1 /* $NetBSD: lfs_subr.c,v 1.102 2020/02/23 15:23:08 riastradh Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Konrad E. Schroder <perseant@hhhh.org>. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 /* 32 * Copyright (c) 1991, 1993 33 * The Regents of the University of California. All rights reserved. 34 * 35 * Redistribution and use in source and binary forms, with or without 36 * modification, are permitted provided that the following conditions 37 * are met: 38 * 1. Redistributions of source code must retain the above copyright 39 * notice, this list of conditions and the following disclaimer. 40 * 2. Redistributions in binary form must reproduce the above copyright 41 * notice, this list of conditions and the following disclaimer in the 42 * documentation and/or other materials provided with the distribution. 43 * 3. Neither the name of the University nor the names of its contributors 44 * may be used to endorse or promote products derived from this software 45 * without specific prior written permission. 46 * 47 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 48 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 49 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 50 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 51 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 52 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 53 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 54 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 55 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 56 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 57 * SUCH DAMAGE. 58 * 59 * @(#)lfs_subr.c 8.4 (Berkeley) 5/8/95 60 */ 61 62 #include <sys/cdefs.h> 63 __KERNEL_RCSID(0, "$NetBSD: lfs_subr.c,v 1.102 2020/02/23 15:23:08 riastradh Exp $"); 64 65 #include <sys/param.h> 66 #include <sys/systm.h> 67 #include <sys/namei.h> 68 #include <sys/vnode.h> 69 #include <sys/buf.h> 70 #include <sys/mount.h> 71 #include <sys/malloc.h> 72 #include <sys/proc.h> 73 #include <sys/kauth.h> 74 75 #include <ufs/lfs/ulfs_inode.h> 76 #include <ufs/lfs/lfs.h> 77 #include <ufs/lfs/lfs_accessors.h> 78 #include <ufs/lfs/lfs_kernel.h> 79 #include <ufs/lfs/lfs_extern.h> 80 81 #include <uvm/uvm.h> 82 83 #ifdef DEBUG 84 const char *lfs_res_names[LFS_NB_COUNT] = { 85 "summary", 86 "superblock", 87 "file block", 88 "cluster", 89 "clean", 90 "blkiov", 91 }; 92 #endif 93 94 int lfs_res_qty[LFS_NB_COUNT] = { 95 LFS_N_SUMMARIES, 96 LFS_N_SBLOCKS, 97 LFS_N_IBLOCKS, 98 LFS_N_CLUSTERS, 99 LFS_N_CLEAN, 100 LFS_N_BLKIOV, 101 }; 102 103 void 104 lfs_setup_resblks(struct lfs *fs) 105 { 106 int i, j; 107 int maxbpp; 108 109 ASSERT_NO_SEGLOCK(fs); 110 fs->lfs_resblk = malloc(LFS_N_TOTAL * sizeof(res_t), M_SEGMENT, 111 M_WAITOK); 112 for (i = 0; i < LFS_N_TOTAL; i++) { 113 fs->lfs_resblk[i].inuse = 0; 114 fs->lfs_resblk[i].p = NULL; 115 } 116 for (i = 0; i < LFS_RESHASH_WIDTH; i++) 117 LIST_INIT(fs->lfs_reshash + i); 118 119 /* 120 * These types of allocations can be larger than a page, 121 * so we can't use the pool subsystem for them. 122 */ 123 for (i = 0, j = 0; j < LFS_N_SUMMARIES; j++, i++) 124 fs->lfs_resblk[i].size = lfs_sb_getsumsize(fs); 125 for (j = 0; j < LFS_N_SBLOCKS; j++, i++) 126 fs->lfs_resblk[i].size = LFS_SBPAD; 127 for (j = 0; j < LFS_N_IBLOCKS; j++, i++) 128 fs->lfs_resblk[i].size = lfs_sb_getbsize(fs); 129 for (j = 0; j < LFS_N_CLUSTERS; j++, i++) 130 fs->lfs_resblk[i].size = MAXPHYS; 131 for (j = 0; j < LFS_N_CLEAN; j++, i++) 132 fs->lfs_resblk[i].size = MAXPHYS; 133 for (j = 0; j < LFS_N_BLKIOV; j++, i++) 134 fs->lfs_resblk[i].size = LFS_MARKV_MAXBLKCNT * sizeof(BLOCK_INFO); 135 136 for (i = 0; i < LFS_N_TOTAL; i++) { 137 fs->lfs_resblk[i].p = malloc(fs->lfs_resblk[i].size, 138 M_SEGMENT, M_WAITOK); 139 } 140 141 /* 142 * Initialize pools for small types (XXX is BPP small?) 143 */ 144 pool_init(&fs->lfs_clpool, sizeof(struct lfs_cluster), 0, 0, 0, 145 "lfsclpl", &pool_allocator_nointr, IPL_NONE); 146 pool_init(&fs->lfs_segpool, sizeof(struct segment), 0, 0, 0, 147 "lfssegpool", &pool_allocator_nointr, IPL_NONE); 148 /* XXX: should this int32 be 32/64? */ 149 maxbpp = ((lfs_sb_getsumsize(fs) - SEGSUM_SIZE(fs)) / sizeof(int32_t) + 2); 150 maxbpp = MIN(maxbpp, lfs_segsize(fs) / lfs_sb_getfsize(fs) + 2); 151 pool_init(&fs->lfs_bpppool, maxbpp * sizeof(struct buf *), 0, 0, 0, 152 "lfsbpppl", &pool_allocator_nointr, IPL_NONE); 153 } 154 155 void 156 lfs_free_resblks(struct lfs *fs) 157 { 158 int i; 159 160 pool_destroy(&fs->lfs_bpppool); 161 pool_destroy(&fs->lfs_segpool); 162 pool_destroy(&fs->lfs_clpool); 163 164 mutex_enter(&lfs_lock); 165 for (i = 0; i < LFS_N_TOTAL; i++) { 166 while (fs->lfs_resblk[i].inuse) 167 mtsleep(&fs->lfs_resblk, PRIBIO + 1, "lfs_free", 0, 168 &lfs_lock); 169 if (fs->lfs_resblk[i].p != NULL) 170 free(fs->lfs_resblk[i].p, M_SEGMENT); 171 } 172 free(fs->lfs_resblk, M_SEGMENT); 173 mutex_exit(&lfs_lock); 174 } 175 176 static unsigned int 177 lfs_mhash(void *vp) 178 { 179 return (unsigned int)(((unsigned long)vp) >> 2) % LFS_RESHASH_WIDTH; 180 } 181 182 /* 183 * Return memory of the given size for the given purpose, or use one of a 184 * number of spare last-resort buffers, if malloc returns NULL. 185 */ 186 void * 187 lfs_malloc(struct lfs *fs, size_t size, int type) 188 { 189 struct lfs_res_blk *re; 190 void *r; 191 int i, start; 192 unsigned int h; 193 194 ASSERT_MAYBE_SEGLOCK(fs); 195 r = NULL; 196 197 /* If no mem allocated for this type, it just waits */ 198 if (lfs_res_qty[type] == 0) { 199 r = malloc(size, M_SEGMENT, M_WAITOK); 200 return r; 201 } 202 203 /* Otherwise try a quick malloc, and if it works, great */ 204 if ((r = malloc(size, M_SEGMENT, M_NOWAIT)) != NULL) { 205 return r; 206 } 207 208 /* 209 * If malloc returned NULL, we are forced to use one of our 210 * reserve blocks. We have on hand at least one summary block, 211 * at least one cluster block, at least one superblock, 212 * and several indirect blocks. 213 */ 214 215 mutex_enter(&lfs_lock); 216 /* skip over blocks of other types */ 217 for (i = 0, start = 0; i < type; i++) 218 start += lfs_res_qty[i]; 219 while (r == NULL) { 220 for (i = 0; i < lfs_res_qty[type]; i++) { 221 if (fs->lfs_resblk[start + i].inuse == 0) { 222 re = fs->lfs_resblk + start + i; 223 re->inuse = 1; 224 r = re->p; 225 KASSERT(re->size >= size); 226 h = lfs_mhash(r); 227 LIST_INSERT_HEAD(&fs->lfs_reshash[h], re, res); 228 mutex_exit(&lfs_lock); 229 return r; 230 } 231 } 232 DLOG((DLOG_MALLOC, "sleeping on %s (%d)\n", 233 lfs_res_names[type], lfs_res_qty[type])); 234 mtsleep(&fs->lfs_resblk, PVM, "lfs_malloc", 0, 235 &lfs_lock); 236 DLOG((DLOG_MALLOC, "done sleeping on %s\n", 237 lfs_res_names[type])); 238 } 239 /* NOTREACHED */ 240 mutex_exit(&lfs_lock); 241 return r; 242 } 243 244 void 245 lfs_free(struct lfs *fs, void *p, int type) 246 { 247 unsigned int h; 248 res_t *re; 249 250 ASSERT_MAYBE_SEGLOCK(fs); 251 h = lfs_mhash(p); 252 mutex_enter(&lfs_lock); 253 LIST_FOREACH(re, &fs->lfs_reshash[h], res) { 254 if (re->p == p) { 255 KASSERT(re->inuse == 1); 256 LIST_REMOVE(re, res); 257 re->inuse = 0; 258 wakeup(&fs->lfs_resblk); 259 mutex_exit(&lfs_lock); 260 return; 261 } 262 } 263 264 #ifdef notyet /* XXX this assert fires */ 265 for (int i = 0; i < LFS_N_TOTAL; i++) { 266 KDASSERTMSG(fs->lfs_resblk[i].p == p, 267 "lfs_free: inconsistent reserved block"); 268 } 269 #endif 270 271 mutex_exit(&lfs_lock); 272 273 /* 274 * If we didn't find it, free it. 275 */ 276 free(p, M_SEGMENT); 277 } 278 279 /* 280 * lfs_seglock -- 281 * Single thread the segment writer. 282 */ 283 int 284 lfs_seglock(struct lfs *fs, unsigned long flags) 285 { 286 struct segment *sp; 287 288 mutex_enter(&lfs_lock); 289 if (fs->lfs_seglock) { 290 if (fs->lfs_lockpid == curproc->p_pid && 291 fs->lfs_locklwp == curlwp->l_lid) { 292 ++fs->lfs_seglock; 293 fs->lfs_sp->seg_flags |= flags; 294 mutex_exit(&lfs_lock); 295 return 0; 296 } else if (flags & SEGM_PAGEDAEMON) { 297 mutex_exit(&lfs_lock); 298 return EWOULDBLOCK; 299 } else { 300 while (fs->lfs_seglock) { 301 (void)mtsleep(&fs->lfs_seglock, PRIBIO + 1, 302 "lfs_seglock", 0, &lfs_lock); 303 } 304 } 305 } 306 307 fs->lfs_seglock = 1; 308 fs->lfs_lockpid = curproc->p_pid; 309 fs->lfs_locklwp = curlwp->l_lid; 310 mutex_exit(&lfs_lock); 311 fs->lfs_cleanind = 0; 312 313 LFS_ENTER_LOG("seglock", __FILE__, __LINE__, 0, flags, curproc->p_pid); 314 315 /* Drain fragment size changes out */ 316 rw_enter(&fs->lfs_fraglock, RW_WRITER); 317 318 sp = fs->lfs_sp = pool_get(&fs->lfs_segpool, PR_WAITOK); 319 sp->bpp = pool_get(&fs->lfs_bpppool, PR_WAITOK); 320 sp->seg_flags = flags; 321 sp->vp = NULL; 322 sp->seg_iocount = 0; 323 (void) lfs_initseg(fs); 324 325 /* 326 * Keep a cumulative count of the outstanding I/O operations. If the 327 * disk drive catches up with us it could go to zero before we finish, 328 * so we artificially increment it by one until we've scheduled all of 329 * the writes we intend to do. 330 */ 331 mutex_enter(&lfs_lock); 332 ++fs->lfs_iocount; 333 fs->lfs_startseg = lfs_sb_getcurseg(fs); 334 mutex_exit(&lfs_lock); 335 return 0; 336 } 337 338 static void lfs_unmark_dirop(struct lfs *); 339 340 static void 341 lfs_unmark_dirop(struct lfs *fs) 342 { 343 struct inode *ip, *marker; 344 struct vnode *vp; 345 int doit; 346 347 ASSERT_NO_SEGLOCK(fs); 348 mutex_enter(&lfs_lock); 349 doit = !(fs->lfs_flags & LFS_UNDIROP); 350 if (doit) 351 fs->lfs_flags |= LFS_UNDIROP; 352 mutex_exit(&lfs_lock); 353 354 if (!doit) 355 return; 356 357 marker = pool_get(&lfs_inode_pool, PR_WAITOK); 358 KASSERT(fs != NULL); 359 memset(marker, 0, sizeof(*marker)); 360 marker->inode_ext.lfs = pool_get(&lfs_inoext_pool, PR_WAITOK); 361 memset(marker->inode_ext.lfs, 0, sizeof(*marker->inode_ext.lfs)); 362 marker->i_state |= IN_MARKER; 363 364 mutex_enter(&lfs_lock); 365 TAILQ_INSERT_HEAD(&fs->lfs_dchainhd, marker, i_lfs_dchain); 366 while ((ip = TAILQ_NEXT(marker, i_lfs_dchain)) != NULL) { 367 TAILQ_REMOVE(&fs->lfs_dchainhd, marker, i_lfs_dchain); 368 TAILQ_INSERT_AFTER(&fs->lfs_dchainhd, ip, marker, 369 i_lfs_dchain); 370 if (ip->i_state & IN_MARKER) 371 continue; 372 vp = ITOV(ip); 373 if ((ip->i_state & (IN_ADIROP | IN_CDIROP)) == IN_CDIROP) { 374 --lfs_dirvcount; 375 --fs->lfs_dirvcount; 376 vp->v_uflag &= ~VU_DIROP; 377 TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain); 378 wakeup(&lfs_dirvcount); 379 fs->lfs_unlockvp = vp; 380 mutex_exit(&lfs_lock); 381 vrele(vp); 382 mutex_enter(&lfs_lock); 383 fs->lfs_unlockvp = NULL; 384 ip->i_state &= ~IN_CDIROP; 385 } 386 } 387 TAILQ_REMOVE(&fs->lfs_dchainhd, marker, i_lfs_dchain); 388 fs->lfs_flags &= ~LFS_UNDIROP; 389 wakeup(&fs->lfs_flags); 390 mutex_exit(&lfs_lock); 391 392 pool_put(&lfs_inoext_pool, marker->inode_ext.lfs); 393 pool_put(&lfs_inode_pool, marker); 394 } 395 396 static void 397 lfs_auto_segclean(struct lfs *fs) 398 { 399 int i, error, waited; 400 401 ASSERT_SEGLOCK(fs); 402 /* 403 * Now that we've swapped lfs_activesb, but while we still 404 * hold the segment lock, run through the segment list marking 405 * the empty ones clean. 406 * XXX - do we really need to do them all at once? 407 */ 408 waited = 0; 409 for (i = 0; i < lfs_sb_getnseg(fs); i++) { 410 if ((fs->lfs_suflags[0][i] & 411 (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) == 412 (SEGUSE_DIRTY | SEGUSE_EMPTY) && 413 (fs->lfs_suflags[1][i] & 414 (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) == 415 (SEGUSE_DIRTY | SEGUSE_EMPTY)) { 416 417 /* Make sure the sb is written before we clean */ 418 mutex_enter(&lfs_lock); 419 while (waited == 0 && fs->lfs_sbactive) 420 mtsleep(&fs->lfs_sbactive, PRIBIO+1, "lfs asb", 421 0, &lfs_lock); 422 mutex_exit(&lfs_lock); 423 waited = 1; 424 425 if ((error = lfs_do_segclean(fs, i)) != 0) { 426 DLOG((DLOG_CLEAN, "lfs_auto_segclean: lfs_do_segclean returned %d for seg %d\n", error, i)); 427 } 428 } 429 fs->lfs_suflags[1 - fs->lfs_activesb][i] = 430 fs->lfs_suflags[fs->lfs_activesb][i]; 431 } 432 } 433 434 /* 435 * lfs_segunlock -- 436 * Single thread the segment writer. 437 */ 438 void 439 lfs_segunlock(struct lfs *fs) 440 { 441 struct segment *sp; 442 unsigned long sync, ckp; 443 struct buf *bp; 444 int do_unmark_dirop = 0; 445 446 sp = fs->lfs_sp; 447 448 mutex_enter(&lfs_lock); 449 450 if (!LFS_SEGLOCK_HELD(fs)) 451 panic("lfs seglock not held"); 452 453 if (fs->lfs_seglock == 1) { 454 if ((sp->seg_flags & (SEGM_PROT | SEGM_CLEAN)) == 0) 455 do_unmark_dirop = 1; 456 mutex_exit(&lfs_lock); 457 sync = sp->seg_flags & SEGM_SYNC; 458 ckp = sp->seg_flags & SEGM_CKP; 459 460 /* We should have a segment summary, and nothing else */ 461 KASSERT(sp->cbpp == sp->bpp + 1); 462 463 /* Free allocated segment summary */ 464 lfs_sb_suboffset(fs, lfs_btofsb(fs, lfs_sb_getsumsize(fs))); 465 bp = *sp->bpp; 466 lfs_freebuf(fs, bp); 467 468 pool_put(&fs->lfs_bpppool, sp->bpp); 469 sp->bpp = NULL; 470 471 /* 472 * If we're not sync, we're done with sp, get rid of it. 473 * Otherwise, we keep a local copy around but free 474 * fs->lfs_sp so another process can use it (we have to 475 * wait but they don't have to wait for us). 476 */ 477 if (!sync) 478 pool_put(&fs->lfs_segpool, sp); 479 fs->lfs_sp = NULL; 480 481 /* 482 * If the I/O count is non-zero, sleep until it reaches zero. 483 * At the moment, the user's process hangs around so we can 484 * sleep. 485 */ 486 mutex_enter(&lfs_lock); 487 if (--fs->lfs_iocount <= 1) 488 wakeup(&fs->lfs_iocount); 489 mutex_exit(&lfs_lock); 490 491 /* 492 * If we're not checkpointing, we don't have to block 493 * other processes to wait for a synchronous write 494 * to complete. 495 */ 496 if (!ckp) { 497 LFS_ENTER_LOG("segunlock_std", __FILE__, __LINE__, 0, 0, curproc->p_pid); 498 499 mutex_enter(&lfs_lock); 500 --fs->lfs_seglock; 501 fs->lfs_lockpid = 0; 502 fs->lfs_locklwp = 0; 503 mutex_exit(&lfs_lock); 504 wakeup(&fs->lfs_seglock); 505 } 506 /* 507 * We let checkpoints happen asynchronously. That means 508 * that during recovery, we have to roll forward between 509 * the two segments described by the first and second 510 * superblocks to make sure that the checkpoint described 511 * by a superblock completed. 512 */ 513 mutex_enter(&lfs_lock); 514 while (ckp && sync && fs->lfs_iocount) { 515 (void)mtsleep(&fs->lfs_iocount, PRIBIO + 1, 516 "lfs_iocount", 0, &lfs_lock); 517 DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", fs, fs->lfs_iocount)); 518 } 519 while (sync && sp->seg_iocount) { 520 (void)mtsleep(&sp->seg_iocount, PRIBIO + 1, 521 "seg_iocount", 0, &lfs_lock); 522 DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n", sp, sp->seg_iocount)); 523 } 524 mutex_exit(&lfs_lock); 525 if (sync) 526 pool_put(&fs->lfs_segpool, sp); 527 528 if (ckp) { 529 fs->lfs_nactive = 0; 530 /* If we *know* everything's on disk, write both sbs */ 531 /* XXX should wait for this one */ 532 if (sync) 533 lfs_writesuper(fs, lfs_sb_getsboff(fs, fs->lfs_activesb)); 534 lfs_writesuper(fs, lfs_sb_getsboff(fs, 1 - fs->lfs_activesb)); 535 if (!(fs->lfs_ivnode->v_mount->mnt_iflag & IMNT_UNMOUNT)) { 536 lfs_auto_segclean(fs); 537 /* If sync, we can clean the remainder too */ 538 if (sync) 539 lfs_auto_segclean(fs); 540 } 541 fs->lfs_activesb = 1 - fs->lfs_activesb; 542 543 LFS_ENTER_LOG("segunlock_ckp", __FILE__, __LINE__, 0, 0, curproc->p_pid); 544 545 mutex_enter(&lfs_lock); 546 --fs->lfs_seglock; 547 fs->lfs_lockpid = 0; 548 fs->lfs_locklwp = 0; 549 mutex_exit(&lfs_lock); 550 wakeup(&fs->lfs_seglock); 551 } 552 /* Reenable fragment size changes */ 553 rw_exit(&fs->lfs_fraglock); 554 if (do_unmark_dirop) 555 lfs_unmark_dirop(fs); 556 } else { 557 --fs->lfs_seglock; 558 KASSERT(fs->lfs_seglock != 0); 559 mutex_exit(&lfs_lock); 560 } 561 } 562 563 /* 564 * Drain dirops and start writer. 565 * 566 * No simple_locks are held when we enter and none are held when we return. 567 */ 568 void 569 lfs_writer_enter(struct lfs *fs, const char *wmesg) 570 { 571 int error __diagused; 572 573 ASSERT_NO_SEGLOCK(fs); 574 mutex_enter(&lfs_lock); 575 576 /* disallow dirops during flush */ 577 fs->lfs_writer++; 578 579 while (fs->lfs_dirops > 0) { 580 ++fs->lfs_diropwait; 581 error = mtsleep(&fs->lfs_writer, PRIBIO+1, wmesg, 0, 582 &lfs_lock); 583 KASSERT(error == 0); 584 --fs->lfs_diropwait; 585 } 586 587 mutex_exit(&lfs_lock); 588 } 589 590 int 591 lfs_writer_tryenter(struct lfs *fs) 592 { 593 int writer_set; 594 595 ASSERT_MAYBE_SEGLOCK(fs); 596 mutex_enter(&lfs_lock); 597 writer_set = (fs->lfs_dirops == 0); 598 if (writer_set) 599 fs->lfs_writer++; 600 mutex_exit(&lfs_lock); 601 602 return writer_set; 603 } 604 605 void 606 lfs_writer_leave(struct lfs *fs) 607 { 608 bool dowakeup; 609 610 ASSERT_MAYBE_SEGLOCK(fs); 611 mutex_enter(&lfs_lock); 612 dowakeup = !(--fs->lfs_writer); 613 if (dowakeup) 614 cv_broadcast(&fs->lfs_diropscv); 615 mutex_exit(&lfs_lock); 616 } 617 618 /* 619 * Unlock, wait for the cleaner, then relock to where we were before. 620 * To be used only at a fairly high level, to address a paucity of free 621 * segments propagated back from lfs_gop_write(). 622 */ 623 void 624 lfs_segunlock_relock(struct lfs *fs) 625 { 626 int n = fs->lfs_seglock; 627 u_int16_t seg_flags; 628 CLEANERINFO *cip; 629 struct buf *bp; 630 631 if (n == 0) 632 return; 633 634 /* Write anything we've already gathered to disk */ 635 lfs_writeseg(fs, fs->lfs_sp); 636 637 /* Tell cleaner */ 638 LFS_CLEANERINFO(cip, fs, bp); 639 lfs_ci_setflags(fs, cip, 640 lfs_ci_getflags(fs, cip) | LFS_CLEANER_MUST_CLEAN); 641 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 642 643 /* Save segment flags for later */ 644 seg_flags = fs->lfs_sp->seg_flags; 645 646 fs->lfs_sp->seg_flags |= SEGM_PROT; /* Don't unmark dirop nodes */ 647 while(fs->lfs_seglock) 648 lfs_segunlock(fs); 649 650 /* Wait for the cleaner */ 651 lfs_wakeup_cleaner(fs); 652 mutex_enter(&lfs_lock); 653 while (LFS_STARVED_FOR_SEGS(fs)) 654 mtsleep(&fs->lfs_availsleep, PRIBIO, "relock", 0, 655 &lfs_lock); 656 mutex_exit(&lfs_lock); 657 658 /* Put the segment lock back the way it was. */ 659 while(n--) 660 lfs_seglock(fs, seg_flags); 661 662 /* Cleaner can relax now */ 663 LFS_CLEANERINFO(cip, fs, bp); 664 lfs_ci_setflags(fs, cip, 665 lfs_ci_getflags(fs, cip) & ~LFS_CLEANER_MUST_CLEAN); 666 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 667 668 return; 669 } 670 671 /* 672 * Wake up the cleaner, provided that nowrap is not set. 673 */ 674 void 675 lfs_wakeup_cleaner(struct lfs *fs) 676 { 677 if (fs->lfs_nowrap > 0) 678 return; 679 680 cv_broadcast(&fs->lfs_nextsegsleep); 681 cv_broadcast(&lfs_allclean_wakeup); 682 } 683