| 1 | /* $NetBSD: lfs_subr.c,v 1.86 2015/10/03 08:28:16 dholland 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 1.86 2015/10/03 08:28:16 dholland 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, s, 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 | s = splbio(); |
| 228 | LIST_INSERT_HEAD(&fs->lfs_reshash[h], re, res); |
| 229 | splx(s); |
| 230 | mutex_exit(&lfs_lock); |
| 231 | return r; |
| 232 | } |
| 233 | } |
| 234 | DLOG((DLOG_MALLOC, "sleeping on %s (%d)\n" , |
| 235 | lfs_res_names[type], lfs_res_qty[type])); |
| 236 | mtsleep(&fs->lfs_resblk, PVM, "lfs_malloc" , 0, |
| 237 | &lfs_lock); |
| 238 | DLOG((DLOG_MALLOC, "done sleeping on %s\n" , |
| 239 | lfs_res_names[type])); |
| 240 | } |
| 241 | /* NOTREACHED */ |
| 242 | mutex_exit(&lfs_lock); |
| 243 | return r; |
| 244 | } |
| 245 | |
| 246 | void |
| 247 | lfs_free(struct lfs *fs, void *p, int type) |
| 248 | { |
| 249 | int s; |
| 250 | unsigned int h; |
| 251 | res_t *re; |
| 252 | #ifdef DEBUG |
| 253 | int i; |
| 254 | #endif |
| 255 | |
| 256 | ASSERT_MAYBE_SEGLOCK(fs); |
| 257 | h = lfs_mhash(p); |
| 258 | mutex_enter(&lfs_lock); |
| 259 | s = splbio(); |
| 260 | LIST_FOREACH(re, &fs->lfs_reshash[h], res) { |
| 261 | if (re->p == p) { |
| 262 | KASSERT(re->inuse == 1); |
| 263 | LIST_REMOVE(re, res); |
| 264 | re->inuse = 0; |
| 265 | wakeup(&fs->lfs_resblk); |
| 266 | splx(s); |
| 267 | mutex_exit(&lfs_lock); |
| 268 | return; |
| 269 | } |
| 270 | } |
| 271 | #ifdef DEBUG |
| 272 | for (i = 0; i < LFS_N_TOTAL; i++) { |
| 273 | if (fs->lfs_resblk[i].p == p) |
| 274 | panic("lfs_free: inconsistent reserved block" ); |
| 275 | } |
| 276 | #endif |
| 277 | splx(s); |
| 278 | mutex_exit(&lfs_lock); |
| 279 | |
| 280 | /* |
| 281 | * If we didn't find it, free it. |
| 282 | */ |
| 283 | free(p, M_SEGMENT); |
| 284 | } |
| 285 | |
| 286 | /* |
| 287 | * lfs_seglock -- |
| 288 | * Single thread the segment writer. |
| 289 | */ |
| 290 | int |
| 291 | lfs_seglock(struct lfs *fs, unsigned long flags) |
| 292 | { |
| 293 | struct segment *sp; |
| 294 | |
| 295 | mutex_enter(&lfs_lock); |
| 296 | if (fs->lfs_seglock) { |
| 297 | if (fs->lfs_lockpid == curproc->p_pid && |
| 298 | fs->lfs_locklwp == curlwp->l_lid) { |
| 299 | ++fs->lfs_seglock; |
| 300 | fs->lfs_sp->seg_flags |= flags; |
| 301 | mutex_exit(&lfs_lock); |
| 302 | return 0; |
| 303 | } else if (flags & SEGM_PAGEDAEMON) { |
| 304 | mutex_exit(&lfs_lock); |
| 305 | return EWOULDBLOCK; |
| 306 | } else { |
| 307 | while (fs->lfs_seglock) { |
| 308 | (void)mtsleep(&fs->lfs_seglock, PRIBIO + 1, |
| 309 | "lfs_seglock" , 0, &lfs_lock); |
| 310 | } |
| 311 | } |
| 312 | } |
| 313 | |
| 314 | fs->lfs_seglock = 1; |
| 315 | fs->lfs_lockpid = curproc->p_pid; |
| 316 | fs->lfs_locklwp = curlwp->l_lid; |
| 317 | mutex_exit(&lfs_lock); |
| 318 | fs->lfs_cleanind = 0; |
| 319 | |
| 320 | #ifdef DEBUG |
| 321 | LFS_ENTER_LOG("seglock" , __FILE__, __LINE__, 0, flags, curproc->p_pid); |
| 322 | #endif |
| 323 | /* Drain fragment size changes out */ |
| 324 | rw_enter(&fs->lfs_fraglock, RW_WRITER); |
| 325 | |
| 326 | sp = fs->lfs_sp = pool_get(&fs->lfs_segpool, PR_WAITOK); |
| 327 | sp->bpp = pool_get(&fs->lfs_bpppool, PR_WAITOK); |
| 328 | sp->seg_flags = flags; |
| 329 | sp->vp = NULL; |
| 330 | sp->seg_iocount = 0; |
| 331 | (void) lfs_initseg(fs); |
| 332 | |
| 333 | /* |
| 334 | * Keep a cumulative count of the outstanding I/O operations. If the |
| 335 | * disk drive catches up with us it could go to zero before we finish, |
| 336 | * so we artificially increment it by one until we've scheduled all of |
| 337 | * the writes we intend to do. |
| 338 | */ |
| 339 | mutex_enter(&lfs_lock); |
| 340 | ++fs->lfs_iocount; |
| 341 | fs->lfs_startseg = lfs_sb_getcurseg(fs); |
| 342 | mutex_exit(&lfs_lock); |
| 343 | return 0; |
| 344 | } |
| 345 | |
| 346 | static void lfs_unmark_dirop(struct lfs *); |
| 347 | |
| 348 | static void |
| 349 | lfs_unmark_dirop(struct lfs *fs) |
| 350 | { |
| 351 | struct inode *ip, *nip; |
| 352 | struct vnode *vp; |
| 353 | int doit; |
| 354 | |
| 355 | ASSERT_NO_SEGLOCK(fs); |
| 356 | mutex_enter(&lfs_lock); |
| 357 | doit = !(fs->lfs_flags & LFS_UNDIROP); |
| 358 | if (doit) |
| 359 | fs->lfs_flags |= LFS_UNDIROP; |
| 360 | if (!doit) { |
| 361 | mutex_exit(&lfs_lock); |
| 362 | return; |
| 363 | } |
| 364 | |
| 365 | for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { |
| 366 | nip = TAILQ_NEXT(ip, i_lfs_dchain); |
| 367 | vp = ITOV(ip); |
| 368 | if ((ip->i_flag & (IN_ADIROP | IN_CDIROP)) == IN_CDIROP) { |
| 369 | --lfs_dirvcount; |
| 370 | --fs->lfs_dirvcount; |
| 371 | vp->v_uflag &= ~VU_DIROP; |
| 372 | TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain); |
| 373 | wakeup(&lfs_dirvcount); |
| 374 | fs->lfs_unlockvp = vp; |
| 375 | mutex_exit(&lfs_lock); |
| 376 | vrele(vp); |
| 377 | mutex_enter(&lfs_lock); |
| 378 | fs->lfs_unlockvp = NULL; |
| 379 | ip->i_flag &= ~IN_CDIROP; |
| 380 | } |
| 381 | } |
| 382 | |
| 383 | fs->lfs_flags &= ~LFS_UNDIROP; |
| 384 | wakeup(&fs->lfs_flags); |
| 385 | mutex_exit(&lfs_lock); |
| 386 | } |
| 387 | |
| 388 | static void |
| 389 | lfs_auto_segclean(struct lfs *fs) |
| 390 | { |
| 391 | int i, error, s, waited; |
| 392 | |
| 393 | ASSERT_SEGLOCK(fs); |
| 394 | /* |
| 395 | * Now that we've swapped lfs_activesb, but while we still |
| 396 | * hold the segment lock, run through the segment list marking |
| 397 | * the empty ones clean. |
| 398 | * XXX - do we really need to do them all at once? |
| 399 | */ |
| 400 | waited = 0; |
| 401 | for (i = 0; i < lfs_sb_getnseg(fs); i++) { |
| 402 | if ((fs->lfs_suflags[0][i] & |
| 403 | (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) == |
| 404 | (SEGUSE_DIRTY | SEGUSE_EMPTY) && |
| 405 | (fs->lfs_suflags[1][i] & |
| 406 | (SEGUSE_ACTIVE | SEGUSE_DIRTY | SEGUSE_EMPTY)) == |
| 407 | (SEGUSE_DIRTY | SEGUSE_EMPTY)) { |
| 408 | |
| 409 | /* Make sure the sb is written before we clean */ |
| 410 | mutex_enter(&lfs_lock); |
| 411 | s = splbio(); |
| 412 | while (waited == 0 && fs->lfs_sbactive) |
| 413 | mtsleep(&fs->lfs_sbactive, PRIBIO+1, "lfs asb" , |
| 414 | 0, &lfs_lock); |
| 415 | splx(s); |
| 416 | mutex_exit(&lfs_lock); |
| 417 | waited = 1; |
| 418 | |
| 419 | if ((error = lfs_do_segclean(fs, i)) != 0) { |
| 420 | DLOG((DLOG_CLEAN, "lfs_auto_segclean: lfs_do_segclean returned %d for seg %d\n" , error, i)); |
| 421 | } |
| 422 | } |
| 423 | fs->lfs_suflags[1 - fs->lfs_activesb][i] = |
| 424 | fs->lfs_suflags[fs->lfs_activesb][i]; |
| 425 | } |
| 426 | } |
| 427 | |
| 428 | /* |
| 429 | * lfs_segunlock -- |
| 430 | * Single thread the segment writer. |
| 431 | */ |
| 432 | void |
| 433 | lfs_segunlock(struct lfs *fs) |
| 434 | { |
| 435 | struct segment *sp; |
| 436 | unsigned long sync, ckp; |
| 437 | struct buf *bp; |
| 438 | int do_unmark_dirop = 0; |
| 439 | |
| 440 | sp = fs->lfs_sp; |
| 441 | |
| 442 | mutex_enter(&lfs_lock); |
| 443 | KASSERT(LFS_SEGLOCK_HELD(fs)); |
| 444 | if (fs->lfs_seglock == 1) { |
| 445 | if ((sp->seg_flags & (SEGM_PROT | SEGM_CLEAN)) == 0) |
| 446 | do_unmark_dirop = 1; |
| 447 | mutex_exit(&lfs_lock); |
| 448 | sync = sp->seg_flags & SEGM_SYNC; |
| 449 | ckp = sp->seg_flags & SEGM_CKP; |
| 450 | |
| 451 | /* We should have a segment summary, and nothing else */ |
| 452 | KASSERT(sp->cbpp == sp->bpp + 1); |
| 453 | |
| 454 | /* Free allocated segment summary */ |
| 455 | lfs_sb_suboffset(fs, lfs_btofsb(fs, lfs_sb_getsumsize(fs))); |
| 456 | bp = *sp->bpp; |
| 457 | lfs_freebuf(fs, bp); |
| 458 | |
| 459 | pool_put(&fs->lfs_bpppool, sp->bpp); |
| 460 | sp->bpp = NULL; |
| 461 | |
| 462 | /* |
| 463 | * If we're not sync, we're done with sp, get rid of it. |
| 464 | * Otherwise, we keep a local copy around but free |
| 465 | * fs->lfs_sp so another process can use it (we have to |
| 466 | * wait but they don't have to wait for us). |
| 467 | */ |
| 468 | if (!sync) |
| 469 | pool_put(&fs->lfs_segpool, sp); |
| 470 | fs->lfs_sp = NULL; |
| 471 | |
| 472 | /* |
| 473 | * If the I/O count is non-zero, sleep until it reaches zero. |
| 474 | * At the moment, the user's process hangs around so we can |
| 475 | * sleep. |
| 476 | */ |
| 477 | mutex_enter(&lfs_lock); |
| 478 | if (--fs->lfs_iocount == 0) { |
| 479 | LFS_DEBUG_COUNTLOCKED("lfs_segunlock" ); |
| 480 | } |
| 481 | if (fs->lfs_iocount <= 1) |
| 482 | wakeup(&fs->lfs_iocount); |
| 483 | mutex_exit(&lfs_lock); |
| 484 | /* |
| 485 | * If we're not checkpointing, we don't have to block |
| 486 | * other processes to wait for a synchronous write |
| 487 | * to complete. |
| 488 | */ |
| 489 | if (!ckp) { |
| 490 | #ifdef DEBUG |
| 491 | LFS_ENTER_LOG("segunlock_std" , __FILE__, __LINE__, 0, 0, curproc->p_pid); |
| 492 | #endif |
| 493 | mutex_enter(&lfs_lock); |
| 494 | --fs->lfs_seglock; |
| 495 | fs->lfs_lockpid = 0; |
| 496 | fs->lfs_locklwp = 0; |
| 497 | mutex_exit(&lfs_lock); |
| 498 | wakeup(&fs->lfs_seglock); |
| 499 | } |
| 500 | /* |
| 501 | * We let checkpoints happen asynchronously. That means |
| 502 | * that during recovery, we have to roll forward between |
| 503 | * the two segments described by the first and second |
| 504 | * superblocks to make sure that the checkpoint described |
| 505 | * by a superblock completed. |
| 506 | */ |
| 507 | mutex_enter(&lfs_lock); |
| 508 | while (ckp && sync && fs->lfs_iocount) { |
| 509 | (void)mtsleep(&fs->lfs_iocount, PRIBIO + 1, |
| 510 | "lfs_iocount" , 0, &lfs_lock); |
| 511 | DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n" , fs, fs->lfs_iocount)); |
| 512 | } |
| 513 | while (sync && sp->seg_iocount) { |
| 514 | (void)mtsleep(&sp->seg_iocount, PRIBIO + 1, |
| 515 | "seg_iocount" , 0, &lfs_lock); |
| 516 | DLOG((DLOG_SEG, "sleeping on iocount %x == %d\n" , sp, sp->seg_iocount)); |
| 517 | } |
| 518 | mutex_exit(&lfs_lock); |
| 519 | if (sync) |
| 520 | pool_put(&fs->lfs_segpool, sp); |
| 521 | |
| 522 | if (ckp) { |
| 523 | fs->lfs_nactive = 0; |
| 524 | /* If we *know* everything's on disk, write both sbs */ |
| 525 | /* XXX should wait for this one */ |
| 526 | if (sync) |
| 527 | lfs_writesuper(fs, lfs_sb_getsboff(fs, fs->lfs_activesb)); |
| 528 | lfs_writesuper(fs, lfs_sb_getsboff(fs, 1 - fs->lfs_activesb)); |
| 529 | if (!(fs->lfs_ivnode->v_mount->mnt_iflag & IMNT_UNMOUNT)) { |
| 530 | lfs_auto_segclean(fs); |
| 531 | /* If sync, we can clean the remainder too */ |
| 532 | if (sync) |
| 533 | lfs_auto_segclean(fs); |
| 534 | } |
| 535 | fs->lfs_activesb = 1 - fs->lfs_activesb; |
| 536 | #ifdef DEBUG |
| 537 | LFS_ENTER_LOG("segunlock_ckp" , __FILE__, __LINE__, 0, 0, curproc->p_pid); |
| 538 | #endif |
| 539 | mutex_enter(&lfs_lock); |
| 540 | --fs->lfs_seglock; |
| 541 | fs->lfs_lockpid = 0; |
| 542 | fs->lfs_locklwp = 0; |
| 543 | mutex_exit(&lfs_lock); |
| 544 | wakeup(&fs->lfs_seglock); |
| 545 | } |
| 546 | /* Reenable fragment size changes */ |
| 547 | rw_exit(&fs->lfs_fraglock); |
| 548 | if (do_unmark_dirop) |
| 549 | lfs_unmark_dirop(fs); |
| 550 | } else if (fs->lfs_seglock == 0) { |
| 551 | mutex_exit(&lfs_lock); |
| 552 | panic ("Seglock not held" ); |
| 553 | } else { |
| 554 | --fs->lfs_seglock; |
| 555 | mutex_exit(&lfs_lock); |
| 556 | } |
| 557 | } |
| 558 | |
| 559 | /* |
| 560 | * Drain dirops and start writer. |
| 561 | * |
| 562 | * No simple_locks are held when we enter and none are held when we return. |
| 563 | */ |
| 564 | int |
| 565 | lfs_writer_enter(struct lfs *fs, const char *wmesg) |
| 566 | { |
| 567 | int error = 0; |
| 568 | |
| 569 | ASSERT_MAYBE_SEGLOCK(fs); |
| 570 | mutex_enter(&lfs_lock); |
| 571 | |
| 572 | /* disallow dirops during flush */ |
| 573 | fs->lfs_writer++; |
| 574 | |
| 575 | while (fs->lfs_dirops > 0) { |
| 576 | ++fs->lfs_diropwait; |
| 577 | error = mtsleep(&fs->lfs_writer, PRIBIO+1, wmesg, 0, |
| 578 | &lfs_lock); |
| 579 | --fs->lfs_diropwait; |
| 580 | } |
| 581 | |
| 582 | if (error) |
| 583 | fs->lfs_writer--; |
| 584 | |
| 585 | mutex_exit(&lfs_lock); |
| 586 | |
| 587 | return error; |
| 588 | } |
| 589 | |
| 590 | void |
| 591 | lfs_writer_leave(struct lfs *fs) |
| 592 | { |
| 593 | bool dowakeup; |
| 594 | |
| 595 | ASSERT_MAYBE_SEGLOCK(fs); |
| 596 | mutex_enter(&lfs_lock); |
| 597 | dowakeup = !(--fs->lfs_writer); |
| 598 | mutex_exit(&lfs_lock); |
| 599 | if (dowakeup) |
| 600 | wakeup(&fs->lfs_dirops); |
| 601 | } |
| 602 | |
| 603 | /* |
| 604 | * Unlock, wait for the cleaner, then relock to where we were before. |
| 605 | * To be used only at a fairly high level, to address a paucity of free |
| 606 | * segments propagated back from lfs_gop_write(). |
| 607 | */ |
| 608 | void |
| 609 | lfs_segunlock_relock(struct lfs *fs) |
| 610 | { |
| 611 | int n = fs->lfs_seglock; |
| 612 | u_int16_t seg_flags; |
| 613 | CLEANERINFO *cip; |
| 614 | struct buf *bp; |
| 615 | |
| 616 | if (n == 0) |
| 617 | return; |
| 618 | |
| 619 | /* Write anything we've already gathered to disk */ |
| 620 | lfs_writeseg(fs, fs->lfs_sp); |
| 621 | |
| 622 | /* Tell cleaner */ |
| 623 | LFS_CLEANERINFO(cip, fs, bp); |
| 624 | lfs_ci_setflags(fs, cip, |
| 625 | lfs_ci_getflags(fs, cip) | LFS_CLEANER_MUST_CLEAN); |
| 626 | LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); |
| 627 | |
| 628 | /* Save segment flags for later */ |
| 629 | seg_flags = fs->lfs_sp->seg_flags; |
| 630 | |
| 631 | fs->lfs_sp->seg_flags |= SEGM_PROT; /* Don't unmark dirop nodes */ |
| 632 | while(fs->lfs_seglock) |
| 633 | lfs_segunlock(fs); |
| 634 | |
| 635 | /* Wait for the cleaner */ |
| 636 | lfs_wakeup_cleaner(fs); |
| 637 | mutex_enter(&lfs_lock); |
| 638 | while (LFS_STARVED_FOR_SEGS(fs)) |
| 639 | mtsleep(&fs->lfs_availsleep, PRIBIO, "relock" , 0, |
| 640 | &lfs_lock); |
| 641 | mutex_exit(&lfs_lock); |
| 642 | |
| 643 | /* Put the segment lock back the way it was. */ |
| 644 | while(n--) |
| 645 | lfs_seglock(fs, seg_flags); |
| 646 | |
| 647 | /* Cleaner can relax now */ |
| 648 | LFS_CLEANERINFO(cip, fs, bp); |
| 649 | lfs_ci_setflags(fs, cip, |
| 650 | lfs_ci_getflags(fs, cip) & ~LFS_CLEANER_MUST_CLEAN); |
| 651 | LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); |
| 652 | |
| 653 | return; |
| 654 | } |
| 655 | |
| 656 | /* |
| 657 | * Wake up the cleaner, provided that nowrap is not set. |
| 658 | */ |
| 659 | void |
| 660 | lfs_wakeup_cleaner(struct lfs *fs) |
| 661 | { |
| 662 | if (fs->lfs_nowrap > 0) |
| 663 | return; |
| 664 | |
| 665 | wakeup(&fs->lfs_nextsegsleep); |
| 666 | wakeup(&lfs_allclean_wakeup); |
| 667 | } |
| 668 | |