lfs_vnops.c revision 1.213
1 /* $NetBSD: lfs_vnops.c,v 1.213 2007/11/26 19:02:32 pooka 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 (at) 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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 /* 39 * Copyright (c) 1986, 1989, 1991, 1993, 1995 40 * The Regents of the University of California. All rights reserved. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)lfs_vnops.c 8.13 (Berkeley) 6/10/95 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: lfs_vnops.c,v 1.213 2007/11/26 19:02:32 pooka Exp $"); 71 72 #ifdef _KERNEL_OPT 73 #include "opt_compat_netbsd.h" 74 #endif 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/namei.h> 79 #include <sys/resourcevar.h> 80 #include <sys/kernel.h> 81 #include <sys/file.h> 82 #include <sys/stat.h> 83 #include <sys/buf.h> 84 #include <sys/proc.h> 85 #include <sys/mount.h> 86 #include <sys/vnode.h> 87 #include <sys/pool.h> 88 #include <sys/signalvar.h> 89 #include <sys/kauth.h> 90 #include <sys/syslog.h> 91 #include <sys/fstrans.h> 92 93 #include <miscfs/fifofs/fifo.h> 94 #include <miscfs/genfs/genfs.h> 95 #include <miscfs/specfs/specdev.h> 96 97 #include <ufs/ufs/inode.h> 98 #include <ufs/ufs/dir.h> 99 #include <ufs/ufs/ufsmount.h> 100 #include <ufs/ufs/ufs_extern.h> 101 102 #include <uvm/uvm.h> 103 #include <uvm/uvm_pmap.h> 104 #include <uvm/uvm_stat.h> 105 #include <uvm/uvm_pager.h> 106 107 #include <ufs/lfs/lfs.h> 108 #include <ufs/lfs/lfs_extern.h> 109 110 extern pid_t lfs_writer_daemon; 111 int lfs_ignore_lazy_sync = 1; 112 113 /* Global vfs data structures for lfs. */ 114 int (**lfs_vnodeop_p)(void *); 115 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 116 { &vop_default_desc, vn_default_error }, 117 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 118 { &vop_create_desc, lfs_create }, /* create */ 119 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 120 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 121 { &vop_open_desc, ufs_open }, /* open */ 122 { &vop_close_desc, lfs_close }, /* close */ 123 { &vop_access_desc, ufs_access }, /* access */ 124 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 125 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 126 { &vop_read_desc, lfs_read }, /* read */ 127 { &vop_write_desc, lfs_write }, /* write */ 128 { &vop_lease_desc, ufs_lease_check }, /* lease */ 129 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 130 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 131 { &vop_poll_desc, ufs_poll }, /* poll */ 132 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 133 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 134 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 135 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 136 { &vop_seek_desc, ufs_seek }, /* seek */ 137 { &vop_remove_desc, lfs_remove }, /* remove */ 138 { &vop_link_desc, lfs_link }, /* link */ 139 { &vop_rename_desc, lfs_rename }, /* rename */ 140 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 141 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 142 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 143 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 144 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 145 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 146 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 147 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 148 { &vop_lock_desc, ufs_lock }, /* lock */ 149 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 150 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 151 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 152 { &vop_print_desc, ufs_print }, /* print */ 153 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 154 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 155 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 156 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 157 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 158 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 159 { NULL, NULL } 160 }; 161 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 162 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 163 164 int (**lfs_specop_p)(void *); 165 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 166 { &vop_default_desc, vn_default_error }, 167 { &vop_lookup_desc, spec_lookup }, /* lookup */ 168 { &vop_create_desc, spec_create }, /* create */ 169 { &vop_mknod_desc, spec_mknod }, /* mknod */ 170 { &vop_open_desc, spec_open }, /* open */ 171 { &vop_close_desc, lfsspec_close }, /* close */ 172 { &vop_access_desc, ufs_access }, /* access */ 173 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 174 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 175 { &vop_read_desc, ufsspec_read }, /* read */ 176 { &vop_write_desc, ufsspec_write }, /* write */ 177 { &vop_lease_desc, spec_lease_check }, /* lease */ 178 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 179 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 180 { &vop_poll_desc, spec_poll }, /* poll */ 181 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 182 { &vop_revoke_desc, spec_revoke }, /* revoke */ 183 { &vop_mmap_desc, spec_mmap }, /* mmap */ 184 { &vop_fsync_desc, spec_fsync }, /* fsync */ 185 { &vop_seek_desc, spec_seek }, /* seek */ 186 { &vop_remove_desc, spec_remove }, /* remove */ 187 { &vop_link_desc, spec_link }, /* link */ 188 { &vop_rename_desc, spec_rename }, /* rename */ 189 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 190 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 191 { &vop_symlink_desc, spec_symlink }, /* symlink */ 192 { &vop_readdir_desc, spec_readdir }, /* readdir */ 193 { &vop_readlink_desc, spec_readlink }, /* readlink */ 194 { &vop_abortop_desc, spec_abortop }, /* abortop */ 195 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 196 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 197 { &vop_lock_desc, ufs_lock }, /* lock */ 198 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 199 { &vop_bmap_desc, spec_bmap }, /* bmap */ 200 { &vop_strategy_desc, spec_strategy }, /* strategy */ 201 { &vop_print_desc, ufs_print }, /* print */ 202 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 203 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 204 { &vop_advlock_desc, spec_advlock }, /* advlock */ 205 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 206 { &vop_getpages_desc, spec_getpages }, /* getpages */ 207 { &vop_putpages_desc, spec_putpages }, /* putpages */ 208 { NULL, NULL } 209 }; 210 const struct vnodeopv_desc lfs_specop_opv_desc = 211 { &lfs_specop_p, lfs_specop_entries }; 212 213 int (**lfs_fifoop_p)(void *); 214 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 215 { &vop_default_desc, vn_default_error }, 216 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 217 { &vop_create_desc, fifo_create }, /* create */ 218 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 219 { &vop_open_desc, fifo_open }, /* open */ 220 { &vop_close_desc, lfsfifo_close }, /* close */ 221 { &vop_access_desc, ufs_access }, /* access */ 222 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 223 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 224 { &vop_read_desc, ufsfifo_read }, /* read */ 225 { &vop_write_desc, ufsfifo_write }, /* write */ 226 { &vop_lease_desc, fifo_lease_check }, /* lease */ 227 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 228 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 229 { &vop_poll_desc, fifo_poll }, /* poll */ 230 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 231 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 232 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 233 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 234 { &vop_seek_desc, fifo_seek }, /* seek */ 235 { &vop_remove_desc, fifo_remove }, /* remove */ 236 { &vop_link_desc, fifo_link }, /* link */ 237 { &vop_rename_desc, fifo_rename }, /* rename */ 238 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 239 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 240 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 241 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 242 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 243 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 244 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 245 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 246 { &vop_lock_desc, ufs_lock }, /* lock */ 247 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 248 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 249 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 250 { &vop_print_desc, ufs_print }, /* print */ 251 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 252 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 253 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 254 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 255 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 256 { NULL, NULL } 257 }; 258 const struct vnodeopv_desc lfs_fifoop_opv_desc = 259 { &lfs_fifoop_p, lfs_fifoop_entries }; 260 261 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int, struct vm_page **); 262 263 #define LFS_READWRITE 264 #include <ufs/ufs/ufs_readwrite.c> 265 #undef LFS_READWRITE 266 267 /* 268 * Synch an open file. 269 */ 270 /* ARGSUSED */ 271 int 272 lfs_fsync(void *v) 273 { 274 struct vop_fsync_args /* { 275 struct vnode *a_vp; 276 kauth_cred_t a_cred; 277 int a_flags; 278 off_t offlo; 279 off_t offhi; 280 } */ *ap = v; 281 struct vnode *vp = ap->a_vp; 282 int error, wait; 283 struct inode *ip = VTOI(vp); 284 struct lfs *fs = ip->i_lfs; 285 286 /* If we're mounted read-only, don't try to sync. */ 287 if (fs->lfs_ronly) 288 return 0; 289 290 /* 291 * Trickle sync simply adds this vnode to the pager list, as if 292 * the pagedaemon had requested a pageout. 293 */ 294 if (ap->a_flags & FSYNC_LAZY) { 295 if (lfs_ignore_lazy_sync == 0) { 296 simple_lock(&fs->lfs_interlock); 297 if (!(ip->i_flags & IN_PAGING)) { 298 ip->i_flags |= IN_PAGING; 299 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, 300 i_lfs_pchain); 301 } 302 simple_unlock(&fs->lfs_interlock); 303 simple_lock(&lfs_subsys_lock); 304 wakeup(&lfs_writer_daemon); 305 simple_unlock(&lfs_subsys_lock); 306 } 307 return 0; 308 } 309 310 /* 311 * If a vnode is bring cleaned, flush it out before we try to 312 * reuse it. This prevents the cleaner from writing files twice 313 * in the same partial segment, causing an accounting underflow. 314 */ 315 if (ap->a_flags & FSYNC_RECLAIM && ip->i_flags & IN_CLEANING) { 316 lfs_vflush(vp); 317 } 318 319 wait = (ap->a_flags & FSYNC_WAIT); 320 do { 321 simple_lock(&vp->v_interlock); 322 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 323 round_page(ap->a_offhi), 324 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 325 if (error == EAGAIN) { 326 simple_lock(&fs->lfs_interlock); 327 ltsleep(&fs->lfs_avail, PCATCH | PUSER, "lfs_fsync", 328 hz / 100 + 1, &fs->lfs_interlock); 329 simple_unlock(&fs->lfs_interlock); 330 } 331 } while (error == EAGAIN); 332 if (error) 333 return error; 334 335 if ((ap->a_flags & FSYNC_DATAONLY) == 0) 336 error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 337 338 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 339 int l = 0; 340 error = VOP_IOCTL(ip->i_devvp, DIOCCACHESYNC, &l, FWRITE, 341 curlwp->l_cred); 342 } 343 if (wait && !VPISEMPTY(vp)) 344 LFS_SET_UINO(ip, IN_MODIFIED); 345 346 return error; 347 } 348 349 /* 350 * Take IN_ADIROP off, then call ufs_inactive. 351 */ 352 int 353 lfs_inactive(void *v) 354 { 355 struct vop_inactive_args /* { 356 struct vnode *a_vp; 357 } */ *ap = v; 358 359 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 360 361 lfs_unmark_vnode(ap->a_vp); 362 363 /* 364 * The Ifile is only ever inactivated on unmount. 365 * Streamline this process by not giving it more dirty blocks. 366 */ 367 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 368 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 369 VOP_UNLOCK(ap->a_vp, 0); 370 return 0; 371 } 372 373 return ufs_inactive(v); 374 } 375 376 /* 377 * These macros are used to bracket UFS directory ops, so that we can 378 * identify all the pages touched during directory ops which need to 379 * be ordered and flushed atomically, so that they may be recovered. 380 * 381 * Because we have to mark nodes VU_DIROP in order to prevent 382 * the cache from reclaiming them while a dirop is in progress, we must 383 * also manage the number of nodes so marked (otherwise we can run out). 384 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 385 * is decremented during segment write, when VU_DIROP is taken off. 386 */ 387 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 388 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 389 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 390 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 391 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 392 static int lfs_set_dirop(struct vnode *, struct vnode *); 393 394 static int 395 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 396 { 397 struct lfs *fs; 398 int error; 399 400 KASSERT(VOP_ISLOCKED(dvp)); 401 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 402 403 fs = VTOI(dvp)->i_lfs; 404 405 ASSERT_NO_SEGLOCK(fs); 406 /* 407 * LFS_NRESERVE calculates direct and indirect blocks as well 408 * as an inode block; an overestimate in most cases. 409 */ 410 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 411 return (error); 412 413 restart: 414 simple_lock(&fs->lfs_interlock); 415 if (fs->lfs_dirops == 0) { 416 simple_unlock(&fs->lfs_interlock); 417 lfs_check(dvp, LFS_UNUSED_LBN, 0); 418 simple_lock(&fs->lfs_interlock); 419 } 420 while (fs->lfs_writer) { 421 error = ltsleep(&fs->lfs_dirops, (PRIBIO + 1) | PCATCH, 422 "lfs_sdirop", 0, &fs->lfs_interlock); 423 if (error == EINTR) { 424 simple_unlock(&fs->lfs_interlock); 425 goto unreserve; 426 } 427 } 428 simple_lock(&lfs_subsys_lock); 429 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 430 wakeup(&lfs_writer_daemon); 431 simple_unlock(&lfs_subsys_lock); 432 simple_unlock(&fs->lfs_interlock); 433 preempt(); 434 goto restart; 435 } 436 437 if (lfs_dirvcount > LFS_MAX_DIROP) { 438 simple_unlock(&fs->lfs_interlock); 439 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 440 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 441 if ((error = ltsleep(&lfs_dirvcount, 442 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 443 &lfs_subsys_lock)) != 0) { 444 goto unreserve; 445 } 446 goto restart; 447 } 448 simple_unlock(&lfs_subsys_lock); 449 450 ++fs->lfs_dirops; 451 fs->lfs_doifile = 1; 452 simple_unlock(&fs->lfs_interlock); 453 454 /* Hold a reference so SET_ENDOP will be happy */ 455 vref(dvp); 456 if (vp) { 457 vref(vp); 458 MARK_VNODE(vp); 459 } 460 461 MARK_VNODE(dvp); 462 return 0; 463 464 unreserve: 465 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 466 return error; 467 } 468 469 /* 470 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 471 * in getnewvnode(), if we have a stacked filesystem mounted on top 472 * of us. 473 * 474 * NB: this means we have to clear the new vnodes on error. Fortunately 475 * SET_ENDOP is there to do that for us. 476 */ 477 static int 478 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 479 { 480 int error; 481 struct lfs *fs; 482 483 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 484 ASSERT_NO_SEGLOCK(fs); 485 if (fs->lfs_ronly) 486 return EROFS; 487 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 488 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 489 dvp, error)); 490 return error; 491 } 492 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 493 if (vpp) { 494 ungetnewvnode(*vpp); 495 *vpp = NULL; 496 } 497 return error; 498 } 499 return 0; 500 } 501 502 #define SET_ENDOP_BASE(fs, dvp, str) \ 503 do { \ 504 simple_lock(&(fs)->lfs_interlock); \ 505 --(fs)->lfs_dirops; \ 506 if (!(fs)->lfs_dirops) { \ 507 if ((fs)->lfs_nadirop) { \ 508 panic("SET_ENDOP: %s: no dirops but " \ 509 " nadirop=%d", (str), \ 510 (fs)->lfs_nadirop); \ 511 } \ 512 wakeup(&(fs)->lfs_writer); \ 513 simple_unlock(&(fs)->lfs_interlock); \ 514 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 515 } else \ 516 simple_unlock(&(fs)->lfs_interlock); \ 517 } while(0) 518 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 519 do { \ 520 UNMARK_VNODE(dvp); \ 521 if (nvpp && *nvpp) \ 522 UNMARK_VNODE(*nvpp); \ 523 /* Check for error return to stem vnode leakage */ \ 524 if (nvpp && *nvpp && !((*nvpp)->v_uflag & VU_DIROP)) \ 525 ungetnewvnode(*(nvpp)); \ 526 SET_ENDOP_BASE((fs), (dvp), (str)); \ 527 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 528 vrele(dvp); \ 529 } while(0) 530 #define SET_ENDOP_CREATE_AP(ap, str) \ 531 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 532 (ap)->a_vpp, (str)) 533 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 534 do { \ 535 UNMARK_VNODE(dvp); \ 536 if (ovp) \ 537 UNMARK_VNODE(ovp); \ 538 SET_ENDOP_BASE((fs), (dvp), (str)); \ 539 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 540 vrele(dvp); \ 541 if (ovp) \ 542 vrele(ovp); \ 543 } while(0) 544 545 void 546 lfs_mark_vnode(struct vnode *vp) 547 { 548 struct inode *ip = VTOI(vp); 549 struct lfs *fs = ip->i_lfs; 550 551 simple_lock(&fs->lfs_interlock); 552 if (!(ip->i_flag & IN_ADIROP)) { 553 if (!(vp->v_uflag & VU_DIROP)) { 554 (void)lfs_vref(vp); 555 simple_lock(&lfs_subsys_lock); 556 ++lfs_dirvcount; 557 ++fs->lfs_dirvcount; 558 simple_unlock(&lfs_subsys_lock); 559 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 560 vp->v_uflag |= VU_DIROP; 561 } 562 ++fs->lfs_nadirop; 563 ip->i_flag |= IN_ADIROP; 564 } else 565 KASSERT(vp->v_uflag & VU_DIROP); 566 simple_unlock(&fs->lfs_interlock); 567 } 568 569 void 570 lfs_unmark_vnode(struct vnode *vp) 571 { 572 struct inode *ip = VTOI(vp); 573 574 if (ip && (ip->i_flag & IN_ADIROP)) { 575 KASSERT(vp->v_uflag & VU_DIROP); 576 simple_lock(&ip->i_lfs->lfs_interlock); 577 --ip->i_lfs->lfs_nadirop; 578 simple_unlock(&ip->i_lfs->lfs_interlock); 579 ip->i_flag &= ~IN_ADIROP; 580 } 581 } 582 583 int 584 lfs_symlink(void *v) 585 { 586 struct vop_symlink_args /* { 587 struct vnode *a_dvp; 588 struct vnode **a_vpp; 589 struct componentname *a_cnp; 590 struct vattr *a_vap; 591 char *a_target; 592 } */ *ap = v; 593 int error; 594 595 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 596 vput(ap->a_dvp); 597 return error; 598 } 599 error = ufs_symlink(ap); 600 SET_ENDOP_CREATE_AP(ap, "symlink"); 601 return (error); 602 } 603 604 int 605 lfs_mknod(void *v) 606 { 607 struct vop_mknod_args /* { 608 struct vnode *a_dvp; 609 struct vnode **a_vpp; 610 struct componentname *a_cnp; 611 struct vattr *a_vap; 612 } */ *ap = v; 613 struct vattr *vap = ap->a_vap; 614 struct vnode **vpp = ap->a_vpp; 615 struct inode *ip; 616 int error; 617 struct mount *mp; 618 ino_t ino; 619 620 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 621 vput(ap->a_dvp); 622 return error; 623 } 624 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 625 ap->a_dvp, vpp, ap->a_cnp); 626 627 /* Either way we're done with the dirop at this point */ 628 SET_ENDOP_CREATE_AP(ap, "mknod"); 629 630 if (error) 631 return (error); 632 633 ip = VTOI(*vpp); 634 mp = (*vpp)->v_mount; 635 ino = ip->i_number; 636 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 637 if (vap->va_rdev != VNOVAL) { 638 /* 639 * Want to be able to use this to make badblock 640 * inodes, so don't truncate the dev number. 641 */ 642 #if 0 643 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 644 UFS_MPNEEDSWAP((*vpp)->v_mount)); 645 #else 646 ip->i_ffs1_rdev = vap->va_rdev; 647 #endif 648 } 649 650 /* 651 * Call fsync to write the vnode so that we don't have to deal with 652 * flushing it when it's marked VU_DIROP|VI_XLOCK. 653 * 654 * XXX KS - If we can't flush we also can't call vgone(), so must 655 * return. But, that leaves this vnode in limbo, also not good. 656 * Can this ever happen (barring hardware failure)? 657 */ 658 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0)) != 0) { 659 panic("lfs_mknod: couldn't fsync (ino %llu)", 660 (unsigned long long)ino); 661 /* return (error); */ 662 } 663 /* 664 * Remove vnode so that it will be reloaded by VFS_VGET and 665 * checked to see if it is an alias of an existing entry in 666 * the inode cache. 667 */ 668 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 669 670 VOP_UNLOCK(*vpp, 0); 671 lfs_vunref(*vpp); 672 (*vpp)->v_type = VNON; 673 vgone(*vpp); 674 error = VFS_VGET(mp, ino, vpp); 675 676 if (error != 0) { 677 *vpp = NULL; 678 return (error); 679 } 680 return (0); 681 } 682 683 int 684 lfs_create(void *v) 685 { 686 struct vop_create_args /* { 687 struct vnode *a_dvp; 688 struct vnode **a_vpp; 689 struct componentname *a_cnp; 690 struct vattr *a_vap; 691 } */ *ap = v; 692 int error; 693 694 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 695 vput(ap->a_dvp); 696 return error; 697 } 698 error = ufs_create(ap); 699 SET_ENDOP_CREATE_AP(ap, "create"); 700 return (error); 701 } 702 703 int 704 lfs_mkdir(void *v) 705 { 706 struct vop_mkdir_args /* { 707 struct vnode *a_dvp; 708 struct vnode **a_vpp; 709 struct componentname *a_cnp; 710 struct vattr *a_vap; 711 } */ *ap = v; 712 int error; 713 714 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 715 vput(ap->a_dvp); 716 return error; 717 } 718 error = ufs_mkdir(ap); 719 SET_ENDOP_CREATE_AP(ap, "mkdir"); 720 return (error); 721 } 722 723 int 724 lfs_remove(void *v) 725 { 726 struct vop_remove_args /* { 727 struct vnode *a_dvp; 728 struct vnode *a_vp; 729 struct componentname *a_cnp; 730 } */ *ap = v; 731 struct vnode *dvp, *vp; 732 struct inode *ip; 733 int error; 734 735 dvp = ap->a_dvp; 736 vp = ap->a_vp; 737 ip = VTOI(vp); 738 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 739 if (dvp == vp) 740 vrele(vp); 741 else 742 vput(vp); 743 vput(dvp); 744 return error; 745 } 746 error = ufs_remove(ap); 747 if (ip->i_nlink == 0) 748 lfs_orphan(ip->i_lfs, ip->i_number); 749 SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove"); 750 return (error); 751 } 752 753 int 754 lfs_rmdir(void *v) 755 { 756 struct vop_rmdir_args /* { 757 struct vnodeop_desc *a_desc; 758 struct vnode *a_dvp; 759 struct vnode *a_vp; 760 struct componentname *a_cnp; 761 } */ *ap = v; 762 struct vnode *vp; 763 struct inode *ip; 764 int error; 765 766 vp = ap->a_vp; 767 ip = VTOI(vp); 768 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 769 if (ap->a_dvp == vp) 770 vrele(ap->a_dvp); 771 else 772 vput(ap->a_dvp); 773 vput(vp); 774 return error; 775 } 776 error = ufs_rmdir(ap); 777 if (ip->i_nlink == 0) 778 lfs_orphan(ip->i_lfs, ip->i_number); 779 SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 780 return (error); 781 } 782 783 int 784 lfs_link(void *v) 785 { 786 struct vop_link_args /* { 787 struct vnode *a_dvp; 788 struct vnode *a_vp; 789 struct componentname *a_cnp; 790 } */ *ap = v; 791 int error; 792 struct vnode **vpp = NULL; 793 794 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 795 vput(ap->a_dvp); 796 return error; 797 } 798 error = ufs_link(ap); 799 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 800 return (error); 801 } 802 803 int 804 lfs_rename(void *v) 805 { 806 struct vop_rename_args /* { 807 struct vnode *a_fdvp; 808 struct vnode *a_fvp; 809 struct componentname *a_fcnp; 810 struct vnode *a_tdvp; 811 struct vnode *a_tvp; 812 struct componentname *a_tcnp; 813 } */ *ap = v; 814 struct vnode *tvp, *fvp, *tdvp, *fdvp; 815 struct componentname *tcnp, *fcnp; 816 int error; 817 struct lfs *fs; 818 819 fs = VTOI(ap->a_fdvp)->i_lfs; 820 tvp = ap->a_tvp; 821 tdvp = ap->a_tdvp; 822 tcnp = ap->a_tcnp; 823 fvp = ap->a_fvp; 824 fdvp = ap->a_fdvp; 825 fcnp = ap->a_fcnp; 826 827 /* 828 * Check for cross-device rename. 829 * If it is, we don't want to set dirops, just error out. 830 * (In particular note that MARK_VNODE(tdvp) will DTWT on 831 * a cross-device rename.) 832 * 833 * Copied from ufs_rename. 834 */ 835 if ((fvp->v_mount != tdvp->v_mount) || 836 (tvp && (fvp->v_mount != tvp->v_mount))) { 837 error = EXDEV; 838 goto errout; 839 } 840 841 /* 842 * Check to make sure we're not renaming a vnode onto itself 843 * (deleting a hard link by renaming one name onto another); 844 * if we are we can't recursively call VOP_REMOVE since that 845 * would leave us with an unaccounted-for number of live dirops. 846 * 847 * Inline the relevant section of ufs_rename here, *before* 848 * calling SET_DIROP_REMOVE. 849 */ 850 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 851 (VTOI(tdvp)->i_flags & APPEND))) { 852 error = EPERM; 853 goto errout; 854 } 855 if (fvp == tvp) { 856 if (fvp->v_type == VDIR) { 857 error = EINVAL; 858 goto errout; 859 } 860 861 /* Release destination completely. */ 862 VOP_ABORTOP(tdvp, tcnp); 863 vput(tdvp); 864 vput(tvp); 865 866 /* Delete source. */ 867 vrele(fvp); 868 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 869 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 870 fcnp->cn_nameiop = DELETE; 871 vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY); 872 if ((error = relookup(fdvp, &fvp, fcnp))) { 873 vput(fdvp); 874 return (error); 875 } 876 return (VOP_REMOVE(fdvp, fvp, fcnp)); 877 } 878 879 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 880 goto errout; 881 MARK_VNODE(fdvp); 882 MARK_VNODE(fvp); 883 884 error = ufs_rename(ap); 885 UNMARK_VNODE(fdvp); 886 UNMARK_VNODE(fvp); 887 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 888 return (error); 889 890 errout: 891 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 892 if (tdvp == tvp) 893 vrele(tdvp); 894 else 895 vput(tdvp); 896 if (tvp) 897 vput(tvp); 898 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 899 vrele(fdvp); 900 vrele(fvp); 901 return (error); 902 } 903 904 /* XXX hack to avoid calling ITIMES in getattr */ 905 int 906 lfs_getattr(void *v) 907 { 908 struct vop_getattr_args /* { 909 struct vnode *a_vp; 910 struct vattr *a_vap; 911 kauth_cred_t a_cred; 912 } */ *ap = v; 913 struct vnode *vp = ap->a_vp; 914 struct inode *ip = VTOI(vp); 915 struct vattr *vap = ap->a_vap; 916 struct lfs *fs = ip->i_lfs; 917 /* 918 * Copy from inode table 919 */ 920 vap->va_fsid = ip->i_dev; 921 vap->va_fileid = ip->i_number; 922 vap->va_mode = ip->i_mode & ~IFMT; 923 vap->va_nlink = ip->i_nlink; 924 vap->va_uid = ip->i_uid; 925 vap->va_gid = ip->i_gid; 926 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 927 vap->va_size = vp->v_size; 928 vap->va_atime.tv_sec = ip->i_ffs1_atime; 929 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 930 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 931 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 932 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 933 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 934 vap->va_flags = ip->i_flags; 935 vap->va_gen = ip->i_gen; 936 /* this doesn't belong here */ 937 if (vp->v_type == VBLK) 938 vap->va_blocksize = BLKDEV_IOSIZE; 939 else if (vp->v_type == VCHR) 940 vap->va_blocksize = MAXBSIZE; 941 else 942 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 943 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 944 vap->va_type = vp->v_type; 945 vap->va_filerev = ip->i_modrev; 946 return (0); 947 } 948 949 /* 950 * Check to make sure the inode blocks won't choke the buffer 951 * cache, then call ufs_setattr as usual. 952 */ 953 int 954 lfs_setattr(void *v) 955 { 956 struct vop_setattr_args /* { 957 struct vnode *a_vp; 958 struct vattr *a_vap; 959 kauth_cred_t a_cred; 960 } */ *ap = v; 961 struct vnode *vp = ap->a_vp; 962 963 lfs_check(vp, LFS_UNUSED_LBN, 0); 964 return ufs_setattr(v); 965 } 966 967 /* 968 * Release the block we hold on lfs_newseg wrapping. Called on file close, 969 * or explicitly from LFCNWRAPGO. Called with the interlock held. 970 */ 971 static int 972 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor) 973 { 974 if (lockstatus(&fs->lfs_stoplock) != LK_EXCLUSIVE) 975 return EBUSY; 976 977 lockmgr(&fs->lfs_stoplock, LK_RELEASE, &fs->lfs_interlock); 978 979 KASSERT(fs->lfs_nowrap > 0); 980 if (fs->lfs_nowrap <= 0) { 981 return 0; 982 } 983 984 if (--fs->lfs_nowrap == 0) { 985 log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt); 986 wakeup(&fs->lfs_wrappass); 987 lfs_wakeup_cleaner(fs); 988 } 989 if (waitfor) { 990 ltsleep(&fs->lfs_nextseg, PCATCH | PUSER, 991 "segment", 0, &fs->lfs_interlock); 992 } 993 994 return 0; 995 } 996 997 /* 998 * Close called 999 */ 1000 /* ARGSUSED */ 1001 int 1002 lfs_close(void *v) 1003 { 1004 struct vop_close_args /* { 1005 struct vnode *a_vp; 1006 int a_fflag; 1007 kauth_cred_t a_cred; 1008 } */ *ap = v; 1009 struct vnode *vp = ap->a_vp; 1010 struct inode *ip = VTOI(vp); 1011 struct lfs *fs = ip->i_lfs; 1012 1013 if ((ip->i_number == ROOTINO || ip->i_number == LFS_IFILE_INUM) && 1014 lockstatus(&fs->lfs_stoplock) == LK_EXCLUSIVE) { 1015 simple_lock(&fs->lfs_interlock); 1016 log(LOG_NOTICE, "lfs_close: releasing log wrap control\n"); 1017 lfs_wrapgo(fs, ip, 0); 1018 simple_unlock(&fs->lfs_interlock); 1019 } 1020 1021 if (vp == ip->i_lfs->lfs_ivnode && 1022 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 1023 return 0; 1024 1025 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 1026 LFS_ITIMES(ip, NULL, NULL, NULL); 1027 } 1028 return (0); 1029 } 1030 1031 /* 1032 * Close wrapper for special devices. 1033 * 1034 * Update the times on the inode then do device close. 1035 */ 1036 int 1037 lfsspec_close(void *v) 1038 { 1039 struct vop_close_args /* { 1040 struct vnode *a_vp; 1041 int a_fflag; 1042 kauth_cred_t a_cred; 1043 } */ *ap = v; 1044 struct vnode *vp; 1045 struct inode *ip; 1046 1047 vp = ap->a_vp; 1048 ip = VTOI(vp); 1049 if (vp->v_usecount > 1) { 1050 LFS_ITIMES(ip, NULL, NULL, NULL); 1051 } 1052 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1053 } 1054 1055 /* 1056 * Close wrapper for fifo's. 1057 * 1058 * Update the times on the inode then do device close. 1059 */ 1060 int 1061 lfsfifo_close(void *v) 1062 { 1063 struct vop_close_args /* { 1064 struct vnode *a_vp; 1065 int a_fflag; 1066 kauth_cred_ a_cred; 1067 } */ *ap = v; 1068 struct vnode *vp; 1069 struct inode *ip; 1070 1071 vp = ap->a_vp; 1072 ip = VTOI(vp); 1073 if (ap->a_vp->v_usecount > 1) { 1074 LFS_ITIMES(ip, NULL, NULL, NULL); 1075 } 1076 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1077 } 1078 1079 /* 1080 * Reclaim an inode so that it can be used for other purposes. 1081 */ 1082 1083 int 1084 lfs_reclaim(void *v) 1085 { 1086 struct vop_reclaim_args /* { 1087 struct vnode *a_vp; 1088 } */ *ap = v; 1089 struct vnode *vp = ap->a_vp; 1090 struct inode *ip = VTOI(vp); 1091 struct lfs *fs = ip->i_lfs; 1092 int error; 1093 1094 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1095 1096 LFS_CLR_UINO(ip, IN_ALLMOD); 1097 if ((error = ufs_reclaim(vp))) 1098 return (error); 1099 1100 /* 1101 * Take us off the paging and/or dirop queues if we were on them. 1102 * We shouldn't be on them. 1103 */ 1104 simple_lock(&fs->lfs_interlock); 1105 if (ip->i_flags & IN_PAGING) { 1106 log(LOG_WARNING, "%s: reclaimed vnode is IN_PAGING\n", 1107 fs->lfs_fsmnt); 1108 ip->i_flags &= ~IN_PAGING; 1109 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1110 } 1111 if (vp->v_uflag & VU_DIROP) { 1112 panic("reclaimed vnode is VU_DIROP"); 1113 vp->v_uflag &= ~VU_DIROP; 1114 TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain); 1115 } 1116 simple_unlock(&fs->lfs_interlock); 1117 1118 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1119 lfs_deregister_all(vp); 1120 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1121 ip->inode_ext.lfs = NULL; 1122 genfs_node_destroy(vp); 1123 pool_put(&lfs_inode_pool, vp->v_data); 1124 vp->v_data = NULL; 1125 return (0); 1126 } 1127 1128 /* 1129 * Read a block from a storage device. 1130 * In order to avoid reading blocks that are in the process of being 1131 * written by the cleaner---and hence are not mutexed by the normal 1132 * buffer cache / page cache mechanisms---check for collisions before 1133 * reading. 1134 * 1135 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1136 * the active cleaner test. 1137 * 1138 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1139 */ 1140 int 1141 lfs_strategy(void *v) 1142 { 1143 struct vop_strategy_args /* { 1144 struct vnode *a_vp; 1145 struct buf *a_bp; 1146 } */ *ap = v; 1147 struct buf *bp; 1148 struct lfs *fs; 1149 struct vnode *vp; 1150 struct inode *ip; 1151 daddr_t tbn; 1152 int i, sn, error, slept; 1153 1154 bp = ap->a_bp; 1155 vp = ap->a_vp; 1156 ip = VTOI(vp); 1157 fs = ip->i_lfs; 1158 1159 /* lfs uses its strategy routine only for read */ 1160 KASSERT(bp->b_flags & B_READ); 1161 1162 if (vp->v_type == VBLK || vp->v_type == VCHR) 1163 panic("lfs_strategy: spec"); 1164 KASSERT(bp->b_bcount != 0); 1165 if (bp->b_blkno == bp->b_lblkno) { 1166 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1167 NULL); 1168 if (error) { 1169 bp->b_error = error; 1170 biodone(bp); 1171 return (error); 1172 } 1173 if ((long)bp->b_blkno == -1) /* no valid data */ 1174 clrbuf(bp); 1175 } 1176 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1177 biodone(bp); 1178 return (0); 1179 } 1180 1181 slept = 1; 1182 simple_lock(&fs->lfs_interlock); 1183 while (slept && fs->lfs_seglock) { 1184 simple_unlock(&fs->lfs_interlock); 1185 /* 1186 * Look through list of intervals. 1187 * There will only be intervals to look through 1188 * if the cleaner holds the seglock. 1189 * Since the cleaner is synchronous, we can trust 1190 * the list of intervals to be current. 1191 */ 1192 tbn = dbtofsb(fs, bp->b_blkno); 1193 sn = dtosn(fs, tbn); 1194 slept = 0; 1195 for (i = 0; i < fs->lfs_cleanind; i++) { 1196 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1197 tbn >= fs->lfs_cleanint[i]) { 1198 DLOG((DLOG_CLEAN, 1199 "lfs_strategy: ino %d lbn %" PRId64 1200 " ind %d sn %d fsb %" PRIx32 1201 " given sn %d fsb %" PRIx64 "\n", 1202 ip->i_number, bp->b_lblkno, i, 1203 dtosn(fs, fs->lfs_cleanint[i]), 1204 fs->lfs_cleanint[i], sn, tbn)); 1205 DLOG((DLOG_CLEAN, 1206 "lfs_strategy: sleeping on ino %d lbn %" 1207 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1208 simple_lock(&fs->lfs_interlock); 1209 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1210 /* Cleaner can't wait for itself */ 1211 ltsleep(&fs->lfs_iocount, 1212 (PRIBIO + 1) | PNORELOCK, 1213 "clean2", 0, 1214 &fs->lfs_interlock); 1215 slept = 1; 1216 break; 1217 } else if (fs->lfs_seglock) { 1218 ltsleep(&fs->lfs_seglock, 1219 (PRIBIO + 1) | PNORELOCK, 1220 "clean1", 0, 1221 &fs->lfs_interlock); 1222 slept = 1; 1223 break; 1224 } 1225 simple_unlock(&fs->lfs_interlock); 1226 } 1227 } 1228 simple_lock(&fs->lfs_interlock); 1229 } 1230 simple_unlock(&fs->lfs_interlock); 1231 1232 vp = ip->i_devvp; 1233 VOP_STRATEGY(vp, bp); 1234 return (0); 1235 } 1236 1237 void 1238 lfs_flush_dirops(struct lfs *fs) 1239 { 1240 struct inode *ip, *nip; 1241 struct vnode *vp; 1242 extern int lfs_dostats; 1243 struct segment *sp; 1244 int waslocked; 1245 1246 ASSERT_MAYBE_SEGLOCK(fs); 1247 KASSERT(fs->lfs_nadirop == 0); 1248 1249 if (fs->lfs_ronly) 1250 return; 1251 1252 simple_lock(&fs->lfs_interlock); 1253 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1254 simple_unlock(&fs->lfs_interlock); 1255 return; 1256 } else 1257 simple_unlock(&fs->lfs_interlock); 1258 1259 if (lfs_dostats) 1260 ++lfs_stats.flush_invoked; 1261 1262 /* 1263 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1264 * Technically this is a checkpoint (the on-disk state is valid) 1265 * even though we are leaving out all the file data. 1266 */ 1267 lfs_imtime(fs); 1268 lfs_seglock(fs, SEGM_CKP); 1269 sp = fs->lfs_sp; 1270 1271 /* 1272 * lfs_writevnodes, optimized to get dirops out of the way. 1273 * Only write dirops, and don't flush files' pages, only 1274 * blocks from the directories. 1275 * 1276 * We don't need to vref these files because they are 1277 * dirops and so hold an extra reference until the 1278 * segunlock clears them of that status. 1279 * 1280 * We don't need to check for IN_ADIROP because we know that 1281 * no dirops are active. 1282 * 1283 */ 1284 simple_lock(&fs->lfs_interlock); 1285 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1286 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1287 simple_unlock(&fs->lfs_interlock); 1288 vp = ITOV(ip); 1289 1290 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1291 1292 /* 1293 * All writes to directories come from dirops; all 1294 * writes to files' direct blocks go through the page 1295 * cache, which we're not touching. Reads to files 1296 * and/or directories will not be affected by writing 1297 * directory blocks inodes and file inodes. So we don't 1298 * really need to lock. If we don't lock, though, 1299 * make sure that we don't clear IN_MODIFIED 1300 * unnecessarily. 1301 */ 1302 if (vp->v_iflag & (VI_XLOCK | VI_FREEING)) { 1303 simple_lock(&fs->lfs_interlock); 1304 continue; 1305 } 1306 waslocked = VOP_ISLOCKED(vp); 1307 if (vp->v_type != VREG && 1308 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1309 lfs_writefile(fs, sp, vp); 1310 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1311 !(ip->i_flag & IN_ALLMOD)) { 1312 LFS_SET_UINO(ip, IN_MODIFIED); 1313 } 1314 } 1315 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1316 (void) lfs_writeinode(fs, sp, ip); 1317 if (waslocked == LK_EXCLOTHER) 1318 LFS_SET_UINO(ip, IN_MODIFIED); 1319 simple_lock(&fs->lfs_interlock); 1320 } 1321 simple_unlock(&fs->lfs_interlock); 1322 /* We've written all the dirops there are */ 1323 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1324 lfs_finalize_fs_seguse(fs); 1325 (void) lfs_writeseg(fs, sp); 1326 lfs_segunlock(fs); 1327 } 1328 1329 /* 1330 * Flush all vnodes for which the pagedaemon has requested pageouts. 1331 * Skip over any files that are marked VU_DIROP (since lfs_flush_dirop() 1332 * has just run, this would be an error). If we have to skip a vnode 1333 * for any reason, just skip it; if we have to wait for the cleaner, 1334 * abort. The writer daemon will call us again later. 1335 */ 1336 void 1337 lfs_flush_pchain(struct lfs *fs) 1338 { 1339 struct inode *ip, *nip; 1340 struct vnode *vp; 1341 extern int lfs_dostats; 1342 struct segment *sp; 1343 int error; 1344 1345 ASSERT_NO_SEGLOCK(fs); 1346 1347 if (fs->lfs_ronly) 1348 return; 1349 1350 simple_lock(&fs->lfs_interlock); 1351 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1352 simple_unlock(&fs->lfs_interlock); 1353 return; 1354 } else 1355 simple_unlock(&fs->lfs_interlock); 1356 1357 /* Get dirops out of the way */ 1358 lfs_flush_dirops(fs); 1359 1360 if (lfs_dostats) 1361 ++lfs_stats.flush_invoked; 1362 1363 /* 1364 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1365 */ 1366 lfs_imtime(fs); 1367 lfs_seglock(fs, 0); 1368 sp = fs->lfs_sp; 1369 1370 /* 1371 * lfs_writevnodes, optimized to clear pageout requests. 1372 * Only write non-dirop files that are in the pageout queue. 1373 * We're very conservative about what we write; we want to be 1374 * fast and async. 1375 */ 1376 simple_lock(&fs->lfs_interlock); 1377 top: 1378 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1379 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1380 vp = ITOV(ip); 1381 1382 if (!(ip->i_flags & IN_PAGING)) 1383 goto top; 1384 1385 if ((vp->v_iflag|vp->v_uflag) & (VI_XLOCK|VU_DIROP)) 1386 continue; 1387 if (vp->v_type != VREG) 1388 continue; 1389 if (lfs_vref(vp)) 1390 continue; 1391 simple_unlock(&fs->lfs_interlock); 1392 1393 if (VOP_ISLOCKED(vp)) { 1394 lfs_vunref(vp); 1395 simple_lock(&fs->lfs_interlock); 1396 continue; 1397 } 1398 1399 error = lfs_writefile(fs, sp, vp); 1400 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1401 !(ip->i_flag & IN_ALLMOD)) { 1402 LFS_SET_UINO(ip, IN_MODIFIED); 1403 } 1404 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1405 (void) lfs_writeinode(fs, sp, ip); 1406 1407 lfs_vunref(vp); 1408 1409 if (error == EAGAIN) { 1410 lfs_writeseg(fs, sp); 1411 simple_lock(&fs->lfs_interlock); 1412 break; 1413 } 1414 simple_lock(&fs->lfs_interlock); 1415 } 1416 simple_unlock(&fs->lfs_interlock); 1417 (void) lfs_writeseg(fs, sp); 1418 lfs_segunlock(fs); 1419 } 1420 1421 /* 1422 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1423 */ 1424 int 1425 lfs_fcntl(void *v) 1426 { 1427 struct vop_fcntl_args /* { 1428 struct vnode *a_vp; 1429 u_long a_command; 1430 void * a_data; 1431 int a_fflag; 1432 kauth_cred_t a_cred; 1433 } */ *ap = v; 1434 struct timeval *tvp; 1435 BLOCK_INFO *blkiov; 1436 CLEANERINFO *cip; 1437 SEGUSE *sup; 1438 int blkcnt, error, oclean; 1439 size_t fh_size; 1440 struct lfs_fcntl_markv blkvp; 1441 struct lwp *l; 1442 fsid_t *fsidp; 1443 struct lfs *fs; 1444 struct buf *bp; 1445 fhandle_t *fhp; 1446 daddr_t off; 1447 1448 /* Only respect LFS fcntls on fs root or Ifile */ 1449 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1450 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1451 return ufs_fcntl(v); 1452 } 1453 1454 /* Avoid locking a draining lock */ 1455 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1456 return ESHUTDOWN; 1457 } 1458 1459 /* LFS control and monitoring fcntls are available only to root */ 1460 l = curlwp; 1461 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1462 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1463 NULL)) != 0) 1464 return (error); 1465 1466 fs = VTOI(ap->a_vp)->i_lfs; 1467 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1468 1469 error = 0; 1470 switch (ap->a_command) { 1471 case LFCNSEGWAITALL: 1472 case LFCNSEGWAITALL_COMPAT: 1473 fsidp = NULL; 1474 /* FALLSTHROUGH */ 1475 case LFCNSEGWAIT: 1476 case LFCNSEGWAIT_COMPAT: 1477 tvp = (struct timeval *)ap->a_data; 1478 simple_lock(&fs->lfs_interlock); 1479 ++fs->lfs_sleepers; 1480 simple_unlock(&fs->lfs_interlock); 1481 1482 error = lfs_segwait(fsidp, tvp); 1483 1484 simple_lock(&fs->lfs_interlock); 1485 if (--fs->lfs_sleepers == 0) 1486 wakeup(&fs->lfs_sleepers); 1487 simple_unlock(&fs->lfs_interlock); 1488 return error; 1489 1490 case LFCNBMAPV: 1491 case LFCNMARKV: 1492 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1493 1494 blkcnt = blkvp.blkcnt; 1495 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1496 return (EINVAL); 1497 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1498 if ((error = copyin(blkvp.blkiov, blkiov, 1499 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1500 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1501 return error; 1502 } 1503 1504 simple_lock(&fs->lfs_interlock); 1505 ++fs->lfs_sleepers; 1506 simple_unlock(&fs->lfs_interlock); 1507 if (ap->a_command == LFCNBMAPV) 1508 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1509 else /* LFCNMARKV */ 1510 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1511 if (error == 0) 1512 error = copyout(blkiov, blkvp.blkiov, 1513 blkcnt * sizeof(BLOCK_INFO)); 1514 simple_lock(&fs->lfs_interlock); 1515 if (--fs->lfs_sleepers == 0) 1516 wakeup(&fs->lfs_sleepers); 1517 simple_unlock(&fs->lfs_interlock); 1518 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1519 return error; 1520 1521 case LFCNRECLAIM: 1522 /* 1523 * Flush dirops and write Ifile, allowing empty segments 1524 * to be immediately reclaimed. 1525 */ 1526 lfs_writer_enter(fs, "pndirop"); 1527 off = fs->lfs_offset; 1528 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1529 lfs_flush_dirops(fs); 1530 LFS_CLEANERINFO(cip, fs, bp); 1531 oclean = cip->clean; 1532 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1533 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1534 fs->lfs_sp->seg_flags |= SEGM_PROT; 1535 lfs_segunlock(fs); 1536 lfs_writer_leave(fs); 1537 1538 #ifdef DEBUG 1539 LFS_CLEANERINFO(cip, fs, bp); 1540 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1541 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1542 fs->lfs_offset - off, cip->clean - oclean, 1543 fs->lfs_activesb)); 1544 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1545 #endif 1546 1547 return 0; 1548 1549 #ifdef COMPAT_30 1550 case LFCNIFILEFH_COMPAT: 1551 /* Return the filehandle of the Ifile */ 1552 if ((error = kauth_authorize_generic(l->l_cred, 1553 KAUTH_GENERIC_ISSUSER, NULL)) != 0) 1554 return (error); 1555 fhp = (struct fhandle *)ap->a_data; 1556 fhp->fh_fsid = *fsidp; 1557 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1558 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1559 #endif 1560 1561 case LFCNIFILEFH_COMPAT2: 1562 case LFCNIFILEFH: 1563 /* Return the filehandle of the Ifile */ 1564 fhp = (struct fhandle *)ap->a_data; 1565 fhp->fh_fsid = *fsidp; 1566 fh_size = sizeof(struct lfs_fhandle) - 1567 offsetof(fhandle_t, fh_fid); 1568 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1569 1570 case LFCNREWIND: 1571 /* Move lfs_offset to the lowest-numbered segment */ 1572 return lfs_rewind(fs, *(int *)ap->a_data); 1573 1574 case LFCNINVAL: 1575 /* Mark a segment SEGUSE_INVAL */ 1576 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1577 if (sup->su_nbytes > 0) { 1578 brelse(bp, 0); 1579 lfs_unset_inval_all(fs); 1580 return EBUSY; 1581 } 1582 sup->su_flags |= SEGUSE_INVAL; 1583 VOP_BWRITE(bp); 1584 return 0; 1585 1586 case LFCNRESIZE: 1587 /* Resize the filesystem */ 1588 return lfs_resize_fs(fs, *(int *)ap->a_data); 1589 1590 case LFCNWRAPSTOP: 1591 case LFCNWRAPSTOP_COMPAT: 1592 /* 1593 * Hold lfs_newseg at segment 0; if requested, sleep until 1594 * the filesystem wraps around. To support external agents 1595 * (dump, fsck-based regression test) that need to look at 1596 * a snapshot of the filesystem, without necessarily 1597 * requiring that all fs activity stops. 1598 */ 1599 if (lockstatus(&fs->lfs_stoplock)) 1600 return EALREADY; 1601 1602 simple_lock(&fs->lfs_interlock); 1603 lockmgr(&fs->lfs_stoplock, LK_EXCLUSIVE, &fs->lfs_interlock); 1604 if (fs->lfs_nowrap == 0) 1605 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1606 ++fs->lfs_nowrap; 1607 if (*(int *)ap->a_data == 1 || 1608 ap->a_command == LFCNWRAPSTOP_COMPAT) { 1609 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1610 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1611 "segwrap", 0, &fs->lfs_interlock); 1612 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1613 if (error) { 1614 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1615 } 1616 } 1617 simple_unlock(&fs->lfs_interlock); 1618 return 0; 1619 1620 case LFCNWRAPGO: 1621 case LFCNWRAPGO_COMPAT: 1622 /* 1623 * Having done its work, the agent wakes up the writer. 1624 * If the argument is 1, it sleeps until a new segment 1625 * is selected. 1626 */ 1627 simple_lock(&fs->lfs_interlock); 1628 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1629 (ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1630 *((int *)ap->a_data))); 1631 simple_unlock(&fs->lfs_interlock); 1632 return error; 1633 1634 case LFCNWRAPPASS: 1635 if (lockstatus(&fs->lfs_stoplock) != LK_EXCLUSIVE) 1636 return EALREADY; 1637 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1638 return EALREADY; 1639 simple_lock(&fs->lfs_interlock); 1640 if (fs->lfs_nowrap == 0) { 1641 simple_unlock(&fs->lfs_interlock); 1642 return EBUSY; 1643 } 1644 fs->lfs_wrappass = 1; 1645 wakeup(&fs->lfs_wrappass); 1646 /* Wait for the log to wrap, if asked */ 1647 if (*(int *)ap->a_data) { 1648 lfs_vref(ap->a_vp); 1649 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1650 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1651 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1652 "segwrap", 0, &fs->lfs_interlock); 1653 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1654 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1655 lfs_vunref(ap->a_vp); 1656 } 1657 simple_unlock(&fs->lfs_interlock); 1658 return error; 1659 1660 case LFCNWRAPSTATUS: 1661 simple_lock(&fs->lfs_interlock); 1662 *(int *)ap->a_data = fs->lfs_wrapstatus; 1663 simple_unlock(&fs->lfs_interlock); 1664 return 0; 1665 1666 default: 1667 return ufs_fcntl(v); 1668 } 1669 return 0; 1670 } 1671 1672 int 1673 lfs_getpages(void *v) 1674 { 1675 struct vop_getpages_args /* { 1676 struct vnode *a_vp; 1677 voff_t a_offset; 1678 struct vm_page **a_m; 1679 int *a_count; 1680 int a_centeridx; 1681 vm_prot_t a_access_type; 1682 int a_advice; 1683 int a_flags; 1684 } */ *ap = v; 1685 1686 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1687 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1688 return EPERM; 1689 } 1690 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1691 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1692 } 1693 1694 /* 1695 * we're relying on the fact that genfs_getpages() always read in 1696 * entire filesystem blocks. 1697 */ 1698 return genfs_getpages(v); 1699 } 1700 1701 /* 1702 * Wait for a page to become unbusy, possibly printing diagnostic messages 1703 * as well. 1704 * 1705 * Called with vp->v_interlock held; return with it held. 1706 */ 1707 static void 1708 wait_for_page(struct vnode *vp, struct vm_page *pg, const char *label) 1709 { 1710 if ((pg->flags & PG_BUSY) == 0) 1711 return; /* Nothing to wait for! */ 1712 1713 #if defined(DEBUG) && defined(UVM_PAGE_TRKOWN) 1714 static struct vm_page *lastpg; 1715 1716 if (label != NULL && pg != lastpg) { 1717 if (pg->owner_tag) { 1718 printf("lfs_putpages[%d.%d]: %s: page %p owner %d.%d [%s]\n", 1719 curproc->p_pid, curlwp->l_lid, label, 1720 pg, pg->owner, pg->lowner, pg->owner_tag); 1721 } else { 1722 printf("lfs_putpages[%d.%d]: %s: page %p unowned?!\n", 1723 curproc->p_pid, curlwp->l_lid, label, pg); 1724 } 1725 } 1726 lastpg = pg; 1727 #endif 1728 1729 pg->flags |= PG_WANTED; 1730 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, "lfsput", 0); 1731 simple_lock(&vp->v_interlock); 1732 } 1733 1734 /* 1735 * This routine is called by lfs_putpages() when it can't complete the 1736 * write because a page is busy. This means that either (1) someone, 1737 * possibly the pagedaemon, is looking at this page, and will give it up 1738 * presently; or (2) we ourselves are holding the page busy in the 1739 * process of being written (either gathered or actually on its way to 1740 * disk). We don't need to give up the segment lock, but we might need 1741 * to call lfs_writeseg() to expedite the page's journey to disk. 1742 * 1743 * Called with vp->v_interlock held; return with it held. 1744 */ 1745 /* #define BUSYWAIT */ 1746 static void 1747 write_and_wait(struct lfs *fs, struct vnode *vp, struct vm_page *pg, 1748 int seglocked, const char *label) 1749 { 1750 #ifndef BUSYWAIT 1751 struct inode *ip = VTOI(vp); 1752 struct segment *sp = fs->lfs_sp; 1753 int count = 0; 1754 1755 if (pg == NULL) 1756 return; 1757 1758 while (pg->flags & PG_BUSY) { 1759 simple_unlock(&vp->v_interlock); 1760 if (sp->cbpp - sp->bpp > 1) { 1761 /* Write gathered pages */ 1762 lfs_updatemeta(sp); 1763 lfs_release_finfo(fs); 1764 (void) lfs_writeseg(fs, sp); 1765 1766 /* 1767 * Reinitialize FIP 1768 */ 1769 KASSERT(sp->vp == vp); 1770 lfs_acquire_finfo(fs, ip->i_number, 1771 ip->i_gen); 1772 } 1773 ++count; 1774 simple_lock(&vp->v_interlock); 1775 wait_for_page(vp, pg, label); 1776 } 1777 if (label != NULL && count > 1) 1778 printf("lfs_putpages[%d]: %s: %sn = %d\n", curproc->p_pid, 1779 label, (count > 0 ? "looping, " : ""), count); 1780 #else 1781 preempt(1); 1782 #endif 1783 } 1784 1785 /* 1786 * Make sure that for all pages in every block in the given range, 1787 * either all are dirty or all are clean. If any of the pages 1788 * we've seen so far are dirty, put the vnode on the paging chain, 1789 * and mark it IN_PAGING. 1790 * 1791 * If checkfirst != 0, don't check all the pages but return at the 1792 * first dirty page. 1793 */ 1794 static int 1795 check_dirty(struct lfs *fs, struct vnode *vp, 1796 off_t startoffset, off_t endoffset, off_t blkeof, 1797 int flags, int checkfirst, struct vm_page **pgp) 1798 { 1799 int by_list; 1800 struct vm_page *curpg = NULL; /* XXX: gcc */ 1801 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1802 off_t soff = 0; /* XXX: gcc */ 1803 voff_t off; 1804 int i; 1805 int nonexistent; 1806 int any_dirty; /* number of dirty pages */ 1807 int dirty; /* number of dirty pages in a block */ 1808 int tdirty; 1809 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1810 int pagedaemon = (curlwp == uvm.pagedaemon_lwp); 1811 1812 ASSERT_MAYBE_SEGLOCK(fs); 1813 top: 1814 by_list = (vp->v_uobj.uo_npages <= 1815 ((endoffset - startoffset) >> PAGE_SHIFT) * 1816 UVM_PAGE_HASH_PENALTY); 1817 any_dirty = 0; 1818 1819 if (by_list) { 1820 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1821 } else { 1822 soff = startoffset; 1823 } 1824 while (by_list || soff < MIN(blkeof, endoffset)) { 1825 if (by_list) { 1826 /* 1827 * Find the first page in a block. Skip 1828 * blocks outside our area of interest or beyond 1829 * the end of file. 1830 */ 1831 if (pages_per_block > 1) { 1832 while (curpg && 1833 ((curpg->offset & fs->lfs_bmask) || 1834 curpg->offset >= vp->v_size || 1835 curpg->offset >= endoffset)) 1836 curpg = TAILQ_NEXT(curpg, listq); 1837 } 1838 if (curpg == NULL) 1839 break; 1840 soff = curpg->offset; 1841 } 1842 1843 /* 1844 * Mark all pages in extended range busy; find out if any 1845 * of them are dirty. 1846 */ 1847 nonexistent = dirty = 0; 1848 for (i = 0; i == 0 || i < pages_per_block; i++) { 1849 if (by_list && pages_per_block <= 1) { 1850 pgs[i] = pg = curpg; 1851 } else { 1852 off = soff + (i << PAGE_SHIFT); 1853 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1854 if (pg == NULL) { 1855 ++nonexistent; 1856 continue; 1857 } 1858 } 1859 KASSERT(pg != NULL); 1860 1861 /* 1862 * If we're holding the segment lock, we can deadlock 1863 * against a process that has our page and is waiting 1864 * for the cleaner, while the cleaner waits for the 1865 * segment lock. Just bail in that case. 1866 */ 1867 if ((pg->flags & PG_BUSY) && 1868 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1869 if (i > 0) 1870 uvm_page_unbusy(pgs, i); 1871 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1872 if (pgp) 1873 *pgp = pg; 1874 return -1; 1875 } 1876 1877 while (pg->flags & PG_BUSY) { 1878 wait_for_page(vp, pg, NULL); 1879 if (i > 0) 1880 uvm_page_unbusy(pgs, i); 1881 goto top; 1882 } 1883 pg->flags |= PG_BUSY; 1884 UVM_PAGE_OWN(pg, "lfs_putpages"); 1885 1886 pmap_page_protect(pg, VM_PROT_NONE); 1887 tdirty = (pmap_clear_modify(pg) || 1888 (pg->flags & PG_CLEAN) == 0); 1889 dirty += tdirty; 1890 } 1891 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1892 if (by_list) { 1893 curpg = TAILQ_NEXT(curpg, listq); 1894 } else { 1895 soff += fs->lfs_bsize; 1896 } 1897 continue; 1898 } 1899 1900 any_dirty += dirty; 1901 KASSERT(nonexistent == 0); 1902 1903 /* 1904 * If any are dirty make all dirty; unbusy them, 1905 * but if we were asked to clean, wire them so that 1906 * the pagedaemon doesn't bother us about them while 1907 * they're on their way to disk. 1908 */ 1909 for (i = 0; i == 0 || i < pages_per_block; i++) { 1910 pg = pgs[i]; 1911 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1912 if (dirty) { 1913 pg->flags &= ~PG_CLEAN; 1914 if (flags & PGO_FREE) { 1915 /* 1916 * Wire the page so that 1917 * pdaemon doesn't see it again. 1918 */ 1919 uvm_lock_pageq(); 1920 uvm_pagewire(pg); 1921 uvm_unlock_pageq(); 1922 1923 /* Suspended write flag */ 1924 pg->flags |= PG_DELWRI; 1925 } 1926 } 1927 if (pg->flags & PG_WANTED) 1928 wakeup(pg); 1929 pg->flags &= ~(PG_WANTED|PG_BUSY); 1930 UVM_PAGE_OWN(pg, NULL); 1931 } 1932 1933 if (checkfirst && any_dirty) 1934 break; 1935 1936 if (by_list) { 1937 curpg = TAILQ_NEXT(curpg, listq); 1938 } else { 1939 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1940 } 1941 } 1942 1943 return any_dirty; 1944 } 1945 1946 /* 1947 * lfs_putpages functions like genfs_putpages except that 1948 * 1949 * (1) It needs to bounds-check the incoming requests to ensure that 1950 * they are block-aligned; if they are not, expand the range and 1951 * do the right thing in case, e.g., the requested range is clean 1952 * but the expanded range is dirty. 1953 * 1954 * (2) It needs to explicitly send blocks to be written when it is done. 1955 * If VOP_PUTPAGES is called without the seglock held, we simply take 1956 * the seglock and let lfs_segunlock wait for us. 1957 * XXX There might be a bad situation if we have to flush a vnode while 1958 * XXX lfs_markv is in operation. As of this writing we panic in this 1959 * XXX case. 1960 * 1961 * Assumptions: 1962 * 1963 * (1) The caller does not hold any pages in this vnode busy. If it does, 1964 * there is a danger that when we expand the page range and busy the 1965 * pages we will deadlock. 1966 * 1967 * (2) We are called with vp->v_interlock held; we must return with it 1968 * released. 1969 * 1970 * (3) We don't absolutely have to free pages right away, provided that 1971 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1972 * us a request with PGO_FREE, we take the pages out of the paging 1973 * queue and wake up the writer, which will handle freeing them for us. 1974 * 1975 * We ensure that for any filesystem block, all pages for that 1976 * block are either resident or not, even if those pages are higher 1977 * than EOF; that means that we will be getting requests to free 1978 * "unused" pages above EOF all the time, and should ignore them. 1979 * 1980 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 1981 * into has been set up for us by lfs_writefile. If not, we will 1982 * have to handle allocating and/or freeing an finfo entry. 1983 * 1984 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1985 */ 1986 1987 /* How many times to loop before we should start to worry */ 1988 #define TOOMANY 4 1989 1990 int 1991 lfs_putpages(void *v) 1992 { 1993 int error; 1994 struct vop_putpages_args /* { 1995 struct vnode *a_vp; 1996 voff_t a_offlo; 1997 voff_t a_offhi; 1998 int a_flags; 1999 } */ *ap = v; 2000 struct vnode *vp; 2001 struct inode *ip; 2002 struct lfs *fs; 2003 struct segment *sp; 2004 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 2005 off_t off, max_endoffset; 2006 int s; 2007 bool seglocked, sync, pagedaemon; 2008 struct vm_page *pg, *busypg; 2009 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 2010 #ifdef DEBUG 2011 int debug_n_again, debug_n_dirtyclean; 2012 #endif 2013 2014 vp = ap->a_vp; 2015 ip = VTOI(vp); 2016 fs = ip->i_lfs; 2017 sync = (ap->a_flags & PGO_SYNCIO) != 0; 2018 pagedaemon = (curlwp == uvm.pagedaemon_lwp); 2019 2020 /* Putpages does nothing for metadata. */ 2021 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 2022 simple_unlock(&vp->v_interlock); 2023 return 0; 2024 } 2025 2026 /* 2027 * If there are no pages, don't do anything. 2028 */ 2029 if (vp->v_uobj.uo_npages == 0) { 2030 s = splbio(); 2031 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 2032 (vp->v_iflag & VI_ONWORKLST) && 2033 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 2034 vp->v_iflag &= ~VI_WRMAPDIRTY; 2035 vn_syncer_remove_from_worklist(vp); 2036 } 2037 splx(s); 2038 simple_unlock(&vp->v_interlock); 2039 2040 /* Remove us from paging queue, if we were on it */ 2041 simple_lock(&fs->lfs_interlock); 2042 if (ip->i_flags & IN_PAGING) { 2043 ip->i_flags &= ~IN_PAGING; 2044 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2045 } 2046 simple_unlock(&fs->lfs_interlock); 2047 return 0; 2048 } 2049 2050 blkeof = blkroundup(fs, ip->i_size); 2051 2052 /* 2053 * Ignore requests to free pages past EOF but in the same block 2054 * as EOF, unless the request is synchronous. (If the request is 2055 * sync, it comes from lfs_truncate.) 2056 * XXXUBC Make these pages look "active" so the pagedaemon won't 2057 * XXXUBC bother us with them again. 2058 */ 2059 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 2060 origoffset = ap->a_offlo; 2061 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 2062 pg = uvm_pagelookup(&vp->v_uobj, off); 2063 KASSERT(pg != NULL); 2064 while (pg->flags & PG_BUSY) { 2065 pg->flags |= PG_WANTED; 2066 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 2067 "lfsput2", 0); 2068 simple_lock(&vp->v_interlock); 2069 } 2070 uvm_lock_pageq(); 2071 uvm_pageactivate(pg); 2072 uvm_unlock_pageq(); 2073 } 2074 ap->a_offlo = blkeof; 2075 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 2076 simple_unlock(&vp->v_interlock); 2077 return 0; 2078 } 2079 } 2080 2081 /* 2082 * Extend page range to start and end at block boundaries. 2083 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 2084 */ 2085 origoffset = ap->a_offlo; 2086 origendoffset = ap->a_offhi; 2087 startoffset = origoffset & ~(fs->lfs_bmask); 2088 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 2089 << fs->lfs_bshift; 2090 2091 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2092 endoffset = max_endoffset; 2093 origendoffset = endoffset; 2094 } else { 2095 origendoffset = round_page(ap->a_offhi); 2096 endoffset = round_page(blkroundup(fs, origendoffset)); 2097 } 2098 2099 KASSERT(startoffset > 0 || endoffset >= startoffset); 2100 if (startoffset == endoffset) { 2101 /* Nothing to do, why were we called? */ 2102 simple_unlock(&vp->v_interlock); 2103 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 2104 PRId64 "\n", startoffset)); 2105 return 0; 2106 } 2107 2108 ap->a_offlo = startoffset; 2109 ap->a_offhi = endoffset; 2110 2111 /* 2112 * If not cleaning, just send the pages through genfs_putpages 2113 * to be returned to the pool. 2114 */ 2115 if (!(ap->a_flags & PGO_CLEANIT)) 2116 return genfs_putpages(v); 2117 2118 /* Set PGO_BUSYFAIL to avoid deadlocks */ 2119 ap->a_flags |= PGO_BUSYFAIL; 2120 2121 /* 2122 * Likewise, if we are asked to clean but the pages are not 2123 * dirty, we can just free them using genfs_putpages. 2124 */ 2125 #ifdef DEBUG 2126 debug_n_dirtyclean = 0; 2127 #endif 2128 do { 2129 int r; 2130 2131 /* Count the number of dirty pages */ 2132 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2133 ap->a_flags, 1, NULL); 2134 if (r < 0) { 2135 /* Pages are busy with another process */ 2136 simple_unlock(&vp->v_interlock); 2137 return EDEADLK; 2138 } 2139 if (r > 0) /* Some pages are dirty */ 2140 break; 2141 2142 /* 2143 * Sometimes pages are dirtied between the time that 2144 * we check and the time we try to clean them. 2145 * Instruct lfs_gop_write to return EDEADLK in this case 2146 * so we can write them properly. 2147 */ 2148 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2149 r = genfs_do_putpages(vp, startoffset, endoffset, 2150 ap->a_flags, &busypg); 2151 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2152 if (r != EDEADLK) 2153 return r; 2154 2155 /* One of the pages was busy. Start over. */ 2156 simple_lock(&vp->v_interlock); 2157 wait_for_page(vp, busypg, "dirtyclean"); 2158 #ifdef DEBUG 2159 ++debug_n_dirtyclean; 2160 #endif 2161 } while(1); 2162 2163 #ifdef DEBUG 2164 if (debug_n_dirtyclean > TOOMANY) 2165 printf("lfs_putpages: dirtyclean: looping, n = %d\n", 2166 debug_n_dirtyclean); 2167 #endif 2168 2169 /* 2170 * Dirty and asked to clean. 2171 * 2172 * Pagedaemon can't actually write LFS pages; wake up 2173 * the writer to take care of that. The writer will 2174 * notice the pager inode queue and act on that. 2175 */ 2176 if (pagedaemon) { 2177 simple_lock(&fs->lfs_interlock); 2178 if (!(ip->i_flags & IN_PAGING)) { 2179 ip->i_flags |= IN_PAGING; 2180 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2181 } 2182 simple_lock(&lfs_subsys_lock); 2183 wakeup(&lfs_writer_daemon); 2184 simple_unlock(&lfs_subsys_lock); 2185 simple_unlock(&fs->lfs_interlock); 2186 simple_unlock(&vp->v_interlock); 2187 preempt(); 2188 return EWOULDBLOCK; 2189 } 2190 2191 /* 2192 * If this is a file created in a recent dirop, we can't flush its 2193 * inode until the dirop is complete. Drain dirops, then flush the 2194 * filesystem (taking care of any other pending dirops while we're 2195 * at it). 2196 */ 2197 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2198 (vp->v_uflag & VU_DIROP)) { 2199 int locked; 2200 2201 DLOG((DLOG_PAGE, "lfs_putpages: flushing VU_DIROP\n")); 2202 locked = (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 2203 simple_unlock(&vp->v_interlock); 2204 lfs_writer_enter(fs, "ppdirop"); 2205 if (locked) 2206 VOP_UNLOCK(vp, 0); /* XXX why? */ 2207 2208 simple_lock(&fs->lfs_interlock); 2209 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2210 simple_unlock(&fs->lfs_interlock); 2211 2212 simple_lock(&vp->v_interlock); 2213 if (locked) { 2214 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2215 simple_lock(&vp->v_interlock); 2216 } 2217 lfs_writer_leave(fs); 2218 2219 /* XXX the flush should have taken care of this one too! */ 2220 } 2221 2222 /* 2223 * This is it. We are going to write some pages. From here on 2224 * down it's all just mechanics. 2225 * 2226 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2227 */ 2228 ap->a_flags &= ~PGO_SYNCIO; 2229 2230 /* 2231 * If we've already got the seglock, flush the node and return. 2232 * The FIP has already been set up for us by lfs_writefile, 2233 * and FIP cleanup and lfs_updatemeta will also be done there, 2234 * unless genfs_putpages returns EDEADLK; then we must flush 2235 * what we have, and correct FIP and segment header accounting. 2236 */ 2237 get_seglock: 2238 /* 2239 * If we are not called with the segment locked, lock it. 2240 * Account for a new FIP in the segment header, and set sp->vp. 2241 * (This should duplicate the setup at the top of lfs_writefile().) 2242 */ 2243 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2244 if (!seglocked) { 2245 simple_unlock(&vp->v_interlock); 2246 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2247 if (error != 0) 2248 return error; 2249 simple_lock(&vp->v_interlock); 2250 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2251 } 2252 sp = fs->lfs_sp; 2253 KASSERT(sp->vp == NULL); 2254 sp->vp = vp; 2255 2256 /* 2257 * Ensure that the partial segment is marked SS_DIROP if this 2258 * vnode is a DIROP. 2259 */ 2260 if (!seglocked && vp->v_uflag & VU_DIROP) 2261 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2262 2263 /* 2264 * Loop over genfs_putpages until all pages are gathered. 2265 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2266 * Whenever we lose the interlock we have to rerun check_dirty, as 2267 * well, since more pages might have been dirtied in our absence. 2268 */ 2269 #ifdef DEBUG 2270 debug_n_again = 0; 2271 #endif 2272 do { 2273 busypg = NULL; 2274 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2275 ap->a_flags, 0, &busypg) < 0) { 2276 simple_unlock(&vp->v_interlock); 2277 2278 simple_lock(&vp->v_interlock); 2279 write_and_wait(fs, vp, busypg, seglocked, NULL); 2280 if (!seglocked) { 2281 lfs_release_finfo(fs); 2282 lfs_segunlock(fs); 2283 } 2284 sp->vp = NULL; 2285 goto get_seglock; 2286 } 2287 2288 busypg = NULL; 2289 error = genfs_do_putpages(vp, startoffset, endoffset, 2290 ap->a_flags, &busypg); 2291 2292 if (error == EDEADLK || error == EAGAIN) { 2293 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2294 " %d ino %d off %x (seg %d)\n", error, 2295 ip->i_number, fs->lfs_offset, 2296 dtosn(fs, fs->lfs_offset))); 2297 2298 simple_lock(&vp->v_interlock); 2299 write_and_wait(fs, vp, busypg, seglocked, "again"); 2300 } 2301 #ifdef DEBUG 2302 ++debug_n_again; 2303 #endif 2304 } while (error == EDEADLK); 2305 #ifdef DEBUG 2306 if (debug_n_again > TOOMANY) 2307 printf("lfs_putpages: again: looping, n = %d\n", debug_n_again); 2308 #endif 2309 2310 KASSERT(sp != NULL && sp->vp == vp); 2311 if (!seglocked) { 2312 sp->vp = NULL; 2313 2314 /* Write indirect blocks as well */ 2315 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2316 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2317 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2318 2319 KASSERT(sp->vp == NULL); 2320 sp->vp = vp; 2321 } 2322 2323 /* 2324 * Blocks are now gathered into a segment waiting to be written. 2325 * All that's left to do is update metadata, and write them. 2326 */ 2327 lfs_updatemeta(sp); 2328 KASSERT(sp->vp == vp); 2329 sp->vp = NULL; 2330 2331 /* 2332 * If we were called from lfs_writefile, we don't need to clean up 2333 * the FIP or unlock the segment lock. We're done. 2334 */ 2335 if (seglocked) 2336 return error; 2337 2338 /* Clean up FIP and send it to disk. */ 2339 lfs_release_finfo(fs); 2340 lfs_writeseg(fs, fs->lfs_sp); 2341 2342 /* 2343 * Remove us from paging queue if we wrote all our pages. 2344 */ 2345 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2346 simple_lock(&fs->lfs_interlock); 2347 if (ip->i_flags & IN_PAGING) { 2348 ip->i_flags &= ~IN_PAGING; 2349 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2350 } 2351 simple_unlock(&fs->lfs_interlock); 2352 } 2353 2354 /* 2355 * XXX - with the malloc/copy writeseg, the pages are freed by now 2356 * even if we don't wait (e.g. if we hold a nested lock). This 2357 * will not be true if we stop using malloc/copy. 2358 */ 2359 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2360 lfs_segunlock(fs); 2361 2362 /* 2363 * Wait for v_numoutput to drop to zero. The seglock should 2364 * take care of this, but there is a slight possibility that 2365 * aiodoned might not have got around to our buffers yet. 2366 */ 2367 if (sync) { 2368 s = splbio(); 2369 simple_lock(&global_v_numoutput_slock); 2370 while (vp->v_numoutput > 0) { 2371 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2372 " num %d\n", ip->i_number, vp->v_numoutput)); 2373 vp->v_iflag |= VI_BWAIT; 2374 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 2375 &global_v_numoutput_slock); 2376 } 2377 simple_unlock(&global_v_numoutput_slock); 2378 splx(s); 2379 } 2380 return error; 2381 } 2382 2383 /* 2384 * Return the last logical file offset that should be written for this file 2385 * if we're doing a write that ends at "size". If writing, we need to know 2386 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2387 * to know about entire blocks. 2388 */ 2389 void 2390 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2391 { 2392 struct inode *ip = VTOI(vp); 2393 struct lfs *fs = ip->i_lfs; 2394 daddr_t olbn, nlbn; 2395 2396 olbn = lblkno(fs, ip->i_size); 2397 nlbn = lblkno(fs, size); 2398 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2399 *eobp = fragroundup(fs, size); 2400 } else { 2401 *eobp = blkroundup(fs, size); 2402 } 2403 } 2404 2405 #ifdef DEBUG 2406 void lfs_dump_vop(void *); 2407 2408 void 2409 lfs_dump_vop(void *v) 2410 { 2411 struct vop_putpages_args /* { 2412 struct vnode *a_vp; 2413 voff_t a_offlo; 2414 voff_t a_offhi; 2415 int a_flags; 2416 } */ *ap = v; 2417 2418 #ifdef DDB 2419 vfs_vnode_print(ap->a_vp, 0, printf); 2420 #endif 2421 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2422 } 2423 #endif 2424 2425 int 2426 lfs_mmap(void *v) 2427 { 2428 struct vop_mmap_args /* { 2429 const struct vnodeop_desc *a_desc; 2430 struct vnode *a_vp; 2431 vm_prot_t a_prot; 2432 kauth_cred_t a_cred; 2433 } */ *ap = v; 2434 2435 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2436 return EOPNOTSUPP; 2437 return ufs_mmap(v); 2438 } 2439
Indexes created Tue Sep 30 20:09:53 GMT 2025