| 1 | /* $NetBSD: lfs_bio.c,v 1.135 2015/10/03 09:31:29 hannken Exp $ */ |
| 2 | |
| 3 | /*- |
| 4 | * Copyright (c) 1999, 2000, 2001, 2002, 2003, 2008 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_bio.c 8.10 (Berkeley) 6/10/95 |
| 60 | */ |
| 61 | |
| 62 | #include <sys/cdefs.h> |
| 63 | __KERNEL_RCSID(0, "$NetBSD: lfs_bio.c,v 1.135 2015/10/03 09:31:29 hannken Exp $" ); |
| 64 | |
| 65 | #include <sys/param.h> |
| 66 | #include <sys/systm.h> |
| 67 | #include <sys/proc.h> |
| 68 | #include <sys/buf.h> |
| 69 | #include <sys/vnode.h> |
| 70 | #include <sys/resourcevar.h> |
| 71 | #include <sys/mount.h> |
| 72 | #include <sys/kernel.h> |
| 73 | #include <sys/kauth.h> |
| 74 | |
| 75 | #include <ufs/lfs/ulfs_inode.h> |
| 76 | #include <ufs/lfs/ulfsmount.h> |
| 77 | #include <ufs/lfs/ulfs_extern.h> |
| 78 | |
| 79 | #include <ufs/lfs/lfs.h> |
| 80 | #include <ufs/lfs/lfs_accessors.h> |
| 81 | #include <ufs/lfs/lfs_extern.h> |
| 82 | #include <ufs/lfs/lfs_kernel.h> |
| 83 | |
| 84 | #include <uvm/uvm.h> |
| 85 | |
| 86 | /* |
| 87 | * LFS block write function. |
| 88 | * |
| 89 | * XXX |
| 90 | * No write cost accounting is done. |
| 91 | * This is almost certainly wrong for synchronous operations and NFS. |
| 92 | * |
| 93 | * protected by lfs_lock. |
| 94 | */ |
| 95 | int locked_queue_count = 0; /* Count of locked-down buffers. */ |
| 96 | long locked_queue_bytes = 0L; /* Total size of locked buffers. */ |
| 97 | int lfs_subsys_pages = 0L; /* Total number LFS-written pages */ |
| 98 | int lfs_fs_pagetrip = 0; /* # of pages to trip per-fs write */ |
| 99 | int lfs_writing = 0; /* Set if already kicked off a writer |
| 100 | because of buffer space */ |
| 101 | int locked_queue_waiters = 0; /* Number of processes waiting on lq */ |
| 102 | |
| 103 | /* Lock and condition variables for above. */ |
| 104 | kcondvar_t locked_queue_cv; |
| 105 | kcondvar_t lfs_writing_cv; |
| 106 | kmutex_t lfs_lock; |
| 107 | |
| 108 | extern int lfs_dostats; |
| 109 | |
| 110 | /* |
| 111 | * reserved number/bytes of locked buffers |
| 112 | */ |
| 113 | int locked_queue_rcount = 0; |
| 114 | long locked_queue_rbytes = 0L; |
| 115 | |
| 116 | static int lfs_fits_buf(struct lfs *, int, int); |
| 117 | static int lfs_reservebuf(struct lfs *, struct vnode *vp, struct vnode *vp2, |
| 118 | int, int); |
| 119 | static int lfs_reserveavail(struct lfs *, struct vnode *vp, struct vnode *vp2, |
| 120 | int); |
| 121 | |
| 122 | static int |
| 123 | lfs_fits_buf(struct lfs *fs, int n, int bytes) |
| 124 | { |
| 125 | int count_fit, bytes_fit; |
| 126 | |
| 127 | ASSERT_NO_SEGLOCK(fs); |
| 128 | KASSERT(mutex_owned(&lfs_lock)); |
| 129 | |
| 130 | count_fit = |
| 131 | (locked_queue_count + locked_queue_rcount + n <= LFS_WAIT_BUFS); |
| 132 | bytes_fit = |
| 133 | (locked_queue_bytes + locked_queue_rbytes + bytes <= LFS_WAIT_BYTES); |
| 134 | |
| 135 | #ifdef DEBUG |
| 136 | if (!count_fit) { |
| 137 | DLOG((DLOG_AVAIL, "lfs_fits_buf: no fit count: %d + %d + %d >= %d\n" , |
| 138 | locked_queue_count, locked_queue_rcount, |
| 139 | n, LFS_WAIT_BUFS)); |
| 140 | } |
| 141 | if (!bytes_fit) { |
| 142 | DLOG((DLOG_AVAIL, "lfs_fits_buf: no fit bytes: %ld + %ld + %d >= %ld\n" , |
| 143 | locked_queue_bytes, locked_queue_rbytes, |
| 144 | bytes, LFS_WAIT_BYTES)); |
| 145 | } |
| 146 | #endif /* DEBUG */ |
| 147 | |
| 148 | return (count_fit && bytes_fit); |
| 149 | } |
| 150 | |
| 151 | /* ARGSUSED */ |
| 152 | static int |
| 153 | lfs_reservebuf(struct lfs *fs, struct vnode *vp, |
| 154 | struct vnode *vp2, int n, int bytes) |
| 155 | { |
| 156 | int cantwait; |
| 157 | |
| 158 | ASSERT_MAYBE_SEGLOCK(fs); |
| 159 | KASSERT(locked_queue_rcount >= 0); |
| 160 | KASSERT(locked_queue_rbytes >= 0); |
| 161 | |
| 162 | cantwait = (VTOI(vp)->i_flag & IN_ADIROP) || fs->lfs_unlockvp == vp; |
| 163 | mutex_enter(&lfs_lock); |
| 164 | while (!cantwait && n > 0 && !lfs_fits_buf(fs, n, bytes)) { |
| 165 | int error; |
| 166 | |
| 167 | lfs_flush(fs, 0, 0); |
| 168 | |
| 169 | DLOG((DLOG_AVAIL, "lfs_reservebuf: waiting: count=%d, bytes=%ld\n" , |
| 170 | locked_queue_count, locked_queue_bytes)); |
| 171 | ++locked_queue_waiters; |
| 172 | error = cv_timedwait_sig(&locked_queue_cv, &lfs_lock, |
| 173 | hz * LFS_BUFWAIT); |
| 174 | --locked_queue_waiters; |
| 175 | if (error && error != EWOULDBLOCK) { |
| 176 | mutex_exit(&lfs_lock); |
| 177 | return error; |
| 178 | } |
| 179 | } |
| 180 | |
| 181 | locked_queue_rcount += n; |
| 182 | locked_queue_rbytes += bytes; |
| 183 | |
| 184 | if (n < 0 && locked_queue_waiters > 0) { |
| 185 | DLOG((DLOG_AVAIL, "lfs_reservebuf: broadcast: count=%d, bytes=%ld\n" , |
| 186 | locked_queue_count, locked_queue_bytes)); |
| 187 | cv_broadcast(&locked_queue_cv); |
| 188 | } |
| 189 | |
| 190 | mutex_exit(&lfs_lock); |
| 191 | |
| 192 | KASSERT(locked_queue_rcount >= 0); |
| 193 | KASSERT(locked_queue_rbytes >= 0); |
| 194 | |
| 195 | return 0; |
| 196 | } |
| 197 | |
| 198 | /* |
| 199 | * Try to reserve some blocks, prior to performing a sensitive operation that |
| 200 | * requires the vnode lock to be honored. If there is not enough space, wait |
| 201 | * for the space to become available. |
| 202 | * |
| 203 | * Called with vp locked. (Note nowever that if fsb < 0, vp is ignored.) |
| 204 | */ |
| 205 | static int |
| 206 | lfs_reserveavail(struct lfs *fs, struct vnode *vp, |
| 207 | struct vnode *vp2, int fsb) |
| 208 | { |
| 209 | CLEANERINFO *cip; |
| 210 | struct buf *bp; |
| 211 | int error, slept; |
| 212 | int cantwait; |
| 213 | |
| 214 | ASSERT_MAYBE_SEGLOCK(fs); |
| 215 | slept = 0; |
| 216 | mutex_enter(&lfs_lock); |
| 217 | cantwait = (VTOI(vp)->i_flag & IN_ADIROP) || fs->lfs_unlockvp == vp; |
| 218 | while (!cantwait && fsb > 0 && |
| 219 | !lfs_fits(fs, fsb + fs->lfs_ravail + fs->lfs_favail)) { |
| 220 | mutex_exit(&lfs_lock); |
| 221 | |
| 222 | if (!slept) { |
| 223 | DLOG((DLOG_AVAIL, "lfs_reserve: waiting for %ld (bfree = %jd," |
| 224 | " est_bfree = %jd)\n" , |
| 225 | fsb + fs->lfs_ravail + fs->lfs_favail, |
| 226 | (intmax_t)lfs_sb_getbfree(fs), |
| 227 | (intmax_t)LFS_EST_BFREE(fs))); |
| 228 | } |
| 229 | ++slept; |
| 230 | |
| 231 | /* Wake up the cleaner */ |
| 232 | LFS_CLEANERINFO(cip, fs, bp); |
| 233 | LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); |
| 234 | lfs_wakeup_cleaner(fs); |
| 235 | |
| 236 | mutex_enter(&lfs_lock); |
| 237 | /* Cleaner might have run while we were reading, check again */ |
| 238 | if (lfs_fits(fs, fsb + fs->lfs_ravail + fs->lfs_favail)) |
| 239 | break; |
| 240 | |
| 241 | error = mtsleep(&fs->lfs_availsleep, PCATCH | PUSER, |
| 242 | "lfs_reserve" , 0, &lfs_lock); |
| 243 | if (error) { |
| 244 | mutex_exit(&lfs_lock); |
| 245 | return error; |
| 246 | } |
| 247 | } |
| 248 | #ifdef DEBUG |
| 249 | if (slept) { |
| 250 | DLOG((DLOG_AVAIL, "lfs_reserve: woke up\n" )); |
| 251 | } |
| 252 | #endif |
| 253 | fs->lfs_ravail += fsb; |
| 254 | mutex_exit(&lfs_lock); |
| 255 | |
| 256 | return 0; |
| 257 | } |
| 258 | |
| 259 | #ifdef DIAGNOSTIC |
| 260 | int lfs_rescount; |
| 261 | int lfs_rescountdirop; |
| 262 | #endif |
| 263 | |
| 264 | int |
| 265 | lfs_reserve(struct lfs *fs, struct vnode *vp, struct vnode *vp2, int fsb) |
| 266 | { |
| 267 | int error; |
| 268 | |
| 269 | ASSERT_MAYBE_SEGLOCK(fs); |
| 270 | if (vp2) { |
| 271 | /* Make sure we're not in the process of reclaiming vp2 */ |
| 272 | mutex_enter(&lfs_lock); |
| 273 | while(fs->lfs_flags & LFS_UNDIROP) { |
| 274 | mtsleep(&fs->lfs_flags, PRIBIO + 1, "lfsrundirop" , 0, |
| 275 | &lfs_lock); |
| 276 | } |
| 277 | mutex_exit(&lfs_lock); |
| 278 | } |
| 279 | |
| 280 | KASSERT(fsb < 0 || VOP_ISLOCKED(vp)); |
| 281 | KASSERT(vp2 == NULL || fsb < 0 || VOP_ISLOCKED(vp2)); |
| 282 | KASSERT(vp2 == NULL || vp2 != fs->lfs_unlockvp); |
| 283 | |
| 284 | #ifdef DIAGNOSTIC |
| 285 | mutex_enter(&lfs_lock); |
| 286 | if (fsb > 0) |
| 287 | lfs_rescount++; |
| 288 | else if (fsb < 0) |
| 289 | lfs_rescount--; |
| 290 | if (lfs_rescount < 0) |
| 291 | panic("lfs_rescount" ); |
| 292 | mutex_exit(&lfs_lock); |
| 293 | #endif |
| 294 | |
| 295 | error = lfs_reserveavail(fs, vp, vp2, fsb); |
| 296 | if (error) |
| 297 | return error; |
| 298 | |
| 299 | /* |
| 300 | * XXX just a guess. should be more precise. |
| 301 | */ |
| 302 | error = lfs_reservebuf(fs, vp, vp2, fsb, lfs_fsbtob(fs, fsb)); |
| 303 | if (error) |
| 304 | lfs_reserveavail(fs, vp, vp2, -fsb); |
| 305 | |
| 306 | return error; |
| 307 | } |
| 308 | |
| 309 | int |
| 310 | lfs_bwrite(void *v) |
| 311 | { |
| 312 | struct vop_bwrite_args /* { |
| 313 | struct vnode *a_vp; |
| 314 | struct buf *a_bp; |
| 315 | } */ *ap = v; |
| 316 | struct buf *bp = ap->a_bp; |
| 317 | |
| 318 | #ifdef DIAGNOSTIC |
| 319 | if (VTOI(bp->b_vp)->i_lfs->lfs_ronly == 0 && (bp->b_flags & B_ASYNC)) { |
| 320 | panic("bawrite LFS buffer" ); |
| 321 | } |
| 322 | #endif /* DIAGNOSTIC */ |
| 323 | return lfs_bwrite_ext(bp, 0); |
| 324 | } |
| 325 | |
| 326 | /* |
| 327 | * Determine if there is enough room currently available to write fsb |
| 328 | * blocks. We need enough blocks for the new blocks, the current |
| 329 | * inode blocks (including potentially the ifile inode), a summary block, |
| 330 | * and the segment usage table, plus an ifile block. |
| 331 | */ |
| 332 | int |
| 333 | lfs_fits(struct lfs *fs, int fsb) |
| 334 | { |
| 335 | int64_t needed; |
| 336 | |
| 337 | ASSERT_NO_SEGLOCK(fs); |
| 338 | needed = fsb + lfs_btofsb(fs, lfs_sb_getsumsize(fs)) + |
| 339 | ((howmany(lfs_sb_getuinodes(fs) + 1, LFS_INOPB(fs)) + |
| 340 | lfs_sb_getsegtabsz(fs) + |
| 341 | 1) << (lfs_sb_getbshift(fs) - lfs_sb_getffshift(fs))); |
| 342 | |
| 343 | if (needed >= lfs_sb_getavail(fs)) { |
| 344 | #ifdef DEBUG |
| 345 | DLOG((DLOG_AVAIL, "lfs_fits: no fit: fsb = %ld, uinodes = %ld, " |
| 346 | "needed = %jd, avail = %jd\n" , |
| 347 | (long)fsb, (long)lfs_sb_getuinodes(fs), (intmax_t)needed, |
| 348 | (intmax_t)lfs_sb_getavail(fs))); |
| 349 | #endif |
| 350 | return 0; |
| 351 | } |
| 352 | return 1; |
| 353 | } |
| 354 | |
| 355 | int |
| 356 | lfs_availwait(struct lfs *fs, int fsb) |
| 357 | { |
| 358 | int error; |
| 359 | CLEANERINFO *cip; |
| 360 | struct buf *cbp; |
| 361 | |
| 362 | ASSERT_NO_SEGLOCK(fs); |
| 363 | /* Push cleaner blocks through regardless */ |
| 364 | mutex_enter(&lfs_lock); |
| 365 | if (LFS_SEGLOCK_HELD(fs) && |
| 366 | fs->lfs_sp->seg_flags & (SEGM_CLEAN | SEGM_FORCE_CKP)) { |
| 367 | mutex_exit(&lfs_lock); |
| 368 | return 0; |
| 369 | } |
| 370 | mutex_exit(&lfs_lock); |
| 371 | |
| 372 | while (!lfs_fits(fs, fsb)) { |
| 373 | /* |
| 374 | * Out of space, need cleaner to run. |
| 375 | * Update the cleaner info, then wake it up. |
| 376 | * Note the cleanerinfo block is on the ifile |
| 377 | * so it CANT_WAIT. |
| 378 | */ |
| 379 | LFS_CLEANERINFO(cip, fs, cbp); |
| 380 | LFS_SYNC_CLEANERINFO(cip, fs, cbp, 0); |
| 381 | |
| 382 | #ifdef DEBUG |
| 383 | DLOG((DLOG_AVAIL, "lfs_availwait: out of available space, " |
| 384 | "waiting on cleaner\n" )); |
| 385 | #endif |
| 386 | |
| 387 | lfs_wakeup_cleaner(fs); |
| 388 | #ifdef DIAGNOSTIC |
| 389 | if (LFS_SEGLOCK_HELD(fs)) |
| 390 | panic("lfs_availwait: deadlock" ); |
| 391 | #endif |
| 392 | error = tsleep(&fs->lfs_availsleep, PCATCH | PUSER, |
| 393 | "cleaner" , 0); |
| 394 | if (error) |
| 395 | return (error); |
| 396 | } |
| 397 | return 0; |
| 398 | } |
| 399 | |
| 400 | int |
| 401 | lfs_bwrite_ext(struct buf *bp, int flags) |
| 402 | { |
| 403 | struct lfs *fs; |
| 404 | struct inode *ip; |
| 405 | struct vnode *vp; |
| 406 | int fsb; |
| 407 | |
| 408 | vp = bp->b_vp; |
| 409 | fs = VFSTOULFS(vp->v_mount)->um_lfs; |
| 410 | |
| 411 | ASSERT_MAYBE_SEGLOCK(fs); |
| 412 | KASSERT(bp->b_cflags & BC_BUSY); |
| 413 | KASSERT(flags & BW_CLEAN || !LFS_IS_MALLOC_BUF(bp)); |
| 414 | KASSERT(((bp->b_oflags | bp->b_flags) & (BO_DELWRI|B_LOCKED)) |
| 415 | != BO_DELWRI); |
| 416 | |
| 417 | /* |
| 418 | * Don't write *any* blocks if we're mounted read-only, or |
| 419 | * if we are "already unmounted". |
| 420 | * |
| 421 | * In particular the cleaner can't write blocks either. |
| 422 | */ |
| 423 | if (fs->lfs_ronly || (lfs_sb_getpflags(fs) & LFS_PF_CLEAN)) { |
| 424 | bp->b_oflags &= ~BO_DELWRI; |
| 425 | bp->b_flags |= B_READ; /* XXX is this right? --ks */ |
| 426 | bp->b_error = 0; |
| 427 | mutex_enter(&bufcache_lock); |
| 428 | LFS_UNLOCK_BUF(bp); |
| 429 | if (LFS_IS_MALLOC_BUF(bp)) |
| 430 | bp->b_cflags &= ~BC_BUSY; |
| 431 | else |
| 432 | brelsel(bp, 0); |
| 433 | mutex_exit(&bufcache_lock); |
| 434 | return (fs->lfs_ronly ? EROFS : 0); |
| 435 | } |
| 436 | |
| 437 | /* |
| 438 | * Set the delayed write flag and use reassignbuf to move the buffer |
| 439 | * from the clean list to the dirty one. |
| 440 | * |
| 441 | * Set the B_LOCKED flag and unlock the buffer, causing brelse to move |
| 442 | * the buffer onto the LOCKED free list. This is necessary, otherwise |
| 443 | * getnewbuf() would try to reclaim the buffers using bawrite, which |
| 444 | * isn't going to work. |
| 445 | * |
| 446 | * XXX we don't let meta-data writes run out of space because they can |
| 447 | * come from the segment writer. We need to make sure that there is |
| 448 | * enough space reserved so that there's room to write meta-data |
| 449 | * blocks. |
| 450 | */ |
| 451 | if ((bp->b_flags & B_LOCKED) == 0) { |
| 452 | fsb = lfs_numfrags(fs, bp->b_bcount); |
| 453 | |
| 454 | ip = VTOI(vp); |
| 455 | mutex_enter(&lfs_lock); |
| 456 | if (flags & BW_CLEAN) { |
| 457 | LFS_SET_UINO(ip, IN_CLEANING); |
| 458 | } else { |
| 459 | LFS_SET_UINO(ip, IN_MODIFIED); |
| 460 | } |
| 461 | mutex_exit(&lfs_lock); |
| 462 | lfs_sb_subavail(fs, fsb); |
| 463 | |
| 464 | mutex_enter(&bufcache_lock); |
| 465 | mutex_enter(vp->v_interlock); |
| 466 | bp->b_oflags = (bp->b_oflags | BO_DELWRI) & ~BO_DONE; |
| 467 | LFS_LOCK_BUF(bp); |
| 468 | bp->b_flags &= ~B_READ; |
| 469 | bp->b_error = 0; |
| 470 | reassignbuf(bp, bp->b_vp); |
| 471 | mutex_exit(vp->v_interlock); |
| 472 | } else { |
| 473 | mutex_enter(&bufcache_lock); |
| 474 | } |
| 475 | |
| 476 | if (bp->b_iodone != NULL) |
| 477 | bp->b_cflags &= ~BC_BUSY; |
| 478 | else |
| 479 | brelsel(bp, 0); |
| 480 | mutex_exit(&bufcache_lock); |
| 481 | |
| 482 | return (0); |
| 483 | } |
| 484 | |
| 485 | /* |
| 486 | * Called and return with the lfs_lock held. |
| 487 | */ |
| 488 | void |
| 489 | lfs_flush_fs(struct lfs *fs, int flags) |
| 490 | { |
| 491 | ASSERT_NO_SEGLOCK(fs); |
| 492 | KASSERT(mutex_owned(&lfs_lock)); |
| 493 | if (fs->lfs_ronly) |
| 494 | return; |
| 495 | |
| 496 | if (lfs_dostats) |
| 497 | ++lfs_stats.flush_invoked; |
| 498 | |
| 499 | fs->lfs_pdflush = 0; |
| 500 | mutex_exit(&lfs_lock); |
| 501 | lfs_writer_enter(fs, "fldirop" ); |
| 502 | lfs_segwrite(fs->lfs_ivnode->v_mount, flags); |
| 503 | lfs_writer_leave(fs); |
| 504 | mutex_enter(&lfs_lock); |
| 505 | fs->lfs_favail = 0; /* XXX */ |
| 506 | } |
| 507 | |
| 508 | /* |
| 509 | * This routine initiates segment writes when LFS is consuming too many |
| 510 | * resources. Ideally the pageout daemon would be able to direct LFS |
| 511 | * more subtly. |
| 512 | * XXX We have one static count of locked buffers; |
| 513 | * XXX need to think more about the multiple filesystem case. |
| 514 | * |
| 515 | * Called and return with lfs_lock held. |
| 516 | * If fs != NULL, we hold the segment lock for fs. |
| 517 | */ |
| 518 | void |
| 519 | lfs_flush(struct lfs *fs, int flags, int only_onefs) |
| 520 | { |
| 521 | extern u_int64_t locked_fakequeue_count; |
| 522 | struct mount *mp, *nmp; |
| 523 | struct lfs *tfs; |
| 524 | |
| 525 | KASSERT(mutex_owned(&lfs_lock)); |
| 526 | KDASSERT(fs == NULL || !LFS_SEGLOCK_HELD(fs)); |
| 527 | |
| 528 | if (lfs_dostats) |
| 529 | ++lfs_stats.write_exceeded; |
| 530 | /* XXX should we include SEGM_CKP here? */ |
| 531 | if (lfs_writing && !(flags & SEGM_SYNC)) { |
| 532 | DLOG((DLOG_FLUSH, "lfs_flush: not flushing because another flush is active\n" )); |
| 533 | return; |
| 534 | } |
| 535 | while (lfs_writing) |
| 536 | cv_wait(&lfs_writing_cv, &lfs_lock); |
| 537 | lfs_writing = 1; |
| 538 | |
| 539 | mutex_exit(&lfs_lock); |
| 540 | |
| 541 | if (only_onefs) { |
| 542 | KASSERT(fs != NULL); |
| 543 | if (vfs_busy(fs->lfs_ivnode->v_mount, NULL)) |
| 544 | goto errout; |
| 545 | mutex_enter(&lfs_lock); |
| 546 | lfs_flush_fs(fs, flags); |
| 547 | mutex_exit(&lfs_lock); |
| 548 | vfs_unbusy(fs->lfs_ivnode->v_mount, false, NULL); |
| 549 | } else { |
| 550 | locked_fakequeue_count = 0; |
| 551 | mutex_enter(&mountlist_lock); |
| 552 | for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { |
| 553 | if (vfs_busy(mp, &nmp)) { |
| 554 | DLOG((DLOG_FLUSH, "lfs_flush: fs vfs_busy\n" )); |
| 555 | continue; |
| 556 | } |
| 557 | if (strncmp(&mp->mnt_stat.f_fstypename[0], MOUNT_LFS, |
| 558 | sizeof(mp->mnt_stat.f_fstypename)) == 0) { |
| 559 | tfs = VFSTOULFS(mp)->um_lfs; |
| 560 | mutex_enter(&lfs_lock); |
| 561 | lfs_flush_fs(tfs, flags); |
| 562 | mutex_exit(&lfs_lock); |
| 563 | } |
| 564 | vfs_unbusy(mp, false, &nmp); |
| 565 | } |
| 566 | mutex_exit(&mountlist_lock); |
| 567 | } |
| 568 | LFS_DEBUG_COUNTLOCKED("flush" ); |
| 569 | wakeup(&lfs_subsys_pages); |
| 570 | |
| 571 | errout: |
| 572 | mutex_enter(&lfs_lock); |
| 573 | KASSERT(lfs_writing); |
| 574 | lfs_writing = 0; |
| 575 | wakeup(&lfs_writing); |
| 576 | } |
| 577 | |
| 578 | #define INOCOUNT(fs) howmany(lfs_sb_getuinodes(fs), LFS_INOPB(fs)) |
| 579 | #define INOBYTES(fs) (lfs_sb_getuinodes(fs) * DINOSIZE(fs)) |
| 580 | |
| 581 | /* |
| 582 | * make sure that we don't have too many locked buffers. |
| 583 | * flush buffers if needed. |
| 584 | */ |
| 585 | int |
| 586 | lfs_check(struct vnode *vp, daddr_t blkno, int flags) |
| 587 | { |
| 588 | int error; |
| 589 | struct lfs *fs; |
| 590 | struct inode *ip; |
| 591 | extern pid_t lfs_writer_daemon; |
| 592 | |
| 593 | error = 0; |
| 594 | ip = VTOI(vp); |
| 595 | |
| 596 | /* If out of buffers, wait on writer */ |
| 597 | /* XXX KS - if it's the Ifile, we're probably the cleaner! */ |
| 598 | if (ip->i_number == LFS_IFILE_INUM) |
| 599 | return 0; |
| 600 | /* If we're being called from inside a dirop, don't sleep */ |
| 601 | if (ip->i_flag & IN_ADIROP) |
| 602 | return 0; |
| 603 | |
| 604 | fs = ip->i_lfs; |
| 605 | |
| 606 | ASSERT_NO_SEGLOCK(fs); |
| 607 | |
| 608 | /* |
| 609 | * If we would flush below, but dirops are active, sleep. |
| 610 | * Note that a dirop cannot ever reach this code! |
| 611 | */ |
| 612 | mutex_enter(&lfs_lock); |
| 613 | while (fs->lfs_dirops > 0 && |
| 614 | (locked_queue_count + INOCOUNT(fs) > LFS_MAX_BUFS || |
| 615 | locked_queue_bytes + INOBYTES(fs) > LFS_MAX_BYTES || |
| 616 | lfs_subsys_pages > LFS_MAX_PAGES || |
| 617 | fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) || |
| 618 | lfs_dirvcount > LFS_MAX_DIROP || fs->lfs_diropwait > 0)) |
| 619 | { |
| 620 | ++fs->lfs_diropwait; |
| 621 | mtsleep(&fs->lfs_writer, PRIBIO+1, "bufdirop" , 0, |
| 622 | &lfs_lock); |
| 623 | --fs->lfs_diropwait; |
| 624 | } |
| 625 | |
| 626 | #ifdef DEBUG |
| 627 | if (locked_queue_count + INOCOUNT(fs) > LFS_MAX_BUFS) |
| 628 | DLOG((DLOG_FLUSH, "lfs_check: lqc = %d, max %d\n" , |
| 629 | locked_queue_count + INOCOUNT(fs), LFS_MAX_BUFS)); |
| 630 | if (locked_queue_bytes + INOBYTES(fs) > LFS_MAX_BYTES) |
| 631 | DLOG((DLOG_FLUSH, "lfs_check: lqb = %ld, max %ld\n" , |
| 632 | locked_queue_bytes + INOBYTES(fs), LFS_MAX_BYTES)); |
| 633 | if (lfs_subsys_pages > LFS_MAX_PAGES) |
| 634 | DLOG((DLOG_FLUSH, "lfs_check: lssp = %d, max %d\n" , |
| 635 | lfs_subsys_pages, LFS_MAX_PAGES)); |
| 636 | if (lfs_fs_pagetrip && fs->lfs_pages > lfs_fs_pagetrip) |
| 637 | DLOG((DLOG_FLUSH, "lfs_check: fssp = %d, trip at %d\n" , |
| 638 | fs->lfs_pages, lfs_fs_pagetrip)); |
| 639 | if (lfs_dirvcount > LFS_MAX_DIROP) |
| 640 | DLOG((DLOG_FLUSH, "lfs_check: ldvc = %d, max %d\n" , |
| 641 | lfs_dirvcount, LFS_MAX_DIROP)); |
| 642 | if (fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs)) |
| 643 | DLOG((DLOG_FLUSH, "lfs_check: lfdvc = %d, max %d\n" , |
| 644 | fs->lfs_dirvcount, LFS_MAX_FSDIROP(fs))); |
| 645 | if (fs->lfs_diropwait > 0) |
| 646 | DLOG((DLOG_FLUSH, "lfs_check: ldvw = %d\n" , |
| 647 | fs->lfs_diropwait)); |
| 648 | #endif |
| 649 | |
| 650 | /* If there are too many pending dirops, we have to flush them. */ |
| 651 | if (fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) || |
| 652 | lfs_dirvcount > LFS_MAX_DIROP || fs->lfs_diropwait > 0) { |
| 653 | mutex_exit(&lfs_lock); |
| 654 | lfs_flush_dirops(fs); |
| 655 | mutex_enter(&lfs_lock); |
| 656 | } else if (locked_queue_count + INOCOUNT(fs) > LFS_MAX_BUFS || |
| 657 | locked_queue_bytes + INOBYTES(fs) > LFS_MAX_BYTES || |
| 658 | lfs_subsys_pages > LFS_MAX_PAGES || |
| 659 | fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) || |
| 660 | lfs_dirvcount > LFS_MAX_DIROP || fs->lfs_diropwait > 0) { |
| 661 | lfs_flush(fs, flags, 0); |
| 662 | } else if (lfs_fs_pagetrip && fs->lfs_pages > lfs_fs_pagetrip) { |
| 663 | /* |
| 664 | * If we didn't flush the whole thing, some filesystems |
| 665 | * still might want to be flushed. |
| 666 | */ |
| 667 | ++fs->lfs_pdflush; |
| 668 | wakeup(&lfs_writer_daemon); |
| 669 | } |
| 670 | |
| 671 | while (locked_queue_count + INOCOUNT(fs) >= LFS_WAIT_BUFS || |
| 672 | locked_queue_bytes + INOBYTES(fs) >= LFS_WAIT_BYTES || |
| 673 | lfs_subsys_pages > LFS_WAIT_PAGES || |
| 674 | fs->lfs_dirvcount > LFS_MAX_FSDIROP(fs) || |
| 675 | lfs_dirvcount > LFS_MAX_DIROP) { |
| 676 | |
| 677 | if (lfs_dostats) |
| 678 | ++lfs_stats.wait_exceeded; |
| 679 | DLOG((DLOG_AVAIL, "lfs_check: waiting: count=%d, bytes=%ld\n" , |
| 680 | locked_queue_count, locked_queue_bytes)); |
| 681 | ++locked_queue_waiters; |
| 682 | error = cv_timedwait_sig(&locked_queue_cv, &lfs_lock, |
| 683 | hz * LFS_BUFWAIT); |
| 684 | --locked_queue_waiters; |
| 685 | if (error != EWOULDBLOCK) |
| 686 | break; |
| 687 | |
| 688 | /* |
| 689 | * lfs_flush might not flush all the buffers, if some of the |
| 690 | * inodes were locked or if most of them were Ifile blocks |
| 691 | * and we weren't asked to checkpoint. Try flushing again |
| 692 | * to keep us from blocking indefinitely. |
| 693 | */ |
| 694 | if (locked_queue_count + INOCOUNT(fs) >= LFS_MAX_BUFS || |
| 695 | locked_queue_bytes + INOBYTES(fs) >= LFS_MAX_BYTES) { |
| 696 | lfs_flush(fs, flags | SEGM_CKP, 0); |
| 697 | } |
| 698 | } |
| 699 | mutex_exit(&lfs_lock); |
| 700 | return (error); |
| 701 | } |
| 702 | |
| 703 | /* |
| 704 | * Allocate a new buffer header. |
| 705 | */ |
| 706 | struct buf * |
| 707 | lfs_newbuf(struct lfs *fs, struct vnode *vp, daddr_t daddr, size_t size, int type) |
| 708 | { |
| 709 | struct buf *bp; |
| 710 | size_t nbytes; |
| 711 | |
| 712 | ASSERT_MAYBE_SEGLOCK(fs); |
| 713 | nbytes = roundup(size, lfs_fsbtob(fs, 1)); |
| 714 | |
| 715 | bp = getiobuf(NULL, true); |
| 716 | if (nbytes) { |
| 717 | bp->b_data = lfs_malloc(fs, nbytes, type); |
| 718 | /* memset(bp->b_data, 0, nbytes); */ |
| 719 | } |
| 720 | #ifdef DIAGNOSTIC |
| 721 | if (vp == NULL) |
| 722 | panic("vp is NULL in lfs_newbuf" ); |
| 723 | if (bp == NULL) |
| 724 | panic("bp is NULL after malloc in lfs_newbuf" ); |
| 725 | #endif |
| 726 | |
| 727 | bp->b_bufsize = size; |
| 728 | bp->b_bcount = size; |
| 729 | bp->b_lblkno = daddr; |
| 730 | bp->b_blkno = daddr; |
| 731 | bp->b_error = 0; |
| 732 | bp->b_resid = 0; |
| 733 | bp->b_iodone = lfs_callback; |
| 734 | bp->b_cflags = BC_BUSY | BC_NOCACHE; |
| 735 | bp->b_private = fs; |
| 736 | |
| 737 | mutex_enter(&bufcache_lock); |
| 738 | mutex_enter(vp->v_interlock); |
| 739 | bgetvp(vp, bp); |
| 740 | mutex_exit(vp->v_interlock); |
| 741 | mutex_exit(&bufcache_lock); |
| 742 | |
| 743 | return (bp); |
| 744 | } |
| 745 | |
| 746 | void |
| 747 | lfs_freebuf(struct lfs *fs, struct buf *bp) |
| 748 | { |
| 749 | struct vnode *vp; |
| 750 | |
| 751 | if ((vp = bp->b_vp) != NULL) { |
| 752 | mutex_enter(&bufcache_lock); |
| 753 | mutex_enter(vp->v_interlock); |
| 754 | brelvp(bp); |
| 755 | mutex_exit(vp->v_interlock); |
| 756 | mutex_exit(&bufcache_lock); |
| 757 | } |
| 758 | if (!(bp->b_cflags & BC_INVAL)) { /* BC_INVAL indicates a "fake" buffer */ |
| 759 | lfs_free(fs, bp->b_data, LFS_NB_UNKNOWN); |
| 760 | bp->b_data = NULL; |
| 761 | } |
| 762 | putiobuf(bp); |
| 763 | } |
| 764 | |
| 765 | /* |
| 766 | * Count buffers on the "locked" queue, and compare it to a pro-forma count. |
| 767 | * Don't count malloced buffers, since they don't detract from the total. |
| 768 | */ |
| 769 | void |
| 770 | lfs_countlocked(int *count, long *bytes, const char *msg) |
| 771 | { |
| 772 | struct buf *bp; |
| 773 | int n = 0; |
| 774 | long int size = 0L; |
| 775 | |
| 776 | mutex_enter(&bufcache_lock); |
| 777 | TAILQ_FOREACH(bp, &bufqueues[BQ_LOCKED].bq_queue, b_freelist) { |
| 778 | KASSERT(bp->b_iodone == NULL); |
| 779 | n++; |
| 780 | size += bp->b_bufsize; |
| 781 | #ifdef DIAGNOSTIC |
| 782 | if (n > nbuf) |
| 783 | panic("lfs_countlocked: this can't happen: more" |
| 784 | " buffers locked than exist" ); |
| 785 | #endif |
| 786 | } |
| 787 | /* |
| 788 | * Theoretically this function never really does anything. |
| 789 | * Give a warning if we have to fix the accounting. |
| 790 | */ |
| 791 | if (n != *count) { |
| 792 | DLOG((DLOG_LLIST, "lfs_countlocked: %s: adjusted buf count" |
| 793 | " from %d to %d\n" , msg, *count, n)); |
| 794 | } |
| 795 | if (size != *bytes) { |
| 796 | DLOG((DLOG_LLIST, "lfs_countlocked: %s: adjusted byte count" |
| 797 | " from %ld to %ld\n" , msg, *bytes, size)); |
| 798 | } |
| 799 | *count = n; |
| 800 | *bytes = size; |
| 801 | mutex_exit(&bufcache_lock); |
| 802 | return; |
| 803 | } |
| 804 | |
| 805 | int |
| 806 | lfs_wait_pages(void) |
| 807 | { |
| 808 | int active, inactive; |
| 809 | |
| 810 | uvm_estimatepageable(&active, &inactive); |
| 811 | return LFS_WAIT_RESOURCE(active + inactive + uvmexp.free, 1); |
| 812 | } |
| 813 | |
| 814 | int |
| 815 | lfs_max_pages(void) |
| 816 | { |
| 817 | int active, inactive; |
| 818 | |
| 819 | uvm_estimatepageable(&active, &inactive); |
| 820 | return LFS_MAX_RESOURCE(active + inactive + uvmexp.free, 1); |
| 821 | } |
| 822 | |