1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * This file is part of UBIFS. 4 * 5 * Copyright (C) 2006-2008 Nokia Corporation. 6 * 7 * Authors: Adrian Hunter 8 * Artem Bityutskiy (Битюцкий Артём) 9 */ 10 11 /* 12 * This file implements commit-related functionality of the LEB properties 13 * subsystem. 14 */ 15 16 #include <linux/crc16.h> 17 #include <linux/slab.h> 18 #include <linux/random.h> 19 #include "ubifs.h" 20 21 static int dbg_populate_lsave(struct ubifs_info *c); 22 23 /** 24 * first_dirty_cnode - find first dirty cnode. 25 * @c: UBIFS file-system description object 26 * @nnode: nnode at which to start 27 * 28 * This function returns the first dirty cnode or %NULL if there is not one. 29 */ 30 static struct ubifs_cnode *first_dirty_cnode(const struct ubifs_info *c, struct ubifs_nnode *nnode) 31 { 32 ubifs_assert(c, nnode); 33 while (1) { 34 int i, cont = 0; 35 36 for (i = 0; i < UBIFS_LPT_FANOUT; i++) { 37 struct ubifs_cnode *cnode; 38 39 cnode = nnode->nbranch[i].cnode; 40 if (cnode && 41 test_bit(DIRTY_CNODE, &cnode->flags)) { 42 if (cnode->level == 0) 43 return cnode; 44 nnode = (struct ubifs_nnode *)cnode; 45 cont = 1; 46 break; 47 } 48 } 49 if (!cont) 50 return (struct ubifs_cnode *)nnode; 51 } 52 } 53 54 /** 55 * next_dirty_cnode - find next dirty cnode. 56 * @c: UBIFS file-system description object 57 * @cnode: cnode from which to begin searching 58 * 59 * This function returns the next dirty cnode or %NULL if there is not one. 60 */ 61 static struct ubifs_cnode *next_dirty_cnode(const struct ubifs_info *c, struct ubifs_cnode *cnode) 62 { 63 struct ubifs_nnode *nnode; 64 int i; 65 66 ubifs_assert(c, cnode); 67 nnode = cnode->parent; 68 if (!nnode) 69 return NULL; 70 for (i = cnode->iip + 1; i < UBIFS_LPT_FANOUT; i++) { 71 cnode = nnode->nbranch[i].cnode; 72 if (cnode && test_bit(DIRTY_CNODE, &cnode->flags)) { 73 if (cnode->level == 0) 74 return cnode; /* cnode is a pnode */ 75 /* cnode is a nnode */ 76 return first_dirty_cnode(c, (struct ubifs_nnode *)cnode); 77 } 78 } 79 return (struct ubifs_cnode *)nnode; 80 } 81 82 /** 83 * get_cnodes_to_commit - create list of dirty cnodes to commit. 84 * @c: UBIFS file-system description object 85 * 86 * This function returns the number of cnodes to commit. 87 */ 88 static int get_cnodes_to_commit(struct ubifs_info *c) 89 { 90 struct ubifs_cnode *cnode, *cnext; 91 int cnt = 0; 92 93 if (!c->nroot) 94 return 0; 95 96 if (!test_bit(DIRTY_CNODE, &c->nroot->flags)) 97 return 0; 98 99 c->lpt_cnext = first_dirty_cnode(c, c->nroot); 100 cnode = c->lpt_cnext; 101 if (!cnode) 102 return 0; 103 cnt += 1; 104 while (1) { 105 ubifs_assert(c, !test_bit(COW_CNODE, &cnode->flags)); 106 __set_bit(COW_CNODE, &cnode->flags); 107 cnext = next_dirty_cnode(c, cnode); 108 if (!cnext) { 109 cnode->cnext = c->lpt_cnext; 110 break; 111 } 112 cnode->cnext = cnext; 113 cnode = cnext; 114 cnt += 1; 115 } 116 dbg_cmt("committing %d cnodes", cnt); 117 dbg_lp("committing %d cnodes", cnt); 118 ubifs_assert(c, cnt == c->dirty_nn_cnt + c->dirty_pn_cnt); 119 return cnt; 120 } 121 122 /** 123 * upd_ltab - update LPT LEB properties. 124 * @c: UBIFS file-system description object 125 * @lnum: LEB number 126 * @free: amount of free space 127 * @dirty: amount of dirty space to add 128 */ 129 static void upd_ltab(struct ubifs_info *c, int lnum, int free, int dirty) 130 { 131 dbg_lp("LEB %d free %d dirty %d to %d +%d", 132 lnum, c->ltab[lnum - c->lpt_first].free, 133 c->ltab[lnum - c->lpt_first].dirty, free, dirty); 134 ubifs_assert(c, lnum >= c->lpt_first && lnum <= c->lpt_last); 135 c->ltab[lnum - c->lpt_first].free = free; 136 c->ltab[lnum - c->lpt_first].dirty += dirty; 137 } 138 139 /** 140 * alloc_lpt_leb - allocate an LPT LEB that is empty. 141 * @c: UBIFS file-system description object 142 * @lnum: LEB number is passed and returned here 143 * 144 * This function finds the next empty LEB in the ltab starting from @lnum. If a 145 * an empty LEB is found it is returned in @lnum and the function returns %0. 146 * Otherwise the function returns -ENOSPC. Note however, that LPT is designed 147 * never to run out of space. 148 */ 149 static int alloc_lpt_leb(struct ubifs_info *c, int *lnum) 150 { 151 int i, n; 152 153 n = *lnum - c->lpt_first + 1; 154 for (i = n; i < c->lpt_lebs; i++) { 155 if (c->ltab[i].tgc || c->ltab[i].cmt) 156 continue; 157 if (c->ltab[i].free == c->leb_size) { 158 c->ltab[i].cmt = 1; 159 *lnum = i + c->lpt_first; 160 return 0; 161 } 162 } 163 164 for (i = 0; i < n; i++) { 165 if (c->ltab[i].tgc || c->ltab[i].cmt) 166 continue; 167 if (c->ltab[i].free == c->leb_size) { 168 c->ltab[i].cmt = 1; 169 *lnum = i + c->lpt_first; 170 return 0; 171 } 172 } 173 return -ENOSPC; 174 } 175 176 /** 177 * layout_cnodes - layout cnodes for commit. 178 * @c: UBIFS file-system description object 179 * 180 * This function returns %0 on success and a negative error code on failure. 181 */ 182 static int layout_cnodes(struct ubifs_info *c) 183 { 184 int lnum, offs, len, alen, done_lsave, done_ltab, err; 185 struct ubifs_cnode *cnode; 186 187 err = dbg_chk_lpt_sz(c, 0, 0); 188 if (err) 189 return err; 190 cnode = c->lpt_cnext; 191 if (!cnode) 192 return 0; 193 lnum = c->nhead_lnum; 194 offs = c->nhead_offs; 195 /* Try to place lsave and ltab nicely */ 196 done_lsave = !c->big_lpt; 197 done_ltab = 0; 198 if (!done_lsave && offs + c->lsave_sz <= c->leb_size) { 199 done_lsave = 1; 200 c->lsave_lnum = lnum; 201 c->lsave_offs = offs; 202 offs += c->lsave_sz; 203 dbg_chk_lpt_sz(c, 1, c->lsave_sz); 204 } 205 206 if (offs + c->ltab_sz <= c->leb_size) { 207 done_ltab = 1; 208 c->ltab_lnum = lnum; 209 c->ltab_offs = offs; 210 offs += c->ltab_sz; 211 dbg_chk_lpt_sz(c, 1, c->ltab_sz); 212 } 213 214 do { 215 if (cnode->level) { 216 len = c->nnode_sz; 217 c->dirty_nn_cnt -= 1; 218 } else { 219 len = c->pnode_sz; 220 c->dirty_pn_cnt -= 1; 221 } 222 while (offs + len > c->leb_size) { 223 alen = ALIGN(offs, c->min_io_size); 224 upd_ltab(c, lnum, c->leb_size - alen, alen - offs); 225 dbg_chk_lpt_sz(c, 2, c->leb_size - offs); 226 err = alloc_lpt_leb(c, &lnum); 227 if (err) 228 goto no_space; 229 offs = 0; 230 ubifs_assert(c, lnum >= c->lpt_first && 231 lnum <= c->lpt_last); 232 /* Try to place lsave and ltab nicely */ 233 if (!done_lsave) { 234 done_lsave = 1; 235 c->lsave_lnum = lnum; 236 c->lsave_offs = offs; 237 offs += c->lsave_sz; 238 dbg_chk_lpt_sz(c, 1, c->lsave_sz); 239 continue; 240 } 241 if (!done_ltab) { 242 done_ltab = 1; 243 c->ltab_lnum = lnum; 244 c->ltab_offs = offs; 245 offs += c->ltab_sz; 246 dbg_chk_lpt_sz(c, 1, c->ltab_sz); 247 continue; 248 } 249 break; 250 } 251 if (cnode->parent) { 252 cnode->parent->nbranch[cnode->iip].lnum = lnum; 253 cnode->parent->nbranch[cnode->iip].offs = offs; 254 } else { 255 c->lpt_lnum = lnum; 256 c->lpt_offs = offs; 257 } 258 offs += len; 259 dbg_chk_lpt_sz(c, 1, len); 260 cnode = cnode->cnext; 261 } while (cnode && cnode != c->lpt_cnext); 262 263 /* Make sure to place LPT's save table */ 264 if (!done_lsave) { 265 if (offs + c->lsave_sz > c->leb_size) { 266 alen = ALIGN(offs, c->min_io_size); 267 upd_ltab(c, lnum, c->leb_size - alen, alen - offs); 268 dbg_chk_lpt_sz(c, 2, c->leb_size - offs); 269 err = alloc_lpt_leb(c, &lnum); 270 if (err) 271 goto no_space; 272 offs = 0; 273 ubifs_assert(c, lnum >= c->lpt_first && 274 lnum <= c->lpt_last); 275 } 276 done_lsave = 1; 277 c->lsave_lnum = lnum; 278 c->lsave_offs = offs; 279 offs += c->lsave_sz; 280 dbg_chk_lpt_sz(c, 1, c->lsave_sz); 281 } 282 283 /* Make sure to place LPT's own lprops table */ 284 if (!done_ltab) { 285 if (offs + c->ltab_sz > c->leb_size) { 286 alen = ALIGN(offs, c->min_io_size); 287 upd_ltab(c, lnum, c->leb_size - alen, alen - offs); 288 dbg_chk_lpt_sz(c, 2, c->leb_size - offs); 289 err = alloc_lpt_leb(c, &lnum); 290 if (err) 291 goto no_space; 292 offs = 0; 293 ubifs_assert(c, lnum >= c->lpt_first && 294 lnum <= c->lpt_last); 295 } 296 c->ltab_lnum = lnum; 297 c->ltab_offs = offs; 298 offs += c->ltab_sz; 299 dbg_chk_lpt_sz(c, 1, c->ltab_sz); 300 } 301 302 alen = ALIGN(offs, c->min_io_size); 303 upd_ltab(c, lnum, c->leb_size - alen, alen - offs); 304 dbg_chk_lpt_sz(c, 4, alen - offs); 305 err = dbg_chk_lpt_sz(c, 3, alen); 306 if (err) 307 return err; 308 return 0; 309 310 no_space: 311 ubifs_err(c, "LPT out of space at LEB %d:%d needing %d, done_ltab %d, done_lsave %d", 312 lnum, offs, len, done_ltab, done_lsave); 313 ubifs_dump_lpt_info(c); 314 ubifs_dump_lpt_lebs(c); 315 dump_stack(); 316 return err; 317 } 318 319 /** 320 * realloc_lpt_leb - allocate an LPT LEB that is empty. 321 * @c: UBIFS file-system description object 322 * @lnum: LEB number is passed and returned here 323 * 324 * This function duplicates exactly the results of the function alloc_lpt_leb. 325 * It is used during end commit to reallocate the same LEB numbers that were 326 * allocated by alloc_lpt_leb during start commit. 327 * 328 * This function finds the next LEB that was allocated by the alloc_lpt_leb 329 * function starting from @lnum. If a LEB is found it is returned in @lnum and 330 * the function returns %0. Otherwise the function returns -ENOSPC. 331 * Note however, that LPT is designed never to run out of space. 332 */ 333 static int realloc_lpt_leb(struct ubifs_info *c, int *lnum) 334 { 335 int i, n; 336 337 n = *lnum - c->lpt_first + 1; 338 for (i = n; i < c->lpt_lebs; i++) 339 if (c->ltab[i].cmt) { 340 c->ltab[i].cmt = 0; 341 *lnum = i + c->lpt_first; 342 return 0; 343 } 344 345 for (i = 0; i < n; i++) 346 if (c->ltab[i].cmt) { 347 c->ltab[i].cmt = 0; 348 *lnum = i + c->lpt_first; 349 return 0; 350 } 351 return -ENOSPC; 352 } 353 354 /** 355 * write_cnodes - write cnodes for commit. 356 * @c: UBIFS file-system description object 357 * 358 * This function returns %0 on success and a negative error code on failure. 359 */ 360 static int write_cnodes(struct ubifs_info *c) 361 { 362 int lnum, offs, len, from, err, wlen, alen, done_ltab, done_lsave; 363 struct ubifs_cnode *cnode; 364 void *buf = c->lpt_buf; 365 366 cnode = c->lpt_cnext; 367 if (!cnode) 368 return 0; 369 lnum = c->nhead_lnum; 370 offs = c->nhead_offs; 371 from = offs; 372 /* Ensure empty LEB is unmapped */ 373 if (offs == 0) { 374 err = ubifs_leb_unmap(c, lnum); 375 if (err) 376 return err; 377 } 378 /* Try to place lsave and ltab nicely */ 379 done_lsave = !c->big_lpt; 380 done_ltab = 0; 381 if (!done_lsave && offs + c->lsave_sz <= c->leb_size) { 382 done_lsave = 1; 383 ubifs_pack_lsave(c, buf + offs, c->lsave); 384 offs += c->lsave_sz; 385 dbg_chk_lpt_sz(c, 1, c->lsave_sz); 386 } 387 388 if (offs + c->ltab_sz <= c->leb_size) { 389 done_ltab = 1; 390 ubifs_pack_ltab(c, buf + offs, c->ltab_cmt); 391 offs += c->ltab_sz; 392 dbg_chk_lpt_sz(c, 1, c->ltab_sz); 393 } 394 395 /* Loop for each cnode */ 396 do { 397 if (cnode->level) 398 len = c->nnode_sz; 399 else 400 len = c->pnode_sz; 401 while (offs + len > c->leb_size) { 402 wlen = offs - from; 403 if (wlen) { 404 alen = ALIGN(wlen, c->min_io_size); 405 memset(buf + offs, 0xff, alen - wlen); 406 err = ubifs_leb_write(c, lnum, buf + from, from, 407 alen); 408 if (err) 409 return err; 410 } 411 dbg_chk_lpt_sz(c, 2, c->leb_size - offs); 412 err = realloc_lpt_leb(c, &lnum); 413 if (err) 414 goto no_space; 415 offs = from = 0; 416 ubifs_assert(c, lnum >= c->lpt_first && 417 lnum <= c->lpt_last); 418 err = ubifs_leb_unmap(c, lnum); 419 if (err) 420 return err; 421 /* Try to place lsave and ltab nicely */ 422 if (!done_lsave) { 423 done_lsave = 1; 424 ubifs_pack_lsave(c, buf + offs, c->lsave); 425 offs += c->lsave_sz; 426 dbg_chk_lpt_sz(c, 1, c->lsave_sz); 427 continue; 428 } 429 if (!done_ltab) { 430 done_ltab = 1; 431 ubifs_pack_ltab(c, buf + offs, c->ltab_cmt); 432 offs += c->ltab_sz; 433 dbg_chk_lpt_sz(c, 1, c->ltab_sz); 434 continue; 435 } 436 break; 437 } 438 if (cnode->level) 439 ubifs_pack_nnode(c, buf + offs, 440 (struct ubifs_nnode *)cnode); 441 else 442 ubifs_pack_pnode(c, buf + offs, 443 (struct ubifs_pnode *)cnode); 444 /* 445 * The reason for the barriers is the same as in case of TNC. 446 * See comment in 'write_index()'. 'dirty_cow_nnode()' and 447 * 'dirty_cow_pnode()' are the functions for which this is 448 * important. 449 */ 450 clear_bit(DIRTY_CNODE, &cnode->flags); 451 smp_mb__before_atomic(); 452 clear_bit(COW_CNODE, &cnode->flags); 453 smp_mb__after_atomic(); 454 offs += len; 455 dbg_chk_lpt_sz(c, 1, len); 456 cnode = cnode->cnext; 457 } while (cnode && cnode != c->lpt_cnext); 458 459 /* Make sure to place LPT's save table */ 460 if (!done_lsave) { 461 if (offs + c->lsave_sz > c->leb_size) { 462 wlen = offs - from; 463 alen = ALIGN(wlen, c->min_io_size); 464 memset(buf + offs, 0xff, alen - wlen); 465 err = ubifs_leb_write(c, lnum, buf + from, from, alen); 466 if (err) 467 return err; 468 dbg_chk_lpt_sz(c, 2, c->leb_size - offs); 469 err = realloc_lpt_leb(c, &lnum); 470 if (err) 471 goto no_space; 472 offs = from = 0; 473 ubifs_assert(c, lnum >= c->lpt_first && 474 lnum <= c->lpt_last); 475 err = ubifs_leb_unmap(c, lnum); 476 if (err) 477 return err; 478 } 479 done_lsave = 1; 480 ubifs_pack_lsave(c, buf + offs, c->lsave); 481 offs += c->lsave_sz; 482 dbg_chk_lpt_sz(c, 1, c->lsave_sz); 483 } 484 485 /* Make sure to place LPT's own lprops table */ 486 if (!done_ltab) { 487 if (offs + c->ltab_sz > c->leb_size) { 488 wlen = offs - from; 489 alen = ALIGN(wlen, c->min_io_size); 490 memset(buf + offs, 0xff, alen - wlen); 491 err = ubifs_leb_write(c, lnum, buf + from, from, alen); 492 if (err) 493 return err; 494 dbg_chk_lpt_sz(c, 2, c->leb_size - offs); 495 err = realloc_lpt_leb(c, &lnum); 496 if (err) 497 goto no_space; 498 offs = from = 0; 499 ubifs_assert(c, lnum >= c->lpt_first && 500 lnum <= c->lpt_last); 501 err = ubifs_leb_unmap(c, lnum); 502 if (err) 503 return err; 504 } 505 ubifs_pack_ltab(c, buf + offs, c->ltab_cmt); 506 offs += c->ltab_sz; 507 dbg_chk_lpt_sz(c, 1, c->ltab_sz); 508 } 509 510 /* Write remaining data in buffer */ 511 wlen = offs - from; 512 alen = ALIGN(wlen, c->min_io_size); 513 memset(buf + offs, 0xff, alen - wlen); 514 err = ubifs_leb_write(c, lnum, buf + from, from, alen); 515 if (err) 516 return err; 517 518 dbg_chk_lpt_sz(c, 4, alen - wlen); 519 err = dbg_chk_lpt_sz(c, 3, ALIGN(offs, c->min_io_size)); 520 if (err) 521 return err; 522 523 c->nhead_lnum = lnum; 524 c->nhead_offs = ALIGN(offs, c->min_io_size); 525 526 dbg_lp("LPT root is at %d:%d", c->lpt_lnum, c->lpt_offs); 527 dbg_lp("LPT head is at %d:%d", c->nhead_lnum, c->nhead_offs); 528 dbg_lp("LPT ltab is at %d:%d", c->ltab_lnum, c->ltab_offs); 529 if (c->big_lpt) 530 dbg_lp("LPT lsave is at %d:%d", c->lsave_lnum, c->lsave_offs); 531 532 return 0; 533 534 no_space: 535 ubifs_err(c, "LPT out of space mismatch at LEB %d:%d needing %d, done_ltab %d, done_lsave %d", 536 lnum, offs, len, done_ltab, done_lsave); 537 ubifs_dump_lpt_info(c); 538 ubifs_dump_lpt_lebs(c); 539 dump_stack(); 540 return err; 541 } 542 543 /** 544 * next_pnode_to_dirty - find next pnode to dirty. 545 * @c: UBIFS file-system description object 546 * @pnode: pnode 547 * 548 * This function returns the next pnode to dirty or %NULL if there are no more 549 * pnodes. Note that pnodes that have never been written (lnum == 0) are 550 * skipped. 551 */ 552 static struct ubifs_pnode *next_pnode_to_dirty(struct ubifs_info *c, 553 struct ubifs_pnode *pnode) 554 { 555 struct ubifs_nnode *nnode; 556 int iip; 557 558 /* Try to go right */ 559 nnode = pnode->parent; 560 for (iip = pnode->iip + 1; iip < UBIFS_LPT_FANOUT; iip++) { 561 if (nnode->nbranch[iip].lnum) 562 return ubifs_get_pnode(c, nnode, iip); 563 } 564 565 /* Go up while can't go right */ 566 do { 567 iip = nnode->iip + 1; 568 nnode = nnode->parent; 569 if (!nnode) 570 return NULL; 571 for (; iip < UBIFS_LPT_FANOUT; iip++) { 572 if (nnode->nbranch[iip].lnum) 573 break; 574 } 575 } while (iip >= UBIFS_LPT_FANOUT); 576 577 /* Go right */ 578 nnode = ubifs_get_nnode(c, nnode, iip); 579 if (IS_ERR(nnode)) 580 return (void *)nnode; 581 582 /* Go down to level 1 */ 583 while (nnode->level > 1) { 584 for (iip = 0; iip < UBIFS_LPT_FANOUT; iip++) { 585 if (nnode->nbranch[iip].lnum) 586 break; 587 } 588 if (iip >= UBIFS_LPT_FANOUT) { 589 /* 590 * Should not happen, but we need to keep going 591 * if it does. 592 */ 593 iip = 0; 594 } 595 nnode = ubifs_get_nnode(c, nnode, iip); 596 if (IS_ERR(nnode)) 597 return (void *)nnode; 598 } 599 600 for (iip = 0; iip < UBIFS_LPT_FANOUT; iip++) 601 if (nnode->nbranch[iip].lnum) 602 break; 603 if (iip >= UBIFS_LPT_FANOUT) 604 /* Should not happen, but we need to keep going if it does */ 605 iip = 0; 606 return ubifs_get_pnode(c, nnode, iip); 607 } 608 609 /** 610 * add_pnode_dirt - add dirty space to LPT LEB properties. 611 * @c: UBIFS file-system description object 612 * @pnode: pnode for which to add dirt 613 */ 614 static void add_pnode_dirt(struct ubifs_info *c, struct ubifs_pnode *pnode) 615 { 616 ubifs_add_lpt_dirt(c, pnode->parent->nbranch[pnode->iip].lnum, 617 c->pnode_sz); 618 } 619 620 /** 621 * do_make_pnode_dirty - mark a pnode dirty. 622 * @c: UBIFS file-system description object 623 * @pnode: pnode to mark dirty 624 */ 625 static void do_make_pnode_dirty(struct ubifs_info *c, struct ubifs_pnode *pnode) 626 { 627 /* Assumes cnext list is empty i.e. not called during commit */ 628 if (!test_and_set_bit(DIRTY_CNODE, &pnode->flags)) { 629 struct ubifs_nnode *nnode; 630 631 c->dirty_pn_cnt += 1; 632 add_pnode_dirt(c, pnode); 633 /* Mark parent and ancestors dirty too */ 634 nnode = pnode->parent; 635 while (nnode) { 636 if (!test_and_set_bit(DIRTY_CNODE, &nnode->flags)) { 637 c->dirty_nn_cnt += 1; 638 ubifs_add_nnode_dirt(c, nnode); 639 nnode = nnode->parent; 640 } else 641 break; 642 } 643 } 644 } 645 646 /** 647 * make_tree_dirty - mark the entire LEB properties tree dirty. 648 * @c: UBIFS file-system description object 649 * 650 * This function is used by the "small" LPT model to cause the entire LEB 651 * properties tree to be written. The "small" LPT model does not use LPT 652 * garbage collection because it is more efficient to write the entire tree 653 * (because it is small). 654 * 655 * This function returns %0 on success and a negative error code on failure. 656 */ 657 static int make_tree_dirty(struct ubifs_info *c) 658 { 659 struct ubifs_pnode *pnode; 660 661 pnode = ubifs_pnode_lookup(c, 0); 662 if (IS_ERR(pnode)) 663 return PTR_ERR(pnode); 664 665 while (pnode) { 666 do_make_pnode_dirty(c, pnode); 667 pnode = next_pnode_to_dirty(c, pnode); 668 if (IS_ERR(pnode)) 669 return PTR_ERR(pnode); 670 } 671 return 0; 672 } 673 674 /** 675 * need_write_all - determine if the LPT area is running out of free space. 676 * @c: UBIFS file-system description object 677 * 678 * This function returns %1 if the LPT area is running out of free space and %0 679 * if it is not. 680 */ 681 static int need_write_all(struct ubifs_info *c) 682 { 683 long long free = 0; 684 int i; 685 686 for (i = 0; i < c->lpt_lebs; i++) { 687 if (i + c->lpt_first == c->nhead_lnum) 688 free += c->leb_size - c->nhead_offs; 689 else if (c->ltab[i].free == c->leb_size) 690 free += c->leb_size; 691 else if (c->ltab[i].free + c->ltab[i].dirty == c->leb_size) 692 free += c->leb_size; 693 } 694 /* Less than twice the size left */ 695 if (free <= c->lpt_sz * 2) 696 return 1; 697 return 0; 698 } 699 700 /** 701 * lpt_tgc_start - start trivial garbage collection of LPT LEBs. 702 * @c: UBIFS file-system description object 703 * 704 * LPT trivial garbage collection is where a LPT LEB contains only dirty and 705 * free space and so may be reused as soon as the next commit is completed. 706 * This function is called during start commit to mark LPT LEBs for trivial GC. 707 */ 708 static void lpt_tgc_start(struct ubifs_info *c) 709 { 710 int i; 711 712 for (i = 0; i < c->lpt_lebs; i++) { 713 if (i + c->lpt_first == c->nhead_lnum) 714 continue; 715 if (c->ltab[i].dirty > 0 && 716 c->ltab[i].free + c->ltab[i].dirty == c->leb_size) { 717 c->ltab[i].tgc = 1; 718 c->ltab[i].free = c->leb_size; 719 c->ltab[i].dirty = 0; 720 dbg_lp("LEB %d", i + c->lpt_first); 721 } 722 } 723 } 724 725 /** 726 * lpt_tgc_end - end trivial garbage collection of LPT LEBs. 727 * @c: UBIFS file-system description object 728 * 729 * LPT trivial garbage collection is where a LPT LEB contains only dirty and 730 * free space and so may be reused as soon as the next commit is completed. 731 * This function is called after the commit is completed (master node has been 732 * written) and un-maps LPT LEBs that were marked for trivial GC. 733 */ 734 static int lpt_tgc_end(struct ubifs_info *c) 735 { 736 int i, err; 737 738 for (i = 0; i < c->lpt_lebs; i++) 739 if (c->ltab[i].tgc) { 740 err = ubifs_leb_unmap(c, i + c->lpt_first); 741 if (err) 742 return err; 743 c->ltab[i].tgc = 0; 744 dbg_lp("LEB %d", i + c->lpt_first); 745 } 746 return 0; 747 } 748 749 /** 750 * populate_lsave - fill the lsave array with important LEB numbers. 751 * @c: the UBIFS file-system description object 752 * 753 * This function is only called for the "big" model. It records a small number 754 * of LEB numbers of important LEBs. Important LEBs are ones that are (from 755 * most important to least important): empty, freeable, freeable index, dirty 756 * index, dirty or free. Upon mount, we read this list of LEB numbers and bring 757 * their pnodes into memory. That will stop us from having to scan the LPT 758 * straight away. For the "small" model we assume that scanning the LPT is no 759 * big deal. 760 */ 761 static void populate_lsave(struct ubifs_info *c) 762 { 763 struct ubifs_lprops *lprops; 764 struct ubifs_lpt_heap *heap; 765 int i, cnt = 0; 766 767 ubifs_assert(c, c->big_lpt); 768 if (!(c->lpt_drty_flgs & LSAVE_DIRTY)) { 769 c->lpt_drty_flgs |= LSAVE_DIRTY; 770 ubifs_add_lpt_dirt(c, c->lsave_lnum, c->lsave_sz); 771 } 772 773 if (dbg_populate_lsave(c)) 774 return; 775 776 list_for_each_entry(lprops, &c->empty_list, list) { 777 c->lsave[cnt++] = lprops->lnum; 778 if (cnt >= c->lsave_cnt) 779 return; 780 } 781 list_for_each_entry(lprops, &c->freeable_list, list) { 782 c->lsave[cnt++] = lprops->lnum; 783 if (cnt >= c->lsave_cnt) 784 return; 785 } 786 list_for_each_entry(lprops, &c->frdi_idx_list, list) { 787 c->lsave[cnt++] = lprops->lnum; 788 if (cnt >= c->lsave_cnt) 789 return; 790 } 791 heap = &c->lpt_heap[LPROPS_DIRTY_IDX - 1]; 792 for (i = 0; i < heap->cnt; i++) { 793 c->lsave[cnt++] = heap->arr[i]->lnum; 794 if (cnt >= c->lsave_cnt) 795 return; 796 } 797 heap = &c->lpt_heap[LPROPS_DIRTY - 1]; 798 for (i = 0; i < heap->cnt; i++) { 799 c->lsave[cnt++] = heap->arr[i]->lnum; 800 if (cnt >= c->lsave_cnt) 801 return; 802 } 803 heap = &c->lpt_heap[LPROPS_FREE - 1]; 804 for (i = 0; i < heap->cnt; i++) { 805 c->lsave[cnt++] = heap->arr[i]->lnum; 806 if (cnt >= c->lsave_cnt) 807 return; 808 } 809 /* Fill it up completely */ 810 while (cnt < c->lsave_cnt) 811 c->lsave[cnt++] = c->main_first; 812 } 813 814 /** 815 * nnode_lookup - lookup a nnode in the LPT. 816 * @c: UBIFS file-system description object 817 * @i: nnode number 818 * 819 * This function returns a pointer to the nnode on success or a negative 820 * error code on failure. 821 */ 822 static struct ubifs_nnode *nnode_lookup(struct ubifs_info *c, int i) 823 { 824 int err, iip; 825 struct ubifs_nnode *nnode; 826 827 if (!c->nroot) { 828 err = ubifs_read_nnode(c, NULL, 0); 829 if (err) 830 return ERR_PTR(err); 831 } 832 nnode = c->nroot; 833 while (1) { 834 iip = i & (UBIFS_LPT_FANOUT - 1); 835 i >>= UBIFS_LPT_FANOUT_SHIFT; 836 if (!i) 837 break; 838 nnode = ubifs_get_nnode(c, nnode, iip); 839 if (IS_ERR(nnode)) 840 return nnode; 841 } 842 return nnode; 843 } 844 845 /** 846 * make_nnode_dirty - find a nnode and, if found, make it dirty. 847 * @c: UBIFS file-system description object 848 * @node_num: nnode number of nnode to make dirty 849 * @lnum: LEB number where nnode was written 850 * @offs: offset where nnode was written 851 * 852 * This function is used by LPT garbage collection. LPT garbage collection is 853 * used only for the "big" LPT model (c->big_lpt == 1). Garbage collection 854 * simply involves marking all the nodes in the LEB being garbage-collected as 855 * dirty. The dirty nodes are written next commit, after which the LEB is free 856 * to be reused. 857 * 858 * This function returns %0 on success and a negative error code on failure. 859 */ 860 static int make_nnode_dirty(struct ubifs_info *c, int node_num, int lnum, 861 int offs) 862 { 863 struct ubifs_nnode *nnode; 864 865 nnode = nnode_lookup(c, node_num); 866 if (IS_ERR(nnode)) 867 return PTR_ERR(nnode); 868 if (nnode->parent) { 869 struct ubifs_nbranch *branch; 870 871 branch = &nnode->parent->nbranch[nnode->iip]; 872 if (branch->lnum != lnum || branch->offs != offs) 873 return 0; /* nnode is obsolete */ 874 } else if (c->lpt_lnum != lnum || c->lpt_offs != offs) 875 return 0; /* nnode is obsolete */ 876 /* Assumes cnext list is empty i.e. not called during commit */ 877 if (!test_and_set_bit(DIRTY_CNODE, &nnode->flags)) { 878 c->dirty_nn_cnt += 1; 879 ubifs_add_nnode_dirt(c, nnode); 880 /* Mark parent and ancestors dirty too */ 881 nnode = nnode->parent; 882 while (nnode) { 883 if (!test_and_set_bit(DIRTY_CNODE, &nnode->flags)) { 884 c->dirty_nn_cnt += 1; 885 ubifs_add_nnode_dirt(c, nnode); 886 nnode = nnode->parent; 887 } else 888 break; 889 } 890 } 891 return 0; 892 } 893 894 /** 895 * make_pnode_dirty - find a pnode and, if found, make it dirty. 896 * @c: UBIFS file-system description object 897 * @node_num: pnode number of pnode to make dirty 898 * @lnum: LEB number where pnode was written 899 * @offs: offset where pnode was written 900 * 901 * This function is used by LPT garbage collection. LPT garbage collection is 902 * used only for the "big" LPT model (c->big_lpt == 1). Garbage collection 903 * simply involves marking all the nodes in the LEB being garbage-collected as 904 * dirty. The dirty nodes are written next commit, after which the LEB is free 905 * to be reused. 906 * 907 * This function returns %0 on success and a negative error code on failure. 908 */ 909 static int make_pnode_dirty(struct ubifs_info *c, int node_num, int lnum, 910 int offs) 911 { 912 struct ubifs_pnode *pnode; 913 struct ubifs_nbranch *branch; 914 915 pnode = ubifs_pnode_lookup(c, node_num); 916 if (IS_ERR(pnode)) 917 return PTR_ERR(pnode); 918 branch = &pnode->parent->nbranch[pnode->iip]; 919 if (branch->lnum != lnum || branch->offs != offs) 920 return 0; 921 do_make_pnode_dirty(c, pnode); 922 return 0; 923 } 924 925 /** 926 * make_ltab_dirty - make ltab node dirty. 927 * @c: UBIFS file-system description object 928 * @lnum: LEB number where ltab was written 929 * @offs: offset where ltab was written 930 * 931 * This function is used by LPT garbage collection. LPT garbage collection is 932 * used only for the "big" LPT model (c->big_lpt == 1). Garbage collection 933 * simply involves marking all the nodes in the LEB being garbage-collected as 934 * dirty. The dirty nodes are written next commit, after which the LEB is free 935 * to be reused. 936 * 937 * This function returns %0 on success and a negative error code on failure. 938 */ 939 static int make_ltab_dirty(struct ubifs_info *c, int lnum, int offs) 940 { 941 if (lnum != c->ltab_lnum || offs != c->ltab_offs) 942 return 0; /* This ltab node is obsolete */ 943 if (!(c->lpt_drty_flgs & LTAB_DIRTY)) { 944 c->lpt_drty_flgs |= LTAB_DIRTY; 945 ubifs_add_lpt_dirt(c, c->ltab_lnum, c->ltab_sz); 946 } 947 return 0; 948 } 949 950 /** 951 * make_lsave_dirty - make lsave node dirty. 952 * @c: UBIFS file-system description object 953 * @lnum: LEB number where lsave was written 954 * @offs: offset where lsave was written 955 * 956 * This function is used by LPT garbage collection. LPT garbage collection is 957 * used only for the "big" LPT model (c->big_lpt == 1). Garbage collection 958 * simply involves marking all the nodes in the LEB being garbage-collected as 959 * dirty. The dirty nodes are written next commit, after which the LEB is free 960 * to be reused. 961 * 962 * This function returns %0 on success and a negative error code on failure. 963 */ 964 static int make_lsave_dirty(struct ubifs_info *c, int lnum, int offs) 965 { 966 if (lnum != c->lsave_lnum || offs != c->lsave_offs) 967 return 0; /* This lsave node is obsolete */ 968 if (!(c->lpt_drty_flgs & LSAVE_DIRTY)) { 969 c->lpt_drty_flgs |= LSAVE_DIRTY; 970 ubifs_add_lpt_dirt(c, c->lsave_lnum, c->lsave_sz); 971 } 972 return 0; 973 } 974 975 /** 976 * make_node_dirty - make node dirty. 977 * @c: UBIFS file-system description object 978 * @node_type: LPT node type 979 * @node_num: node number 980 * @lnum: LEB number where node was written 981 * @offs: offset where node was written 982 * 983 * This function is used by LPT garbage collection. LPT garbage collection is 984 * used only for the "big" LPT model (c->big_lpt == 1). Garbage collection 985 * simply involves marking all the nodes in the LEB being garbage-collected as 986 * dirty. The dirty nodes are written next commit, after which the LEB is free 987 * to be reused. 988 * 989 * This function returns %0 on success and a negative error code on failure. 990 */ 991 static int make_node_dirty(struct ubifs_info *c, int node_type, int node_num, 992 int lnum, int offs) 993 { 994 switch (node_type) { 995 case UBIFS_LPT_NNODE: 996 return make_nnode_dirty(c, node_num, lnum, offs); 997 case UBIFS_LPT_PNODE: 998 return make_pnode_dirty(c, node_num, lnum, offs); 999 case UBIFS_LPT_LTAB: 1000 return make_ltab_dirty(c, lnum, offs); 1001 case UBIFS_LPT_LSAVE: 1002 return make_lsave_dirty(c, lnum, offs); 1003 } 1004 return -EINVAL; 1005 } 1006 1007 /** 1008 * get_lpt_node_len - return the length of a node based on its type. 1009 * @c: UBIFS file-system description object 1010 * @node_type: LPT node type 1011 */ 1012 static int get_lpt_node_len(const struct ubifs_info *c, int node_type) 1013 { 1014 switch (node_type) { 1015 case UBIFS_LPT_NNODE: 1016 return c->nnode_sz; 1017 case UBIFS_LPT_PNODE: 1018 return c->pnode_sz; 1019 case UBIFS_LPT_LTAB: 1020 return c->ltab_sz; 1021 case UBIFS_LPT_LSAVE: 1022 return c->lsave_sz; 1023 } 1024 return 0; 1025 } 1026 1027 /** 1028 * get_pad_len - return the length of padding in a buffer. 1029 * @c: UBIFS file-system description object 1030 * @buf: buffer 1031 * @len: length of buffer 1032 */ 1033 static int get_pad_len(const struct ubifs_info *c, uint8_t *buf, int len) 1034 { 1035 int offs, pad_len; 1036 1037 if (c->min_io_size == 1) 1038 return 0; 1039 offs = c->leb_size - len; 1040 pad_len = ALIGN(offs, c->min_io_size) - offs; 1041 return pad_len; 1042 } 1043 1044 /** 1045 * get_lpt_node_type - return type (and node number) of a node in a buffer. 1046 * @c: UBIFS file-system description object 1047 * @buf: buffer 1048 * @node_num: node number is returned here 1049 */ 1050 static int get_lpt_node_type(const struct ubifs_info *c, uint8_t *buf, 1051 int *node_num) 1052 { 1053 uint8_t *addr = buf + UBIFS_LPT_CRC_BYTES; 1054 int pos = 0, node_type; 1055 1056 node_type = ubifs_unpack_bits(c, &addr, &pos, UBIFS_LPT_TYPE_BITS); 1057 *node_num = ubifs_unpack_bits(c, &addr, &pos, c->pcnt_bits); 1058 return node_type; 1059 } 1060 1061 /** 1062 * is_a_node - determine if a buffer contains a node. 1063 * @c: UBIFS file-system description object 1064 * @buf: buffer 1065 * @len: length of buffer 1066 * 1067 * This function returns %1 if the buffer contains a node or %0 if it does not. 1068 */ 1069 static int is_a_node(const struct ubifs_info *c, uint8_t *buf, int len) 1070 { 1071 uint8_t *addr = buf + UBIFS_LPT_CRC_BYTES; 1072 int pos = 0, node_type, node_len; 1073 uint16_t crc, calc_crc; 1074 1075 if (len < UBIFS_LPT_CRC_BYTES + (UBIFS_LPT_TYPE_BITS + 7) / 8) 1076 return 0; 1077 node_type = ubifs_unpack_bits(c, &addr, &pos, UBIFS_LPT_TYPE_BITS); 1078 if (node_type == UBIFS_LPT_NOT_A_NODE) 1079 return 0; 1080 node_len = get_lpt_node_len(c, node_type); 1081 if (!node_len || node_len > len) 1082 return 0; 1083 pos = 0; 1084 addr = buf; 1085 crc = ubifs_unpack_bits(c, &addr, &pos, UBIFS_LPT_CRC_BITS); 1086 calc_crc = crc16(-1, buf + UBIFS_LPT_CRC_BYTES, 1087 node_len - UBIFS_LPT_CRC_BYTES); 1088 if (crc != calc_crc) 1089 return 0; 1090 return 1; 1091 } 1092 1093 /** 1094 * lpt_gc_lnum - garbage collect a LPT LEB. 1095 * @c: UBIFS file-system description object 1096 * @lnum: LEB number to garbage collect 1097 * 1098 * LPT garbage collection is used only for the "big" LPT model 1099 * (c->big_lpt == 1). Garbage collection simply involves marking all the nodes 1100 * in the LEB being garbage-collected as dirty. The dirty nodes are written 1101 * next commit, after which the LEB is free to be reused. 1102 * 1103 * This function returns %0 on success and a negative error code on failure. 1104 */ 1105 static int lpt_gc_lnum(struct ubifs_info *c, int lnum) 1106 { 1107 int err, len = c->leb_size, node_type, node_num, node_len, offs; 1108 void *buf = c->lpt_buf; 1109 1110 dbg_lp("LEB %d", lnum); 1111 1112 err = ubifs_leb_read(c, lnum, buf, 0, c->leb_size, 1); 1113 if (err) 1114 return err; 1115 1116 while (1) { 1117 if (!is_a_node(c, buf, len)) { 1118 int pad_len; 1119 1120 pad_len = get_pad_len(c, buf, len); 1121 if (pad_len) { 1122 buf += pad_len; 1123 len -= pad_len; 1124 continue; 1125 } 1126 return 0; 1127 } 1128 node_type = get_lpt_node_type(c, buf, &node_num); 1129 node_len = get_lpt_node_len(c, node_type); 1130 offs = c->leb_size - len; 1131 ubifs_assert(c, node_len != 0); 1132 mutex_lock(&c->lp_mutex); 1133 err = make_node_dirty(c, node_type, node_num, lnum, offs); 1134 mutex_unlock(&c->lp_mutex); 1135 if (err) 1136 return err; 1137 buf += node_len; 1138 len -= node_len; 1139 } 1140 return 0; 1141 } 1142 1143 /** 1144 * lpt_gc - LPT garbage collection. 1145 * @c: UBIFS file-system description object 1146 * 1147 * Select a LPT LEB for LPT garbage collection and call 'lpt_gc_lnum()'. 1148 * Returns %0 on success and a negative error code on failure. 1149 */ 1150 static int lpt_gc(struct ubifs_info *c) 1151 { 1152 int i, lnum = -1, dirty = 0; 1153 1154 mutex_lock(&c->lp_mutex); 1155 for (i = 0; i < c->lpt_lebs; i++) { 1156 ubifs_assert(c, !c->ltab[i].tgc); 1157 if (i + c->lpt_first == c->nhead_lnum || 1158 c->ltab[i].free + c->ltab[i].dirty == c->leb_size) 1159 continue; 1160 if (c->ltab[i].dirty > dirty) { 1161 dirty = c->ltab[i].dirty; 1162 lnum = i + c->lpt_first; 1163 } 1164 } 1165 mutex_unlock(&c->lp_mutex); 1166 if (lnum == -1) 1167 return -ENOSPC; 1168 return lpt_gc_lnum(c, lnum); 1169 } 1170 1171 /** 1172 * ubifs_lpt_start_commit - UBIFS commit starts. 1173 * @c: the UBIFS file-system description object 1174 * 1175 * This function has to be called when UBIFS starts the commit operation. 1176 * This function "freezes" all currently dirty LEB properties and does not 1177 * change them anymore. Further changes are saved and tracked separately 1178 * because they are not part of this commit. This function returns zero in case 1179 * of success and a negative error code in case of failure. 1180 */ 1181 int ubifs_lpt_start_commit(struct ubifs_info *c) 1182 { 1183 int err, cnt; 1184 1185 dbg_lp(""); 1186 1187 mutex_lock(&c->lp_mutex); 1188 err = dbg_chk_lpt_free_spc(c); 1189 if (err) 1190 goto out; 1191 err = dbg_check_ltab(c); 1192 if (err) 1193 goto out; 1194 1195 if (c->check_lpt_free) { 1196 /* 1197 * We ensure there is enough free space in 1198 * ubifs_lpt_post_commit() by marking nodes dirty. That 1199 * information is lost when we unmount, so we also need 1200 * to check free space once after mounting also. 1201 */ 1202 c->check_lpt_free = 0; 1203 while (need_write_all(c)) { 1204 mutex_unlock(&c->lp_mutex); 1205 err = lpt_gc(c); 1206 if (err) 1207 return err; 1208 mutex_lock(&c->lp_mutex); 1209 } 1210 } 1211 1212 lpt_tgc_start(c); 1213 1214 if (!c->dirty_pn_cnt) { 1215 dbg_cmt("no cnodes to commit"); 1216 err = 0; 1217 goto out; 1218 } 1219 1220 if (!c->big_lpt && need_write_all(c)) { 1221 /* If needed, write everything */ 1222 err = make_tree_dirty(c); 1223 if (err) 1224 goto out; 1225 lpt_tgc_start(c); 1226 } 1227 1228 if (c->big_lpt) 1229 populate_lsave(c); 1230 1231 cnt = get_cnodes_to_commit(c); 1232 ubifs_assert(c, cnt != 0); 1233 1234 err = layout_cnodes(c); 1235 if (err) 1236 goto out; 1237 1238 err = ubifs_lpt_calc_hash(c, c->mst_node->hash_lpt); 1239 if (err) 1240 goto out; 1241 1242 /* Copy the LPT's own lprops for end commit to write */ 1243 memcpy(c->ltab_cmt, c->ltab, 1244 sizeof(struct ubifs_lpt_lprops) * c->lpt_lebs); 1245 c->lpt_drty_flgs &= ~(LTAB_DIRTY | LSAVE_DIRTY); 1246 1247 out: 1248 mutex_unlock(&c->lp_mutex); 1249 return err; 1250 } 1251 1252 /** 1253 * free_obsolete_cnodes - free obsolete cnodes for commit end. 1254 * @c: UBIFS file-system description object 1255 */ 1256 static void free_obsolete_cnodes(struct ubifs_info *c) 1257 { 1258 struct ubifs_cnode *cnode, *cnext; 1259 1260 cnext = c->lpt_cnext; 1261 if (!cnext) 1262 return; 1263 do { 1264 cnode = cnext; 1265 cnext = cnode->cnext; 1266 if (test_bit(OBSOLETE_CNODE, &cnode->flags)) 1267 kfree(cnode); 1268 else 1269 cnode->cnext = NULL; 1270 } while (cnext != c->lpt_cnext); 1271 c->lpt_cnext = NULL; 1272 } 1273 1274 /** 1275 * ubifs_lpt_end_commit - finish the commit operation. 1276 * @c: the UBIFS file-system description object 1277 * 1278 * This function has to be called when the commit operation finishes. It 1279 * flushes the changes which were "frozen" by 'ubifs_lprops_start_commit()' to 1280 * the media. Returns zero in case of success and a negative error code in case 1281 * of failure. 1282 */ 1283 int ubifs_lpt_end_commit(struct ubifs_info *c) 1284 { 1285 int err; 1286 1287 dbg_lp(""); 1288 1289 if (!c->lpt_cnext) 1290 return 0; 1291 1292 err = write_cnodes(c); 1293 if (err) 1294 return err; 1295 1296 mutex_lock(&c->lp_mutex); 1297 free_obsolete_cnodes(c); 1298 mutex_unlock(&c->lp_mutex); 1299 1300 return 0; 1301 } 1302 1303 /** 1304 * ubifs_lpt_post_commit - post commit LPT trivial GC and LPT GC. 1305 * @c: UBIFS file-system description object 1306 * 1307 * LPT trivial GC is completed after a commit. Also LPT GC is done after a 1308 * commit for the "big" LPT model. 1309 */ 1310 int ubifs_lpt_post_commit(struct ubifs_info *c) 1311 { 1312 int err; 1313 1314 mutex_lock(&c->lp_mutex); 1315 err = lpt_tgc_end(c); 1316 if (err) 1317 goto out; 1318 if (c->big_lpt) 1319 while (need_write_all(c)) { 1320 mutex_unlock(&c->lp_mutex); 1321 err = lpt_gc(c); 1322 if (err) 1323 return err; 1324 mutex_lock(&c->lp_mutex); 1325 } 1326 out: 1327 mutex_unlock(&c->lp_mutex); 1328 return err; 1329 } 1330 1331 /** 1332 * first_nnode - find the first nnode in memory. 1333 * @c: UBIFS file-system description object 1334 * @hght: height of tree where nnode found is returned here 1335 * 1336 * This function returns a pointer to the nnode found or %NULL if no nnode is 1337 * found. This function is a helper to 'ubifs_lpt_free()'. 1338 */ 1339 static struct ubifs_nnode *first_nnode(struct ubifs_info *c, int *hght) 1340 { 1341 struct ubifs_nnode *nnode; 1342 int h, i, found; 1343 1344 nnode = c->nroot; 1345 *hght = 0; 1346 if (!nnode) 1347 return NULL; 1348 for (h = 1; h < c->lpt_hght; h++) { 1349 found = 0; 1350 for (i = 0; i < UBIFS_LPT_FANOUT; i++) { 1351 if (nnode->nbranch[i].nnode) { 1352 found = 1; 1353 nnode = nnode->nbranch[i].nnode; 1354 *hght = h; 1355 break; 1356 } 1357 } 1358 if (!found) 1359 break; 1360 } 1361 return nnode; 1362 } 1363 1364 /** 1365 * next_nnode - find the next nnode in memory. 1366 * @c: UBIFS file-system description object 1367 * @nnode: nnode from which to start. 1368 * @hght: height of tree where nnode is, is passed and returned here 1369 * 1370 * This function returns a pointer to the nnode found or %NULL if no nnode is 1371 * found. This function is a helper to 'ubifs_lpt_free()'. 1372 */ 1373 static struct ubifs_nnode *next_nnode(struct ubifs_info *c, 1374 struct ubifs_nnode *nnode, int *hght) 1375 { 1376 struct ubifs_nnode *parent; 1377 int iip, h, i, found; 1378 1379 parent = nnode->parent; 1380 if (!parent) 1381 return NULL; 1382 if (nnode->iip == UBIFS_LPT_FANOUT - 1) { 1383 *hght -= 1; 1384 return parent; 1385 } 1386 for (iip = nnode->iip + 1; iip < UBIFS_LPT_FANOUT; iip++) { 1387 nnode = parent->nbranch[iip].nnode; 1388 if (nnode) 1389 break; 1390 } 1391 if (!nnode) { 1392 *hght -= 1; 1393 return parent; 1394 } 1395 for (h = *hght + 1; h < c->lpt_hght; h++) { 1396 found = 0; 1397 for (i = 0; i < UBIFS_LPT_FANOUT; i++) { 1398 if (nnode->nbranch[i].nnode) { 1399 found = 1; 1400 nnode = nnode->nbranch[i].nnode; 1401 *hght = h; 1402 break; 1403 } 1404 } 1405 if (!found) 1406 break; 1407 } 1408 return nnode; 1409 } 1410 1411 /** 1412 * ubifs_lpt_free - free resources owned by the LPT. 1413 * @c: UBIFS file-system description object 1414 * @wr_only: free only resources used for writing 1415 */ 1416 void ubifs_lpt_free(struct ubifs_info *c, int wr_only) 1417 { 1418 struct ubifs_nnode *nnode; 1419 int i, hght; 1420 1421 /* Free write-only things first */ 1422 1423 free_obsolete_cnodes(c); /* Leftover from a failed commit */ 1424 1425 vfree(c->ltab_cmt); 1426 c->ltab_cmt = NULL; 1427 vfree(c->lpt_buf); 1428 c->lpt_buf = NULL; 1429 kfree(c->lsave); 1430 c->lsave = NULL; 1431 1432 if (wr_only) 1433 return; 1434 1435 /* Now free the rest */ 1436 1437 nnode = first_nnode(c, &hght); 1438 while (nnode) { 1439 for (i = 0; i < UBIFS_LPT_FANOUT; i++) 1440 kfree(nnode->nbranch[i].nnode); 1441 nnode = next_nnode(c, nnode, &hght); 1442 } 1443 for (i = 0; i < LPROPS_HEAP_CNT; i++) 1444 kfree(c->lpt_heap[i].arr); 1445 kfree(c->dirty_idx.arr); 1446 kfree(c->nroot); 1447 vfree(c->ltab); 1448 kfree(c->lpt_nod_buf); 1449 } 1450 1451 /* 1452 * Everything below is related to debugging. 1453 */ 1454 1455 /** 1456 * dbg_is_all_ff - determine if a buffer contains only 0xFF bytes. 1457 * @buf: buffer 1458 * @len: buffer length 1459 */ 1460 static int dbg_is_all_ff(uint8_t *buf, int len) 1461 { 1462 int i; 1463 1464 for (i = 0; i < len; i++) 1465 if (buf[i] != 0xff) 1466 return 0; 1467 return 1; 1468 } 1469 1470 /** 1471 * dbg_is_nnode_dirty - determine if a nnode is dirty. 1472 * @c: the UBIFS file-system description object 1473 * @lnum: LEB number where nnode was written 1474 * @offs: offset where nnode was written 1475 */ 1476 static int dbg_is_nnode_dirty(struct ubifs_info *c, int lnum, int offs) 1477 { 1478 struct ubifs_nnode *nnode; 1479 int hght; 1480 1481 /* Entire tree is in memory so first_nnode / next_nnode are OK */ 1482 nnode = first_nnode(c, &hght); 1483 for (; nnode; nnode = next_nnode(c, nnode, &hght)) { 1484 struct ubifs_nbranch *branch; 1485 1486 cond_resched(); 1487 if (nnode->parent) { 1488 branch = &nnode->parent->nbranch[nnode->iip]; 1489 if (branch->lnum != lnum || branch->offs != offs) 1490 continue; 1491 if (test_bit(DIRTY_CNODE, &nnode->flags)) 1492 return 1; 1493 return 0; 1494 } else { 1495 if (c->lpt_lnum != lnum || c->lpt_offs != offs) 1496 continue; 1497 if (test_bit(DIRTY_CNODE, &nnode->flags)) 1498 return 1; 1499 return 0; 1500 } 1501 } 1502 return 1; 1503 } 1504 1505 /** 1506 * dbg_is_pnode_dirty - determine if a pnode is dirty. 1507 * @c: the UBIFS file-system description object 1508 * @lnum: LEB number where pnode was written 1509 * @offs: offset where pnode was written 1510 */ 1511 static int dbg_is_pnode_dirty(struct ubifs_info *c, int lnum, int offs) 1512 { 1513 int i, cnt; 1514 1515 cnt = DIV_ROUND_UP(c->main_lebs, UBIFS_LPT_FANOUT); 1516 for (i = 0; i < cnt; i++) { 1517 struct ubifs_pnode *pnode; 1518 struct ubifs_nbranch *branch; 1519 1520 cond_resched(); 1521 pnode = ubifs_pnode_lookup(c, i); 1522 if (IS_ERR(pnode)) 1523 return PTR_ERR(pnode); 1524 branch = &pnode->parent->nbranch[pnode->iip]; 1525 if (branch->lnum != lnum || branch->offs != offs) 1526 continue; 1527 if (test_bit(DIRTY_CNODE, &pnode->flags)) 1528 return 1; 1529 return 0; 1530 } 1531 return 1; 1532 } 1533 1534 /** 1535 * dbg_is_ltab_dirty - determine if a ltab node is dirty. 1536 * @c: the UBIFS file-system description object 1537 * @lnum: LEB number where ltab node was written 1538 * @offs: offset where ltab node was written 1539 */ 1540 static int dbg_is_ltab_dirty(struct ubifs_info *c, int lnum, int offs) 1541 { 1542 if (lnum != c->ltab_lnum || offs != c->ltab_offs) 1543 return 1; 1544 return (c->lpt_drty_flgs & LTAB_DIRTY) != 0; 1545 } 1546 1547 /** 1548 * dbg_is_lsave_dirty - determine if a lsave node is dirty. 1549 * @c: the UBIFS file-system description object 1550 * @lnum: LEB number where lsave node was written 1551 * @offs: offset where lsave node was written 1552 */ 1553 static int dbg_is_lsave_dirty(struct ubifs_info *c, int lnum, int offs) 1554 { 1555 if (lnum != c->lsave_lnum || offs != c->lsave_offs) 1556 return 1; 1557 return (c->lpt_drty_flgs & LSAVE_DIRTY) != 0; 1558 } 1559 1560 /** 1561 * dbg_is_node_dirty - determine if a node is dirty. 1562 * @c: the UBIFS file-system description object 1563 * @node_type: node type 1564 * @lnum: LEB number where node was written 1565 * @offs: offset where node was written 1566 */ 1567 static int dbg_is_node_dirty(struct ubifs_info *c, int node_type, int lnum, 1568 int offs) 1569 { 1570 switch (node_type) { 1571 case UBIFS_LPT_NNODE: 1572 return dbg_is_nnode_dirty(c, lnum, offs); 1573 case UBIFS_LPT_PNODE: 1574 return dbg_is_pnode_dirty(c, lnum, offs); 1575 case UBIFS_LPT_LTAB: 1576 return dbg_is_ltab_dirty(c, lnum, offs); 1577 case UBIFS_LPT_LSAVE: 1578 return dbg_is_lsave_dirty(c, lnum, offs); 1579 } 1580 return 1; 1581 } 1582 1583 /** 1584 * dbg_check_ltab_lnum - check the ltab for a LPT LEB number. 1585 * @c: the UBIFS file-system description object 1586 * @lnum: LEB number where node was written 1587 * 1588 * This function returns %0 on success and a negative error code on failure. 1589 */ 1590 static int dbg_check_ltab_lnum(struct ubifs_info *c, int lnum) 1591 { 1592 int err, len = c->leb_size, dirty = 0, node_type, node_num, node_len; 1593 int ret; 1594 void *buf, *p; 1595 1596 if (!dbg_is_chk_lprops(c)) 1597 return 0; 1598 1599 buf = p = __vmalloc(c->leb_size, GFP_NOFS); 1600 if (!buf) { 1601 ubifs_err(c, "cannot allocate memory for ltab checking"); 1602 return 0; 1603 } 1604 1605 dbg_lp("LEB %d", lnum); 1606 1607 err = ubifs_leb_read(c, lnum, buf, 0, c->leb_size, 1); 1608 if (err) 1609 goto out; 1610 1611 while (1) { 1612 if (!is_a_node(c, p, len)) { 1613 int i, pad_len; 1614 1615 pad_len = get_pad_len(c, p, len); 1616 if (pad_len) { 1617 p += pad_len; 1618 len -= pad_len; 1619 dirty += pad_len; 1620 continue; 1621 } 1622 if (!dbg_is_all_ff(p, len)) { 1623 ubifs_err(c, "invalid empty space in LEB %d at %d", 1624 lnum, c->leb_size - len); 1625 err = -EINVAL; 1626 } 1627 i = lnum - c->lpt_first; 1628 if (len != c->ltab[i].free) { 1629 ubifs_err(c, "invalid free space in LEB %d (free %d, expected %d)", 1630 lnum, len, c->ltab[i].free); 1631 err = -EINVAL; 1632 } 1633 if (dirty != c->ltab[i].dirty) { 1634 ubifs_err(c, "invalid dirty space in LEB %d (dirty %d, expected %d)", 1635 lnum, dirty, c->ltab[i].dirty); 1636 err = -EINVAL; 1637 } 1638 goto out; 1639 } 1640 node_type = get_lpt_node_type(c, p, &node_num); 1641 node_len = get_lpt_node_len(c, node_type); 1642 ret = dbg_is_node_dirty(c, node_type, lnum, c->leb_size - len); 1643 if (ret == 1) 1644 dirty += node_len; 1645 p += node_len; 1646 len -= node_len; 1647 } 1648 1649 err = 0; 1650 out: 1651 vfree(buf); 1652 return err; 1653 } 1654 1655 /** 1656 * dbg_check_ltab - check the free and dirty space in the ltab. 1657 * @c: the UBIFS file-system description object 1658 * 1659 * This function returns %0 on success and a negative error code on failure. 1660 */ 1661 int dbg_check_ltab(struct ubifs_info *c) 1662 { 1663 int lnum, err, i, cnt; 1664 1665 if (!dbg_is_chk_lprops(c)) 1666 return 0; 1667 1668 /* Bring the entire tree into memory */ 1669 cnt = DIV_ROUND_UP(c->main_lebs, UBIFS_LPT_FANOUT); 1670 for (i = 0; i < cnt; i++) { 1671 struct ubifs_pnode *pnode; 1672 1673 pnode = ubifs_pnode_lookup(c, i); 1674 if (IS_ERR(pnode)) 1675 return PTR_ERR(pnode); 1676 cond_resched(); 1677 } 1678 1679 /* Check nodes */ 1680 err = dbg_check_lpt_nodes(c, (struct ubifs_cnode *)c->nroot, 0, 0); 1681 if (err) 1682 return err; 1683 1684 /* Check each LEB */ 1685 for (lnum = c->lpt_first; lnum <= c->lpt_last; lnum++) { 1686 err = dbg_check_ltab_lnum(c, lnum); 1687 if (err) { 1688 ubifs_err(c, "failed at LEB %d", lnum); 1689 return err; 1690 } 1691 } 1692 1693 dbg_lp("succeeded"); 1694 return 0; 1695 } 1696 1697 /** 1698 * dbg_chk_lpt_free_spc - check LPT free space is enough to write entire LPT. 1699 * @c: the UBIFS file-system description object 1700 * 1701 * This function returns %0 on success and a negative error code on failure. 1702 */ 1703 int dbg_chk_lpt_free_spc(struct ubifs_info *c) 1704 { 1705 long long free = 0; 1706 int i; 1707 1708 if (!dbg_is_chk_lprops(c)) 1709 return 0; 1710 1711 for (i = 0; i < c->lpt_lebs; i++) { 1712 if (c->ltab[i].tgc || c->ltab[i].cmt) 1713 continue; 1714 if (i + c->lpt_first == c->nhead_lnum) 1715 free += c->leb_size - c->nhead_offs; 1716 else if (c->ltab[i].free == c->leb_size) 1717 free += c->leb_size; 1718 } 1719 if (free < c->lpt_sz) { 1720 ubifs_err(c, "LPT space error: free %lld lpt_sz %lld", 1721 free, c->lpt_sz); 1722 ubifs_dump_lpt_info(c); 1723 ubifs_dump_lpt_lebs(c); 1724 dump_stack(); 1725 return -EINVAL; 1726 } 1727 return 0; 1728 } 1729 1730 /** 1731 * dbg_chk_lpt_sz - check LPT does not write more than LPT size. 1732 * @c: the UBIFS file-system description object 1733 * @action: what to do 1734 * @len: length written 1735 * 1736 * This function returns %0 on success and a negative error code on failure. 1737 * The @action argument may be one of: 1738 * o %0 - LPT debugging checking starts, initialize debugging variables; 1739 * o %1 - wrote an LPT node, increase LPT size by @len bytes; 1740 * o %2 - switched to a different LEB and wasted @len bytes; 1741 * o %3 - check that we've written the right number of bytes. 1742 * o %4 - wasted @len bytes; 1743 */ 1744 int dbg_chk_lpt_sz(struct ubifs_info *c, int action, int len) 1745 { 1746 struct ubifs_debug_info *d = c->dbg; 1747 long long chk_lpt_sz, lpt_sz; 1748 int err = 0; 1749 1750 if (!dbg_is_chk_lprops(c)) 1751 return 0; 1752 1753 switch (action) { 1754 case 0: 1755 d->chk_lpt_sz = 0; 1756 d->chk_lpt_sz2 = 0; 1757 d->chk_lpt_lebs = 0; 1758 d->chk_lpt_wastage = 0; 1759 if (c->dirty_pn_cnt > c->pnode_cnt) { 1760 ubifs_err(c, "dirty pnodes %d exceed max %d", 1761 c->dirty_pn_cnt, c->pnode_cnt); 1762 err = -EINVAL; 1763 } 1764 if (c->dirty_nn_cnt > c->nnode_cnt) { 1765 ubifs_err(c, "dirty nnodes %d exceed max %d", 1766 c->dirty_nn_cnt, c->nnode_cnt); 1767 err = -EINVAL; 1768 } 1769 return err; 1770 case 1: 1771 d->chk_lpt_sz += len; 1772 return 0; 1773 case 2: 1774 d->chk_lpt_sz += len; 1775 d->chk_lpt_wastage += len; 1776 d->chk_lpt_lebs += 1; 1777 return 0; 1778 case 3: 1779 chk_lpt_sz = c->leb_size; 1780 chk_lpt_sz *= d->chk_lpt_lebs; 1781 chk_lpt_sz += len - c->nhead_offs; 1782 if (d->chk_lpt_sz != chk_lpt_sz) { 1783 ubifs_err(c, "LPT wrote %lld but space used was %lld", 1784 d->chk_lpt_sz, chk_lpt_sz); 1785 err = -EINVAL; 1786 } 1787 if (d->chk_lpt_sz > c->lpt_sz) { 1788 ubifs_err(c, "LPT wrote %lld but lpt_sz is %lld", 1789 d->chk_lpt_sz, c->lpt_sz); 1790 err = -EINVAL; 1791 } 1792 if (d->chk_lpt_sz2 && d->chk_lpt_sz != d->chk_lpt_sz2) { 1793 ubifs_err(c, "LPT layout size %lld but wrote %lld", 1794 d->chk_lpt_sz, d->chk_lpt_sz2); 1795 err = -EINVAL; 1796 } 1797 if (d->chk_lpt_sz2 && d->new_nhead_offs != len) { 1798 ubifs_err(c, "LPT new nhead offs: expected %d was %d", 1799 d->new_nhead_offs, len); 1800 err = -EINVAL; 1801 } 1802 lpt_sz = (long long)c->pnode_cnt * c->pnode_sz; 1803 lpt_sz += (long long)c->nnode_cnt * c->nnode_sz; 1804 lpt_sz += c->ltab_sz; 1805 if (c->big_lpt) 1806 lpt_sz += c->lsave_sz; 1807 if (d->chk_lpt_sz - d->chk_lpt_wastage > lpt_sz) { 1808 ubifs_err(c, "LPT chk_lpt_sz %lld + waste %lld exceeds %lld", 1809 d->chk_lpt_sz, d->chk_lpt_wastage, lpt_sz); 1810 err = -EINVAL; 1811 } 1812 if (err) { 1813 ubifs_dump_lpt_info(c); 1814 ubifs_dump_lpt_lebs(c); 1815 dump_stack(); 1816 } 1817 d->chk_lpt_sz2 = d->chk_lpt_sz; 1818 d->chk_lpt_sz = 0; 1819 d->chk_lpt_wastage = 0; 1820 d->chk_lpt_lebs = 0; 1821 d->new_nhead_offs = len; 1822 return err; 1823 case 4: 1824 d->chk_lpt_sz += len; 1825 d->chk_lpt_wastage += len; 1826 return 0; 1827 default: 1828 return -EINVAL; 1829 } 1830 } 1831 1832 /** 1833 * dump_lpt_leb - dump an LPT LEB. 1834 * @c: UBIFS file-system description object 1835 * @lnum: LEB number to dump 1836 * 1837 * This function dumps an LEB from LPT area. Nodes in this area are very 1838 * different to nodes in the main area (e.g., they do not have common headers, 1839 * they do not have 8-byte alignments, etc), so we have a separate function to 1840 * dump LPT area LEBs. Note, LPT has to be locked by the caller. 1841 */ 1842 static void dump_lpt_leb(const struct ubifs_info *c, int lnum) 1843 { 1844 int err, len = c->leb_size, node_type, node_num, node_len, offs; 1845 void *buf, *p; 1846 1847 pr_err("(pid %d) start dumping LEB %d\n", current->pid, lnum); 1848 buf = p = __vmalloc(c->leb_size, GFP_NOFS); 1849 if (!buf) { 1850 ubifs_err(c, "cannot allocate memory to dump LPT"); 1851 return; 1852 } 1853 1854 err = ubifs_leb_read(c, lnum, buf, 0, c->leb_size, 1); 1855 if (err) 1856 goto out; 1857 1858 while (1) { 1859 offs = c->leb_size - len; 1860 if (!is_a_node(c, p, len)) { 1861 int pad_len; 1862 1863 pad_len = get_pad_len(c, p, len); 1864 if (pad_len) { 1865 pr_err("LEB %d:%d, pad %d bytes\n", 1866 lnum, offs, pad_len); 1867 p += pad_len; 1868 len -= pad_len; 1869 continue; 1870 } 1871 if (len) 1872 pr_err("LEB %d:%d, free %d bytes\n", 1873 lnum, offs, len); 1874 break; 1875 } 1876 1877 node_type = get_lpt_node_type(c, p, &node_num); 1878 switch (node_type) { 1879 case UBIFS_LPT_PNODE: 1880 { 1881 node_len = c->pnode_sz; 1882 if (c->big_lpt) 1883 pr_err("LEB %d:%d, pnode num %d\n", 1884 lnum, offs, node_num); 1885 else 1886 pr_err("LEB %d:%d, pnode\n", lnum, offs); 1887 break; 1888 } 1889 case UBIFS_LPT_NNODE: 1890 { 1891 int i; 1892 struct ubifs_nnode nnode; 1893 1894 node_len = c->nnode_sz; 1895 if (c->big_lpt) 1896 pr_err("LEB %d:%d, nnode num %d, ", 1897 lnum, offs, node_num); 1898 else 1899 pr_err("LEB %d:%d, nnode, ", 1900 lnum, offs); 1901 err = ubifs_unpack_nnode(c, p, &nnode); 1902 if (err) { 1903 pr_err("failed to unpack_node, error %d\n", 1904 err); 1905 break; 1906 } 1907 for (i = 0; i < UBIFS_LPT_FANOUT; i++) { 1908 pr_cont("%d:%d", nnode.nbranch[i].lnum, 1909 nnode.nbranch[i].offs); 1910 if (i != UBIFS_LPT_FANOUT - 1) 1911 pr_cont(", "); 1912 } 1913 pr_cont("\n"); 1914 break; 1915 } 1916 case UBIFS_LPT_LTAB: 1917 node_len = c->ltab_sz; 1918 pr_err("LEB %d:%d, ltab\n", lnum, offs); 1919 break; 1920 case UBIFS_LPT_LSAVE: 1921 node_len = c->lsave_sz; 1922 pr_err("LEB %d:%d, lsave len\n", lnum, offs); 1923 break; 1924 default: 1925 ubifs_err(c, "LPT node type %d not recognized", node_type); 1926 goto out; 1927 } 1928 1929 p += node_len; 1930 len -= node_len; 1931 } 1932 1933 pr_err("(pid %d) finish dumping LEB %d\n", current->pid, lnum); 1934 out: 1935 vfree(buf); 1936 return; 1937 } 1938 1939 /** 1940 * ubifs_dump_lpt_lebs - dump LPT lebs. 1941 * @c: UBIFS file-system description object 1942 * 1943 * This function dumps all LPT LEBs. The caller has to make sure the LPT is 1944 * locked. 1945 */ 1946 void ubifs_dump_lpt_lebs(const struct ubifs_info *c) 1947 { 1948 int i; 1949 1950 pr_err("(pid %d) start dumping all LPT LEBs\n", current->pid); 1951 for (i = 0; i < c->lpt_lebs; i++) 1952 dump_lpt_leb(c, i + c->lpt_first); 1953 pr_err("(pid %d) finish dumping all LPT LEBs\n", current->pid); 1954 } 1955 1956 /** 1957 * dbg_populate_lsave - debugging version of 'populate_lsave()' 1958 * @c: UBIFS file-system description object 1959 * 1960 * This is a debugging version for 'populate_lsave()' which populates lsave 1961 * with random LEBs instead of useful LEBs, which is good for test coverage. 1962 * Returns zero if lsave has not been populated (this debugging feature is 1963 * disabled) an non-zero if lsave has been populated. 1964 */ 1965 static int dbg_populate_lsave(struct ubifs_info *c) 1966 { 1967 struct ubifs_lprops *lprops; 1968 struct ubifs_lpt_heap *heap; 1969 int i; 1970 1971 if (!dbg_is_chk_gen(c)) 1972 return 0; 1973 if (prandom_u32() & 3) 1974 return 0; 1975 1976 for (i = 0; i < c->lsave_cnt; i++) 1977 c->lsave[i] = c->main_first; 1978 1979 list_for_each_entry(lprops, &c->empty_list, list) 1980 c->lsave[prandom_u32() % c->lsave_cnt] = lprops->lnum; 1981 list_for_each_entry(lprops, &c->freeable_list, list) 1982 c->lsave[prandom_u32() % c->lsave_cnt] = lprops->lnum; 1983 list_for_each_entry(lprops, &c->frdi_idx_list, list) 1984 c->lsave[prandom_u32() % c->lsave_cnt] = lprops->lnum; 1985 1986 heap = &c->lpt_heap[LPROPS_DIRTY_IDX - 1]; 1987 for (i = 0; i < heap->cnt; i++) 1988 c->lsave[prandom_u32() % c->lsave_cnt] = heap->arr[i]->lnum; 1989 heap = &c->lpt_heap[LPROPS_DIRTY - 1]; 1990 for (i = 0; i < heap->cnt; i++) 1991 c->lsave[prandom_u32() % c->lsave_cnt] = heap->arr[i]->lnum; 1992 heap = &c->lpt_heap[LPROPS_FREE - 1]; 1993 for (i = 0; i < heap->cnt; i++) 1994 c->lsave[prandom_u32() % c->lsave_cnt] = heap->arr[i]->lnum; 1995 1996 return 1; 1997 } 1998