lfs_vnops.c revision 1.211
1 /* $NetBSD: lfs_vnops.c,v 1.211 2007/10/08 18:01:31 ad 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 1.211 2007/10/08 18:01:31 ad 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 struct lwp *a_l; 281 } */ *ap = v; 282 struct vnode *vp = ap->a_vp; 283 int error, wait; 284 struct inode *ip = VTOI(vp); 285 struct lfs *fs = ip->i_lfs; 286 287 /* If we're mounted read-only, don't try to sync. */ 288 if (fs->lfs_ronly) 289 return 0; 290 291 /* 292 * Trickle sync simply adds this vnode to the pager list, as if 293 * the pagedaemon had requested a pageout. 294 */ 295 if (ap->a_flags & FSYNC_LAZY) { 296 if (lfs_ignore_lazy_sync == 0) { 297 simple_lock(&fs->lfs_interlock); 298 if (!(ip->i_flags & IN_PAGING)) { 299 ip->i_flags |= IN_PAGING; 300 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, 301 i_lfs_pchain); 302 } 303 simple_unlock(&fs->lfs_interlock); 304 simple_lock(&lfs_subsys_lock); 305 wakeup(&lfs_writer_daemon); 306 simple_unlock(&lfs_subsys_lock); 307 } 308 return 0; 309 } 310 311 /* 312 * If a vnode is bring cleaned, flush it out before we try to 313 * reuse it. This prevents the cleaner from writing files twice 314 * in the same partial segment, causing an accounting underflow. 315 */ 316 if (ap->a_flags & FSYNC_RECLAIM && ip->i_flags & IN_CLEANING) { 317 lfs_vflush(vp); 318 } 319 320 wait = (ap->a_flags & FSYNC_WAIT); 321 do { 322 simple_lock(&vp->v_interlock); 323 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 324 round_page(ap->a_offhi), 325 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 326 if (error == EAGAIN) { 327 simple_lock(&fs->lfs_interlock); 328 ltsleep(&fs->lfs_avail, PCATCH | PUSER, "lfs_fsync", 329 hz / 100 + 1, &fs->lfs_interlock); 330 simple_unlock(&fs->lfs_interlock); 331 } 332 } while (error == EAGAIN); 333 if (error) 334 return error; 335 336 if ((ap->a_flags & FSYNC_DATAONLY) == 0) 337 error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 338 339 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 340 int l = 0; 341 error = VOP_IOCTL(ip->i_devvp, DIOCCACHESYNC, &l, FWRITE, 342 ap->a_l->l_cred, ap->a_l); 343 } 344 if (wait && !VPISEMPTY(vp)) 345 LFS_SET_UINO(ip, IN_MODIFIED); 346 347 return error; 348 } 349 350 /* 351 * Take IN_ADIROP off, then call ufs_inactive. 352 */ 353 int 354 lfs_inactive(void *v) 355 { 356 struct vop_inactive_args /* { 357 struct vnode *a_vp; 358 struct lwp *a_l; 359 } */ *ap = v; 360 361 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 362 363 lfs_unmark_vnode(ap->a_vp); 364 365 /* 366 * The Ifile is only ever inactivated on unmount. 367 * Streamline this process by not giving it more dirty blocks. 368 */ 369 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 370 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 371 VOP_UNLOCK(ap->a_vp, 0); 372 return 0; 373 } 374 375 return ufs_inactive(v); 376 } 377 378 /* 379 * These macros are used to bracket UFS directory ops, so that we can 380 * identify all the pages touched during directory ops which need to 381 * be ordered and flushed atomically, so that they may be recovered. 382 * 383 * Because we have to mark nodes VDIROP in order to prevent 384 * the cache from reclaiming them while a dirop is in progress, we must 385 * also manage the number of nodes so marked (otherwise we can run out). 386 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 387 * is decremented during segment write, when VDIROP is taken off. 388 */ 389 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 390 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 391 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 392 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 393 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 394 static int lfs_set_dirop(struct vnode *, struct vnode *); 395 396 static int 397 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 398 { 399 struct lfs *fs; 400 int error; 401 402 KASSERT(VOP_ISLOCKED(dvp)); 403 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 404 405 fs = VTOI(dvp)->i_lfs; 406 407 ASSERT_NO_SEGLOCK(fs); 408 /* 409 * LFS_NRESERVE calculates direct and indirect blocks as well 410 * as an inode block; an overestimate in most cases. 411 */ 412 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 413 return (error); 414 415 restart: 416 simple_lock(&fs->lfs_interlock); 417 if (fs->lfs_dirops == 0) { 418 simple_unlock(&fs->lfs_interlock); 419 lfs_check(dvp, LFS_UNUSED_LBN, 0); 420 simple_lock(&fs->lfs_interlock); 421 } 422 while (fs->lfs_writer) { 423 error = ltsleep(&fs->lfs_dirops, (PRIBIO + 1) | PCATCH, 424 "lfs_sdirop", 0, &fs->lfs_interlock); 425 if (error == EINTR) { 426 simple_unlock(&fs->lfs_interlock); 427 goto unreserve; 428 } 429 } 430 simple_lock(&lfs_subsys_lock); 431 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 432 wakeup(&lfs_writer_daemon); 433 simple_unlock(&lfs_subsys_lock); 434 simple_unlock(&fs->lfs_interlock); 435 preempt(); 436 goto restart; 437 } 438 439 if (lfs_dirvcount > LFS_MAX_DIROP) { 440 simple_unlock(&fs->lfs_interlock); 441 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 442 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 443 if ((error = ltsleep(&lfs_dirvcount, 444 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 445 &lfs_subsys_lock)) != 0) { 446 goto unreserve; 447 } 448 goto restart; 449 } 450 simple_unlock(&lfs_subsys_lock); 451 452 ++fs->lfs_dirops; 453 fs->lfs_doifile = 1; 454 simple_unlock(&fs->lfs_interlock); 455 456 /* Hold a reference so SET_ENDOP will be happy */ 457 vref(dvp); 458 if (vp) { 459 vref(vp); 460 MARK_VNODE(vp); 461 } 462 463 MARK_VNODE(dvp); 464 return 0; 465 466 unreserve: 467 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 468 return error; 469 } 470 471 /* 472 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 473 * in getnewvnode(), if we have a stacked filesystem mounted on top 474 * of us. 475 * 476 * NB: this means we have to clear the new vnodes on error. Fortunately 477 * SET_ENDOP is there to do that for us. 478 */ 479 static int 480 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 481 { 482 int error; 483 struct lfs *fs; 484 485 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 486 ASSERT_NO_SEGLOCK(fs); 487 if (fs->lfs_ronly) 488 return EROFS; 489 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 490 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 491 dvp, error)); 492 return error; 493 } 494 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 495 if (vpp) { 496 ungetnewvnode(*vpp); 497 *vpp = NULL; 498 } 499 return error; 500 } 501 return 0; 502 } 503 504 #define SET_ENDOP_BASE(fs, dvp, str) \ 505 do { \ 506 simple_lock(&(fs)->lfs_interlock); \ 507 --(fs)->lfs_dirops; \ 508 if (!(fs)->lfs_dirops) { \ 509 if ((fs)->lfs_nadirop) { \ 510 panic("SET_ENDOP: %s: no dirops but " \ 511 " nadirop=%d", (str), \ 512 (fs)->lfs_nadirop); \ 513 } \ 514 wakeup(&(fs)->lfs_writer); \ 515 simple_unlock(&(fs)->lfs_interlock); \ 516 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 517 } else \ 518 simple_unlock(&(fs)->lfs_interlock); \ 519 } while(0) 520 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 521 do { \ 522 UNMARK_VNODE(dvp); \ 523 if (nvpp && *nvpp) \ 524 UNMARK_VNODE(*nvpp); \ 525 /* Check for error return to stem vnode leakage */ \ 526 if (nvpp && *nvpp && !((*nvpp)->v_flag & VDIROP)) \ 527 ungetnewvnode(*(nvpp)); \ 528 SET_ENDOP_BASE((fs), (dvp), (str)); \ 529 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 530 vrele(dvp); \ 531 } while(0) 532 #define SET_ENDOP_CREATE_AP(ap, str) \ 533 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 534 (ap)->a_vpp, (str)) 535 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 536 do { \ 537 UNMARK_VNODE(dvp); \ 538 if (ovp) \ 539 UNMARK_VNODE(ovp); \ 540 SET_ENDOP_BASE((fs), (dvp), (str)); \ 541 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 542 vrele(dvp); \ 543 if (ovp) \ 544 vrele(ovp); \ 545 } while(0) 546 547 void 548 lfs_mark_vnode(struct vnode *vp) 549 { 550 struct inode *ip = VTOI(vp); 551 struct lfs *fs = ip->i_lfs; 552 553 simple_lock(&fs->lfs_interlock); 554 if (!(ip->i_flag & IN_ADIROP)) { 555 if (!(vp->v_flag & VDIROP)) { 556 (void)lfs_vref(vp); 557 simple_lock(&lfs_subsys_lock); 558 ++lfs_dirvcount; 559 ++fs->lfs_dirvcount; 560 simple_unlock(&lfs_subsys_lock); 561 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 562 vp->v_flag |= VDIROP; 563 } 564 ++fs->lfs_nadirop; 565 ip->i_flag |= IN_ADIROP; 566 } else 567 KASSERT(vp->v_flag & VDIROP); 568 simple_unlock(&fs->lfs_interlock); 569 } 570 571 void 572 lfs_unmark_vnode(struct vnode *vp) 573 { 574 struct inode *ip = VTOI(vp); 575 576 if (ip && (ip->i_flag & IN_ADIROP)) { 577 KASSERT(vp->v_flag & VDIROP); 578 simple_lock(&ip->i_lfs->lfs_interlock); 579 --ip->i_lfs->lfs_nadirop; 580 simple_unlock(&ip->i_lfs->lfs_interlock); 581 ip->i_flag &= ~IN_ADIROP; 582 } 583 } 584 585 int 586 lfs_symlink(void *v) 587 { 588 struct vop_symlink_args /* { 589 struct vnode *a_dvp; 590 struct vnode **a_vpp; 591 struct componentname *a_cnp; 592 struct vattr *a_vap; 593 char *a_target; 594 } */ *ap = v; 595 int error; 596 597 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 598 vput(ap->a_dvp); 599 return error; 600 } 601 error = ufs_symlink(ap); 602 SET_ENDOP_CREATE_AP(ap, "symlink"); 603 return (error); 604 } 605 606 int 607 lfs_mknod(void *v) 608 { 609 struct vop_mknod_args /* { 610 struct vnode *a_dvp; 611 struct vnode **a_vpp; 612 struct componentname *a_cnp; 613 struct vattr *a_vap; 614 } */ *ap = v; 615 struct vattr *vap = ap->a_vap; 616 struct vnode **vpp = ap->a_vpp; 617 struct inode *ip; 618 int error; 619 struct mount *mp; 620 ino_t ino; 621 622 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 623 vput(ap->a_dvp); 624 return error; 625 } 626 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 627 ap->a_dvp, vpp, ap->a_cnp); 628 629 /* Either way we're done with the dirop at this point */ 630 SET_ENDOP_CREATE_AP(ap, "mknod"); 631 632 if (error) 633 return (error); 634 635 ip = VTOI(*vpp); 636 mp = (*vpp)->v_mount; 637 ino = ip->i_number; 638 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 639 if (vap->va_rdev != VNOVAL) { 640 /* 641 * Want to be able to use this to make badblock 642 * inodes, so don't truncate the dev number. 643 */ 644 #if 0 645 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 646 UFS_MPNEEDSWAP((*vpp)->v_mount)); 647 #else 648 ip->i_ffs1_rdev = vap->va_rdev; 649 #endif 650 } 651 652 /* 653 * Call fsync to write the vnode so that we don't have to deal with 654 * flushing it when it's marked VDIROP|VXLOCK. 655 * 656 * XXX KS - If we can't flush we also can't call vgone(), so must 657 * return. But, that leaves this vnode in limbo, also not good. 658 * Can this ever happen (barring hardware failure)? 659 */ 660 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 661 curlwp)) != 0) { 662 panic("lfs_mknod: couldn't fsync (ino %llu)", 663 (unsigned long long)ino); 664 /* return (error); */ 665 } 666 /* 667 * Remove vnode so that it will be reloaded by VFS_VGET and 668 * checked to see if it is an alias of an existing entry in 669 * the inode cache. 670 */ 671 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 672 673 VOP_UNLOCK(*vpp, 0); 674 lfs_vunref(*vpp); 675 (*vpp)->v_type = VNON; 676 vgone(*vpp); 677 error = VFS_VGET(mp, ino, vpp); 678 679 if (error != 0) { 680 *vpp = NULL; 681 return (error); 682 } 683 return (0); 684 } 685 686 int 687 lfs_create(void *v) 688 { 689 struct vop_create_args /* { 690 struct vnode *a_dvp; 691 struct vnode **a_vpp; 692 struct componentname *a_cnp; 693 struct vattr *a_vap; 694 } */ *ap = v; 695 int error; 696 697 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 698 vput(ap->a_dvp); 699 return error; 700 } 701 error = ufs_create(ap); 702 SET_ENDOP_CREATE_AP(ap, "create"); 703 return (error); 704 } 705 706 int 707 lfs_mkdir(void *v) 708 { 709 struct vop_mkdir_args /* { 710 struct vnode *a_dvp; 711 struct vnode **a_vpp; 712 struct componentname *a_cnp; 713 struct vattr *a_vap; 714 } */ *ap = v; 715 int error; 716 717 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 718 vput(ap->a_dvp); 719 return error; 720 } 721 error = ufs_mkdir(ap); 722 SET_ENDOP_CREATE_AP(ap, "mkdir"); 723 return (error); 724 } 725 726 int 727 lfs_remove(void *v) 728 { 729 struct vop_remove_args /* { 730 struct vnode *a_dvp; 731 struct vnode *a_vp; 732 struct componentname *a_cnp; 733 } */ *ap = v; 734 struct vnode *dvp, *vp; 735 struct inode *ip; 736 int error; 737 738 dvp = ap->a_dvp; 739 vp = ap->a_vp; 740 ip = VTOI(vp); 741 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 742 if (dvp == vp) 743 vrele(vp); 744 else 745 vput(vp); 746 vput(dvp); 747 return error; 748 } 749 error = ufs_remove(ap); 750 if (ip->i_nlink == 0) 751 lfs_orphan(ip->i_lfs, ip->i_number); 752 SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove"); 753 return (error); 754 } 755 756 int 757 lfs_rmdir(void *v) 758 { 759 struct vop_rmdir_args /* { 760 struct vnodeop_desc *a_desc; 761 struct vnode *a_dvp; 762 struct vnode *a_vp; 763 struct componentname *a_cnp; 764 } */ *ap = v; 765 struct vnode *vp; 766 struct inode *ip; 767 int error; 768 769 vp = ap->a_vp; 770 ip = VTOI(vp); 771 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 772 if (ap->a_dvp == vp) 773 vrele(ap->a_dvp); 774 else 775 vput(ap->a_dvp); 776 vput(vp); 777 return error; 778 } 779 error = ufs_rmdir(ap); 780 if (ip->i_nlink == 0) 781 lfs_orphan(ip->i_lfs, ip->i_number); 782 SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 783 return (error); 784 } 785 786 int 787 lfs_link(void *v) 788 { 789 struct vop_link_args /* { 790 struct vnode *a_dvp; 791 struct vnode *a_vp; 792 struct componentname *a_cnp; 793 } */ *ap = v; 794 int error; 795 struct vnode **vpp = NULL; 796 797 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 798 vput(ap->a_dvp); 799 return error; 800 } 801 error = ufs_link(ap); 802 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 803 return (error); 804 } 805 806 int 807 lfs_rename(void *v) 808 { 809 struct vop_rename_args /* { 810 struct vnode *a_fdvp; 811 struct vnode *a_fvp; 812 struct componentname *a_fcnp; 813 struct vnode *a_tdvp; 814 struct vnode *a_tvp; 815 struct componentname *a_tcnp; 816 } */ *ap = v; 817 struct vnode *tvp, *fvp, *tdvp, *fdvp; 818 struct componentname *tcnp, *fcnp; 819 int error; 820 struct lfs *fs; 821 822 fs = VTOI(ap->a_fdvp)->i_lfs; 823 tvp = ap->a_tvp; 824 tdvp = ap->a_tdvp; 825 tcnp = ap->a_tcnp; 826 fvp = ap->a_fvp; 827 fdvp = ap->a_fdvp; 828 fcnp = ap->a_fcnp; 829 830 /* 831 * Check for cross-device rename. 832 * If it is, we don't want to set dirops, just error out. 833 * (In particular note that MARK_VNODE(tdvp) will DTWT on 834 * a cross-device rename.) 835 * 836 * Copied from ufs_rename. 837 */ 838 if ((fvp->v_mount != tdvp->v_mount) || 839 (tvp && (fvp->v_mount != tvp->v_mount))) { 840 error = EXDEV; 841 goto errout; 842 } 843 844 /* 845 * Check to make sure we're not renaming a vnode onto itself 846 * (deleting a hard link by renaming one name onto another); 847 * if we are we can't recursively call VOP_REMOVE since that 848 * would leave us with an unaccounted-for number of live dirops. 849 * 850 * Inline the relevant section of ufs_rename here, *before* 851 * calling SET_DIROP_REMOVE. 852 */ 853 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 854 (VTOI(tdvp)->i_flags & APPEND))) { 855 error = EPERM; 856 goto errout; 857 } 858 if (fvp == tvp) { 859 if (fvp->v_type == VDIR) { 860 error = EINVAL; 861 goto errout; 862 } 863 864 /* Release destination completely. */ 865 VOP_ABORTOP(tdvp, tcnp); 866 vput(tdvp); 867 vput(tvp); 868 869 /* Delete source. */ 870 vrele(fvp); 871 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 872 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 873 fcnp->cn_nameiop = DELETE; 874 vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY); 875 if ((error = relookup(fdvp, &fvp, fcnp))) { 876 vput(fdvp); 877 return (error); 878 } 879 return (VOP_REMOVE(fdvp, fvp, fcnp)); 880 } 881 882 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 883 goto errout; 884 MARK_VNODE(fdvp); 885 MARK_VNODE(fvp); 886 887 error = ufs_rename(ap); 888 UNMARK_VNODE(fdvp); 889 UNMARK_VNODE(fvp); 890 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 891 return (error); 892 893 errout: 894 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 895 if (tdvp == tvp) 896 vrele(tdvp); 897 else 898 vput(tdvp); 899 if (tvp) 900 vput(tvp); 901 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 902 vrele(fdvp); 903 vrele(fvp); 904 return (error); 905 } 906 907 /* XXX hack to avoid calling ITIMES in getattr */ 908 int 909 lfs_getattr(void *v) 910 { 911 struct vop_getattr_args /* { 912 struct vnode *a_vp; 913 struct vattr *a_vap; 914 kauth_cred_t a_cred; 915 struct lwp *a_l; 916 } */ *ap = v; 917 struct vnode *vp = ap->a_vp; 918 struct inode *ip = VTOI(vp); 919 struct vattr *vap = ap->a_vap; 920 struct lfs *fs = ip->i_lfs; 921 /* 922 * Copy from inode table 923 */ 924 vap->va_fsid = ip->i_dev; 925 vap->va_fileid = ip->i_number; 926 vap->va_mode = ip->i_mode & ~IFMT; 927 vap->va_nlink = ip->i_nlink; 928 vap->va_uid = ip->i_uid; 929 vap->va_gid = ip->i_gid; 930 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 931 vap->va_size = vp->v_size; 932 vap->va_atime.tv_sec = ip->i_ffs1_atime; 933 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 934 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 935 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 936 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 937 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 938 vap->va_flags = ip->i_flags; 939 vap->va_gen = ip->i_gen; 940 /* this doesn't belong here */ 941 if (vp->v_type == VBLK) 942 vap->va_blocksize = BLKDEV_IOSIZE; 943 else if (vp->v_type == VCHR) 944 vap->va_blocksize = MAXBSIZE; 945 else 946 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 947 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 948 vap->va_type = vp->v_type; 949 vap->va_filerev = ip->i_modrev; 950 return (0); 951 } 952 953 /* 954 * Check to make sure the inode blocks won't choke the buffer 955 * cache, then call ufs_setattr as usual. 956 */ 957 int 958 lfs_setattr(void *v) 959 { 960 struct vop_setattr_args /* { 961 struct vnode *a_vp; 962 struct vattr *a_vap; 963 kauth_cred_t a_cred; 964 struct lwp *a_l; 965 } */ *ap = v; 966 struct vnode *vp = ap->a_vp; 967 968 lfs_check(vp, LFS_UNUSED_LBN, 0); 969 return ufs_setattr(v); 970 } 971 972 /* 973 * Release the block we hold on lfs_newseg wrapping. Called on file close, 974 * or explicitly from LFCNWRAPGO. Called with the interlock held. 975 */ 976 static int 977 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor) 978 { 979 if (lockstatus(&fs->lfs_stoplock) != LK_EXCLUSIVE) 980 return EBUSY; 981 982 lockmgr(&fs->lfs_stoplock, LK_RELEASE, &fs->lfs_interlock); 983 984 KASSERT(fs->lfs_nowrap > 0); 985 if (fs->lfs_nowrap <= 0) { 986 return 0; 987 } 988 989 if (--fs->lfs_nowrap == 0) { 990 log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt); 991 wakeup(&fs->lfs_wrappass); 992 lfs_wakeup_cleaner(fs); 993 } 994 if (waitfor) { 995 ltsleep(&fs->lfs_nextseg, PCATCH | PUSER, 996 "segment", 0, &fs->lfs_interlock); 997 } 998 999 return 0; 1000 } 1001 1002 /* 1003 * Close called 1004 */ 1005 /* ARGSUSED */ 1006 int 1007 lfs_close(void *v) 1008 { 1009 struct vop_close_args /* { 1010 struct vnode *a_vp; 1011 int a_fflag; 1012 kauth_cred_t a_cred; 1013 struct lwp *a_l; 1014 } */ *ap = v; 1015 struct vnode *vp = ap->a_vp; 1016 struct inode *ip = VTOI(vp); 1017 struct lfs *fs = ip->i_lfs; 1018 1019 if ((ip->i_number == ROOTINO || ip->i_number == LFS_IFILE_INUM) && 1020 lockstatus(&fs->lfs_stoplock) == LK_EXCLUSIVE) { 1021 simple_lock(&fs->lfs_interlock); 1022 log(LOG_NOTICE, "lfs_close: releasing log wrap control\n"); 1023 lfs_wrapgo(fs, ip, 0); 1024 simple_unlock(&fs->lfs_interlock); 1025 } 1026 1027 if (vp == ip->i_lfs->lfs_ivnode && 1028 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 1029 return 0; 1030 1031 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 1032 LFS_ITIMES(ip, NULL, NULL, NULL); 1033 } 1034 return (0); 1035 } 1036 1037 /* 1038 * Close wrapper for special devices. 1039 * 1040 * Update the times on the inode then do device close. 1041 */ 1042 int 1043 lfsspec_close(void *v) 1044 { 1045 struct vop_close_args /* { 1046 struct vnode *a_vp; 1047 int a_fflag; 1048 kauth_cred_t a_cred; 1049 struct lwp *a_l; 1050 } */ *ap = v; 1051 struct vnode *vp; 1052 struct inode *ip; 1053 1054 vp = ap->a_vp; 1055 ip = VTOI(vp); 1056 if (vp->v_usecount > 1) { 1057 LFS_ITIMES(ip, NULL, NULL, NULL); 1058 } 1059 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1060 } 1061 1062 /* 1063 * Close wrapper for fifo's. 1064 * 1065 * Update the times on the inode then do device close. 1066 */ 1067 int 1068 lfsfifo_close(void *v) 1069 { 1070 struct vop_close_args /* { 1071 struct vnode *a_vp; 1072 int a_fflag; 1073 kauth_cred_ a_cred; 1074 struct lwp *a_l; 1075 } */ *ap = v; 1076 struct vnode *vp; 1077 struct inode *ip; 1078 1079 vp = ap->a_vp; 1080 ip = VTOI(vp); 1081 if (ap->a_vp->v_usecount > 1) { 1082 LFS_ITIMES(ip, NULL, NULL, NULL); 1083 } 1084 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1085 } 1086 1087 /* 1088 * Reclaim an inode so that it can be used for other purposes. 1089 */ 1090 1091 int 1092 lfs_reclaim(void *v) 1093 { 1094 struct vop_reclaim_args /* { 1095 struct vnode *a_vp; 1096 struct lwp *a_l; 1097 } */ *ap = v; 1098 struct vnode *vp = ap->a_vp; 1099 struct inode *ip = VTOI(vp); 1100 struct lfs *fs = ip->i_lfs; 1101 int error; 1102 1103 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1104 1105 LFS_CLR_UINO(ip, IN_ALLMOD); 1106 if ((error = ufs_reclaim(vp, ap->a_l))) 1107 return (error); 1108 1109 /* 1110 * Take us off the paging and/or dirop queues if we were on them. 1111 * We shouldn't be on them. 1112 */ 1113 simple_lock(&fs->lfs_interlock); 1114 if (ip->i_flags & IN_PAGING) { 1115 log(LOG_WARNING, "%s: reclaimed vnode is IN_PAGING\n", 1116 fs->lfs_fsmnt); 1117 ip->i_flags &= ~IN_PAGING; 1118 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1119 } 1120 if (vp->v_flag & VDIROP) { 1121 panic("reclaimed vnode is VDIROP"); 1122 vp->v_flag &= ~VDIROP; 1123 TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain); 1124 } 1125 simple_unlock(&fs->lfs_interlock); 1126 1127 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1128 lfs_deregister_all(vp); 1129 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1130 ip->inode_ext.lfs = NULL; 1131 genfs_node_destroy(vp); 1132 pool_put(&lfs_inode_pool, vp->v_data); 1133 vp->v_data = NULL; 1134 return (0); 1135 } 1136 1137 /* 1138 * Read a block from a storage device. 1139 * In order to avoid reading blocks that are in the process of being 1140 * written by the cleaner---and hence are not mutexed by the normal 1141 * buffer cache / page cache mechanisms---check for collisions before 1142 * reading. 1143 * 1144 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1145 * the active cleaner test. 1146 * 1147 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1148 */ 1149 int 1150 lfs_strategy(void *v) 1151 { 1152 struct vop_strategy_args /* { 1153 struct vnode *a_vp; 1154 struct buf *a_bp; 1155 } */ *ap = v; 1156 struct buf *bp; 1157 struct lfs *fs; 1158 struct vnode *vp; 1159 struct inode *ip; 1160 daddr_t tbn; 1161 int i, sn, error, slept; 1162 1163 bp = ap->a_bp; 1164 vp = ap->a_vp; 1165 ip = VTOI(vp); 1166 fs = ip->i_lfs; 1167 1168 /* lfs uses its strategy routine only for read */ 1169 KASSERT(bp->b_flags & B_READ); 1170 1171 if (vp->v_type == VBLK || vp->v_type == VCHR) 1172 panic("lfs_strategy: spec"); 1173 KASSERT(bp->b_bcount != 0); 1174 if (bp->b_blkno == bp->b_lblkno) { 1175 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1176 NULL); 1177 if (error) { 1178 bp->b_error = error; 1179 biodone(bp); 1180 return (error); 1181 } 1182 if ((long)bp->b_blkno == -1) /* no valid data */ 1183 clrbuf(bp); 1184 } 1185 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1186 biodone(bp); 1187 return (0); 1188 } 1189 1190 slept = 1; 1191 simple_lock(&fs->lfs_interlock); 1192 while (slept && fs->lfs_seglock) { 1193 simple_unlock(&fs->lfs_interlock); 1194 /* 1195 * Look through list of intervals. 1196 * There will only be intervals to look through 1197 * if the cleaner holds the seglock. 1198 * Since the cleaner is synchronous, we can trust 1199 * the list of intervals to be current. 1200 */ 1201 tbn = dbtofsb(fs, bp->b_blkno); 1202 sn = dtosn(fs, tbn); 1203 slept = 0; 1204 for (i = 0; i < fs->lfs_cleanind; i++) { 1205 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1206 tbn >= fs->lfs_cleanint[i]) { 1207 DLOG((DLOG_CLEAN, 1208 "lfs_strategy: ino %d lbn %" PRId64 1209 " ind %d sn %d fsb %" PRIx32 1210 " given sn %d fsb %" PRIx64 "\n", 1211 ip->i_number, bp->b_lblkno, i, 1212 dtosn(fs, fs->lfs_cleanint[i]), 1213 fs->lfs_cleanint[i], sn, tbn)); 1214 DLOG((DLOG_CLEAN, 1215 "lfs_strategy: sleeping on ino %d lbn %" 1216 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1217 simple_lock(&fs->lfs_interlock); 1218 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1219 /* Cleaner can't wait for itself */ 1220 ltsleep(&fs->lfs_iocount, 1221 (PRIBIO + 1) | PNORELOCK, 1222 "clean2", 0, 1223 &fs->lfs_interlock); 1224 slept = 1; 1225 break; 1226 } else if (fs->lfs_seglock) { 1227 ltsleep(&fs->lfs_seglock, 1228 (PRIBIO + 1) | PNORELOCK, 1229 "clean1", 0, 1230 &fs->lfs_interlock); 1231 slept = 1; 1232 break; 1233 } 1234 simple_unlock(&fs->lfs_interlock); 1235 } 1236 } 1237 simple_lock(&fs->lfs_interlock); 1238 } 1239 simple_unlock(&fs->lfs_interlock); 1240 1241 vp = ip->i_devvp; 1242 VOP_STRATEGY(vp, bp); 1243 return (0); 1244 } 1245 1246 void 1247 lfs_flush_dirops(struct lfs *fs) 1248 { 1249 struct inode *ip, *nip; 1250 struct vnode *vp; 1251 extern int lfs_dostats; 1252 struct segment *sp; 1253 int waslocked; 1254 1255 ASSERT_MAYBE_SEGLOCK(fs); 1256 KASSERT(fs->lfs_nadirop == 0); 1257 1258 if (fs->lfs_ronly) 1259 return; 1260 1261 simple_lock(&fs->lfs_interlock); 1262 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1263 simple_unlock(&fs->lfs_interlock); 1264 return; 1265 } else 1266 simple_unlock(&fs->lfs_interlock); 1267 1268 if (lfs_dostats) 1269 ++lfs_stats.flush_invoked; 1270 1271 /* 1272 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1273 * Technically this is a checkpoint (the on-disk state is valid) 1274 * even though we are leaving out all the file data. 1275 */ 1276 lfs_imtime(fs); 1277 lfs_seglock(fs, SEGM_CKP); 1278 sp = fs->lfs_sp; 1279 1280 /* 1281 * lfs_writevnodes, optimized to get dirops out of the way. 1282 * Only write dirops, and don't flush files' pages, only 1283 * blocks from the directories. 1284 * 1285 * We don't need to vref these files because they are 1286 * dirops and so hold an extra reference until the 1287 * segunlock clears them of that status. 1288 * 1289 * We don't need to check for IN_ADIROP because we know that 1290 * no dirops are active. 1291 * 1292 */ 1293 simple_lock(&fs->lfs_interlock); 1294 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1295 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1296 simple_unlock(&fs->lfs_interlock); 1297 vp = ITOV(ip); 1298 1299 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1300 1301 /* 1302 * All writes to directories come from dirops; all 1303 * writes to files' direct blocks go through the page 1304 * cache, which we're not touching. Reads to files 1305 * and/or directories will not be affected by writing 1306 * directory blocks inodes and file inodes. So we don't 1307 * really need to lock. If we don't lock, though, 1308 * make sure that we don't clear IN_MODIFIED 1309 * unnecessarily. 1310 */ 1311 if (vp->v_flag & (VXLOCK | VFREEING)) { 1312 simple_lock(&fs->lfs_interlock); 1313 continue; 1314 } 1315 waslocked = VOP_ISLOCKED(vp); 1316 if (vp->v_type != VREG && 1317 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1318 lfs_writefile(fs, sp, vp); 1319 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1320 !(ip->i_flag & IN_ALLMOD)) { 1321 LFS_SET_UINO(ip, IN_MODIFIED); 1322 } 1323 } 1324 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1325 (void) lfs_writeinode(fs, sp, ip); 1326 if (waslocked == LK_EXCLOTHER) 1327 LFS_SET_UINO(ip, IN_MODIFIED); 1328 simple_lock(&fs->lfs_interlock); 1329 } 1330 simple_unlock(&fs->lfs_interlock); 1331 /* We've written all the dirops there are */ 1332 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1333 lfs_finalize_fs_seguse(fs); 1334 (void) lfs_writeseg(fs, sp); 1335 lfs_segunlock(fs); 1336 } 1337 1338 /* 1339 * Flush all vnodes for which the pagedaemon has requested pageouts. 1340 * Skip over any files that are marked VDIROP (since lfs_flush_dirop() 1341 * has just run, this would be an error). If we have to skip a vnode 1342 * for any reason, just skip it; if we have to wait for the cleaner, 1343 * abort. The writer daemon will call us again later. 1344 */ 1345 void 1346 lfs_flush_pchain(struct lfs *fs) 1347 { 1348 struct inode *ip, *nip; 1349 struct vnode *vp; 1350 extern int lfs_dostats; 1351 struct segment *sp; 1352 int error; 1353 1354 ASSERT_NO_SEGLOCK(fs); 1355 1356 if (fs->lfs_ronly) 1357 return; 1358 1359 simple_lock(&fs->lfs_interlock); 1360 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1361 simple_unlock(&fs->lfs_interlock); 1362 return; 1363 } else 1364 simple_unlock(&fs->lfs_interlock); 1365 1366 /* Get dirops out of the way */ 1367 lfs_flush_dirops(fs); 1368 1369 if (lfs_dostats) 1370 ++lfs_stats.flush_invoked; 1371 1372 /* 1373 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1374 */ 1375 lfs_imtime(fs); 1376 lfs_seglock(fs, 0); 1377 sp = fs->lfs_sp; 1378 1379 /* 1380 * lfs_writevnodes, optimized to clear pageout requests. 1381 * Only write non-dirop files that are in the pageout queue. 1382 * We're very conservative about what we write; we want to be 1383 * fast and async. 1384 */ 1385 simple_lock(&fs->lfs_interlock); 1386 top: 1387 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1388 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1389 vp = ITOV(ip); 1390 1391 if (!(ip->i_flags & IN_PAGING)) 1392 goto top; 1393 1394 if (vp->v_flag & (VXLOCK|VDIROP)) 1395 continue; 1396 if (vp->v_type != VREG) 1397 continue; 1398 if (lfs_vref(vp)) 1399 continue; 1400 simple_unlock(&fs->lfs_interlock); 1401 1402 if (VOP_ISLOCKED(vp)) { 1403 lfs_vunref(vp); 1404 simple_lock(&fs->lfs_interlock); 1405 continue; 1406 } 1407 1408 error = lfs_writefile(fs, sp, vp); 1409 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1410 !(ip->i_flag & IN_ALLMOD)) { 1411 LFS_SET_UINO(ip, IN_MODIFIED); 1412 } 1413 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1414 (void) lfs_writeinode(fs, sp, ip); 1415 1416 lfs_vunref(vp); 1417 1418 if (error == EAGAIN) { 1419 lfs_writeseg(fs, sp); 1420 simple_lock(&fs->lfs_interlock); 1421 break; 1422 } 1423 simple_lock(&fs->lfs_interlock); 1424 } 1425 simple_unlock(&fs->lfs_interlock); 1426 (void) lfs_writeseg(fs, sp); 1427 lfs_segunlock(fs); 1428 } 1429 1430 /* 1431 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1432 */ 1433 int 1434 lfs_fcntl(void *v) 1435 { 1436 struct vop_fcntl_args /* { 1437 struct vnode *a_vp; 1438 u_long a_command; 1439 void * a_data; 1440 int a_fflag; 1441 kauth_cred_t a_cred; 1442 struct lwp *a_l; 1443 } */ *ap = v; 1444 struct timeval *tvp; 1445 BLOCK_INFO *blkiov; 1446 CLEANERINFO *cip; 1447 SEGUSE *sup; 1448 int blkcnt, error, oclean; 1449 size_t fh_size; 1450 struct lfs_fcntl_markv blkvp; 1451 struct lwp *l; 1452 fsid_t *fsidp; 1453 struct lfs *fs; 1454 struct buf *bp; 1455 fhandle_t *fhp; 1456 daddr_t off; 1457 1458 /* Only respect LFS fcntls on fs root or Ifile */ 1459 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1460 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1461 return ufs_fcntl(v); 1462 } 1463 1464 /* Avoid locking a draining lock */ 1465 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1466 return ESHUTDOWN; 1467 } 1468 1469 /* LFS control and monitoring fcntls are available only to root */ 1470 l = ap->a_l; 1471 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1472 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1473 NULL)) != 0) 1474 return (error); 1475 1476 fs = VTOI(ap->a_vp)->i_lfs; 1477 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1478 1479 error = 0; 1480 switch (ap->a_command) { 1481 case LFCNSEGWAITALL: 1482 case LFCNSEGWAITALL_COMPAT: 1483 fsidp = NULL; 1484 /* FALLSTHROUGH */ 1485 case LFCNSEGWAIT: 1486 case LFCNSEGWAIT_COMPAT: 1487 tvp = (struct timeval *)ap->a_data; 1488 simple_lock(&fs->lfs_interlock); 1489 ++fs->lfs_sleepers; 1490 simple_unlock(&fs->lfs_interlock); 1491 1492 error = lfs_segwait(fsidp, tvp); 1493 1494 simple_lock(&fs->lfs_interlock); 1495 if (--fs->lfs_sleepers == 0) 1496 wakeup(&fs->lfs_sleepers); 1497 simple_unlock(&fs->lfs_interlock); 1498 return error; 1499 1500 case LFCNBMAPV: 1501 case LFCNMARKV: 1502 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1503 1504 blkcnt = blkvp.blkcnt; 1505 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1506 return (EINVAL); 1507 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1508 if ((error = copyin(blkvp.blkiov, blkiov, 1509 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1510 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1511 return error; 1512 } 1513 1514 simple_lock(&fs->lfs_interlock); 1515 ++fs->lfs_sleepers; 1516 simple_unlock(&fs->lfs_interlock); 1517 if (ap->a_command == LFCNBMAPV) 1518 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1519 else /* LFCNMARKV */ 1520 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1521 if (error == 0) 1522 error = copyout(blkiov, blkvp.blkiov, 1523 blkcnt * sizeof(BLOCK_INFO)); 1524 simple_lock(&fs->lfs_interlock); 1525 if (--fs->lfs_sleepers == 0) 1526 wakeup(&fs->lfs_sleepers); 1527 simple_unlock(&fs->lfs_interlock); 1528 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1529 return error; 1530 1531 case LFCNRECLAIM: 1532 /* 1533 * Flush dirops and write Ifile, allowing empty segments 1534 * to be immediately reclaimed. 1535 */ 1536 lfs_writer_enter(fs, "pndirop"); 1537 off = fs->lfs_offset; 1538 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1539 lfs_flush_dirops(fs); 1540 LFS_CLEANERINFO(cip, fs, bp); 1541 oclean = cip->clean; 1542 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1543 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1544 fs->lfs_sp->seg_flags |= SEGM_PROT; 1545 lfs_segunlock(fs); 1546 lfs_writer_leave(fs); 1547 1548 #ifdef DEBUG 1549 LFS_CLEANERINFO(cip, fs, bp); 1550 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1551 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1552 fs->lfs_offset - off, cip->clean - oclean, 1553 fs->lfs_activesb)); 1554 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1555 #endif 1556 1557 return 0; 1558 1559 #ifdef COMPAT_30 1560 case LFCNIFILEFH_COMPAT: 1561 /* Return the filehandle of the Ifile */ 1562 if ((error = kauth_authorize_generic(l->l_cred, 1563 KAUTH_GENERIC_ISSUSER, NULL)) != 0) 1564 return (error); 1565 fhp = (struct fhandle *)ap->a_data; 1566 fhp->fh_fsid = *fsidp; 1567 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1568 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1569 #endif 1570 1571 case LFCNIFILEFH_COMPAT2: 1572 case LFCNIFILEFH: 1573 /* Return the filehandle of the Ifile */ 1574 fhp = (struct fhandle *)ap->a_data; 1575 fhp->fh_fsid = *fsidp; 1576 fh_size = sizeof(struct lfs_fhandle) - 1577 offsetof(fhandle_t, fh_fid); 1578 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1579 1580 case LFCNREWIND: 1581 /* Move lfs_offset to the lowest-numbered segment */ 1582 return lfs_rewind(fs, *(int *)ap->a_data); 1583 1584 case LFCNINVAL: 1585 /* Mark a segment SEGUSE_INVAL */ 1586 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1587 if (sup->su_nbytes > 0) { 1588 brelse(bp, 0); 1589 lfs_unset_inval_all(fs); 1590 return EBUSY; 1591 } 1592 sup->su_flags |= SEGUSE_INVAL; 1593 VOP_BWRITE(bp); 1594 return 0; 1595 1596 case LFCNRESIZE: 1597 /* Resize the filesystem */ 1598 return lfs_resize_fs(fs, *(int *)ap->a_data); 1599 1600 case LFCNWRAPSTOP: 1601 case LFCNWRAPSTOP_COMPAT: 1602 /* 1603 * Hold lfs_newseg at segment 0; if requested, sleep until 1604 * the filesystem wraps around. To support external agents 1605 * (dump, fsck-based regression test) that need to look at 1606 * a snapshot of the filesystem, without necessarily 1607 * requiring that all fs activity stops. 1608 */ 1609 if (lockstatus(&fs->lfs_stoplock)) 1610 return EALREADY; 1611 1612 simple_lock(&fs->lfs_interlock); 1613 lockmgr(&fs->lfs_stoplock, LK_EXCLUSIVE, &fs->lfs_interlock); 1614 if (fs->lfs_nowrap == 0) 1615 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1616 ++fs->lfs_nowrap; 1617 if (*(int *)ap->a_data == 1 || 1618 ap->a_command == LFCNWRAPSTOP_COMPAT) { 1619 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1620 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1621 "segwrap", 0, &fs->lfs_interlock); 1622 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1623 if (error) { 1624 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1625 } 1626 } 1627 simple_unlock(&fs->lfs_interlock); 1628 return 0; 1629 1630 case LFCNWRAPGO: 1631 case LFCNWRAPGO_COMPAT: 1632 /* 1633 * Having done its work, the agent wakes up the writer. 1634 * If the argument is 1, it sleeps until a new segment 1635 * is selected. 1636 */ 1637 simple_lock(&fs->lfs_interlock); 1638 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1639 (ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1640 *((int *)ap->a_data))); 1641 simple_unlock(&fs->lfs_interlock); 1642 return error; 1643 1644 case LFCNWRAPPASS: 1645 if (lockstatus(&fs->lfs_stoplock) != LK_EXCLUSIVE) 1646 return EALREADY; 1647 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1648 return EALREADY; 1649 simple_lock(&fs->lfs_interlock); 1650 if (fs->lfs_nowrap == 0) { 1651 simple_unlock(&fs->lfs_interlock); 1652 return EBUSY; 1653 } 1654 fs->lfs_wrappass = 1; 1655 wakeup(&fs->lfs_wrappass); 1656 /* Wait for the log to wrap, if asked */ 1657 if (*(int *)ap->a_data) { 1658 lfs_vref(ap->a_vp); 1659 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1660 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1661 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1662 "segwrap", 0, &fs->lfs_interlock); 1663 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1664 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1665 lfs_vunref(ap->a_vp); 1666 } 1667 simple_unlock(&fs->lfs_interlock); 1668 return error; 1669 1670 case LFCNWRAPSTATUS: 1671 simple_lock(&fs->lfs_interlock); 1672 *(int *)ap->a_data = fs->lfs_wrapstatus; 1673 simple_unlock(&fs->lfs_interlock); 1674 return 0; 1675 1676 default: 1677 return ufs_fcntl(v); 1678 } 1679 return 0; 1680 } 1681 1682 int 1683 lfs_getpages(void *v) 1684 { 1685 struct vop_getpages_args /* { 1686 struct vnode *a_vp; 1687 voff_t a_offset; 1688 struct vm_page **a_m; 1689 int *a_count; 1690 int a_centeridx; 1691 vm_prot_t a_access_type; 1692 int a_advice; 1693 int a_flags; 1694 } */ *ap = v; 1695 1696 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1697 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1698 return EPERM; 1699 } 1700 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1701 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1702 } 1703 1704 /* 1705 * we're relying on the fact that genfs_getpages() always read in 1706 * entire filesystem blocks. 1707 */ 1708 return genfs_getpages(v); 1709 } 1710 1711 /* 1712 * Wait for a page to become unbusy, possibly printing diagnostic messages 1713 * as well. 1714 * 1715 * Called with vp->v_interlock held; return with it held. 1716 */ 1717 static void 1718 wait_for_page(struct vnode *vp, struct vm_page *pg, const char *label) 1719 { 1720 if ((pg->flags & PG_BUSY) == 0) 1721 return; /* Nothing to wait for! */ 1722 1723 #if defined(DEBUG) && defined(UVM_PAGE_TRKOWN) 1724 static struct vm_page *lastpg; 1725 1726 if (label != NULL && pg != lastpg) { 1727 if (pg->owner_tag) { 1728 printf("lfs_putpages[%d.%d]: %s: page %p owner %d.%d [%s]\n", 1729 curproc->p_pid, curlwp->l_lid, label, 1730 pg, pg->owner, pg->lowner, pg->owner_tag); 1731 } else { 1732 printf("lfs_putpages[%d.%d]: %s: page %p unowned?!\n", 1733 curproc->p_pid, curlwp->l_lid, label, pg); 1734 } 1735 } 1736 lastpg = pg; 1737 #endif 1738 1739 pg->flags |= PG_WANTED; 1740 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, "lfsput", 0); 1741 simple_lock(&vp->v_interlock); 1742 } 1743 1744 /* 1745 * This routine is called by lfs_putpages() when it can't complete the 1746 * write because a page is busy. This means that either (1) someone, 1747 * possibly the pagedaemon, is looking at this page, and will give it up 1748 * presently; or (2) we ourselves are holding the page busy in the 1749 * process of being written (either gathered or actually on its way to 1750 * disk). We don't need to give up the segment lock, but we might need 1751 * to call lfs_writeseg() to expedite the page's journey to disk. 1752 * 1753 * Called with vp->v_interlock held; return with it held. 1754 */ 1755 /* #define BUSYWAIT */ 1756 static void 1757 write_and_wait(struct lfs *fs, struct vnode *vp, struct vm_page *pg, 1758 int seglocked, const char *label) 1759 { 1760 #ifndef BUSYWAIT 1761 struct inode *ip = VTOI(vp); 1762 struct segment *sp = fs->lfs_sp; 1763 int count = 0; 1764 1765 if (pg == NULL) 1766 return; 1767 1768 while (pg->flags & PG_BUSY) { 1769 simple_unlock(&vp->v_interlock); 1770 if (sp->cbpp - sp->bpp > 1) { 1771 /* Write gathered pages */ 1772 lfs_updatemeta(sp); 1773 lfs_release_finfo(fs); 1774 (void) lfs_writeseg(fs, sp); 1775 1776 /* 1777 * Reinitialize FIP 1778 */ 1779 KASSERT(sp->vp == vp); 1780 lfs_acquire_finfo(fs, ip->i_number, 1781 ip->i_gen); 1782 } 1783 ++count; 1784 simple_lock(&vp->v_interlock); 1785 wait_for_page(vp, pg, label); 1786 } 1787 if (label != NULL && count > 1) 1788 printf("lfs_putpages[%d]: %s: %sn = %d\n", curproc->p_pid, 1789 label, (count > 0 ? "looping, " : ""), count); 1790 #else 1791 preempt(1); 1792 #endif 1793 } 1794 1795 /* 1796 * Make sure that for all pages in every block in the given range, 1797 * either all are dirty or all are clean. If any of the pages 1798 * we've seen so far are dirty, put the vnode on the paging chain, 1799 * and mark it IN_PAGING. 1800 * 1801 * If checkfirst != 0, don't check all the pages but return at the 1802 * first dirty page. 1803 */ 1804 static int 1805 check_dirty(struct lfs *fs, struct vnode *vp, 1806 off_t startoffset, off_t endoffset, off_t blkeof, 1807 int flags, int checkfirst, struct vm_page **pgp) 1808 { 1809 int by_list; 1810 struct vm_page *curpg = NULL; /* XXX: gcc */ 1811 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1812 off_t soff = 0; /* XXX: gcc */ 1813 voff_t off; 1814 int i; 1815 int nonexistent; 1816 int any_dirty; /* number of dirty pages */ 1817 int dirty; /* number of dirty pages in a block */ 1818 int tdirty; 1819 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1820 int pagedaemon = (curlwp == uvm.pagedaemon_lwp); 1821 1822 ASSERT_MAYBE_SEGLOCK(fs); 1823 top: 1824 by_list = (vp->v_uobj.uo_npages <= 1825 ((endoffset - startoffset) >> PAGE_SHIFT) * 1826 UVM_PAGE_HASH_PENALTY); 1827 any_dirty = 0; 1828 1829 if (by_list) { 1830 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1831 } else { 1832 soff = startoffset; 1833 } 1834 while (by_list || soff < MIN(blkeof, endoffset)) { 1835 if (by_list) { 1836 /* 1837 * Find the first page in a block. Skip 1838 * blocks outside our area of interest or beyond 1839 * the end of file. 1840 */ 1841 if (pages_per_block > 1) { 1842 while (curpg && 1843 ((curpg->offset & fs->lfs_bmask) || 1844 curpg->offset >= vp->v_size || 1845 curpg->offset >= endoffset)) 1846 curpg = TAILQ_NEXT(curpg, listq); 1847 } 1848 if (curpg == NULL) 1849 break; 1850 soff = curpg->offset; 1851 } 1852 1853 /* 1854 * Mark all pages in extended range busy; find out if any 1855 * of them are dirty. 1856 */ 1857 nonexistent = dirty = 0; 1858 for (i = 0; i == 0 || i < pages_per_block; i++) { 1859 if (by_list && pages_per_block <= 1) { 1860 pgs[i] = pg = curpg; 1861 } else { 1862 off = soff + (i << PAGE_SHIFT); 1863 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1864 if (pg == NULL) { 1865 ++nonexistent; 1866 continue; 1867 } 1868 } 1869 KASSERT(pg != NULL); 1870 1871 /* 1872 * If we're holding the segment lock, we can deadlock 1873 * against a process that has our page and is waiting 1874 * for the cleaner, while the cleaner waits for the 1875 * segment lock. Just bail in that case. 1876 */ 1877 if ((pg->flags & PG_BUSY) && 1878 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1879 if (i > 0) 1880 uvm_page_unbusy(pgs, i); 1881 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1882 if (pgp) 1883 *pgp = pg; 1884 return -1; 1885 } 1886 1887 while (pg->flags & PG_BUSY) { 1888 wait_for_page(vp, pg, NULL); 1889 if (i > 0) 1890 uvm_page_unbusy(pgs, i); 1891 goto top; 1892 } 1893 pg->flags |= PG_BUSY; 1894 UVM_PAGE_OWN(pg, "lfs_putpages"); 1895 1896 pmap_page_protect(pg, VM_PROT_NONE); 1897 tdirty = (pmap_clear_modify(pg) || 1898 (pg->flags & PG_CLEAN) == 0); 1899 dirty += tdirty; 1900 } 1901 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1902 if (by_list) { 1903 curpg = TAILQ_NEXT(curpg, listq); 1904 } else { 1905 soff += fs->lfs_bsize; 1906 } 1907 continue; 1908 } 1909 1910 any_dirty += dirty; 1911 KASSERT(nonexistent == 0); 1912 1913 /* 1914 * If any are dirty make all dirty; unbusy them, 1915 * but if we were asked to clean, wire them so that 1916 * the pagedaemon doesn't bother us about them while 1917 * they're on their way to disk. 1918 */ 1919 for (i = 0; i == 0 || i < pages_per_block; i++) { 1920 pg = pgs[i]; 1921 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1922 if (dirty) { 1923 pg->flags &= ~PG_CLEAN; 1924 if (flags & PGO_FREE) { 1925 /* 1926 * Wire the page so that 1927 * pdaemon doesn't see it again. 1928 */ 1929 uvm_lock_pageq(); 1930 uvm_pagewire(pg); 1931 uvm_unlock_pageq(); 1932 1933 /* Suspended write flag */ 1934 pg->flags |= PG_DELWRI; 1935 } 1936 } 1937 if (pg->flags & PG_WANTED) 1938 wakeup(pg); 1939 pg->flags &= ~(PG_WANTED|PG_BUSY); 1940 UVM_PAGE_OWN(pg, NULL); 1941 } 1942 1943 if (checkfirst && any_dirty) 1944 break; 1945 1946 if (by_list) { 1947 curpg = TAILQ_NEXT(curpg, listq); 1948 } else { 1949 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1950 } 1951 } 1952 1953 return any_dirty; 1954 } 1955 1956 /* 1957 * lfs_putpages functions like genfs_putpages except that 1958 * 1959 * (1) It needs to bounds-check the incoming requests to ensure that 1960 * they are block-aligned; if they are not, expand the range and 1961 * do the right thing in case, e.g., the requested range is clean 1962 * but the expanded range is dirty. 1963 * 1964 * (2) It needs to explicitly send blocks to be written when it is done. 1965 * If VOP_PUTPAGES is called without the seglock held, we simply take 1966 * the seglock and let lfs_segunlock wait for us. 1967 * XXX There might be a bad situation if we have to flush a vnode while 1968 * XXX lfs_markv is in operation. As of this writing we panic in this 1969 * XXX case. 1970 * 1971 * Assumptions: 1972 * 1973 * (1) The caller does not hold any pages in this vnode busy. If it does, 1974 * there is a danger that when we expand the page range and busy the 1975 * pages we will deadlock. 1976 * 1977 * (2) We are called with vp->v_interlock held; we must return with it 1978 * released. 1979 * 1980 * (3) We don't absolutely have to free pages right away, provided that 1981 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1982 * us a request with PGO_FREE, we take the pages out of the paging 1983 * queue and wake up the writer, which will handle freeing them for us. 1984 * 1985 * We ensure that for any filesystem block, all pages for that 1986 * block are either resident or not, even if those pages are higher 1987 * than EOF; that means that we will be getting requests to free 1988 * "unused" pages above EOF all the time, and should ignore them. 1989 * 1990 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 1991 * into has been set up for us by lfs_writefile. If not, we will 1992 * have to handle allocating and/or freeing an finfo entry. 1993 * 1994 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1995 */ 1996 1997 /* How many times to loop before we should start to worry */ 1998 #define TOOMANY 4 1999 2000 int 2001 lfs_putpages(void *v) 2002 { 2003 int error; 2004 struct vop_putpages_args /* { 2005 struct vnode *a_vp; 2006 voff_t a_offlo; 2007 voff_t a_offhi; 2008 int a_flags; 2009 } */ *ap = v; 2010 struct vnode *vp; 2011 struct inode *ip; 2012 struct lfs *fs; 2013 struct segment *sp; 2014 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 2015 off_t off, max_endoffset; 2016 int s; 2017 bool seglocked, sync, pagedaemon; 2018 struct vm_page *pg, *busypg; 2019 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 2020 #ifdef DEBUG 2021 int debug_n_again, debug_n_dirtyclean; 2022 #endif 2023 2024 vp = ap->a_vp; 2025 ip = VTOI(vp); 2026 fs = ip->i_lfs; 2027 sync = (ap->a_flags & PGO_SYNCIO) != 0; 2028 pagedaemon = (curlwp == uvm.pagedaemon_lwp); 2029 2030 /* Putpages does nothing for metadata. */ 2031 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 2032 simple_unlock(&vp->v_interlock); 2033 return 0; 2034 } 2035 2036 /* 2037 * If there are no pages, don't do anything. 2038 */ 2039 if (vp->v_uobj.uo_npages == 0) { 2040 s = splbio(); 2041 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 2042 (vp->v_flag & VONWORKLST) && 2043 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 2044 vp->v_flag &= ~VWRITEMAPDIRTY; 2045 vn_syncer_remove_from_worklist(vp); 2046 } 2047 splx(s); 2048 simple_unlock(&vp->v_interlock); 2049 2050 /* Remove us from paging queue, if we were on it */ 2051 simple_lock(&fs->lfs_interlock); 2052 if (ip->i_flags & IN_PAGING) { 2053 ip->i_flags &= ~IN_PAGING; 2054 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2055 } 2056 simple_unlock(&fs->lfs_interlock); 2057 return 0; 2058 } 2059 2060 blkeof = blkroundup(fs, ip->i_size); 2061 2062 /* 2063 * Ignore requests to free pages past EOF but in the same block 2064 * as EOF, unless the request is synchronous. (If the request is 2065 * sync, it comes from lfs_truncate.) 2066 * XXXUBC Make these pages look "active" so the pagedaemon won't 2067 * XXXUBC bother us with them again. 2068 */ 2069 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 2070 origoffset = ap->a_offlo; 2071 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 2072 pg = uvm_pagelookup(&vp->v_uobj, off); 2073 KASSERT(pg != NULL); 2074 while (pg->flags & PG_BUSY) { 2075 pg->flags |= PG_WANTED; 2076 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 2077 "lfsput2", 0); 2078 simple_lock(&vp->v_interlock); 2079 } 2080 uvm_lock_pageq(); 2081 uvm_pageactivate(pg); 2082 uvm_unlock_pageq(); 2083 } 2084 ap->a_offlo = blkeof; 2085 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 2086 simple_unlock(&vp->v_interlock); 2087 return 0; 2088 } 2089 } 2090 2091 /* 2092 * Extend page range to start and end at block boundaries. 2093 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 2094 */ 2095 origoffset = ap->a_offlo; 2096 origendoffset = ap->a_offhi; 2097 startoffset = origoffset & ~(fs->lfs_bmask); 2098 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 2099 << fs->lfs_bshift; 2100 2101 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2102 endoffset = max_endoffset; 2103 origendoffset = endoffset; 2104 } else { 2105 origendoffset = round_page(ap->a_offhi); 2106 endoffset = round_page(blkroundup(fs, origendoffset)); 2107 } 2108 2109 KASSERT(startoffset > 0 || endoffset >= startoffset); 2110 if (startoffset == endoffset) { 2111 /* Nothing to do, why were we called? */ 2112 simple_unlock(&vp->v_interlock); 2113 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 2114 PRId64 "\n", startoffset)); 2115 return 0; 2116 } 2117 2118 ap->a_offlo = startoffset; 2119 ap->a_offhi = endoffset; 2120 2121 /* 2122 * If not cleaning, just send the pages through genfs_putpages 2123 * to be returned to the pool. 2124 */ 2125 if (!(ap->a_flags & PGO_CLEANIT)) 2126 return genfs_putpages(v); 2127 2128 /* Set PGO_BUSYFAIL to avoid deadlocks */ 2129 ap->a_flags |= PGO_BUSYFAIL; 2130 2131 /* 2132 * Likewise, if we are asked to clean but the pages are not 2133 * dirty, we can just free them using genfs_putpages. 2134 */ 2135 #ifdef DEBUG 2136 debug_n_dirtyclean = 0; 2137 #endif 2138 do { 2139 int r; 2140 2141 /* Count the number of dirty pages */ 2142 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2143 ap->a_flags, 1, NULL); 2144 if (r < 0) { 2145 /* Pages are busy with another process */ 2146 simple_unlock(&vp->v_interlock); 2147 return EDEADLK; 2148 } 2149 if (r > 0) /* Some pages are dirty */ 2150 break; 2151 2152 /* 2153 * Sometimes pages are dirtied between the time that 2154 * we check and the time we try to clean them. 2155 * Instruct lfs_gop_write to return EDEADLK in this case 2156 * so we can write them properly. 2157 */ 2158 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2159 r = genfs_do_putpages(vp, startoffset, endoffset, 2160 ap->a_flags, &busypg); 2161 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2162 if (r != EDEADLK) 2163 return r; 2164 2165 /* One of the pages was busy. Start over. */ 2166 simple_lock(&vp->v_interlock); 2167 wait_for_page(vp, busypg, "dirtyclean"); 2168 #ifdef DEBUG 2169 ++debug_n_dirtyclean; 2170 #endif 2171 } while(1); 2172 2173 #ifdef DEBUG 2174 if (debug_n_dirtyclean > TOOMANY) 2175 printf("lfs_putpages: dirtyclean: looping, n = %d\n", 2176 debug_n_dirtyclean); 2177 #endif 2178 2179 /* 2180 * Dirty and asked to clean. 2181 * 2182 * Pagedaemon can't actually write LFS pages; wake up 2183 * the writer to take care of that. The writer will 2184 * notice the pager inode queue and act on that. 2185 */ 2186 if (pagedaemon) { 2187 simple_lock(&fs->lfs_interlock); 2188 if (!(ip->i_flags & IN_PAGING)) { 2189 ip->i_flags |= IN_PAGING; 2190 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2191 } 2192 simple_lock(&lfs_subsys_lock); 2193 wakeup(&lfs_writer_daemon); 2194 simple_unlock(&lfs_subsys_lock); 2195 simple_unlock(&fs->lfs_interlock); 2196 simple_unlock(&vp->v_interlock); 2197 preempt(); 2198 return EWOULDBLOCK; 2199 } 2200 2201 /* 2202 * If this is a file created in a recent dirop, we can't flush its 2203 * inode until the dirop is complete. Drain dirops, then flush the 2204 * filesystem (taking care of any other pending dirops while we're 2205 * at it). 2206 */ 2207 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2208 (vp->v_flag & VDIROP)) { 2209 int locked; 2210 2211 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 2212 locked = (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 2213 simple_unlock(&vp->v_interlock); 2214 lfs_writer_enter(fs, "ppdirop"); 2215 if (locked) 2216 VOP_UNLOCK(vp, 0); /* XXX why? */ 2217 2218 simple_lock(&fs->lfs_interlock); 2219 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2220 simple_unlock(&fs->lfs_interlock); 2221 2222 simple_lock(&vp->v_interlock); 2223 if (locked) { 2224 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2225 simple_lock(&vp->v_interlock); 2226 } 2227 lfs_writer_leave(fs); 2228 2229 /* XXX the flush should have taken care of this one too! */ 2230 } 2231 2232 /* 2233 * This is it. We are going to write some pages. From here on 2234 * down it's all just mechanics. 2235 * 2236 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2237 */ 2238 ap->a_flags &= ~PGO_SYNCIO; 2239 2240 /* 2241 * If we've already got the seglock, flush the node and return. 2242 * The FIP has already been set up for us by lfs_writefile, 2243 * and FIP cleanup and lfs_updatemeta will also be done there, 2244 * unless genfs_putpages returns EDEADLK; then we must flush 2245 * what we have, and correct FIP and segment header accounting. 2246 */ 2247 get_seglock: 2248 /* 2249 * If we are not called with the segment locked, lock it. 2250 * Account for a new FIP in the segment header, and set sp->vp. 2251 * (This should duplicate the setup at the top of lfs_writefile().) 2252 */ 2253 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2254 if (!seglocked) { 2255 simple_unlock(&vp->v_interlock); 2256 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2257 if (error != 0) 2258 return error; 2259 simple_lock(&vp->v_interlock); 2260 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2261 } 2262 sp = fs->lfs_sp; 2263 KASSERT(sp->vp == NULL); 2264 sp->vp = vp; 2265 2266 /* 2267 * Ensure that the partial segment is marked SS_DIROP if this 2268 * vnode is a DIROP. 2269 */ 2270 if (!seglocked && vp->v_flag & VDIROP) 2271 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2272 2273 /* 2274 * Loop over genfs_putpages until all pages are gathered. 2275 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2276 * Whenever we lose the interlock we have to rerun check_dirty, as 2277 * well, since more pages might have been dirtied in our absence. 2278 */ 2279 #ifdef DEBUG 2280 debug_n_again = 0; 2281 #endif 2282 do { 2283 busypg = NULL; 2284 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2285 ap->a_flags, 0, &busypg) < 0) { 2286 simple_unlock(&vp->v_interlock); 2287 2288 simple_lock(&vp->v_interlock); 2289 write_and_wait(fs, vp, busypg, seglocked, NULL); 2290 if (!seglocked) { 2291 lfs_release_finfo(fs); 2292 lfs_segunlock(fs); 2293 } 2294 sp->vp = NULL; 2295 goto get_seglock; 2296 } 2297 2298 busypg = NULL; 2299 error = genfs_do_putpages(vp, startoffset, endoffset, 2300 ap->a_flags, &busypg); 2301 2302 if (error == EDEADLK || error == EAGAIN) { 2303 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2304 " %d ino %d off %x (seg %d)\n", error, 2305 ip->i_number, fs->lfs_offset, 2306 dtosn(fs, fs->lfs_offset))); 2307 2308 simple_lock(&vp->v_interlock); 2309 write_and_wait(fs, vp, busypg, seglocked, "again"); 2310 } 2311 #ifdef DEBUG 2312 ++debug_n_again; 2313 #endif 2314 } while (error == EDEADLK); 2315 #ifdef DEBUG 2316 if (debug_n_again > TOOMANY) 2317 printf("lfs_putpages: again: looping, n = %d\n", debug_n_again); 2318 #endif 2319 2320 KASSERT(sp != NULL && sp->vp == vp); 2321 if (!seglocked) { 2322 sp->vp = NULL; 2323 2324 /* Write indirect blocks as well */ 2325 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2326 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2327 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2328 2329 KASSERT(sp->vp == NULL); 2330 sp->vp = vp; 2331 } 2332 2333 /* 2334 * Blocks are now gathered into a segment waiting to be written. 2335 * All that's left to do is update metadata, and write them. 2336 */ 2337 lfs_updatemeta(sp); 2338 KASSERT(sp->vp == vp); 2339 sp->vp = NULL; 2340 2341 /* 2342 * If we were called from lfs_writefile, we don't need to clean up 2343 * the FIP or unlock the segment lock. We're done. 2344 */ 2345 if (seglocked) 2346 return error; 2347 2348 /* Clean up FIP and send it to disk. */ 2349 lfs_release_finfo(fs); 2350 lfs_writeseg(fs, fs->lfs_sp); 2351 2352 /* 2353 * Remove us from paging queue if we wrote all our pages. 2354 */ 2355 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2356 simple_lock(&fs->lfs_interlock); 2357 if (ip->i_flags & IN_PAGING) { 2358 ip->i_flags &= ~IN_PAGING; 2359 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2360 } 2361 simple_unlock(&fs->lfs_interlock); 2362 } 2363 2364 /* 2365 * XXX - with the malloc/copy writeseg, the pages are freed by now 2366 * even if we don't wait (e.g. if we hold a nested lock). This 2367 * will not be true if we stop using malloc/copy. 2368 */ 2369 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2370 lfs_segunlock(fs); 2371 2372 /* 2373 * Wait for v_numoutput to drop to zero. The seglock should 2374 * take care of this, but there is a slight possibility that 2375 * aiodoned might not have got around to our buffers yet. 2376 */ 2377 if (sync) { 2378 s = splbio(); 2379 simple_lock(&global_v_numoutput_slock); 2380 while (vp->v_numoutput > 0) { 2381 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2382 " num %d\n", ip->i_number, vp->v_numoutput)); 2383 vp->v_flag |= VBWAIT; 2384 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 2385 &global_v_numoutput_slock); 2386 } 2387 simple_unlock(&global_v_numoutput_slock); 2388 splx(s); 2389 } 2390 return error; 2391 } 2392 2393 /* 2394 * Return the last logical file offset that should be written for this file 2395 * if we're doing a write that ends at "size". If writing, we need to know 2396 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2397 * to know about entire blocks. 2398 */ 2399 void 2400 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2401 { 2402 struct inode *ip = VTOI(vp); 2403 struct lfs *fs = ip->i_lfs; 2404 daddr_t olbn, nlbn; 2405 2406 olbn = lblkno(fs, ip->i_size); 2407 nlbn = lblkno(fs, size); 2408 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2409 *eobp = fragroundup(fs, size); 2410 } else { 2411 *eobp = blkroundup(fs, size); 2412 } 2413 } 2414 2415 #ifdef DEBUG 2416 void lfs_dump_vop(void *); 2417 2418 void 2419 lfs_dump_vop(void *v) 2420 { 2421 struct vop_putpages_args /* { 2422 struct vnode *a_vp; 2423 voff_t a_offlo; 2424 voff_t a_offhi; 2425 int a_flags; 2426 } */ *ap = v; 2427 2428 #ifdef DDB 2429 vfs_vnode_print(ap->a_vp, 0, printf); 2430 #endif 2431 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2432 } 2433 #endif 2434 2435 int 2436 lfs_mmap(void *v) 2437 { 2438 struct vop_mmap_args /* { 2439 const struct vnodeop_desc *a_desc; 2440 struct vnode *a_vp; 2441 vm_prot_t a_prot; 2442 kauth_cred_t a_cred; 2443 struct lwp *a_l; 2444 } */ *ap = v; 2445 2446 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2447 return EOPNOTSUPP; 2448 return ufs_mmap(v); 2449 } 2450
Indexes created Tue Sep 30 20:09:53 GMT 2025