lfs_vnops.c revision 1.157
1 /* $NetBSD: lfs_vnops.c,v 1.157 2005/12/11 12:25:26 christos 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 1.157 2005/12/11 12:25:26 christos 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 /* 278 * Trickle sync checks for need to do a checkpoint after possible 279 * activity from the pagedaemon. 280 */ 281 if (ap->a_flags & FSYNC_LAZY) { 282 simple_lock(&lfs_subsys_lock); 283 wakeup(&lfs_writer_daemon); 284 simple_unlock(&lfs_subsys_lock); 285 return 0; 286 } 287 288 wait = (ap->a_flags & FSYNC_WAIT); 289 simple_lock(&vp->v_interlock); 290 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 291 round_page(ap->a_offhi), 292 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 293 if (error) 294 return error; 295 error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 296 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 297 int l = 0; 298 error = VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE, 299 ap->a_l->l_proc->p_ucred, ap->a_l); 300 } 301 if (wait && !VPISEMPTY(vp)) 302 LFS_SET_UINO(VTOI(vp), IN_MODIFIED); 303 304 return error; 305 } 306 307 /* 308 * Take IN_ADIROP off, then call ufs_inactive. 309 */ 310 int 311 lfs_inactive(void *v) 312 { 313 struct vop_inactive_args /* { 314 struct vnode *a_vp; 315 struct lwp *a_l; 316 } */ *ap = v; 317 318 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 319 320 lfs_unmark_vnode(ap->a_vp); 321 322 /* 323 * The Ifile is only ever inactivated on unmount. 324 * Streamline this process by not giving it more dirty blocks. 325 */ 326 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 327 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 328 VOP_UNLOCK(ap->a_vp, 0); 329 return 0; 330 } 331 332 return ufs_inactive(v); 333 } 334 335 /* 336 * These macros are used to bracket UFS directory ops, so that we can 337 * identify all the pages touched during directory ops which need to 338 * be ordered and flushed atomically, so that they may be recovered. 339 * 340 * Because we have to mark nodes VDIROP in order to prevent 341 * the cache from reclaiming them while a dirop is in progress, we must 342 * also manage the number of nodes so marked (otherwise we can run out). 343 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 344 * is decremented during segment write, when VDIROP is taken off. 345 */ 346 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 347 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 348 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 349 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 350 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 351 static int lfs_set_dirop(struct vnode *, struct vnode *); 352 353 static int 354 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 355 { 356 struct lfs *fs; 357 int error; 358 359 KASSERT(VOP_ISLOCKED(dvp)); 360 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 361 362 fs = VTOI(dvp)->i_lfs; 363 364 ASSERT_NO_SEGLOCK(fs); 365 /* 366 * LFS_NRESERVE calculates direct and indirect blocks as well 367 * as an inode block; an overestimate in most cases. 368 */ 369 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 370 return (error); 371 372 restart: 373 simple_lock(&fs->lfs_interlock); 374 if (fs->lfs_dirops == 0) { 375 simple_unlock(&fs->lfs_interlock); 376 lfs_check(dvp, LFS_UNUSED_LBN, 0); 377 simple_lock(&fs->lfs_interlock); 378 } 379 while (fs->lfs_writer) 380 ltsleep(&fs->lfs_dirops, (PRIBIO + 1), "lfs_sdirop", 0, 381 &fs->lfs_interlock); 382 simple_lock(&lfs_subsys_lock); 383 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 384 wakeup(&lfs_writer_daemon); 385 simple_unlock(&lfs_subsys_lock); 386 simple_unlock(&fs->lfs_interlock); 387 preempt(1); 388 goto restart; 389 } 390 391 if (lfs_dirvcount > LFS_MAX_DIROP) { 392 simple_unlock(&fs->lfs_interlock); 393 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 394 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 395 if ((error = ltsleep(&lfs_dirvcount, 396 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 397 &lfs_subsys_lock)) != 0) { 398 goto unreserve; 399 } 400 goto restart; 401 } 402 simple_unlock(&lfs_subsys_lock); 403 404 ++fs->lfs_dirops; 405 fs->lfs_doifile = 1; 406 simple_unlock(&fs->lfs_interlock); 407 408 /* Hold a reference so SET_ENDOP will be happy */ 409 vref(dvp); 410 if (vp) { 411 vref(vp); 412 MARK_VNODE(vp); 413 } 414 415 MARK_VNODE(dvp); 416 return 0; 417 418 unreserve: 419 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 420 return error; 421 } 422 423 /* 424 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 425 * in getnewvnode(), if we have a stacked filesystem mounted on top 426 * of us. 427 * 428 * NB: this means we have to clear the new vnodes on error. Fortunately 429 * SET_ENDOP is there to do that for us. 430 */ 431 static int 432 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 433 { 434 int error; 435 struct lfs *fs; 436 437 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 438 ASSERT_NO_SEGLOCK(fs); 439 if (fs->lfs_ronly) 440 return EROFS; 441 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 442 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 443 dvp, error)); 444 return error; 445 } 446 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 447 if (vpp) { 448 ungetnewvnode(*vpp); 449 *vpp = NULL; 450 } 451 return error; 452 } 453 return 0; 454 } 455 456 #define SET_ENDOP_BASE(fs, dvp, str) \ 457 do { \ 458 simple_lock(&(fs)->lfs_interlock); \ 459 --(fs)->lfs_dirops; \ 460 if (!(fs)->lfs_dirops) { \ 461 if ((fs)->lfs_nadirop) { \ 462 panic("SET_ENDOP: %s: no dirops but " \ 463 " nadirop=%d", (str), \ 464 (fs)->lfs_nadirop); \ 465 } \ 466 wakeup(&(fs)->lfs_writer); \ 467 simple_unlock(&(fs)->lfs_interlock); \ 468 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 469 } else \ 470 simple_unlock(&(fs)->lfs_interlock); \ 471 } while(0) 472 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 473 do { \ 474 UNMARK_VNODE(dvp); \ 475 if (nvpp && *nvpp) \ 476 UNMARK_VNODE(*nvpp); \ 477 /* Check for error return to stem vnode leakage */ \ 478 if (nvpp && *nvpp && !((*nvpp)->v_flag & VDIROP)) \ 479 ungetnewvnode(*(nvpp)); \ 480 SET_ENDOP_BASE((fs), (dvp), (str)); \ 481 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 482 vrele(dvp); \ 483 } while(0) 484 #define SET_ENDOP_CREATE_AP(ap, str) \ 485 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 486 (ap)->a_vpp, (str)) 487 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 488 do { \ 489 UNMARK_VNODE(dvp); \ 490 if (ovp) \ 491 UNMARK_VNODE(ovp); \ 492 SET_ENDOP_BASE((fs), (dvp), (str)); \ 493 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 494 vrele(dvp); \ 495 if (ovp) \ 496 vrele(ovp); \ 497 } while(0) 498 499 void 500 lfs_mark_vnode(struct vnode *vp) 501 { 502 struct inode *ip = VTOI(vp); 503 struct lfs *fs = ip->i_lfs; 504 505 simple_lock(&fs->lfs_interlock); 506 if (!(ip->i_flag & IN_ADIROP)) { 507 if (!(vp->v_flag & VDIROP)) { 508 (void)lfs_vref(vp); 509 simple_lock(&lfs_subsys_lock); 510 ++lfs_dirvcount; 511 simple_unlock(&lfs_subsys_lock); 512 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 513 vp->v_flag |= VDIROP; 514 } 515 ++fs->lfs_nadirop; 516 ip->i_flag |= IN_ADIROP; 517 } else 518 KASSERT(vp->v_flag & VDIROP); 519 simple_unlock(&fs->lfs_interlock); 520 } 521 522 void 523 lfs_unmark_vnode(struct vnode *vp) 524 { 525 struct inode *ip = VTOI(vp); 526 527 if (ip && (ip->i_flag & IN_ADIROP)) { 528 KASSERT(vp->v_flag & VDIROP); 529 simple_lock(&ip->i_lfs->lfs_interlock); 530 --ip->i_lfs->lfs_nadirop; 531 simple_unlock(&ip->i_lfs->lfs_interlock); 532 ip->i_flag &= ~IN_ADIROP; 533 } 534 } 535 536 int 537 lfs_symlink(void *v) 538 { 539 struct vop_symlink_args /* { 540 struct vnode *a_dvp; 541 struct vnode **a_vpp; 542 struct componentname *a_cnp; 543 struct vattr *a_vap; 544 char *a_target; 545 } */ *ap = v; 546 int error; 547 548 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 549 vput(ap->a_dvp); 550 return error; 551 } 552 error = ufs_symlink(ap); 553 SET_ENDOP_CREATE_AP(ap, "symlink"); 554 return (error); 555 } 556 557 int 558 lfs_mknod(void *v) 559 { 560 struct vop_mknod_args /* { 561 struct vnode *a_dvp; 562 struct vnode **a_vpp; 563 struct componentname *a_cnp; 564 struct vattr *a_vap; 565 } */ *ap = v; 566 struct vattr *vap = ap->a_vap; 567 struct vnode **vpp = ap->a_vpp; 568 struct inode *ip; 569 int error; 570 struct mount *mp; 571 ino_t ino; 572 573 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 574 vput(ap->a_dvp); 575 return error; 576 } 577 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 578 ap->a_dvp, vpp, ap->a_cnp); 579 580 /* Either way we're done with the dirop at this point */ 581 SET_ENDOP_CREATE_AP(ap, "mknod"); 582 583 if (error) 584 return (error); 585 586 ip = VTOI(*vpp); 587 mp = (*vpp)->v_mount; 588 ino = ip->i_number; 589 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 590 if (vap->va_rdev != VNOVAL) { 591 /* 592 * Want to be able to use this to make badblock 593 * inodes, so don't truncate the dev number. 594 */ 595 #if 0 596 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 597 UFS_MPNEEDSWAP((*vpp)->v_mount)); 598 #else 599 ip->i_ffs1_rdev = vap->va_rdev; 600 #endif 601 } 602 603 /* 604 * Call fsync to write the vnode so that we don't have to deal with 605 * flushing it when it's marked VDIROP|VXLOCK. 606 * 607 * XXX KS - If we can't flush we also can't call vgone(), so must 608 * return. But, that leaves this vnode in limbo, also not good. 609 * Can this ever happen (barring hardware failure)? 610 */ 611 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 612 curlwp)) != 0) { 613 panic("lfs_mknod: couldn't fsync (ino %llu)", 614 (unsigned long long)ino); 615 /* return (error); */ 616 } 617 /* 618 * Remove vnode so that it will be reloaded by VFS_VGET and 619 * checked to see if it is an alias of an existing entry in 620 * the inode cache. 621 */ 622 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 623 624 VOP_UNLOCK(*vpp, 0); 625 lfs_vunref(*vpp); 626 (*vpp)->v_type = VNON; 627 vgone(*vpp); 628 error = VFS_VGET(mp, ino, vpp); 629 630 if (error != 0) { 631 *vpp = NULL; 632 return (error); 633 } 634 return (0); 635 } 636 637 int 638 lfs_create(void *v) 639 { 640 struct vop_create_args /* { 641 struct vnode *a_dvp; 642 struct vnode **a_vpp; 643 struct componentname *a_cnp; 644 struct vattr *a_vap; 645 } */ *ap = v; 646 int error; 647 648 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 649 vput(ap->a_dvp); 650 return error; 651 } 652 error = ufs_create(ap); 653 SET_ENDOP_CREATE_AP(ap, "create"); 654 return (error); 655 } 656 657 int 658 lfs_mkdir(void *v) 659 { 660 struct vop_mkdir_args /* { 661 struct vnode *a_dvp; 662 struct vnode **a_vpp; 663 struct componentname *a_cnp; 664 struct vattr *a_vap; 665 } */ *ap = v; 666 int error; 667 668 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 669 vput(ap->a_dvp); 670 return error; 671 } 672 error = ufs_mkdir(ap); 673 SET_ENDOP_CREATE_AP(ap, "mkdir"); 674 return (error); 675 } 676 677 int 678 lfs_remove(void *v) 679 { 680 struct vop_remove_args /* { 681 struct vnode *a_dvp; 682 struct vnode *a_vp; 683 struct componentname *a_cnp; 684 } */ *ap = v; 685 struct vnode *dvp, *vp; 686 int error; 687 688 dvp = ap->a_dvp; 689 vp = ap->a_vp; 690 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 691 if (dvp == vp) 692 vrele(vp); 693 else 694 vput(vp); 695 vput(dvp); 696 return error; 697 } 698 error = ufs_remove(ap); 699 SET_ENDOP_REMOVE(VTOI(dvp)->i_lfs, dvp, vp, "remove"); 700 return (error); 701 } 702 703 int 704 lfs_rmdir(void *v) 705 { 706 struct vop_rmdir_args /* { 707 struct vnodeop_desc *a_desc; 708 struct vnode *a_dvp; 709 struct vnode *a_vp; 710 struct componentname *a_cnp; 711 } */ *ap = v; 712 struct vnode *vp; 713 int error; 714 715 vp = ap->a_vp; 716 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 717 vrele(ap->a_dvp); 718 if (ap->a_vp != ap->a_dvp) 719 VOP_UNLOCK(ap->a_dvp, 0); 720 vput(vp); 721 return error; 722 } 723 error = ufs_rmdir(ap); 724 SET_ENDOP_REMOVE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vp, "rmdir"); 725 return (error); 726 } 727 728 int 729 lfs_link(void *v) 730 { 731 struct vop_link_args /* { 732 struct vnode *a_dvp; 733 struct vnode *a_vp; 734 struct componentname *a_cnp; 735 } */ *ap = v; 736 int error; 737 struct vnode **vpp = NULL; 738 739 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 740 vput(ap->a_dvp); 741 return error; 742 } 743 error = ufs_link(ap); 744 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 745 return (error); 746 } 747 748 int 749 lfs_rename(void *v) 750 { 751 struct vop_rename_args /* { 752 struct vnode *a_fdvp; 753 struct vnode *a_fvp; 754 struct componentname *a_fcnp; 755 struct vnode *a_tdvp; 756 struct vnode *a_tvp; 757 struct componentname *a_tcnp; 758 } */ *ap = v; 759 struct vnode *tvp, *fvp, *tdvp, *fdvp; 760 struct componentname *tcnp, *fcnp; 761 int error; 762 struct lfs *fs; 763 764 fs = VTOI(ap->a_fdvp)->i_lfs; 765 tvp = ap->a_tvp; 766 tdvp = ap->a_tdvp; 767 tcnp = ap->a_tcnp; 768 fvp = ap->a_fvp; 769 fdvp = ap->a_fdvp; 770 fcnp = ap->a_fcnp; 771 772 /* 773 * Check for cross-device rename. 774 * If it is, we don't want to set dirops, just error out. 775 * (In particular note that MARK_VNODE(tdvp) will DTWT on 776 * a cross-device rename.) 777 * 778 * Copied from ufs_rename. 779 */ 780 if ((fvp->v_mount != tdvp->v_mount) || 781 (tvp && (fvp->v_mount != tvp->v_mount))) { 782 error = EXDEV; 783 goto errout; 784 } 785 786 /* 787 * Check to make sure we're not renaming a vnode onto itself 788 * (deleting a hard link by renaming one name onto another); 789 * if we are we can't recursively call VOP_REMOVE since that 790 * would leave us with an unaccounted-for number of live dirops. 791 * 792 * Inline the relevant section of ufs_rename here, *before* 793 * calling SET_DIROP_REMOVE. 794 */ 795 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 796 (VTOI(tdvp)->i_flags & APPEND))) { 797 error = EPERM; 798 goto errout; 799 } 800 if (fvp == tvp) { 801 if (fvp->v_type == VDIR) { 802 error = EINVAL; 803 goto errout; 804 } 805 806 /* Release destination completely. */ 807 VOP_ABORTOP(tdvp, tcnp); 808 vput(tdvp); 809 vput(tvp); 810 811 /* Delete source. */ 812 vrele(fvp); 813 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 814 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 815 fcnp->cn_nameiop = DELETE; 816 if ((error = relookup(fdvp, &fvp, fcnp))){ 817 /* relookup blew away fdvp */ 818 return (error); 819 } 820 return (VOP_REMOVE(fdvp, fvp, fcnp)); 821 } 822 823 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 824 goto errout; 825 MARK_VNODE(fdvp); 826 MARK_VNODE(fvp); 827 828 error = ufs_rename(ap); 829 UNMARK_VNODE(fdvp); 830 UNMARK_VNODE(fvp); 831 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 832 return (error); 833 834 errout: 835 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 836 if (tdvp == tvp) 837 vrele(tdvp); 838 else 839 vput(tdvp); 840 if (tvp) 841 vput(tvp); 842 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 843 vrele(fdvp); 844 vrele(fvp); 845 return (error); 846 } 847 848 /* XXX hack to avoid calling ITIMES in getattr */ 849 int 850 lfs_getattr(void *v) 851 { 852 struct vop_getattr_args /* { 853 struct vnode *a_vp; 854 struct vattr *a_vap; 855 struct ucred *a_cred; 856 struct lwp *a_l; 857 } */ *ap = v; 858 struct vnode *vp = ap->a_vp; 859 struct inode *ip = VTOI(vp); 860 struct vattr *vap = ap->a_vap; 861 struct lfs *fs = ip->i_lfs; 862 /* 863 * Copy from inode table 864 */ 865 vap->va_fsid = ip->i_dev; 866 vap->va_fileid = ip->i_number; 867 vap->va_mode = ip->i_mode & ~IFMT; 868 vap->va_nlink = ip->i_nlink; 869 vap->va_uid = ip->i_uid; 870 vap->va_gid = ip->i_gid; 871 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 872 vap->va_size = vp->v_size; 873 vap->va_atime.tv_sec = ip->i_ffs1_atime; 874 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 875 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 876 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 877 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 878 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 879 vap->va_flags = ip->i_flags; 880 vap->va_gen = ip->i_gen; 881 /* this doesn't belong here */ 882 if (vp->v_type == VBLK) 883 vap->va_blocksize = BLKDEV_IOSIZE; 884 else if (vp->v_type == VCHR) 885 vap->va_blocksize = MAXBSIZE; 886 else 887 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 888 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 889 vap->va_type = vp->v_type; 890 vap->va_filerev = ip->i_modrev; 891 return (0); 892 } 893 894 /* 895 * Check to make sure the inode blocks won't choke the buffer 896 * cache, then call ufs_setattr as usual. 897 */ 898 int 899 lfs_setattr(void *v) 900 { 901 struct vop_setattr_args /* { 902 struct vnode *a_vp; 903 struct vattr *a_vap; 904 struct ucred *a_cred; 905 struct lwp *a_l; 906 } */ *ap = v; 907 struct vnode *vp = ap->a_vp; 908 909 lfs_check(vp, LFS_UNUSED_LBN, 0); 910 return ufs_setattr(v); 911 } 912 913 /* 914 * Close called 915 * 916 * XXX -- we were using ufs_close, but since it updates the 917 * times on the inode, we might need to bump the uinodes 918 * count. 919 */ 920 /* ARGSUSED */ 921 int 922 lfs_close(void *v) 923 { 924 struct vop_close_args /* { 925 struct vnode *a_vp; 926 int a_fflag; 927 struct ucred *a_cred; 928 struct lwp *a_l; 929 } */ *ap = v; 930 struct vnode *vp = ap->a_vp; 931 struct inode *ip = VTOI(vp); 932 933 if (vp == ip->i_lfs->lfs_ivnode && 934 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 935 return 0; 936 937 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 938 LFS_ITIMES(ip, NULL, NULL, NULL); 939 } 940 return (0); 941 } 942 943 /* 944 * Close wrapper for special devices. 945 * 946 * Update the times on the inode then do device close. 947 */ 948 int 949 lfsspec_close(void *v) 950 { 951 struct vop_close_args /* { 952 struct vnode *a_vp; 953 int a_fflag; 954 struct ucred *a_cred; 955 struct lwp *a_l; 956 } */ *ap = v; 957 struct vnode *vp; 958 struct inode *ip; 959 960 vp = ap->a_vp; 961 ip = VTOI(vp); 962 if (vp->v_usecount > 1) { 963 LFS_ITIMES(ip, NULL, NULL, NULL); 964 } 965 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 966 } 967 968 /* 969 * Close wrapper for fifo's. 970 * 971 * Update the times on the inode then do device close. 972 */ 973 int 974 lfsfifo_close(void *v) 975 { 976 struct vop_close_args /* { 977 struct vnode *a_vp; 978 int a_fflag; 979 struct ucred *a_cred; 980 struct lwp *a_l; 981 } */ *ap = v; 982 struct vnode *vp; 983 struct inode *ip; 984 985 vp = ap->a_vp; 986 ip = VTOI(vp); 987 if (ap->a_vp->v_usecount > 1) { 988 LFS_ITIMES(ip, NULL, NULL, NULL); 989 } 990 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 991 } 992 993 /* 994 * Reclaim an inode so that it can be used for other purposes. 995 */ 996 997 int 998 lfs_reclaim(void *v) 999 { 1000 struct vop_reclaim_args /* { 1001 struct vnode *a_vp; 1002 struct lwp *a_l; 1003 } */ *ap = v; 1004 struct vnode *vp = ap->a_vp; 1005 struct inode *ip = VTOI(vp); 1006 int error; 1007 1008 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1009 1010 LFS_CLR_UINO(ip, IN_ALLMOD); 1011 if ((error = ufs_reclaim(vp, ap->a_l))) 1012 return (error); 1013 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1014 lfs_deregister_all(vp); 1015 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1016 ip->inode_ext.lfs = NULL; 1017 pool_put(&lfs_inode_pool, vp->v_data); 1018 vp->v_data = NULL; 1019 return (0); 1020 } 1021 1022 /* 1023 * Read a block from a storage device. 1024 * In order to avoid reading blocks that are in the process of being 1025 * written by the cleaner---and hence are not mutexed by the normal 1026 * buffer cache / page cache mechanisms---check for collisions before 1027 * reading. 1028 * 1029 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1030 * the active cleaner test. 1031 * 1032 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1033 */ 1034 int 1035 lfs_strategy(void *v) 1036 { 1037 struct vop_strategy_args /* { 1038 struct vnode *a_vp; 1039 struct buf *a_bp; 1040 } */ *ap = v; 1041 struct buf *bp; 1042 struct lfs *fs; 1043 struct vnode *vp; 1044 struct inode *ip; 1045 daddr_t tbn; 1046 int i, sn, error, slept; 1047 1048 bp = ap->a_bp; 1049 vp = ap->a_vp; 1050 ip = VTOI(vp); 1051 fs = ip->i_lfs; 1052 1053 /* lfs uses its strategy routine only for read */ 1054 KASSERT(bp->b_flags & B_READ); 1055 1056 if (vp->v_type == VBLK || vp->v_type == VCHR) 1057 panic("lfs_strategy: spec"); 1058 KASSERT(bp->b_bcount != 0); 1059 if (bp->b_blkno == bp->b_lblkno) { 1060 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1061 NULL); 1062 if (error) { 1063 bp->b_error = error; 1064 bp->b_flags |= B_ERROR; 1065 biodone(bp); 1066 return (error); 1067 } 1068 if ((long)bp->b_blkno == -1) /* no valid data */ 1069 clrbuf(bp); 1070 } 1071 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1072 biodone(bp); 1073 return (0); 1074 } 1075 1076 slept = 1; 1077 simple_lock(&fs->lfs_interlock); 1078 while (slept && fs->lfs_seglock) { 1079 simple_unlock(&fs->lfs_interlock); 1080 /* 1081 * Look through list of intervals. 1082 * There will only be intervals to look through 1083 * if the cleaner holds the seglock. 1084 * Since the cleaner is synchronous, we can trust 1085 * the list of intervals to be current. 1086 */ 1087 tbn = dbtofsb(fs, bp->b_blkno); 1088 sn = dtosn(fs, tbn); 1089 slept = 0; 1090 for (i = 0; i < fs->lfs_cleanind; i++) { 1091 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1092 tbn >= fs->lfs_cleanint[i]) { 1093 DLOG((DLOG_CLEAN, 1094 "lfs_strategy: ino %d lbn %" PRId64 1095 " ind %d sn %d fsb %" PRIx32 1096 " given sn %d fsb %" PRIx64 "\n", 1097 ip->i_number, bp->b_lblkno, i, 1098 dtosn(fs, fs->lfs_cleanint[i]), 1099 fs->lfs_cleanint[i], sn, tbn)); 1100 DLOG((DLOG_CLEAN, 1101 "lfs_strategy: sleeping on ino %d lbn %" 1102 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1103 simple_lock(&fs->lfs_interlock); 1104 if (fs->lfs_seglock) 1105 ltsleep(&fs->lfs_seglock, 1106 (PRIBIO + 1) | PNORELOCK, 1107 "lfs_strategy", 0, 1108 &fs->lfs_interlock); 1109 /* Things may be different now; start over. */ 1110 slept = 1; 1111 break; 1112 } 1113 } 1114 simple_lock(&fs->lfs_interlock); 1115 } 1116 simple_unlock(&fs->lfs_interlock); 1117 1118 vp = ip->i_devvp; 1119 VOP_STRATEGY(vp, bp); 1120 return (0); 1121 } 1122 1123 static void 1124 lfs_flush_dirops(struct lfs *fs) 1125 { 1126 struct inode *ip, *nip; 1127 struct vnode *vp; 1128 extern int lfs_dostats; 1129 struct segment *sp; 1130 int needunlock; 1131 1132 ASSERT_NO_SEGLOCK(fs); 1133 1134 if (fs->lfs_ronly) 1135 return; 1136 1137 simple_lock(&fs->lfs_interlock); 1138 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1139 simple_unlock(&fs->lfs_interlock); 1140 return; 1141 } else 1142 simple_unlock(&fs->lfs_interlock); 1143 1144 if (lfs_dostats) 1145 ++lfs_stats.flush_invoked; 1146 1147 /* 1148 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1149 * Technically this is a checkpoint (the on-disk state is valid) 1150 * even though we are leaving out all the file data. 1151 */ 1152 lfs_imtime(fs); 1153 lfs_seglock(fs, SEGM_CKP); 1154 sp = fs->lfs_sp; 1155 1156 /* 1157 * lfs_writevnodes, optimized to get dirops out of the way. 1158 * Only write dirops, and don't flush files' pages, only 1159 * blocks from the directories. 1160 * 1161 * We don't need to vref these files because they are 1162 * dirops and so hold an extra reference until the 1163 * segunlock clears them of that status. 1164 * 1165 * We don't need to check for IN_ADIROP because we know that 1166 * no dirops are active. 1167 * 1168 */ 1169 simple_lock(&fs->lfs_interlock); 1170 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1171 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1172 simple_unlock(&fs->lfs_interlock); 1173 vp = ITOV(ip); 1174 1175 /* 1176 * All writes to directories come from dirops; all 1177 * writes to files' direct blocks go through the page 1178 * cache, which we're not touching. Reads to files 1179 * and/or directories will not be affected by writing 1180 * directory blocks inodes and file inodes. So we don't 1181 * really need to lock. If we don't lock, though, 1182 * make sure that we don't clear IN_MODIFIED 1183 * unnecessarily. 1184 */ 1185 if (vp->v_flag & VXLOCK) 1186 continue; 1187 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT) == 0) { 1188 needunlock = 1; 1189 } else { 1190 DLOG((DLOG_VNODE, "lfs_flush_dirops: flushing locked ino %d\n", 1191 VTOI(vp)->i_number)); 1192 needunlock = 0; 1193 } 1194 if (vp->v_type != VREG && 1195 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1196 lfs_writefile(fs, sp, vp); 1197 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1198 !(ip->i_flag & IN_ALLMOD)) { 1199 LFS_SET_UINO(ip, IN_MODIFIED); 1200 } 1201 } 1202 (void) lfs_writeinode(fs, sp, ip); 1203 if (needunlock) 1204 VOP_UNLOCK(vp, 0); 1205 else 1206 LFS_SET_UINO(ip, IN_MODIFIED); 1207 simple_lock(&fs->lfs_interlock); 1208 } 1209 simple_unlock(&fs->lfs_interlock); 1210 /* We've written all the dirops there are */ 1211 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1212 (void) lfs_writeseg(fs, sp); 1213 lfs_segunlock(fs); 1214 } 1215 1216 /* 1217 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1218 */ 1219 int 1220 lfs_fcntl(void *v) 1221 { 1222 struct vop_fcntl_args /* { 1223 struct vnode *a_vp; 1224 u_long a_command; 1225 caddr_t a_data; 1226 int a_fflag; 1227 struct ucred *a_cred; 1228 struct lwp *a_l; 1229 } */ *ap = v; 1230 struct timeval *tvp; 1231 BLOCK_INFO *blkiov; 1232 CLEANERINFO *cip; 1233 SEGUSE *sup; 1234 int blkcnt, error, oclean; 1235 struct lfs_fcntl_markv blkvp; 1236 struct proc *p; 1237 fsid_t *fsidp; 1238 struct lfs *fs; 1239 struct buf *bp; 1240 fhandle_t *fhp; 1241 daddr_t off; 1242 1243 /* Only respect LFS fcntls on fs root or Ifile */ 1244 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1245 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1246 return ufs_fcntl(v); 1247 } 1248 1249 /* Avoid locking a draining lock */ 1250 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1251 return ESHUTDOWN; 1252 } 1253 1254 p = ap->a_l->l_proc; 1255 fs = VTOI(ap->a_vp)->i_lfs; 1256 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1257 1258 switch (ap->a_command) { 1259 case LFCNSEGWAITALL: 1260 case LFCNSEGWAITALL_COMPAT: 1261 fsidp = NULL; 1262 /* FALLSTHROUGH */ 1263 case LFCNSEGWAIT: 1264 case LFCNSEGWAIT_COMPAT: 1265 tvp = (struct timeval *)ap->a_data; 1266 simple_lock(&fs->lfs_interlock); 1267 ++fs->lfs_sleepers; 1268 simple_unlock(&fs->lfs_interlock); 1269 VOP_UNLOCK(ap->a_vp, 0); 1270 1271 error = lfs_segwait(fsidp, tvp); 1272 1273 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1274 simple_lock(&fs->lfs_interlock); 1275 if (--fs->lfs_sleepers == 0) 1276 wakeup(&fs->lfs_sleepers); 1277 simple_unlock(&fs->lfs_interlock); 1278 return error; 1279 1280 case LFCNBMAPV: 1281 case LFCNMARKV: 1282 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0) 1283 return (error); 1284 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1285 1286 blkcnt = blkvp.blkcnt; 1287 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1288 return (EINVAL); 1289 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1290 if ((error = copyin(blkvp.blkiov, blkiov, 1291 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1292 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1293 return error; 1294 } 1295 1296 simple_lock(&fs->lfs_interlock); 1297 ++fs->lfs_sleepers; 1298 simple_unlock(&fs->lfs_interlock); 1299 VOP_UNLOCK(ap->a_vp, 0); 1300 if (ap->a_command == LFCNBMAPV) 1301 error = lfs_bmapv(p, fsidp, blkiov, blkcnt); 1302 else /* LFCNMARKV */ 1303 error = lfs_markv(p, fsidp, blkiov, blkcnt); 1304 if (error == 0) 1305 error = copyout(blkiov, blkvp.blkiov, 1306 blkcnt * sizeof(BLOCK_INFO)); 1307 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1308 simple_lock(&fs->lfs_interlock); 1309 if (--fs->lfs_sleepers == 0) 1310 wakeup(&fs->lfs_sleepers); 1311 simple_unlock(&fs->lfs_interlock); 1312 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1313 return error; 1314 1315 case LFCNRECLAIM: 1316 /* 1317 * Flush dirops and write Ifile, allowing empty segments 1318 * to be immediately reclaimed. 1319 */ 1320 VOP_UNLOCK(ap->a_vp, 0); 1321 lfs_writer_enter(fs, "pndirop"); 1322 off = fs->lfs_offset; 1323 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1324 lfs_flush_dirops(fs); 1325 LFS_CLEANERINFO(cip, fs, bp); 1326 oclean = cip->clean; 1327 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1328 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1329 lfs_segunlock(fs); 1330 lfs_writer_leave(fs); 1331 1332 #ifdef DEBUG 1333 LFS_CLEANERINFO(cip, fs, bp); 1334 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1335 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1336 fs->lfs_offset - off, cip->clean - oclean, 1337 fs->lfs_activesb)); 1338 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1339 #endif 1340 1341 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1342 return 0; 1343 1344 case LFCNIFILEFH: 1345 /* Return the filehandle of the Ifile */ 1346 if ((error = suser(ap->a_l->l_proc->p_ucred, &ap->a_l->l_proc->p_acflag)) != 0) 1347 return (error); 1348 fhp = (struct fhandle *)ap->a_data; 1349 fhp->fh_fsid = *fsidp; 1350 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid)); 1351 1352 case LFCNREWIND: 1353 /* Move lfs_offset to the lowest-numbered segment */ 1354 return lfs_rewind(fs, *(int *)ap->a_data); 1355 1356 case LFCNINVAL: 1357 /* Mark a segment SEGUSE_INVAL */ 1358 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1359 if (sup->su_nbytes > 0) { 1360 brelse(bp); 1361 lfs_unset_inval_all(fs); 1362 return EBUSY; 1363 } 1364 sup->su_flags |= SEGUSE_INVAL; 1365 VOP_BWRITE(bp); 1366 return 0; 1367 1368 case LFCNRESIZE: 1369 /* Resize the filesystem */ 1370 return lfs_resize_fs(fs, *(int *)ap->a_data); 1371 1372 default: 1373 return ufs_fcntl(v); 1374 } 1375 return 0; 1376 } 1377 1378 int 1379 lfs_getpages(void *v) 1380 { 1381 struct vop_getpages_args /* { 1382 struct vnode *a_vp; 1383 voff_t a_offset; 1384 struct vm_page **a_m; 1385 int *a_count; 1386 int a_centeridx; 1387 vm_prot_t a_access_type; 1388 int a_advice; 1389 int a_flags; 1390 } */ *ap = v; 1391 1392 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1393 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1394 return EPERM; 1395 } 1396 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1397 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1398 } 1399 1400 /* 1401 * we're relying on the fact that genfs_getpages() always read in 1402 * entire filesystem blocks. 1403 */ 1404 return genfs_getpages(v); 1405 } 1406 1407 /* 1408 * Make sure that for all pages in every block in the given range, 1409 * either all are dirty or all are clean. If any of the pages 1410 * we've seen so far are dirty, put the vnode on the paging chain, 1411 * and mark it IN_PAGING. 1412 * 1413 * If checkfirst != 0, don't check all the pages but return at the 1414 * first dirty page. 1415 */ 1416 static int 1417 check_dirty(struct lfs *fs, struct vnode *vp, 1418 off_t startoffset, off_t endoffset, off_t blkeof, 1419 int flags, int checkfirst) 1420 { 1421 int by_list; 1422 struct vm_page *curpg = NULL; /* XXX: gcc */ 1423 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1424 off_t soff = 0; /* XXX: gcc */ 1425 voff_t off; 1426 int i; 1427 int nonexistent; 1428 int any_dirty; /* number of dirty pages */ 1429 int dirty; /* number of dirty pages in a block */ 1430 int tdirty; 1431 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1432 1433 ASSERT_MAYBE_SEGLOCK(fs); 1434 top: 1435 by_list = (vp->v_uobj.uo_npages <= 1436 ((endoffset - startoffset) >> PAGE_SHIFT) * 1437 UVM_PAGE_HASH_PENALTY); 1438 any_dirty = 0; 1439 1440 if (by_list) { 1441 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1442 } else { 1443 soff = startoffset; 1444 } 1445 while (by_list || soff < MIN(blkeof, endoffset)) { 1446 if (by_list) { 1447 /* 1448 * Find the first page in a block. Skip 1449 * blocks outside our area of interest or beyond 1450 * the end of file. 1451 */ 1452 if (pages_per_block > 1) { 1453 while (curpg && 1454 ((curpg->offset & fs->lfs_bmask) || 1455 curpg->offset >= vp->v_size || 1456 curpg->offset >= endoffset)) 1457 curpg = TAILQ_NEXT(curpg, listq); 1458 } 1459 if (curpg == NULL) 1460 break; 1461 soff = curpg->offset; 1462 } 1463 1464 /* 1465 * Mark all pages in extended range busy; find out if any 1466 * of them are dirty. 1467 */ 1468 nonexistent = dirty = 0; 1469 for (i = 0; i == 0 || i < pages_per_block; i++) { 1470 if (by_list && pages_per_block <= 1) { 1471 pgs[i] = pg = curpg; 1472 } else { 1473 off = soff + (i << PAGE_SHIFT); 1474 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1475 if (pg == NULL) { 1476 ++nonexistent; 1477 continue; 1478 } 1479 } 1480 KASSERT(pg != NULL); 1481 while (pg->flags & PG_BUSY) { 1482 pg->flags |= PG_WANTED; 1483 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1484 "lfsput", 0); 1485 simple_lock(&vp->v_interlock); 1486 if (by_list) { 1487 if (i > 0) 1488 uvm_page_unbusy(pgs, i); 1489 goto top; 1490 } 1491 } 1492 pg->flags |= PG_BUSY; 1493 UVM_PAGE_OWN(pg, "lfs_putpages"); 1494 1495 pmap_page_protect(pg, VM_PROT_NONE); 1496 tdirty = (pmap_clear_modify(pg) || 1497 (pg->flags & PG_CLEAN) == 0); 1498 dirty += tdirty; 1499 } 1500 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1501 if (by_list) { 1502 curpg = TAILQ_NEXT(curpg, listq); 1503 } else { 1504 soff += fs->lfs_bsize; 1505 } 1506 continue; 1507 } 1508 1509 any_dirty += dirty; 1510 KASSERT(nonexistent == 0); 1511 1512 /* 1513 * If any are dirty make all dirty; unbusy them, 1514 * but if we were asked to clean, wire them so that 1515 * the pagedaemon doesn't bother us about them while 1516 * they're on their way to disk. 1517 */ 1518 for (i = 0; i == 0 || i < pages_per_block; i++) { 1519 pg = pgs[i]; 1520 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1521 if (dirty) { 1522 pg->flags &= ~PG_CLEAN; 1523 if (flags & PGO_FREE) { 1524 /* 1525 * Wire the page so that 1526 * pdaemon doesn't see it again. 1527 */ 1528 uvm_lock_pageq(); 1529 uvm_pagewire(pg); 1530 uvm_unlock_pageq(); 1531 1532 /* Suspended write flag */ 1533 pg->flags |= PG_DELWRI; 1534 } 1535 } 1536 if (pg->flags & PG_WANTED) 1537 wakeup(pg); 1538 pg->flags &= ~(PG_WANTED|PG_BUSY); 1539 UVM_PAGE_OWN(pg, NULL); 1540 } 1541 1542 if (checkfirst && any_dirty) 1543 break; 1544 1545 if (by_list) { 1546 curpg = TAILQ_NEXT(curpg, listq); 1547 } else { 1548 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1549 } 1550 } 1551 1552 /* 1553 * If any pages were dirty, mark this inode as "pageout requested", 1554 * and put it on the paging queue. 1555 * XXXUBC locking (check locking on dchainhd too) 1556 */ 1557 #ifdef notyet 1558 if (any_dirty) { 1559 if (!(ip->i_flags & IN_PAGING)) { 1560 ip->i_flags |= IN_PAGING; 1561 simple_lock(&fs->lfs_interlock); 1562 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1563 simple_unlock(&fs->lfs_interlock); 1564 } 1565 } 1566 #endif 1567 return any_dirty; 1568 } 1569 1570 /* 1571 * lfs_putpages functions like genfs_putpages except that 1572 * 1573 * (1) It needs to bounds-check the incoming requests to ensure that 1574 * they are block-aligned; if they are not, expand the range and 1575 * do the right thing in case, e.g., the requested range is clean 1576 * but the expanded range is dirty. 1577 * (2) It needs to explicitly send blocks to be written when it is done. 1578 * VOP_PUTPAGES is not ever called with the seglock held, so 1579 * we simply take the seglock and let lfs_segunlock wait for us. 1580 * XXX Actually we can be called with the seglock held, if we have 1581 * XXX to flush a vnode while lfs_markv is in operation. As of this 1582 * XXX writing we panic in this case. 1583 * 1584 * Assumptions: 1585 * 1586 * (1) The caller does not hold any pages in this vnode busy. If it does, 1587 * there is a danger that when we expand the page range and busy the 1588 * pages we will deadlock. 1589 * (2) We are called with vp->v_interlock held; we must return with it 1590 * released. 1591 * (3) We don't absolutely have to free pages right away, provided that 1592 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1593 * us a request with PGO_FREE, we take the pages out of the paging 1594 * queue and wake up the writer, which will handle freeing them for us. 1595 * 1596 * We ensure that for any filesystem block, all pages for that 1597 * block are either resident or not, even if those pages are higher 1598 * than EOF; that means that we will be getting requests to free 1599 * "unused" pages above EOF all the time, and should ignore them. 1600 * 1601 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1602 */ 1603 1604 int 1605 lfs_putpages(void *v) 1606 { 1607 int error; 1608 struct vop_putpages_args /* { 1609 struct vnode *a_vp; 1610 voff_t a_offlo; 1611 voff_t a_offhi; 1612 int a_flags; 1613 } */ *ap = v; 1614 struct vnode *vp; 1615 struct inode *ip; 1616 struct lfs *fs; 1617 struct segment *sp; 1618 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1619 off_t off, max_endoffset; 1620 int s; 1621 boolean_t seglocked, sync, pagedaemon; 1622 struct vm_page *pg; 1623 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1624 1625 vp = ap->a_vp; 1626 ip = VTOI(vp); 1627 fs = ip->i_lfs; 1628 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1629 pagedaemon = (curproc == uvm.pagedaemon_proc); 1630 1631 /* Putpages does nothing for metadata. */ 1632 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1633 simple_unlock(&vp->v_interlock); 1634 return 0; 1635 } 1636 1637 /* 1638 * If there are no pages, don't do anything. 1639 */ 1640 if (vp->v_uobj.uo_npages == 0) { 1641 s = splbio(); 1642 if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL && 1643 (vp->v_flag & VONWORKLST)) { 1644 vp->v_flag &= ~VONWORKLST; 1645 LIST_REMOVE(vp, v_synclist); 1646 } 1647 splx(s); 1648 simple_unlock(&vp->v_interlock); 1649 return 0; 1650 } 1651 1652 blkeof = blkroundup(fs, ip->i_size); 1653 1654 /* 1655 * Ignore requests to free pages past EOF but in the same block 1656 * as EOF, unless the request is synchronous. (XXX why sync?) 1657 * XXXUBC Make these pages look "active" so the pagedaemon won't 1658 * XXXUBC bother us with them again. 1659 */ 1660 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1661 origoffset = ap->a_offlo; 1662 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1663 pg = uvm_pagelookup(&vp->v_uobj, off); 1664 KASSERT(pg != NULL); 1665 while (pg->flags & PG_BUSY) { 1666 pg->flags |= PG_WANTED; 1667 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1668 "lfsput2", 0); 1669 simple_lock(&vp->v_interlock); 1670 } 1671 uvm_lock_pageq(); 1672 uvm_pageactivate(pg); 1673 uvm_unlock_pageq(); 1674 } 1675 ap->a_offlo = blkeof; 1676 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1677 simple_unlock(&vp->v_interlock); 1678 return 0; 1679 } 1680 } 1681 1682 /* 1683 * Extend page range to start and end at block boundaries. 1684 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1685 */ 1686 origoffset = ap->a_offlo; 1687 origendoffset = ap->a_offhi; 1688 startoffset = origoffset & ~(fs->lfs_bmask); 1689 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1690 << fs->lfs_bshift; 1691 1692 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1693 endoffset = max_endoffset; 1694 origendoffset = endoffset; 1695 } else { 1696 origendoffset = round_page(ap->a_offhi); 1697 endoffset = round_page(blkroundup(fs, origendoffset)); 1698 } 1699 1700 KASSERT(startoffset > 0 || endoffset >= startoffset); 1701 if (startoffset == endoffset) { 1702 /* Nothing to do, why were we called? */ 1703 simple_unlock(&vp->v_interlock); 1704 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 1705 PRId64 "\n", startoffset)); 1706 return 0; 1707 } 1708 1709 ap->a_offlo = startoffset; 1710 ap->a_offhi = endoffset; 1711 1712 if (!(ap->a_flags & PGO_CLEANIT)) 1713 return genfs_putpages(v); 1714 1715 /* 1716 * If there are more than one page per block, we don't want 1717 * to get caught locking them backwards; so set PGO_BUSYFAIL 1718 * to avoid deadlocks. 1719 */ 1720 ap->a_flags |= PGO_BUSYFAIL; 1721 1722 do { 1723 int r; 1724 1725 /* If no pages are dirty, we can just use genfs_putpages. */ 1726 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 1727 ap->a_flags, 1) != 0) 1728 break; 1729 1730 /* 1731 * Sometimes pages are dirtied between the time that 1732 * we check and the time we try to clean them. 1733 * Instruct lfs_gop_write to return EDEADLK in this case 1734 * so we can write them properly. 1735 */ 1736 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 1737 r = genfs_putpages(v); 1738 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 1739 if (r != EDEADLK) 1740 return r; 1741 1742 /* Start over. */ 1743 preempt(1); 1744 simple_lock(&vp->v_interlock); 1745 } while(1); 1746 1747 /* 1748 * Dirty and asked to clean. 1749 * 1750 * Pagedaemon can't actually write LFS pages; wake up 1751 * the writer to take care of that. The writer will 1752 * notice the pager inode queue and act on that. 1753 */ 1754 if (pagedaemon) { 1755 simple_lock(&fs->lfs_interlock); 1756 ++fs->lfs_pdflush; 1757 simple_unlock(&fs->lfs_interlock); 1758 wakeup(&lfs_writer_daemon); 1759 simple_unlock(&vp->v_interlock); 1760 return EWOULDBLOCK; 1761 } 1762 1763 /* 1764 * If this is a file created in a recent dirop, we can't flush its 1765 * inode until the dirop is complete. Drain dirops, then flush the 1766 * filesystem (taking care of any other pending dirops while we're 1767 * at it). 1768 */ 1769 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 1770 (vp->v_flag & VDIROP)) { 1771 int locked; 1772 1773 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 1774 locked = VOP_ISLOCKED(vp) && /* XXX */ 1775 vp->v_lock.lk_lockholder == curproc->p_pid; 1776 simple_unlock(&vp->v_interlock); 1777 lfs_writer_enter(fs, "ppdirop"); 1778 if (locked) 1779 VOP_UNLOCK(vp, 0); 1780 1781 simple_lock(&fs->lfs_interlock); 1782 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 1783 simple_unlock(&fs->lfs_interlock); 1784 1785 simple_lock(&vp->v_interlock); 1786 if (locked) { 1787 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 1788 simple_lock(&vp->v_interlock); 1789 } 1790 lfs_writer_leave(fs); 1791 1792 /* XXX the flush should have taken care of this one too! */ 1793 } 1794 1795 /* 1796 * This is it. We are going to write some pages. From here on 1797 * down it's all just mechanics. 1798 * 1799 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 1800 */ 1801 ap->a_flags &= ~PGO_SYNCIO; 1802 1803 /* 1804 * If we've already got the seglock, flush the node and return. 1805 * The FIP has already been set up for us by lfs_writefile, 1806 * and FIP cleanup and lfs_updatemeta will also be done there, 1807 * unless genfs_putpages returns EDEADLK; then we must flush 1808 * what we have, and correct FIP and segment header accounting. 1809 */ 1810 1811 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 1812 if (!seglocked) { 1813 simple_unlock(&vp->v_interlock); 1814 /* 1815 * Take the seglock, because we are going to be writing pages. 1816 */ 1817 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 1818 if (error != 0) 1819 return error; 1820 simple_lock(&vp->v_interlock); 1821 } 1822 1823 /* 1824 * VOP_PUTPAGES should not be called while holding the seglock. 1825 * XXXUBC fix lfs_markv, or do this properly. 1826 */ 1827 #ifdef notyet 1828 KASSERT(fs->lfs_seglock == 1); 1829 #endif /* notyet */ 1830 1831 /* 1832 * We assume we're being called with sp->fip pointing at blank space. 1833 * Account for a new FIP in the segment header, and set sp->vp. 1834 * (This should duplicate the setup at the top of lfs_writefile().) 1835 */ 1836 sp = fs->lfs_sp; 1837 if (!seglocked) { 1838 if (sp->seg_bytes_left < fs->lfs_bsize || 1839 sp->sum_bytes_left < sizeof(struct finfo)) 1840 (void) lfs_writeseg(fs, fs->lfs_sp); 1841 1842 sp->sum_bytes_left -= FINFOSIZE; 1843 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1844 } 1845 KASSERT(sp->vp == NULL); 1846 sp->vp = vp; 1847 1848 if (!seglocked) { 1849 if (vp->v_flag & VDIROP) 1850 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 1851 } 1852 1853 sp->fip->fi_nblocks = 0; 1854 sp->fip->fi_ino = ip->i_number; 1855 sp->fip->fi_version = ip->i_gen; 1856 1857 /* 1858 * Loop through genfs_putpages until all pages are gathered. 1859 * genfs_putpages() drops the interlock, so reacquire it if necessary. 1860 * Whenever we lose the interlock we have to rerun check_dirty, as 1861 * well. 1862 */ 1863 again: 1864 check_dirty(fs, vp, startoffset, endoffset, blkeof, ap->a_flags, 0); 1865 1866 if ((error = genfs_putpages(v)) == EDEADLK) { 1867 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 1868 " EDEADLK [2] ino %d off %x (seg %d)\n", 1869 ip->i_number, fs->lfs_offset, 1870 dtosn(fs, fs->lfs_offset))); 1871 /* If nothing to write, short-circuit */ 1872 if (sp->cbpp - sp->bpp > 1) { 1873 /* Write gathered pages */ 1874 lfs_updatemeta(sp); 1875 (void) lfs_writeseg(fs, sp); 1876 1877 /* 1878 * Reinitialize brand new FIP and add us to it. 1879 * (This should duplicate the fixup in 1880 * lfs_gatherpages().) 1881 */ 1882 KASSERT(sp->vp == vp); 1883 sp->fip->fi_version = ip->i_gen; 1884 sp->fip->fi_ino = ip->i_number; 1885 /* Add us to the new segment summary. */ 1886 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1887 sp->sum_bytes_left -= FINFOSIZE; 1888 } 1889 1890 /* Give the write a chance to complete */ 1891 preempt(1); 1892 1893 /* We've lost the interlock. Start over. */ 1894 simple_lock(&vp->v_interlock); 1895 goto again; 1896 } 1897 1898 KASSERT(sp->vp == vp); 1899 if (!seglocked) { 1900 sp->vp = NULL; /* XXX lfs_gather below will set this */ 1901 1902 /* Write indirect blocks as well */ 1903 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 1904 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 1905 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 1906 1907 KASSERT(sp->vp == NULL); 1908 sp->vp = vp; 1909 } 1910 1911 /* 1912 * Blocks are now gathered into a segment waiting to be written. 1913 * All that's left to do is update metadata, and write them. 1914 */ 1915 lfs_updatemeta(sp); 1916 KASSERT(sp->vp == vp); 1917 sp->vp = NULL; 1918 1919 if (seglocked) { 1920 /* we're called by lfs_writefile. */ 1921 return error; 1922 } 1923 1924 /* 1925 * Clean up FIP, since we're done writing this file. 1926 * This should duplicate cleanup at the end of lfs_writefile(). 1927 */ 1928 if (sp->fip->fi_nblocks != 0) { 1929 sp->fip = (FINFO*)((caddr_t)sp->fip + FINFOSIZE + 1930 sizeof(int32_t) * sp->fip->fi_nblocks); 1931 sp->start_lbp = &sp->fip->fi_blocks[0]; 1932 } else { 1933 sp->sum_bytes_left += FINFOSIZE; 1934 --((SEGSUM *)(sp->segsum))->ss_nfinfo; 1935 } 1936 lfs_writeseg(fs, fs->lfs_sp); 1937 1938 /* 1939 * XXX - with the malloc/copy writeseg, the pages are freed by now 1940 * even if we don't wait (e.g. if we hold a nested lock). This 1941 * will not be true if we stop using malloc/copy. 1942 */ 1943 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 1944 lfs_segunlock(fs); 1945 1946 /* 1947 * Wait for v_numoutput to drop to zero. The seglock should 1948 * take care of this, but there is a slight possibility that 1949 * aiodoned might not have got around to our buffers yet. 1950 */ 1951 if (sync) { 1952 s = splbio(); 1953 simple_lock(&global_v_numoutput_slock); 1954 while (vp->v_numoutput > 0) { 1955 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 1956 " num %d\n", ip->i_number, vp->v_numoutput)); 1957 vp->v_flag |= VBWAIT; 1958 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 1959 &global_v_numoutput_slock); 1960 } 1961 simple_unlock(&global_v_numoutput_slock); 1962 splx(s); 1963 } 1964 return error; 1965 } 1966 1967 /* 1968 * Return the last logical file offset that should be written for this file 1969 * if we're doing a write that ends at "size". If writing, we need to know 1970 * about sizes on disk, i.e. fragments if there are any; if reading, we need 1971 * to know about entire blocks. 1972 */ 1973 void 1974 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 1975 { 1976 struct inode *ip = VTOI(vp); 1977 struct lfs *fs = ip->i_lfs; 1978 daddr_t olbn, nlbn; 1979 1980 KASSERT(flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)); 1981 KASSERT((flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)) 1982 != (GOP_SIZE_READ | GOP_SIZE_WRITE)); 1983 1984 olbn = lblkno(fs, ip->i_size); 1985 nlbn = lblkno(fs, size); 1986 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 1987 *eobp = fragroundup(fs, size); 1988 } else { 1989 *eobp = blkroundup(fs, size); 1990 } 1991 } 1992 1993 #ifdef DEBUG 1994 void lfs_dump_vop(void *); 1995 1996 void 1997 lfs_dump_vop(void *v) 1998 { 1999 struct vop_putpages_args /* { 2000 struct vnode *a_vp; 2001 voff_t a_offlo; 2002 voff_t a_offhi; 2003 int a_flags; 2004 } */ *ap = v; 2005 2006 #ifdef DDB 2007 vfs_vnode_print(ap->a_vp, 0, printf); 2008 #endif 2009 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2010 } 2011 #endif 2012 2013 int 2014 lfs_mmap(void *v) 2015 { 2016 struct vop_mmap_args /* { 2017 const struct vnodeop_desc *a_desc; 2018 struct vnode *a_vp; 2019 int a_fflags; 2020 struct ucred *a_cred; 2021 struct lwp *a_l; 2022 } */ *ap = v; 2023 2024 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2025 return EOPNOTSUPP; 2026 return ufs_mmap(v); 2027 } 2028
Indexes created Mon Sep 29 13:09:55 GMT 2025