 1791 | struct ocfs2_chain_list *cl)
|
| 1792 | {
|
| 1793 | unsigned int bpc = le16_to_cpu(cl->cl_bpc);
|
| 1794 | unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
|
| 1795 | unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
|
| 1796 |
|
| 1797 | if (res->sr_bit_offset < bitoff)
|
| 1798 | return 0;
|
| 1799 | if (res->sr_bit_offset >= (bitoff + bitcount))
|
| 1800 | return 0;
|
| 1801 | res->sr_blkno = le64_to_cpu(rec->e_blkno) +
|
| 1802 | (res->sr_bit_offset - bitoff);
|
| 1803 | if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
|
| 1804 | res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
|
| 1805 | return 1;
|
| 1806 | }
|
| 1807 |
|
| 1808 | static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
|
| 1809 | struct ocfs2_group_desc *bg,
|
| 1810 | struct ocfs2_suballoc_result *res)
|
| 1811 | {
|
| 1812 | int i;
|
| 1813 | u64 bg_blkno = res->sr_bg_blkno;
|
| 1814 | struct ocfs2_extent_rec *rec;
|
| 1815 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
|
| 1816 | struct ocfs2_chain_list *cl = &di->id2.i_chain;
|
| 1817 |
|
| 1818 | if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
|
| 1819 | res->sr_blkno = 0;
|
| 1820 | return;
|
| 1821 | }
|
| 1822 |
|
| 1823 | res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
|
| 1824 | res->sr_bg_blkno = 0;
|
| 1825 | if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
|
| 1826 | !bg->bg_list.l_next_free_rec)
|
| 1827 | return;
|
| 1828 |
|
| 1829 | for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
|
| 1830 | rec = &bg->bg_list.l_recs[i];
|
| 1831 | if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
|
| 1832 | res->sr_bg_blkno = bg_blkno;
|
| 1833 | break;
|
| 1834 | }
|
| 1835 | }
|
| 1836 | }
|
| 1837 |
|
| 1838 | static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
|
| 1839 | handle_t *handle,
|
| 1840 | u32 bits_wanted,
|
| 1841 | u32 min_bits,
|
| 1842 | struct ocfs2_suballoc_result *res,
|
| 1843 | u16 *bits_left, int *released)
|
| 1844 | {
|
| 1845 | int ret;
|
| 1846 | struct buffer_head *group_bh = NULL;
|
| 1847 | struct ocfs2_group_desc *gd;
|
| 1848 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
|
| 1849 | struct inode *alloc_inode = ac->ac_inode;
|
| 1850 |
|
| 1851 | ret = ocfs2_read_hint_group_descriptor(alloc_inode, di,
|
| 1852 | res->sr_bg_blkno, &group_bh, released);
|
| 1853 | if (*released) {
|
| 1854 | return 0;
|
| 1855 | } else if (ret < 0) {
|
| 1856 | mlog_errno(ret);
|
| 1857 | return ret;
|
| 1858 | }
|
| 1859 |
|
| 1860 | gd = (struct ocfs2_group_desc *) group_bh->b_data;
|
| 1861 | ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
|
| 1862 | ac->ac_max_block, res);
|
| 1863 | if (ret < 0) {
|
| 1864 | if (ret != -ENOSPC)
|
| 1865 | mlog_errno(ret);
|
| 1866 | goto out;
|
| 1867 | }
|
| 1868 |
|
| 1869 | if (!ret)
|
| 1870 | ocfs2_bg_discontig_fix_result(ac, gd, res);
|
| 1871 |
|
| 1872 | |
| 1873 | * sr_bg_blkno might have been changed by
|
| 1874 | * ocfs2_bg_discontig_fix_result
|
| 1875 |
|
| 1876 | res->sr_bg_stable_blkno = group_bh->b_blocknr;
|
| 1877 |
|
| 1878 | if (ac->ac_find_loc_only)
|
| 1879 | goto out_loc_only;
|
| 1880 |
|
| 1881 | ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
|
| 1882 | res->sr_bits,
|
| 1883 | le16_to_cpu(gd->bg_chain));
|
| 1884 | if (ret < 0) {
|
| 1885 | mlog_errno(ret);
|
| 1886 | goto out;
|
| 1887 | }
|
| 1888 |
|
| 1889 | ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
|
| 1890 | res->sr_bit_offset, res->sr_bits,
|
| 1891 | res->sr_max_contig_bits, 0);
|
| 1892 | if (ret < 0) {
|
| 1893 | ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
|
| 1894 | res->sr_bits,
|
| 1895 | le16_to_cpu(gd->bg_chain));
|
| 1896 | mlog_errno(ret);
|
| 1897 | }
|
| 1898 |
|
| 1899 | out_loc_only:
|
| 1900 | *bits_left = le16_to_cpu(gd->bg_free_bits_count);
|
| 1901 |
|
| 1902 | out:
|
| 1903 | brelse(group_bh);
|
| 1904 |
|
| 1905 | return ret;
|
| 1906 | }
|
| 1907 |
|
| 1908 | static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
|
| 1909 | handle_t *handle,
|
| 1910 | u32 bits_wanted,
|
| 1911 | u32 min_bits,
|
| 1912 | struct ocfs2_suballoc_result *res,
|
| 1913 | u16 *bits_left)
|
| 1914 | {
|
| 1915 | int status;
|
| 1916 | u16 chain;
|
| 1917 | u32 contig_bits;
|
| 1918 | u64 next_group;
|
| 1919 | struct inode *alloc_inode = ac->ac_inode;
|
| 1920 | struct buffer_head *group_bh = NULL;
|
| 1921 | struct buffer_head *prev_group_bh = NULL;
|
| 1922 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
|
| 1923 | struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
|
| 1924 | struct ocfs2_group_desc *bg;
|
| 1925 |
|
| 1926 | chain = ac->ac_chain;
|
| 1927 | trace_ocfs2_search_chain_begin(
|
| 1928 | (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
|
| 1929 | bits_wanted, chain);
|
| 1930 |
|
| 1931 | status = ocfs2_read_group_descriptor(alloc_inode, fe,
|
| 1932 | le64_to_cpu(cl->cl_recs[chain].c_blkno),
|
| 1933 | &group_bh);
|
| 1934 | if (status < 0) {
|
| 1935 | mlog_errno(status);
|
| 1936 | goto bail;
|
| 1937 | }
|
| 1938 | bg = (struct ocfs2_group_desc *) group_bh->b_data;
|
| 1939 |
|
| 1940 | status = -ENOSPC;
|
| 1941 | |
| 1942 | * the 1st group with any empty bits.
|
| 1943 | while (1) {
|
| 1944 | if (ac->ac_which == OCFS2_AC_USE_MAIN_DISCONTIG) {
|
| 1945 | contig_bits = le16_to_cpu(bg->bg_contig_free_bits);
|
| 1946 | if (!contig_bits)
|
| 1947 | contig_bits = ocfs2_find_max_contig_free_bits(bg->bg_bitmap,
|
| 1948 | le16_to_cpu(bg->bg_bits), 0);
|
| 1949 | if (bits_wanted > contig_bits && contig_bits >= min_bits)
|
| 1950 | bits_wanted = contig_bits;
|
| 1951 | }
|
| 1952 |
|
| 1953 | status = ac->ac_group_search(alloc_inode, group_bh,
|
| 1954 | bits_wanted, min_bits,
|
| 1955 | ac->ac_max_block, res);
|
| 1956 | if (status != -ENOSPC)
|
| 1957 | break;
|
| 1958 | if (!bg->bg_next_group)
|
| 1959 | break;
|
| 1960 |
|
| 1961 | brelse(prev_group_bh);
|
| 1962 | prev_group_bh = NULL;
|
| 1963 |
|
| 1964 | next_group = le64_to_cpu(bg->bg_next_group);
|
| 1965 | prev_group_bh = group_bh;
|
| 1966 | group_bh = NULL;
|
| 1967 | status = ocfs2_read_group_descriptor(alloc_inode, fe,
|
| 1968 | next_group, &group_bh);
|
| 1969 | if (status < 0) {
|
| 1970 | mlog_errno(status);
|
| 1971 | goto bail;
|
| 1972 | }
|
| 1973 | bg = (struct ocfs2_group_desc *) group_bh->b_data;
|
| 1974 | }
|
| 1975 | if (status < 0) {
|
| 1976 | if (status != -ENOSPC)
|
| 1977 | mlog_errno(status);
|
| 1978 | goto bail;
|
| 1979 | }
|
| 1980 |
|
| 1981 | trace_ocfs2_search_chain_succ(
|
| 1982 | (unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
|
| 1983 |
|
| 1984 | res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
|
| 1985 |
|
| 1986 | BUG_ON(res->sr_bits == 0);
|
| 1987 | if (!status)
|
| 1988 | ocfs2_bg_discontig_fix_result(ac, bg, res);
|
| 1989 |
|
| 1990 | |
| 1991 | * sr_bg_blkno might have been changed by
|
| 1992 | * ocfs2_bg_discontig_fix_result
|
| 1993 |
|
| 1994 | res->sr_bg_stable_blkno = group_bh->b_blocknr;
|
| 1995 |
|
| 1996 | |
| 1997 | * Keep track of previous block descriptor read. When
|
| 1998 | * we find a target, if we have read more than X
|
| 1999 | * number of descriptors, and the target is reasonably
|
| 2000 | * empty, relink him to top of his chain.
|
| 2001 | *
|
| 2002 | * We've read 0 extra blocks and only send one more to
|
| 2003 | * the transaction, yet the next guy to search has a
|
| 2004 | * much easier time.
|
| 2005 | *
|
| 2006 | * Do this *after* figuring out how many bits we're taking out
|
| 2007 | * of our target group.
|
| 2008 |
|
| 2009 | if (!ac->ac_disable_chain_relink &&
|
| 2010 | (prev_group_bh) &&
|
| 2011 | (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
|
| 2012 | status = ocfs2_relink_block_group(handle, alloc_inode,
|
| 2013 | ac->ac_bh, group_bh,
|
| 2014 | prev_group_bh, chain);
|
| 2015 | if (status < 0) {
|
| 2016 | mlog_errno(status);
|
| 2017 | goto bail;
|
| 2018 | }
|
| 2019 | }
|
| 2020 |
|
| 2021 | if (ac->ac_find_loc_only)
|
| 2022 | goto out_loc_only;
|
| 2023 |
|
| 2024 | status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
|
| 2025 | ac->ac_bh, res->sr_bits,
|
| 2026 | chain);
|
| 2027 | if (status) {
|
| 2028 | mlog_errno(status);
|
| 2029 | goto bail;
|
| 2030 | }
|
| 2031 |
|
| 2032 | status = ocfs2_block_group_set_bits(handle,
|
| 2033 | alloc_inode,
|
| 2034 | bg,
|
| 2035 | group_bh,
|
| 2036 | res->sr_bit_offset,
|
| 2037 | res->sr_bits,
|
| 2038 | res->sr_max_contig_bits,
|
| 2039 | 0);
|
| 2040 | if (status < 0) {
|
| 2041 | ocfs2_rollback_alloc_dinode_counts(alloc_inode,
|
| 2042 | ac->ac_bh, res->sr_bits, chain);
|
| 2043 | mlog_errno(status);
|
| 2044 | goto bail;
|
| 2045 | }
|
| 2046 |
|
| 2047 | trace_ocfs2_search_chain_end(
|
| 2048 | (unsigned long long)le64_to_cpu(fe->i_blkno),
|
| 2049 | res->sr_bits);
|
| 2050 |
|
| 2051 | out_loc_only:
|
| 2052 | *bits_left = le16_to_cpu(bg->bg_free_bits_count);
|
| 2053 | bail:
|
| 2054 | brelse(group_bh);
|
| 2055 | brelse(prev_group_bh);
|
| 2056 |
|
| 2057 | if (status)
|
| 2058 | mlog_errno(status);
|
| 2059 | return status;
|
| 2060 | }
|
| 2061 |
|
| 2062 |
|
| 2063 | static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
|
| 2064 | handle_t *handle,
|
| 2065 | u32 bits_wanted,
|
| 2066 | u32 min_bits,
|
| 2067 | struct ocfs2_suballoc_result *res)
|
| 2068 | {
|
| 2069 | int status;
|
| 2070 | int released = 0;
|
| 2071 | u16 victim, i;
|
| 2072 | u16 bits_left = 0;
|
| 2073 | u64 hint = ac->ac_last_group;
|
| 2074 | struct ocfs2_chain_list *cl;
|
| 2075 | struct ocfs2_dinode *fe;
|
| 2076 |
|
| 2077 | BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
|
| 2078 | BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
|
| 2079 | BUG_ON(!ac->ac_bh);
|
| 2080 |
|
| 2081 | fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
|
| 2082 |
|
| 2083 | |
| 2084 | * ocfs2_reserve_suballoc_bits(). Any corruption is a code bug.
|
| 2085 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
|
| 2086 |
|
| 2087 | if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
|
| 2088 | le32_to_cpu(fe->id1.bitmap1.i_total)) {
|
| 2089 | status = ocfs2_error(ac->ac_inode->i_sb,
|
| 2090 | "Chain allocator dinode %llu has %u used bits but only %u total\n",
|
| 2091 | (unsigned long long)le64_to_cpu(fe->i_blkno),
|
| 2092 | le32_to_cpu(fe->id1.bitmap1.i_used),
|
| 2093 | le32_to_cpu(fe->id1.bitmap1.i_total));
|
| 2094 | goto bail;
|
| 2095 | }
|
| 2096 |
|
| 2097 |
|
| 2098 | res->sr_bg_blkno = hint;
|
| 2099 | if (res->sr_bg_blkno) {
|
| 2100 | |
| 2101 | * by jumping straight to the most recently used
|
| 2102 | * allocation group. This helps us maintain some
|
| 2103 | * contiguousness across allocations.
|
| 2104 | status = ocfs2_search_one_group(ac, handle, bits_wanted,
|
| 2105 | min_bits, res, &bits_left,
|
| 2106 | &released);
|
| 2107 | if (released) {
|
| 2108 | res->sr_bg_blkno = 0;
|
| 2109 | goto chain_search;
|
| 2110 | }
|
| 2111 | if (!status)
|
| 2112 | goto set_hint;
|
| 2113 | if (status < 0 && status != -ENOSPC) {
|
| 2114 | mlog_errno(status);
|
| 2115 | goto bail;
|
| 2116 | }
|
| 2117 | }
|
| 2118 | chain_search:
|
| 2119 | cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
|
| 2120 | if (!le16_to_cpu(cl->cl_next_free_rec) ||
|
| 2121 | le16_to_cpu(cl->cl_next_free_rec) > le16_to_cpu(cl->cl_count)) {
|
| 2122 | status = ocfs2_error(ac->ac_inode->i_sb,
|
| 2123 | "Chain allocator dinode %llu has invalid next "
|
| 2124 | "free chain record %u, but only %u total\n",
|
| 2125 | (unsigned long long)le64_to_cpu(fe->i_blkno),
|
| 2126 | le16_to_cpu(cl->cl_next_free_rec),
|
| 2127 | le16_to_cpu(cl->cl_count));
|
| 2128 | goto bail;
|
| 2129 | }
|
| 2130 |
|
| 2131 | victim = ocfs2_find_victim_chain(cl);
|
| 2132 | ac->ac_chain = victim;
|
| 2133 |
|
| 2134 | search:
|
| 2135 | status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
|
| 2136 | res, &bits_left);
|
| 2137 | if (!status) {
|
| 2138 | if (ocfs2_is_cluster_bitmap(ac->ac_inode))
|
| 2139 | hint = res->sr_bg_blkno;
|
| 2140 | else
|
| 2141 | hint = ocfs2_group_from_res(res);
|
| 2142 | goto set_hint;
|
| 2143 | }
|
| 2144 | if (status < 0 && status != -ENOSPC) {
|
| 2145 | mlog_errno(status);
|
| 2146 | goto bail;
|
| 2147 | }
|
| 2148 |
|
| 2149 | trace_ocfs2_claim_suballoc_bits(victim);
|
| 2150 |
|
| 2151 | |
| 2152 | * searching each chain in order. Don't allow chain relinking
|
| 2153 | * because we only calculate enough journal credits for one
|
| 2154 | * relink per alloc.
|
| 2155 | ac->ac_disable_chain_relink = 1;
|
| 2156 | for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
|
| 2157 | if (i == victim)
|
| 2158 | continue;
|
| 2159 | if (le32_to_cpu(cl->cl_recs[i].c_free) < bits_wanted)
|
| 2160 | continue;
|
| 2161 |
|
| 2162 | ac->ac_chain = i;
|
| 2163 | status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
|
| 2164 | res, &bits_left);
|
| 2165 | if (!status) {
|
| 2166 | hint = ocfs2_group_from_res(res);
|
| 2167 | break;
|
| 2168 | }
|
| 2169 | if (status < 0 && status != -ENOSPC) {
|
| 2170 | mlog_errno(status);
|
| 2171 | goto bail;
|
| 2172 | }
|
| 2173 | }
|
| 2174 |
|
| 2175 | |
| 2176 | * We should switch to using every single bit when allocating
|
| 2177 | * from the global bitmap.
|
| 2178 | if (i == le16_to_cpu(cl->cl_next_free_rec) &&
|
| 2179 | status == -ENOSPC && ac->ac_which == OCFS2_AC_USE_MAIN) {
|
| 2180 | ac->ac_which = OCFS2_AC_USE_MAIN_DISCONTIG;
|
| 2181 | ac->ac_chain = victim;
|
| 2182 | goto search;
|
| 2183 | }
|
| 2184 |
|
| 2185 | set_hint:
|
| 2186 | if (status != -ENOSPC) {
|
| 2187 | |
| 2188 | * yield a suitable extent, then we reset the last
|
| 2189 | * group hint so as to not waste a disk read
|
| 2190 | if (bits_left < min_bits)
|
| 2191 | ac->ac_last_group = 0;
|
| 2192 | else
|
| 2193 | ac->ac_last_group = hint;
|
| 2194 | }
|
| 2195 |
|
| 2196 | bail:
|
| 2197 | if (status)
|
| 2198 | mlog_errno(status);
|
| 2199 | return status;
|
| 2200 | }
|
| 2201 |
|
| 2202 | int ocfs2_claim_metadata(handle_t *handle,
|
| 2203 | struct ocfs2_alloc_context *ac,
|
| 2204 | u32 bits_wanted,
|
| 2205 | u64 *suballoc_loc,
|
| 2206 | u16 *suballoc_bit_start,
|
| 2207 | unsigned int *num_bits,
|
| 2208 | u64 *blkno_start)
|
| 2209 | {
|
| 2210 | int status;
|
| 2211 | struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
|
| 2212 |
|
| 2213 | BUG_ON(!ac);
|
| 2214 | BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
|
| 2215 | BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
|
| 2216 |
|
| 2217 | status = ocfs2_claim_suballoc_bits(ac,
|
| 2218 | handle,
|
| 2219 | bits_wanted,
|
| 2220 | 1,
|
| 2221 | &res);
|
| 2222 | if (status < 0) {
|
| 2223 | mlog_errno(status);
|
| 2224 | goto bail;
|
| 2225 | }
|
| 2226 | atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
|
| 2227 |
|
| 2228 | *suballoc_loc = res.sr_bg_blkno;
|
| 2229 | *suballoc_bit_start = res.sr_bit_offset;
|
| 2230 | *blkno_start = res.sr_blkno;
|
| 2231 | ac->ac_bits_given += res.sr_bits;
|
| 2232 | *num_bits = res.sr_bits;
|
| 2233 | status = 0;
|
| 2234 | bail:
|
| 2235 | if (status)
|
| 2236 | mlog_errno(status);
|
| 2237 | return status;
|
| 2238 | }
|
| 2239 |
|
| 2240 | |
| 2241 | * after ocfs2 has the ability to release block group unused space,
|
| 2242 | * the ->ip_last_used_group may be invalid. so this function returns
|
| 2243 | * ac->ac_last_group need to verify.
|
| 2244 | * refer the 'hint' in ocfs2_claim_suballoc_bits() for more details.
|
| 2245 |
|
| 2246 | static void ocfs2_init_inode_ac_group(struct inode *dir,
|
| 2247 | struct buffer_head *parent_di_bh,
|
| 2248 | struct ocfs2_alloc_context *ac)
|
| 2249 | {
|
| 2250 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
|
| 2251 | |
| 2252 | * Try to allocate inodes from some specific group.
|
| 2253 | *
|
| 2254 | * If the parent dir has recorded the last group used in allocation,
|
| 2255 | * cool, use it. Otherwise if we try to allocate new inode from the
|
| 2256 | * same slot the parent dir belongs to, use the same chunk.
|
| 2257 | *
|
| 2258 | * We are very careful here to avoid the mistake of setting
|
| 2259 | * ac_last_group to a group descriptor from a different (unlocked) slot.
|
| 2260 |
|
| 2261 | if (OCFS2_I(dir)->ip_last_used_group &&
|
| 2262 | OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
|
| 2263 | ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
|
| 2264 | else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
|
| 2265 | if (di->i_suballoc_loc)
|
| 2266 | ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
|
| 2267 | else
|
| 2268 | ac->ac_last_group = ocfs2_which_suballoc_group(
|
| 2269 | le64_to_cpu(di->i_blkno),
|
| 2270 | le16_to_cpu(di->i_suballoc_bit));
|
| 2271 | }
|
| 2272 | }
|
| 2273 |
|
| 2274 | static inline void ocfs2_save_inode_ac_group(struct inode *dir,
|
| 2275 | struct ocfs2_alloc_context *ac)
|
| 2276 | {
|
| 2277 | OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
|
| 2278 | OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
|
| 2279 | }
|
| 2280 |
|
| 2281 | int ocfs2_find_new_inode_loc(struct inode *dir,
|
| 2282 | struct buffer_head *parent_fe_bh,
|
| 2283 | struct ocfs2_alloc_context *ac,
|
| 2284 | u64 *fe_blkno)
|
| 2285 | {
|
| 2286 | int ret;
|
| 2287 | handle_t *handle = NULL;
|
| 2288 | struct ocfs2_suballoc_result *res;
|
| 2289 |
|
| 2290 | BUG_ON(!ac);
|
| 2291 | BUG_ON(ac->ac_bits_given != 0);
|
| 2292 | BUG_ON(ac->ac_bits_wanted != 1);
|
| 2293 | BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
|
| 2294 |
|
| 2295 | res = kzalloc_obj(*res, GFP_NOFS);
|
| 2296 | if (res == NULL) {
|
| 2297 | ret = -ENOMEM;
|
| 2298 | mlog_errno(ret);
|
| 2299 | goto out;
|
| 2300 | }
|
| 2301 |
|
| 2302 | ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
|
| 2303 |
|
| 2304 | |
| 2305 | * The handle started here is for chain relink. Alternatively,
|
| 2306 | * we could just disable relink for these calls.
|
| 2307 |
|
| 2308 | handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
|
| 2309 | if (IS_ERR(handle)) {
|
| 2310 | ret = PTR_ERR(handle);
|
| 2311 | handle = NULL;
|
| 2312 | mlog_errno(ret);
|
| 2313 | goto out;
|
| 2314 | }
|
| 2315 |
|
| 2316 | |
| 2317 | * This will instruct ocfs2_claim_suballoc_bits and
|
| 2318 | * ocfs2_search_one_group to search but save actual allocation
|
| 2319 | * for later.
|
| 2320 |
|
| 2321 | ac->ac_find_loc_only = 1;
|
| 2322 |
|
| 2323 | ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
|
| 2324 | if (ret < 0) {
|
| 2325 | mlog_errno(ret);
|
| 2326 | goto out;
|
| 2327 | }
|
| 2328 |
|
| 2329 | ac->ac_find_loc_priv = res;
|
| 2330 | *fe_blkno = res->sr_blkno;
|
| 2331 | ocfs2_update_inode_fsync_trans(handle, dir, 0);
|
| 2332 | out:
|
| 2333 | if (handle)
|
| 2334 | ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
|
| 2335 |
|
| 2336 | if (ret)
|
| 2337 | kfree(res);
|
| 2338 |
|
| 2339 | return ret;
|
| 2340 | }
|
| 2341 |
|
| 2342 | int ocfs2_claim_new_inode_at_loc(handle_t *handle,
|
| 2343 | struct inode *dir,
|
| 2344 | struct ocfs2_alloc_context *ac,
|
| 2345 | u64 *suballoc_loc,
|
| 2346 | u16 *suballoc_bit,
|
| 2347 | u64 di_blkno)
|
| 2348 | {
|
| 2349 | int ret;
|
| 2350 | u16 chain;
|
| 2351 | struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
|
| 2352 | struct buffer_head *bg_bh = NULL;
|
| 2353 | struct ocfs2_group_desc *bg;
|
| 2354 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
|
| 2355 |
|
| 2356 | |
| 2357 | * Since di_blkno is being passed back in, we check for any
|
| 2358 | * inconsistencies which may have happened between
|
| 2359 | * calls. These are code bugs as di_blkno is not expected to
|
| 2360 | * change once returned from ocfs2_find_new_inode_loc()
|
| 2361 |
|
| 2362 | BUG_ON(res->sr_blkno != di_blkno);
|
| 2363 |
|
| 2364 | ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
|
| 2365 | res->sr_bg_stable_blkno, &bg_bh);
|
| 2366 | if (ret) {
|
| 2367 | mlog_errno(ret);
|
| 2368 | goto out;
|
| 2369 | }
|
| 2370 |
|
| 2371 | bg = (struct ocfs2_group_desc *) bg_bh->b_data;
|
| 2372 | chain = le16_to_cpu(bg->bg_chain);
|
| 2373 |
|
| 2374 | ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
|
| 2375 | ac->ac_bh, res->sr_bits,
|
| 2376 | chain);
|
| 2377 | if (ret) {
|
| 2378 | mlog_errno(ret);
|
| 2379 | goto out;
|
| 2380 | }
|
| 2381 |
|
| 2382 | ret = ocfs2_block_group_set_bits(handle,
|
| 2383 | ac->ac_inode,
|
| 2384 | bg,
|
| 2385 | bg_bh,
|
| 2386 | res->sr_bit_offset,
|
| 2387 | res->sr_bits,
|
| 2388 | res->sr_max_contig_bits,
|
| 2389 | 0);
|
| 2390 | if (ret < 0) {
|
| 2391 | ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
|
| 2392 | ac->ac_bh, res->sr_bits, chain);
|
| 2393 | mlog_errno(ret);
|
| 2394 | goto out;
|
| 2395 | }
|
| 2396 |
|
| 2397 | trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
|
| 2398 | res->sr_bits);
|
| 2399 |
|
| 2400 | atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
|
| 2401 |
|
| 2402 | BUG_ON(res->sr_bits != 1);
|
| 2403 |
|
| 2404 | *suballoc_loc = res->sr_bg_blkno;
|
| 2405 | *suballoc_bit = res->sr_bit_offset;
|
| 2406 | ac->ac_bits_given++;
|
| 2407 | ocfs2_save_inode_ac_group(dir, ac);
|
| 2408 |
|
| 2409 | out:
|
| 2410 | brelse(bg_bh);
|
| 2411 |
|
| 2412 | return ret;
|
| 2413 | }
|
| 2414 |
|
| 2415 | int ocfs2_claim_new_inode(handle_t *handle,
|
| 2416 | struct inode *dir,
|
| 2417 | struct buffer_head *parent_fe_bh,
|
| 2418 | struct ocfs2_alloc_context *ac,
|
| 2419 | u64 *suballoc_loc,
|
| 2420 | u16 *suballoc_bit,
|
| 2421 | u64 *fe_blkno)
|
| 2422 | {
|
| 2423 | int status;
|
| 2424 | struct ocfs2_suballoc_result res;
|
| 2425 |
|
| 2426 | BUG_ON(!ac);
|
| 2427 | BUG_ON(ac->ac_bits_given != 0);
|
| 2428 | BUG_ON(ac->ac_bits_wanted != 1);
|
| 2429 | BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
|
| 2430 |
|
| 2431 | ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
|
| 2432 |
|
| 2433 | status = ocfs2_claim_suballoc_bits(ac,
|
| 2434 | handle,
|
| 2435 | 1,
|
| 2436 | 1,
|
| 2437 | &res);
|
| 2438 | if (status < 0) {
|
| 2439 | mlog_errno(status);
|
| 2440 | goto bail;
|
| 2441 | }
|
| 2442 | atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
|
| 2443 |
|
| 2444 | BUG_ON(res.sr_bits != 1);
|
| 2445 |
|
| 2446 | *suballoc_loc = res.sr_bg_blkno;
|
| 2447 | *suballoc_bit = res.sr_bit_offset;
|
| 2448 | *fe_blkno = res.sr_blkno;
|
| 2449 | ac->ac_bits_given++;
|
| 2450 | ocfs2_save_inode_ac_group(dir, ac);
|
| 2451 | status = 0;
|
| 2452 | bail:
|
| 2453 | if (status)
|
| 2454 | mlog_errno(status);
|
| 2455 | return status;
|
| 2456 | }
|
| 2457 |
|
| 2458 | |
| 2459 | * disk cluster offset.
|
| 2460 | static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
|
| 2461 | u64 bg_blkno,
|
| 2462 | u16 bg_bit_off)
|
| 2463 | {
|
| 2464 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
|
| 2465 | u32 cluster = 0;
|
| 2466 |
|
| 2467 | BUG_ON(!ocfs2_is_cluster_bitmap(inode));
|
| 2468 |
|
| 2469 | if (bg_blkno != osb->first_cluster_group_blkno)
|
| 2470 | cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
|
| 2471 | cluster += (u32) bg_bit_off;
|
| 2472 | return cluster;
|
| 2473 | }
|
| 2474 |
|
| 2475 | |
| 2476 | * and return that block offset.
|
| 2477 | u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
|
| 2478 | {
|
| 2479 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
|
| 2480 | u32 group_no;
|
| 2481 |
|
| 2482 | BUG_ON(!ocfs2_is_cluster_bitmap(inode));
|
| 2483 |
|
| 2484 | group_no = cluster / osb->bitmap_cpg;
|
| 2485 | if (!group_no)
|
| 2486 | return osb->first_cluster_group_blkno;
|
| 2487 | return ocfs2_clusters_to_blocks(inode->i_sb,
|
| 2488 | group_no * osb->bitmap_cpg);
|
| 2489 | }
|
| 2490 |
|
| 2491 | |
| 2492 | * group and descriptor bitmap offset that corresponds to.
|
| 2493 | static inline void ocfs2_block_to_cluster_group(struct inode *inode,
|
| 2494 | u64 data_blkno,
|
| 2495 | u64 *bg_blkno,
|
| 2496 | u16 *bg_bit_off)
|
| 2497 | {
|
| 2498 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
|
| 2499 | u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
|
| 2500 |
|
| 2501 | BUG_ON(!ocfs2_is_cluster_bitmap(inode));
|
| 2502 |
|
| 2503 | *bg_blkno = ocfs2_which_cluster_group(inode,
|
| 2504 | data_cluster);
|
| 2505 |
|
| 2506 | if (*bg_blkno == osb->first_cluster_group_blkno)
|
| 2507 | *bg_bit_off = (u16) data_cluster;
|
| 2508 | else
|
| 2509 | *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
|
| 2510 | data_blkno - *bg_blkno);
|
| 2511 | }
|
| 2512 |
|
| 2513 | |
| 2514 | * min_bits - minimum contiguous chunk from this total allocation we
|
| 2515 | * can handle. set to what we asked for originally for a full
|
| 2516 | * contig. allocation, set to '1' to indicate we can deal with extents
|
| 2517 | * of any size.
|
| 2518 |
|
| 2519 | int __ocfs2_claim_clusters(handle_t *handle,
|
| 2520 | struct ocfs2_alloc_context *ac,
|
| 2521 | u32 min_clusters,
|
| 2522 | u32 max_clusters,
|
| 2523 | u32 *cluster_start,
|
| 2524 | u32 *num_clusters)
|
| 2525 | {
|
| 2526 | int status;
|
| 2527 | unsigned int bits_wanted = max_clusters;
|
| 2528 | struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
|
| 2529 | struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
|
| 2530 |
|
| 2531 | BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
|
| 2532 |
|
| 2533 | BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
|
| 2534 | && ac->ac_which != OCFS2_AC_USE_MAIN
|
| 2535 | && ac->ac_which != OCFS2_AC_USE_MAIN_DISCONTIG);
|
| 2536 |
|
| 2537 | if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
|
| 2538 | WARN_ON(min_clusters > 1);
|
| 2539 |
|
| 2540 | status = ocfs2_claim_local_alloc_bits(osb,
|
| 2541 | handle,
|
| 2542 | ac,
|
| 2543 | bits_wanted,
|
| 2544 | cluster_start,
|
| 2545 | num_clusters);
|
| 2546 | if (!status)
|
| 2547 | atomic_inc(&osb->alloc_stats.local_data);
|
| 2548 | } else {
|
| 2549 | if (min_clusters > (osb->bitmap_cpg - 1)) {
|
| 2550 | |
| 2551 | * should know about this already.
|
| 2552 | mlog(ML_ERROR, "minimum allocation requested %u exceeds "
|
| 2553 | "group bitmap size %u!\n", min_clusters,
|
| 2554 | osb->bitmap_cpg);
|
| 2555 | status = -ENOSPC;
|
| 2556 | goto bail;
|
| 2557 | }
|
| 2558 |
|
| 2559 | if (bits_wanted > (osb->bitmap_cpg - 1))
|
| 2560 | bits_wanted = osb->bitmap_cpg - 1;
|
| 2561 |
|
| 2562 | status = ocfs2_claim_suballoc_bits(ac,
|
| 2563 | handle,
|
| 2564 | bits_wanted,
|
| 2565 | min_clusters,
|
| 2566 | &res);
|
| 2567 | if (!status) {
|
| 2568 | BUG_ON(res.sr_blkno);
|
| 2569 | *cluster_start =
|
| 2570 | ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
|
| 2571 | res.sr_bg_blkno,
|
| 2572 | res.sr_bit_offset);
|
| 2573 | atomic_inc(&osb->alloc_stats.bitmap_data);
|
| 2574 | *num_clusters = res.sr_bits;
|
| 2575 | }
|
| 2576 | }
|
| 2577 | if (status < 0) {
|
| 2578 | if (status != -ENOSPC)
|
| 2579 | mlog_errno(status);
|
| 2580 | goto bail;
|
| 2581 | }
|
| 2582 |
|
| 2583 | ac->ac_bits_given += *num_clusters;
|
| 2584 |
|
| 2585 | bail:
|
| 2586 | if (status)
|
| 2587 | mlog_errno(status);
|
| 2588 | return status;
|
| 2589 | }
|
| 2590 |
|
| 2591 | int ocfs2_claim_clusters(handle_t *handle,
|
| 2592 | struct ocfs2_alloc_context *ac,
|
| 2593 | u32 min_clusters,
|
| 2594 | u32 *cluster_start,
|
| 2595 | u32 *num_clusters)
|
| 2596 | {
|
| 2597 | unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
|
| 2598 |
|
| 2599 | return __ocfs2_claim_clusters(handle, ac, min_clusters,
|
| 2600 | bits_wanted, cluster_start, num_clusters);
|
| 2601 | }
|
| 2602 |
|
| 2603 | static int ocfs2_block_group_clear_bits(handle_t *handle,
|
| 2604 | struct inode *alloc_inode,
|
| 2605 | struct ocfs2_group_desc *bg,
|
| 2606 | struct buffer_head *group_bh,
|
| 2607 | unsigned int bit_off,
|
| 2608 | unsigned int num_bits,
|
| 2609 | unsigned int max_contig_bits,
|
| 2610 | void (*undo_fn)(unsigned int bit,
|
| 2611 | unsigned long *bmap))
|
| 2612 | {
|
| 2613 | int status;
|
| 2614 | unsigned int tmp;
|
| 2615 | u16 contig_bits;
|
| 2616 | struct ocfs2_group_desc *undo_bg = NULL;
|
| 2617 | struct journal_head *jh;
|
| 2618 |
|
| 2619 | |
| 2620 | * ocfs2_read_group_descriptor(). Any corruption is a code bug.
|
| 2621 | BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
|
| 2622 |
|
| 2623 | trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
|
| 2624 |
|
| 2625 | BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
|
| 2626 | status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
|
| 2627 | group_bh,
|
| 2628 | undo_fn ?
|
| 2629 | OCFS2_JOURNAL_ACCESS_UNDO :
|
| 2630 | OCFS2_JOURNAL_ACCESS_WRITE);
|
| 2631 | if (status < 0) {
|
| 2632 | mlog_errno(status);
|
| 2633 | goto bail;
|
| 2634 | }
|
| 2635 |
|
| 2636 | jh = bh2jh(group_bh);
|
| 2637 | if (undo_fn) {
|
| 2638 | spin_lock(&jh->b_state_lock);
|
| 2639 | undo_bg = (struct ocfs2_group_desc *) jh->b_committed_data;
|
| 2640 | BUG_ON(!undo_bg);
|
| 2641 | }
|
| 2642 |
|
| 2643 | tmp = num_bits;
|
| 2644 | while(tmp--) {
|
| 2645 | ocfs2_clear_bit((bit_off + tmp),
|
| 2646 | (unsigned long *) bg->bg_bitmap);
|
| 2647 | if (undo_fn)
|
| 2648 | undo_fn(bit_off + tmp,
|
| 2649 | (unsigned long *) undo_bg->bg_bitmap);
|
| 2650 | }
|
| 2651 | le16_add_cpu(&bg->bg_free_bits_count, num_bits);
|
| 2652 | if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
|
| 2653 | if (undo_fn)
|
| 2654 | spin_unlock(&jh->b_state_lock);
|
| 2655 | return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
|
| 2656 | (unsigned long long)le64_to_cpu(bg->bg_blkno),
|
| 2657 | le16_to_cpu(bg->bg_bits),
|
| 2658 | le16_to_cpu(bg->bg_free_bits_count),
|
| 2659 | num_bits);
|
| 2660 | }
|
| 2661 |
|
| 2662 | |
| 2663 | * TODO: even 'num_bits == 1' (the worst case, release 1 cluster),
|
| 2664 | * we still need to rescan whole bitmap.
|
| 2665 |
|
| 2666 | if (ocfs2_is_cluster_bitmap(alloc_inode)) {
|
| 2667 | contig_bits = ocfs2_find_max_contig_free_bits(bg->bg_bitmap,
|
| 2668 | le16_to_cpu(bg->bg_bits), 0);
|
| 2669 | if (contig_bits > max_contig_bits)
|
| 2670 | max_contig_bits = contig_bits;
|
| 2671 | bg->bg_contig_free_bits = cpu_to_le16(max_contig_bits);
|
| 2672 | } else {
|
| 2673 | bg->bg_contig_free_bits = 0;
|
| 2674 | }
|
| 2675 |
|
| 2676 | if (undo_fn)
|
| 2677 | spin_unlock(&jh->b_state_lock);
|
| 2678 |
|
| 2679 | ocfs2_journal_dirty(handle, group_bh);
|
| 2680 | bail:
|
| 2681 | return status;
|
| 2682 | }
|
| 2683 |
|
| 2684 | |
| 2685 | * Reclaim the suballocator managed space to main bitmap.
|
| 2686 | * This function first works on the suballocator to perform the
|
| 2687 | * cleanup rec/alloc_inode job, then switches to the main bitmap
|
| 2688 | * to reclaim released space.
|
| 2689 | *
|
| 2690 | * handle: The transaction handle
|
| 2691 | * alloc_inode: The suballoc inode
|
| 2692 | * alloc_bh: The buffer_head of suballoc inode
|
| 2693 | * group_bh: The group descriptor buffer_head of suballocator managed.
|
| 2694 | * Caller should release the input group_bh.
|
| 2695 |
|
| 2696 | static int _ocfs2_reclaim_suballoc_to_main(handle_t *handle,
|
| 2697 | struct inode *alloc_inode,
|
| 2698 | struct buffer_head *alloc_bh,
|
| 2699 | struct buffer_head *group_bh)
|
| 2700 | {
|
| 2701 | int idx, status = 0;
|
| 2702 | int i, next_free_rec, len = 0;
|
| 2703 | __le16 old_bg_contig_free_bits = 0;
|
| 2704 | u16 start_bit;
|
| 2705 | u32 tmp_used;
|
| 2706 | u64 bg_blkno, start_blk;
|
| 2707 | unsigned int count;
|
| 2708 | struct ocfs2_chain_rec *rec;
|
| 2709 | struct buffer_head *main_bm_bh = NULL;
|
| 2710 | struct inode *main_bm_inode = NULL;
|
| 2711 | struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
|
| 2712 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
|
| 2713 | struct ocfs2_chain_list *cl = &fe->id2.i_chain;
|
| 2714 | struct ocfs2_group_desc *group = (struct ocfs2_group_desc *) group_bh->b_data;
|
| 2715 |
|
| 2716 | idx = le16_to_cpu(group->bg_chain);
|
| 2717 | rec = &(cl->cl_recs[idx]);
|
| 2718 |
|
| 2719 | status = ocfs2_extend_trans(handle,
|
| 2720 | ocfs2_calc_group_alloc_credits(osb->sb,
|
| 2721 | le16_to_cpu(cl->cl_cpg)));
|
| 2722 | if (status) {
|
| 2723 | mlog_errno(status);
|
| 2724 | goto bail;
|
| 2725 | }
|
| 2726 | status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
|
| 2727 | alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
|
| 2728 | if (status < 0) {
|
| 2729 | mlog_errno(status);
|
| 2730 | goto bail;
|
| 2731 | }
|
| 2732 |
|
| 2733 | |
| 2734 | * Only clear the suballocator rec item in-place.
|
| 2735 | *
|
| 2736 | * If idx is not the last, we don't compress (remove the empty item)
|
| 2737 | * the cl_recs[]. If not, we need to do lots jobs.
|
| 2738 | *
|
| 2739 | * Compress cl_recs[] code example:
|
| 2740 | * if (idx != cl->cl_next_free_rec - 1)
|
| 2741 | * memmove(&cl->cl_recs[idx], &cl->cl_recs[idx + 1],
|
| 2742 | * sizeof(struct ocfs2_chain_rec) *
|
| 2743 | * (cl->cl_next_free_rec - idx - 1));
|
| 2744 | * for(i = idx; i < cl->cl_next_free_rec-1; i++) {
|
| 2745 | * group->bg_chain = "later group->bg_chain";
|
| 2746 | * group->bg_blkno = xxx;
|
| 2747 | * ... ...
|
| 2748 | * }
|
| 2749 |
|
| 2750 |
|
| 2751 | tmp_used = le32_to_cpu(fe->id1.bitmap1.i_total);
|
| 2752 | fe->id1.bitmap1.i_total = cpu_to_le32(tmp_used - le32_to_cpu(rec->c_total));
|
| 2753 |
|
| 2754 |
|
| 2755 | tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
|
| 2756 | fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - 1);
|
| 2757 |
|
| 2758 | tmp_used = le32_to_cpu(fe->i_clusters);
|
| 2759 | fe->i_clusters = cpu_to_le32(tmp_used - le16_to_cpu(cl->cl_cpg));
|
| 2760 |
|
| 2761 | spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
|
| 2762 | OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
|
| 2763 | fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
|
| 2764 | le32_to_cpu(fe->i_clusters)));
|
| 2765 | spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
|
| 2766 | i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
|
| 2767 | alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
|
| 2768 |
|
| 2769 | ocfs2_journal_dirty(handle, alloc_bh);
|
| 2770 | ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
|
| 2771 |
|
| 2772 | start_blk = le64_to_cpu(rec->c_blkno);
|
| 2773 | count = le32_to_cpu(rec->c_total) / le16_to_cpu(cl->cl_bpc);
|
| 2774 |
|
| 2775 | |
| 2776 | * If the rec is the last one, let's compress the chain list by
|
| 2777 | * removing the empty cl_recs[] at the end.
|
| 2778 |
|
| 2779 | next_free_rec = le16_to_cpu(cl->cl_next_free_rec);
|
| 2780 | if (idx == (next_free_rec - 1)) {
|
| 2781 | len++;
|
| 2782 | for (i = (next_free_rec - 2); i > 0; i--) {
|
| 2783 | if (cl->cl_recs[i].c_free == cl->cl_recs[i].c_total)
|
| 2784 | len++;
|
| 2785 | else
|
| 2786 | break;
|
| 2787 | }
|
| 2788 | }
|
| 2789 | le16_add_cpu(&cl->cl_next_free_rec, -len);
|
| 2790 |
|
| 2791 | rec->c_free = 0;
|
| 2792 | rec->c_total = 0;
|
| 2793 | rec->c_blkno = 0;
|
| 2794 | ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), group_bh);
|
| 2795 | memset(group, 0, sizeof(struct ocfs2_group_desc));
|
| 2796 |
|
| 2797 |
|
| 2798 | main_bm_inode = ocfs2_get_system_file_inode(osb,
|
| 2799 | GLOBAL_BITMAP_SYSTEM_INODE,
|
| 2800 | OCFS2_INVALID_SLOT);
|
| 2801 | if (!main_bm_inode)
|
| 2802 | goto bail;
|
| 2803 |
|
| 2804 | inode_lock(main_bm_inode);
|
| 2805 |
|
| 2806 | status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
|
| 2807 | if (status < 0)
|
| 2808 | goto free_bm_inode;
|
| 2809 |
|
| 2810 | ocfs2_block_to_cluster_group(main_bm_inode, start_blk, &bg_blkno,
|
| 2811 | &start_bit);
|
| 2812 | fe = (struct ocfs2_dinode *) main_bm_bh->b_data;
|
| 2813 | cl = &fe->id2.i_chain;
|
| 2814 |
|
| 2815 | group_bh = NULL;
|
| 2816 |
|
| 2817 |
|
| 2818 | status = ocfs2_read_group_descriptor(main_bm_inode, fe, bg_blkno,
|
| 2819 | &group_bh);
|
| 2820 | if (status < 0) {
|
| 2821 | mlog_errno(status);
|
| 2822 | goto free_bm_bh;
|
| 2823 | }
|
| 2824 | group = (struct ocfs2_group_desc *) group_bh->b_data;
|
| 2825 |
|
| 2826 | if ((count + start_bit) > le16_to_cpu(group->bg_bits)) {
|
| 2827 | ocfs2_error(alloc_inode->i_sb,
|
| 2828 | "reclaim length (%d) beyands block group length (%d)",
|
| 2829 | count + start_bit, le16_to_cpu(group->bg_bits));
|
| 2830 | goto free_group_bh;
|
| 2831 | }
|
| 2832 |
|
| 2833 | old_bg_contig_free_bits = group->bg_contig_free_bits;
|
| 2834 | status = ocfs2_block_group_clear_bits(handle, main_bm_inode,
|
| 2835 | group, group_bh,
|
| 2836 | start_bit, count, 0,
|
| 2837 | _ocfs2_clear_bit);
|
| 2838 | if (status < 0) {
|
| 2839 | mlog_errno(status);
|
| 2840 | goto free_group_bh;
|
| 2841 | }
|
| 2842 |
|
| 2843 | status = ocfs2_journal_access_di(handle, INODE_CACHE(main_bm_inode),
|
| 2844 | main_bm_bh, OCFS2_JOURNAL_ACCESS_WRITE);
|
| 2845 | if (status < 0) {
|
| 2846 | mlog_errno(status);
|
| 2847 | ocfs2_block_group_set_bits(handle, main_bm_inode, group, group_bh,
|
| 2848 | start_bit, count,
|
| 2849 | le16_to_cpu(old_bg_contig_free_bits), 1);
|
| 2850 | goto free_group_bh;
|
| 2851 | }
|
| 2852 |
|
| 2853 | idx = le16_to_cpu(group->bg_chain);
|
| 2854 | rec = &(cl->cl_recs[idx]);
|
| 2855 |
|
| 2856 | le32_add_cpu(&rec->c_free, count);
|
| 2857 | tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
|
| 2858 | fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
|
| 2859 | ocfs2_journal_dirty(handle, main_bm_bh);
|
| 2860 |
|
| 2861 | free_group_bh:
|
| 2862 | brelse(group_bh);
|
| 2863 |
|
| 2864 | free_bm_bh:
|
| 2865 | ocfs2_inode_unlock(main_bm_inode, 1);
|
| 2866 | brelse(main_bm_bh);
|
| 2867 |
|
| 2868 | free_bm_inode:
|
| 2869 | inode_unlock(main_bm_inode);
|
| 2870 | iput(main_bm_inode);
|
| 2871 |
|
| 2872 | bail:
|
| 2873 | return status;
|
| 2874 | }
|
| 2875 |
|
| 2876 | |
| 2877 | * expects the suballoc inode to already be locked.
|
| 2878 |
|
| 2879 | static int _ocfs2_free_suballoc_bits(handle_t *handle,
|
| 2880 | struct inode *alloc_inode,
|
| 2881 | struct buffer_head *alloc_bh,
|
| 2882 | unsigned int start_bit,
|
| 2883 | u64 bg_blkno,
|
| 2884 | unsigned int count,
|
| 2885 | void (*undo_fn)(unsigned int bit,
|
| 2886 | unsigned long *bitmap))
|
| 2887 | {
|
| 2888 | int idx, status = 0;
|
| 2889 | u32 tmp_used;
|
| 2890 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
|
| 2891 | struct ocfs2_chain_list *cl = &fe->id2.i_chain;
|
| 2892 | struct buffer_head *group_bh = NULL;
|
| 2893 | struct ocfs2_group_desc *group;
|
| 2894 | struct ocfs2_chain_rec *rec;
|
| 2895 | __le16 old_bg_contig_free_bits = 0;
|
| 2896 |
|
| 2897 | |
| 2898 | * ocfs2_free_clusters(). The callers have all locked the
|
| 2899 | * allocator and gotten alloc_bh from the lock call. This
|
| 2900 | * validates the dinode buffer. Any corruption that has happened
|
| 2901 | * is a code bug.
|
| 2902 | BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
|
| 2903 | BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
|
| 2904 |
|
| 2905 | trace_ocfs2_free_suballoc_bits(
|
| 2906 | (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
|
| 2907 | (unsigned long long)bg_blkno,
|
| 2908 | start_bit, count);
|
| 2909 |
|
| 2910 | status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
|
| 2911 | &group_bh);
|
| 2912 | if (status < 0) {
|
| 2913 | mlog_errno(status);
|
| 2914 | goto bail;
|
| 2915 | }
|
| 2916 | group = (struct ocfs2_group_desc *) group_bh->b_data;
|
| 2917 |
|
| 2918 | BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
|
| 2919 |
|
| 2920 | if (ocfs2_is_cluster_bitmap(alloc_inode))
|
| 2921 | old_bg_contig_free_bits = group->bg_contig_free_bits;
|
| 2922 | status = ocfs2_block_group_clear_bits(handle, alloc_inode,
|
| 2923 | group, group_bh,
|
| 2924 | start_bit, count, 0, undo_fn);
|
| 2925 | if (status < 0) {
|
| 2926 | mlog_errno(status);
|
| 2927 | goto bail;
|
| 2928 | }
|
| 2929 |
|
| 2930 | status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
|
| 2931 | alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
|
| 2932 | if (status < 0) {
|
| 2933 | mlog_errno(status);
|
| 2934 | ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
|
| 2935 | start_bit, count,
|
| 2936 | le16_to_cpu(old_bg_contig_free_bits), 1);
|
| 2937 | goto bail;
|
| 2938 | }
|
| 2939 |
|
| 2940 | idx = le16_to_cpu(group->bg_chain);
|
| 2941 | rec = &(cl->cl_recs[idx]);
|
| 2942 |
|
| 2943 | le32_add_cpu(&rec->c_free, count);
|
| 2944 | tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
|
| 2945 | fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
|
| 2946 | ocfs2_journal_dirty(handle, alloc_bh);
|
| 2947 |
|
| 2948 | |
| 2949 | * Reclaim suballocator free space.
|
| 2950 | * Bypass: global_bitmap, non empty rec, first rec in cl_recs[]
|
| 2951 |
|
| 2952 | if (ocfs2_is_cluster_bitmap(alloc_inode) ||
|
| 2953 | (le32_to_cpu(rec->c_free) != (le32_to_cpu(rec->c_total) - 1)) ||
|
| 2954 | (le16_to_cpu(cl->cl_next_free_rec) == 1)) {
|
| 2955 | goto bail;
|
| 2956 | }
|
| 2957 |
|
| 2958 | _ocfs2_reclaim_suballoc_to_main(handle, alloc_inode, alloc_bh, group_bh);
|
| 2959 |
|
| 2960 | bail:
|
| 2961 | brelse(group_bh);
|
| 2962 | return status;
|
| 2963 | }
|
| 2964 |
|
| 2965 | int ocfs2_free_suballoc_bits(handle_t *handle,
|
| 2966 | struct inode *alloc_inode,
|
| 2967 | struct buffer_head *alloc_bh,
|
| 2968 | unsigned int start_bit,
|
| 2969 | u64 bg_blkno,
|
| 2970 | unsigned int count)
|
| 2971 | {
|
| 2972 | return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
|
| 2973 | start_bit, bg_blkno, count, NULL);
|
| 2974 | }
|
| 2975 |
|
| 2976 | int ocfs2_free_dinode(handle_t *handle,
|
| 2977 | struct inode *inode_alloc_inode,
|
| 2978 | struct buffer_head *inode_alloc_bh,
|
| 2979 | struct ocfs2_dinode *di)
|
| 2980 | {
|
| 2981 | u64 blk = le64_to_cpu(di->i_blkno);
|
| 2982 | u16 bit = le16_to_cpu(di->i_suballoc_bit);
|
| 2983 | u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
|
| 2984 |
|
| 2985 | if (di->i_suballoc_loc)
|
| 2986 | bg_blkno = le64_to_cpu(di->i_suballoc_loc);
|
| 2987 | return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
|
| 2988 | inode_alloc_bh, bit, bg_blkno, 1);
|
| 2989 | }
|
| 2990 |
|
| 2991 | static int _ocfs2_free_clusters(handle_t *handle,
|
| 2992 | struct inode *bitmap_inode,
|
| 2993 | struct buffer_head *bitmap_bh,
|
| 2994 | u64 start_blk,
|
| 2995 | unsigned int num_clusters,
|
| 2996 | void (*undo_fn)(unsigned int bit,
|
| 2997 | unsigned long *bitmap))
|
| 2998 | {
|
| 2999 | int status;
|
| 3000 | u16 bg_start_bit;
|
| 3001 | u64 bg_blkno;
|
| 3002 |
|
| 3003 | |
| 3004 | * bigger than a block group bitmap so we never have to worry
|
| 3005 | * about looping on them.
|
| 3006 | * This is expensive. We can safely remove once this stuff has
|
| 3007 | * gotten tested really well.
|
| 3008 | BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb,
|
| 3009 | ocfs2_blocks_to_clusters(bitmap_inode->i_sb,
|
| 3010 | start_blk)));
|
| 3011 |
|
| 3012 |
|
| 3013 | ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
|
| 3014 | &bg_start_bit);
|
| 3015 |
|
| 3016 | trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
|
| 3017 | (unsigned long long)start_blk,
|
| 3018 | bg_start_bit, num_clusters);
|
| 3019 |
|
| 3020 | status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
|
| 3021 | bg_start_bit, bg_blkno,
|
| 3022 | num_clusters, undo_fn);
|
| 3023 | if (status < 0) {
|
| 3024 | mlog_errno(status);
|
| 3025 | goto out;
|
| 3026 | }
|
| 3027 |
|
| 3028 | ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
|
| 3029 | num_clusters);
|
| 3030 |
|
| 3031 | out:
|
| 3032 | return status;
|
| 3033 | }
|
| 3034 |
|
| 3035 | int ocfs2_free_clusters(handle_t *handle,
|
| 3036 | struct inode *bitmap_inode,
|
| 3037 | struct buffer_head *bitmap_bh,
|
| 3038 | u64 start_blk,
|
| 3039 | unsigned int num_clusters)
|
| 3040 | {
|
| 3041 | return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
|
| 3042 | start_blk, num_clusters,
|
| 3043 | _ocfs2_set_bit);
|
| 3044 | }
|
| 3045 |
|
| 3046 | |
| 3047 | * Give never-used clusters back to the global bitmap. We don't need
|
| 3048 | * to protect these bits in the undo buffer.
|
| 3049 |
|
| 3050 | int ocfs2_release_clusters(handle_t *handle,
|
| 3051 | struct inode *bitmap_inode,
|
| 3052 | struct buffer_head *bitmap_bh,
|
| 3053 | u64 start_blk,
|
| 3054 | unsigned int num_clusters)
|
| 3055 | {
|
| 3056 | return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
|
| 3057 | start_blk, num_clusters,
|
| 3058 | _ocfs2_clear_bit);
|
| 3059 | }
|
| 3060 |
|
| 3061 | |
| 3062 | * For a given allocation, determine which allocators will need to be
|
| 3063 | * accessed, and lock them, reserving the appropriate number of bits.
|
| 3064 | *
|
| 3065 | * Sparse file systems call this from ocfs2_write_begin_nolock()
|
| 3066 | * and ocfs2_allocate_unwritten_extents().
|
| 3067 | *
|
| 3068 | * File systems which don't support holes call this from
|
| 3069 | * ocfs2_extend_allocation().
|
| 3070 |
|
| 3071 | int ocfs2_lock_allocators(struct inode *inode,
|
| 3072 | struct ocfs2_extent_tree *et,
|
| 3073 | u32 clusters_to_add, u32 extents_to_split,
|
| 3074 | struct ocfs2_alloc_context **data_ac,
|
| 3075 | struct ocfs2_alloc_context **meta_ac)
|
| 3076 | {
|
| 3077 | int ret = 0, num_free_extents;
|
| 3078 | unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
|
| 3079 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
|
| 3080 |
|
| 3081 | *meta_ac = NULL;
|
| 3082 | if (data_ac)
|
| 3083 | *data_ac = NULL;
|
| 3084 |
|
| 3085 | BUG_ON(clusters_to_add != 0 && data_ac == NULL);
|
| 3086 |
|
| 3087 | num_free_extents = ocfs2_num_free_extents(et);
|
| 3088 | if (num_free_extents < 0) {
|
| 3089 | ret = num_free_extents;
|
| 3090 | mlog_errno(ret);
|
| 3091 | goto out;
|
| 3092 | }
|
| 3093 |
|
| 3094 | |
| 3095 | * Sparse allocation file systems need to be more conservative
|
| 3096 | * with reserving room for expansion - the actual allocation
|
| 3097 | * happens while we've got a journal handle open so re-taking
|
| 3098 | * a cluster lock (because we ran out of room for another
|
| 3099 | * extent) will violate ordering rules.
|
| 3100 | *
|
| 3101 | * Most of the time we'll only be seeing this 1 cluster at a time
|
| 3102 | * anyway.
|
| 3103 | *
|
| 3104 | * Always lock for any unwritten extents - we might want to
|
| 3105 | * add blocks during a split.
|
| 3106 |
|
| 3107 | if (!num_free_extents ||
|
| 3108 | (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
|
| 3109 | ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
|
| 3110 | if (ret < 0) {
|
| 3111 | if (ret != -ENOSPC)
|
| 3112 | mlog_errno(ret);
|
| 3113 | goto out;
|
| 3114 | }
|
| 3115 | }
|
| 3116 |
|
| 3117 | if (clusters_to_add == 0)
|
| 3118 | goto out;
|
| 3119 |
|
| 3120 | ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
|
| 3121 | if (ret < 0) {
|
| 3122 | if (ret != -ENOSPC)
|
| 3123 | mlog_errno(ret);
|
| 3124 | goto out;
|
| 3125 | }
|
| 3126 |
|
| 3127 | out:
|
| 3128 | if (ret) {
|
| 3129 | if (*meta_ac) {
|
| 3130 | ocfs2_free_alloc_context(*meta_ac);
|
| 3131 | *meta_ac = NULL;
|
| 3132 | }
|
| 3133 |
|
| 3134 | |
| 3135 | * We cannot have an error and a non null *data_ac.
|
| 3136 |
|
| 3137 | }
|
| 3138 |
|
| 3139 | return ret;
|
| 3140 | }
|
| 3141 |
|
| 3142 | |
| 3143 | * Read the inode specified by blkno to get suballoc_slot and
|
| 3144 | * suballoc_bit.
|
| 3145 |
|
| 3146 | static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
|
| 3147 | u16 *suballoc_slot, u64 *group_blkno,
|
| 3148 | u16 *suballoc_bit)
|
| 3149 | {
|
| 3150 | int status;
|
| 3151 | struct buffer_head *inode_bh = NULL;
|
| 3152 | struct ocfs2_dinode *inode_fe;
|
| 3153 |
|
| 3154 | trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
|
| 3155 |
|
| 3156 |
|
| 3157 | status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
|
| 3158 | if (status < 0) {
|
| 3159 | mlog(ML_ERROR, "read block %llu failed %d\n",
|
| 3160 | (unsigned long long)blkno, status);
|
| 3161 | goto bail;
|
| 3162 | }
|
| 3163 |
|
| 3164 | inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
|
| 3165 | if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
|
| 3166 | mlog(ML_ERROR, "invalid inode %llu requested\n",
|
| 3167 | (unsigned long long)blkno);
|
| 3168 | status = -EINVAL;
|
| 3169 | goto bail;
|
| 3170 | }
|
| 3171 |
|
| 3172 | if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
|
| 3173 | (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
|
| 3174 | mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
|
| 3175 | (unsigned long long)blkno,
|
| 3176 | (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
|
| 3177 | status = -EINVAL;
|
| 3178 | goto bail;
|
| 3179 | }
|
| 3180 |
|
| 3181 | if (suballoc_slot)
|
| 3182 | *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
|
| 3183 | if (suballoc_bit)
|
| 3184 | *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
|
| 3185 | if (group_blkno)
|
| 3186 | *group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
|
| 3187 |
|
| 3188 | bail:
|
| 3189 | brelse(inode_bh);
|
| 3190 |
|
| 3191 | if (status)
|
| 3192 | mlog_errno(status);
|
| 3193 | return status;
|
| 3194 | }
|
| 3195 |
|
| 3196 | |
| 3197 | * test whether bit is SET in allocator bitmap or not. on success, 0
|
| 3198 | * is returned and *res is 1 for SET; 0 otherwise. when fails, errno
|
| 3199 | * is returned and *res is meaningless. Call this after you have
|
| 3200 | * cluster locked against suballoc, or you may get a result based on
|
| 3201 | * non-up2date contents
|
| 3202 |
|
| 3203 | static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
|
| 3204 | struct inode *suballoc,
|
| 3205 | struct buffer_head *alloc_bh,
|
| 3206 | u64 group_blkno, u64 blkno,
|
| 3207 | u16 bit, int *res)
|
| 3208 | {
|
| 3209 | struct ocfs2_dinode *alloc_di;
|
| 3210 | struct ocfs2_group_desc *group;
|
| 3211 | struct buffer_head *group_bh = NULL;
|
| 3212 | u64 bg_blkno;
|
| 3213 | int status, quiet = 0, released = 0;
|
| 3214 |
|
| 3215 | trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
|
| 3216 | (unsigned int)bit);
|
| 3217 |
|
| 3218 | alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
|
| 3219 | if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
|
| 3220 | mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
|
| 3221 | (unsigned int)bit,
|
| 3222 | ocfs2_bits_per_group(&alloc_di->id2.i_chain));
|
| 3223 | status = -EINVAL;
|
| 3224 | goto bail;
|
| 3225 | }
|
| 3226 |
|
| 3227 | bg_blkno = group_blkno ? group_blkno :
|
| 3228 | ocfs2_which_suballoc_group(blkno, bit);
|
| 3229 | status = ocfs2_read_hint_group_descriptor(suballoc, alloc_di, bg_blkno,
|
| 3230 | &group_bh, &released);
|
| 3231 | if (released) {
|
| 3232 | quiet = 1;
|
| 3233 | status = -ESTALE;
|
| 3234 | goto bail;
|
| 3235 | } else if (status < 0) {
|
| 3236 | mlog(ML_ERROR, "read group %llu failed %d\n",
|
| 3237 | (unsigned long long)bg_blkno, status);
|
| 3238 | goto bail;
|
| 3239 | }
|
| 3240 |
|
| 3241 | group = (struct ocfs2_group_desc *) group_bh->b_data;
|
| 3242 | *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
|
| 3243 |
|
| 3244 | bail:
|
| 3245 | brelse(group_bh);
|
| 3246 |
|
| 3247 | if (status && !quiet)
|
| 3248 | mlog_errno(status);
|
| 3249 | return status;
|
| 3250 | }
|
| 3251 |
|
| 3252 | |
| 3253 | * Test if the bit representing this inode (blkno) is set in the
|
| 3254 | * suballocator.
|
| 3255 | *
|
| 3256 | * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
|
| 3257 | *
|
| 3258 | * In the event of failure, a negative value is returned and *res is
|
| 3259 | * meaningless.
|
| 3260 | *
|
| 3261 | * Callers must make sure to hold nfs_sync_lock to prevent
|
| 3262 | * ocfs2_delete_inode() on another node from accessing the same
|
| 3263 | * suballocator concurrently.
|
| 3264 |
|
| 3265 | int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
|
| 3266 | {
|
| 3267 | int status, quiet = 0;
|
| 3268 | u64 group_blkno = 0;
|
| 3269 | u16 suballoc_bit = 0, suballoc_slot = 0;
|
| 3270 | struct inode *inode_alloc_inode;
|
| 3271 | struct buffer_head *alloc_bh = NULL;
|
| 3272 |
|
| 3273 | trace_ocfs2_test_inode_bit((unsigned long long)blkno);
|
| 3274 |
|
| 3275 | status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
|
| 3276 | &group_blkno, &suballoc_bit);
|
| 3277 | if (status < 0) {
|
| 3278 | mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
|
| 3279 | goto bail;
|
| 3280 | }
|
| 3281 |
|
| 3282 | if (suballoc_slot == (u16)OCFS2_INVALID_SLOT)
|
| 3283 | inode_alloc_inode = ocfs2_get_system_file_inode(osb,
|
| 3284 | GLOBAL_INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
|
| 3285 | else
|
| 3286 | inode_alloc_inode = ocfs2_get_system_file_inode(osb,
|
| 3287 | INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
|
| 3288 | if (!inode_alloc_inode) {
|
| 3289 | |
| 3290 | * get the correct one.
|
| 3291 | status = -EINVAL;
|
| 3292 | mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
|
| 3293 | (u32)suballoc_slot);
|
| 3294 | goto bail;
|
| 3295 | }
|
| 3296 |
|
| 3297 | inode_lock(inode_alloc_inode);
|
| 3298 | status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
|
| 3299 | if (status < 0) {
|
| 3300 | inode_unlock(inode_alloc_inode);
|
| 3301 | iput(inode_alloc_inode);
|
| 3302 | mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
|
| 3303 | (u32)suballoc_slot, status);
|
| 3304 | goto bail;
|
| 3305 | }
|
| 3306 |
|
| 3307 | status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
|
| 3308 | group_blkno, blkno, suballoc_bit, res);
|
| 3309 | if (status < 0) {
|
| 3310 | if (status == -ESTALE)
|
| 3311 | quiet = 1;
|
| 3312 | else
|
| 3313 | mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
|
| 3314 | }
|
| 3315 |
|
| 3316 | ocfs2_inode_unlock(inode_alloc_inode, 0);
|
| 3317 | inode_unlock(inode_alloc_inode);
|
| 3318 |
|
| 3319 | iput(inode_alloc_inode);
|
| 3320 | brelse(alloc_bh);
|
| 3321 | bail:
|
| 3322 | if (status && !quiet)
|
| 3323 | mlog_errno(status);
|
| 3324 | return status;
|
| 3325 | }
|