lfs_vnops.c revision 1.209
1 /* $NetBSD: lfs_vnops.c,v 1.209 2007/07/27 08:26:39 pooka Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Konrad E. Schroder <perseant (at) hhhh.org>. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 /* 39 * Copyright (c) 1986, 1989, 1991, 1993, 1995 40 * The Regents of the University of California. All rights reserved. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)lfs_vnops.c 8.13 (Berkeley) 6/10/95 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: lfs_vnops.c,v 1.209 2007/07/27 08:26:39 pooka Exp $"); 71 72 #ifdef _KERNEL_OPT 73 #include "opt_compat_netbsd.h" 74 #endif 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/namei.h> 79 #include <sys/resourcevar.h> 80 #include <sys/kernel.h> 81 #include <sys/file.h> 82 #include <sys/stat.h> 83 #include <sys/buf.h> 84 #include <sys/proc.h> 85 #include <sys/mount.h> 86 #include <sys/vnode.h> 87 #include <sys/pool.h> 88 #include <sys/signalvar.h> 89 #include <sys/kauth.h> 90 #include <sys/syslog.h> 91 #include <sys/fstrans.h> 92 93 #include <miscfs/fifofs/fifo.h> 94 #include <miscfs/genfs/genfs.h> 95 #include <miscfs/specfs/specdev.h> 96 97 #include <ufs/ufs/inode.h> 98 #include <ufs/ufs/dir.h> 99 #include <ufs/ufs/ufsmount.h> 100 #include <ufs/ufs/ufs_extern.h> 101 102 #include <uvm/uvm.h> 103 #include <uvm/uvm_pmap.h> 104 #include <uvm/uvm_stat.h> 105 #include <uvm/uvm_pager.h> 106 107 #include <ufs/lfs/lfs.h> 108 #include <ufs/lfs/lfs_extern.h> 109 110 extern pid_t lfs_writer_daemon; 111 int lfs_ignore_lazy_sync = 1; 112 113 /* Global vfs data structures for lfs. */ 114 int (**lfs_vnodeop_p)(void *); 115 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 116 { &vop_default_desc, vn_default_error }, 117 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 118 { &vop_create_desc, lfs_create }, /* create */ 119 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 120 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 121 { &vop_open_desc, ufs_open }, /* open */ 122 { &vop_close_desc, lfs_close }, /* close */ 123 { &vop_access_desc, ufs_access }, /* access */ 124 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 125 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 126 { &vop_read_desc, lfs_read }, /* read */ 127 { &vop_write_desc, lfs_write }, /* write */ 128 { &vop_lease_desc, ufs_lease_check }, /* lease */ 129 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 130 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 131 { &vop_poll_desc, ufs_poll }, /* poll */ 132 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 133 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 134 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 135 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 136 { &vop_seek_desc, ufs_seek }, /* seek */ 137 { &vop_remove_desc, lfs_remove }, /* remove */ 138 { &vop_link_desc, lfs_link }, /* link */ 139 { &vop_rename_desc, lfs_rename }, /* rename */ 140 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 141 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 142 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 143 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 144 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 145 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 146 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 147 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 148 { &vop_lock_desc, ufs_lock }, /* lock */ 149 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 150 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 151 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 152 { &vop_print_desc, ufs_print }, /* print */ 153 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 154 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 155 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 156 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 157 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 158 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 159 { NULL, NULL } 160 }; 161 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 162 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 163 164 int (**lfs_specop_p)(void *); 165 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 166 { &vop_default_desc, vn_default_error }, 167 { &vop_lookup_desc, spec_lookup }, /* lookup */ 168 { &vop_create_desc, spec_create }, /* create */ 169 { &vop_mknod_desc, spec_mknod }, /* mknod */ 170 { &vop_open_desc, spec_open }, /* open */ 171 { &vop_close_desc, lfsspec_close }, /* close */ 172 { &vop_access_desc, ufs_access }, /* access */ 173 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 174 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 175 { &vop_read_desc, ufsspec_read }, /* read */ 176 { &vop_write_desc, ufsspec_write }, /* write */ 177 { &vop_lease_desc, spec_lease_check }, /* lease */ 178 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 179 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 180 { &vop_poll_desc, spec_poll }, /* poll */ 181 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 182 { &vop_revoke_desc, spec_revoke }, /* revoke */ 183 { &vop_mmap_desc, spec_mmap }, /* mmap */ 184 { &vop_fsync_desc, spec_fsync }, /* fsync */ 185 { &vop_seek_desc, spec_seek }, /* seek */ 186 { &vop_remove_desc, spec_remove }, /* remove */ 187 { &vop_link_desc, spec_link }, /* link */ 188 { &vop_rename_desc, spec_rename }, /* rename */ 189 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 190 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 191 { &vop_symlink_desc, spec_symlink }, /* symlink */ 192 { &vop_readdir_desc, spec_readdir }, /* readdir */ 193 { &vop_readlink_desc, spec_readlink }, /* readlink */ 194 { &vop_abortop_desc, spec_abortop }, /* abortop */ 195 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 196 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 197 { &vop_lock_desc, ufs_lock }, /* lock */ 198 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 199 { &vop_bmap_desc, spec_bmap }, /* bmap */ 200 { &vop_strategy_desc, spec_strategy }, /* strategy */ 201 { &vop_print_desc, ufs_print }, /* print */ 202 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 203 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 204 { &vop_advlock_desc, spec_advlock }, /* advlock */ 205 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 206 { &vop_getpages_desc, spec_getpages }, /* getpages */ 207 { &vop_putpages_desc, spec_putpages }, /* putpages */ 208 { NULL, NULL } 209 }; 210 const struct vnodeopv_desc lfs_specop_opv_desc = 211 { &lfs_specop_p, lfs_specop_entries }; 212 213 int (**lfs_fifoop_p)(void *); 214 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 215 { &vop_default_desc, vn_default_error }, 216 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 217 { &vop_create_desc, fifo_create }, /* create */ 218 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 219 { &vop_open_desc, fifo_open }, /* open */ 220 { &vop_close_desc, lfsfifo_close }, /* close */ 221 { &vop_access_desc, ufs_access }, /* access */ 222 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 223 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 224 { &vop_read_desc, ufsfifo_read }, /* read */ 225 { &vop_write_desc, ufsfifo_write }, /* write */ 226 { &vop_lease_desc, fifo_lease_check }, /* lease */ 227 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 228 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 229 { &vop_poll_desc, fifo_poll }, /* poll */ 230 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 231 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 232 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 233 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 234 { &vop_seek_desc, fifo_seek }, /* seek */ 235 { &vop_remove_desc, fifo_remove }, /* remove */ 236 { &vop_link_desc, fifo_link }, /* link */ 237 { &vop_rename_desc, fifo_rename }, /* rename */ 238 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 239 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 240 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 241 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 242 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 243 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 244 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 245 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 246 { &vop_lock_desc, ufs_lock }, /* lock */ 247 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 248 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 249 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 250 { &vop_print_desc, ufs_print }, /* print */ 251 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 252 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 253 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 254 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 255 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 256 { NULL, NULL } 257 }; 258 const struct vnodeopv_desc lfs_fifoop_opv_desc = 259 { &lfs_fifoop_p, lfs_fifoop_entries }; 260 261 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int, struct vm_page **); 262 263 #define LFS_READWRITE 264 #include <ufs/ufs/ufs_readwrite.c> 265 #undef LFS_READWRITE 266 267 /* 268 * Synch an open file. 269 */ 270 /* ARGSUSED */ 271 int 272 lfs_fsync(void *v) 273 { 274 struct vop_fsync_args /* { 275 struct vnode *a_vp; 276 kauth_cred_t a_cred; 277 int a_flags; 278 off_t offlo; 279 off_t offhi; 280 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 bp->b_flags |= B_ERROR; 1180 biodone(bp); 1181 return (error); 1182 } 1183 if ((long)bp->b_blkno == -1) /* no valid data */ 1184 clrbuf(bp); 1185 } 1186 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1187 biodone(bp); 1188 return (0); 1189 } 1190 1191 slept = 1; 1192 simple_lock(&fs->lfs_interlock); 1193 while (slept && fs->lfs_seglock) { 1194 simple_unlock(&fs->lfs_interlock); 1195 /* 1196 * Look through list of intervals. 1197 * There will only be intervals to look through 1198 * if the cleaner holds the seglock. 1199 * Since the cleaner is synchronous, we can trust 1200 * the list of intervals to be current. 1201 */ 1202 tbn = dbtofsb(fs, bp->b_blkno); 1203 sn = dtosn(fs, tbn); 1204 slept = 0; 1205 for (i = 0; i < fs->lfs_cleanind; i++) { 1206 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1207 tbn >= fs->lfs_cleanint[i]) { 1208 DLOG((DLOG_CLEAN, 1209 "lfs_strategy: ino %d lbn %" PRId64 1210 " ind %d sn %d fsb %" PRIx32 1211 " given sn %d fsb %" PRIx64 "\n", 1212 ip->i_number, bp->b_lblkno, i, 1213 dtosn(fs, fs->lfs_cleanint[i]), 1214 fs->lfs_cleanint[i], sn, tbn)); 1215 DLOG((DLOG_CLEAN, 1216 "lfs_strategy: sleeping on ino %d lbn %" 1217 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1218 simple_lock(&fs->lfs_interlock); 1219 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1220 /* Cleaner can't wait for itself */ 1221 ltsleep(&fs->lfs_iocount, 1222 (PRIBIO + 1) | PNORELOCK, 1223 "clean2", 0, 1224 &fs->lfs_interlock); 1225 slept = 1; 1226 break; 1227 } else if (fs->lfs_seglock) { 1228 ltsleep(&fs->lfs_seglock, 1229 (PRIBIO + 1) | PNORELOCK, 1230 "clean1", 0, 1231 &fs->lfs_interlock); 1232 slept = 1; 1233 break; 1234 } 1235 simple_unlock(&fs->lfs_interlock); 1236 } 1237 } 1238 simple_lock(&fs->lfs_interlock); 1239 } 1240 simple_unlock(&fs->lfs_interlock); 1241 1242 vp = ip->i_devvp; 1243 VOP_STRATEGY(vp, bp); 1244 return (0); 1245 } 1246 1247 void 1248 lfs_flush_dirops(struct lfs *fs) 1249 { 1250 struct inode *ip, *nip; 1251 struct vnode *vp; 1252 extern int lfs_dostats; 1253 struct segment *sp; 1254 int waslocked; 1255 1256 ASSERT_MAYBE_SEGLOCK(fs); 1257 KASSERT(fs->lfs_nadirop == 0); 1258 1259 if (fs->lfs_ronly) 1260 return; 1261 1262 simple_lock(&fs->lfs_interlock); 1263 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1264 simple_unlock(&fs->lfs_interlock); 1265 return; 1266 } else 1267 simple_unlock(&fs->lfs_interlock); 1268 1269 if (lfs_dostats) 1270 ++lfs_stats.flush_invoked; 1271 1272 /* 1273 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1274 * Technically this is a checkpoint (the on-disk state is valid) 1275 * even though we are leaving out all the file data. 1276 */ 1277 lfs_imtime(fs); 1278 lfs_seglock(fs, SEGM_CKP); 1279 sp = fs->lfs_sp; 1280 1281 /* 1282 * lfs_writevnodes, optimized to get dirops out of the way. 1283 * Only write dirops, and don't flush files' pages, only 1284 * blocks from the directories. 1285 * 1286 * We don't need to vref these files because they are 1287 * dirops and so hold an extra reference until the 1288 * segunlock clears them of that status. 1289 * 1290 * We don't need to check for IN_ADIROP because we know that 1291 * no dirops are active. 1292 * 1293 */ 1294 simple_lock(&fs->lfs_interlock); 1295 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1296 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1297 simple_unlock(&fs->lfs_interlock); 1298 vp = ITOV(ip); 1299 1300 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1301 1302 /* 1303 * All writes to directories come from dirops; all 1304 * writes to files' direct blocks go through the page 1305 * cache, which we're not touching. Reads to files 1306 * and/or directories will not be affected by writing 1307 * directory blocks inodes and file inodes. So we don't 1308 * really need to lock. If we don't lock, though, 1309 * make sure that we don't clear IN_MODIFIED 1310 * unnecessarily. 1311 */ 1312 if (vp->v_flag & (VXLOCK | VFREEING)) { 1313 simple_lock(&fs->lfs_interlock); 1314 continue; 1315 } 1316 waslocked = VOP_ISLOCKED(vp); 1317 if (vp->v_type != VREG && 1318 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1319 lfs_writefile(fs, sp, vp); 1320 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1321 !(ip->i_flag & IN_ALLMOD)) { 1322 LFS_SET_UINO(ip, IN_MODIFIED); 1323 } 1324 } 1325 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1326 (void) lfs_writeinode(fs, sp, ip); 1327 if (waslocked == LK_EXCLOTHER) 1328 LFS_SET_UINO(ip, IN_MODIFIED); 1329 simple_lock(&fs->lfs_interlock); 1330 } 1331 simple_unlock(&fs->lfs_interlock); 1332 /* We've written all the dirops there are */ 1333 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1334 lfs_finalize_fs_seguse(fs); 1335 (void) lfs_writeseg(fs, sp); 1336 lfs_segunlock(fs); 1337 } 1338 1339 /* 1340 * Flush all vnodes for which the pagedaemon has requested pageouts. 1341 * Skip over any files that are marked VDIROP (since lfs_flush_dirop() 1342 * has just run, this would be an error). If we have to skip a vnode 1343 * for any reason, just skip it; if we have to wait for the cleaner, 1344 * abort. The writer daemon will call us again later. 1345 */ 1346 void 1347 lfs_flush_pchain(struct lfs *fs) 1348 { 1349 struct inode *ip, *nip; 1350 struct vnode *vp; 1351 extern int lfs_dostats; 1352 struct segment *sp; 1353 int error; 1354 1355 ASSERT_NO_SEGLOCK(fs); 1356 1357 if (fs->lfs_ronly) 1358 return; 1359 1360 simple_lock(&fs->lfs_interlock); 1361 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1362 simple_unlock(&fs->lfs_interlock); 1363 return; 1364 } else 1365 simple_unlock(&fs->lfs_interlock); 1366 1367 /* Get dirops out of the way */ 1368 lfs_flush_dirops(fs); 1369 1370 if (lfs_dostats) 1371 ++lfs_stats.flush_invoked; 1372 1373 /* 1374 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1375 */ 1376 lfs_imtime(fs); 1377 lfs_seglock(fs, 0); 1378 sp = fs->lfs_sp; 1379 1380 /* 1381 * lfs_writevnodes, optimized to clear pageout requests. 1382 * Only write non-dirop files that are in the pageout queue. 1383 * We're very conservative about what we write; we want to be 1384 * fast and async. 1385 */ 1386 simple_lock(&fs->lfs_interlock); 1387 top: 1388 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1389 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1390 vp = ITOV(ip); 1391 1392 if (!(ip->i_flags & IN_PAGING)) 1393 goto top; 1394 1395 if (vp->v_flag & (VXLOCK|VDIROP)) 1396 continue; 1397 if (vp->v_type != VREG) 1398 continue; 1399 if (lfs_vref(vp)) 1400 continue; 1401 simple_unlock(&fs->lfs_interlock); 1402 1403 if (VOP_ISLOCKED(vp)) { 1404 lfs_vunref(vp); 1405 simple_lock(&fs->lfs_interlock); 1406 continue; 1407 } 1408 1409 error = lfs_writefile(fs, sp, vp); 1410 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1411 !(ip->i_flag & IN_ALLMOD)) { 1412 LFS_SET_UINO(ip, IN_MODIFIED); 1413 } 1414 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1415 (void) lfs_writeinode(fs, sp, ip); 1416 1417 lfs_vunref(vp); 1418 1419 if (error == EAGAIN) { 1420 lfs_writeseg(fs, sp); 1421 simple_lock(&fs->lfs_interlock); 1422 break; 1423 } 1424 simple_lock(&fs->lfs_interlock); 1425 } 1426 simple_unlock(&fs->lfs_interlock); 1427 (void) lfs_writeseg(fs, sp); 1428 lfs_segunlock(fs); 1429 } 1430 1431 /* 1432 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1433 */ 1434 int 1435 lfs_fcntl(void *v) 1436 { 1437 struct vop_fcntl_args /* { 1438 struct vnode *a_vp; 1439 u_long a_command; 1440 void * a_data; 1441 int a_fflag; 1442 kauth_cred_t a_cred; 1443 struct lwp *a_l; 1444 } */ *ap = v; 1445 struct timeval *tvp; 1446 BLOCK_INFO *blkiov; 1447 CLEANERINFO *cip; 1448 SEGUSE *sup; 1449 int blkcnt, error, oclean; 1450 size_t fh_size; 1451 struct lfs_fcntl_markv blkvp; 1452 struct lwp *l; 1453 fsid_t *fsidp; 1454 struct lfs *fs; 1455 struct buf *bp; 1456 fhandle_t *fhp; 1457 daddr_t off; 1458 1459 /* Only respect LFS fcntls on fs root or Ifile */ 1460 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1461 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1462 return ufs_fcntl(v); 1463 } 1464 1465 /* Avoid locking a draining lock */ 1466 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1467 return ESHUTDOWN; 1468 } 1469 1470 /* LFS control and monitoring fcntls are available only to root */ 1471 l = ap->a_l; 1472 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1473 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1474 NULL)) != 0) 1475 return (error); 1476 1477 fs = VTOI(ap->a_vp)->i_lfs; 1478 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1479 1480 error = 0; 1481 switch (ap->a_command) { 1482 case LFCNSEGWAITALL: 1483 case LFCNSEGWAITALL_COMPAT: 1484 fsidp = NULL; 1485 /* FALLSTHROUGH */ 1486 case LFCNSEGWAIT: 1487 case LFCNSEGWAIT_COMPAT: 1488 tvp = (struct timeval *)ap->a_data; 1489 simple_lock(&fs->lfs_interlock); 1490 ++fs->lfs_sleepers; 1491 simple_unlock(&fs->lfs_interlock); 1492 1493 error = lfs_segwait(fsidp, tvp); 1494 1495 simple_lock(&fs->lfs_interlock); 1496 if (--fs->lfs_sleepers == 0) 1497 wakeup(&fs->lfs_sleepers); 1498 simple_unlock(&fs->lfs_interlock); 1499 return error; 1500 1501 case LFCNBMAPV: 1502 case LFCNMARKV: 1503 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1504 1505 blkcnt = blkvp.blkcnt; 1506 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1507 return (EINVAL); 1508 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1509 if ((error = copyin(blkvp.blkiov, blkiov, 1510 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1511 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1512 return error; 1513 } 1514 1515 simple_lock(&fs->lfs_interlock); 1516 ++fs->lfs_sleepers; 1517 simple_unlock(&fs->lfs_interlock); 1518 if (ap->a_command == LFCNBMAPV) 1519 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1520 else /* LFCNMARKV */ 1521 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1522 if (error == 0) 1523 error = copyout(blkiov, blkvp.blkiov, 1524 blkcnt * sizeof(BLOCK_INFO)); 1525 simple_lock(&fs->lfs_interlock); 1526 if (--fs->lfs_sleepers == 0) 1527 wakeup(&fs->lfs_sleepers); 1528 simple_unlock(&fs->lfs_interlock); 1529 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1530 return error; 1531 1532 case LFCNRECLAIM: 1533 /* 1534 * Flush dirops and write Ifile, allowing empty segments 1535 * to be immediately reclaimed. 1536 */ 1537 lfs_writer_enter(fs, "pndirop"); 1538 off = fs->lfs_offset; 1539 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1540 lfs_flush_dirops(fs); 1541 LFS_CLEANERINFO(cip, fs, bp); 1542 oclean = cip->clean; 1543 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1544 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1545 fs->lfs_sp->seg_flags |= SEGM_PROT; 1546 lfs_segunlock(fs); 1547 lfs_writer_leave(fs); 1548 1549 #ifdef DEBUG 1550 LFS_CLEANERINFO(cip, fs, bp); 1551 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1552 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1553 fs->lfs_offset - off, cip->clean - oclean, 1554 fs->lfs_activesb)); 1555 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1556 #endif 1557 1558 return 0; 1559 1560 #ifdef COMPAT_30 1561 case LFCNIFILEFH_COMPAT: 1562 /* Return the filehandle of the Ifile */ 1563 if ((error = kauth_authorize_generic(l->l_cred, 1564 KAUTH_GENERIC_ISSUSER, NULL)) != 0) 1565 return (error); 1566 fhp = (struct fhandle *)ap->a_data; 1567 fhp->fh_fsid = *fsidp; 1568 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1569 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1570 #endif 1571 1572 case LFCNIFILEFH_COMPAT2: 1573 case LFCNIFILEFH: 1574 /* Return the filehandle of the Ifile */ 1575 fhp = (struct fhandle *)ap->a_data; 1576 fhp->fh_fsid = *fsidp; 1577 fh_size = sizeof(struct lfs_fhandle) - 1578 offsetof(fhandle_t, fh_fid); 1579 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1580 1581 case LFCNREWIND: 1582 /* Move lfs_offset to the lowest-numbered segment */ 1583 return lfs_rewind(fs, *(int *)ap->a_data); 1584 1585 case LFCNINVAL: 1586 /* Mark a segment SEGUSE_INVAL */ 1587 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1588 if (sup->su_nbytes > 0) { 1589 brelse(bp); 1590 lfs_unset_inval_all(fs); 1591 return EBUSY; 1592 } 1593 sup->su_flags |= SEGUSE_INVAL; 1594 VOP_BWRITE(bp); 1595 return 0; 1596 1597 case LFCNRESIZE: 1598 /* Resize the filesystem */ 1599 return lfs_resize_fs(fs, *(int *)ap->a_data); 1600 1601 case LFCNWRAPSTOP: 1602 case LFCNWRAPSTOP_COMPAT: 1603 /* 1604 * Hold lfs_newseg at segment 0; if requested, sleep until 1605 * the filesystem wraps around. To support external agents 1606 * (dump, fsck-based regression test) that need to look at 1607 * a snapshot of the filesystem, without necessarily 1608 * requiring that all fs activity stops. 1609 */ 1610 if (lockstatus(&fs->lfs_stoplock)) 1611 return EALREADY; 1612 1613 simple_lock(&fs->lfs_interlock); 1614 lockmgr(&fs->lfs_stoplock, LK_EXCLUSIVE, &fs->lfs_interlock); 1615 if (fs->lfs_nowrap == 0) 1616 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1617 ++fs->lfs_nowrap; 1618 if (*(int *)ap->a_data == 1 || 1619 ap->a_command == LFCNWRAPSTOP_COMPAT) { 1620 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1621 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1622 "segwrap", 0, &fs->lfs_interlock); 1623 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1624 if (error) { 1625 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1626 } 1627 } 1628 simple_unlock(&fs->lfs_interlock); 1629 return 0; 1630 1631 case LFCNWRAPGO: 1632 case LFCNWRAPGO_COMPAT: 1633 /* 1634 * Having done its work, the agent wakes up the writer. 1635 * If the argument is 1, it sleeps until a new segment 1636 * is selected. 1637 */ 1638 simple_lock(&fs->lfs_interlock); 1639 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1640 (ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1641 *((int *)ap->a_data))); 1642 simple_unlock(&fs->lfs_interlock); 1643 return error; 1644 1645 case LFCNWRAPPASS: 1646 if (lockstatus(&fs->lfs_stoplock) != LK_EXCLUSIVE) 1647 return EALREADY; 1648 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1649 return EALREADY; 1650 simple_lock(&fs->lfs_interlock); 1651 if (fs->lfs_nowrap == 0) { 1652 simple_unlock(&fs->lfs_interlock); 1653 return EBUSY; 1654 } 1655 fs->lfs_wrappass = 1; 1656 wakeup(&fs->lfs_wrappass); 1657 /* Wait for the log to wrap, if asked */ 1658 if (*(int *)ap->a_data) { 1659 lfs_vref(ap->a_vp); 1660 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1661 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1662 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1663 "segwrap", 0, &fs->lfs_interlock); 1664 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1665 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1666 lfs_vunref(ap->a_vp); 1667 } 1668 simple_unlock(&fs->lfs_interlock); 1669 return error; 1670 1671 case LFCNWRAPSTATUS: 1672 simple_lock(&fs->lfs_interlock); 1673 *(int *)ap->a_data = fs->lfs_wrapstatus; 1674 simple_unlock(&fs->lfs_interlock); 1675 return 0; 1676 1677 default: 1678 return ufs_fcntl(v); 1679 } 1680 return 0; 1681 } 1682 1683 int 1684 lfs_getpages(void *v) 1685 { 1686 struct vop_getpages_args /* { 1687 struct vnode *a_vp; 1688 voff_t a_offset; 1689 struct vm_page **a_m; 1690 int *a_count; 1691 int a_centeridx; 1692 vm_prot_t a_access_type; 1693 int a_advice; 1694 int a_flags; 1695 } */ *ap = v; 1696 1697 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1698 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1699 return EPERM; 1700 } 1701 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1702 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1703 } 1704 1705 /* 1706 * we're relying on the fact that genfs_getpages() always read in 1707 * entire filesystem blocks. 1708 */ 1709 return genfs_getpages(v); 1710 } 1711 1712 /* 1713 * Wait for a page to become unbusy, possibly printing diagnostic messages 1714 * as well. 1715 * 1716 * Called with vp->v_interlock held; return with it held. 1717 */ 1718 static void 1719 wait_for_page(struct vnode *vp, struct vm_page *pg, const char *label) 1720 { 1721 if ((pg->flags & PG_BUSY) == 0) 1722 return; /* Nothing to wait for! */ 1723 1724 #if defined(DEBUG) && defined(UVM_PAGE_TRKOWN) 1725 static struct vm_page *lastpg; 1726 1727 if (label != NULL && pg != lastpg) { 1728 if (pg->owner_tag) { 1729 printf("lfs_putpages[%d.%d]: %s: page %p owner %d.%d [%s]\n", 1730 curproc->p_pid, curlwp->l_lid, label, 1731 pg, pg->owner, pg->lowner, pg->owner_tag); 1732 } else { 1733 printf("lfs_putpages[%d.%d]: %s: page %p unowned?!\n", 1734 curproc->p_pid, curlwp->l_lid, label, pg); 1735 } 1736 } 1737 lastpg = pg; 1738 #endif 1739 1740 pg->flags |= PG_WANTED; 1741 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, "lfsput", 0); 1742 simple_lock(&vp->v_interlock); 1743 } 1744 1745 /* 1746 * This routine is called by lfs_putpages() when it can't complete the 1747 * write because a page is busy. This means that either (1) someone, 1748 * possibly the pagedaemon, is looking at this page, and will give it up 1749 * presently; or (2) we ourselves are holding the page busy in the 1750 * process of being written (either gathered or actually on its way to 1751 * disk). We don't need to give up the segment lock, but we might need 1752 * to call lfs_writeseg() to expedite the page's journey to disk. 1753 * 1754 * Called with vp->v_interlock held; return with it held. 1755 */ 1756 /* #define BUSYWAIT */ 1757 static void 1758 write_and_wait(struct lfs *fs, struct vnode *vp, struct vm_page *pg, 1759 int seglocked, const char *label) 1760 { 1761 #ifndef BUSYWAIT 1762 struct inode *ip = VTOI(vp); 1763 struct segment *sp = fs->lfs_sp; 1764 int count = 0; 1765 1766 if (pg == NULL) 1767 return; 1768 1769 while (pg->flags & PG_BUSY) { 1770 simple_unlock(&vp->v_interlock); 1771 if (sp->cbpp - sp->bpp > 1) { 1772 /* Write gathered pages */ 1773 lfs_updatemeta(sp); 1774 lfs_release_finfo(fs); 1775 (void) lfs_writeseg(fs, sp); 1776 1777 /* 1778 * Reinitialize FIP 1779 */ 1780 KASSERT(sp->vp == vp); 1781 lfs_acquire_finfo(fs, ip->i_number, 1782 ip->i_gen); 1783 } 1784 ++count; 1785 simple_lock(&vp->v_interlock); 1786 wait_for_page(vp, pg, label); 1787 } 1788 if (label != NULL && count > 1) 1789 printf("lfs_putpages[%d]: %s: %sn = %d\n", curproc->p_pid, 1790 label, (count > 0 ? "looping, " : ""), count); 1791 #else 1792 preempt(1); 1793 #endif 1794 } 1795 1796 /* 1797 * Make sure that for all pages in every block in the given range, 1798 * either all are dirty or all are clean. If any of the pages 1799 * we've seen so far are dirty, put the vnode on the paging chain, 1800 * and mark it IN_PAGING. 1801 * 1802 * If checkfirst != 0, don't check all the pages but return at the 1803 * first dirty page. 1804 */ 1805 static int 1806 check_dirty(struct lfs *fs, struct vnode *vp, 1807 off_t startoffset, off_t endoffset, off_t blkeof, 1808 int flags, int checkfirst, struct vm_page **pgp) 1809 { 1810 int by_list; 1811 struct vm_page *curpg = NULL; /* XXX: gcc */ 1812 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1813 off_t soff = 0; /* XXX: gcc */ 1814 voff_t off; 1815 int i; 1816 int nonexistent; 1817 int any_dirty; /* number of dirty pages */ 1818 int dirty; /* number of dirty pages in a block */ 1819 int tdirty; 1820 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1821 int pagedaemon = (curlwp == uvm.pagedaemon_lwp); 1822 1823 ASSERT_MAYBE_SEGLOCK(fs); 1824 top: 1825 by_list = (vp->v_uobj.uo_npages <= 1826 ((endoffset - startoffset) >> PAGE_SHIFT) * 1827 UVM_PAGE_HASH_PENALTY); 1828 any_dirty = 0; 1829 1830 if (by_list) { 1831 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1832 } else { 1833 soff = startoffset; 1834 } 1835 while (by_list || soff < MIN(blkeof, endoffset)) { 1836 if (by_list) { 1837 /* 1838 * Find the first page in a block. Skip 1839 * blocks outside our area of interest or beyond 1840 * the end of file. 1841 */ 1842 if (pages_per_block > 1) { 1843 while (curpg && 1844 ((curpg->offset & fs->lfs_bmask) || 1845 curpg->offset >= vp->v_size || 1846 curpg->offset >= endoffset)) 1847 curpg = TAILQ_NEXT(curpg, listq); 1848 } 1849 if (curpg == NULL) 1850 break; 1851 soff = curpg->offset; 1852 } 1853 1854 /* 1855 * Mark all pages in extended range busy; find out if any 1856 * of them are dirty. 1857 */ 1858 nonexistent = dirty = 0; 1859 for (i = 0; i == 0 || i < pages_per_block; i++) { 1860 if (by_list && pages_per_block <= 1) { 1861 pgs[i] = pg = curpg; 1862 } else { 1863 off = soff + (i << PAGE_SHIFT); 1864 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1865 if (pg == NULL) { 1866 ++nonexistent; 1867 continue; 1868 } 1869 } 1870 KASSERT(pg != NULL); 1871 1872 /* 1873 * If we're holding the segment lock, we can deadlock 1874 * against a process that has our page and is waiting 1875 * for the cleaner, while the cleaner waits for the 1876 * segment lock. Just bail in that case. 1877 */ 1878 if ((pg->flags & PG_BUSY) && 1879 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1880 if (i > 0) 1881 uvm_page_unbusy(pgs, i); 1882 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1883 if (pgp) 1884 *pgp = pg; 1885 return -1; 1886 } 1887 1888 while (pg->flags & PG_BUSY) { 1889 wait_for_page(vp, pg, NULL); 1890 if (i > 0) 1891 uvm_page_unbusy(pgs, i); 1892 goto top; 1893 } 1894 pg->flags |= PG_BUSY; 1895 UVM_PAGE_OWN(pg, "lfs_putpages"); 1896 1897 pmap_page_protect(pg, VM_PROT_NONE); 1898 tdirty = (pmap_clear_modify(pg) || 1899 (pg->flags & PG_CLEAN) == 0); 1900 dirty += tdirty; 1901 } 1902 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1903 if (by_list) { 1904 curpg = TAILQ_NEXT(curpg, listq); 1905 } else { 1906 soff += fs->lfs_bsize; 1907 } 1908 continue; 1909 } 1910 1911 any_dirty += dirty; 1912 KASSERT(nonexistent == 0); 1913 1914 /* 1915 * If any are dirty make all dirty; unbusy them, 1916 * but if we were asked to clean, wire them so that 1917 * the pagedaemon doesn't bother us about them while 1918 * they're on their way to disk. 1919 */ 1920 for (i = 0; i == 0 || i < pages_per_block; i++) { 1921 pg = pgs[i]; 1922 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1923 if (dirty) { 1924 pg->flags &= ~PG_CLEAN; 1925 if (flags & PGO_FREE) { 1926 /* 1927 * Wire the page so that 1928 * pdaemon doesn't see it again. 1929 */ 1930 uvm_lock_pageq(); 1931 uvm_pagewire(pg); 1932 uvm_unlock_pageq(); 1933 1934 /* Suspended write flag */ 1935 pg->flags |= PG_DELWRI; 1936 } 1937 } 1938 if (pg->flags & PG_WANTED) 1939 wakeup(pg); 1940 pg->flags &= ~(PG_WANTED|PG_BUSY); 1941 UVM_PAGE_OWN(pg, NULL); 1942 } 1943 1944 if (checkfirst && any_dirty) 1945 break; 1946 1947 if (by_list) { 1948 curpg = TAILQ_NEXT(curpg, listq); 1949 } else { 1950 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1951 } 1952 } 1953 1954 return any_dirty; 1955 } 1956 1957 /* 1958 * lfs_putpages functions like genfs_putpages except that 1959 * 1960 * (1) It needs to bounds-check the incoming requests to ensure that 1961 * they are block-aligned; if they are not, expand the range and 1962 * do the right thing in case, e.g., the requested range is clean 1963 * but the expanded range is dirty. 1964 * 1965 * (2) It needs to explicitly send blocks to be written when it is done. 1966 * If VOP_PUTPAGES is called without the seglock held, we simply take 1967 * the seglock and let lfs_segunlock wait for us. 1968 * XXX There might be a bad situation if we have to flush a vnode while 1969 * XXX lfs_markv is in operation. As of this writing we panic in this 1970 * XXX case. 1971 * 1972 * Assumptions: 1973 * 1974 * (1) The caller does not hold any pages in this vnode busy. If it does, 1975 * there is a danger that when we expand the page range and busy the 1976 * pages we will deadlock. 1977 * 1978 * (2) We are called with vp->v_interlock held; we must return with it 1979 * released. 1980 * 1981 * (3) We don't absolutely have to free pages right away, provided that 1982 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1983 * us a request with PGO_FREE, we take the pages out of the paging 1984 * queue and wake up the writer, which will handle freeing them for us. 1985 * 1986 * We ensure that for any filesystem block, all pages for that 1987 * block are either resident or not, even if those pages are higher 1988 * than EOF; that means that we will be getting requests to free 1989 * "unused" pages above EOF all the time, and should ignore them. 1990 * 1991 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 1992 * into has been set up for us by lfs_writefile. If not, we will 1993 * have to handle allocating and/or freeing an finfo entry. 1994 * 1995 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1996 */ 1997 1998 /* How many times to loop before we should start to worry */ 1999 #define TOOMANY 4 2000 2001 int 2002 lfs_putpages(void *v) 2003 { 2004 int error; 2005 struct vop_putpages_args /* { 2006 struct vnode *a_vp; 2007 voff_t a_offlo; 2008 voff_t a_offhi; 2009 int a_flags; 2010 } */ *ap = v; 2011 struct vnode *vp; 2012 struct inode *ip; 2013 struct lfs *fs; 2014 struct segment *sp; 2015 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 2016 off_t off, max_endoffset; 2017 int s; 2018 bool seglocked, sync, pagedaemon; 2019 struct vm_page *pg, *busypg; 2020 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 2021 #ifdef DEBUG 2022 int debug_n_again, debug_n_dirtyclean; 2023 #endif 2024 2025 vp = ap->a_vp; 2026 ip = VTOI(vp); 2027 fs = ip->i_lfs; 2028 sync = (ap->a_flags & PGO_SYNCIO) != 0; 2029 pagedaemon = (curlwp == uvm.pagedaemon_lwp); 2030 2031 /* Putpages does nothing for metadata. */ 2032 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 2033 simple_unlock(&vp->v_interlock); 2034 return 0; 2035 } 2036 2037 /* 2038 * If there are no pages, don't do anything. 2039 */ 2040 if (vp->v_uobj.uo_npages == 0) { 2041 s = splbio(); 2042 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 2043 (vp->v_flag & VONWORKLST) && 2044 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 2045 vp->v_flag &= ~VWRITEMAPDIRTY; 2046 vn_syncer_remove_from_worklist(vp); 2047 } 2048 splx(s); 2049 simple_unlock(&vp->v_interlock); 2050 2051 /* Remove us from paging queue, if we were on it */ 2052 simple_lock(&fs->lfs_interlock); 2053 if (ip->i_flags & IN_PAGING) { 2054 ip->i_flags &= ~IN_PAGING; 2055 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2056 } 2057 simple_unlock(&fs->lfs_interlock); 2058 return 0; 2059 } 2060 2061 blkeof = blkroundup(fs, ip->i_size); 2062 2063 /* 2064 * Ignore requests to free pages past EOF but in the same block 2065 * as EOF, unless the request is synchronous. (If the request is 2066 * sync, it comes from lfs_truncate.) 2067 * XXXUBC Make these pages look "active" so the pagedaemon won't 2068 * XXXUBC bother us with them again. 2069 */ 2070 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 2071 origoffset = ap->a_offlo; 2072 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 2073 pg = uvm_pagelookup(&vp->v_uobj, off); 2074 KASSERT(pg != NULL); 2075 while (pg->flags & PG_BUSY) { 2076 pg->flags |= PG_WANTED; 2077 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 2078 "lfsput2", 0); 2079 simple_lock(&vp->v_interlock); 2080 } 2081 uvm_lock_pageq(); 2082 uvm_pageactivate(pg); 2083 uvm_unlock_pageq(); 2084 } 2085 ap->a_offlo = blkeof; 2086 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 2087 simple_unlock(&vp->v_interlock); 2088 return 0; 2089 } 2090 } 2091 2092 /* 2093 * Extend page range to start and end at block boundaries. 2094 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 2095 */ 2096 origoffset = ap->a_offlo; 2097 origendoffset = ap->a_offhi; 2098 startoffset = origoffset & ~(fs->lfs_bmask); 2099 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 2100 << fs->lfs_bshift; 2101 2102 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2103 endoffset = max_endoffset; 2104 origendoffset = endoffset; 2105 } else { 2106 origendoffset = round_page(ap->a_offhi); 2107 endoffset = round_page(blkroundup(fs, origendoffset)); 2108 } 2109 2110 KASSERT(startoffset > 0 || endoffset >= startoffset); 2111 if (startoffset == endoffset) { 2112 /* Nothing to do, why were we called? */ 2113 simple_unlock(&vp->v_interlock); 2114 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 2115 PRId64 "\n", startoffset)); 2116 return 0; 2117 } 2118 2119 ap->a_offlo = startoffset; 2120 ap->a_offhi = endoffset; 2121 2122 /* 2123 * If not cleaning, just send the pages through genfs_putpages 2124 * to be returned to the pool. 2125 */ 2126 if (!(ap->a_flags & PGO_CLEANIT)) 2127 return genfs_putpages(v); 2128 2129 /* Set PGO_BUSYFAIL to avoid deadlocks */ 2130 ap->a_flags |= PGO_BUSYFAIL; 2131 2132 /* 2133 * Likewise, if we are asked to clean but the pages are not 2134 * dirty, we can just free them using genfs_putpages. 2135 */ 2136 #ifdef DEBUG 2137 debug_n_dirtyclean = 0; 2138 #endif 2139 do { 2140 int r; 2141 2142 /* Count the number of dirty pages */ 2143 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2144 ap->a_flags, 1, NULL); 2145 if (r < 0) { 2146 /* Pages are busy with another process */ 2147 simple_unlock(&vp->v_interlock); 2148 return EDEADLK; 2149 } 2150 if (r > 0) /* Some pages are dirty */ 2151 break; 2152 2153 /* 2154 * Sometimes pages are dirtied between the time that 2155 * we check and the time we try to clean them. 2156 * Instruct lfs_gop_write to return EDEADLK in this case 2157 * so we can write them properly. 2158 */ 2159 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2160 r = genfs_do_putpages(vp, startoffset, endoffset, 2161 ap->a_flags, &busypg); 2162 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2163 if (r != EDEADLK) 2164 return r; 2165 2166 /* One of the pages was busy. Start over. */ 2167 simple_lock(&vp->v_interlock); 2168 wait_for_page(vp, busypg, "dirtyclean"); 2169 #ifdef DEBUG 2170 ++debug_n_dirtyclean; 2171 #endif 2172 } while(1); 2173 2174 #ifdef DEBUG 2175 if (debug_n_dirtyclean > TOOMANY) 2176 printf("lfs_putpages: dirtyclean: looping, n = %d\n", 2177 debug_n_dirtyclean); 2178 #endif 2179 2180 /* 2181 * Dirty and asked to clean. 2182 * 2183 * Pagedaemon can't actually write LFS pages; wake up 2184 * the writer to take care of that. The writer will 2185 * notice the pager inode queue and act on that. 2186 */ 2187 if (pagedaemon) { 2188 simple_lock(&fs->lfs_interlock); 2189 if (!(ip->i_flags & IN_PAGING)) { 2190 ip->i_flags |= IN_PAGING; 2191 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2192 } 2193 simple_lock(&lfs_subsys_lock); 2194 wakeup(&lfs_writer_daemon); 2195 simple_unlock(&lfs_subsys_lock); 2196 simple_unlock(&fs->lfs_interlock); 2197 simple_unlock(&vp->v_interlock); 2198 preempt(); 2199 return EWOULDBLOCK; 2200 } 2201 2202 /* 2203 * If this is a file created in a recent dirop, we can't flush its 2204 * inode until the dirop is complete. Drain dirops, then flush the 2205 * filesystem (taking care of any other pending dirops while we're 2206 * at it). 2207 */ 2208 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2209 (vp->v_flag & VDIROP)) { 2210 int locked; 2211 2212 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 2213 locked = (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 2214 simple_unlock(&vp->v_interlock); 2215 lfs_writer_enter(fs, "ppdirop"); 2216 if (locked) 2217 VOP_UNLOCK(vp, 0); /* XXX why? */ 2218 2219 simple_lock(&fs->lfs_interlock); 2220 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2221 simple_unlock(&fs->lfs_interlock); 2222 2223 simple_lock(&vp->v_interlock); 2224 if (locked) { 2225 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2226 simple_lock(&vp->v_interlock); 2227 } 2228 lfs_writer_leave(fs); 2229 2230 /* XXX the flush should have taken care of this one too! */ 2231 } 2232 2233 /* 2234 * This is it. We are going to write some pages. From here on 2235 * down it's all just mechanics. 2236 * 2237 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2238 */ 2239 ap->a_flags &= ~PGO_SYNCIO; 2240 2241 /* 2242 * If we've already got the seglock, flush the node and return. 2243 * The FIP has already been set up for us by lfs_writefile, 2244 * and FIP cleanup and lfs_updatemeta will also be done there, 2245 * unless genfs_putpages returns EDEADLK; then we must flush 2246 * what we have, and correct FIP and segment header accounting. 2247 */ 2248 get_seglock: 2249 /* 2250 * If we are not called with the segment locked, lock it. 2251 * Account for a new FIP in the segment header, and set sp->vp. 2252 * (This should duplicate the setup at the top of lfs_writefile().) 2253 */ 2254 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2255 if (!seglocked) { 2256 simple_unlock(&vp->v_interlock); 2257 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2258 if (error != 0) 2259 return error; 2260 simple_lock(&vp->v_interlock); 2261 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2262 } 2263 sp = fs->lfs_sp; 2264 KASSERT(sp->vp == NULL); 2265 sp->vp = vp; 2266 2267 /* 2268 * Ensure that the partial segment is marked SS_DIROP if this 2269 * vnode is a DIROP. 2270 */ 2271 if (!seglocked && vp->v_flag & VDIROP) 2272 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2273 2274 /* 2275 * Loop over genfs_putpages until all pages are gathered. 2276 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2277 * Whenever we lose the interlock we have to rerun check_dirty, as 2278 * well, since more pages might have been dirtied in our absence. 2279 */ 2280 #ifdef DEBUG 2281 debug_n_again = 0; 2282 #endif 2283 do { 2284 busypg = NULL; 2285 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2286 ap->a_flags, 0, &busypg) < 0) { 2287 simple_unlock(&vp->v_interlock); 2288 2289 simple_lock(&vp->v_interlock); 2290 write_and_wait(fs, vp, busypg, seglocked, NULL); 2291 if (!seglocked) { 2292 lfs_release_finfo(fs); 2293 lfs_segunlock(fs); 2294 } 2295 sp->vp = NULL; 2296 goto get_seglock; 2297 } 2298 2299 busypg = NULL; 2300 error = genfs_do_putpages(vp, startoffset, endoffset, 2301 ap->a_flags, &busypg); 2302 2303 if (error == EDEADLK || error == EAGAIN) { 2304 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2305 " %d ino %d off %x (seg %d)\n", error, 2306 ip->i_number, fs->lfs_offset, 2307 dtosn(fs, fs->lfs_offset))); 2308 2309 simple_lock(&vp->v_interlock); 2310 write_and_wait(fs, vp, busypg, seglocked, "again"); 2311 } 2312 #ifdef DEBUG 2313 ++debug_n_again; 2314 #endif 2315 } while (error == EDEADLK); 2316 #ifdef DEBUG 2317 if (debug_n_again > TOOMANY) 2318 printf("lfs_putpages: again: looping, n = %d\n", debug_n_again); 2319 #endif 2320 2321 KASSERT(sp != NULL && sp->vp == vp); 2322 if (!seglocked) { 2323 sp->vp = NULL; 2324 2325 /* Write indirect blocks as well */ 2326 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2327 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2328 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2329 2330 KASSERT(sp->vp == NULL); 2331 sp->vp = vp; 2332 } 2333 2334 /* 2335 * Blocks are now gathered into a segment waiting to be written. 2336 * All that's left to do is update metadata, and write them. 2337 */ 2338 lfs_updatemeta(sp); 2339 KASSERT(sp->vp == vp); 2340 sp->vp = NULL; 2341 2342 /* 2343 * If we were called from lfs_writefile, we don't need to clean up 2344 * the FIP or unlock the segment lock. We're done. 2345 */ 2346 if (seglocked) 2347 return error; 2348 2349 /* Clean up FIP and send it to disk. */ 2350 lfs_release_finfo(fs); 2351 lfs_writeseg(fs, fs->lfs_sp); 2352 2353 /* 2354 * Remove us from paging queue if we wrote all our pages. 2355 */ 2356 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2357 simple_lock(&fs->lfs_interlock); 2358 if (ip->i_flags & IN_PAGING) { 2359 ip->i_flags &= ~IN_PAGING; 2360 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2361 } 2362 simple_unlock(&fs->lfs_interlock); 2363 } 2364 2365 /* 2366 * XXX - with the malloc/copy writeseg, the pages are freed by now 2367 * even if we don't wait (e.g. if we hold a nested lock). This 2368 * will not be true if we stop using malloc/copy. 2369 */ 2370 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2371 lfs_segunlock(fs); 2372 2373 /* 2374 * Wait for v_numoutput to drop to zero. The seglock should 2375 * take care of this, but there is a slight possibility that 2376 * aiodoned might not have got around to our buffers yet. 2377 */ 2378 if (sync) { 2379 s = splbio(); 2380 simple_lock(&global_v_numoutput_slock); 2381 while (vp->v_numoutput > 0) { 2382 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2383 " num %d\n", ip->i_number, vp->v_numoutput)); 2384 vp->v_flag |= VBWAIT; 2385 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 2386 &global_v_numoutput_slock); 2387 } 2388 simple_unlock(&global_v_numoutput_slock); 2389 splx(s); 2390 } 2391 return error; 2392 } 2393 2394 /* 2395 * Return the last logical file offset that should be written for this file 2396 * if we're doing a write that ends at "size". If writing, we need to know 2397 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2398 * to know about entire blocks. 2399 */ 2400 void 2401 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2402 { 2403 struct inode *ip = VTOI(vp); 2404 struct lfs *fs = ip->i_lfs; 2405 daddr_t olbn, nlbn; 2406 2407 olbn = lblkno(fs, ip->i_size); 2408 nlbn = lblkno(fs, size); 2409 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2410 *eobp = fragroundup(fs, size); 2411 } else { 2412 *eobp = blkroundup(fs, size); 2413 } 2414 } 2415 2416 #ifdef DEBUG 2417 void lfs_dump_vop(void *); 2418 2419 void 2420 lfs_dump_vop(void *v) 2421 { 2422 struct vop_putpages_args /* { 2423 struct vnode *a_vp; 2424 voff_t a_offlo; 2425 voff_t a_offhi; 2426 int a_flags; 2427 } */ *ap = v; 2428 2429 #ifdef DDB 2430 vfs_vnode_print(ap->a_vp, 0, printf); 2431 #endif 2432 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2433 } 2434 #endif 2435 2436 int 2437 lfs_mmap(void *v) 2438 { 2439 struct vop_mmap_args /* { 2440 const struct vnodeop_desc *a_desc; 2441 struct vnode *a_vp; 2442 vm_prot_t a_prot; 2443 kauth_cred_t a_cred; 2444 struct lwp *a_l; 2445 } */ *ap = v; 2446 2447 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2448 return EOPNOTSUPP; 2449 return ufs_mmap(v); 2450 } 2451
Indexes created Tue Sep 30 17:09:57 GMT 2025