lfs_vnops.c revision 1.219
1 /* $NetBSD: lfs_vnops.c,v 1.219 2009/01/16 05:15:29 yamt 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 * 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) 1986, 1989, 1991, 1993, 1995 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_vnops.c 8.13 (Berkeley) 6/10/95 60 */ 61 62 #include <sys/cdefs.h> 63 __KERNEL_RCSID(0, "$NetBSD: lfs_vnops.c,v 1.219 2009/01/16 05:15:29 yamt Exp $"); 64 65 #ifdef _KERNEL_OPT 66 #include "opt_compat_netbsd.h" 67 #endif 68 69 #include <sys/param.h> 70 #include <sys/systm.h> 71 #include <sys/namei.h> 72 #include <sys/resourcevar.h> 73 #include <sys/kernel.h> 74 #include <sys/file.h> 75 #include <sys/stat.h> 76 #include <sys/buf.h> 77 #include <sys/proc.h> 78 #include <sys/mount.h> 79 #include <sys/vnode.h> 80 #include <sys/pool.h> 81 #include <sys/signalvar.h> 82 #include <sys/kauth.h> 83 #include <sys/syslog.h> 84 #include <sys/fstrans.h> 85 86 #include <miscfs/fifofs/fifo.h> 87 #include <miscfs/genfs/genfs.h> 88 #include <miscfs/specfs/specdev.h> 89 90 #include <ufs/ufs/inode.h> 91 #include <ufs/ufs/dir.h> 92 #include <ufs/ufs/ufsmount.h> 93 #include <ufs/ufs/ufs_extern.h> 94 95 #include <uvm/uvm.h> 96 #include <uvm/uvm_pmap.h> 97 #include <uvm/uvm_stat.h> 98 #include <uvm/uvm_pager.h> 99 100 #include <ufs/lfs/lfs.h> 101 #include <ufs/lfs/lfs_extern.h> 102 103 extern pid_t lfs_writer_daemon; 104 int lfs_ignore_lazy_sync = 1; 105 106 /* Global vfs data structures for lfs. */ 107 int (**lfs_vnodeop_p)(void *); 108 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 109 { &vop_default_desc, vn_default_error }, 110 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 111 { &vop_create_desc, lfs_create }, /* create */ 112 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 113 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 114 { &vop_open_desc, ufs_open }, /* open */ 115 { &vop_close_desc, lfs_close }, /* close */ 116 { &vop_access_desc, ufs_access }, /* access */ 117 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 118 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 119 { &vop_read_desc, lfs_read }, /* read */ 120 { &vop_write_desc, lfs_write }, /* write */ 121 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 122 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 123 { &vop_poll_desc, ufs_poll }, /* poll */ 124 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 125 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 126 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 127 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 128 { &vop_seek_desc, ufs_seek }, /* seek */ 129 { &vop_remove_desc, lfs_remove }, /* remove */ 130 { &vop_link_desc, lfs_link }, /* link */ 131 { &vop_rename_desc, lfs_rename }, /* rename */ 132 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 133 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 134 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 135 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 136 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 137 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 138 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 139 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 140 { &vop_lock_desc, ufs_lock }, /* lock */ 141 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 142 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 143 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 144 { &vop_print_desc, ufs_print }, /* print */ 145 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 146 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 147 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 148 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 149 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 150 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 151 { NULL, NULL } 152 }; 153 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 154 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 155 156 int (**lfs_specop_p)(void *); 157 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 158 { &vop_default_desc, vn_default_error }, 159 { &vop_lookup_desc, spec_lookup }, /* lookup */ 160 { &vop_create_desc, spec_create }, /* create */ 161 { &vop_mknod_desc, spec_mknod }, /* mknod */ 162 { &vop_open_desc, spec_open }, /* open */ 163 { &vop_close_desc, lfsspec_close }, /* close */ 164 { &vop_access_desc, ufs_access }, /* access */ 165 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 166 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 167 { &vop_read_desc, ufsspec_read }, /* read */ 168 { &vop_write_desc, ufsspec_write }, /* write */ 169 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 170 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 171 { &vop_poll_desc, spec_poll }, /* poll */ 172 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 173 { &vop_revoke_desc, spec_revoke }, /* revoke */ 174 { &vop_mmap_desc, spec_mmap }, /* mmap */ 175 { &vop_fsync_desc, spec_fsync }, /* fsync */ 176 { &vop_seek_desc, spec_seek }, /* seek */ 177 { &vop_remove_desc, spec_remove }, /* remove */ 178 { &vop_link_desc, spec_link }, /* link */ 179 { &vop_rename_desc, spec_rename }, /* rename */ 180 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 181 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 182 { &vop_symlink_desc, spec_symlink }, /* symlink */ 183 { &vop_readdir_desc, spec_readdir }, /* readdir */ 184 { &vop_readlink_desc, spec_readlink }, /* readlink */ 185 { &vop_abortop_desc, spec_abortop }, /* abortop */ 186 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 187 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 188 { &vop_lock_desc, ufs_lock }, /* lock */ 189 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 190 { &vop_bmap_desc, spec_bmap }, /* bmap */ 191 { &vop_strategy_desc, spec_strategy }, /* strategy */ 192 { &vop_print_desc, ufs_print }, /* print */ 193 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 194 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 195 { &vop_advlock_desc, spec_advlock }, /* advlock */ 196 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 197 { &vop_getpages_desc, spec_getpages }, /* getpages */ 198 { &vop_putpages_desc, spec_putpages }, /* putpages */ 199 { NULL, NULL } 200 }; 201 const struct vnodeopv_desc lfs_specop_opv_desc = 202 { &lfs_specop_p, lfs_specop_entries }; 203 204 int (**lfs_fifoop_p)(void *); 205 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 206 { &vop_default_desc, vn_default_error }, 207 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 208 { &vop_create_desc, fifo_create }, /* create */ 209 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 210 { &vop_open_desc, fifo_open }, /* open */ 211 { &vop_close_desc, lfsfifo_close }, /* close */ 212 { &vop_access_desc, ufs_access }, /* access */ 213 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 214 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 215 { &vop_read_desc, ufsfifo_read }, /* read */ 216 { &vop_write_desc, ufsfifo_write }, /* write */ 217 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 218 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 219 { &vop_poll_desc, fifo_poll }, /* poll */ 220 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 221 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 222 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 223 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 224 { &vop_seek_desc, fifo_seek }, /* seek */ 225 { &vop_remove_desc, fifo_remove }, /* remove */ 226 { &vop_link_desc, fifo_link }, /* link */ 227 { &vop_rename_desc, fifo_rename }, /* rename */ 228 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 229 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 230 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 231 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 232 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 233 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 234 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 235 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 236 { &vop_lock_desc, ufs_lock }, /* lock */ 237 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 238 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 239 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 240 { &vop_print_desc, ufs_print }, /* print */ 241 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 242 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 243 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 244 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 245 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 246 { NULL, NULL } 247 }; 248 const struct vnodeopv_desc lfs_fifoop_opv_desc = 249 { &lfs_fifoop_p, lfs_fifoop_entries }; 250 251 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int, struct vm_page **); 252 253 #define LFS_READWRITE 254 #include <ufs/ufs/ufs_readwrite.c> 255 #undef LFS_READWRITE 256 257 /* 258 * Synch an open file. 259 */ 260 /* ARGSUSED */ 261 int 262 lfs_fsync(void *v) 263 { 264 struct vop_fsync_args /* { 265 struct vnode *a_vp; 266 kauth_cred_t a_cred; 267 int a_flags; 268 off_t offlo; 269 off_t offhi; 270 } */ *ap = v; 271 struct vnode *vp = ap->a_vp; 272 int error, wait; 273 struct inode *ip = VTOI(vp); 274 struct lfs *fs = ip->i_lfs; 275 276 /* If we're mounted read-only, don't try to sync. */ 277 if (fs->lfs_ronly) 278 return 0; 279 280 /* 281 * Trickle sync simply adds this vnode to the pager list, as if 282 * the pagedaemon had requested a pageout. 283 */ 284 if (ap->a_flags & FSYNC_LAZY) { 285 if (lfs_ignore_lazy_sync == 0) { 286 mutex_enter(&lfs_lock); 287 if (!(ip->i_flags & IN_PAGING)) { 288 ip->i_flags |= IN_PAGING; 289 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, 290 i_lfs_pchain); 291 } 292 wakeup(&lfs_writer_daemon); 293 mutex_exit(&lfs_lock); 294 } 295 return 0; 296 } 297 298 /* 299 * If a vnode is bring cleaned, flush it out before we try to 300 * reuse it. This prevents the cleaner from writing files twice 301 * in the same partial segment, causing an accounting underflow. 302 */ 303 if (ap->a_flags & FSYNC_RECLAIM && ip->i_flags & IN_CLEANING) { 304 lfs_vflush(vp); 305 } 306 307 wait = (ap->a_flags & FSYNC_WAIT); 308 do { 309 mutex_enter(&vp->v_interlock); 310 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 311 round_page(ap->a_offhi), 312 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 313 if (error == EAGAIN) { 314 mutex_enter(&lfs_lock); 315 mtsleep(&fs->lfs_avail, PCATCH | PUSER, "lfs_fsync", 316 hz / 100 + 1, &lfs_lock); 317 mutex_exit(&lfs_lock); 318 } 319 } while (error == EAGAIN); 320 if (error) 321 return error; 322 323 if ((ap->a_flags & FSYNC_DATAONLY) == 0) 324 error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 325 326 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 327 int l = 0; 328 error = VOP_IOCTL(ip->i_devvp, DIOCCACHESYNC, &l, FWRITE, 329 curlwp->l_cred); 330 } 331 if (wait && !VPISEMPTY(vp)) 332 LFS_SET_UINO(ip, IN_MODIFIED); 333 334 return error; 335 } 336 337 /* 338 * Take IN_ADIROP off, then call ufs_inactive. 339 */ 340 int 341 lfs_inactive(void *v) 342 { 343 struct vop_inactive_args /* { 344 struct vnode *a_vp; 345 } */ *ap = v; 346 347 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 348 349 lfs_unmark_vnode(ap->a_vp); 350 351 /* 352 * The Ifile is only ever inactivated on unmount. 353 * Streamline this process by not giving it more dirty blocks. 354 */ 355 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 356 mutex_enter(&lfs_lock); 357 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 358 mutex_exit(&lfs_lock); 359 VOP_UNLOCK(ap->a_vp, 0); 360 return 0; 361 } 362 363 return ufs_inactive(v); 364 } 365 366 /* 367 * These macros are used to bracket UFS directory ops, so that we can 368 * identify all the pages touched during directory ops which need to 369 * be ordered and flushed atomically, so that they may be recovered. 370 * 371 * Because we have to mark nodes VU_DIROP in order to prevent 372 * the cache from reclaiming them while a dirop is in progress, we must 373 * also manage the number of nodes so marked (otherwise we can run out). 374 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 375 * is decremented during segment write, when VU_DIROP is taken off. 376 */ 377 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 378 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 379 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 380 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 381 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 382 static int lfs_set_dirop(struct vnode *, struct vnode *); 383 384 static int 385 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 386 { 387 struct lfs *fs; 388 int error; 389 390 KASSERT(VOP_ISLOCKED(dvp)); 391 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 392 393 fs = VTOI(dvp)->i_lfs; 394 395 ASSERT_NO_SEGLOCK(fs); 396 /* 397 * LFS_NRESERVE calculates direct and indirect blocks as well 398 * as an inode block; an overestimate in most cases. 399 */ 400 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 401 return (error); 402 403 restart: 404 mutex_enter(&lfs_lock); 405 if (fs->lfs_dirops == 0) { 406 mutex_exit(&lfs_lock); 407 lfs_check(dvp, LFS_UNUSED_LBN, 0); 408 mutex_enter(&lfs_lock); 409 } 410 while (fs->lfs_writer) { 411 error = mtsleep(&fs->lfs_dirops, (PRIBIO + 1) | PCATCH, 412 "lfs_sdirop", 0, &lfs_lock); 413 if (error == EINTR) { 414 mutex_exit(&lfs_lock); 415 goto unreserve; 416 } 417 } 418 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 419 wakeup(&lfs_writer_daemon); 420 mutex_exit(&lfs_lock); 421 preempt(); 422 goto restart; 423 } 424 425 if (lfs_dirvcount > LFS_MAX_DIROP) { 426 mutex_exit(&lfs_lock); 427 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 428 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 429 if ((error = mtsleep(&lfs_dirvcount, 430 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 431 &lfs_lock)) != 0) { 432 goto unreserve; 433 } 434 goto restart; 435 } 436 437 ++fs->lfs_dirops; 438 fs->lfs_doifile = 1; 439 mutex_exit(&lfs_lock); 440 441 /* Hold a reference so SET_ENDOP will be happy */ 442 vref(dvp); 443 if (vp) { 444 vref(vp); 445 MARK_VNODE(vp); 446 } 447 448 MARK_VNODE(dvp); 449 return 0; 450 451 unreserve: 452 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 453 return error; 454 } 455 456 /* 457 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 458 * in getnewvnode(), if we have a stacked filesystem mounted on top 459 * of us. 460 * 461 * NB: this means we have to clear the new vnodes on error. Fortunately 462 * SET_ENDOP is there to do that for us. 463 */ 464 static int 465 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 466 { 467 int error; 468 struct lfs *fs; 469 470 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 471 ASSERT_NO_SEGLOCK(fs); 472 if (fs->lfs_ronly) 473 return EROFS; 474 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 475 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 476 dvp, error)); 477 return error; 478 } 479 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 480 if (vpp) { 481 ungetnewvnode(*vpp); 482 *vpp = NULL; 483 } 484 return error; 485 } 486 return 0; 487 } 488 489 #define SET_ENDOP_BASE(fs, dvp, str) \ 490 do { \ 491 mutex_enter(&lfs_lock); \ 492 --(fs)->lfs_dirops; \ 493 if (!(fs)->lfs_dirops) { \ 494 if ((fs)->lfs_nadirop) { \ 495 panic("SET_ENDOP: %s: no dirops but " \ 496 " nadirop=%d", (str), \ 497 (fs)->lfs_nadirop); \ 498 } \ 499 wakeup(&(fs)->lfs_writer); \ 500 mutex_exit(&lfs_lock); \ 501 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 502 } else \ 503 mutex_exit(&lfs_lock); \ 504 } while(0) 505 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 506 do { \ 507 UNMARK_VNODE(dvp); \ 508 if (nvpp && *nvpp) \ 509 UNMARK_VNODE(*nvpp); \ 510 /* Check for error return to stem vnode leakage */ \ 511 if (nvpp && *nvpp && !((*nvpp)->v_uflag & VU_DIROP)) \ 512 ungetnewvnode(*(nvpp)); \ 513 SET_ENDOP_BASE((fs), (dvp), (str)); \ 514 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 515 vrele(dvp); \ 516 } while(0) 517 #define SET_ENDOP_CREATE_AP(ap, str) \ 518 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 519 (ap)->a_vpp, (str)) 520 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 521 do { \ 522 UNMARK_VNODE(dvp); \ 523 if (ovp) \ 524 UNMARK_VNODE(ovp); \ 525 SET_ENDOP_BASE((fs), (dvp), (str)); \ 526 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 527 vrele(dvp); \ 528 if (ovp) \ 529 vrele(ovp); \ 530 } while(0) 531 532 void 533 lfs_mark_vnode(struct vnode *vp) 534 { 535 struct inode *ip = VTOI(vp); 536 struct lfs *fs = ip->i_lfs; 537 538 mutex_enter(&lfs_lock); 539 if (!(ip->i_flag & IN_ADIROP)) { 540 if (!(vp->v_uflag & VU_DIROP)) { 541 mutex_enter(&vp->v_interlock); 542 (void)lfs_vref(vp); 543 ++lfs_dirvcount; 544 ++fs->lfs_dirvcount; 545 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 546 vp->v_uflag |= VU_DIROP; 547 } 548 ++fs->lfs_nadirop; 549 ip->i_flag |= IN_ADIROP; 550 } else 551 KASSERT(vp->v_uflag & VU_DIROP); 552 mutex_exit(&lfs_lock); 553 } 554 555 void 556 lfs_unmark_vnode(struct vnode *vp) 557 { 558 struct inode *ip = VTOI(vp); 559 560 if (ip && (ip->i_flag & IN_ADIROP)) { 561 KASSERT(vp->v_uflag & VU_DIROP); 562 mutex_enter(&lfs_lock); 563 --ip->i_lfs->lfs_nadirop; 564 mutex_exit(&lfs_lock); 565 ip->i_flag &= ~IN_ADIROP; 566 } 567 } 568 569 int 570 lfs_symlink(void *v) 571 { 572 struct vop_symlink_args /* { 573 struct vnode *a_dvp; 574 struct vnode **a_vpp; 575 struct componentname *a_cnp; 576 struct vattr *a_vap; 577 char *a_target; 578 } */ *ap = v; 579 int error; 580 581 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 582 vput(ap->a_dvp); 583 return error; 584 } 585 error = ufs_symlink(ap); 586 SET_ENDOP_CREATE_AP(ap, "symlink"); 587 return (error); 588 } 589 590 int 591 lfs_mknod(void *v) 592 { 593 struct vop_mknod_args /* { 594 struct vnode *a_dvp; 595 struct vnode **a_vpp; 596 struct componentname *a_cnp; 597 struct vattr *a_vap; 598 } */ *ap = v; 599 struct vattr *vap = ap->a_vap; 600 struct vnode **vpp = ap->a_vpp; 601 struct inode *ip; 602 int error; 603 struct mount *mp; 604 ino_t ino; 605 606 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 607 vput(ap->a_dvp); 608 return error; 609 } 610 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 611 ap->a_dvp, vpp, ap->a_cnp); 612 613 /* Either way we're done with the dirop at this point */ 614 SET_ENDOP_CREATE_AP(ap, "mknod"); 615 616 if (error) 617 return (error); 618 619 ip = VTOI(*vpp); 620 mp = (*vpp)->v_mount; 621 ino = ip->i_number; 622 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 623 if (vap->va_rdev != VNOVAL) { 624 /* 625 * Want to be able to use this to make badblock 626 * inodes, so don't truncate the dev number. 627 */ 628 #if 0 629 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 630 UFS_MPNEEDSWAP((*vpp)->v_mount)); 631 #else 632 ip->i_ffs1_rdev = vap->va_rdev; 633 #endif 634 } 635 636 /* 637 * Call fsync to write the vnode so that we don't have to deal with 638 * flushing it when it's marked VU_DIROP|VI_XLOCK. 639 * 640 * XXX KS - If we can't flush we also can't call vgone(), so must 641 * return. But, that leaves this vnode in limbo, also not good. 642 * Can this ever happen (barring hardware failure)? 643 */ 644 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0)) != 0) { 645 panic("lfs_mknod: couldn't fsync (ino %llu)", 646 (unsigned long long)ino); 647 /* return (error); */ 648 } 649 /* 650 * Remove vnode so that it will be reloaded by VFS_VGET and 651 * checked to see if it is an alias of an existing entry in 652 * the inode cache. 653 */ 654 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 655 656 VOP_UNLOCK(*vpp, 0); 657 (*vpp)->v_type = VNON; 658 vgone(*vpp); 659 error = VFS_VGET(mp, ino, vpp); 660 661 if (error != 0) { 662 *vpp = NULL; 663 return (error); 664 } 665 return (0); 666 } 667 668 int 669 lfs_create(void *v) 670 { 671 struct vop_create_args /* { 672 struct vnode *a_dvp; 673 struct vnode **a_vpp; 674 struct componentname *a_cnp; 675 struct vattr *a_vap; 676 } */ *ap = v; 677 int error; 678 679 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 680 vput(ap->a_dvp); 681 return error; 682 } 683 error = ufs_create(ap); 684 SET_ENDOP_CREATE_AP(ap, "create"); 685 return (error); 686 } 687 688 int 689 lfs_mkdir(void *v) 690 { 691 struct vop_mkdir_args /* { 692 struct vnode *a_dvp; 693 struct vnode **a_vpp; 694 struct componentname *a_cnp; 695 struct vattr *a_vap; 696 } */ *ap = v; 697 int error; 698 699 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 700 vput(ap->a_dvp); 701 return error; 702 } 703 error = ufs_mkdir(ap); 704 SET_ENDOP_CREATE_AP(ap, "mkdir"); 705 return (error); 706 } 707 708 int 709 lfs_remove(void *v) 710 { 711 struct vop_remove_args /* { 712 struct vnode *a_dvp; 713 struct vnode *a_vp; 714 struct componentname *a_cnp; 715 } */ *ap = v; 716 struct vnode *dvp, *vp; 717 struct inode *ip; 718 int error; 719 720 dvp = ap->a_dvp; 721 vp = ap->a_vp; 722 ip = VTOI(vp); 723 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 724 if (dvp == vp) 725 vrele(vp); 726 else 727 vput(vp); 728 vput(dvp); 729 return error; 730 } 731 error = ufs_remove(ap); 732 if (ip->i_nlink == 0) 733 lfs_orphan(ip->i_lfs, ip->i_number); 734 SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove"); 735 return (error); 736 } 737 738 int 739 lfs_rmdir(void *v) 740 { 741 struct vop_rmdir_args /* { 742 struct vnodeop_desc *a_desc; 743 struct vnode *a_dvp; 744 struct vnode *a_vp; 745 struct componentname *a_cnp; 746 } */ *ap = v; 747 struct vnode *vp; 748 struct inode *ip; 749 int error; 750 751 vp = ap->a_vp; 752 ip = VTOI(vp); 753 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 754 if (ap->a_dvp == vp) 755 vrele(ap->a_dvp); 756 else 757 vput(ap->a_dvp); 758 vput(vp); 759 return error; 760 } 761 error = ufs_rmdir(ap); 762 if (ip->i_nlink == 0) 763 lfs_orphan(ip->i_lfs, ip->i_number); 764 SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 765 return (error); 766 } 767 768 int 769 lfs_link(void *v) 770 { 771 struct vop_link_args /* { 772 struct vnode *a_dvp; 773 struct vnode *a_vp; 774 struct componentname *a_cnp; 775 } */ *ap = v; 776 int error; 777 struct vnode **vpp = NULL; 778 779 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 780 vput(ap->a_dvp); 781 return error; 782 } 783 error = ufs_link(ap); 784 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 785 return (error); 786 } 787 788 int 789 lfs_rename(void *v) 790 { 791 struct vop_rename_args /* { 792 struct vnode *a_fdvp; 793 struct vnode *a_fvp; 794 struct componentname *a_fcnp; 795 struct vnode *a_tdvp; 796 struct vnode *a_tvp; 797 struct componentname *a_tcnp; 798 } */ *ap = v; 799 struct vnode *tvp, *fvp, *tdvp, *fdvp; 800 struct componentname *tcnp, *fcnp; 801 int error; 802 struct lfs *fs; 803 804 fs = VTOI(ap->a_fdvp)->i_lfs; 805 tvp = ap->a_tvp; 806 tdvp = ap->a_tdvp; 807 tcnp = ap->a_tcnp; 808 fvp = ap->a_fvp; 809 fdvp = ap->a_fdvp; 810 fcnp = ap->a_fcnp; 811 812 /* 813 * Check for cross-device rename. 814 * If it is, we don't want to set dirops, just error out. 815 * (In particular note that MARK_VNODE(tdvp) will DTWT on 816 * a cross-device rename.) 817 * 818 * Copied from ufs_rename. 819 */ 820 if ((fvp->v_mount != tdvp->v_mount) || 821 (tvp && (fvp->v_mount != tvp->v_mount))) { 822 error = EXDEV; 823 goto errout; 824 } 825 826 /* 827 * Check to make sure we're not renaming a vnode onto itself 828 * (deleting a hard link by renaming one name onto another); 829 * if we are we can't recursively call VOP_REMOVE since that 830 * would leave us with an unaccounted-for number of live dirops. 831 * 832 * Inline the relevant section of ufs_rename here, *before* 833 * calling SET_DIROP_REMOVE. 834 */ 835 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 836 (VTOI(tdvp)->i_flags & APPEND))) { 837 error = EPERM; 838 goto errout; 839 } 840 if (fvp == tvp) { 841 if (fvp->v_type == VDIR) { 842 error = EINVAL; 843 goto errout; 844 } 845 846 /* Release destination completely. */ 847 VOP_ABORTOP(tdvp, tcnp); 848 vput(tdvp); 849 vput(tvp); 850 851 /* Delete source. */ 852 vrele(fvp); 853 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 854 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 855 fcnp->cn_nameiop = DELETE; 856 vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY); 857 if ((error = relookup(fdvp, &fvp, fcnp))) { 858 vput(fdvp); 859 return (error); 860 } 861 return (VOP_REMOVE(fdvp, fvp, fcnp)); 862 } 863 864 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 865 goto errout; 866 MARK_VNODE(fdvp); 867 MARK_VNODE(fvp); 868 869 error = ufs_rename(ap); 870 UNMARK_VNODE(fdvp); 871 UNMARK_VNODE(fvp); 872 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 873 return (error); 874 875 errout: 876 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 877 if (tdvp == tvp) 878 vrele(tdvp); 879 else 880 vput(tdvp); 881 if (tvp) 882 vput(tvp); 883 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 884 vrele(fdvp); 885 vrele(fvp); 886 return (error); 887 } 888 889 /* XXX hack to avoid calling ITIMES in getattr */ 890 int 891 lfs_getattr(void *v) 892 { 893 struct vop_getattr_args /* { 894 struct vnode *a_vp; 895 struct vattr *a_vap; 896 kauth_cred_t a_cred; 897 } */ *ap = v; 898 struct vnode *vp = ap->a_vp; 899 struct inode *ip = VTOI(vp); 900 struct vattr *vap = ap->a_vap; 901 struct lfs *fs = ip->i_lfs; 902 /* 903 * Copy from inode table 904 */ 905 vap->va_fsid = ip->i_dev; 906 vap->va_fileid = ip->i_number; 907 vap->va_mode = ip->i_mode & ~IFMT; 908 vap->va_nlink = ip->i_nlink; 909 vap->va_uid = ip->i_uid; 910 vap->va_gid = ip->i_gid; 911 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 912 vap->va_size = vp->v_size; 913 vap->va_atime.tv_sec = ip->i_ffs1_atime; 914 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 915 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 916 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 917 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 918 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 919 vap->va_flags = ip->i_flags; 920 vap->va_gen = ip->i_gen; 921 /* this doesn't belong here */ 922 if (vp->v_type == VBLK) 923 vap->va_blocksize = BLKDEV_IOSIZE; 924 else if (vp->v_type == VCHR) 925 vap->va_blocksize = MAXBSIZE; 926 else 927 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 928 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 929 vap->va_type = vp->v_type; 930 vap->va_filerev = ip->i_modrev; 931 return (0); 932 } 933 934 /* 935 * Check to make sure the inode blocks won't choke the buffer 936 * cache, then call ufs_setattr as usual. 937 */ 938 int 939 lfs_setattr(void *v) 940 { 941 struct vop_setattr_args /* { 942 struct vnode *a_vp; 943 struct vattr *a_vap; 944 kauth_cred_t a_cred; 945 } */ *ap = v; 946 struct vnode *vp = ap->a_vp; 947 948 lfs_check(vp, LFS_UNUSED_LBN, 0); 949 return ufs_setattr(v); 950 } 951 952 /* 953 * Release the block we hold on lfs_newseg wrapping. Called on file close, 954 * or explicitly from LFCNWRAPGO. Called with the interlock held. 955 */ 956 static int 957 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor) 958 { 959 if (fs->lfs_stoplwp != curlwp) 960 return EBUSY; 961 962 fs->lfs_stoplwp = NULL; 963 cv_signal(&fs->lfs_stopcv); 964 965 KASSERT(fs->lfs_nowrap > 0); 966 if (fs->lfs_nowrap <= 0) { 967 return 0; 968 } 969 970 if (--fs->lfs_nowrap == 0) { 971 log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt); 972 wakeup(&fs->lfs_wrappass); 973 lfs_wakeup_cleaner(fs); 974 } 975 if (waitfor) { 976 mtsleep(&fs->lfs_nextseg, PCATCH | PUSER, "segment", 977 0, &lfs_lock); 978 } 979 980 return 0; 981 } 982 983 /* 984 * Close called 985 */ 986 /* ARGSUSED */ 987 int 988 lfs_close(void *v) 989 { 990 struct vop_close_args /* { 991 struct vnode *a_vp; 992 int a_fflag; 993 kauth_cred_t a_cred; 994 } */ *ap = v; 995 struct vnode *vp = ap->a_vp; 996 struct inode *ip = VTOI(vp); 997 struct lfs *fs = ip->i_lfs; 998 999 if ((ip->i_number == ROOTINO || ip->i_number == LFS_IFILE_INUM) && 1000 fs->lfs_stoplwp == curlwp) { 1001 mutex_enter(&lfs_lock); 1002 log(LOG_NOTICE, "lfs_close: releasing log wrap control\n"); 1003 lfs_wrapgo(fs, ip, 0); 1004 mutex_exit(&lfs_lock); 1005 } 1006 1007 if (vp == ip->i_lfs->lfs_ivnode && 1008 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 1009 return 0; 1010 1011 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 1012 LFS_ITIMES(ip, NULL, NULL, NULL); 1013 } 1014 return (0); 1015 } 1016 1017 /* 1018 * Close wrapper for special devices. 1019 * 1020 * Update the times on the inode then do device close. 1021 */ 1022 int 1023 lfsspec_close(void *v) 1024 { 1025 struct vop_close_args /* { 1026 struct vnode *a_vp; 1027 int a_fflag; 1028 kauth_cred_t a_cred; 1029 } */ *ap = v; 1030 struct vnode *vp; 1031 struct inode *ip; 1032 1033 vp = ap->a_vp; 1034 ip = VTOI(vp); 1035 if (vp->v_usecount > 1) { 1036 LFS_ITIMES(ip, NULL, NULL, NULL); 1037 } 1038 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1039 } 1040 1041 /* 1042 * Close wrapper for fifo's. 1043 * 1044 * Update the times on the inode then do device close. 1045 */ 1046 int 1047 lfsfifo_close(void *v) 1048 { 1049 struct vop_close_args /* { 1050 struct vnode *a_vp; 1051 int a_fflag; 1052 kauth_cred_ a_cred; 1053 } */ *ap = v; 1054 struct vnode *vp; 1055 struct inode *ip; 1056 1057 vp = ap->a_vp; 1058 ip = VTOI(vp); 1059 if (ap->a_vp->v_usecount > 1) { 1060 LFS_ITIMES(ip, NULL, NULL, NULL); 1061 } 1062 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1063 } 1064 1065 /* 1066 * Reclaim an inode so that it can be used for other purposes. 1067 */ 1068 1069 int 1070 lfs_reclaim(void *v) 1071 { 1072 struct vop_reclaim_args /* { 1073 struct vnode *a_vp; 1074 } */ *ap = v; 1075 struct vnode *vp = ap->a_vp; 1076 struct inode *ip = VTOI(vp); 1077 struct lfs *fs = ip->i_lfs; 1078 int error; 1079 1080 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1081 1082 mutex_enter(&lfs_lock); 1083 LFS_CLR_UINO(ip, IN_ALLMOD); 1084 mutex_exit(&lfs_lock); 1085 if ((error = ufs_reclaim(vp))) 1086 return (error); 1087 1088 /* 1089 * Take us off the paging and/or dirop queues if we were on them. 1090 * We shouldn't be on them. 1091 */ 1092 mutex_enter(&lfs_lock); 1093 if (ip->i_flags & IN_PAGING) { 1094 log(LOG_WARNING, "%s: reclaimed vnode is IN_PAGING\n", 1095 fs->lfs_fsmnt); 1096 ip->i_flags &= ~IN_PAGING; 1097 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1098 } 1099 if (vp->v_uflag & VU_DIROP) { 1100 panic("reclaimed vnode is VU_DIROP"); 1101 vp->v_uflag &= ~VU_DIROP; 1102 TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain); 1103 } 1104 mutex_exit(&lfs_lock); 1105 1106 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1107 lfs_deregister_all(vp); 1108 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1109 ip->inode_ext.lfs = NULL; 1110 genfs_node_destroy(vp); 1111 pool_put(&lfs_inode_pool, vp->v_data); 1112 vp->v_data = NULL; 1113 return (0); 1114 } 1115 1116 /* 1117 * Read a block from a storage device. 1118 * In order to avoid reading blocks that are in the process of being 1119 * written by the cleaner---and hence are not mutexed by the normal 1120 * buffer cache / page cache mechanisms---check for collisions before 1121 * reading. 1122 * 1123 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1124 * the active cleaner test. 1125 * 1126 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1127 */ 1128 int 1129 lfs_strategy(void *v) 1130 { 1131 struct vop_strategy_args /* { 1132 struct vnode *a_vp; 1133 struct buf *a_bp; 1134 } */ *ap = v; 1135 struct buf *bp; 1136 struct lfs *fs; 1137 struct vnode *vp; 1138 struct inode *ip; 1139 daddr_t tbn; 1140 int i, sn, error, slept; 1141 1142 bp = ap->a_bp; 1143 vp = ap->a_vp; 1144 ip = VTOI(vp); 1145 fs = ip->i_lfs; 1146 1147 /* lfs uses its strategy routine only for read */ 1148 KASSERT(bp->b_flags & B_READ); 1149 1150 if (vp->v_type == VBLK || vp->v_type == VCHR) 1151 panic("lfs_strategy: spec"); 1152 KASSERT(bp->b_bcount != 0); 1153 if (bp->b_blkno == bp->b_lblkno) { 1154 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1155 NULL); 1156 if (error) { 1157 bp->b_error = error; 1158 bp->b_resid = bp->b_bcount; 1159 biodone(bp); 1160 return (error); 1161 } 1162 if ((long)bp->b_blkno == -1) /* no valid data */ 1163 clrbuf(bp); 1164 } 1165 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1166 bp->b_resid = bp->b_bcount; 1167 biodone(bp); 1168 return (0); 1169 } 1170 1171 slept = 1; 1172 mutex_enter(&lfs_lock); 1173 while (slept && fs->lfs_seglock) { 1174 mutex_exit(&lfs_lock); 1175 /* 1176 * Look through list of intervals. 1177 * There will only be intervals to look through 1178 * if the cleaner holds the seglock. 1179 * Since the cleaner is synchronous, we can trust 1180 * the list of intervals to be current. 1181 */ 1182 tbn = dbtofsb(fs, bp->b_blkno); 1183 sn = dtosn(fs, tbn); 1184 slept = 0; 1185 for (i = 0; i < fs->lfs_cleanind; i++) { 1186 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1187 tbn >= fs->lfs_cleanint[i]) { 1188 DLOG((DLOG_CLEAN, 1189 "lfs_strategy: ino %d lbn %" PRId64 1190 " ind %d sn %d fsb %" PRIx32 1191 " given sn %d fsb %" PRIx64 "\n", 1192 ip->i_number, bp->b_lblkno, i, 1193 dtosn(fs, fs->lfs_cleanint[i]), 1194 fs->lfs_cleanint[i], sn, tbn)); 1195 DLOG((DLOG_CLEAN, 1196 "lfs_strategy: sleeping on ino %d lbn %" 1197 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1198 mutex_enter(&lfs_lock); 1199 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1200 /* Cleaner can't wait for itself */ 1201 mtsleep(&fs->lfs_iocount, 1202 (PRIBIO + 1) | PNORELOCK, 1203 "clean2", 0, 1204 &lfs_lock); 1205 slept = 1; 1206 break; 1207 } else if (fs->lfs_seglock) { 1208 mtsleep(&fs->lfs_seglock, 1209 (PRIBIO + 1) | PNORELOCK, 1210 "clean1", 0, 1211 &lfs_lock); 1212 slept = 1; 1213 break; 1214 } 1215 mutex_exit(&lfs_lock); 1216 } 1217 } 1218 mutex_enter(&lfs_lock); 1219 } 1220 mutex_exit(&lfs_lock); 1221 1222 vp = ip->i_devvp; 1223 VOP_STRATEGY(vp, bp); 1224 return (0); 1225 } 1226 1227 void 1228 lfs_flush_dirops(struct lfs *fs) 1229 { 1230 struct inode *ip, *nip; 1231 struct vnode *vp; 1232 extern int lfs_dostats; 1233 struct segment *sp; 1234 int waslocked; 1235 1236 ASSERT_MAYBE_SEGLOCK(fs); 1237 KASSERT(fs->lfs_nadirop == 0); 1238 1239 if (fs->lfs_ronly) 1240 return; 1241 1242 mutex_enter(&lfs_lock); 1243 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1244 mutex_exit(&lfs_lock); 1245 return; 1246 } else 1247 mutex_exit(&lfs_lock); 1248 1249 if (lfs_dostats) 1250 ++lfs_stats.flush_invoked; 1251 1252 /* 1253 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1254 * Technically this is a checkpoint (the on-disk state is valid) 1255 * even though we are leaving out all the file data. 1256 */ 1257 lfs_imtime(fs); 1258 lfs_seglock(fs, SEGM_CKP); 1259 sp = fs->lfs_sp; 1260 1261 /* 1262 * lfs_writevnodes, optimized to get dirops out of the way. 1263 * Only write dirops, and don't flush files' pages, only 1264 * blocks from the directories. 1265 * 1266 * We don't need to vref these files because they are 1267 * dirops and so hold an extra reference until the 1268 * segunlock clears them of that status. 1269 * 1270 * We don't need to check for IN_ADIROP because we know that 1271 * no dirops are active. 1272 * 1273 */ 1274 mutex_enter(&lfs_lock); 1275 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1276 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1277 mutex_exit(&lfs_lock); 1278 vp = ITOV(ip); 1279 1280 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1281 1282 /* 1283 * All writes to directories come from dirops; all 1284 * writes to files' direct blocks go through the page 1285 * cache, which we're not touching. Reads to files 1286 * and/or directories will not be affected by writing 1287 * directory blocks inodes and file inodes. So we don't 1288 * really need to lock. If we don't lock, though, 1289 * make sure that we don't clear IN_MODIFIED 1290 * unnecessarily. 1291 */ 1292 if (vp->v_iflag & VI_XLOCK) { 1293 mutex_enter(&lfs_lock); 1294 continue; 1295 } 1296 waslocked = VOP_ISLOCKED(vp); 1297 if (vp->v_type != VREG && 1298 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1299 lfs_writefile(fs, sp, vp); 1300 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1301 !(ip->i_flag & IN_ALLMOD)) { 1302 mutex_enter(&lfs_lock); 1303 LFS_SET_UINO(ip, IN_MODIFIED); 1304 mutex_exit(&lfs_lock); 1305 } 1306 } 1307 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1308 (void) lfs_writeinode(fs, sp, ip); 1309 mutex_enter(&lfs_lock); 1310 if (waslocked == LK_EXCLOTHER) 1311 LFS_SET_UINO(ip, IN_MODIFIED); 1312 } 1313 mutex_exit(&lfs_lock); 1314 /* We've written all the dirops there are */ 1315 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1316 lfs_finalize_fs_seguse(fs); 1317 (void) lfs_writeseg(fs, sp); 1318 lfs_segunlock(fs); 1319 } 1320 1321 /* 1322 * Flush all vnodes for which the pagedaemon has requested pageouts. 1323 * Skip over any files that are marked VU_DIROP (since lfs_flush_dirop() 1324 * has just run, this would be an error). If we have to skip a vnode 1325 * for any reason, just skip it; if we have to wait for the cleaner, 1326 * abort. The writer daemon will call us again later. 1327 */ 1328 void 1329 lfs_flush_pchain(struct lfs *fs) 1330 { 1331 struct inode *ip, *nip; 1332 struct vnode *vp; 1333 extern int lfs_dostats; 1334 struct segment *sp; 1335 int error; 1336 1337 ASSERT_NO_SEGLOCK(fs); 1338 1339 if (fs->lfs_ronly) 1340 return; 1341 1342 mutex_enter(&lfs_lock); 1343 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1344 mutex_exit(&lfs_lock); 1345 return; 1346 } else 1347 mutex_exit(&lfs_lock); 1348 1349 /* Get dirops out of the way */ 1350 lfs_flush_dirops(fs); 1351 1352 if (lfs_dostats) 1353 ++lfs_stats.flush_invoked; 1354 1355 /* 1356 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1357 */ 1358 lfs_imtime(fs); 1359 lfs_seglock(fs, 0); 1360 sp = fs->lfs_sp; 1361 1362 /* 1363 * lfs_writevnodes, optimized to clear pageout requests. 1364 * Only write non-dirop files that are in the pageout queue. 1365 * We're very conservative about what we write; we want to be 1366 * fast and async. 1367 */ 1368 mutex_enter(&lfs_lock); 1369 top: 1370 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1371 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1372 vp = ITOV(ip); 1373 1374 if (!(ip->i_flags & IN_PAGING)) 1375 goto top; 1376 1377 mutex_enter(&vp->v_interlock); 1378 if ((vp->v_iflag & VI_XLOCK) || (vp->v_uflag & VU_DIROP) != 0) { 1379 mutex_exit(&vp->v_interlock); 1380 continue; 1381 } 1382 if (vp->v_type != VREG) { 1383 mutex_exit(&vp->v_interlock); 1384 continue; 1385 } 1386 if (lfs_vref(vp)) 1387 continue; 1388 mutex_exit(&lfs_lock); 1389 1390 if (VOP_ISLOCKED(vp)) { 1391 lfs_vunref(vp); 1392 mutex_enter(&lfs_lock); 1393 continue; 1394 } 1395 1396 error = lfs_writefile(fs, sp, vp); 1397 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1398 !(ip->i_flag & IN_ALLMOD)) { 1399 mutex_enter(&lfs_lock); 1400 LFS_SET_UINO(ip, IN_MODIFIED); 1401 mutex_exit(&lfs_lock); 1402 } 1403 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1404 (void) lfs_writeinode(fs, sp, ip); 1405 1406 lfs_vunref(vp); 1407 1408 if (error == EAGAIN) { 1409 lfs_writeseg(fs, sp); 1410 mutex_enter(&lfs_lock); 1411 break; 1412 } 1413 mutex_enter(&lfs_lock); 1414 } 1415 mutex_exit(&lfs_lock); 1416 (void) lfs_writeseg(fs, sp); 1417 lfs_segunlock(fs); 1418 } 1419 1420 /* 1421 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1422 */ 1423 int 1424 lfs_fcntl(void *v) 1425 { 1426 struct vop_fcntl_args /* { 1427 struct vnode *a_vp; 1428 u_int a_command; 1429 void * a_data; 1430 int a_fflag; 1431 kauth_cred_t a_cred; 1432 } */ *ap = v; 1433 struct timeval *tvp; 1434 BLOCK_INFO *blkiov; 1435 CLEANERINFO *cip; 1436 SEGUSE *sup; 1437 int blkcnt, error, oclean; 1438 size_t fh_size; 1439 struct lfs_fcntl_markv blkvp; 1440 struct lwp *l; 1441 fsid_t *fsidp; 1442 struct lfs *fs; 1443 struct buf *bp; 1444 fhandle_t *fhp; 1445 daddr_t off; 1446 1447 /* Only respect LFS fcntls on fs root or Ifile */ 1448 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1449 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1450 return ufs_fcntl(v); 1451 } 1452 1453 /* Avoid locking a draining lock */ 1454 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1455 return ESHUTDOWN; 1456 } 1457 1458 /* LFS control and monitoring fcntls are available only to root */ 1459 l = curlwp; 1460 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1461 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1462 NULL)) != 0) 1463 return (error); 1464 1465 fs = VTOI(ap->a_vp)->i_lfs; 1466 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1467 1468 error = 0; 1469 switch ((int)ap->a_command) { 1470 case LFCNSEGWAITALL: 1471 case LFCNSEGWAITALL_COMPAT: 1472 fsidp = NULL; 1473 /* FALLSTHROUGH */ 1474 case LFCNSEGWAIT: 1475 case LFCNSEGWAIT_COMPAT: 1476 tvp = (struct timeval *)ap->a_data; 1477 mutex_enter(&lfs_lock); 1478 ++fs->lfs_sleepers; 1479 mutex_exit(&lfs_lock); 1480 1481 error = lfs_segwait(fsidp, tvp); 1482 1483 mutex_enter(&lfs_lock); 1484 if (--fs->lfs_sleepers == 0) 1485 wakeup(&fs->lfs_sleepers); 1486 mutex_exit(&lfs_lock); 1487 return error; 1488 1489 case LFCNBMAPV: 1490 case LFCNMARKV: 1491 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1492 1493 blkcnt = blkvp.blkcnt; 1494 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1495 return (EINVAL); 1496 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1497 if ((error = copyin(blkvp.blkiov, blkiov, 1498 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1499 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1500 return error; 1501 } 1502 1503 mutex_enter(&lfs_lock); 1504 ++fs->lfs_sleepers; 1505 mutex_exit(&lfs_lock); 1506 if (ap->a_command == LFCNBMAPV) 1507 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1508 else /* LFCNMARKV */ 1509 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1510 if (error == 0) 1511 error = copyout(blkiov, blkvp.blkiov, 1512 blkcnt * sizeof(BLOCK_INFO)); 1513 mutex_enter(&lfs_lock); 1514 if (--fs->lfs_sleepers == 0) 1515 wakeup(&fs->lfs_sleepers); 1516 mutex_exit(&lfs_lock); 1517 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1518 return error; 1519 1520 case LFCNRECLAIM: 1521 /* 1522 * Flush dirops and write Ifile, allowing empty segments 1523 * to be immediately reclaimed. 1524 */ 1525 lfs_writer_enter(fs, "pndirop"); 1526 off = fs->lfs_offset; 1527 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1528 lfs_flush_dirops(fs); 1529 LFS_CLEANERINFO(cip, fs, bp); 1530 oclean = cip->clean; 1531 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1532 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1533 fs->lfs_sp->seg_flags |= SEGM_PROT; 1534 lfs_segunlock(fs); 1535 lfs_writer_leave(fs); 1536 1537 #ifdef DEBUG 1538 LFS_CLEANERINFO(cip, fs, bp); 1539 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1540 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1541 fs->lfs_offset - off, cip->clean - oclean, 1542 fs->lfs_activesb)); 1543 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1544 #endif 1545 1546 return 0; 1547 1548 #ifdef COMPAT_30 1549 case LFCNIFILEFH_COMPAT: 1550 /* Return the filehandle of the Ifile */ 1551 if ((error = kauth_authorize_generic(l->l_cred, 1552 KAUTH_GENERIC_ISSUSER, NULL)) != 0) 1553 return (error); 1554 fhp = (struct fhandle *)ap->a_data; 1555 fhp->fh_fsid = *fsidp; 1556 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1557 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1558 #endif 1559 1560 case LFCNIFILEFH_COMPAT2: 1561 case LFCNIFILEFH: 1562 /* Return the filehandle of the Ifile */ 1563 fhp = (struct fhandle *)ap->a_data; 1564 fhp->fh_fsid = *fsidp; 1565 fh_size = sizeof(struct lfs_fhandle) - 1566 offsetof(fhandle_t, fh_fid); 1567 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1568 1569 case LFCNREWIND: 1570 /* Move lfs_offset to the lowest-numbered segment */ 1571 return lfs_rewind(fs, *(int *)ap->a_data); 1572 1573 case LFCNINVAL: 1574 /* Mark a segment SEGUSE_INVAL */ 1575 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1576 if (sup->su_nbytes > 0) { 1577 brelse(bp, 0); 1578 lfs_unset_inval_all(fs); 1579 return EBUSY; 1580 } 1581 sup->su_flags |= SEGUSE_INVAL; 1582 VOP_BWRITE(bp); 1583 return 0; 1584 1585 case LFCNRESIZE: 1586 /* Resize the filesystem */ 1587 return lfs_resize_fs(fs, *(int *)ap->a_data); 1588 1589 case LFCNWRAPSTOP: 1590 case LFCNWRAPSTOP_COMPAT: 1591 /* 1592 * Hold lfs_newseg at segment 0; if requested, sleep until 1593 * the filesystem wraps around. To support external agents 1594 * (dump, fsck-based regression test) that need to look at 1595 * a snapshot of the filesystem, without necessarily 1596 * requiring that all fs activity stops. 1597 */ 1598 if (fs->lfs_stoplwp == curlwp) 1599 return EALREADY; 1600 1601 mutex_enter(&lfs_lock); 1602 while (fs->lfs_stoplwp != NULL) 1603 cv_wait(&fs->lfs_stopcv, &lfs_lock); 1604 fs->lfs_stoplwp = curlwp; 1605 if (fs->lfs_nowrap == 0) 1606 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1607 ++fs->lfs_nowrap; 1608 if (*(int *)ap->a_data == 1 || 1609 ap->a_command == LFCNWRAPSTOP_COMPAT) { 1610 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1611 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1612 "segwrap", 0, &lfs_lock); 1613 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1614 if (error) { 1615 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1616 } 1617 } 1618 mutex_exit(&lfs_lock); 1619 return 0; 1620 1621 case LFCNWRAPGO: 1622 case LFCNWRAPGO_COMPAT: 1623 /* 1624 * Having done its work, the agent wakes up the writer. 1625 * If the argument is 1, it sleeps until a new segment 1626 * is selected. 1627 */ 1628 mutex_enter(&lfs_lock); 1629 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1630 (ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1631 *((int *)ap->a_data))); 1632 mutex_exit(&lfs_lock); 1633 return error; 1634 1635 case LFCNWRAPPASS: 1636 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1637 return EALREADY; 1638 mutex_enter(&lfs_lock); 1639 if (fs->lfs_stoplwp != curlwp) { 1640 mutex_exit(&lfs_lock); 1641 return EALREADY; 1642 } 1643 if (fs->lfs_nowrap == 0) { 1644 mutex_exit(&lfs_lock); 1645 return EBUSY; 1646 } 1647 fs->lfs_wrappass = 1; 1648 wakeup(&fs->lfs_wrappass); 1649 /* Wait for the log to wrap, if asked */ 1650 if (*(int *)ap->a_data) { 1651 mutex_enter(&ap->a_vp->v_interlock); 1652 lfs_vref(ap->a_vp); 1653 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1654 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1655 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1656 "segwrap", 0, &lfs_lock); 1657 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1658 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1659 lfs_vunref(ap->a_vp); 1660 } 1661 mutex_exit(&lfs_lock); 1662 return error; 1663 1664 case LFCNWRAPSTATUS: 1665 mutex_enter(&lfs_lock); 1666 *(int *)ap->a_data = fs->lfs_wrapstatus; 1667 mutex_exit(&lfs_lock); 1668 return 0; 1669 1670 default: 1671 return ufs_fcntl(v); 1672 } 1673 return 0; 1674 } 1675 1676 int 1677 lfs_getpages(void *v) 1678 { 1679 struct vop_getpages_args /* { 1680 struct vnode *a_vp; 1681 voff_t a_offset; 1682 struct vm_page **a_m; 1683 int *a_count; 1684 int a_centeridx; 1685 vm_prot_t a_access_type; 1686 int a_advice; 1687 int a_flags; 1688 } */ *ap = v; 1689 1690 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1691 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1692 return EPERM; 1693 } 1694 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1695 mutex_enter(&lfs_lock); 1696 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1697 mutex_exit(&lfs_lock); 1698 } 1699 1700 /* 1701 * we're relying on the fact that genfs_getpages() always read in 1702 * entire filesystem blocks. 1703 */ 1704 return genfs_getpages(v); 1705 } 1706 1707 /* 1708 * Wait for a page to become unbusy, possibly printing diagnostic messages 1709 * as well. 1710 * 1711 * Called with vp->v_interlock held; return with it held. 1712 */ 1713 static void 1714 wait_for_page(struct vnode *vp, struct vm_page *pg, const char *label) 1715 { 1716 if ((pg->flags & PG_BUSY) == 0) 1717 return; /* Nothing to wait for! */ 1718 1719 #if defined(DEBUG) && defined(UVM_PAGE_TRKOWN) 1720 static struct vm_page *lastpg; 1721 1722 if (label != NULL && pg != lastpg) { 1723 if (pg->owner_tag) { 1724 printf("lfs_putpages[%d.%d]: %s: page %p owner %d.%d [%s]\n", 1725 curproc->p_pid, curlwp->l_lid, label, 1726 pg, pg->owner, pg->lowner, pg->owner_tag); 1727 } else { 1728 printf("lfs_putpages[%d.%d]: %s: page %p unowned?!\n", 1729 curproc->p_pid, curlwp->l_lid, label, pg); 1730 } 1731 } 1732 lastpg = pg; 1733 #endif 1734 1735 pg->flags |= PG_WANTED; 1736 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, "lfsput", 0); 1737 mutex_enter(&vp->v_interlock); 1738 } 1739 1740 /* 1741 * This routine is called by lfs_putpages() when it can't complete the 1742 * write because a page is busy. This means that either (1) someone, 1743 * possibly the pagedaemon, is looking at this page, and will give it up 1744 * presently; or (2) we ourselves are holding the page busy in the 1745 * process of being written (either gathered or actually on its way to 1746 * disk). We don't need to give up the segment lock, but we might need 1747 * to call lfs_writeseg() to expedite the page's journey to disk. 1748 * 1749 * Called with vp->v_interlock held; return with it held. 1750 */ 1751 /* #define BUSYWAIT */ 1752 static void 1753 write_and_wait(struct lfs *fs, struct vnode *vp, struct vm_page *pg, 1754 int seglocked, const char *label) 1755 { 1756 #ifndef BUSYWAIT 1757 struct inode *ip = VTOI(vp); 1758 struct segment *sp = fs->lfs_sp; 1759 int count = 0; 1760 1761 if (pg == NULL) 1762 return; 1763 1764 while (pg->flags & PG_BUSY) { 1765 mutex_exit(&vp->v_interlock); 1766 if (sp->cbpp - sp->bpp > 1) { 1767 /* Write gathered pages */ 1768 lfs_updatemeta(sp); 1769 lfs_release_finfo(fs); 1770 (void) lfs_writeseg(fs, sp); 1771 1772 /* 1773 * Reinitialize FIP 1774 */ 1775 KASSERT(sp->vp == vp); 1776 lfs_acquire_finfo(fs, ip->i_number, 1777 ip->i_gen); 1778 } 1779 ++count; 1780 mutex_enter(&vp->v_interlock); 1781 wait_for_page(vp, pg, label); 1782 } 1783 if (label != NULL && count > 1) 1784 printf("lfs_putpages[%d]: %s: %sn = %d\n", curproc->p_pid, 1785 label, (count > 0 ? "looping, " : ""), count); 1786 #else 1787 preempt(1); 1788 #endif 1789 } 1790 1791 /* 1792 * Make sure that for all pages in every block in the given range, 1793 * either all are dirty or all are clean. If any of the pages 1794 * we've seen so far are dirty, put the vnode on the paging chain, 1795 * and mark it IN_PAGING. 1796 * 1797 * If checkfirst != 0, don't check all the pages but return at the 1798 * first dirty page. 1799 */ 1800 static int 1801 check_dirty(struct lfs *fs, struct vnode *vp, 1802 off_t startoffset, off_t endoffset, off_t blkeof, 1803 int flags, int checkfirst, struct vm_page **pgp) 1804 { 1805 int by_list; 1806 struct vm_page *curpg = NULL; /* XXX: gcc */ 1807 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1808 off_t soff = 0; /* XXX: gcc */ 1809 voff_t off; 1810 int i; 1811 int nonexistent; 1812 int any_dirty; /* number of dirty pages */ 1813 int dirty; /* number of dirty pages in a block */ 1814 int tdirty; 1815 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1816 int pagedaemon = (curlwp == uvm.pagedaemon_lwp); 1817 1818 ASSERT_MAYBE_SEGLOCK(fs); 1819 top: 1820 by_list = (vp->v_uobj.uo_npages <= 1821 ((endoffset - startoffset) >> PAGE_SHIFT) * 1822 UVM_PAGE_TREE_PENALTY); 1823 any_dirty = 0; 1824 1825 if (by_list) { 1826 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1827 } else { 1828 soff = startoffset; 1829 } 1830 while (by_list || soff < MIN(blkeof, endoffset)) { 1831 if (by_list) { 1832 /* 1833 * Find the first page in a block. Skip 1834 * blocks outside our area of interest or beyond 1835 * the end of file. 1836 */ 1837 if (pages_per_block > 1) { 1838 while (curpg && 1839 ((curpg->offset & fs->lfs_bmask) || 1840 curpg->offset >= vp->v_size || 1841 curpg->offset >= endoffset)) 1842 curpg = TAILQ_NEXT(curpg, listq.queue); 1843 } 1844 if (curpg == NULL) 1845 break; 1846 soff = curpg->offset; 1847 } 1848 1849 /* 1850 * Mark all pages in extended range busy; find out if any 1851 * of them are dirty. 1852 */ 1853 nonexistent = dirty = 0; 1854 for (i = 0; i == 0 || i < pages_per_block; i++) { 1855 if (by_list && pages_per_block <= 1) { 1856 pgs[i] = pg = curpg; 1857 } else { 1858 off = soff + (i << PAGE_SHIFT); 1859 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1860 if (pg == NULL) { 1861 ++nonexistent; 1862 continue; 1863 } 1864 } 1865 KASSERT(pg != NULL); 1866 1867 /* 1868 * If we're holding the segment lock, we can deadlock 1869 * against a process that has our page and is waiting 1870 * for the cleaner, while the cleaner waits for the 1871 * segment lock. Just bail in that case. 1872 */ 1873 if ((pg->flags & PG_BUSY) && 1874 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1875 if (i > 0) 1876 uvm_page_unbusy(pgs, i); 1877 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1878 if (pgp) 1879 *pgp = pg; 1880 return -1; 1881 } 1882 1883 while (pg->flags & PG_BUSY) { 1884 wait_for_page(vp, pg, NULL); 1885 if (i > 0) 1886 uvm_page_unbusy(pgs, i); 1887 goto top; 1888 } 1889 pg->flags |= PG_BUSY; 1890 UVM_PAGE_OWN(pg, "lfs_putpages"); 1891 1892 pmap_page_protect(pg, VM_PROT_NONE); 1893 tdirty = (pmap_clear_modify(pg) || 1894 (pg->flags & PG_CLEAN) == 0); 1895 dirty += tdirty; 1896 } 1897 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1898 if (by_list) { 1899 curpg = TAILQ_NEXT(curpg, listq.queue); 1900 } else { 1901 soff += fs->lfs_bsize; 1902 } 1903 continue; 1904 } 1905 1906 any_dirty += dirty; 1907 KASSERT(nonexistent == 0); 1908 1909 /* 1910 * If any are dirty make all dirty; unbusy them, 1911 * but if we were asked to clean, wire them so that 1912 * the pagedaemon doesn't bother us about them while 1913 * they're on their way to disk. 1914 */ 1915 for (i = 0; i == 0 || i < pages_per_block; i++) { 1916 pg = pgs[i]; 1917 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1918 if (dirty) { 1919 pg->flags &= ~PG_CLEAN; 1920 if (flags & PGO_FREE) { 1921 /* 1922 * Wire the page so that 1923 * pdaemon doesn't see it again. 1924 */ 1925 mutex_enter(&uvm_pageqlock); 1926 uvm_pagewire(pg); 1927 mutex_exit(&uvm_pageqlock); 1928 1929 /* Suspended write flag */ 1930 pg->flags |= PG_DELWRI; 1931 } 1932 } 1933 if (pg->flags & PG_WANTED) 1934 wakeup(pg); 1935 pg->flags &= ~(PG_WANTED|PG_BUSY); 1936 UVM_PAGE_OWN(pg, NULL); 1937 } 1938 1939 if (checkfirst && any_dirty) 1940 break; 1941 1942 if (by_list) { 1943 curpg = TAILQ_NEXT(curpg, listq.queue); 1944 } else { 1945 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1946 } 1947 } 1948 1949 return any_dirty; 1950 } 1951 1952 /* 1953 * lfs_putpages functions like genfs_putpages except that 1954 * 1955 * (1) It needs to bounds-check the incoming requests to ensure that 1956 * they are block-aligned; if they are not, expand the range and 1957 * do the right thing in case, e.g., the requested range is clean 1958 * but the expanded range is dirty. 1959 * 1960 * (2) It needs to explicitly send blocks to be written when it is done. 1961 * If VOP_PUTPAGES is called without the seglock held, we simply take 1962 * the seglock and let lfs_segunlock wait for us. 1963 * XXX There might be a bad situation if we have to flush a vnode while 1964 * XXX lfs_markv is in operation. As of this writing we panic in this 1965 * XXX case. 1966 * 1967 * Assumptions: 1968 * 1969 * (1) The caller does not hold any pages in this vnode busy. If it does, 1970 * there is a danger that when we expand the page range and busy the 1971 * pages we will deadlock. 1972 * 1973 * (2) We are called with vp->v_interlock held; we must return with it 1974 * released. 1975 * 1976 * (3) We don't absolutely have to free pages right away, provided that 1977 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1978 * us a request with PGO_FREE, we take the pages out of the paging 1979 * queue and wake up the writer, which will handle freeing them for us. 1980 * 1981 * We ensure that for any filesystem block, all pages for that 1982 * block are either resident or not, even if those pages are higher 1983 * than EOF; that means that we will be getting requests to free 1984 * "unused" pages above EOF all the time, and should ignore them. 1985 * 1986 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 1987 * into has been set up for us by lfs_writefile. If not, we will 1988 * have to handle allocating and/or freeing an finfo entry. 1989 * 1990 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1991 */ 1992 1993 /* How many times to loop before we should start to worry */ 1994 #define TOOMANY 4 1995 1996 int 1997 lfs_putpages(void *v) 1998 { 1999 int error; 2000 struct vop_putpages_args /* { 2001 struct vnode *a_vp; 2002 voff_t a_offlo; 2003 voff_t a_offhi; 2004 int a_flags; 2005 } */ *ap = v; 2006 struct vnode *vp; 2007 struct inode *ip; 2008 struct lfs *fs; 2009 struct segment *sp; 2010 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 2011 off_t off, max_endoffset; 2012 bool seglocked, sync, pagedaemon; 2013 struct vm_page *pg, *busypg; 2014 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 2015 #ifdef DEBUG 2016 int debug_n_again, debug_n_dirtyclean; 2017 #endif 2018 2019 vp = ap->a_vp; 2020 ip = VTOI(vp); 2021 fs = ip->i_lfs; 2022 sync = (ap->a_flags & PGO_SYNCIO) != 0; 2023 pagedaemon = (curlwp == uvm.pagedaemon_lwp); 2024 2025 /* Putpages does nothing for metadata. */ 2026 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 2027 mutex_exit(&vp->v_interlock); 2028 return 0; 2029 } 2030 2031 /* 2032 * If there are no pages, don't do anything. 2033 */ 2034 if (vp->v_uobj.uo_npages == 0) { 2035 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 2036 (vp->v_iflag & VI_ONWORKLST) && 2037 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 2038 vp->v_iflag &= ~VI_WRMAPDIRTY; 2039 vn_syncer_remove_from_worklist(vp); 2040 } 2041 mutex_exit(&vp->v_interlock); 2042 2043 /* Remove us from paging queue, if we were on it */ 2044 mutex_enter(&lfs_lock); 2045 if (ip->i_flags & IN_PAGING) { 2046 ip->i_flags &= ~IN_PAGING; 2047 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2048 } 2049 mutex_exit(&lfs_lock); 2050 return 0; 2051 } 2052 2053 blkeof = blkroundup(fs, ip->i_size); 2054 2055 /* 2056 * Ignore requests to free pages past EOF but in the same block 2057 * as EOF, unless the request is synchronous. (If the request is 2058 * sync, it comes from lfs_truncate.) 2059 * XXXUBC Make these pages look "active" so the pagedaemon won't 2060 * XXXUBC bother us with them again. 2061 */ 2062 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 2063 origoffset = ap->a_offlo; 2064 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 2065 pg = uvm_pagelookup(&vp->v_uobj, off); 2066 KASSERT(pg != NULL); 2067 while (pg->flags & PG_BUSY) { 2068 pg->flags |= PG_WANTED; 2069 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 2070 "lfsput2", 0); 2071 mutex_enter(&vp->v_interlock); 2072 } 2073 mutex_enter(&uvm_pageqlock); 2074 uvm_pageactivate(pg); 2075 mutex_exit(&uvm_pageqlock); 2076 } 2077 ap->a_offlo = blkeof; 2078 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 2079 mutex_exit(&vp->v_interlock); 2080 return 0; 2081 } 2082 } 2083 2084 /* 2085 * Extend page range to start and end at block boundaries. 2086 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 2087 */ 2088 origoffset = ap->a_offlo; 2089 origendoffset = ap->a_offhi; 2090 startoffset = origoffset & ~(fs->lfs_bmask); 2091 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 2092 << fs->lfs_bshift; 2093 2094 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2095 endoffset = max_endoffset; 2096 origendoffset = endoffset; 2097 } else { 2098 origendoffset = round_page(ap->a_offhi); 2099 endoffset = round_page(blkroundup(fs, origendoffset)); 2100 } 2101 2102 KASSERT(startoffset > 0 || endoffset >= startoffset); 2103 if (startoffset == endoffset) { 2104 /* Nothing to do, why were we called? */ 2105 mutex_exit(&vp->v_interlock); 2106 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 2107 PRId64 "\n", startoffset)); 2108 return 0; 2109 } 2110 2111 ap->a_offlo = startoffset; 2112 ap->a_offhi = endoffset; 2113 2114 /* 2115 * If not cleaning, just send the pages through genfs_putpages 2116 * to be returned to the pool. 2117 */ 2118 if (!(ap->a_flags & PGO_CLEANIT)) 2119 return genfs_putpages(v); 2120 2121 /* Set PGO_BUSYFAIL to avoid deadlocks */ 2122 ap->a_flags |= PGO_BUSYFAIL; 2123 2124 /* 2125 * Likewise, if we are asked to clean but the pages are not 2126 * dirty, we can just free them using genfs_putpages. 2127 */ 2128 #ifdef DEBUG 2129 debug_n_dirtyclean = 0; 2130 #endif 2131 do { 2132 int r; 2133 2134 /* Count the number of dirty pages */ 2135 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2136 ap->a_flags, 1, NULL); 2137 if (r < 0) { 2138 /* Pages are busy with another process */ 2139 mutex_exit(&vp->v_interlock); 2140 return EDEADLK; 2141 } 2142 if (r > 0) /* Some pages are dirty */ 2143 break; 2144 2145 /* 2146 * Sometimes pages are dirtied between the time that 2147 * we check and the time we try to clean them. 2148 * Instruct lfs_gop_write to return EDEADLK in this case 2149 * so we can write them properly. 2150 */ 2151 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2152 r = genfs_do_putpages(vp, startoffset, endoffset, 2153 ap->a_flags, &busypg); 2154 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2155 if (r != EDEADLK) 2156 return r; 2157 2158 /* One of the pages was busy. Start over. */ 2159 mutex_enter(&vp->v_interlock); 2160 wait_for_page(vp, busypg, "dirtyclean"); 2161 #ifdef DEBUG 2162 ++debug_n_dirtyclean; 2163 #endif 2164 } while(1); 2165 2166 #ifdef DEBUG 2167 if (debug_n_dirtyclean > TOOMANY) 2168 printf("lfs_putpages: dirtyclean: looping, n = %d\n", 2169 debug_n_dirtyclean); 2170 #endif 2171 2172 /* 2173 * Dirty and asked to clean. 2174 * 2175 * Pagedaemon can't actually write LFS pages; wake up 2176 * the writer to take care of that. The writer will 2177 * notice the pager inode queue and act on that. 2178 */ 2179 if (pagedaemon) { 2180 mutex_enter(&lfs_lock); 2181 if (!(ip->i_flags & IN_PAGING)) { 2182 ip->i_flags |= IN_PAGING; 2183 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2184 } 2185 wakeup(&lfs_writer_daemon); 2186 mutex_exit(&lfs_lock); 2187 mutex_exit(&vp->v_interlock); 2188 preempt(); 2189 return EWOULDBLOCK; 2190 } 2191 2192 /* 2193 * If this is a file created in a recent dirop, we can't flush its 2194 * inode until the dirop is complete. Drain dirops, then flush the 2195 * filesystem (taking care of any other pending dirops while we're 2196 * at it). 2197 */ 2198 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2199 (vp->v_uflag & VU_DIROP)) { 2200 int locked; 2201 2202 DLOG((DLOG_PAGE, "lfs_putpages: flushing VU_DIROP\n")); 2203 locked = (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 2204 mutex_exit(&vp->v_interlock); 2205 lfs_writer_enter(fs, "ppdirop"); 2206 if (locked) 2207 VOP_UNLOCK(vp, 0); /* XXX why? */ 2208 2209 mutex_enter(&lfs_lock); 2210 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2211 mutex_exit(&lfs_lock); 2212 2213 mutex_enter(&vp->v_interlock); 2214 if (locked) { 2215 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2216 mutex_enter(&vp->v_interlock); 2217 } 2218 lfs_writer_leave(fs); 2219 2220 /* XXX the flush should have taken care of this one too! */ 2221 } 2222 2223 /* 2224 * This is it. We are going to write some pages. From here on 2225 * down it's all just mechanics. 2226 * 2227 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2228 */ 2229 ap->a_flags &= ~PGO_SYNCIO; 2230 2231 /* 2232 * If we've already got the seglock, flush the node and return. 2233 * The FIP has already been set up for us by lfs_writefile, 2234 * and FIP cleanup and lfs_updatemeta will also be done there, 2235 * unless genfs_putpages returns EDEADLK; then we must flush 2236 * what we have, and correct FIP and segment header accounting. 2237 */ 2238 get_seglock: 2239 /* 2240 * If we are not called with the segment locked, lock it. 2241 * Account for a new FIP in the segment header, and set sp->vp. 2242 * (This should duplicate the setup at the top of lfs_writefile().) 2243 */ 2244 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2245 if (!seglocked) { 2246 mutex_exit(&vp->v_interlock); 2247 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2248 if (error != 0) 2249 return error; 2250 mutex_enter(&vp->v_interlock); 2251 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2252 } 2253 sp = fs->lfs_sp; 2254 KASSERT(sp->vp == NULL); 2255 sp->vp = vp; 2256 2257 /* 2258 * Ensure that the partial segment is marked SS_DIROP if this 2259 * vnode is a DIROP. 2260 */ 2261 if (!seglocked && vp->v_uflag & VU_DIROP) 2262 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2263 2264 /* 2265 * Loop over genfs_putpages until all pages are gathered. 2266 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2267 * Whenever we lose the interlock we have to rerun check_dirty, as 2268 * well, since more pages might have been dirtied in our absence. 2269 */ 2270 #ifdef DEBUG 2271 debug_n_again = 0; 2272 #endif 2273 do { 2274 busypg = NULL; 2275 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2276 ap->a_flags, 0, &busypg) < 0) { 2277 mutex_exit(&vp->v_interlock); 2278 2279 mutex_enter(&vp->v_interlock); 2280 write_and_wait(fs, vp, busypg, seglocked, NULL); 2281 if (!seglocked) { 2282 lfs_release_finfo(fs); 2283 lfs_segunlock(fs); 2284 } 2285 sp->vp = NULL; 2286 goto get_seglock; 2287 } 2288 2289 busypg = NULL; 2290 error = genfs_do_putpages(vp, startoffset, endoffset, 2291 ap->a_flags, &busypg); 2292 2293 if (error == EDEADLK || error == EAGAIN) { 2294 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2295 " %d ino %d off %x (seg %d)\n", error, 2296 ip->i_number, fs->lfs_offset, 2297 dtosn(fs, fs->lfs_offset))); 2298 2299 mutex_enter(&vp->v_interlock); 2300 write_and_wait(fs, vp, busypg, seglocked, "again"); 2301 } 2302 #ifdef DEBUG 2303 ++debug_n_again; 2304 #endif 2305 } while (error == EDEADLK); 2306 #ifdef DEBUG 2307 if (debug_n_again > TOOMANY) 2308 printf("lfs_putpages: again: looping, n = %d\n", debug_n_again); 2309 #endif 2310 2311 KASSERT(sp != NULL && sp->vp == vp); 2312 if (!seglocked) { 2313 sp->vp = NULL; 2314 2315 /* Write indirect blocks as well */ 2316 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2317 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2318 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2319 2320 KASSERT(sp->vp == NULL); 2321 sp->vp = vp; 2322 } 2323 2324 /* 2325 * Blocks are now gathered into a segment waiting to be written. 2326 * All that's left to do is update metadata, and write them. 2327 */ 2328 lfs_updatemeta(sp); 2329 KASSERT(sp->vp == vp); 2330 sp->vp = NULL; 2331 2332 /* 2333 * If we were called from lfs_writefile, we don't need to clean up 2334 * the FIP or unlock the segment lock. We're done. 2335 */ 2336 if (seglocked) 2337 return error; 2338 2339 /* Clean up FIP and send it to disk. */ 2340 lfs_release_finfo(fs); 2341 lfs_writeseg(fs, fs->lfs_sp); 2342 2343 /* 2344 * Remove us from paging queue if we wrote all our pages. 2345 */ 2346 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2347 mutex_enter(&lfs_lock); 2348 if (ip->i_flags & IN_PAGING) { 2349 ip->i_flags &= ~IN_PAGING; 2350 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2351 } 2352 mutex_exit(&lfs_lock); 2353 } 2354 2355 /* 2356 * XXX - with the malloc/copy writeseg, the pages are freed by now 2357 * even if we don't wait (e.g. if we hold a nested lock). This 2358 * will not be true if we stop using malloc/copy. 2359 */ 2360 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2361 lfs_segunlock(fs); 2362 2363 /* 2364 * Wait for v_numoutput to drop to zero. The seglock should 2365 * take care of this, but there is a slight possibility that 2366 * aiodoned might not have got around to our buffers yet. 2367 */ 2368 if (sync) { 2369 mutex_enter(&vp->v_interlock); 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 cv_wait(&vp->v_cv, &vp->v_interlock); 2374 } 2375 mutex_exit(&vp->v_interlock); 2376 } 2377 return error; 2378 } 2379 2380 /* 2381 * Return the last logical file offset that should be written for this file 2382 * if we're doing a write that ends at "size". If writing, we need to know 2383 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2384 * to know about entire blocks. 2385 */ 2386 void 2387 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2388 { 2389 struct inode *ip = VTOI(vp); 2390 struct lfs *fs = ip->i_lfs; 2391 daddr_t olbn, nlbn; 2392 2393 olbn = lblkno(fs, ip->i_size); 2394 nlbn = lblkno(fs, size); 2395 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2396 *eobp = fragroundup(fs, size); 2397 } else { 2398 *eobp = blkroundup(fs, size); 2399 } 2400 } 2401 2402 #ifdef DEBUG 2403 void lfs_dump_vop(void *); 2404 2405 void 2406 lfs_dump_vop(void *v) 2407 { 2408 struct vop_putpages_args /* { 2409 struct vnode *a_vp; 2410 voff_t a_offlo; 2411 voff_t a_offhi; 2412 int a_flags; 2413 } */ *ap = v; 2414 2415 #ifdef DDB 2416 vfs_vnode_print(ap->a_vp, 0, printf); 2417 #endif 2418 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2419 } 2420 #endif 2421 2422 int 2423 lfs_mmap(void *v) 2424 { 2425 struct vop_mmap_args /* { 2426 const struct vnodeop_desc *a_desc; 2427 struct vnode *a_vp; 2428 vm_prot_t a_prot; 2429 kauth_cred_t a_cred; 2430 } */ *ap = v; 2431 2432 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2433 return EOPNOTSUPP; 2434 return ufs_mmap(v); 2435 } 2436
Indexes created Tue Sep 30 20:09:53 GMT 2025