lfs_vnops.c revision 1.173
1 /* $NetBSD: lfs_vnops.c,v 1.173 2006/05/04 04:22:57 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 1.173 2006/05/04 04:22:57 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 ++fs->lfs_dirvcount; 516 simple_unlock(&lfs_subsys_lock); 517 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 518 vp->v_flag |= VDIROP; 519 } 520 ++fs->lfs_nadirop; 521 ip->i_flag |= IN_ADIROP; 522 } else 523 KASSERT(vp->v_flag & VDIROP); 524 simple_unlock(&fs->lfs_interlock); 525 } 526 527 void 528 lfs_unmark_vnode(struct vnode *vp) 529 { 530 struct inode *ip = VTOI(vp); 531 532 if (ip && (ip->i_flag & IN_ADIROP)) { 533 KASSERT(vp->v_flag & VDIROP); 534 simple_lock(&ip->i_lfs->lfs_interlock); 535 --ip->i_lfs->lfs_nadirop; 536 simple_unlock(&ip->i_lfs->lfs_interlock); 537 ip->i_flag &= ~IN_ADIROP; 538 } 539 } 540 541 int 542 lfs_symlink(void *v) 543 { 544 struct vop_symlink_args /* { 545 struct vnode *a_dvp; 546 struct vnode **a_vpp; 547 struct componentname *a_cnp; 548 struct vattr *a_vap; 549 char *a_target; 550 } */ *ap = v; 551 int error; 552 553 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 554 vput(ap->a_dvp); 555 return error; 556 } 557 error = ufs_symlink(ap); 558 SET_ENDOP_CREATE_AP(ap, "symlink"); 559 return (error); 560 } 561 562 int 563 lfs_mknod(void *v) 564 { 565 struct vop_mknod_args /* { 566 struct vnode *a_dvp; 567 struct vnode **a_vpp; 568 struct componentname *a_cnp; 569 struct vattr *a_vap; 570 } */ *ap = v; 571 struct vattr *vap = ap->a_vap; 572 struct vnode **vpp = ap->a_vpp; 573 struct inode *ip; 574 int error; 575 struct mount *mp; 576 ino_t ino; 577 578 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 579 vput(ap->a_dvp); 580 return error; 581 } 582 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 583 ap->a_dvp, vpp, ap->a_cnp); 584 585 /* Either way we're done with the dirop at this point */ 586 SET_ENDOP_CREATE_AP(ap, "mknod"); 587 588 if (error) 589 return (error); 590 591 ip = VTOI(*vpp); 592 mp = (*vpp)->v_mount; 593 ino = ip->i_number; 594 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 595 if (vap->va_rdev != VNOVAL) { 596 /* 597 * Want to be able to use this to make badblock 598 * inodes, so don't truncate the dev number. 599 */ 600 #if 0 601 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 602 UFS_MPNEEDSWAP((*vpp)->v_mount)); 603 #else 604 ip->i_ffs1_rdev = vap->va_rdev; 605 #endif 606 } 607 608 /* 609 * Call fsync to write the vnode so that we don't have to deal with 610 * flushing it when it's marked VDIROP|VXLOCK. 611 * 612 * XXX KS - If we can't flush we also can't call vgone(), so must 613 * return. But, that leaves this vnode in limbo, also not good. 614 * Can this ever happen (barring hardware failure)? 615 */ 616 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 617 curlwp)) != 0) { 618 panic("lfs_mknod: couldn't fsync (ino %llu)", 619 (unsigned long long)ino); 620 /* return (error); */ 621 } 622 /* 623 * Remove vnode so that it will be reloaded by VFS_VGET and 624 * checked to see if it is an alias of an existing entry in 625 * the inode cache. 626 */ 627 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 628 629 VOP_UNLOCK(*vpp, 0); 630 lfs_vunref(*vpp); 631 (*vpp)->v_type = VNON; 632 vgone(*vpp); 633 error = VFS_VGET(mp, ino, vpp); 634 635 if (error != 0) { 636 *vpp = NULL; 637 return (error); 638 } 639 return (0); 640 } 641 642 int 643 lfs_create(void *v) 644 { 645 struct vop_create_args /* { 646 struct vnode *a_dvp; 647 struct vnode **a_vpp; 648 struct componentname *a_cnp; 649 struct vattr *a_vap; 650 } */ *ap = v; 651 int error; 652 653 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 654 vput(ap->a_dvp); 655 return error; 656 } 657 error = ufs_create(ap); 658 SET_ENDOP_CREATE_AP(ap, "create"); 659 return (error); 660 } 661 662 int 663 lfs_mkdir(void *v) 664 { 665 struct vop_mkdir_args /* { 666 struct vnode *a_dvp; 667 struct vnode **a_vpp; 668 struct componentname *a_cnp; 669 struct vattr *a_vap; 670 } */ *ap = v; 671 int error; 672 673 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 674 vput(ap->a_dvp); 675 return error; 676 } 677 error = ufs_mkdir(ap); 678 SET_ENDOP_CREATE_AP(ap, "mkdir"); 679 return (error); 680 } 681 682 int 683 lfs_remove(void *v) 684 { 685 struct vop_remove_args /* { 686 struct vnode *a_dvp; 687 struct vnode *a_vp; 688 struct componentname *a_cnp; 689 } */ *ap = v; 690 struct vnode *dvp, *vp; 691 int error; 692 693 dvp = ap->a_dvp; 694 vp = ap->a_vp; 695 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 696 if (dvp == vp) 697 vrele(vp); 698 else 699 vput(vp); 700 vput(dvp); 701 return error; 702 } 703 error = ufs_remove(ap); 704 SET_ENDOP_REMOVE(VTOI(dvp)->i_lfs, dvp, ap->a_vp, "remove"); 705 return (error); 706 } 707 708 int 709 lfs_rmdir(void *v) 710 { 711 struct vop_rmdir_args /* { 712 struct vnodeop_desc *a_desc; 713 struct vnode *a_dvp; 714 struct vnode *a_vp; 715 struct componentname *a_cnp; 716 } */ *ap = v; 717 struct vnode *vp; 718 int error; 719 720 vp = ap->a_vp; 721 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 722 vrele(ap->a_dvp); 723 if (ap->a_vp != ap->a_dvp) 724 VOP_UNLOCK(ap->a_dvp, 0); 725 vput(vp); 726 return error; 727 } 728 error = ufs_rmdir(ap); 729 SET_ENDOP_REMOVE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 730 return (error); 731 } 732 733 int 734 lfs_link(void *v) 735 { 736 struct vop_link_args /* { 737 struct vnode *a_dvp; 738 struct vnode *a_vp; 739 struct componentname *a_cnp; 740 } */ *ap = v; 741 int error; 742 struct vnode **vpp = NULL; 743 744 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 745 vput(ap->a_dvp); 746 return error; 747 } 748 error = ufs_link(ap); 749 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 750 return (error); 751 } 752 753 int 754 lfs_rename(void *v) 755 { 756 struct vop_rename_args /* { 757 struct vnode *a_fdvp; 758 struct vnode *a_fvp; 759 struct componentname *a_fcnp; 760 struct vnode *a_tdvp; 761 struct vnode *a_tvp; 762 struct componentname *a_tcnp; 763 } */ *ap = v; 764 struct vnode *tvp, *fvp, *tdvp, *fdvp; 765 struct componentname *tcnp, *fcnp; 766 int error; 767 struct lfs *fs; 768 769 fs = VTOI(ap->a_fdvp)->i_lfs; 770 tvp = ap->a_tvp; 771 tdvp = ap->a_tdvp; 772 tcnp = ap->a_tcnp; 773 fvp = ap->a_fvp; 774 fdvp = ap->a_fdvp; 775 fcnp = ap->a_fcnp; 776 777 /* 778 * Check for cross-device rename. 779 * If it is, we don't want to set dirops, just error out. 780 * (In particular note that MARK_VNODE(tdvp) will DTWT on 781 * a cross-device rename.) 782 * 783 * Copied from ufs_rename. 784 */ 785 if ((fvp->v_mount != tdvp->v_mount) || 786 (tvp && (fvp->v_mount != tvp->v_mount))) { 787 error = EXDEV; 788 goto errout; 789 } 790 791 /* 792 * Check to make sure we're not renaming a vnode onto itself 793 * (deleting a hard link by renaming one name onto another); 794 * if we are we can't recursively call VOP_REMOVE since that 795 * would leave us with an unaccounted-for number of live dirops. 796 * 797 * Inline the relevant section of ufs_rename here, *before* 798 * calling SET_DIROP_REMOVE. 799 */ 800 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 801 (VTOI(tdvp)->i_flags & APPEND))) { 802 error = EPERM; 803 goto errout; 804 } 805 if (fvp == tvp) { 806 if (fvp->v_type == VDIR) { 807 error = EINVAL; 808 goto errout; 809 } 810 811 /* Release destination completely. */ 812 VOP_ABORTOP(tdvp, tcnp); 813 vput(tdvp); 814 vput(tvp); 815 816 /* Delete source. */ 817 vrele(fvp); 818 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 819 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 820 fcnp->cn_nameiop = DELETE; 821 if ((error = relookup(fdvp, &fvp, fcnp))){ 822 /* relookup blew away fdvp */ 823 return (error); 824 } 825 return (VOP_REMOVE(fdvp, fvp, fcnp)); 826 } 827 828 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 829 goto errout; 830 MARK_VNODE(fdvp); 831 MARK_VNODE(fvp); 832 833 error = ufs_rename(ap); 834 UNMARK_VNODE(fdvp); 835 UNMARK_VNODE(fvp); 836 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 837 return (error); 838 839 errout: 840 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 841 if (tdvp == tvp) 842 vrele(tdvp); 843 else 844 vput(tdvp); 845 if (tvp) 846 vput(tvp); 847 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 848 vrele(fdvp); 849 vrele(fvp); 850 return (error); 851 } 852 853 /* XXX hack to avoid calling ITIMES in getattr */ 854 int 855 lfs_getattr(void *v) 856 { 857 struct vop_getattr_args /* { 858 struct vnode *a_vp; 859 struct vattr *a_vap; 860 struct ucred *a_cred; 861 struct lwp *a_l; 862 } */ *ap = v; 863 struct vnode *vp = ap->a_vp; 864 struct inode *ip = VTOI(vp); 865 struct vattr *vap = ap->a_vap; 866 struct lfs *fs = ip->i_lfs; 867 /* 868 * Copy from inode table 869 */ 870 vap->va_fsid = ip->i_dev; 871 vap->va_fileid = ip->i_number; 872 vap->va_mode = ip->i_mode & ~IFMT; 873 vap->va_nlink = ip->i_nlink; 874 vap->va_uid = ip->i_uid; 875 vap->va_gid = ip->i_gid; 876 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 877 vap->va_size = vp->v_size; 878 vap->va_atime.tv_sec = ip->i_ffs1_atime; 879 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 880 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 881 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 882 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 883 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 884 vap->va_flags = ip->i_flags; 885 vap->va_gen = ip->i_gen; 886 /* this doesn't belong here */ 887 if (vp->v_type == VBLK) 888 vap->va_blocksize = BLKDEV_IOSIZE; 889 else if (vp->v_type == VCHR) 890 vap->va_blocksize = MAXBSIZE; 891 else 892 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 893 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 894 vap->va_type = vp->v_type; 895 vap->va_filerev = ip->i_modrev; 896 return (0); 897 } 898 899 /* 900 * Check to make sure the inode blocks won't choke the buffer 901 * cache, then call ufs_setattr as usual. 902 */ 903 int 904 lfs_setattr(void *v) 905 { 906 struct vop_setattr_args /* { 907 struct vnode *a_vp; 908 struct vattr *a_vap; 909 struct ucred *a_cred; 910 struct lwp *a_l; 911 } */ *ap = v; 912 struct vnode *vp = ap->a_vp; 913 914 lfs_check(vp, LFS_UNUSED_LBN, 0); 915 return ufs_setattr(v); 916 } 917 918 /* 919 * Close called 920 * 921 * XXX -- we were using ufs_close, but since it updates the 922 * times on the inode, we might need to bump the uinodes 923 * count. 924 */ 925 /* ARGSUSED */ 926 int 927 lfs_close(void *v) 928 { 929 struct vop_close_args /* { 930 struct vnode *a_vp; 931 int a_fflag; 932 struct ucred *a_cred; 933 struct lwp *a_l; 934 } */ *ap = v; 935 struct vnode *vp = ap->a_vp; 936 struct inode *ip = VTOI(vp); 937 938 if (vp == ip->i_lfs->lfs_ivnode && 939 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 940 return 0; 941 942 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 943 LFS_ITIMES(ip, NULL, NULL, NULL); 944 } 945 return (0); 946 } 947 948 /* 949 * Close wrapper for special devices. 950 * 951 * Update the times on the inode then do device close. 952 */ 953 int 954 lfsspec_close(void *v) 955 { 956 struct vop_close_args /* { 957 struct vnode *a_vp; 958 int a_fflag; 959 struct ucred *a_cred; 960 struct lwp *a_l; 961 } */ *ap = v; 962 struct vnode *vp; 963 struct inode *ip; 964 965 vp = ap->a_vp; 966 ip = VTOI(vp); 967 if (vp->v_usecount > 1) { 968 LFS_ITIMES(ip, NULL, NULL, NULL); 969 } 970 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 971 } 972 973 /* 974 * Close wrapper for fifo's. 975 * 976 * Update the times on the inode then do device close. 977 */ 978 int 979 lfsfifo_close(void *v) 980 { 981 struct vop_close_args /* { 982 struct vnode *a_vp; 983 int a_fflag; 984 struct ucred *a_cred; 985 struct lwp *a_l; 986 } */ *ap = v; 987 struct vnode *vp; 988 struct inode *ip; 989 990 vp = ap->a_vp; 991 ip = VTOI(vp); 992 if (ap->a_vp->v_usecount > 1) { 993 LFS_ITIMES(ip, NULL, NULL, NULL); 994 } 995 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 996 } 997 998 /* 999 * Reclaim an inode so that it can be used for other purposes. 1000 */ 1001 1002 int 1003 lfs_reclaim(void *v) 1004 { 1005 struct vop_reclaim_args /* { 1006 struct vnode *a_vp; 1007 struct lwp *a_l; 1008 } */ *ap = v; 1009 struct vnode *vp = ap->a_vp; 1010 struct inode *ip = VTOI(vp); 1011 int error; 1012 1013 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1014 1015 LFS_CLR_UINO(ip, IN_ALLMOD); 1016 if ((error = ufs_reclaim(vp, ap->a_l))) 1017 return (error); 1018 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1019 lfs_deregister_all(vp); 1020 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1021 ip->inode_ext.lfs = NULL; 1022 pool_put(&lfs_inode_pool, vp->v_data); 1023 vp->v_data = NULL; 1024 return (0); 1025 } 1026 1027 /* 1028 * Read a block from a storage device. 1029 * In order to avoid reading blocks that are in the process of being 1030 * written by the cleaner---and hence are not mutexed by the normal 1031 * buffer cache / page cache mechanisms---check for collisions before 1032 * reading. 1033 * 1034 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1035 * the active cleaner test. 1036 * 1037 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1038 */ 1039 int 1040 lfs_strategy(void *v) 1041 { 1042 struct vop_strategy_args /* { 1043 struct vnode *a_vp; 1044 struct buf *a_bp; 1045 } */ *ap = v; 1046 struct buf *bp; 1047 struct lfs *fs; 1048 struct vnode *vp; 1049 struct inode *ip; 1050 daddr_t tbn; 1051 int i, sn, error, slept; 1052 1053 bp = ap->a_bp; 1054 vp = ap->a_vp; 1055 ip = VTOI(vp); 1056 fs = ip->i_lfs; 1057 1058 /* lfs uses its strategy routine only for read */ 1059 KASSERT(bp->b_flags & B_READ); 1060 1061 if (vp->v_type == VBLK || vp->v_type == VCHR) 1062 panic("lfs_strategy: spec"); 1063 KASSERT(bp->b_bcount != 0); 1064 if (bp->b_blkno == bp->b_lblkno) { 1065 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1066 NULL); 1067 if (error) { 1068 bp->b_error = error; 1069 bp->b_flags |= B_ERROR; 1070 biodone(bp); 1071 return (error); 1072 } 1073 if ((long)bp->b_blkno == -1) /* no valid data */ 1074 clrbuf(bp); 1075 } 1076 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1077 biodone(bp); 1078 return (0); 1079 } 1080 1081 slept = 1; 1082 simple_lock(&fs->lfs_interlock); 1083 while (slept && fs->lfs_seglock) { 1084 simple_unlock(&fs->lfs_interlock); 1085 /* 1086 * Look through list of intervals. 1087 * There will only be intervals to look through 1088 * if the cleaner holds the seglock. 1089 * Since the cleaner is synchronous, we can trust 1090 * the list of intervals to be current. 1091 */ 1092 tbn = dbtofsb(fs, bp->b_blkno); 1093 sn = dtosn(fs, tbn); 1094 slept = 0; 1095 for (i = 0; i < fs->lfs_cleanind; i++) { 1096 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1097 tbn >= fs->lfs_cleanint[i]) { 1098 DLOG((DLOG_CLEAN, 1099 "lfs_strategy: ino %d lbn %" PRId64 1100 " ind %d sn %d fsb %" PRIx32 1101 " given sn %d fsb %" PRIx64 "\n", 1102 ip->i_number, bp->b_lblkno, i, 1103 dtosn(fs, fs->lfs_cleanint[i]), 1104 fs->lfs_cleanint[i], sn, tbn)); 1105 DLOG((DLOG_CLEAN, 1106 "lfs_strategy: sleeping on ino %d lbn %" 1107 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1108 simple_lock(&fs->lfs_interlock); 1109 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1110 /* Cleaner can't wait for itself */ 1111 ltsleep(&fs->lfs_iocount, 1112 (PRIBIO + 1) | PNORELOCK, 1113 "clean2", 0, 1114 &fs->lfs_interlock); 1115 slept = 1; 1116 break; 1117 } else if (fs->lfs_seglock) { 1118 ltsleep(&fs->lfs_seglock, 1119 (PRIBIO + 1) | PNORELOCK, 1120 "clean1", 0, 1121 &fs->lfs_interlock); 1122 slept = 1; 1123 break; 1124 } 1125 simple_unlock(&fs->lfs_interlock); 1126 } 1127 } 1128 simple_lock(&fs->lfs_interlock); 1129 } 1130 simple_unlock(&fs->lfs_interlock); 1131 1132 vp = ip->i_devvp; 1133 VOP_STRATEGY(vp, bp); 1134 return (0); 1135 } 1136 1137 void 1138 lfs_flush_dirops(struct lfs *fs) 1139 { 1140 struct inode *ip, *nip; 1141 struct vnode *vp; 1142 extern int lfs_dostats; 1143 struct segment *sp; 1144 int waslocked; 1145 1146 ASSERT_MAYBE_SEGLOCK(fs); 1147 KASSERT(fs->lfs_nadirop == 0); 1148 1149 if (fs->lfs_ronly) 1150 return; 1151 1152 simple_lock(&fs->lfs_interlock); 1153 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1154 simple_unlock(&fs->lfs_interlock); 1155 return; 1156 } else 1157 simple_unlock(&fs->lfs_interlock); 1158 1159 if (lfs_dostats) 1160 ++lfs_stats.flush_invoked; 1161 1162 /* 1163 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1164 * Technically this is a checkpoint (the on-disk state is valid) 1165 * even though we are leaving out all the file data. 1166 */ 1167 lfs_imtime(fs); 1168 lfs_seglock(fs, SEGM_CKP); 1169 sp = fs->lfs_sp; 1170 1171 /* 1172 * lfs_writevnodes, optimized to get dirops out of the way. 1173 * Only write dirops, and don't flush files' pages, only 1174 * blocks from the directories. 1175 * 1176 * We don't need to vref these files because they are 1177 * dirops and so hold an extra reference until the 1178 * segunlock clears them of that status. 1179 * 1180 * We don't need to check for IN_ADIROP because we know that 1181 * no dirops are active. 1182 * 1183 */ 1184 simple_lock(&fs->lfs_interlock); 1185 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1186 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1187 simple_unlock(&fs->lfs_interlock); 1188 vp = ITOV(ip); 1189 1190 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1191 1192 /* 1193 * All writes to directories come from dirops; all 1194 * writes to files' direct blocks go through the page 1195 * cache, which we're not touching. Reads to files 1196 * and/or directories will not be affected by writing 1197 * directory blocks inodes and file inodes. So we don't 1198 * really need to lock. If we don't lock, though, 1199 * make sure that we don't clear IN_MODIFIED 1200 * unnecessarily. 1201 */ 1202 if (vp->v_flag & (VXLOCK | VFREEING)) { 1203 simple_lock(&fs->lfs_interlock); 1204 continue; 1205 } 1206 waslocked = VOP_ISLOCKED(vp); 1207 if (vp->v_type != VREG && 1208 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1209 lfs_writefile(fs, sp, vp); 1210 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1211 !(ip->i_flag & IN_ALLMOD)) { 1212 LFS_SET_UINO(ip, IN_MODIFIED); 1213 } 1214 } 1215 (void) lfs_writeinode(fs, sp, ip); 1216 if (waslocked) 1217 LFS_SET_UINO(ip, IN_MODIFIED); 1218 simple_lock(&fs->lfs_interlock); 1219 } 1220 simple_unlock(&fs->lfs_interlock); 1221 /* We've written all the dirops there are */ 1222 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1223 lfs_finalize_fs_seguse(fs); 1224 (void) lfs_writeseg(fs, sp); 1225 lfs_segunlock(fs); 1226 } 1227 1228 /* 1229 * Flush all vnodes for which the pagedaemon has requested pageouts. 1230 * Skip over any files that are marked VDIROP (since lfs_flush_dirop() 1231 * has just run, this would be an error). If we have to skip a vnode 1232 * for any reason, just skip it; if we have to wait for the cleaner, 1233 * abort. The writer daemon will call us again later. 1234 */ 1235 void 1236 lfs_flush_pchain(struct lfs *fs) 1237 { 1238 struct inode *ip, *nip; 1239 struct vnode *vp; 1240 extern int lfs_dostats; 1241 struct segment *sp; 1242 int error; 1243 1244 ASSERT_NO_SEGLOCK(fs); 1245 1246 if (fs->lfs_ronly) 1247 return; 1248 1249 simple_lock(&fs->lfs_interlock); 1250 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1251 simple_unlock(&fs->lfs_interlock); 1252 return; 1253 } else 1254 simple_unlock(&fs->lfs_interlock); 1255 1256 /* Get dirops out of the way */ 1257 lfs_flush_dirops(fs); 1258 1259 if (lfs_dostats) 1260 ++lfs_stats.flush_invoked; 1261 1262 /* 1263 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1264 */ 1265 lfs_imtime(fs); 1266 lfs_seglock(fs, 0); 1267 sp = fs->lfs_sp; 1268 1269 /* 1270 * lfs_writevnodes, optimized to clear pageout requests. 1271 * Only write non-dirop files that are in the pageout queue. 1272 * We're very conservative about what we write; we want to be 1273 * fast and async. 1274 */ 1275 simple_lock(&fs->lfs_interlock); 1276 top: 1277 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1278 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1279 vp = ITOV(ip); 1280 1281 if (!(ip->i_flags & IN_PAGING)) 1282 goto top; 1283 1284 if (vp->v_flag & (VXLOCK|VDIROP)) 1285 continue; 1286 if (vp->v_type != VREG) 1287 continue; 1288 if (lfs_vref(vp)) 1289 continue; 1290 simple_unlock(&fs->lfs_interlock); 1291 1292 if (VOP_ISLOCKED(vp)) { 1293 lfs_vunref(vp); 1294 simple_lock(&fs->lfs_interlock); 1295 continue; 1296 } 1297 1298 error = lfs_writefile(fs, sp, vp); 1299 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1300 !(ip->i_flag & IN_ALLMOD)) { 1301 LFS_SET_UINO(ip, IN_MODIFIED); 1302 } 1303 (void) lfs_writeinode(fs, sp, ip); 1304 1305 lfs_vunref(vp); 1306 1307 if (error == EAGAIN) { 1308 lfs_writeseg(fs, sp); 1309 simple_lock(&fs->lfs_interlock); 1310 break; 1311 } 1312 simple_lock(&fs->lfs_interlock); 1313 } 1314 simple_unlock(&fs->lfs_interlock); 1315 (void) lfs_writeseg(fs, sp); 1316 lfs_segunlock(fs); 1317 } 1318 1319 /* 1320 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1321 */ 1322 int 1323 lfs_fcntl(void *v) 1324 { 1325 struct vop_fcntl_args /* { 1326 struct vnode *a_vp; 1327 u_long a_command; 1328 caddr_t a_data; 1329 int a_fflag; 1330 struct ucred *a_cred; 1331 struct lwp *a_l; 1332 } */ *ap = v; 1333 struct timeval *tvp; 1334 BLOCK_INFO *blkiov; 1335 CLEANERINFO *cip; 1336 SEGUSE *sup; 1337 int blkcnt, error, oclean; 1338 struct lfs_fcntl_markv blkvp; 1339 struct proc *p; 1340 fsid_t *fsidp; 1341 struct lfs *fs; 1342 struct buf *bp; 1343 fhandle_t *fhp; 1344 daddr_t off; 1345 1346 /* Only respect LFS fcntls on fs root or Ifile */ 1347 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1348 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1349 return ufs_fcntl(v); 1350 } 1351 1352 /* Avoid locking a draining lock */ 1353 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1354 return ESHUTDOWN; 1355 } 1356 1357 p = ap->a_l->l_proc; 1358 fs = VTOI(ap->a_vp)->i_lfs; 1359 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1360 1361 switch (ap->a_command) { 1362 case LFCNSEGWAITALL: 1363 case LFCNSEGWAITALL_COMPAT: 1364 fsidp = NULL; 1365 /* FALLSTHROUGH */ 1366 case LFCNSEGWAIT: 1367 case LFCNSEGWAIT_COMPAT: 1368 tvp = (struct timeval *)ap->a_data; 1369 simple_lock(&fs->lfs_interlock); 1370 ++fs->lfs_sleepers; 1371 simple_unlock(&fs->lfs_interlock); 1372 VOP_UNLOCK(ap->a_vp, 0); 1373 1374 error = lfs_segwait(fsidp, tvp); 1375 1376 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1377 simple_lock(&fs->lfs_interlock); 1378 if (--fs->lfs_sleepers == 0) 1379 wakeup(&fs->lfs_sleepers); 1380 simple_unlock(&fs->lfs_interlock); 1381 return error; 1382 1383 case LFCNBMAPV: 1384 case LFCNMARKV: 1385 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0) 1386 return (error); 1387 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1388 1389 blkcnt = blkvp.blkcnt; 1390 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1391 return (EINVAL); 1392 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1393 if ((error = copyin(blkvp.blkiov, blkiov, 1394 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1395 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1396 return error; 1397 } 1398 1399 simple_lock(&fs->lfs_interlock); 1400 ++fs->lfs_sleepers; 1401 simple_unlock(&fs->lfs_interlock); 1402 VOP_UNLOCK(ap->a_vp, 0); 1403 if (ap->a_command == LFCNBMAPV) 1404 error = lfs_bmapv(p, fsidp, blkiov, blkcnt); 1405 else /* LFCNMARKV */ 1406 error = lfs_markv(p, fsidp, blkiov, blkcnt); 1407 if (error == 0) 1408 error = copyout(blkiov, blkvp.blkiov, 1409 blkcnt * sizeof(BLOCK_INFO)); 1410 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1411 simple_lock(&fs->lfs_interlock); 1412 if (--fs->lfs_sleepers == 0) 1413 wakeup(&fs->lfs_sleepers); 1414 simple_unlock(&fs->lfs_interlock); 1415 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1416 return error; 1417 1418 case LFCNRECLAIM: 1419 /* 1420 * Flush dirops and write Ifile, allowing empty segments 1421 * to be immediately reclaimed. 1422 */ 1423 VOP_UNLOCK(ap->a_vp, 0); 1424 lfs_writer_enter(fs, "pndirop"); 1425 off = fs->lfs_offset; 1426 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1427 lfs_flush_dirops(fs); 1428 LFS_CLEANERINFO(cip, fs, bp); 1429 oclean = cip->clean; 1430 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1431 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1432 fs->lfs_sp->seg_flags |= SEGM_PROT; 1433 lfs_segunlock(fs); 1434 lfs_writer_leave(fs); 1435 1436 #ifdef DEBUG 1437 LFS_CLEANERINFO(cip, fs, bp); 1438 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1439 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1440 fs->lfs_offset - off, cip->clean - oclean, 1441 fs->lfs_activesb)); 1442 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1443 #endif 1444 1445 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1446 return 0; 1447 1448 case LFCNIFILEFH: 1449 /* Return the filehandle of the Ifile */ 1450 if ((error = suser(ap->a_l->l_proc->p_ucred, &ap->a_l->l_proc->p_acflag)) != 0) 1451 return (error); 1452 fhp = (struct fhandle *)ap->a_data; 1453 fhp->fh_fsid = *fsidp; 1454 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid)); 1455 1456 case LFCNREWIND: 1457 /* Move lfs_offset to the lowest-numbered segment */ 1458 return lfs_rewind(fs, *(int *)ap->a_data); 1459 1460 case LFCNINVAL: 1461 /* Mark a segment SEGUSE_INVAL */ 1462 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1463 if (sup->su_nbytes > 0) { 1464 brelse(bp); 1465 lfs_unset_inval_all(fs); 1466 return EBUSY; 1467 } 1468 sup->su_flags |= SEGUSE_INVAL; 1469 VOP_BWRITE(bp); 1470 return 0; 1471 1472 case LFCNRESIZE: 1473 /* Resize the filesystem */ 1474 return lfs_resize_fs(fs, *(int *)ap->a_data); 1475 1476 case LFCNWRAPSTOP: 1477 /* 1478 * Hold lfs_newseg at segment 0; sleep until the filesystem 1479 * wraps around. For debugging purposes, so an external 1480 * agent can log every segment in the filesystem as it 1481 * was written, and we can regression-test checkpoint 1482 * validity in the general case. 1483 */ 1484 VOP_UNLOCK(ap->a_vp, 0); 1485 simple_lock(&fs->lfs_interlock); 1486 fs->lfs_nowrap = 1; 1487 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER | PNORELOCK, 1488 "segwrap", 0, &fs->lfs_interlock); 1489 if (error) { 1490 fs->lfs_nowrap = 0; 1491 wakeup(&fs->lfs_nowrap); 1492 } 1493 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1494 return 0; 1495 1496 case LFCNWRAPGO: 1497 /* 1498 * Having done its work, the agent wakes up the writer. 1499 * It sleeps until a new segment is selected. 1500 */ 1501 VOP_UNLOCK(ap->a_vp, 0); 1502 simple_lock(&fs->lfs_interlock); 1503 fs->lfs_nowrap = 0; 1504 wakeup(&fs->lfs_nowrap); 1505 ltsleep(&fs->lfs_nextseg, PCATCH | PUSER | PNORELOCK, 1506 "segment", 0, &fs->lfs_interlock); 1507 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1508 return 0; 1509 1510 default: 1511 return ufs_fcntl(v); 1512 } 1513 return 0; 1514 } 1515 1516 int 1517 lfs_getpages(void *v) 1518 { 1519 struct vop_getpages_args /* { 1520 struct vnode *a_vp; 1521 voff_t a_offset; 1522 struct vm_page **a_m; 1523 int *a_count; 1524 int a_centeridx; 1525 vm_prot_t a_access_type; 1526 int a_advice; 1527 int a_flags; 1528 } */ *ap = v; 1529 1530 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1531 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1532 return EPERM; 1533 } 1534 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1535 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1536 } 1537 1538 /* 1539 * we're relying on the fact that genfs_getpages() always read in 1540 * entire filesystem blocks. 1541 */ 1542 return genfs_getpages(v); 1543 } 1544 1545 /* 1546 * Make sure that for all pages in every block in the given range, 1547 * either all are dirty or all are clean. If any of the pages 1548 * we've seen so far are dirty, put the vnode on the paging chain, 1549 * and mark it IN_PAGING. 1550 * 1551 * If checkfirst != 0, don't check all the pages but return at the 1552 * first dirty page. 1553 */ 1554 static int 1555 check_dirty(struct lfs *fs, struct vnode *vp, 1556 off_t startoffset, off_t endoffset, off_t blkeof, 1557 int flags, int checkfirst) 1558 { 1559 int by_list; 1560 struct vm_page *curpg = NULL; /* XXX: gcc */ 1561 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1562 off_t soff = 0; /* XXX: gcc */ 1563 voff_t off; 1564 int i; 1565 int nonexistent; 1566 int any_dirty; /* number of dirty pages */ 1567 int dirty; /* number of dirty pages in a block */ 1568 int tdirty; 1569 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1570 int pagedaemon = (curproc == uvm.pagedaemon_proc); 1571 1572 ASSERT_MAYBE_SEGLOCK(fs); 1573 top: 1574 by_list = (vp->v_uobj.uo_npages <= 1575 ((endoffset - startoffset) >> PAGE_SHIFT) * 1576 UVM_PAGE_HASH_PENALTY); 1577 any_dirty = 0; 1578 1579 if (by_list) { 1580 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1581 } else { 1582 soff = startoffset; 1583 } 1584 while (by_list || soff < MIN(blkeof, endoffset)) { 1585 if (by_list) { 1586 /* 1587 * Find the first page in a block. Skip 1588 * blocks outside our area of interest or beyond 1589 * the end of file. 1590 */ 1591 if (pages_per_block > 1) { 1592 while (curpg && 1593 ((curpg->offset & fs->lfs_bmask) || 1594 curpg->offset >= vp->v_size || 1595 curpg->offset >= endoffset)) 1596 curpg = TAILQ_NEXT(curpg, listq); 1597 } 1598 if (curpg == NULL) 1599 break; 1600 soff = curpg->offset; 1601 } 1602 1603 /* 1604 * Mark all pages in extended range busy; find out if any 1605 * of them are dirty. 1606 */ 1607 nonexistent = dirty = 0; 1608 for (i = 0; i == 0 || i < pages_per_block; i++) { 1609 if (by_list && pages_per_block <= 1) { 1610 pgs[i] = pg = curpg; 1611 } else { 1612 off = soff + (i << PAGE_SHIFT); 1613 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1614 if (pg == NULL) { 1615 ++nonexistent; 1616 continue; 1617 } 1618 } 1619 KASSERT(pg != NULL); 1620 1621 /* 1622 * If we're holding the segment lock, we can deadlocked 1623 * against a process that has our page and is waiting 1624 * for the cleaner, while the cleaner waits for the 1625 * segment lock. Just bail in that case. 1626 */ 1627 if ((pg->flags & PG_BUSY) && 1628 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1629 if (by_list && i > 0) 1630 uvm_page_unbusy(pgs, i); 1631 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1632 return -1; 1633 } 1634 1635 while (pg->flags & PG_BUSY) { 1636 pg->flags |= PG_WANTED; 1637 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1638 "lfsput", 0); 1639 simple_lock(&vp->v_interlock); 1640 if (by_list) { 1641 if (i > 0) 1642 uvm_page_unbusy(pgs, i); 1643 goto top; 1644 } 1645 } 1646 pg->flags |= PG_BUSY; 1647 UVM_PAGE_OWN(pg, "lfs_putpages"); 1648 1649 pmap_page_protect(pg, VM_PROT_NONE); 1650 tdirty = (pmap_clear_modify(pg) || 1651 (pg->flags & PG_CLEAN) == 0); 1652 dirty += tdirty; 1653 } 1654 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1655 if (by_list) { 1656 curpg = TAILQ_NEXT(curpg, listq); 1657 } else { 1658 soff += fs->lfs_bsize; 1659 } 1660 continue; 1661 } 1662 1663 any_dirty += dirty; 1664 KASSERT(nonexistent == 0); 1665 1666 /* 1667 * If any are dirty make all dirty; unbusy them, 1668 * but if we were asked to clean, wire them so that 1669 * the pagedaemon doesn't bother us about them while 1670 * they're on their way to disk. 1671 */ 1672 for (i = 0; i == 0 || i < pages_per_block; i++) { 1673 pg = pgs[i]; 1674 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1675 if (dirty) { 1676 pg->flags &= ~PG_CLEAN; 1677 if (flags & PGO_FREE) { 1678 /* 1679 * Wire the page so that 1680 * pdaemon doesn't see it again. 1681 */ 1682 uvm_lock_pageq(); 1683 uvm_pagewire(pg); 1684 uvm_unlock_pageq(); 1685 1686 /* Suspended write flag */ 1687 pg->flags |= PG_DELWRI; 1688 } 1689 } 1690 if (pg->flags & PG_WANTED) 1691 wakeup(pg); 1692 pg->flags &= ~(PG_WANTED|PG_BUSY); 1693 UVM_PAGE_OWN(pg, NULL); 1694 } 1695 1696 if (checkfirst && any_dirty) 1697 break; 1698 1699 if (by_list) { 1700 curpg = TAILQ_NEXT(curpg, listq); 1701 } else { 1702 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1703 } 1704 } 1705 1706 return any_dirty; 1707 } 1708 1709 /* 1710 * lfs_putpages functions like genfs_putpages except that 1711 * 1712 * (1) It needs to bounds-check the incoming requests to ensure that 1713 * they are block-aligned; if they are not, expand the range and 1714 * do the right thing in case, e.g., the requested range is clean 1715 * but the expanded range is dirty. 1716 * (2) It needs to explicitly send blocks to be written when it is done. 1717 * VOP_PUTPAGES is not ever called with the seglock held, so 1718 * we simply take the seglock and let lfs_segunlock wait for us. 1719 * XXX Actually we can be called with the seglock held, if we have 1720 * XXX to flush a vnode while lfs_markv is in operation. As of this 1721 * XXX writing we panic in this case. 1722 * 1723 * Assumptions: 1724 * 1725 * (1) The caller does not hold any pages in this vnode busy. If it does, 1726 * there is a danger that when we expand the page range and busy the 1727 * pages we will deadlock. 1728 * (2) We are called with vp->v_interlock held; we must return with it 1729 * released. 1730 * (3) We don't absolutely have to free pages right away, provided that 1731 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1732 * us a request with PGO_FREE, we take the pages out of the paging 1733 * queue and wake up the writer, which will handle freeing them for us. 1734 * 1735 * We ensure that for any filesystem block, all pages for that 1736 * block are either resident or not, even if those pages are higher 1737 * than EOF; that means that we will be getting requests to free 1738 * "unused" pages above EOF all the time, and should ignore them. 1739 * 1740 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1741 */ 1742 1743 int 1744 lfs_putpages(void *v) 1745 { 1746 int error; 1747 struct vop_putpages_args /* { 1748 struct vnode *a_vp; 1749 voff_t a_offlo; 1750 voff_t a_offhi; 1751 int a_flags; 1752 } */ *ap = v; 1753 struct vnode *vp; 1754 struct inode *ip; 1755 struct lfs *fs; 1756 struct segment *sp; 1757 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1758 off_t off, max_endoffset; 1759 int s; 1760 boolean_t seglocked, sync, pagedaemon; 1761 struct vm_page *pg; 1762 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1763 1764 vp = ap->a_vp; 1765 ip = VTOI(vp); 1766 fs = ip->i_lfs; 1767 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1768 pagedaemon = (curproc == uvm.pagedaemon_proc); 1769 1770 /* Putpages does nothing for metadata. */ 1771 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1772 simple_unlock(&vp->v_interlock); 1773 return 0; 1774 } 1775 1776 /* 1777 * If there are no pages, don't do anything. 1778 */ 1779 if (vp->v_uobj.uo_npages == 0) { 1780 s = splbio(); 1781 if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL && 1782 (vp->v_flag & VONWORKLST)) { 1783 vp->v_flag &= ~VONWORKLST; 1784 LIST_REMOVE(vp, v_synclist); 1785 } 1786 splx(s); 1787 simple_unlock(&vp->v_interlock); 1788 1789 /* Remove us from paging queue, if we were on it */ 1790 simple_lock(&fs->lfs_interlock); 1791 if (ip->i_flags & IN_PAGING) { 1792 ip->i_flags &= ~IN_PAGING; 1793 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1794 } 1795 simple_unlock(&fs->lfs_interlock); 1796 return 0; 1797 } 1798 1799 blkeof = blkroundup(fs, ip->i_size); 1800 1801 /* 1802 * Ignore requests to free pages past EOF but in the same block 1803 * as EOF, unless the request is synchronous. (If the request is 1804 * sync, it comes from lfs_truncate.) 1805 * XXXUBC Make these pages look "active" so the pagedaemon won't 1806 * XXXUBC bother us with them again. 1807 */ 1808 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1809 origoffset = ap->a_offlo; 1810 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1811 pg = uvm_pagelookup(&vp->v_uobj, off); 1812 KASSERT(pg != NULL); 1813 while (pg->flags & PG_BUSY) { 1814 pg->flags |= PG_WANTED; 1815 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1816 "lfsput2", 0); 1817 simple_lock(&vp->v_interlock); 1818 } 1819 uvm_lock_pageq(); 1820 uvm_pageactivate(pg); 1821 uvm_unlock_pageq(); 1822 } 1823 ap->a_offlo = blkeof; 1824 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1825 simple_unlock(&vp->v_interlock); 1826 return 0; 1827 } 1828 } 1829 1830 /* 1831 * Extend page range to start and end at block boundaries. 1832 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1833 */ 1834 origoffset = ap->a_offlo; 1835 origendoffset = ap->a_offhi; 1836 startoffset = origoffset & ~(fs->lfs_bmask); 1837 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1838 << fs->lfs_bshift; 1839 1840 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1841 endoffset = max_endoffset; 1842 origendoffset = endoffset; 1843 } else { 1844 origendoffset = round_page(ap->a_offhi); 1845 endoffset = round_page(blkroundup(fs, origendoffset)); 1846 } 1847 1848 KASSERT(startoffset > 0 || endoffset >= startoffset); 1849 if (startoffset == endoffset) { 1850 /* Nothing to do, why were we called? */ 1851 simple_unlock(&vp->v_interlock); 1852 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 1853 PRId64 "\n", startoffset)); 1854 return 0; 1855 } 1856 1857 ap->a_offlo = startoffset; 1858 ap->a_offhi = endoffset; 1859 1860 if (!(ap->a_flags & PGO_CLEANIT)) 1861 return genfs_putpages(v); 1862 1863 /* 1864 * If there are more than one page per block, we don't want 1865 * to get caught locking them backwards; so set PGO_BUSYFAIL 1866 * to avoid deadlocks. 1867 */ 1868 ap->a_flags |= PGO_BUSYFAIL; 1869 1870 do { 1871 int r; 1872 1873 /* If no pages are dirty, we can just use genfs_putpages. */ 1874 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 1875 ap->a_flags, 1); 1876 if (r < 0) { 1877 simple_unlock(&vp->v_interlock); 1878 return EDEADLK; 1879 } 1880 if (r > 0) 1881 break; 1882 1883 /* 1884 * Sometimes pages are dirtied between the time that 1885 * we check and the time we try to clean them. 1886 * Instruct lfs_gop_write to return EDEADLK in this case 1887 * so we can write them properly. 1888 */ 1889 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 1890 r = genfs_putpages(v); 1891 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 1892 if (r != EDEADLK) 1893 return r; 1894 1895 /* Start over. */ 1896 preempt(1); 1897 simple_lock(&vp->v_interlock); 1898 } while(1); 1899 1900 /* 1901 * Dirty and asked to clean. 1902 * 1903 * Pagedaemon can't actually write LFS pages; wake up 1904 * the writer to take care of that. The writer will 1905 * notice the pager inode queue and act on that. 1906 */ 1907 if (pagedaemon) { 1908 simple_lock(&fs->lfs_interlock); 1909 if (!(ip->i_flags & IN_PAGING)) { 1910 ip->i_flags |= IN_PAGING; 1911 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1912 } 1913 simple_lock(&lfs_subsys_lock); 1914 wakeup(&lfs_writer_daemon); 1915 simple_unlock(&lfs_subsys_lock); 1916 simple_unlock(&fs->lfs_interlock); 1917 simple_unlock(&vp->v_interlock); 1918 preempt(1); 1919 return EWOULDBLOCK; 1920 } 1921 1922 /* 1923 * If this is a file created in a recent dirop, we can't flush its 1924 * inode until the dirop is complete. Drain dirops, then flush the 1925 * filesystem (taking care of any other pending dirops while we're 1926 * at it). 1927 */ 1928 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 1929 (vp->v_flag & VDIROP)) { 1930 int locked; 1931 1932 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 1933 locked = VOP_ISLOCKED(vp) && /* XXX */ 1934 vp->v_lock.lk_lockholder == curproc->p_pid; 1935 simple_unlock(&vp->v_interlock); 1936 lfs_writer_enter(fs, "ppdirop"); 1937 if (locked) 1938 VOP_UNLOCK(vp, 0); 1939 1940 simple_lock(&fs->lfs_interlock); 1941 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 1942 simple_unlock(&fs->lfs_interlock); 1943 1944 simple_lock(&vp->v_interlock); 1945 if (locked) { 1946 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 1947 simple_lock(&vp->v_interlock); 1948 } 1949 lfs_writer_leave(fs); 1950 1951 /* XXX the flush should have taken care of this one too! */ 1952 } 1953 1954 /* 1955 * This is it. We are going to write some pages. From here on 1956 * down it's all just mechanics. 1957 * 1958 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 1959 */ 1960 ap->a_flags &= ~PGO_SYNCIO; 1961 1962 /* 1963 * If we've already got the seglock, flush the node and return. 1964 * The FIP has already been set up for us by lfs_writefile, 1965 * and FIP cleanup and lfs_updatemeta will also be done there, 1966 * unless genfs_putpages returns EDEADLK; then we must flush 1967 * what we have, and correct FIP and segment header accounting. 1968 */ 1969 1970 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 1971 if (!seglocked) { 1972 simple_unlock(&vp->v_interlock); 1973 /* 1974 * Take the seglock, because we are going to be writing pages. 1975 */ 1976 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 1977 if (error != 0) 1978 return error; 1979 simple_lock(&vp->v_interlock); 1980 } 1981 1982 /* 1983 * VOP_PUTPAGES should not be called while holding the seglock. 1984 * XXXUBC fix lfs_markv, or do this properly. 1985 */ 1986 #ifdef notyet 1987 KASSERT(fs->lfs_seglock == 1); 1988 #endif /* notyet */ 1989 1990 /* 1991 * We assume we're being called with sp->fip pointing at blank space. 1992 * Account for a new FIP in the segment header, and set sp->vp. 1993 * (This should duplicate the setup at the top of lfs_writefile().) 1994 */ 1995 sp = fs->lfs_sp; 1996 if (!seglocked) { 1997 if (sp->seg_bytes_left < fs->lfs_bsize || 1998 sp->sum_bytes_left < sizeof(struct finfo)) 1999 (void) lfs_writeseg(fs, fs->lfs_sp); 2000 2001 sp->sum_bytes_left -= FINFOSIZE; 2002 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 2003 } 2004 KASSERT(sp->vp == NULL); 2005 sp->vp = vp; 2006 2007 if (!seglocked) { 2008 if (vp->v_flag & VDIROP) 2009 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2010 } 2011 2012 sp->fip->fi_nblocks = 0; 2013 sp->fip->fi_ino = ip->i_number; 2014 sp->fip->fi_version = ip->i_gen; 2015 2016 /* 2017 * Loop through genfs_putpages until all pages are gathered. 2018 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2019 * Whenever we lose the interlock we have to rerun check_dirty, as 2020 * well. 2021 */ 2022 again: 2023 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2024 ap->a_flags, 0) < 0) { 2025 simple_unlock(&vp->v_interlock); 2026 sp->vp = NULL; 2027 if (!seglocked) 2028 lfs_segunlock(fs); 2029 return EDEADLK; 2030 } 2031 2032 error = genfs_putpages(v); 2033 if (error == EDEADLK || error == EAGAIN) { 2034 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2035 " EDEADLK [2] ino %d off %x (seg %d)\n", 2036 ip->i_number, fs->lfs_offset, 2037 dtosn(fs, fs->lfs_offset))); 2038 /* If nothing to write, short-circuit */ 2039 if (sp->cbpp - sp->bpp > 1) { 2040 /* Write gathered pages */ 2041 lfs_updatemeta(sp); 2042 (void) lfs_writeseg(fs, sp); 2043 2044 /* 2045 * Reinitialize brand new FIP and add us to it. 2046 * (This should duplicate the fixup in 2047 * lfs_gatherpages().) 2048 */ 2049 KASSERT(sp->vp == vp); 2050 sp->fip->fi_version = ip->i_gen; 2051 sp->fip->fi_ino = ip->i_number; 2052 /* Add us to the new segment summary. */ 2053 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 2054 sp->sum_bytes_left -= FINFOSIZE; 2055 } 2056 2057 /* Give the write a chance to complete */ 2058 preempt(1); 2059 2060 /* We've lost the interlock. Start over. */ 2061 if (error == EDEADLK) { 2062 simple_lock(&vp->v_interlock); 2063 goto again; 2064 } 2065 } 2066 2067 KASSERT(sp->vp == vp); 2068 if (!seglocked) { 2069 sp->vp = NULL; /* XXX lfs_gather below will set this */ 2070 2071 /* Write indirect blocks as well */ 2072 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2073 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2074 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2075 2076 KASSERT(sp->vp == NULL); 2077 sp->vp = vp; 2078 } 2079 2080 /* 2081 * Blocks are now gathered into a segment waiting to be written. 2082 * All that's left to do is update metadata, and write them. 2083 */ 2084 lfs_updatemeta(sp); 2085 KASSERT(sp->vp == vp); 2086 sp->vp = NULL; 2087 2088 if (seglocked) { 2089 /* we're called by lfs_writefile. */ 2090 return error; 2091 } 2092 2093 /* 2094 * Clean up FIP, since we're done writing this file. 2095 * This should duplicate cleanup at the end of lfs_writefile(). 2096 */ 2097 if (sp->fip->fi_nblocks != 0) { 2098 sp->fip = (FINFO*)((caddr_t)sp->fip + FINFOSIZE + 2099 sizeof(int32_t) * sp->fip->fi_nblocks); 2100 sp->start_lbp = &sp->fip->fi_blocks[0]; 2101 } else { 2102 sp->sum_bytes_left += FINFOSIZE; 2103 --((SEGSUM *)(sp->segsum))->ss_nfinfo; 2104 } 2105 lfs_writeseg(fs, fs->lfs_sp); 2106 2107 /* 2108 * Remove us from paging queue, since we've now written all our 2109 * pages. 2110 */ 2111 simple_lock(&fs->lfs_interlock); 2112 if (ip->i_flags & IN_PAGING) { 2113 ip->i_flags &= ~IN_PAGING; 2114 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2115 } 2116 simple_unlock(&fs->lfs_interlock); 2117 2118 /* 2119 * XXX - with the malloc/copy writeseg, the pages are freed by now 2120 * even if we don't wait (e.g. if we hold a nested lock). This 2121 * will not be true if we stop using malloc/copy. 2122 */ 2123 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2124 lfs_segunlock(fs); 2125 2126 /* 2127 * Wait for v_numoutput to drop to zero. The seglock should 2128 * take care of this, but there is a slight possibility that 2129 * aiodoned might not have got around to our buffers yet. 2130 */ 2131 if (sync) { 2132 s = splbio(); 2133 simple_lock(&global_v_numoutput_slock); 2134 while (vp->v_numoutput > 0) { 2135 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2136 " num %d\n", ip->i_number, vp->v_numoutput)); 2137 vp->v_flag |= VBWAIT; 2138 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 2139 &global_v_numoutput_slock); 2140 } 2141 simple_unlock(&global_v_numoutput_slock); 2142 splx(s); 2143 } 2144 return error; 2145 } 2146 2147 /* 2148 * Return the last logical file offset that should be written for this file 2149 * if we're doing a write that ends at "size". If writing, we need to know 2150 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2151 * to know about entire blocks. 2152 */ 2153 void 2154 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2155 { 2156 struct inode *ip = VTOI(vp); 2157 struct lfs *fs = ip->i_lfs; 2158 daddr_t olbn, nlbn; 2159 2160 olbn = lblkno(fs, ip->i_size); 2161 nlbn = lblkno(fs, size); 2162 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2163 *eobp = fragroundup(fs, size); 2164 } else { 2165 *eobp = blkroundup(fs, size); 2166 } 2167 } 2168 2169 #ifdef DEBUG 2170 void lfs_dump_vop(void *); 2171 2172 void 2173 lfs_dump_vop(void *v) 2174 { 2175 struct vop_putpages_args /* { 2176 struct vnode *a_vp; 2177 voff_t a_offlo; 2178 voff_t a_offhi; 2179 int a_flags; 2180 } */ *ap = v; 2181 2182 #ifdef DDB 2183 vfs_vnode_print(ap->a_vp, 0, printf); 2184 #endif 2185 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2186 } 2187 #endif 2188 2189 int 2190 lfs_mmap(void *v) 2191 { 2192 struct vop_mmap_args /* { 2193 const struct vnodeop_desc *a_desc; 2194 struct vnode *a_vp; 2195 int a_fflags; 2196 struct ucred *a_cred; 2197 struct lwp *a_l; 2198 } */ *ap = v; 2199 2200 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2201 return EOPNOTSUPP; 2202 return ufs_mmap(v); 2203 } 2204
Indexes created Mon Sep 29 21:09:56 GMT 2025