1 // SPDX-License-Identifier: GPL-2.0+ 2 /* 3 * NILFS disk address translation. 4 * 5 * Copyright (C) 2006-2008 Nippon Telegraph and Telephone Corporation. 6 * 7 * Written by Koji Sato. 8 */ 9 10 #include <linux/types.h> 11 #include <linux/buffer_head.h> 12 #include <linux/string.h> 13 #include <linux/errno.h> 14 #include "nilfs.h" 15 #include "mdt.h" 16 #include "alloc.h" 17 #include "dat.h" 18 19 20 #define NILFS_CNO_MIN ((__u64)1) 21 #define NILFS_CNO_MAX (~(__u64)0) 22 23 /** 24 * struct nilfs_dat_info - on-memory private data of DAT file 25 * @mi: on-memory private data of metadata file 26 * @palloc_cache: persistent object allocator cache of DAT file 27 * @shadow: shadow map of DAT file 28 */ 29 struct nilfs_dat_info { 30 struct nilfs_mdt_info mi; 31 struct nilfs_palloc_cache palloc_cache; 32 struct nilfs_shadow_map shadow; 33 }; 34 35 static inline struct nilfs_dat_info *NILFS_DAT_I(struct inode *dat) 36 { 37 return (struct nilfs_dat_info *)NILFS_MDT(dat); 38 } 39 40 static int nilfs_dat_prepare_entry(struct inode *dat, 41 struct nilfs_palloc_req *req, int create) 42 { 43 return nilfs_palloc_get_entry_block(dat, req->pr_entry_nr, 44 create, &req->pr_entry_bh); 45 } 46 47 static void nilfs_dat_commit_entry(struct inode *dat, 48 struct nilfs_palloc_req *req) 49 { 50 mark_buffer_dirty(req->pr_entry_bh); 51 nilfs_mdt_mark_dirty(dat); 52 brelse(req->pr_entry_bh); 53 } 54 55 static void nilfs_dat_abort_entry(struct inode *dat, 56 struct nilfs_palloc_req *req) 57 { 58 brelse(req->pr_entry_bh); 59 } 60 61 int nilfs_dat_prepare_alloc(struct inode *dat, struct nilfs_palloc_req *req) 62 { 63 int ret; 64 65 ret = nilfs_palloc_prepare_alloc_entry(dat, req); 66 if (ret < 0) 67 return ret; 68 69 ret = nilfs_dat_prepare_entry(dat, req, 1); 70 if (ret < 0) 71 nilfs_palloc_abort_alloc_entry(dat, req); 72 73 return ret; 74 } 75 76 void nilfs_dat_commit_alloc(struct inode *dat, struct nilfs_palloc_req *req) 77 { 78 struct nilfs_dat_entry *entry; 79 void *kaddr; 80 81 kaddr = kmap_atomic(req->pr_entry_bh->b_page); 82 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr, 83 req->pr_entry_bh, kaddr); 84 entry->de_start = cpu_to_le64(NILFS_CNO_MIN); 85 entry->de_end = cpu_to_le64(NILFS_CNO_MAX); 86 entry->de_blocknr = cpu_to_le64(0); 87 kunmap_atomic(kaddr); 88 89 nilfs_palloc_commit_alloc_entry(dat, req); 90 nilfs_dat_commit_entry(dat, req); 91 } 92 93 void nilfs_dat_abort_alloc(struct inode *dat, struct nilfs_palloc_req *req) 94 { 95 nilfs_dat_abort_entry(dat, req); 96 nilfs_palloc_abort_alloc_entry(dat, req); 97 } 98 99 static void nilfs_dat_commit_free(struct inode *dat, 100 struct nilfs_palloc_req *req) 101 { 102 struct nilfs_dat_entry *entry; 103 void *kaddr; 104 105 kaddr = kmap_atomic(req->pr_entry_bh->b_page); 106 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr, 107 req->pr_entry_bh, kaddr); 108 entry->de_start = cpu_to_le64(NILFS_CNO_MIN); 109 entry->de_end = cpu_to_le64(NILFS_CNO_MIN); 110 entry->de_blocknr = cpu_to_le64(0); 111 kunmap_atomic(kaddr); 112 113 nilfs_dat_commit_entry(dat, req); 114 nilfs_palloc_commit_free_entry(dat, req); 115 } 116 117 int nilfs_dat_prepare_start(struct inode *dat, struct nilfs_palloc_req *req) 118 { 119 int ret; 120 121 ret = nilfs_dat_prepare_entry(dat, req, 0); 122 WARN_ON(ret == -ENOENT); 123 return ret; 124 } 125 126 void nilfs_dat_commit_start(struct inode *dat, struct nilfs_palloc_req *req, 127 sector_t blocknr) 128 { 129 struct nilfs_dat_entry *entry; 130 void *kaddr; 131 132 kaddr = kmap_atomic(req->pr_entry_bh->b_page); 133 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr, 134 req->pr_entry_bh, kaddr); 135 entry->de_start = cpu_to_le64(nilfs_mdt_cno(dat)); 136 entry->de_blocknr = cpu_to_le64(blocknr); 137 kunmap_atomic(kaddr); 138 139 nilfs_dat_commit_entry(dat, req); 140 } 141 142 int nilfs_dat_prepare_end(struct inode *dat, struct nilfs_palloc_req *req) 143 { 144 struct nilfs_dat_entry *entry; 145 sector_t blocknr; 146 void *kaddr; 147 int ret; 148 149 ret = nilfs_dat_prepare_entry(dat, req, 0); 150 if (ret < 0) { 151 WARN_ON(ret == -ENOENT); 152 return ret; 153 } 154 155 kaddr = kmap_atomic(req->pr_entry_bh->b_page); 156 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr, 157 req->pr_entry_bh, kaddr); 158 blocknr = le64_to_cpu(entry->de_blocknr); 159 kunmap_atomic(kaddr); 160 161 if (blocknr == 0) { 162 ret = nilfs_palloc_prepare_free_entry(dat, req); 163 if (ret < 0) { 164 nilfs_dat_abort_entry(dat, req); 165 return ret; 166 } 167 } 168 169 return 0; 170 } 171 172 void nilfs_dat_commit_end(struct inode *dat, struct nilfs_palloc_req *req, 173 int dead) 174 { 175 struct nilfs_dat_entry *entry; 176 __u64 start, end; 177 sector_t blocknr; 178 void *kaddr; 179 180 kaddr = kmap_atomic(req->pr_entry_bh->b_page); 181 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr, 182 req->pr_entry_bh, kaddr); 183 end = start = le64_to_cpu(entry->de_start); 184 if (!dead) { 185 end = nilfs_mdt_cno(dat); 186 WARN_ON(start > end); 187 } 188 entry->de_end = cpu_to_le64(end); 189 blocknr = le64_to_cpu(entry->de_blocknr); 190 kunmap_atomic(kaddr); 191 192 if (blocknr == 0) 193 nilfs_dat_commit_free(dat, req); 194 else 195 nilfs_dat_commit_entry(dat, req); 196 } 197 198 void nilfs_dat_abort_end(struct inode *dat, struct nilfs_palloc_req *req) 199 { 200 struct nilfs_dat_entry *entry; 201 __u64 start; 202 sector_t blocknr; 203 void *kaddr; 204 205 kaddr = kmap_atomic(req->pr_entry_bh->b_page); 206 entry = nilfs_palloc_block_get_entry(dat, req->pr_entry_nr, 207 req->pr_entry_bh, kaddr); 208 start = le64_to_cpu(entry->de_start); 209 blocknr = le64_to_cpu(entry->de_blocknr); 210 kunmap_atomic(kaddr); 211 212 if (start == nilfs_mdt_cno(dat) && blocknr == 0) 213 nilfs_palloc_abort_free_entry(dat, req); 214 nilfs_dat_abort_entry(dat, req); 215 } 216 217 int nilfs_dat_prepare_update(struct inode *dat, 218 struct nilfs_palloc_req *oldreq, 219 struct nilfs_palloc_req *newreq) 220 { 221 int ret; 222 223 ret = nilfs_dat_prepare_end(dat, oldreq); 224 if (!ret) { 225 ret = nilfs_dat_prepare_alloc(dat, newreq); 226 if (ret < 0) 227 nilfs_dat_abort_end(dat, oldreq); 228 } 229 return ret; 230 } 231 232 void nilfs_dat_commit_update(struct inode *dat, 233 struct nilfs_palloc_req *oldreq, 234 struct nilfs_palloc_req *newreq, int dead) 235 { 236 nilfs_dat_commit_end(dat, oldreq, dead); 237 nilfs_dat_commit_alloc(dat, newreq); 238 } 239 240 void nilfs_dat_abort_update(struct inode *dat, 241 struct nilfs_palloc_req *oldreq, 242 struct nilfs_palloc_req *newreq) 243 { 244 nilfs_dat_abort_end(dat, oldreq); 245 nilfs_dat_abort_alloc(dat, newreq); 246 } 247 248 /** 249 * nilfs_dat_mark_dirty - 250 * @dat: DAT file inode 251 * @vblocknr: virtual block number 252 * 253 * Description: 254 * 255 * Return Value: On success, 0 is returned. On error, one of the following 256 * negative error codes is returned. 257 * 258 * %-EIO - I/O error. 259 * 260 * %-ENOMEM - Insufficient amount of memory available. 261 */ 262 int nilfs_dat_mark_dirty(struct inode *dat, __u64 vblocknr) 263 { 264 struct nilfs_palloc_req req; 265 int ret; 266 267 req.pr_entry_nr = vblocknr; 268 ret = nilfs_dat_prepare_entry(dat, &req, 0); 269 if (ret == 0) 270 nilfs_dat_commit_entry(dat, &req); 271 return ret; 272 } 273 274 /** 275 * nilfs_dat_freev - free virtual block numbers 276 * @dat: DAT file inode 277 * @vblocknrs: array of virtual block numbers 278 * @nitems: number of virtual block numbers 279 * 280 * Description: nilfs_dat_freev() frees the virtual block numbers specified by 281 * @vblocknrs and @nitems. 282 * 283 * Return Value: On success, 0 is returned. On error, one of the following 284 * negative error codes is returned. 285 * 286 * %-EIO - I/O error. 287 * 288 * %-ENOMEM - Insufficient amount of memory available. 289 * 290 * %-ENOENT - The virtual block number have not been allocated. 291 */ 292 int nilfs_dat_freev(struct inode *dat, __u64 *vblocknrs, size_t nitems) 293 { 294 return nilfs_palloc_freev(dat, vblocknrs, nitems); 295 } 296 297 /** 298 * nilfs_dat_move - change a block number 299 * @dat: DAT file inode 300 * @vblocknr: virtual block number 301 * @blocknr: block number 302 * 303 * Description: nilfs_dat_move() changes the block number associated with 304 * @vblocknr to @blocknr. 305 * 306 * Return Value: On success, 0 is returned. On error, one of the following 307 * negative error codes is returned. 308 * 309 * %-EIO - I/O error. 310 * 311 * %-ENOMEM - Insufficient amount of memory available. 312 */ 313 int nilfs_dat_move(struct inode *dat, __u64 vblocknr, sector_t blocknr) 314 { 315 struct buffer_head *entry_bh; 316 struct nilfs_dat_entry *entry; 317 void *kaddr; 318 int ret; 319 320 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh); 321 if (ret < 0) 322 return ret; 323 324 /* 325 * The given disk block number (blocknr) is not yet written to 326 * the device at this point. 327 * 328 * To prevent nilfs_dat_translate() from returning the 329 * uncommitted block number, this makes a copy of the entry 330 * buffer and redirects nilfs_dat_translate() to the copy. 331 */ 332 if (!buffer_nilfs_redirected(entry_bh)) { 333 ret = nilfs_mdt_freeze_buffer(dat, entry_bh); 334 if (ret) { 335 brelse(entry_bh); 336 return ret; 337 } 338 } 339 340 kaddr = kmap_atomic(entry_bh->b_page); 341 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr); 342 if (unlikely(entry->de_blocknr == cpu_to_le64(0))) { 343 nilfs_crit(dat->i_sb, 344 "%s: invalid vblocknr = %llu, [%llu, %llu)", 345 __func__, (unsigned long long)vblocknr, 346 (unsigned long long)le64_to_cpu(entry->de_start), 347 (unsigned long long)le64_to_cpu(entry->de_end)); 348 kunmap_atomic(kaddr); 349 brelse(entry_bh); 350 return -EINVAL; 351 } 352 WARN_ON(blocknr == 0); 353 entry->de_blocknr = cpu_to_le64(blocknr); 354 kunmap_atomic(kaddr); 355 356 mark_buffer_dirty(entry_bh); 357 nilfs_mdt_mark_dirty(dat); 358 359 brelse(entry_bh); 360 361 return 0; 362 } 363 364 /** 365 * nilfs_dat_translate - translate a virtual block number to a block number 366 * @dat: DAT file inode 367 * @vblocknr: virtual block number 368 * @blocknrp: pointer to a block number 369 * 370 * Description: nilfs_dat_translate() maps the virtual block number @vblocknr 371 * to the corresponding block number. 372 * 373 * Return Value: On success, 0 is returned and the block number associated 374 * with @vblocknr is stored in the place pointed by @blocknrp. On error, one 375 * of the following negative error codes is returned. 376 * 377 * %-EIO - I/O error. 378 * 379 * %-ENOMEM - Insufficient amount of memory available. 380 * 381 * %-ENOENT - A block number associated with @vblocknr does not exist. 382 */ 383 int nilfs_dat_translate(struct inode *dat, __u64 vblocknr, sector_t *blocknrp) 384 { 385 struct buffer_head *entry_bh, *bh; 386 struct nilfs_dat_entry *entry; 387 sector_t blocknr; 388 void *kaddr; 389 int ret; 390 391 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh); 392 if (ret < 0) 393 return ret; 394 395 if (!nilfs_doing_gc() && buffer_nilfs_redirected(entry_bh)) { 396 bh = nilfs_mdt_get_frozen_buffer(dat, entry_bh); 397 if (bh) { 398 WARN_ON(!buffer_uptodate(bh)); 399 brelse(entry_bh); 400 entry_bh = bh; 401 } 402 } 403 404 kaddr = kmap_atomic(entry_bh->b_page); 405 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr); 406 blocknr = le64_to_cpu(entry->de_blocknr); 407 if (blocknr == 0) { 408 ret = -ENOENT; 409 goto out; 410 } 411 *blocknrp = blocknr; 412 413 out: 414 kunmap_atomic(kaddr); 415 brelse(entry_bh); 416 return ret; 417 } 418 419 ssize_t nilfs_dat_get_vinfo(struct inode *dat, void *buf, unsigned int visz, 420 size_t nvi) 421 { 422 struct buffer_head *entry_bh; 423 struct nilfs_dat_entry *entry; 424 struct nilfs_vinfo *vinfo = buf; 425 __u64 first, last; 426 void *kaddr; 427 unsigned long entries_per_block = NILFS_MDT(dat)->mi_entries_per_block; 428 int i, j, n, ret; 429 430 for (i = 0; i < nvi; i += n) { 431 ret = nilfs_palloc_get_entry_block(dat, vinfo->vi_vblocknr, 432 0, &entry_bh); 433 if (ret < 0) 434 return ret; 435 kaddr = kmap_atomic(entry_bh->b_page); 436 /* last virtual block number in this block */ 437 first = vinfo->vi_vblocknr; 438 do_div(first, entries_per_block); 439 first *= entries_per_block; 440 last = first + entries_per_block - 1; 441 for (j = i, n = 0; 442 j < nvi && vinfo->vi_vblocknr >= first && 443 vinfo->vi_vblocknr <= last; 444 j++, n++, vinfo = (void *)vinfo + visz) { 445 entry = nilfs_palloc_block_get_entry( 446 dat, vinfo->vi_vblocknr, entry_bh, kaddr); 447 vinfo->vi_start = le64_to_cpu(entry->de_start); 448 vinfo->vi_end = le64_to_cpu(entry->de_end); 449 vinfo->vi_blocknr = le64_to_cpu(entry->de_blocknr); 450 } 451 kunmap_atomic(kaddr); 452 brelse(entry_bh); 453 } 454 455 return nvi; 456 } 457 458 /** 459 * nilfs_dat_read - read or get dat inode 460 * @sb: super block instance 461 * @entry_size: size of a dat entry 462 * @raw_inode: on-disk dat inode 463 * @inodep: buffer to store the inode 464 */ 465 int nilfs_dat_read(struct super_block *sb, size_t entry_size, 466 struct nilfs_inode *raw_inode, struct inode **inodep) 467 { 468 static struct lock_class_key dat_lock_key; 469 struct inode *dat; 470 struct nilfs_dat_info *di; 471 int err; 472 473 if (entry_size > sb->s_blocksize) { 474 nilfs_err(sb, "too large DAT entry size: %zu bytes", 475 entry_size); 476 return -EINVAL; 477 } else if (entry_size < NILFS_MIN_DAT_ENTRY_SIZE) { 478 nilfs_err(sb, "too small DAT entry size: %zu bytes", 479 entry_size); 480 return -EINVAL; 481 } 482 483 dat = nilfs_iget_locked(sb, NULL, NILFS_DAT_INO); 484 if (unlikely(!dat)) 485 return -ENOMEM; 486 if (!(dat->i_state & I_NEW)) 487 goto out; 488 489 err = nilfs_mdt_init(dat, NILFS_MDT_GFP, sizeof(*di)); 490 if (err) 491 goto failed; 492 493 err = nilfs_palloc_init_blockgroup(dat, entry_size); 494 if (err) 495 goto failed; 496 497 di = NILFS_DAT_I(dat); 498 lockdep_set_class(&di->mi.mi_sem, &dat_lock_key); 499 nilfs_palloc_setup_cache(dat, &di->palloc_cache); 500 err = nilfs_mdt_setup_shadow_map(dat, &di->shadow); 501 if (err) 502 goto failed; 503 504 err = nilfs_read_inode_common(dat, raw_inode); 505 if (err) 506 goto failed; 507 508 unlock_new_inode(dat); 509 out: 510 *inodep = dat; 511 return 0; 512 failed: 513 iget_failed(dat); 514 return err; 515 } 516