lfs_vnops.c revision 1.170
1 /* $NetBSD: lfs_vnops.c,v 1.170 2006/04/30 21:19:42 perseant 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 1.170 2006/04/30 21:19:42 perseant Exp $"); 71 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/namei.h> 75 #include <sys/resourcevar.h> 76 #include <sys/kernel.h> 77 #include <sys/file.h> 78 #include <sys/stat.h> 79 #include <sys/buf.h> 80 #include <sys/proc.h> 81 #include <sys/mount.h> 82 #include <sys/vnode.h> 83 #include <sys/pool.h> 84 #include <sys/signalvar.h> 85 86 #include <miscfs/fifofs/fifo.h> 87 #include <miscfs/genfs/genfs.h> 88 #include <miscfs/specfs/specdev.h> 89 90 #include <ufs/ufs/inode.h> 91 #include <ufs/ufs/dir.h> 92 #include <ufs/ufs/ufsmount.h> 93 #include <ufs/ufs/ufs_extern.h> 94 95 #include <uvm/uvm.h> 96 #include <uvm/uvm_pmap.h> 97 #include <uvm/uvm_stat.h> 98 #include <uvm/uvm_pager.h> 99 100 #include <ufs/lfs/lfs.h> 101 #include <ufs/lfs/lfs_extern.h> 102 103 extern pid_t lfs_writer_daemon; 104 105 /* Global vfs data structures for lfs. */ 106 int (**lfs_vnodeop_p)(void *); 107 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 108 { &vop_default_desc, vn_default_error }, 109 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 110 { &vop_create_desc, lfs_create }, /* create */ 111 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 112 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 113 { &vop_open_desc, ufs_open }, /* open */ 114 { &vop_close_desc, lfs_close }, /* close */ 115 { &vop_access_desc, ufs_access }, /* access */ 116 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 117 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 118 { &vop_read_desc, lfs_read }, /* read */ 119 { &vop_write_desc, lfs_write }, /* write */ 120 { &vop_lease_desc, ufs_lease_check }, /* lease */ 121 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 122 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 123 { &vop_poll_desc, ufs_poll }, /* poll */ 124 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 125 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 126 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 127 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 128 { &vop_seek_desc, ufs_seek }, /* seek */ 129 { &vop_remove_desc, lfs_remove }, /* remove */ 130 { &vop_link_desc, lfs_link }, /* link */ 131 { &vop_rename_desc, lfs_rename }, /* rename */ 132 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 133 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 134 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 135 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 136 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 137 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 138 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 139 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 140 { &vop_lock_desc, ufs_lock }, /* lock */ 141 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 142 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 143 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 144 { &vop_print_desc, ufs_print }, /* print */ 145 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 146 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 147 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 148 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 149 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 150 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 151 { NULL, NULL } 152 }; 153 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 154 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 155 156 int (**lfs_specop_p)(void *); 157 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 158 { &vop_default_desc, vn_default_error }, 159 { &vop_lookup_desc, spec_lookup }, /* lookup */ 160 { &vop_create_desc, spec_create }, /* create */ 161 { &vop_mknod_desc, spec_mknod }, /* mknod */ 162 { &vop_open_desc, spec_open }, /* open */ 163 { &vop_close_desc, lfsspec_close }, /* close */ 164 { &vop_access_desc, ufs_access }, /* access */ 165 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 166 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 167 { &vop_read_desc, ufsspec_read }, /* read */ 168 { &vop_write_desc, ufsspec_write }, /* write */ 169 { &vop_lease_desc, spec_lease_check }, /* lease */ 170 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 171 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 172 { &vop_poll_desc, spec_poll }, /* poll */ 173 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 174 { &vop_revoke_desc, spec_revoke }, /* revoke */ 175 { &vop_mmap_desc, spec_mmap }, /* mmap */ 176 { &vop_fsync_desc, spec_fsync }, /* fsync */ 177 { &vop_seek_desc, spec_seek }, /* seek */ 178 { &vop_remove_desc, spec_remove }, /* remove */ 179 { &vop_link_desc, spec_link }, /* link */ 180 { &vop_rename_desc, spec_rename }, /* rename */ 181 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 182 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 183 { &vop_symlink_desc, spec_symlink }, /* symlink */ 184 { &vop_readdir_desc, spec_readdir }, /* readdir */ 185 { &vop_readlink_desc, spec_readlink }, /* readlink */ 186 { &vop_abortop_desc, spec_abortop }, /* abortop */ 187 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 188 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 189 { &vop_lock_desc, ufs_lock }, /* lock */ 190 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 191 { &vop_bmap_desc, spec_bmap }, /* bmap */ 192 { &vop_strategy_desc, spec_strategy }, /* strategy */ 193 { &vop_print_desc, ufs_print }, /* print */ 194 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 195 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 196 { &vop_advlock_desc, spec_advlock }, /* advlock */ 197 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 198 { &vop_getpages_desc, spec_getpages }, /* getpages */ 199 { &vop_putpages_desc, spec_putpages }, /* putpages */ 200 { NULL, NULL } 201 }; 202 const struct vnodeopv_desc lfs_specop_opv_desc = 203 { &lfs_specop_p, lfs_specop_entries }; 204 205 int (**lfs_fifoop_p)(void *); 206 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 207 { &vop_default_desc, vn_default_error }, 208 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 209 { &vop_create_desc, fifo_create }, /* create */ 210 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 211 { &vop_open_desc, fifo_open }, /* open */ 212 { &vop_close_desc, lfsfifo_close }, /* close */ 213 { &vop_access_desc, ufs_access }, /* access */ 214 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 215 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 216 { &vop_read_desc, ufsfifo_read }, /* read */ 217 { &vop_write_desc, ufsfifo_write }, /* write */ 218 { &vop_lease_desc, fifo_lease_check }, /* lease */ 219 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 220 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 221 { &vop_poll_desc, fifo_poll }, /* poll */ 222 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 223 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 224 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 225 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 226 { &vop_seek_desc, fifo_seek }, /* seek */ 227 { &vop_remove_desc, fifo_remove }, /* remove */ 228 { &vop_link_desc, fifo_link }, /* link */ 229 { &vop_rename_desc, fifo_rename }, /* rename */ 230 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 231 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 232 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 233 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 234 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 235 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 236 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 237 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 238 { &vop_lock_desc, ufs_lock }, /* lock */ 239 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 240 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 241 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 242 { &vop_print_desc, ufs_print }, /* print */ 243 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 244 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 245 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 246 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 247 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 248 { NULL, NULL } 249 }; 250 const struct vnodeopv_desc lfs_fifoop_opv_desc = 251 { &lfs_fifoop_p, lfs_fifoop_entries }; 252 253 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int); 254 255 #define LFS_READWRITE 256 #include <ufs/ufs/ufs_readwrite.c> 257 #undef LFS_READWRITE 258 259 /* 260 * Synch an open file. 261 */ 262 /* ARGSUSED */ 263 int 264 lfs_fsync(void *v) 265 { 266 struct vop_fsync_args /* { 267 struct vnode *a_vp; 268 struct ucred *a_cred; 269 int a_flags; 270 off_t offlo; 271 off_t offhi; 272 struct lwp *a_l; 273 } */ *ap = v; 274 struct vnode *vp = ap->a_vp; 275 int error, wait; 276 277 /* If we're mounted read-only, don't try to sync. */ 278 if (VTOI(vp)->i_lfs->lfs_ronly) 279 return 0; 280 281 /* 282 * Trickle sync checks for need to do a checkpoint after possible 283 * activity from the pagedaemon. 284 */ 285 if (ap->a_flags & FSYNC_LAZY) { 286 simple_lock(&lfs_subsys_lock); 287 wakeup(&lfs_writer_daemon); 288 simple_unlock(&lfs_subsys_lock); 289 return 0; 290 } 291 292 wait = (ap->a_flags & FSYNC_WAIT); 293 simple_lock(&vp->v_interlock); 294 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 295 round_page(ap->a_offhi), 296 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 297 if (error) 298 return error; 299 error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 300 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 301 int l = 0; 302 error = VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE, 303 ap->a_l->l_proc->p_ucred, ap->a_l); 304 } 305 if (wait && !VPISEMPTY(vp)) 306 LFS_SET_UINO(VTOI(vp), IN_MODIFIED); 307 308 return error; 309 } 310 311 /* 312 * Take IN_ADIROP off, then call ufs_inactive. 313 */ 314 int 315 lfs_inactive(void *v) 316 { 317 struct vop_inactive_args /* { 318 struct vnode *a_vp; 319 struct lwp *a_l; 320 } */ *ap = v; 321 322 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 323 324 lfs_unmark_vnode(ap->a_vp); 325 326 /* 327 * The Ifile is only ever inactivated on unmount. 328 * Streamline this process by not giving it more dirty blocks. 329 */ 330 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 331 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 332 VOP_UNLOCK(ap->a_vp, 0); 333 return 0; 334 } 335 336 return ufs_inactive(v); 337 } 338 339 /* 340 * These macros are used to bracket UFS directory ops, so that we can 341 * identify all the pages touched during directory ops which need to 342 * be ordered and flushed atomically, so that they may be recovered. 343 * 344 * Because we have to mark nodes VDIROP in order to prevent 345 * the cache from reclaiming them while a dirop is in progress, we must 346 * also manage the number of nodes so marked (otherwise we can run out). 347 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 348 * is decremented during segment write, when VDIROP is taken off. 349 */ 350 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 351 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 352 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 353 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 354 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 355 static int lfs_set_dirop(struct vnode *, struct vnode *); 356 357 static int 358 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 359 { 360 struct lfs *fs; 361 int error; 362 363 KASSERT(VOP_ISLOCKED(dvp)); 364 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 365 366 fs = VTOI(dvp)->i_lfs; 367 368 ASSERT_NO_SEGLOCK(fs); 369 /* 370 * LFS_NRESERVE calculates direct and indirect blocks as well 371 * as an inode block; an overestimate in most cases. 372 */ 373 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 374 return (error); 375 376 restart: 377 simple_lock(&fs->lfs_interlock); 378 if (fs->lfs_dirops == 0) { 379 simple_unlock(&fs->lfs_interlock); 380 lfs_check(dvp, LFS_UNUSED_LBN, 0); 381 simple_lock(&fs->lfs_interlock); 382 } 383 while (fs->lfs_writer) 384 ltsleep(&fs->lfs_dirops, (PRIBIO + 1), "lfs_sdirop", 0, 385 &fs->lfs_interlock); 386 simple_lock(&lfs_subsys_lock); 387 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 388 wakeup(&lfs_writer_daemon); 389 simple_unlock(&lfs_subsys_lock); 390 simple_unlock(&fs->lfs_interlock); 391 preempt(1); 392 goto restart; 393 } 394 395 if (lfs_dirvcount > LFS_MAX_DIROP) { 396 simple_unlock(&fs->lfs_interlock); 397 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 398 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 399 if ((error = ltsleep(&lfs_dirvcount, 400 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 401 &lfs_subsys_lock)) != 0) { 402 goto unreserve; 403 } 404 goto restart; 405 } 406 simple_unlock(&lfs_subsys_lock); 407 408 ++fs->lfs_dirops; 409 fs->lfs_doifile = 1; 410 simple_unlock(&fs->lfs_interlock); 411 412 /* Hold a reference so SET_ENDOP will be happy */ 413 vref(dvp); 414 if (vp) { 415 vref(vp); 416 MARK_VNODE(vp); 417 } 418 419 MARK_VNODE(dvp); 420 return 0; 421 422 unreserve: 423 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 424 return error; 425 } 426 427 /* 428 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 429 * in getnewvnode(), if we have a stacked filesystem mounted on top 430 * of us. 431 * 432 * NB: this means we have to clear the new vnodes on error. Fortunately 433 * SET_ENDOP is there to do that for us. 434 */ 435 static int 436 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 437 { 438 int error; 439 struct lfs *fs; 440 441 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 442 ASSERT_NO_SEGLOCK(fs); 443 if (fs->lfs_ronly) 444 return EROFS; 445 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 446 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 447 dvp, error)); 448 return error; 449 } 450 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 451 if (vpp) { 452 ungetnewvnode(*vpp); 453 *vpp = NULL; 454 } 455 return error; 456 } 457 return 0; 458 } 459 460 #define SET_ENDOP_BASE(fs, dvp, str) \ 461 do { \ 462 simple_lock(&(fs)->lfs_interlock); \ 463 --(fs)->lfs_dirops; \ 464 if (!(fs)->lfs_dirops) { \ 465 if ((fs)->lfs_nadirop) { \ 466 panic("SET_ENDOP: %s: no dirops but " \ 467 " nadirop=%d", (str), \ 468 (fs)->lfs_nadirop); \ 469 } \ 470 wakeup(&(fs)->lfs_writer); \ 471 simple_unlock(&(fs)->lfs_interlock); \ 472 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 473 } else \ 474 simple_unlock(&(fs)->lfs_interlock); \ 475 } while(0) 476 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 477 do { \ 478 UNMARK_VNODE(dvp); \ 479 if (nvpp && *nvpp) \ 480 UNMARK_VNODE(*nvpp); \ 481 /* Check for error return to stem vnode leakage */ \ 482 if (nvpp && *nvpp && !((*nvpp)->v_flag & VDIROP)) \ 483 ungetnewvnode(*(nvpp)); \ 484 SET_ENDOP_BASE((fs), (dvp), (str)); \ 485 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 486 vrele(dvp); \ 487 } while(0) 488 #define SET_ENDOP_CREATE_AP(ap, str) \ 489 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 490 (ap)->a_vpp, (str)) 491 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 492 do { \ 493 UNMARK_VNODE(dvp); \ 494 if (ovp) \ 495 UNMARK_VNODE(ovp); \ 496 SET_ENDOP_BASE((fs), (dvp), (str)); \ 497 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 498 vrele(dvp); \ 499 if (ovp) \ 500 vrele(ovp); \ 501 } while(0) 502 503 void 504 lfs_mark_vnode(struct vnode *vp) 505 { 506 struct inode *ip = VTOI(vp); 507 struct lfs *fs = ip->i_lfs; 508 509 simple_lock(&fs->lfs_interlock); 510 if (!(ip->i_flag & IN_ADIROP)) { 511 if (!(vp->v_flag & VDIROP)) { 512 (void)lfs_vref(vp); 513 simple_lock(&lfs_subsys_lock); 514 ++lfs_dirvcount; 515 simple_unlock(&lfs_subsys_lock); 516 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 517 vp->v_flag |= VDIROP; 518 } 519 ++fs->lfs_nadirop; 520 ip->i_flag |= IN_ADIROP; 521 } else 522 KASSERT(vp->v_flag & VDIROP); 523 simple_unlock(&fs->lfs_interlock); 524 } 525 526 void 527 lfs_unmark_vnode(struct vnode *vp) 528 { 529 struct inode *ip = VTOI(vp); 530 531 if (ip && (ip->i_flag & IN_ADIROP)) { 532 KASSERT(vp->v_flag & VDIROP); 533 simple_lock(&ip->i_lfs->lfs_interlock); 534 --ip->i_lfs->lfs_nadirop; 535 simple_unlock(&ip->i_lfs->lfs_interlock); 536 ip->i_flag &= ~IN_ADIROP; 537 } 538 } 539 540 int 541 lfs_symlink(void *v) 542 { 543 struct vop_symlink_args /* { 544 struct vnode *a_dvp; 545 struct vnode **a_vpp; 546 struct componentname *a_cnp; 547 struct vattr *a_vap; 548 char *a_target; 549 } */ *ap = v; 550 int error; 551 552 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 553 vput(ap->a_dvp); 554 return error; 555 } 556 error = ufs_symlink(ap); 557 SET_ENDOP_CREATE_AP(ap, "symlink"); 558 return (error); 559 } 560 561 int 562 lfs_mknod(void *v) 563 { 564 struct vop_mknod_args /* { 565 struct vnode *a_dvp; 566 struct vnode **a_vpp; 567 struct componentname *a_cnp; 568 struct vattr *a_vap; 569 } */ *ap = v; 570 struct vattr *vap = ap->a_vap; 571 struct vnode **vpp = ap->a_vpp; 572 struct inode *ip; 573 int error; 574 struct mount *mp; 575 ino_t ino; 576 577 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 578 vput(ap->a_dvp); 579 return error; 580 } 581 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 582 ap->a_dvp, vpp, ap->a_cnp); 583 584 /* Either way we're done with the dirop at this point */ 585 SET_ENDOP_CREATE_AP(ap, "mknod"); 586 587 if (error) 588 return (error); 589 590 ip = VTOI(*vpp); 591 mp = (*vpp)->v_mount; 592 ino = ip->i_number; 593 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 594 if (vap->va_rdev != VNOVAL) { 595 /* 596 * Want to be able to use this to make badblock 597 * inodes, so don't truncate the dev number. 598 */ 599 #if 0 600 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 601 UFS_MPNEEDSWAP((*vpp)->v_mount)); 602 #else 603 ip->i_ffs1_rdev = vap->va_rdev; 604 #endif 605 } 606 607 /* 608 * Call fsync to write the vnode so that we don't have to deal with 609 * flushing it when it's marked VDIROP|VXLOCK. 610 * 611 * XXX KS - If we can't flush we also can't call vgone(), so must 612 * return. But, that leaves this vnode in limbo, also not good. 613 * Can this ever happen (barring hardware failure)? 614 */ 615 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 616 curlwp)) != 0) { 617 panic("lfs_mknod: couldn't fsync (ino %llu)", 618 (unsigned long long)ino); 619 /* return (error); */ 620 } 621 /* 622 * Remove vnode so that it will be reloaded by VFS_VGET and 623 * checked to see if it is an alias of an existing entry in 624 * the inode cache. 625 */ 626 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 627 628 VOP_UNLOCK(*vpp, 0); 629 lfs_vunref(*vpp); 630 (*vpp)->v_type = VNON; 631 vgone(*vpp); 632 error = VFS_VGET(mp, ino, vpp); 633 634 if (error != 0) { 635 *vpp = NULL; 636 return (error); 637 } 638 return (0); 639 } 640 641 int 642 lfs_create(void *v) 643 { 644 struct vop_create_args /* { 645 struct vnode *a_dvp; 646 struct vnode **a_vpp; 647 struct componentname *a_cnp; 648 struct vattr *a_vap; 649 } */ *ap = v; 650 int error; 651 652 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 653 vput(ap->a_dvp); 654 return error; 655 } 656 error = ufs_create(ap); 657 SET_ENDOP_CREATE_AP(ap, "create"); 658 return (error); 659 } 660 661 int 662 lfs_mkdir(void *v) 663 { 664 struct vop_mkdir_args /* { 665 struct vnode *a_dvp; 666 struct vnode **a_vpp; 667 struct componentname *a_cnp; 668 struct vattr *a_vap; 669 } */ *ap = v; 670 int error; 671 672 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 673 vput(ap->a_dvp); 674 return error; 675 } 676 error = ufs_mkdir(ap); 677 SET_ENDOP_CREATE_AP(ap, "mkdir"); 678 return (error); 679 } 680 681 int 682 lfs_remove(void *v) 683 { 684 struct vop_remove_args /* { 685 struct vnode *a_dvp; 686 struct vnode *a_vp; 687 struct componentname *a_cnp; 688 } */ *ap = v; 689 struct vnode *dvp, *vp; 690 int error; 691 692 dvp = ap->a_dvp; 693 vp = ap->a_vp; 694 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 695 if (dvp == vp) 696 vrele(vp); 697 else 698 vput(vp); 699 vput(dvp); 700 return error; 701 } 702 error = ufs_remove(ap); 703 SET_ENDOP_REMOVE(VTOI(dvp)->i_lfs, dvp, ap->a_vp, "remove"); 704 return (error); 705 } 706 707 int 708 lfs_rmdir(void *v) 709 { 710 struct vop_rmdir_args /* { 711 struct vnodeop_desc *a_desc; 712 struct vnode *a_dvp; 713 struct vnode *a_vp; 714 struct componentname *a_cnp; 715 } */ *ap = v; 716 struct vnode *vp; 717 int error; 718 719 vp = ap->a_vp; 720 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 721 vrele(ap->a_dvp); 722 if (ap->a_vp != ap->a_dvp) 723 VOP_UNLOCK(ap->a_dvp, 0); 724 vput(vp); 725 return error; 726 } 727 error = ufs_rmdir(ap); 728 SET_ENDOP_REMOVE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 729 return (error); 730 } 731 732 int 733 lfs_link(void *v) 734 { 735 struct vop_link_args /* { 736 struct vnode *a_dvp; 737 struct vnode *a_vp; 738 struct componentname *a_cnp; 739 } */ *ap = v; 740 int error; 741 struct vnode **vpp = NULL; 742 743 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 744 vput(ap->a_dvp); 745 return error; 746 } 747 error = ufs_link(ap); 748 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 749 return (error); 750 } 751 752 int 753 lfs_rename(void *v) 754 { 755 struct vop_rename_args /* { 756 struct vnode *a_fdvp; 757 struct vnode *a_fvp; 758 struct componentname *a_fcnp; 759 struct vnode *a_tdvp; 760 struct vnode *a_tvp; 761 struct componentname *a_tcnp; 762 } */ *ap = v; 763 struct vnode *tvp, *fvp, *tdvp, *fdvp; 764 struct componentname *tcnp, *fcnp; 765 int error; 766 struct lfs *fs; 767 768 fs = VTOI(ap->a_fdvp)->i_lfs; 769 tvp = ap->a_tvp; 770 tdvp = ap->a_tdvp; 771 tcnp = ap->a_tcnp; 772 fvp = ap->a_fvp; 773 fdvp = ap->a_fdvp; 774 fcnp = ap->a_fcnp; 775 776 /* 777 * Check for cross-device rename. 778 * If it is, we don't want to set dirops, just error out. 779 * (In particular note that MARK_VNODE(tdvp) will DTWT on 780 * a cross-device rename.) 781 * 782 * Copied from ufs_rename. 783 */ 784 if ((fvp->v_mount != tdvp->v_mount) || 785 (tvp && (fvp->v_mount != tvp->v_mount))) { 786 error = EXDEV; 787 goto errout; 788 } 789 790 /* 791 * Check to make sure we're not renaming a vnode onto itself 792 * (deleting a hard link by renaming one name onto another); 793 * if we are we can't recursively call VOP_REMOVE since that 794 * would leave us with an unaccounted-for number of live dirops. 795 * 796 * Inline the relevant section of ufs_rename here, *before* 797 * calling SET_DIROP_REMOVE. 798 */ 799 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 800 (VTOI(tdvp)->i_flags & APPEND))) { 801 error = EPERM; 802 goto errout; 803 } 804 if (fvp == tvp) { 805 if (fvp->v_type == VDIR) { 806 error = EINVAL; 807 goto errout; 808 } 809 810 /* Release destination completely. */ 811 VOP_ABORTOP(tdvp, tcnp); 812 vput(tdvp); 813 vput(tvp); 814 815 /* Delete source. */ 816 vrele(fvp); 817 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 818 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 819 fcnp->cn_nameiop = DELETE; 820 if ((error = relookup(fdvp, &fvp, fcnp))){ 821 /* relookup blew away fdvp */ 822 return (error); 823 } 824 return (VOP_REMOVE(fdvp, fvp, fcnp)); 825 } 826 827 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 828 goto errout; 829 MARK_VNODE(fdvp); 830 MARK_VNODE(fvp); 831 832 error = ufs_rename(ap); 833 UNMARK_VNODE(fdvp); 834 UNMARK_VNODE(fvp); 835 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 836 return (error); 837 838 errout: 839 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 840 if (tdvp == tvp) 841 vrele(tdvp); 842 else 843 vput(tdvp); 844 if (tvp) 845 vput(tvp); 846 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 847 vrele(fdvp); 848 vrele(fvp); 849 return (error); 850 } 851 852 /* XXX hack to avoid calling ITIMES in getattr */ 853 int 854 lfs_getattr(void *v) 855 { 856 struct vop_getattr_args /* { 857 struct vnode *a_vp; 858 struct vattr *a_vap; 859 struct ucred *a_cred; 860 struct lwp *a_l; 861 } */ *ap = v; 862 struct vnode *vp = ap->a_vp; 863 struct inode *ip = VTOI(vp); 864 struct vattr *vap = ap->a_vap; 865 struct lfs *fs = ip->i_lfs; 866 /* 867 * Copy from inode table 868 */ 869 vap->va_fsid = ip->i_dev; 870 vap->va_fileid = ip->i_number; 871 vap->va_mode = ip->i_mode & ~IFMT; 872 vap->va_nlink = ip->i_nlink; 873 vap->va_uid = ip->i_uid; 874 vap->va_gid = ip->i_gid; 875 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 876 vap->va_size = vp->v_size; 877 vap->va_atime.tv_sec = ip->i_ffs1_atime; 878 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 879 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 880 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 881 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 882 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 883 vap->va_flags = ip->i_flags; 884 vap->va_gen = ip->i_gen; 885 /* this doesn't belong here */ 886 if (vp->v_type == VBLK) 887 vap->va_blocksize = BLKDEV_IOSIZE; 888 else if (vp->v_type == VCHR) 889 vap->va_blocksize = MAXBSIZE; 890 else 891 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 892 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 893 vap->va_type = vp->v_type; 894 vap->va_filerev = ip->i_modrev; 895 return (0); 896 } 897 898 /* 899 * Check to make sure the inode blocks won't choke the buffer 900 * cache, then call ufs_setattr as usual. 901 */ 902 int 903 lfs_setattr(void *v) 904 { 905 struct vop_setattr_args /* { 906 struct vnode *a_vp; 907 struct vattr *a_vap; 908 struct ucred *a_cred; 909 struct lwp *a_l; 910 } */ *ap = v; 911 struct vnode *vp = ap->a_vp; 912 913 lfs_check(vp, LFS_UNUSED_LBN, 0); 914 return ufs_setattr(v); 915 } 916 917 /* 918 * Close called 919 * 920 * XXX -- we were using ufs_close, but since it updates the 921 * times on the inode, we might need to bump the uinodes 922 * count. 923 */ 924 /* ARGSUSED */ 925 int 926 lfs_close(void *v) 927 { 928 struct vop_close_args /* { 929 struct vnode *a_vp; 930 int a_fflag; 931 struct ucred *a_cred; 932 struct lwp *a_l; 933 } */ *ap = v; 934 struct vnode *vp = ap->a_vp; 935 struct inode *ip = VTOI(vp); 936 937 if (vp == ip->i_lfs->lfs_ivnode && 938 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 939 return 0; 940 941 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 942 LFS_ITIMES(ip, NULL, NULL, NULL); 943 } 944 return (0); 945 } 946 947 /* 948 * Close wrapper for special devices. 949 * 950 * Update the times on the inode then do device close. 951 */ 952 int 953 lfsspec_close(void *v) 954 { 955 struct vop_close_args /* { 956 struct vnode *a_vp; 957 int a_fflag; 958 struct ucred *a_cred; 959 struct lwp *a_l; 960 } */ *ap = v; 961 struct vnode *vp; 962 struct inode *ip; 963 964 vp = ap->a_vp; 965 ip = VTOI(vp); 966 if (vp->v_usecount > 1) { 967 LFS_ITIMES(ip, NULL, NULL, NULL); 968 } 969 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 970 } 971 972 /* 973 * Close wrapper for fifo's. 974 * 975 * Update the times on the inode then do device close. 976 */ 977 int 978 lfsfifo_close(void *v) 979 { 980 struct vop_close_args /* { 981 struct vnode *a_vp; 982 int a_fflag; 983 struct ucred *a_cred; 984 struct lwp *a_l; 985 } */ *ap = v; 986 struct vnode *vp; 987 struct inode *ip; 988 989 vp = ap->a_vp; 990 ip = VTOI(vp); 991 if (ap->a_vp->v_usecount > 1) { 992 LFS_ITIMES(ip, NULL, NULL, NULL); 993 } 994 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 995 } 996 997 /* 998 * Reclaim an inode so that it can be used for other purposes. 999 */ 1000 1001 int 1002 lfs_reclaim(void *v) 1003 { 1004 struct vop_reclaim_args /* { 1005 struct vnode *a_vp; 1006 struct lwp *a_l; 1007 } */ *ap = v; 1008 struct vnode *vp = ap->a_vp; 1009 struct inode *ip = VTOI(vp); 1010 int error; 1011 1012 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1013 1014 LFS_CLR_UINO(ip, IN_ALLMOD); 1015 if ((error = ufs_reclaim(vp, ap->a_l))) 1016 return (error); 1017 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1018 lfs_deregister_all(vp); 1019 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1020 ip->inode_ext.lfs = NULL; 1021 pool_put(&lfs_inode_pool, vp->v_data); 1022 vp->v_data = NULL; 1023 return (0); 1024 } 1025 1026 /* 1027 * Read a block from a storage device. 1028 * In order to avoid reading blocks that are in the process of being 1029 * written by the cleaner---and hence are not mutexed by the normal 1030 * buffer cache / page cache mechanisms---check for collisions before 1031 * reading. 1032 * 1033 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1034 * the active cleaner test. 1035 * 1036 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1037 */ 1038 int 1039 lfs_strategy(void *v) 1040 { 1041 struct vop_strategy_args /* { 1042 struct vnode *a_vp; 1043 struct buf *a_bp; 1044 } */ *ap = v; 1045 struct buf *bp; 1046 struct lfs *fs; 1047 struct vnode *vp; 1048 struct inode *ip; 1049 daddr_t tbn; 1050 int i, sn, error, slept; 1051 1052 bp = ap->a_bp; 1053 vp = ap->a_vp; 1054 ip = VTOI(vp); 1055 fs = ip->i_lfs; 1056 1057 /* lfs uses its strategy routine only for read */ 1058 KASSERT(bp->b_flags & B_READ); 1059 1060 if (vp->v_type == VBLK || vp->v_type == VCHR) 1061 panic("lfs_strategy: spec"); 1062 KASSERT(bp->b_bcount != 0); 1063 if (bp->b_blkno == bp->b_lblkno) { 1064 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1065 NULL); 1066 if (error) { 1067 bp->b_error = error; 1068 bp->b_flags |= B_ERROR; 1069 biodone(bp); 1070 return (error); 1071 } 1072 if ((long)bp->b_blkno == -1) /* no valid data */ 1073 clrbuf(bp); 1074 } 1075 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1076 biodone(bp); 1077 return (0); 1078 } 1079 1080 slept = 1; 1081 simple_lock(&fs->lfs_interlock); 1082 while (slept && fs->lfs_seglock) { 1083 simple_unlock(&fs->lfs_interlock); 1084 /* 1085 * Look through list of intervals. 1086 * There will only be intervals to look through 1087 * if the cleaner holds the seglock. 1088 * Since the cleaner is synchronous, we can trust 1089 * the list of intervals to be current. 1090 */ 1091 tbn = dbtofsb(fs, bp->b_blkno); 1092 sn = dtosn(fs, tbn); 1093 slept = 0; 1094 for (i = 0; i < fs->lfs_cleanind; i++) { 1095 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1096 tbn >= fs->lfs_cleanint[i]) { 1097 DLOG((DLOG_CLEAN, 1098 "lfs_strategy: ino %d lbn %" PRId64 1099 " ind %d sn %d fsb %" PRIx32 1100 " given sn %d fsb %" PRIx64 "\n", 1101 ip->i_number, bp->b_lblkno, i, 1102 dtosn(fs, fs->lfs_cleanint[i]), 1103 fs->lfs_cleanint[i], sn, tbn)); 1104 DLOG((DLOG_CLEAN, 1105 "lfs_strategy: sleeping on ino %d lbn %" 1106 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1107 simple_lock(&fs->lfs_interlock); 1108 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1109 /* Cleaner can't wait for itself */ 1110 ltsleep(&fs->lfs_iocount, 1111 (PRIBIO + 1) | PNORELOCK, 1112 "clean2", 0, 1113 &fs->lfs_interlock); 1114 slept = 1; 1115 break; 1116 } else if (fs->lfs_seglock) { 1117 ltsleep(&fs->lfs_seglock, 1118 (PRIBIO + 1) | PNORELOCK, 1119 "clean1", 0, 1120 &fs->lfs_interlock); 1121 slept = 1; 1122 break; 1123 } 1124 simple_unlock(&fs->lfs_interlock); 1125 } 1126 } 1127 simple_lock(&fs->lfs_interlock); 1128 } 1129 simple_unlock(&fs->lfs_interlock); 1130 1131 vp = ip->i_devvp; 1132 VOP_STRATEGY(vp, bp); 1133 return (0); 1134 } 1135 1136 static void 1137 lfs_flush_dirops(struct lfs *fs) 1138 { 1139 struct inode *ip, *nip; 1140 struct vnode *vp; 1141 extern int lfs_dostats; 1142 struct segment *sp; 1143 int needunlock; 1144 1145 ASSERT_MAYBE_SEGLOCK(fs); 1146 1147 if (fs->lfs_ronly) 1148 return; 1149 1150 simple_lock(&fs->lfs_interlock); 1151 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1152 simple_unlock(&fs->lfs_interlock); 1153 return; 1154 } else 1155 simple_unlock(&fs->lfs_interlock); 1156 1157 if (lfs_dostats) 1158 ++lfs_stats.flush_invoked; 1159 1160 /* 1161 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1162 * Technically this is a checkpoint (the on-disk state is valid) 1163 * even though we are leaving out all the file data. 1164 */ 1165 lfs_imtime(fs); 1166 lfs_seglock(fs, SEGM_CKP); 1167 sp = fs->lfs_sp; 1168 1169 /* 1170 * lfs_writevnodes, optimized to get dirops out of the way. 1171 * Only write dirops, and don't flush files' pages, only 1172 * blocks from the directories. 1173 * 1174 * We don't need to vref these files because they are 1175 * dirops and so hold an extra reference until the 1176 * segunlock clears them of that status. 1177 * 1178 * We don't need to check for IN_ADIROP because we know that 1179 * no dirops are active. 1180 * 1181 */ 1182 simple_lock(&fs->lfs_interlock); 1183 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1184 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1185 simple_unlock(&fs->lfs_interlock); 1186 vp = ITOV(ip); 1187 1188 /* 1189 * All writes to directories come from dirops; all 1190 * writes to files' direct blocks go through the page 1191 * cache, which we're not touching. Reads to files 1192 * and/or directories will not be affected by writing 1193 * directory blocks inodes and file inodes. So we don't 1194 * really need to lock. If we don't lock, though, 1195 * make sure that we don't clear IN_MODIFIED 1196 * unnecessarily. 1197 */ 1198 if (vp->v_flag & (VXLOCK | VFREEING)) { 1199 simple_lock(&fs->lfs_interlock); 1200 continue; 1201 } 1202 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT) == 0) { 1203 needunlock = 1; 1204 } else { 1205 DLOG((DLOG_VNODE, "lfs_flush_dirops: flushing locked ino %d\n", 1206 VTOI(vp)->i_number)); 1207 needunlock = 0; 1208 } 1209 if (vp->v_type != VREG && 1210 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1211 lfs_writefile(fs, sp, vp); 1212 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1213 !(ip->i_flag & IN_ALLMOD)) { 1214 LFS_SET_UINO(ip, IN_MODIFIED); 1215 } 1216 } 1217 (void) lfs_writeinode(fs, sp, ip); 1218 if (needunlock) 1219 VOP_UNLOCK(vp, 0); 1220 else 1221 LFS_SET_UINO(ip, IN_MODIFIED); 1222 simple_lock(&fs->lfs_interlock); 1223 } 1224 simple_unlock(&fs->lfs_interlock); 1225 /* We've written all the dirops there are */ 1226 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1227 lfs_finalize_fs_seguse(fs); 1228 (void) lfs_writeseg(fs, sp); 1229 lfs_segunlock(fs); 1230 } 1231 1232 /* 1233 * Flush all vnodes for which the pagedaemon has requested pageouts. 1234 * Skip over any files that are marked VDIROP (since lfs_flush_dirop() 1235 * has just run, this would be an error). If we have to skip a vnode 1236 * for any reason, just skip it; if we have to wait for the cleaner, 1237 * abort. The writer daemon will call us again later. 1238 */ 1239 void 1240 lfs_flush_pchain(struct lfs *fs) 1241 { 1242 struct inode *ip, *nip; 1243 struct vnode *vp; 1244 extern int lfs_dostats; 1245 struct segment *sp; 1246 int error; 1247 1248 ASSERT_NO_SEGLOCK(fs); 1249 1250 if (fs->lfs_ronly) 1251 return; 1252 1253 simple_lock(&fs->lfs_interlock); 1254 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1255 simple_unlock(&fs->lfs_interlock); 1256 return; 1257 } else 1258 simple_unlock(&fs->lfs_interlock); 1259 1260 /* Get dirops out of the way */ 1261 lfs_flush_dirops(fs); 1262 1263 if (lfs_dostats) 1264 ++lfs_stats.flush_invoked; 1265 1266 /* 1267 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1268 */ 1269 lfs_imtime(fs); 1270 lfs_seglock(fs, 0); 1271 sp = fs->lfs_sp; 1272 1273 /* 1274 * lfs_writevnodes, optimized to clear pageout requests. 1275 * Only write non-dirop files that are in the pageout queue. 1276 * We're very conservative about what we write; we want to be 1277 * fast and async. 1278 */ 1279 simple_lock(&fs->lfs_interlock); 1280 top: 1281 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1282 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1283 vp = ITOV(ip); 1284 1285 if (!(ip->i_flags & IN_PAGING)) 1286 goto top; 1287 1288 if (vp->v_flag & (VXLOCK|VDIROP)) 1289 continue; 1290 if (vp->v_type != VREG) 1291 continue; 1292 if (lfs_vref(vp)) 1293 continue; 1294 simple_unlock(&fs->lfs_interlock); 1295 1296 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 1297 lfs_vunref(vp); 1298 simple_lock(&fs->lfs_interlock); 1299 continue; 1300 } 1301 1302 error = lfs_writefile(fs, sp, vp); 1303 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1304 !(ip->i_flag & IN_ALLMOD)) { 1305 LFS_SET_UINO(ip, IN_MODIFIED); 1306 } 1307 (void) lfs_writeinode(fs, sp, ip); 1308 1309 VOP_UNLOCK(vp, 0); 1310 lfs_vunref(vp); 1311 1312 if (error == EAGAIN) { 1313 lfs_writeseg(fs, sp); 1314 simple_lock(&fs->lfs_interlock); 1315 break; 1316 } 1317 simple_lock(&fs->lfs_interlock); 1318 } 1319 simple_unlock(&fs->lfs_interlock); 1320 (void) lfs_writeseg(fs, sp); 1321 lfs_segunlock(fs); 1322 } 1323 1324 /* 1325 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1326 */ 1327 int 1328 lfs_fcntl(void *v) 1329 { 1330 struct vop_fcntl_args /* { 1331 struct vnode *a_vp; 1332 u_long a_command; 1333 caddr_t a_data; 1334 int a_fflag; 1335 struct ucred *a_cred; 1336 struct lwp *a_l; 1337 } */ *ap = v; 1338 struct timeval *tvp; 1339 BLOCK_INFO *blkiov; 1340 CLEANERINFO *cip; 1341 SEGUSE *sup; 1342 int blkcnt, error, oclean; 1343 struct lfs_fcntl_markv blkvp; 1344 struct proc *p; 1345 fsid_t *fsidp; 1346 struct lfs *fs; 1347 struct buf *bp; 1348 fhandle_t *fhp; 1349 daddr_t off; 1350 1351 /* Only respect LFS fcntls on fs root or Ifile */ 1352 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1353 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1354 return ufs_fcntl(v); 1355 } 1356 1357 /* Avoid locking a draining lock */ 1358 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1359 return ESHUTDOWN; 1360 } 1361 1362 p = ap->a_l->l_proc; 1363 fs = VTOI(ap->a_vp)->i_lfs; 1364 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1365 1366 switch (ap->a_command) { 1367 case LFCNSEGWAITALL: 1368 case LFCNSEGWAITALL_COMPAT: 1369 fsidp = NULL; 1370 /* FALLSTHROUGH */ 1371 case LFCNSEGWAIT: 1372 case LFCNSEGWAIT_COMPAT: 1373 tvp = (struct timeval *)ap->a_data; 1374 simple_lock(&fs->lfs_interlock); 1375 ++fs->lfs_sleepers; 1376 simple_unlock(&fs->lfs_interlock); 1377 VOP_UNLOCK(ap->a_vp, 0); 1378 1379 error = lfs_segwait(fsidp, tvp); 1380 1381 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1382 simple_lock(&fs->lfs_interlock); 1383 if (--fs->lfs_sleepers == 0) 1384 wakeup(&fs->lfs_sleepers); 1385 simple_unlock(&fs->lfs_interlock); 1386 return error; 1387 1388 case LFCNBMAPV: 1389 case LFCNMARKV: 1390 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0) 1391 return (error); 1392 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1393 1394 blkcnt = blkvp.blkcnt; 1395 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1396 return (EINVAL); 1397 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1398 if ((error = copyin(blkvp.blkiov, blkiov, 1399 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1400 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1401 return error; 1402 } 1403 1404 simple_lock(&fs->lfs_interlock); 1405 ++fs->lfs_sleepers; 1406 simple_unlock(&fs->lfs_interlock); 1407 VOP_UNLOCK(ap->a_vp, 0); 1408 if (ap->a_command == LFCNBMAPV) 1409 error = lfs_bmapv(p, fsidp, blkiov, blkcnt); 1410 else /* LFCNMARKV */ 1411 error = lfs_markv(p, fsidp, blkiov, blkcnt); 1412 if (error == 0) 1413 error = copyout(blkiov, blkvp.blkiov, 1414 blkcnt * sizeof(BLOCK_INFO)); 1415 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1416 simple_lock(&fs->lfs_interlock); 1417 if (--fs->lfs_sleepers == 0) 1418 wakeup(&fs->lfs_sleepers); 1419 simple_unlock(&fs->lfs_interlock); 1420 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1421 return error; 1422 1423 case LFCNRECLAIM: 1424 /* 1425 * Flush dirops and write Ifile, allowing empty segments 1426 * to be immediately reclaimed. 1427 */ 1428 VOP_UNLOCK(ap->a_vp, 0); 1429 lfs_writer_enter(fs, "pndirop"); 1430 off = fs->lfs_offset; 1431 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1432 lfs_flush_dirops(fs); 1433 LFS_CLEANERINFO(cip, fs, bp); 1434 oclean = cip->clean; 1435 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1436 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1437 fs->lfs_sp->seg_flags |= SEGM_PROT; 1438 lfs_segunlock(fs); 1439 lfs_writer_leave(fs); 1440 1441 #ifdef DEBUG 1442 LFS_CLEANERINFO(cip, fs, bp); 1443 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1444 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1445 fs->lfs_offset - off, cip->clean - oclean, 1446 fs->lfs_activesb)); 1447 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1448 #endif 1449 1450 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1451 return 0; 1452 1453 case LFCNIFILEFH: 1454 /* Return the filehandle of the Ifile */ 1455 if ((error = suser(ap->a_l->l_proc->p_ucred, &ap->a_l->l_proc->p_acflag)) != 0) 1456 return (error); 1457 fhp = (struct fhandle *)ap->a_data; 1458 fhp->fh_fsid = *fsidp; 1459 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid)); 1460 1461 case LFCNREWIND: 1462 /* Move lfs_offset to the lowest-numbered segment */ 1463 return lfs_rewind(fs, *(int *)ap->a_data); 1464 1465 case LFCNINVAL: 1466 /* Mark a segment SEGUSE_INVAL */ 1467 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1468 if (sup->su_nbytes > 0) { 1469 brelse(bp); 1470 lfs_unset_inval_all(fs); 1471 return EBUSY; 1472 } 1473 sup->su_flags |= SEGUSE_INVAL; 1474 VOP_BWRITE(bp); 1475 return 0; 1476 1477 case LFCNRESIZE: 1478 /* Resize the filesystem */ 1479 return lfs_resize_fs(fs, *(int *)ap->a_data); 1480 1481 case LFCNWRAPSTOP: 1482 /* 1483 * Hold lfs_newseg at segment 0; sleep until the filesystem 1484 * wraps around. For debugging purposes, so an external 1485 * agent can log every segment in the filesystem as it 1486 * was written, and we can regression-test checkpoint 1487 * validity in the general case. 1488 */ 1489 VOP_UNLOCK(ap->a_vp, 0); 1490 simple_lock(&fs->lfs_interlock); 1491 fs->lfs_nowrap = 1; 1492 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER | PNORELOCK, 1493 "segwrap", 0, &fs->lfs_interlock); 1494 if (error) { 1495 fs->lfs_nowrap = 0; 1496 wakeup(&fs->lfs_nowrap); 1497 } 1498 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1499 return 0; 1500 1501 case LFCNWRAPGO: 1502 /* 1503 * Having done its work, the agent wakes up the writer. 1504 * It sleeps until a new segment is selected. 1505 */ 1506 VOP_UNLOCK(ap->a_vp, 0); 1507 simple_lock(&fs->lfs_interlock); 1508 fs->lfs_nowrap = 0; 1509 wakeup(&fs->lfs_nowrap); 1510 ltsleep(&fs->lfs_nextseg, PCATCH | PUSER | PNORELOCK, 1511 "segment", 0, &fs->lfs_interlock); 1512 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1513 return 0; 1514 1515 default: 1516 return ufs_fcntl(v); 1517 } 1518 return 0; 1519 } 1520 1521 int 1522 lfs_getpages(void *v) 1523 { 1524 struct vop_getpages_args /* { 1525 struct vnode *a_vp; 1526 voff_t a_offset; 1527 struct vm_page **a_m; 1528 int *a_count; 1529 int a_centeridx; 1530 vm_prot_t a_access_type; 1531 int a_advice; 1532 int a_flags; 1533 } */ *ap = v; 1534 1535 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1536 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1537 return EPERM; 1538 } 1539 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1540 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1541 } 1542 1543 /* 1544 * we're relying on the fact that genfs_getpages() always read in 1545 * entire filesystem blocks. 1546 */ 1547 return genfs_getpages(v); 1548 } 1549 1550 /* 1551 * Make sure that for all pages in every block in the given range, 1552 * either all are dirty or all are clean. If any of the pages 1553 * we've seen so far are dirty, put the vnode on the paging chain, 1554 * and mark it IN_PAGING. 1555 * 1556 * If checkfirst != 0, don't check all the pages but return at the 1557 * first dirty page. 1558 */ 1559 static int 1560 check_dirty(struct lfs *fs, struct vnode *vp, 1561 off_t startoffset, off_t endoffset, off_t blkeof, 1562 int flags, int checkfirst) 1563 { 1564 int by_list; 1565 struct vm_page *curpg = NULL; /* XXX: gcc */ 1566 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1567 off_t soff = 0; /* XXX: gcc */ 1568 voff_t off; 1569 int i; 1570 int nonexistent; 1571 int any_dirty; /* number of dirty pages */ 1572 int dirty; /* number of dirty pages in a block */ 1573 int tdirty; 1574 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1575 int pagedaemon = (curproc == uvm.pagedaemon_proc); 1576 1577 ASSERT_MAYBE_SEGLOCK(fs); 1578 top: 1579 by_list = (vp->v_uobj.uo_npages <= 1580 ((endoffset - startoffset) >> PAGE_SHIFT) * 1581 UVM_PAGE_HASH_PENALTY); 1582 any_dirty = 0; 1583 1584 if (by_list) { 1585 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1586 } else { 1587 soff = startoffset; 1588 } 1589 while (by_list || soff < MIN(blkeof, endoffset)) { 1590 if (by_list) { 1591 /* 1592 * Find the first page in a block. Skip 1593 * blocks outside our area of interest or beyond 1594 * the end of file. 1595 */ 1596 if (pages_per_block > 1) { 1597 while (curpg && 1598 ((curpg->offset & fs->lfs_bmask) || 1599 curpg->offset >= vp->v_size || 1600 curpg->offset >= endoffset)) 1601 curpg = TAILQ_NEXT(curpg, listq); 1602 } 1603 if (curpg == NULL) 1604 break; 1605 soff = curpg->offset; 1606 } 1607 1608 /* 1609 * Mark all pages in extended range busy; find out if any 1610 * of them are dirty. 1611 */ 1612 nonexistent = dirty = 0; 1613 for (i = 0; i == 0 || i < pages_per_block; i++) { 1614 if (by_list && pages_per_block <= 1) { 1615 pgs[i] = pg = curpg; 1616 } else { 1617 off = soff + (i << PAGE_SHIFT); 1618 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1619 if (pg == NULL) { 1620 ++nonexistent; 1621 continue; 1622 } 1623 } 1624 KASSERT(pg != NULL); 1625 1626 /* 1627 * If we're holding the segment lock, we can deadlocked 1628 * against a process that has our page and is waiting 1629 * for the cleaner, while the cleaner waits for the 1630 * segment lock. Just bail in that case. 1631 */ 1632 if ((pg->flags & PG_BUSY) && 1633 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1634 if (by_list && i > 0) 1635 uvm_page_unbusy(pgs, i); 1636 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1637 return -1; 1638 } 1639 1640 while (pg->flags & PG_BUSY) { 1641 pg->flags |= PG_WANTED; 1642 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1643 "lfsput", 0); 1644 simple_lock(&vp->v_interlock); 1645 if (by_list) { 1646 if (i > 0) 1647 uvm_page_unbusy(pgs, i); 1648 goto top; 1649 } 1650 } 1651 pg->flags |= PG_BUSY; 1652 UVM_PAGE_OWN(pg, "lfs_putpages"); 1653 1654 pmap_page_protect(pg, VM_PROT_NONE); 1655 tdirty = (pmap_clear_modify(pg) || 1656 (pg->flags & PG_CLEAN) == 0); 1657 dirty += tdirty; 1658 } 1659 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1660 if (by_list) { 1661 curpg = TAILQ_NEXT(curpg, listq); 1662 } else { 1663 soff += fs->lfs_bsize; 1664 } 1665 continue; 1666 } 1667 1668 any_dirty += dirty; 1669 KASSERT(nonexistent == 0); 1670 1671 /* 1672 * If any are dirty make all dirty; unbusy them, 1673 * but if we were asked to clean, wire them so that 1674 * the pagedaemon doesn't bother us about them while 1675 * they're on their way to disk. 1676 */ 1677 for (i = 0; i == 0 || i < pages_per_block; i++) { 1678 pg = pgs[i]; 1679 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1680 if (dirty) { 1681 pg->flags &= ~PG_CLEAN; 1682 if (flags & PGO_FREE) { 1683 /* 1684 * Wire the page so that 1685 * pdaemon doesn't see it again. 1686 */ 1687 uvm_lock_pageq(); 1688 uvm_pagewire(pg); 1689 uvm_unlock_pageq(); 1690 1691 /* Suspended write flag */ 1692 pg->flags |= PG_DELWRI; 1693 } 1694 } 1695 if (pg->flags & PG_WANTED) 1696 wakeup(pg); 1697 pg->flags &= ~(PG_WANTED|PG_BUSY); 1698 UVM_PAGE_OWN(pg, NULL); 1699 } 1700 1701 if (checkfirst && any_dirty) 1702 break; 1703 1704 if (by_list) { 1705 curpg = TAILQ_NEXT(curpg, listq); 1706 } else { 1707 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1708 } 1709 } 1710 1711 return any_dirty; 1712 } 1713 1714 /* 1715 * lfs_putpages functions like genfs_putpages except that 1716 * 1717 * (1) It needs to bounds-check the incoming requests to ensure that 1718 * they are block-aligned; if they are not, expand the range and 1719 * do the right thing in case, e.g., the requested range is clean 1720 * but the expanded range is dirty. 1721 * (2) It needs to explicitly send blocks to be written when it is done. 1722 * VOP_PUTPAGES is not ever called with the seglock held, so 1723 * we simply take the seglock and let lfs_segunlock wait for us. 1724 * XXX Actually we can be called with the seglock held, if we have 1725 * XXX to flush a vnode while lfs_markv is in operation. As of this 1726 * XXX writing we panic in this case. 1727 * 1728 * Assumptions: 1729 * 1730 * (1) The caller does not hold any pages in this vnode busy. If it does, 1731 * there is a danger that when we expand the page range and busy the 1732 * pages we will deadlock. 1733 * (2) We are called with vp->v_interlock held; we must return with it 1734 * released. 1735 * (3) We don't absolutely have to free pages right away, provided that 1736 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1737 * us a request with PGO_FREE, we take the pages out of the paging 1738 * queue and wake up the writer, which will handle freeing them for us. 1739 * 1740 * We ensure that for any filesystem block, all pages for that 1741 * block are either resident or not, even if those pages are higher 1742 * than EOF; that means that we will be getting requests to free 1743 * "unused" pages above EOF all the time, and should ignore them. 1744 * 1745 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1746 */ 1747 1748 int 1749 lfs_putpages(void *v) 1750 { 1751 int error; 1752 struct vop_putpages_args /* { 1753 struct vnode *a_vp; 1754 voff_t a_offlo; 1755 voff_t a_offhi; 1756 int a_flags; 1757 } */ *ap = v; 1758 struct vnode *vp; 1759 struct inode *ip; 1760 struct lfs *fs; 1761 struct segment *sp; 1762 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1763 off_t off, max_endoffset; 1764 int s; 1765 boolean_t seglocked, sync, pagedaemon; 1766 struct vm_page *pg; 1767 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1768 1769 vp = ap->a_vp; 1770 ip = VTOI(vp); 1771 fs = ip->i_lfs; 1772 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1773 pagedaemon = (curproc == uvm.pagedaemon_proc); 1774 1775 /* Putpages does nothing for metadata. */ 1776 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1777 simple_unlock(&vp->v_interlock); 1778 return 0; 1779 } 1780 1781 /* 1782 * If there are no pages, don't do anything. 1783 */ 1784 if (vp->v_uobj.uo_npages == 0) { 1785 s = splbio(); 1786 if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL && 1787 (vp->v_flag & VONWORKLST)) { 1788 vp->v_flag &= ~VONWORKLST; 1789 LIST_REMOVE(vp, v_synclist); 1790 } 1791 splx(s); 1792 simple_unlock(&vp->v_interlock); 1793 1794 /* Remove us from paging queue, if we were on it */ 1795 simple_lock(&fs->lfs_interlock); 1796 if (ip->i_flags & IN_PAGING) { 1797 ip->i_flags &= ~IN_PAGING; 1798 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1799 } 1800 simple_unlock(&fs->lfs_interlock); 1801 return 0; 1802 } 1803 1804 blkeof = blkroundup(fs, ip->i_size); 1805 1806 /* 1807 * Ignore requests to free pages past EOF but in the same block 1808 * as EOF, unless the request is synchronous. (If the request is 1809 * sync, it comes from lfs_truncate.) 1810 * XXXUBC Make these pages look "active" so the pagedaemon won't 1811 * XXXUBC bother us with them again. 1812 */ 1813 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1814 origoffset = ap->a_offlo; 1815 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1816 pg = uvm_pagelookup(&vp->v_uobj, off); 1817 KASSERT(pg != NULL); 1818 while (pg->flags & PG_BUSY) { 1819 pg->flags |= PG_WANTED; 1820 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1821 "lfsput2", 0); 1822 simple_lock(&vp->v_interlock); 1823 } 1824 uvm_lock_pageq(); 1825 uvm_pageactivate(pg); 1826 uvm_unlock_pageq(); 1827 } 1828 ap->a_offlo = blkeof; 1829 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1830 simple_unlock(&vp->v_interlock); 1831 return 0; 1832 } 1833 } 1834 1835 /* 1836 * Extend page range to start and end at block boundaries. 1837 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1838 */ 1839 origoffset = ap->a_offlo; 1840 origendoffset = ap->a_offhi; 1841 startoffset = origoffset & ~(fs->lfs_bmask); 1842 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1843 << fs->lfs_bshift; 1844 1845 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1846 endoffset = max_endoffset; 1847 origendoffset = endoffset; 1848 } else { 1849 origendoffset = round_page(ap->a_offhi); 1850 endoffset = round_page(blkroundup(fs, origendoffset)); 1851 } 1852 1853 KASSERT(startoffset > 0 || endoffset >= startoffset); 1854 if (startoffset == endoffset) { 1855 /* Nothing to do, why were we called? */ 1856 simple_unlock(&vp->v_interlock); 1857 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 1858 PRId64 "\n", startoffset)); 1859 return 0; 1860 } 1861 1862 ap->a_offlo = startoffset; 1863 ap->a_offhi = endoffset; 1864 1865 if (!(ap->a_flags & PGO_CLEANIT)) 1866 return genfs_putpages(v); 1867 1868 /* 1869 * If there are more than one page per block, we don't want 1870 * to get caught locking them backwards; so set PGO_BUSYFAIL 1871 * to avoid deadlocks. 1872 */ 1873 ap->a_flags |= PGO_BUSYFAIL; 1874 1875 do { 1876 int r; 1877 1878 /* If no pages are dirty, we can just use genfs_putpages. */ 1879 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 1880 ap->a_flags, 1); 1881 if (r < 0) { 1882 simple_unlock(&vp->v_interlock); 1883 return EDEADLK; 1884 } 1885 if (r > 0) 1886 break; 1887 1888 /* 1889 * Sometimes pages are dirtied between the time that 1890 * we check and the time we try to clean them. 1891 * Instruct lfs_gop_write to return EDEADLK in this case 1892 * so we can write them properly. 1893 */ 1894 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 1895 r = genfs_putpages(v); 1896 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 1897 if (r != EDEADLK) 1898 return r; 1899 1900 /* Start over. */ 1901 preempt(1); 1902 simple_lock(&vp->v_interlock); 1903 } while(1); 1904 1905 /* 1906 * Dirty and asked to clean. 1907 * 1908 * Pagedaemon can't actually write LFS pages; wake up 1909 * the writer to take care of that. The writer will 1910 * notice the pager inode queue and act on that. 1911 */ 1912 if (pagedaemon) { 1913 simple_lock(&fs->lfs_interlock); 1914 if (!(ip->i_flags & IN_PAGING)) { 1915 ip->i_flags |= IN_PAGING; 1916 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1917 } 1918 simple_lock(&lfs_subsys_lock); 1919 wakeup(&lfs_writer_daemon); 1920 simple_unlock(&lfs_subsys_lock); 1921 simple_unlock(&fs->lfs_interlock); 1922 simple_unlock(&vp->v_interlock); 1923 preempt(1); 1924 return EWOULDBLOCK; 1925 } 1926 1927 /* 1928 * If this is a file created in a recent dirop, we can't flush its 1929 * inode until the dirop is complete. Drain dirops, then flush the 1930 * filesystem (taking care of any other pending dirops while we're 1931 * at it). 1932 */ 1933 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 1934 (vp->v_flag & VDIROP)) { 1935 int locked; 1936 1937 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 1938 locked = VOP_ISLOCKED(vp) && /* XXX */ 1939 vp->v_lock.lk_lockholder == curproc->p_pid; 1940 simple_unlock(&vp->v_interlock); 1941 lfs_writer_enter(fs, "ppdirop"); 1942 if (locked) 1943 VOP_UNLOCK(vp, 0); 1944 1945 simple_lock(&fs->lfs_interlock); 1946 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 1947 simple_unlock(&fs->lfs_interlock); 1948 1949 simple_lock(&vp->v_interlock); 1950 if (locked) { 1951 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 1952 simple_lock(&vp->v_interlock); 1953 } 1954 lfs_writer_leave(fs); 1955 1956 /* XXX the flush should have taken care of this one too! */ 1957 } 1958 1959 /* 1960 * This is it. We are going to write some pages. From here on 1961 * down it's all just mechanics. 1962 * 1963 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 1964 */ 1965 ap->a_flags &= ~PGO_SYNCIO; 1966 1967 /* 1968 * If we've already got the seglock, flush the node and return. 1969 * The FIP has already been set up for us by lfs_writefile, 1970 * and FIP cleanup and lfs_updatemeta will also be done there, 1971 * unless genfs_putpages returns EDEADLK; then we must flush 1972 * what we have, and correct FIP and segment header accounting. 1973 */ 1974 1975 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 1976 if (!seglocked) { 1977 simple_unlock(&vp->v_interlock); 1978 /* 1979 * Take the seglock, because we are going to be writing pages. 1980 */ 1981 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 1982 if (error != 0) 1983 return error; 1984 simple_lock(&vp->v_interlock); 1985 } 1986 1987 /* 1988 * VOP_PUTPAGES should not be called while holding the seglock. 1989 * XXXUBC fix lfs_markv, or do this properly. 1990 */ 1991 #ifdef notyet 1992 KASSERT(fs->lfs_seglock == 1); 1993 #endif /* notyet */ 1994 1995 /* 1996 * We assume we're being called with sp->fip pointing at blank space. 1997 * Account for a new FIP in the segment header, and set sp->vp. 1998 * (This should duplicate the setup at the top of lfs_writefile().) 1999 */ 2000 sp = fs->lfs_sp; 2001 if (!seglocked) { 2002 if (sp->seg_bytes_left < fs->lfs_bsize || 2003 sp->sum_bytes_left < sizeof(struct finfo)) 2004 (void) lfs_writeseg(fs, fs->lfs_sp); 2005 2006 sp->sum_bytes_left -= FINFOSIZE; 2007 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 2008 } 2009 KASSERT(sp->vp == NULL); 2010 sp->vp = vp; 2011 2012 if (!seglocked) { 2013 if (vp->v_flag & VDIROP) 2014 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2015 } 2016 2017 sp->fip->fi_nblocks = 0; 2018 sp->fip->fi_ino = ip->i_number; 2019 sp->fip->fi_version = ip->i_gen; 2020 2021 /* 2022 * Loop through genfs_putpages until all pages are gathered. 2023 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2024 * Whenever we lose the interlock we have to rerun check_dirty, as 2025 * well. 2026 */ 2027 again: 2028 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2029 ap->a_flags, 0) < 0) { 2030 simple_unlock(&vp->v_interlock); 2031 sp->vp = NULL; 2032 if (!seglocked) 2033 lfs_segunlock(fs); 2034 return EDEADLK; 2035 } 2036 2037 error = genfs_putpages(v); 2038 if (error == EDEADLK || error == EAGAIN) { 2039 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2040 " EDEADLK [2] ino %d off %x (seg %d)\n", 2041 ip->i_number, fs->lfs_offset, 2042 dtosn(fs, fs->lfs_offset))); 2043 /* If nothing to write, short-circuit */ 2044 if (sp->cbpp - sp->bpp > 1) { 2045 /* Write gathered pages */ 2046 lfs_updatemeta(sp); 2047 (void) lfs_writeseg(fs, sp); 2048 2049 /* 2050 * Reinitialize brand new FIP and add us to it. 2051 * (This should duplicate the fixup in 2052 * lfs_gatherpages().) 2053 */ 2054 KASSERT(sp->vp == vp); 2055 sp->fip->fi_version = ip->i_gen; 2056 sp->fip->fi_ino = ip->i_number; 2057 /* Add us to the new segment summary. */ 2058 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 2059 sp->sum_bytes_left -= FINFOSIZE; 2060 } 2061 2062 /* Give the write a chance to complete */ 2063 preempt(1); 2064 2065 /* We've lost the interlock. Start over. */ 2066 if (error == EDEADLK) { 2067 simple_lock(&vp->v_interlock); 2068 goto again; 2069 } 2070 } 2071 2072 KASSERT(sp->vp == vp); 2073 if (!seglocked) { 2074 sp->vp = NULL; /* XXX lfs_gather below will set this */ 2075 2076 /* Write indirect blocks as well */ 2077 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2078 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2079 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2080 2081 KASSERT(sp->vp == NULL); 2082 sp->vp = vp; 2083 } 2084 2085 /* 2086 * Blocks are now gathered into a segment waiting to be written. 2087 * All that's left to do is update metadata, and write them. 2088 */ 2089 lfs_updatemeta(sp); 2090 KASSERT(sp->vp == vp); 2091 sp->vp = NULL; 2092 2093 if (seglocked) { 2094 /* we're called by lfs_writefile. */ 2095 return error; 2096 } 2097 2098 /* 2099 * Clean up FIP, since we're done writing this file. 2100 * This should duplicate cleanup at the end of lfs_writefile(). 2101 */ 2102 if (sp->fip->fi_nblocks != 0) { 2103 sp->fip = (FINFO*)((caddr_t)sp->fip + FINFOSIZE + 2104 sizeof(int32_t) * sp->fip->fi_nblocks); 2105 sp->start_lbp = &sp->fip->fi_blocks[0]; 2106 } else { 2107 sp->sum_bytes_left += FINFOSIZE; 2108 --((SEGSUM *)(sp->segsum))->ss_nfinfo; 2109 } 2110 lfs_writeseg(fs, fs->lfs_sp); 2111 2112 /* 2113 * Remove us from paging queue, since we've now written all our 2114 * pages. 2115 */ 2116 simple_lock(&fs->lfs_interlock); 2117 if (ip->i_flags & IN_PAGING) { 2118 ip->i_flags &= ~IN_PAGING; 2119 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2120 } 2121 simple_unlock(&fs->lfs_interlock); 2122 2123 /* 2124 * XXX - with the malloc/copy writeseg, the pages are freed by now 2125 * even if we don't wait (e.g. if we hold a nested lock). This 2126 * will not be true if we stop using malloc/copy. 2127 */ 2128 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2129 lfs_segunlock(fs); 2130 2131 /* 2132 * Wait for v_numoutput to drop to zero. The seglock should 2133 * take care of this, but there is a slight possibility that 2134 * aiodoned might not have got around to our buffers yet. 2135 */ 2136 if (sync) { 2137 s = splbio(); 2138 simple_lock(&global_v_numoutput_slock); 2139 while (vp->v_numoutput > 0) { 2140 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2141 " num %d\n", ip->i_number, vp->v_numoutput)); 2142 vp->v_flag |= VBWAIT; 2143 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 2144 &global_v_numoutput_slock); 2145 } 2146 simple_unlock(&global_v_numoutput_slock); 2147 splx(s); 2148 } 2149 return error; 2150 } 2151 2152 /* 2153 * Return the last logical file offset that should be written for this file 2154 * if we're doing a write that ends at "size". If writing, we need to know 2155 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2156 * to know about entire blocks. 2157 */ 2158 void 2159 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2160 { 2161 struct inode *ip = VTOI(vp); 2162 struct lfs *fs = ip->i_lfs; 2163 daddr_t olbn, nlbn; 2164 2165 olbn = lblkno(fs, ip->i_size); 2166 nlbn = lblkno(fs, size); 2167 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2168 *eobp = fragroundup(fs, size); 2169 } else { 2170 *eobp = blkroundup(fs, size); 2171 } 2172 } 2173 2174 #ifdef DEBUG 2175 void lfs_dump_vop(void *); 2176 2177 void 2178 lfs_dump_vop(void *v) 2179 { 2180 struct vop_putpages_args /* { 2181 struct vnode *a_vp; 2182 voff_t a_offlo; 2183 voff_t a_offhi; 2184 int a_flags; 2185 } */ *ap = v; 2186 2187 #ifdef DDB 2188 vfs_vnode_print(ap->a_vp, 0, printf); 2189 #endif 2190 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2191 } 2192 #endif 2193 2194 int 2195 lfs_mmap(void *v) 2196 { 2197 struct vop_mmap_args /* { 2198 const struct vnodeop_desc *a_desc; 2199 struct vnode *a_vp; 2200 int a_fflags; 2201 struct ucred *a_cred; 2202 struct lwp *a_l; 2203 } */ *ap = v; 2204 2205 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2206 return EOPNOTSUPP; 2207 return ufs_mmap(v); 2208 } 2209
Indexes created Tue Sep 30 11:09:46 GMT 2025