1 // SPDX-License-Identifier: GPL-2.0 2 /* Generic part */ 3 4 typedef struct { 5 block_t *p; 6 block_t key; 7 struct buffer_head *bh; 8 } Indirect; 9 10 static DEFINE_RWLOCK(pointers_lock); 11 12 static inline void add_chain(Indirect *p, struct buffer_head *bh, block_t *v) 13 { 14 p->key = *(p->p = v); 15 p->bh = bh; 16 } 17 18 static inline int verify_chain(Indirect *from, Indirect *to) 19 { 20 while (from <= to && from->key == *from->p) 21 from++; 22 return (from > to); 23 } 24 25 static inline block_t *block_end(struct buffer_head *bh) 26 { 27 return (block_t *)((char*)bh->b_data + bh->b_size); 28 } 29 30 static inline Indirect *get_branch(struct inode *inode, 31 int depth, 32 int *offsets, 33 Indirect chain[DEPTH], 34 int *err) 35 { 36 struct super_block *sb = inode->i_sb; 37 Indirect *p = chain; 38 struct buffer_head *bh; 39 40 *err = 0; 41 /* i_data is not going away, no lock needed */ 42 add_chain (chain, NULL, i_data(inode) + *offsets); 43 if (!p->key) 44 goto no_block; 45 while (--depth) { 46 bh = sb_bread(sb, block_to_cpu(p->key)); 47 if (!bh) 48 goto failure; 49 read_lock(&pointers_lock); 50 if (!verify_chain(chain, p)) 51 goto changed; 52 add_chain(++p, bh, (block_t *)bh->b_data + *++offsets); 53 read_unlock(&pointers_lock); 54 if (!p->key) 55 goto no_block; 56 } 57 return NULL; 58 59 changed: 60 read_unlock(&pointers_lock); 61 brelse(bh); 62 *err = -EAGAIN; 63 goto no_block; 64 failure: 65 *err = -EIO; 66 no_block: 67 return p; 68 } 69 70 static int alloc_branch(struct inode *inode, 71 int num, 72 int *offsets, 73 Indirect *branch) 74 { 75 int n = 0; 76 int i; 77 int parent = minix_new_block(inode); 78 int err = -ENOSPC; 79 80 branch[0].key = cpu_to_block(parent); 81 if (parent) for (n = 1; n < num; n++) { 82 struct buffer_head *bh; 83 /* Allocate the next block */ 84 int nr = minix_new_block(inode); 85 if (!nr) 86 break; 87 branch[n].key = cpu_to_block(nr); 88 bh = sb_getblk(inode->i_sb, parent); 89 if (!bh) { 90 minix_free_block(inode, nr); 91 err = -ENOMEM; 92 break; 93 } 94 lock_buffer(bh); 95 memset(bh->b_data, 0, bh->b_size); 96 branch[n].bh = bh; 97 branch[n].p = (block_t*) bh->b_data + offsets[n]; 98 *branch[n].p = branch[n].key; 99 set_buffer_uptodate(bh); 100 unlock_buffer(bh); 101 mark_buffer_dirty_inode(bh, inode); 102 parent = nr; 103 } 104 if (n == num) 105 return 0; 106 107 /* Allocation failed, free what we already allocated */ 108 for (i = 1; i < n; i++) 109 bforget(branch[i].bh); 110 for (i = 0; i < n; i++) 111 minix_free_block(inode, block_to_cpu(branch[i].key)); 112 return err; 113 } 114 115 static inline int splice_branch(struct inode *inode, 116 Indirect chain[DEPTH], 117 Indirect *where, 118 int num) 119 { 120 int i; 121 122 write_lock(&pointers_lock); 123 124 /* Verify that place we are splicing to is still there and vacant */ 125 if (!verify_chain(chain, where-1) || *where->p) 126 goto changed; 127 128 *where->p = where->key; 129 130 write_unlock(&pointers_lock); 131 132 /* We are done with atomic stuff, now do the rest of housekeeping */ 133 134 inode_set_ctime_current(inode); 135 136 /* had we spliced it onto indirect block? */ 137 if (where->bh) 138 mark_buffer_dirty_inode(where->bh, inode); 139 140 mark_inode_dirty(inode); 141 return 0; 142 143 changed: 144 write_unlock(&pointers_lock); 145 for (i = 1; i < num; i++) 146 bforget(where[i].bh); 147 for (i = 0; i < num; i++) 148 minix_free_block(inode, block_to_cpu(where[i].key)); 149 return -EAGAIN; 150 } 151 152 static int get_block(struct inode * inode, sector_t block, 153 struct buffer_head *bh, int create) 154 { 155 int err = -EIO; 156 int offsets[DEPTH]; 157 Indirect chain[DEPTH]; 158 Indirect *partial; 159 int left; 160 int depth = block_to_path(inode, block, offsets); 161 162 if (depth == 0) 163 goto out; 164 165 reread: 166 partial = get_branch(inode, depth, offsets, chain, &err); 167 168 /* Simplest case - block found, no allocation needed */ 169 if (!partial) { 170 got_it: 171 map_bh(bh, inode->i_sb, block_to_cpu(chain[depth-1].key)); 172 /* Clean up and exit */ 173 partial = chain+depth-1; /* the whole chain */ 174 goto cleanup; 175 } 176 177 /* Next simple case - plain lookup or failed read of indirect block */ 178 if (!create || err == -EIO) { 179 cleanup: 180 while (partial > chain) { 181 brelse(partial->bh); 182 partial--; 183 } 184 out: 185 return err; 186 } 187 188 /* 189 * Indirect block might be removed by truncate while we were 190 * reading it. Handling of that case (forget what we've got and 191 * reread) is taken out of the main path. 192 */ 193 if (err == -EAGAIN) 194 goto changed; 195 196 left = (chain + depth) - partial; 197 err = alloc_branch(inode, left, offsets+(partial-chain), partial); 198 if (err) 199 goto cleanup; 200 201 if (splice_branch(inode, chain, partial, left) < 0) 202 goto changed; 203 204 set_buffer_new(bh); 205 goto got_it; 206 207 changed: 208 while (partial > chain) { 209 brelse(partial->bh); 210 partial--; 211 } 212 goto reread; 213 } 214 215 static inline int all_zeroes(block_t *p, block_t *q) 216 { 217 while (p < q) 218 if (*p++) 219 return 0; 220 return 1; 221 } 222 223 static Indirect *find_shared(struct inode *inode, 224 int depth, 225 int offsets[DEPTH], 226 Indirect chain[DEPTH], 227 block_t *top) 228 { 229 Indirect *partial, *p; 230 int k, err; 231 232 *top = 0; 233 for (k = depth; k > 1 && !offsets[k-1]; k--) 234 ; 235 partial = get_branch(inode, k, offsets, chain, &err); 236 237 write_lock(&pointers_lock); 238 if (!partial) 239 partial = chain + k-1; 240 if (!partial->key && *partial->p) { 241 write_unlock(&pointers_lock); 242 goto no_top; 243 } 244 for (p=partial;p>chain && all_zeroes((block_t*)p->bh->b_data,p->p);p--) 245 ; 246 if (p == chain + k - 1 && p > chain) { 247 p->p--; 248 } else { 249 *top = *p->p; 250 *p->p = 0; 251 } 252 write_unlock(&pointers_lock); 253 254 while(partial > p) 255 { 256 brelse(partial->bh); 257 partial--; 258 } 259 no_top: 260 return partial; 261 } 262 263 static inline void free_data(struct inode *inode, block_t *p, block_t *q) 264 { 265 unsigned long nr; 266 267 for ( ; p < q ; p++) { 268 nr = block_to_cpu(*p); 269 if (nr) { 270 *p = 0; 271 minix_free_block(inode, nr); 272 } 273 } 274 } 275 276 static void free_branches(struct inode *inode, block_t *p, block_t *q, int depth) 277 { 278 struct buffer_head * bh; 279 unsigned long nr; 280 281 if (depth--) { 282 for ( ; p < q ; p++) { 283 nr = block_to_cpu(*p); 284 if (!nr) 285 continue; 286 *p = 0; 287 bh = sb_bread(inode->i_sb, nr); 288 if (!bh) 289 continue; 290 free_branches(inode, (block_t*)bh->b_data, 291 block_end(bh), depth); 292 bforget(bh); 293 minix_free_block(inode, nr); 294 mark_inode_dirty(inode); 295 } 296 } else 297 free_data(inode, p, q); 298 } 299 300 static inline void truncate (struct inode * inode) 301 { 302 struct super_block *sb = inode->i_sb; 303 block_t *idata = i_data(inode); 304 int offsets[DEPTH]; 305 Indirect chain[DEPTH]; 306 Indirect *partial; 307 block_t nr = 0; 308 int n; 309 int first_whole; 310 long iblock; 311 312 iblock = (inode->i_size + sb->s_blocksize -1) >> sb->s_blocksize_bits; 313 block_truncate_page(inode->i_mapping, inode->i_size, get_block); 314 315 n = block_to_path(inode, iblock, offsets); 316 if (!n) 317 return; 318 319 if (n == 1) { 320 free_data(inode, idata+offsets[0], idata + DIRECT); 321 first_whole = 0; 322 goto do_indirects; 323 } 324 325 first_whole = offsets[0] + 1 - DIRECT; 326 partial = find_shared(inode, n, offsets, chain, &nr); 327 if (nr) { 328 if (partial == chain) 329 mark_inode_dirty(inode); 330 else 331 mark_buffer_dirty_inode(partial->bh, inode); 332 free_branches(inode, &nr, &nr+1, (chain+n-1) - partial); 333 } 334 /* Clear the ends of indirect blocks on the shared branch */ 335 while (partial > chain) { 336 free_branches(inode, partial->p + 1, block_end(partial->bh), 337 (chain+n-1) - partial); 338 mark_buffer_dirty_inode(partial->bh, inode); 339 brelse (partial->bh); 340 partial--; 341 } 342 do_indirects: 343 /* Kill the remaining (whole) subtrees */ 344 while (first_whole < DEPTH-1) { 345 nr = idata[DIRECT+first_whole]; 346 if (nr) { 347 idata[DIRECT+first_whole] = 0; 348 mark_inode_dirty(inode); 349 free_branches(inode, &nr, &nr+1, first_whole+1); 350 } 351 first_whole++; 352 } 353 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 354 mark_inode_dirty(inode); 355 } 356 357 static inline unsigned nblocks(loff_t size, struct super_block *sb) 358 { 359 int k = sb->s_blocksize_bits - 10; 360 unsigned blocks, res, direct = DIRECT, i = DEPTH; 361 blocks = (size + sb->s_blocksize - 1) >> (BLOCK_SIZE_BITS + k); 362 res = blocks; 363 while (--i && blocks > direct) { 364 blocks -= direct; 365 blocks += sb->s_blocksize/sizeof(block_t) - 1; 366 blocks /= sb->s_blocksize/sizeof(block_t); 367 res += blocks; 368 direct = 1; 369 } 370 return res; 371 } 372
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.