lfs_vnops.c revision 1.188
1 /* $NetBSD: lfs_vnops.c,v 1.188 2006/09/01 19:41:28 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 1.188 2006/09/01 19:41:28 perseant Exp $"); 71 72 #ifdef _KERNEL_OPT 73 #include "opt_compat_netbsd.h" 74 #endif 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/namei.h> 79 #include <sys/resourcevar.h> 80 #include <sys/kernel.h> 81 #include <sys/file.h> 82 #include <sys/stat.h> 83 #include <sys/buf.h> 84 #include <sys/proc.h> 85 #include <sys/mount.h> 86 #include <sys/vnode.h> 87 #include <sys/pool.h> 88 #include <sys/signalvar.h> 89 #include <sys/kauth.h> 90 #include <sys/syslog.h> 91 92 #include <miscfs/fifofs/fifo.h> 93 #include <miscfs/genfs/genfs.h> 94 #include <miscfs/specfs/specdev.h> 95 96 #include <ufs/ufs/inode.h> 97 #include <ufs/ufs/dir.h> 98 #include <ufs/ufs/ufsmount.h> 99 #include <ufs/ufs/ufs_extern.h> 100 101 #include <uvm/uvm.h> 102 #include <uvm/uvm_pmap.h> 103 #include <uvm/uvm_stat.h> 104 #include <uvm/uvm_pager.h> 105 106 #include <ufs/lfs/lfs.h> 107 #include <ufs/lfs/lfs_extern.h> 108 109 extern pid_t lfs_writer_daemon; 110 111 /* Global vfs data structures for lfs. */ 112 int (**lfs_vnodeop_p)(void *); 113 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 114 { &vop_default_desc, vn_default_error }, 115 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 116 { &vop_create_desc, lfs_create }, /* create */ 117 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 118 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 119 { &vop_open_desc, ufs_open }, /* open */ 120 { &vop_close_desc, lfs_close }, /* close */ 121 { &vop_access_desc, ufs_access }, /* access */ 122 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 123 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 124 { &vop_read_desc, lfs_read }, /* read */ 125 { &vop_write_desc, lfs_write }, /* write */ 126 { &vop_lease_desc, ufs_lease_check }, /* lease */ 127 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 128 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 129 { &vop_poll_desc, ufs_poll }, /* poll */ 130 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 131 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 132 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 133 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 134 { &vop_seek_desc, ufs_seek }, /* seek */ 135 { &vop_remove_desc, lfs_remove }, /* remove */ 136 { &vop_link_desc, lfs_link }, /* link */ 137 { &vop_rename_desc, lfs_rename }, /* rename */ 138 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 139 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 140 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 141 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 142 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 143 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 144 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 145 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 146 { &vop_lock_desc, ufs_lock }, /* lock */ 147 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 148 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 149 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 150 { &vop_print_desc, ufs_print }, /* print */ 151 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 152 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 153 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 154 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 155 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 156 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 157 { NULL, NULL } 158 }; 159 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 160 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 161 162 int (**lfs_specop_p)(void *); 163 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 164 { &vop_default_desc, vn_default_error }, 165 { &vop_lookup_desc, spec_lookup }, /* lookup */ 166 { &vop_create_desc, spec_create }, /* create */ 167 { &vop_mknod_desc, spec_mknod }, /* mknod */ 168 { &vop_open_desc, spec_open }, /* open */ 169 { &vop_close_desc, lfsspec_close }, /* close */ 170 { &vop_access_desc, ufs_access }, /* access */ 171 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 172 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 173 { &vop_read_desc, ufsspec_read }, /* read */ 174 { &vop_write_desc, ufsspec_write }, /* write */ 175 { &vop_lease_desc, spec_lease_check }, /* lease */ 176 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 177 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 178 { &vop_poll_desc, spec_poll }, /* poll */ 179 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 180 { &vop_revoke_desc, spec_revoke }, /* revoke */ 181 { &vop_mmap_desc, spec_mmap }, /* mmap */ 182 { &vop_fsync_desc, spec_fsync }, /* fsync */ 183 { &vop_seek_desc, spec_seek }, /* seek */ 184 { &vop_remove_desc, spec_remove }, /* remove */ 185 { &vop_link_desc, spec_link }, /* link */ 186 { &vop_rename_desc, spec_rename }, /* rename */ 187 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 188 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 189 { &vop_symlink_desc, spec_symlink }, /* symlink */ 190 { &vop_readdir_desc, spec_readdir }, /* readdir */ 191 { &vop_readlink_desc, spec_readlink }, /* readlink */ 192 { &vop_abortop_desc, spec_abortop }, /* abortop */ 193 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 194 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 195 { &vop_lock_desc, ufs_lock }, /* lock */ 196 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 197 { &vop_bmap_desc, spec_bmap }, /* bmap */ 198 { &vop_strategy_desc, spec_strategy }, /* strategy */ 199 { &vop_print_desc, ufs_print }, /* print */ 200 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 201 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 202 { &vop_advlock_desc, spec_advlock }, /* advlock */ 203 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 204 { &vop_getpages_desc, spec_getpages }, /* getpages */ 205 { &vop_putpages_desc, spec_putpages }, /* putpages */ 206 { NULL, NULL } 207 }; 208 const struct vnodeopv_desc lfs_specop_opv_desc = 209 { &lfs_specop_p, lfs_specop_entries }; 210 211 int (**lfs_fifoop_p)(void *); 212 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 213 { &vop_default_desc, vn_default_error }, 214 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 215 { &vop_create_desc, fifo_create }, /* create */ 216 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 217 { &vop_open_desc, fifo_open }, /* open */ 218 { &vop_close_desc, lfsfifo_close }, /* close */ 219 { &vop_access_desc, ufs_access }, /* access */ 220 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 221 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 222 { &vop_read_desc, ufsfifo_read }, /* read */ 223 { &vop_write_desc, ufsfifo_write }, /* write */ 224 { &vop_lease_desc, fifo_lease_check }, /* lease */ 225 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 226 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 227 { &vop_poll_desc, fifo_poll }, /* poll */ 228 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 229 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 230 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 231 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 232 { &vop_seek_desc, fifo_seek }, /* seek */ 233 { &vop_remove_desc, fifo_remove }, /* remove */ 234 { &vop_link_desc, fifo_link }, /* link */ 235 { &vop_rename_desc, fifo_rename }, /* rename */ 236 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 237 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 238 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 239 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 240 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 241 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 242 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 243 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 244 { &vop_lock_desc, ufs_lock }, /* lock */ 245 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 246 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 247 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 248 { &vop_print_desc, ufs_print }, /* print */ 249 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 250 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 251 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 252 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 253 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 254 { NULL, NULL } 255 }; 256 const struct vnodeopv_desc lfs_fifoop_opv_desc = 257 { &lfs_fifoop_p, lfs_fifoop_entries }; 258 259 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int); 260 261 #define LFS_READWRITE 262 #include <ufs/ufs/ufs_readwrite.c> 263 #undef LFS_READWRITE 264 265 /* 266 * Synch an open file. 267 */ 268 /* ARGSUSED */ 269 int 270 lfs_fsync(void *v) 271 { 272 struct vop_fsync_args /* { 273 struct vnode *a_vp; 274 kauth_cred_t a_cred; 275 int a_flags; 276 off_t offlo; 277 off_t offhi; 278 struct lwp *a_l; 279 } */ *ap = v; 280 struct vnode *vp = ap->a_vp; 281 int error, wait; 282 283 /* If we're mounted read-only, don't try to sync. */ 284 if (VTOI(vp)->i_lfs->lfs_ronly) 285 return 0; 286 287 /* 288 * Trickle sync checks for need to do a checkpoint after possible 289 * activity from the pagedaemon. 290 */ 291 if (ap->a_flags & FSYNC_LAZY) { 292 simple_lock(&lfs_subsys_lock); 293 wakeup(&lfs_writer_daemon); 294 simple_unlock(&lfs_subsys_lock); 295 return 0; 296 } 297 298 /* 299 * If a vnode is bring cleaned, flush it out before we try to 300 * reuse it. This prevents the cleaner from writing files twice 301 * in the same partial segment, causing an accounting underflow. 302 */ 303 if (ap->a_flags & FSYNC_RECLAIM && VTOI(vp)->i_flags & IN_CLEANING) { 304 /* printf("avoiding VONWORKLIST panic\n"); */ 305 lfs_vflush(vp); 306 } 307 308 wait = (ap->a_flags & FSYNC_WAIT); 309 simple_lock(&vp->v_interlock); 310 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 311 round_page(ap->a_offhi), 312 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 313 if (error) 314 return error; 315 error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 316 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 317 int l = 0; 318 error = VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE, 319 ap->a_l->l_cred, ap->a_l); 320 } 321 if (wait && !VPISEMPTY(vp)) 322 LFS_SET_UINO(VTOI(vp), IN_MODIFIED); 323 324 return error; 325 } 326 327 /* 328 * Take IN_ADIROP off, then call ufs_inactive. 329 */ 330 int 331 lfs_inactive(void *v) 332 { 333 struct vop_inactive_args /* { 334 struct vnode *a_vp; 335 struct lwp *a_l; 336 } */ *ap = v; 337 338 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 339 340 lfs_unmark_vnode(ap->a_vp); 341 342 /* 343 * The Ifile is only ever inactivated on unmount. 344 * Streamline this process by not giving it more dirty blocks. 345 */ 346 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 347 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 348 VOP_UNLOCK(ap->a_vp, 0); 349 return 0; 350 } 351 352 return ufs_inactive(v); 353 } 354 355 /* 356 * These macros are used to bracket UFS directory ops, so that we can 357 * identify all the pages touched during directory ops which need to 358 * be ordered and flushed atomically, so that they may be recovered. 359 * 360 * Because we have to mark nodes VDIROP in order to prevent 361 * the cache from reclaiming them while a dirop is in progress, we must 362 * also manage the number of nodes so marked (otherwise we can run out). 363 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 364 * is decremented during segment write, when VDIROP is taken off. 365 */ 366 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 367 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 368 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 369 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 370 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 371 static int lfs_set_dirop(struct vnode *, struct vnode *); 372 373 static int 374 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 375 { 376 struct lfs *fs; 377 int error; 378 379 KASSERT(VOP_ISLOCKED(dvp)); 380 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 381 382 fs = VTOI(dvp)->i_lfs; 383 384 ASSERT_NO_SEGLOCK(fs); 385 /* 386 * LFS_NRESERVE calculates direct and indirect blocks as well 387 * as an inode block; an overestimate in most cases. 388 */ 389 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 390 return (error); 391 392 restart: 393 simple_lock(&fs->lfs_interlock); 394 if (fs->lfs_dirops == 0) { 395 simple_unlock(&fs->lfs_interlock); 396 lfs_check(dvp, LFS_UNUSED_LBN, 0); 397 simple_lock(&fs->lfs_interlock); 398 } 399 while (fs->lfs_writer) 400 ltsleep(&fs->lfs_dirops, (PRIBIO + 1), "lfs_sdirop", 0, 401 &fs->lfs_interlock); 402 simple_lock(&lfs_subsys_lock); 403 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 404 wakeup(&lfs_writer_daemon); 405 simple_unlock(&lfs_subsys_lock); 406 simple_unlock(&fs->lfs_interlock); 407 preempt(1); 408 goto restart; 409 } 410 411 if (lfs_dirvcount > LFS_MAX_DIROP) { 412 simple_unlock(&fs->lfs_interlock); 413 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 414 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 415 if ((error = ltsleep(&lfs_dirvcount, 416 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 417 &lfs_subsys_lock)) != 0) { 418 goto unreserve; 419 } 420 goto restart; 421 } 422 simple_unlock(&lfs_subsys_lock); 423 424 ++fs->lfs_dirops; 425 fs->lfs_doifile = 1; 426 simple_unlock(&fs->lfs_interlock); 427 428 /* Hold a reference so SET_ENDOP will be happy */ 429 vref(dvp); 430 if (vp) { 431 vref(vp); 432 MARK_VNODE(vp); 433 } 434 435 MARK_VNODE(dvp); 436 return 0; 437 438 unreserve: 439 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 440 return error; 441 } 442 443 /* 444 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 445 * in getnewvnode(), if we have a stacked filesystem mounted on top 446 * of us. 447 * 448 * NB: this means we have to clear the new vnodes on error. Fortunately 449 * SET_ENDOP is there to do that for us. 450 */ 451 static int 452 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 453 { 454 int error; 455 struct lfs *fs; 456 457 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 458 ASSERT_NO_SEGLOCK(fs); 459 if (fs->lfs_ronly) 460 return EROFS; 461 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 462 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 463 dvp, error)); 464 return error; 465 } 466 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 467 if (vpp) { 468 ungetnewvnode(*vpp); 469 *vpp = NULL; 470 } 471 return error; 472 } 473 return 0; 474 } 475 476 #define SET_ENDOP_BASE(fs, dvp, str) \ 477 do { \ 478 simple_lock(&(fs)->lfs_interlock); \ 479 --(fs)->lfs_dirops; \ 480 if (!(fs)->lfs_dirops) { \ 481 if ((fs)->lfs_nadirop) { \ 482 panic("SET_ENDOP: %s: no dirops but " \ 483 " nadirop=%d", (str), \ 484 (fs)->lfs_nadirop); \ 485 } \ 486 wakeup(&(fs)->lfs_writer); \ 487 simple_unlock(&(fs)->lfs_interlock); \ 488 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 489 } else \ 490 simple_unlock(&(fs)->lfs_interlock); \ 491 } while(0) 492 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 493 do { \ 494 UNMARK_VNODE(dvp); \ 495 if (nvpp && *nvpp) \ 496 UNMARK_VNODE(*nvpp); \ 497 /* Check for error return to stem vnode leakage */ \ 498 if (nvpp && *nvpp && !((*nvpp)->v_flag & VDIROP)) \ 499 ungetnewvnode(*(nvpp)); \ 500 SET_ENDOP_BASE((fs), (dvp), (str)); \ 501 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 502 vrele(dvp); \ 503 } while(0) 504 #define SET_ENDOP_CREATE_AP(ap, str) \ 505 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 506 (ap)->a_vpp, (str)) 507 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 508 do { \ 509 UNMARK_VNODE(dvp); \ 510 if (ovp) \ 511 UNMARK_VNODE(ovp); \ 512 SET_ENDOP_BASE((fs), (dvp), (str)); \ 513 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 514 vrele(dvp); \ 515 if (ovp) \ 516 vrele(ovp); \ 517 } while(0) 518 519 void 520 lfs_mark_vnode(struct vnode *vp) 521 { 522 struct inode *ip = VTOI(vp); 523 struct lfs *fs = ip->i_lfs; 524 525 simple_lock(&fs->lfs_interlock); 526 if (!(ip->i_flag & IN_ADIROP)) { 527 if (!(vp->v_flag & VDIROP)) { 528 (void)lfs_vref(vp); 529 simple_lock(&lfs_subsys_lock); 530 ++lfs_dirvcount; 531 ++fs->lfs_dirvcount; 532 simple_unlock(&lfs_subsys_lock); 533 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 534 vp->v_flag |= VDIROP; 535 } 536 ++fs->lfs_nadirop; 537 ip->i_flag |= IN_ADIROP; 538 } else 539 KASSERT(vp->v_flag & VDIROP); 540 simple_unlock(&fs->lfs_interlock); 541 } 542 543 void 544 lfs_unmark_vnode(struct vnode *vp) 545 { 546 struct inode *ip = VTOI(vp); 547 548 if (ip && (ip->i_flag & IN_ADIROP)) { 549 KASSERT(vp->v_flag & VDIROP); 550 simple_lock(&ip->i_lfs->lfs_interlock); 551 --ip->i_lfs->lfs_nadirop; 552 simple_unlock(&ip->i_lfs->lfs_interlock); 553 ip->i_flag &= ~IN_ADIROP; 554 } 555 } 556 557 int 558 lfs_symlink(void *v) 559 { 560 struct vop_symlink_args /* { 561 struct vnode *a_dvp; 562 struct vnode **a_vpp; 563 struct componentname *a_cnp; 564 struct vattr *a_vap; 565 char *a_target; 566 } */ *ap = v; 567 int error; 568 569 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 570 vput(ap->a_dvp); 571 return error; 572 } 573 error = ufs_symlink(ap); 574 SET_ENDOP_CREATE_AP(ap, "symlink"); 575 return (error); 576 } 577 578 int 579 lfs_mknod(void *v) 580 { 581 struct vop_mknod_args /* { 582 struct vnode *a_dvp; 583 struct vnode **a_vpp; 584 struct componentname *a_cnp; 585 struct vattr *a_vap; 586 } */ *ap = v; 587 struct vattr *vap = ap->a_vap; 588 struct vnode **vpp = ap->a_vpp; 589 struct inode *ip; 590 int error; 591 struct mount *mp; 592 ino_t ino; 593 594 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 595 vput(ap->a_dvp); 596 return error; 597 } 598 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 599 ap->a_dvp, vpp, ap->a_cnp); 600 601 /* Either way we're done with the dirop at this point */ 602 SET_ENDOP_CREATE_AP(ap, "mknod"); 603 604 if (error) 605 return (error); 606 607 ip = VTOI(*vpp); 608 mp = (*vpp)->v_mount; 609 ino = ip->i_number; 610 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 611 if (vap->va_rdev != VNOVAL) { 612 /* 613 * Want to be able to use this to make badblock 614 * inodes, so don't truncate the dev number. 615 */ 616 #if 0 617 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 618 UFS_MPNEEDSWAP((*vpp)->v_mount)); 619 #else 620 ip->i_ffs1_rdev = vap->va_rdev; 621 #endif 622 } 623 624 /* 625 * Call fsync to write the vnode so that we don't have to deal with 626 * flushing it when it's marked VDIROP|VXLOCK. 627 * 628 * XXX KS - If we can't flush we also can't call vgone(), so must 629 * return. But, that leaves this vnode in limbo, also not good. 630 * Can this ever happen (barring hardware failure)? 631 */ 632 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 633 curlwp)) != 0) { 634 panic("lfs_mknod: couldn't fsync (ino %llu)", 635 (unsigned long long)ino); 636 /* return (error); */ 637 } 638 /* 639 * Remove vnode so that it will be reloaded by VFS_VGET and 640 * checked to see if it is an alias of an existing entry in 641 * the inode cache. 642 */ 643 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 644 645 VOP_UNLOCK(*vpp, 0); 646 lfs_vunref(*vpp); 647 (*vpp)->v_type = VNON; 648 vgone(*vpp); 649 error = VFS_VGET(mp, ino, vpp); 650 651 if (error != 0) { 652 *vpp = NULL; 653 return (error); 654 } 655 return (0); 656 } 657 658 int 659 lfs_create(void *v) 660 { 661 struct vop_create_args /* { 662 struct vnode *a_dvp; 663 struct vnode **a_vpp; 664 struct componentname *a_cnp; 665 struct vattr *a_vap; 666 } */ *ap = v; 667 int error; 668 669 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 670 vput(ap->a_dvp); 671 return error; 672 } 673 error = ufs_create(ap); 674 SET_ENDOP_CREATE_AP(ap, "create"); 675 return (error); 676 } 677 678 int 679 lfs_mkdir(void *v) 680 { 681 struct vop_mkdir_args /* { 682 struct vnode *a_dvp; 683 struct vnode **a_vpp; 684 struct componentname *a_cnp; 685 struct vattr *a_vap; 686 } */ *ap = v; 687 int error; 688 689 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 690 vput(ap->a_dvp); 691 return error; 692 } 693 error = ufs_mkdir(ap); 694 SET_ENDOP_CREATE_AP(ap, "mkdir"); 695 return (error); 696 } 697 698 int 699 lfs_remove(void *v) 700 { 701 struct vop_remove_args /* { 702 struct vnode *a_dvp; 703 struct vnode *a_vp; 704 struct componentname *a_cnp; 705 } */ *ap = v; 706 struct vnode *dvp, *vp; 707 struct inode *ip; 708 int error; 709 710 dvp = ap->a_dvp; 711 vp = ap->a_vp; 712 ip = VTOI(vp); 713 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 714 if (dvp == vp) 715 vrele(vp); 716 else 717 vput(vp); 718 vput(dvp); 719 return error; 720 } 721 error = ufs_remove(ap); 722 if (ip->i_nlink == 0) 723 lfs_orphan(ip->i_lfs, ip->i_number); 724 SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove"); 725 return (error); 726 } 727 728 int 729 lfs_rmdir(void *v) 730 { 731 struct vop_rmdir_args /* { 732 struct vnodeop_desc *a_desc; 733 struct vnode *a_dvp; 734 struct vnode *a_vp; 735 struct componentname *a_cnp; 736 } */ *ap = v; 737 struct vnode *vp; 738 struct inode *ip; 739 int error; 740 741 vp = ap->a_vp; 742 ip = VTOI(vp); 743 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 744 vrele(ap->a_dvp); 745 if (ap->a_vp != ap->a_dvp) 746 VOP_UNLOCK(ap->a_dvp, 0); 747 vput(vp); 748 return error; 749 } 750 error = ufs_rmdir(ap); 751 if (ip->i_nlink == 0) 752 lfs_orphan(ip->i_lfs, ip->i_number); 753 SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 754 return (error); 755 } 756 757 int 758 lfs_link(void *v) 759 { 760 struct vop_link_args /* { 761 struct vnode *a_dvp; 762 struct vnode *a_vp; 763 struct componentname *a_cnp; 764 } */ *ap = v; 765 int error; 766 struct vnode **vpp = NULL; 767 768 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 769 vput(ap->a_dvp); 770 return error; 771 } 772 error = ufs_link(ap); 773 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 774 return (error); 775 } 776 777 int 778 lfs_rename(void *v) 779 { 780 struct vop_rename_args /* { 781 struct vnode *a_fdvp; 782 struct vnode *a_fvp; 783 struct componentname *a_fcnp; 784 struct vnode *a_tdvp; 785 struct vnode *a_tvp; 786 struct componentname *a_tcnp; 787 } */ *ap = v; 788 struct vnode *tvp, *fvp, *tdvp, *fdvp; 789 struct componentname *tcnp, *fcnp; 790 int error; 791 struct lfs *fs; 792 793 fs = VTOI(ap->a_fdvp)->i_lfs; 794 tvp = ap->a_tvp; 795 tdvp = ap->a_tdvp; 796 tcnp = ap->a_tcnp; 797 fvp = ap->a_fvp; 798 fdvp = ap->a_fdvp; 799 fcnp = ap->a_fcnp; 800 801 /* 802 * Check for cross-device rename. 803 * If it is, we don't want to set dirops, just error out. 804 * (In particular note that MARK_VNODE(tdvp) will DTWT on 805 * a cross-device rename.) 806 * 807 * Copied from ufs_rename. 808 */ 809 if ((fvp->v_mount != tdvp->v_mount) || 810 (tvp && (fvp->v_mount != tvp->v_mount))) { 811 error = EXDEV; 812 goto errout; 813 } 814 815 /* 816 * Check to make sure we're not renaming a vnode onto itself 817 * (deleting a hard link by renaming one name onto another); 818 * if we are we can't recursively call VOP_REMOVE since that 819 * would leave us with an unaccounted-for number of live dirops. 820 * 821 * Inline the relevant section of ufs_rename here, *before* 822 * calling SET_DIROP_REMOVE. 823 */ 824 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 825 (VTOI(tdvp)->i_flags & APPEND))) { 826 error = EPERM; 827 goto errout; 828 } 829 if (fvp == tvp) { 830 if (fvp->v_type == VDIR) { 831 error = EINVAL; 832 goto errout; 833 } 834 835 /* Release destination completely. */ 836 VOP_ABORTOP(tdvp, tcnp); 837 vput(tdvp); 838 vput(tvp); 839 840 /* Delete source. */ 841 vrele(fvp); 842 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 843 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 844 fcnp->cn_nameiop = DELETE; 845 if ((error = relookup(fdvp, &fvp, fcnp))){ 846 /* relookup blew away fdvp */ 847 return (error); 848 } 849 return (VOP_REMOVE(fdvp, fvp, fcnp)); 850 } 851 852 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 853 goto errout; 854 MARK_VNODE(fdvp); 855 MARK_VNODE(fvp); 856 857 error = ufs_rename(ap); 858 UNMARK_VNODE(fdvp); 859 UNMARK_VNODE(fvp); 860 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 861 return (error); 862 863 errout: 864 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 865 if (tdvp == tvp) 866 vrele(tdvp); 867 else 868 vput(tdvp); 869 if (tvp) 870 vput(tvp); 871 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 872 vrele(fdvp); 873 vrele(fvp); 874 return (error); 875 } 876 877 /* XXX hack to avoid calling ITIMES in getattr */ 878 int 879 lfs_getattr(void *v) 880 { 881 struct vop_getattr_args /* { 882 struct vnode *a_vp; 883 struct vattr *a_vap; 884 kauth_cred_t a_cred; 885 struct lwp *a_l; 886 } */ *ap = v; 887 struct vnode *vp = ap->a_vp; 888 struct inode *ip = VTOI(vp); 889 struct vattr *vap = ap->a_vap; 890 struct lfs *fs = ip->i_lfs; 891 /* 892 * Copy from inode table 893 */ 894 vap->va_fsid = ip->i_dev; 895 vap->va_fileid = ip->i_number; 896 vap->va_mode = ip->i_mode & ~IFMT; 897 vap->va_nlink = ip->i_nlink; 898 vap->va_uid = ip->i_uid; 899 vap->va_gid = ip->i_gid; 900 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 901 vap->va_size = vp->v_size; 902 vap->va_atime.tv_sec = ip->i_ffs1_atime; 903 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 904 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 905 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 906 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 907 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 908 vap->va_flags = ip->i_flags; 909 vap->va_gen = ip->i_gen; 910 /* this doesn't belong here */ 911 if (vp->v_type == VBLK) 912 vap->va_blocksize = BLKDEV_IOSIZE; 913 else if (vp->v_type == VCHR) 914 vap->va_blocksize = MAXBSIZE; 915 else 916 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 917 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 918 vap->va_type = vp->v_type; 919 vap->va_filerev = ip->i_modrev; 920 return (0); 921 } 922 923 /* 924 * Check to make sure the inode blocks won't choke the buffer 925 * cache, then call ufs_setattr as usual. 926 */ 927 int 928 lfs_setattr(void *v) 929 { 930 struct vop_setattr_args /* { 931 struct vnode *a_vp; 932 struct vattr *a_vap; 933 kauth_cred_t a_cred; 934 struct lwp *a_l; 935 } */ *ap = v; 936 struct vnode *vp = ap->a_vp; 937 938 lfs_check(vp, LFS_UNUSED_LBN, 0); 939 return ufs_setattr(v); 940 } 941 942 /* 943 * Release the block we hold on lfs_newseg wrapping. Called on file close, 944 * or explicitly from LFCNWRAPGO. Called with the interlock held. 945 */ 946 static int 947 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor) 948 { 949 if ((ip->i_lfs_iflags & LFSI_WRAPBLOCK) == 0) 950 return EBUSY; 951 952 ip->i_lfs_iflags &= ~LFSI_WRAPBLOCK; 953 954 KASSERT(fs->lfs_nowrap > 0); 955 if (fs->lfs_nowrap <= 0) { 956 return 0; 957 } 958 959 if (--fs->lfs_nowrap == 0) { 960 log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt); 961 wakeup(&fs->lfs_wrappass); 962 lfs_wakeup_cleaner(fs); 963 } 964 if (waitfor) { 965 ltsleep(&fs->lfs_nextseg, PCATCH | PUSER, 966 "segment", 0, &fs->lfs_interlock); 967 } 968 969 return 0; 970 } 971 972 /* 973 * Close called 974 */ 975 /* ARGSUSED */ 976 int 977 lfs_close(void *v) 978 { 979 struct vop_close_args /* { 980 struct vnode *a_vp; 981 int a_fflag; 982 kauth_cred_t a_cred; 983 struct lwp *a_l; 984 } */ *ap = v; 985 struct vnode *vp = ap->a_vp; 986 struct inode *ip = VTOI(vp); 987 #if 0 988 /* XXX fix this */ 989 struct lfs *fs = ip->i_lfs; 990 991 if ((ip->i_lfs_iflags & (LFSI_WRAPBLOCK | LFSI_WRAPWAIT)) == 992 LFSI_WRAPBLOCK) { 993 simple_lock(&fs->lfs_interlock); 994 printf("lfs_close with flags 0x%x\n", (unsigned)ap->a_fflag); 995 log(LOG_NOTICE, "lfs_close: releasing log wrap control\n"); 996 lfs_wrapgo(fs, ip, 0); 997 simple_unlock(&fs->lfs_interlock); 998 } 999 #endif 1000 1001 if (vp == ip->i_lfs->lfs_ivnode && 1002 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 1003 return 0; 1004 1005 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 1006 LFS_ITIMES(ip, NULL, NULL, NULL); 1007 } 1008 return (0); 1009 } 1010 1011 /* 1012 * Close wrapper for special devices. 1013 * 1014 * Update the times on the inode then do device close. 1015 */ 1016 int 1017 lfsspec_close(void *v) 1018 { 1019 struct vop_close_args /* { 1020 struct vnode *a_vp; 1021 int a_fflag; 1022 kauth_cred_t a_cred; 1023 struct lwp *a_l; 1024 } */ *ap = v; 1025 struct vnode *vp; 1026 struct inode *ip; 1027 1028 vp = ap->a_vp; 1029 ip = VTOI(vp); 1030 if (vp->v_usecount > 1) { 1031 LFS_ITIMES(ip, NULL, NULL, NULL); 1032 } 1033 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1034 } 1035 1036 /* 1037 * Close wrapper for fifo's. 1038 * 1039 * Update the times on the inode then do device close. 1040 */ 1041 int 1042 lfsfifo_close(void *v) 1043 { 1044 struct vop_close_args /* { 1045 struct vnode *a_vp; 1046 int a_fflag; 1047 kauth_cred_ a_cred; 1048 struct lwp *a_l; 1049 } */ *ap = v; 1050 struct vnode *vp; 1051 struct inode *ip; 1052 1053 vp = ap->a_vp; 1054 ip = VTOI(vp); 1055 if (ap->a_vp->v_usecount > 1) { 1056 LFS_ITIMES(ip, NULL, NULL, NULL); 1057 } 1058 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1059 } 1060 1061 /* 1062 * Reclaim an inode so that it can be used for other purposes. 1063 */ 1064 1065 int 1066 lfs_reclaim(void *v) 1067 { 1068 struct vop_reclaim_args /* { 1069 struct vnode *a_vp; 1070 struct lwp *a_l; 1071 } */ *ap = v; 1072 struct vnode *vp = ap->a_vp; 1073 struct inode *ip = VTOI(vp); 1074 int error; 1075 1076 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1077 1078 LFS_CLR_UINO(ip, IN_ALLMOD); 1079 if ((error = ufs_reclaim(vp, ap->a_l))) 1080 return (error); 1081 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1082 lfs_deregister_all(vp); 1083 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1084 ip->inode_ext.lfs = NULL; 1085 pool_put(&lfs_inode_pool, vp->v_data); 1086 vp->v_data = NULL; 1087 return (0); 1088 } 1089 1090 /* 1091 * Read a block from a storage device. 1092 * In order to avoid reading blocks that are in the process of being 1093 * written by the cleaner---and hence are not mutexed by the normal 1094 * buffer cache / page cache mechanisms---check for collisions before 1095 * reading. 1096 * 1097 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1098 * the active cleaner test. 1099 * 1100 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1101 */ 1102 int 1103 lfs_strategy(void *v) 1104 { 1105 struct vop_strategy_args /* { 1106 struct vnode *a_vp; 1107 struct buf *a_bp; 1108 } */ *ap = v; 1109 struct buf *bp; 1110 struct lfs *fs; 1111 struct vnode *vp; 1112 struct inode *ip; 1113 daddr_t tbn; 1114 int i, sn, error, slept; 1115 1116 bp = ap->a_bp; 1117 vp = ap->a_vp; 1118 ip = VTOI(vp); 1119 fs = ip->i_lfs; 1120 1121 /* lfs uses its strategy routine only for read */ 1122 KASSERT(bp->b_flags & B_READ); 1123 1124 if (vp->v_type == VBLK || vp->v_type == VCHR) 1125 panic("lfs_strategy: spec"); 1126 KASSERT(bp->b_bcount != 0); 1127 if (bp->b_blkno == bp->b_lblkno) { 1128 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1129 NULL); 1130 if (error) { 1131 bp->b_error = error; 1132 bp->b_flags |= B_ERROR; 1133 biodone(bp); 1134 return (error); 1135 } 1136 if ((long)bp->b_blkno == -1) /* no valid data */ 1137 clrbuf(bp); 1138 } 1139 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1140 biodone(bp); 1141 return (0); 1142 } 1143 1144 slept = 1; 1145 simple_lock(&fs->lfs_interlock); 1146 while (slept && fs->lfs_seglock) { 1147 simple_unlock(&fs->lfs_interlock); 1148 /* 1149 * Look through list of intervals. 1150 * There will only be intervals to look through 1151 * if the cleaner holds the seglock. 1152 * Since the cleaner is synchronous, we can trust 1153 * the list of intervals to be current. 1154 */ 1155 tbn = dbtofsb(fs, bp->b_blkno); 1156 sn = dtosn(fs, tbn); 1157 slept = 0; 1158 for (i = 0; i < fs->lfs_cleanind; i++) { 1159 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1160 tbn >= fs->lfs_cleanint[i]) { 1161 DLOG((DLOG_CLEAN, 1162 "lfs_strategy: ino %d lbn %" PRId64 1163 " ind %d sn %d fsb %" PRIx32 1164 " given sn %d fsb %" PRIx64 "\n", 1165 ip->i_number, bp->b_lblkno, i, 1166 dtosn(fs, fs->lfs_cleanint[i]), 1167 fs->lfs_cleanint[i], sn, tbn)); 1168 DLOG((DLOG_CLEAN, 1169 "lfs_strategy: sleeping on ino %d lbn %" 1170 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1171 simple_lock(&fs->lfs_interlock); 1172 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1173 /* Cleaner can't wait for itself */ 1174 ltsleep(&fs->lfs_iocount, 1175 (PRIBIO + 1) | PNORELOCK, 1176 "clean2", 0, 1177 &fs->lfs_interlock); 1178 slept = 1; 1179 break; 1180 } else if (fs->lfs_seglock) { 1181 ltsleep(&fs->lfs_seglock, 1182 (PRIBIO + 1) | PNORELOCK, 1183 "clean1", 0, 1184 &fs->lfs_interlock); 1185 slept = 1; 1186 break; 1187 } 1188 simple_unlock(&fs->lfs_interlock); 1189 } 1190 } 1191 simple_lock(&fs->lfs_interlock); 1192 } 1193 simple_unlock(&fs->lfs_interlock); 1194 1195 vp = ip->i_devvp; 1196 VOP_STRATEGY(vp, bp); 1197 return (0); 1198 } 1199 1200 void 1201 lfs_flush_dirops(struct lfs *fs) 1202 { 1203 struct inode *ip, *nip; 1204 struct vnode *vp; 1205 extern int lfs_dostats; 1206 struct segment *sp; 1207 int waslocked; 1208 1209 ASSERT_MAYBE_SEGLOCK(fs); 1210 KASSERT(fs->lfs_nadirop == 0); 1211 1212 if (fs->lfs_ronly) 1213 return; 1214 1215 simple_lock(&fs->lfs_interlock); 1216 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1217 simple_unlock(&fs->lfs_interlock); 1218 return; 1219 } else 1220 simple_unlock(&fs->lfs_interlock); 1221 1222 if (lfs_dostats) 1223 ++lfs_stats.flush_invoked; 1224 1225 /* 1226 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1227 * Technically this is a checkpoint (the on-disk state is valid) 1228 * even though we are leaving out all the file data. 1229 */ 1230 lfs_imtime(fs); 1231 lfs_seglock(fs, SEGM_CKP); 1232 sp = fs->lfs_sp; 1233 1234 /* 1235 * lfs_writevnodes, optimized to get dirops out of the way. 1236 * Only write dirops, and don't flush files' pages, only 1237 * blocks from the directories. 1238 * 1239 * We don't need to vref these files because they are 1240 * dirops and so hold an extra reference until the 1241 * segunlock clears them of that status. 1242 * 1243 * We don't need to check for IN_ADIROP because we know that 1244 * no dirops are active. 1245 * 1246 */ 1247 simple_lock(&fs->lfs_interlock); 1248 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1249 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1250 simple_unlock(&fs->lfs_interlock); 1251 vp = ITOV(ip); 1252 1253 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1254 1255 /* 1256 * All writes to directories come from dirops; all 1257 * writes to files' direct blocks go through the page 1258 * cache, which we're not touching. Reads to files 1259 * and/or directories will not be affected by writing 1260 * directory blocks inodes and file inodes. So we don't 1261 * really need to lock. If we don't lock, though, 1262 * make sure that we don't clear IN_MODIFIED 1263 * unnecessarily. 1264 */ 1265 if (vp->v_flag & (VXLOCK | VFREEING)) { 1266 simple_lock(&fs->lfs_interlock); 1267 continue; 1268 } 1269 waslocked = VOP_ISLOCKED(vp); 1270 if (vp->v_type != VREG && 1271 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1272 lfs_writefile(fs, sp, vp); 1273 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1274 !(ip->i_flag & IN_ALLMOD)) { 1275 LFS_SET_UINO(ip, IN_MODIFIED); 1276 } 1277 } 1278 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1279 (void) lfs_writeinode(fs, sp, ip); 1280 if (waslocked) 1281 LFS_SET_UINO(ip, IN_MODIFIED); 1282 simple_lock(&fs->lfs_interlock); 1283 } 1284 simple_unlock(&fs->lfs_interlock); 1285 /* We've written all the dirops there are */ 1286 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1287 lfs_finalize_fs_seguse(fs); 1288 (void) lfs_writeseg(fs, sp); 1289 lfs_segunlock(fs); 1290 } 1291 1292 /* 1293 * Flush all vnodes for which the pagedaemon has requested pageouts. 1294 * Skip over any files that are marked VDIROP (since lfs_flush_dirop() 1295 * has just run, this would be an error). If we have to skip a vnode 1296 * for any reason, just skip it; if we have to wait for the cleaner, 1297 * abort. The writer daemon will call us again later. 1298 */ 1299 void 1300 lfs_flush_pchain(struct lfs *fs) 1301 { 1302 struct inode *ip, *nip; 1303 struct vnode *vp; 1304 extern int lfs_dostats; 1305 struct segment *sp; 1306 int error; 1307 1308 ASSERT_NO_SEGLOCK(fs); 1309 1310 if (fs->lfs_ronly) 1311 return; 1312 1313 simple_lock(&fs->lfs_interlock); 1314 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1315 simple_unlock(&fs->lfs_interlock); 1316 return; 1317 } else 1318 simple_unlock(&fs->lfs_interlock); 1319 1320 /* Get dirops out of the way */ 1321 lfs_flush_dirops(fs); 1322 1323 if (lfs_dostats) 1324 ++lfs_stats.flush_invoked; 1325 1326 /* 1327 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1328 */ 1329 lfs_imtime(fs); 1330 lfs_seglock(fs, 0); 1331 sp = fs->lfs_sp; 1332 1333 /* 1334 * lfs_writevnodes, optimized to clear pageout requests. 1335 * Only write non-dirop files that are in the pageout queue. 1336 * We're very conservative about what we write; we want to be 1337 * fast and async. 1338 */ 1339 simple_lock(&fs->lfs_interlock); 1340 top: 1341 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1342 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1343 vp = ITOV(ip); 1344 1345 if (!(ip->i_flags & IN_PAGING)) 1346 goto top; 1347 1348 if (vp->v_flag & (VXLOCK|VDIROP)) 1349 continue; 1350 if (vp->v_type != VREG) 1351 continue; 1352 if (lfs_vref(vp)) 1353 continue; 1354 simple_unlock(&fs->lfs_interlock); 1355 1356 if (VOP_ISLOCKED(vp)) { 1357 lfs_vunref(vp); 1358 simple_lock(&fs->lfs_interlock); 1359 continue; 1360 } 1361 1362 error = lfs_writefile(fs, sp, vp); 1363 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1364 !(ip->i_flag & IN_ALLMOD)) { 1365 LFS_SET_UINO(ip, IN_MODIFIED); 1366 } 1367 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1368 (void) lfs_writeinode(fs, sp, ip); 1369 1370 lfs_vunref(vp); 1371 1372 if (error == EAGAIN) { 1373 lfs_writeseg(fs, sp); 1374 simple_lock(&fs->lfs_interlock); 1375 break; 1376 } 1377 simple_lock(&fs->lfs_interlock); 1378 } 1379 simple_unlock(&fs->lfs_interlock); 1380 (void) lfs_writeseg(fs, sp); 1381 lfs_segunlock(fs); 1382 } 1383 1384 /* 1385 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1386 */ 1387 int 1388 lfs_fcntl(void *v) 1389 { 1390 struct vop_fcntl_args /* { 1391 struct vnode *a_vp; 1392 u_long a_command; 1393 caddr_t a_data; 1394 int a_fflag; 1395 kauth_cred_t a_cred; 1396 struct lwp *a_l; 1397 } */ *ap = v; 1398 struct timeval *tvp; 1399 BLOCK_INFO *blkiov; 1400 CLEANERINFO *cip; 1401 SEGUSE *sup; 1402 int blkcnt, error, oclean; 1403 size_t fh_size; 1404 struct lfs_fcntl_markv blkvp; 1405 struct lwp *l; 1406 fsid_t *fsidp; 1407 struct lfs *fs; 1408 struct buf *bp; 1409 fhandle_t *fhp; 1410 daddr_t off; 1411 1412 /* Only respect LFS fcntls on fs root or Ifile */ 1413 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1414 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1415 return ufs_fcntl(v); 1416 } 1417 1418 /* Avoid locking a draining lock */ 1419 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1420 return ESHUTDOWN; 1421 } 1422 1423 /* LFS control and monitoring fcntls are available only to root */ 1424 l = ap->a_l; 1425 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1426 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1427 &l->l_acflag)) != 0) 1428 return (error); 1429 1430 fs = VTOI(ap->a_vp)->i_lfs; 1431 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1432 1433 error = 0; 1434 switch (ap->a_command) { 1435 case LFCNSEGWAITALL: 1436 case LFCNSEGWAITALL_COMPAT: 1437 fsidp = NULL; 1438 /* FALLSTHROUGH */ 1439 case LFCNSEGWAIT: 1440 case LFCNSEGWAIT_COMPAT: 1441 tvp = (struct timeval *)ap->a_data; 1442 simple_lock(&fs->lfs_interlock); 1443 ++fs->lfs_sleepers; 1444 simple_unlock(&fs->lfs_interlock); 1445 1446 error = lfs_segwait(fsidp, tvp); 1447 1448 simple_lock(&fs->lfs_interlock); 1449 if (--fs->lfs_sleepers == 0) 1450 wakeup(&fs->lfs_sleepers); 1451 simple_unlock(&fs->lfs_interlock); 1452 return error; 1453 1454 case LFCNBMAPV: 1455 case LFCNMARKV: 1456 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1457 1458 blkcnt = blkvp.blkcnt; 1459 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1460 return (EINVAL); 1461 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1462 if ((error = copyin(blkvp.blkiov, blkiov, 1463 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1464 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1465 return error; 1466 } 1467 1468 simple_lock(&fs->lfs_interlock); 1469 ++fs->lfs_sleepers; 1470 simple_unlock(&fs->lfs_interlock); 1471 if (ap->a_command == LFCNBMAPV) 1472 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1473 else /* LFCNMARKV */ 1474 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1475 if (error == 0) 1476 error = copyout(blkiov, blkvp.blkiov, 1477 blkcnt * sizeof(BLOCK_INFO)); 1478 simple_lock(&fs->lfs_interlock); 1479 if (--fs->lfs_sleepers == 0) 1480 wakeup(&fs->lfs_sleepers); 1481 simple_unlock(&fs->lfs_interlock); 1482 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1483 return error; 1484 1485 case LFCNRECLAIM: 1486 /* 1487 * Flush dirops and write Ifile, allowing empty segments 1488 * to be immediately reclaimed. 1489 */ 1490 lfs_writer_enter(fs, "pndirop"); 1491 off = fs->lfs_offset; 1492 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1493 lfs_flush_dirops(fs); 1494 LFS_CLEANERINFO(cip, fs, bp); 1495 oclean = cip->clean; 1496 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1497 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1498 fs->lfs_sp->seg_flags |= SEGM_PROT; 1499 lfs_segunlock(fs); 1500 lfs_writer_leave(fs); 1501 1502 #ifdef DEBUG 1503 LFS_CLEANERINFO(cip, fs, bp); 1504 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1505 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1506 fs->lfs_offset - off, cip->clean - oclean, 1507 fs->lfs_activesb)); 1508 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1509 #endif 1510 1511 return 0; 1512 1513 #ifdef COMPAT_30 1514 case LFCNIFILEFH_COMPAT: 1515 /* Return the filehandle of the Ifile */ 1516 if ((error = kauth_authorize_generic(l->l_cred, 1517 KAUTH_GENERIC_ISSUSER, &l->l_acflag)) != 0) 1518 return (error); 1519 fhp = (struct fhandle *)ap->a_data; 1520 fhp->fh_fsid = *fsidp; 1521 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1522 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1523 #endif 1524 1525 case LFCNIFILEFH_COMPAT2: 1526 case LFCNIFILEFH: 1527 /* Return the filehandle of the Ifile */ 1528 fhp = (struct fhandle *)ap->a_data; 1529 fhp->fh_fsid = *fsidp; 1530 fh_size = sizeof(struct lfs_fhandle) - 1531 offsetof(fhandle_t, fh_fid); 1532 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1533 1534 case LFCNREWIND: 1535 /* Move lfs_offset to the lowest-numbered segment */ 1536 return lfs_rewind(fs, *(int *)ap->a_data); 1537 1538 case LFCNINVAL: 1539 /* Mark a segment SEGUSE_INVAL */ 1540 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1541 if (sup->su_nbytes > 0) { 1542 brelse(bp); 1543 lfs_unset_inval_all(fs); 1544 return EBUSY; 1545 } 1546 sup->su_flags |= SEGUSE_INVAL; 1547 VOP_BWRITE(bp); 1548 return 0; 1549 1550 case LFCNRESIZE: 1551 /* Resize the filesystem */ 1552 return lfs_resize_fs(fs, *(int *)ap->a_data); 1553 1554 case LFCNWRAPSTOP: 1555 case LFCNWRAPSTOP_COMPAT: 1556 /* 1557 * Hold lfs_newseg at segment 0; if requested, sleep until 1558 * the filesystem wraps around. To support external agents 1559 * (dump, fsck-based regression test) that need to look at 1560 * a snapshot of the filesystem, without necessarily 1561 * requiring that all fs activity stops. 1562 */ 1563 if (VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPBLOCK) 1564 return EALREADY; 1565 1566 simple_lock(&fs->lfs_interlock); 1567 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPBLOCK; 1568 if (fs->lfs_nowrap == 0) 1569 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1570 ++fs->lfs_nowrap; 1571 if (*(int *)ap->a_data == 1 || 1572 ap->a_command == LFCNWRAPSTOP_COMPAT) { 1573 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1574 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1575 "segwrap", 0, &fs->lfs_interlock); 1576 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1577 if (error) { 1578 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1579 } 1580 } 1581 simple_unlock(&fs->lfs_interlock); 1582 return 0; 1583 1584 case LFCNWRAPGO: 1585 case LFCNWRAPGO_COMPAT: 1586 /* 1587 * Having done its work, the agent wakes up the writer. 1588 * If the argument is 1, it sleeps until a new segment 1589 * is selected. 1590 */ 1591 simple_lock(&fs->lfs_interlock); 1592 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1593 (ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1594 *((int *)ap->a_data))); 1595 simple_unlock(&fs->lfs_interlock); 1596 return error; 1597 1598 case LFCNWRAPPASS: 1599 if (!(VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPBLOCK)) 1600 return EALREADY; 1601 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1602 return EALREADY; 1603 simple_lock(&fs->lfs_interlock); 1604 if (fs->lfs_nowrap == 0) { 1605 simple_unlock(&fs->lfs_interlock); 1606 return EBUSY; 1607 } 1608 fs->lfs_wrappass = 1; 1609 wakeup(&fs->lfs_wrappass); 1610 /* Wait for the log to wrap, if asked */ 1611 if (*(int *)ap->a_data) { 1612 lfs_vref(ap->a_vp); 1613 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1614 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1615 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1616 "segwrap", 0, &fs->lfs_interlock); 1617 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1618 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1619 lfs_vunref(ap->a_vp); 1620 } 1621 simple_unlock(&fs->lfs_interlock); 1622 return error; 1623 1624 case LFCNWRAPSTATUS: 1625 simple_lock(&fs->lfs_interlock); 1626 *(int *)ap->a_data = fs->lfs_wrapstatus; 1627 simple_unlock(&fs->lfs_interlock); 1628 return 0; 1629 1630 default: 1631 return ufs_fcntl(v); 1632 } 1633 return 0; 1634 } 1635 1636 int 1637 lfs_getpages(void *v) 1638 { 1639 struct vop_getpages_args /* { 1640 struct vnode *a_vp; 1641 voff_t a_offset; 1642 struct vm_page **a_m; 1643 int *a_count; 1644 int a_centeridx; 1645 vm_prot_t a_access_type; 1646 int a_advice; 1647 int a_flags; 1648 } */ *ap = v; 1649 1650 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1651 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1652 return EPERM; 1653 } 1654 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1655 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1656 } 1657 1658 /* 1659 * we're relying on the fact that genfs_getpages() always read in 1660 * entire filesystem blocks. 1661 */ 1662 return genfs_getpages(v); 1663 } 1664 1665 /* 1666 * Make sure that for all pages in every block in the given range, 1667 * either all are dirty or all are clean. If any of the pages 1668 * we've seen so far are dirty, put the vnode on the paging chain, 1669 * and mark it IN_PAGING. 1670 * 1671 * If checkfirst != 0, don't check all the pages but return at the 1672 * first dirty page. 1673 */ 1674 static int 1675 check_dirty(struct lfs *fs, struct vnode *vp, 1676 off_t startoffset, off_t endoffset, off_t blkeof, 1677 int flags, int checkfirst) 1678 { 1679 int by_list; 1680 struct vm_page *curpg = NULL; /* XXX: gcc */ 1681 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1682 off_t soff = 0; /* XXX: gcc */ 1683 voff_t off; 1684 int i; 1685 int nonexistent; 1686 int any_dirty; /* number of dirty pages */ 1687 int dirty; /* number of dirty pages in a block */ 1688 int tdirty; 1689 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1690 int pagedaemon = (curproc == uvm.pagedaemon_proc); 1691 1692 ASSERT_MAYBE_SEGLOCK(fs); 1693 top: 1694 by_list = (vp->v_uobj.uo_npages <= 1695 ((endoffset - startoffset) >> PAGE_SHIFT) * 1696 UVM_PAGE_HASH_PENALTY); 1697 any_dirty = 0; 1698 1699 if (by_list) { 1700 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1701 } else { 1702 soff = startoffset; 1703 } 1704 while (by_list || soff < MIN(blkeof, endoffset)) { 1705 if (by_list) { 1706 /* 1707 * Find the first page in a block. Skip 1708 * blocks outside our area of interest or beyond 1709 * the end of file. 1710 */ 1711 if (pages_per_block > 1) { 1712 while (curpg && 1713 ((curpg->offset & fs->lfs_bmask) || 1714 curpg->offset >= vp->v_size || 1715 curpg->offset >= endoffset)) 1716 curpg = TAILQ_NEXT(curpg, listq); 1717 } 1718 if (curpg == NULL) 1719 break; 1720 soff = curpg->offset; 1721 } 1722 1723 /* 1724 * Mark all pages in extended range busy; find out if any 1725 * of them are dirty. 1726 */ 1727 nonexistent = dirty = 0; 1728 for (i = 0; i == 0 || i < pages_per_block; i++) { 1729 if (by_list && pages_per_block <= 1) { 1730 pgs[i] = pg = curpg; 1731 } else { 1732 off = soff + (i << PAGE_SHIFT); 1733 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1734 if (pg == NULL) { 1735 ++nonexistent; 1736 continue; 1737 } 1738 } 1739 KASSERT(pg != NULL); 1740 1741 /* 1742 * If we're holding the segment lock, we can deadlock 1743 * against a process that has our page and is waiting 1744 * for the cleaner, while the cleaner waits for the 1745 * segment lock. Just bail in that case. 1746 */ 1747 if ((pg->flags & PG_BUSY) && 1748 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1749 if (by_list && i > 0) 1750 uvm_page_unbusy(pgs, i); 1751 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1752 return -1; 1753 } 1754 1755 while (pg->flags & PG_BUSY) { 1756 pg->flags |= PG_WANTED; 1757 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1758 "lfsput", 0); 1759 simple_lock(&vp->v_interlock); 1760 if (by_list) { 1761 if (i > 0) 1762 uvm_page_unbusy(pgs, i); 1763 goto top; 1764 } 1765 } 1766 pg->flags |= PG_BUSY; 1767 UVM_PAGE_OWN(pg, "lfs_putpages"); 1768 1769 pmap_page_protect(pg, VM_PROT_NONE); 1770 tdirty = (pmap_clear_modify(pg) || 1771 (pg->flags & PG_CLEAN) == 0); 1772 dirty += tdirty; 1773 } 1774 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1775 if (by_list) { 1776 curpg = TAILQ_NEXT(curpg, listq); 1777 } else { 1778 soff += fs->lfs_bsize; 1779 } 1780 continue; 1781 } 1782 1783 any_dirty += dirty; 1784 KASSERT(nonexistent == 0); 1785 1786 /* 1787 * If any are dirty make all dirty; unbusy them, 1788 * but if we were asked to clean, wire them so that 1789 * the pagedaemon doesn't bother us about them while 1790 * they're on their way to disk. 1791 */ 1792 for (i = 0; i == 0 || i < pages_per_block; i++) { 1793 pg = pgs[i]; 1794 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1795 if (dirty) { 1796 pg->flags &= ~PG_CLEAN; 1797 if (flags & PGO_FREE) { 1798 /* 1799 * Wire the page so that 1800 * pdaemon doesn't see it again. 1801 */ 1802 uvm_lock_pageq(); 1803 uvm_pagewire(pg); 1804 uvm_unlock_pageq(); 1805 1806 /* Suspended write flag */ 1807 pg->flags |= PG_DELWRI; 1808 } 1809 } 1810 if (pg->flags & PG_WANTED) 1811 wakeup(pg); 1812 pg->flags &= ~(PG_WANTED|PG_BUSY); 1813 UVM_PAGE_OWN(pg, NULL); 1814 } 1815 1816 if (checkfirst && any_dirty) 1817 break; 1818 1819 if (by_list) { 1820 curpg = TAILQ_NEXT(curpg, listq); 1821 } else { 1822 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1823 } 1824 } 1825 1826 return any_dirty; 1827 } 1828 1829 /* 1830 * lfs_putpages functions like genfs_putpages except that 1831 * 1832 * (1) It needs to bounds-check the incoming requests to ensure that 1833 * they are block-aligned; if they are not, expand the range and 1834 * do the right thing in case, e.g., the requested range is clean 1835 * but the expanded range is dirty. 1836 * 1837 * (2) It needs to explicitly send blocks to be written when it is done. 1838 * VOP_PUTPAGES is not ever called with the seglock held, so 1839 * we simply take the seglock and let lfs_segunlock wait for us. 1840 * XXX Actually we can be called with the seglock held, if we have 1841 * XXX to flush a vnode while lfs_markv is in operation. As of this 1842 * XXX writing we panic in this case. 1843 * 1844 * Assumptions: 1845 * 1846 * (1) The caller does not hold any pages in this vnode busy. If it does, 1847 * there is a danger that when we expand the page range and busy the 1848 * pages we will deadlock. 1849 * 1850 * (2) We are called with vp->v_interlock held; we must return with it 1851 * released. 1852 * 1853 * (3) We don't absolutely have to free pages right away, provided that 1854 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1855 * us a request with PGO_FREE, we take the pages out of the paging 1856 * queue and wake up the writer, which will handle freeing them for us. 1857 * 1858 * We ensure that for any filesystem block, all pages for that 1859 * block are either resident or not, even if those pages are higher 1860 * than EOF; that means that we will be getting requests to free 1861 * "unused" pages above EOF all the time, and should ignore them. 1862 * 1863 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 1864 * into has been set up for us by lfs_writefile. If not, we will 1865 * have to handle allocating and/or freeing an finfo entry. 1866 * 1867 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1868 */ 1869 1870 int 1871 lfs_putpages(void *v) 1872 { 1873 int error; 1874 struct vop_putpages_args /* { 1875 struct vnode *a_vp; 1876 voff_t a_offlo; 1877 voff_t a_offhi; 1878 int a_flags; 1879 } */ *ap = v; 1880 struct vnode *vp; 1881 struct inode *ip; 1882 struct lfs *fs; 1883 struct segment *sp; 1884 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1885 off_t off, max_endoffset; 1886 int s; 1887 boolean_t seglocked, sync, pagedaemon; 1888 struct vm_page *pg; 1889 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1890 1891 vp = ap->a_vp; 1892 ip = VTOI(vp); 1893 fs = ip->i_lfs; 1894 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1895 pagedaemon = (curproc == uvm.pagedaemon_proc); 1896 1897 /* Putpages does nothing for metadata. */ 1898 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1899 simple_unlock(&vp->v_interlock); 1900 return 0; 1901 } 1902 1903 /* 1904 * If there are no pages, don't do anything. 1905 */ 1906 if (vp->v_uobj.uo_npages == 0) { 1907 s = splbio(); 1908 if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL && 1909 (vp->v_flag & VONWORKLST)) { 1910 vp->v_flag &= ~VONWORKLST; 1911 LIST_REMOVE(vp, v_synclist); 1912 } 1913 splx(s); 1914 simple_unlock(&vp->v_interlock); 1915 1916 /* Remove us from paging queue, if we were on it */ 1917 simple_lock(&fs->lfs_interlock); 1918 if (ip->i_flags & IN_PAGING) { 1919 ip->i_flags &= ~IN_PAGING; 1920 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1921 } 1922 simple_unlock(&fs->lfs_interlock); 1923 return 0; 1924 } 1925 1926 blkeof = blkroundup(fs, ip->i_size); 1927 1928 /* 1929 * Ignore requests to free pages past EOF but in the same block 1930 * as EOF, unless the request is synchronous. (If the request is 1931 * sync, it comes from lfs_truncate.) 1932 * XXXUBC Make these pages look "active" so the pagedaemon won't 1933 * XXXUBC bother us with them again. 1934 */ 1935 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1936 origoffset = ap->a_offlo; 1937 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1938 pg = uvm_pagelookup(&vp->v_uobj, off); 1939 KASSERT(pg != NULL); 1940 while (pg->flags & PG_BUSY) { 1941 pg->flags |= PG_WANTED; 1942 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1943 "lfsput2", 0); 1944 simple_lock(&vp->v_interlock); 1945 } 1946 uvm_lock_pageq(); 1947 uvm_pageactivate(pg); 1948 uvm_unlock_pageq(); 1949 } 1950 ap->a_offlo = blkeof; 1951 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1952 simple_unlock(&vp->v_interlock); 1953 return 0; 1954 } 1955 } 1956 1957 /* 1958 * Extend page range to start and end at block boundaries. 1959 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1960 */ 1961 origoffset = ap->a_offlo; 1962 origendoffset = ap->a_offhi; 1963 startoffset = origoffset & ~(fs->lfs_bmask); 1964 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1965 << fs->lfs_bshift; 1966 1967 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1968 endoffset = max_endoffset; 1969 origendoffset = endoffset; 1970 } else { 1971 origendoffset = round_page(ap->a_offhi); 1972 endoffset = round_page(blkroundup(fs, origendoffset)); 1973 } 1974 1975 KASSERT(startoffset > 0 || endoffset >= startoffset); 1976 if (startoffset == endoffset) { 1977 /* Nothing to do, why were we called? */ 1978 simple_unlock(&vp->v_interlock); 1979 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 1980 PRId64 "\n", startoffset)); 1981 return 0; 1982 } 1983 1984 ap->a_offlo = startoffset; 1985 ap->a_offhi = endoffset; 1986 1987 if (!(ap->a_flags & PGO_CLEANIT)) 1988 return genfs_putpages(v); 1989 1990 /* 1991 * If there are more than one page per block, we don't want 1992 * to get caught locking them backwards; so set PGO_BUSYFAIL 1993 * to avoid deadlocks. 1994 */ 1995 ap->a_flags |= PGO_BUSYFAIL; 1996 1997 do { 1998 int r; 1999 2000 /* If no pages are dirty, we can just use genfs_putpages. */ 2001 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2002 ap->a_flags, 1); 2003 if (r < 0) { 2004 simple_unlock(&vp->v_interlock); 2005 return EDEADLK; 2006 } 2007 if (r > 0) 2008 break; 2009 2010 /* 2011 * Sometimes pages are dirtied between the time that 2012 * we check and the time we try to clean them. 2013 * Instruct lfs_gop_write to return EDEADLK in this case 2014 * so we can write them properly. 2015 */ 2016 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2017 r = genfs_putpages(v); 2018 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2019 if (r != EDEADLK) 2020 return r; 2021 2022 /* Start over. */ 2023 preempt(1); 2024 simple_lock(&vp->v_interlock); 2025 } while(1); 2026 2027 /* 2028 * Dirty and asked to clean. 2029 * 2030 * Pagedaemon can't actually write LFS pages; wake up 2031 * the writer to take care of that. The writer will 2032 * notice the pager inode queue and act on that. 2033 */ 2034 if (pagedaemon) { 2035 simple_lock(&fs->lfs_interlock); 2036 if (!(ip->i_flags & IN_PAGING)) { 2037 ip->i_flags |= IN_PAGING; 2038 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2039 } 2040 simple_lock(&lfs_subsys_lock); 2041 wakeup(&lfs_writer_daemon); 2042 simple_unlock(&lfs_subsys_lock); 2043 simple_unlock(&fs->lfs_interlock); 2044 simple_unlock(&vp->v_interlock); 2045 preempt(1); 2046 return EWOULDBLOCK; 2047 } 2048 2049 /* 2050 * If this is a file created in a recent dirop, we can't flush its 2051 * inode until the dirop is complete. Drain dirops, then flush the 2052 * filesystem (taking care of any other pending dirops while we're 2053 * at it). 2054 */ 2055 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2056 (vp->v_flag & VDIROP)) { 2057 int locked; 2058 2059 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 2060 locked = VOP_ISLOCKED(vp) && /* XXX */ 2061 vp->v_lock.lk_lockholder == curproc->p_pid; 2062 simple_unlock(&vp->v_interlock); 2063 lfs_writer_enter(fs, "ppdirop"); 2064 if (locked) 2065 VOP_UNLOCK(vp, 0); 2066 2067 simple_lock(&fs->lfs_interlock); 2068 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2069 simple_unlock(&fs->lfs_interlock); 2070 2071 simple_lock(&vp->v_interlock); 2072 if (locked) { 2073 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2074 simple_lock(&vp->v_interlock); 2075 } 2076 lfs_writer_leave(fs); 2077 2078 /* XXX the flush should have taken care of this one too! */ 2079 } 2080 2081 /* 2082 * This is it. We are going to write some pages. From here on 2083 * down it's all just mechanics. 2084 * 2085 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2086 */ 2087 ap->a_flags &= ~PGO_SYNCIO; 2088 2089 /* 2090 * If we've already got the seglock, flush the node and return. 2091 * The FIP has already been set up for us by lfs_writefile, 2092 * and FIP cleanup and lfs_updatemeta will also be done there, 2093 * unless genfs_putpages returns EDEADLK; then we must flush 2094 * what we have, and correct FIP and segment header accounting. 2095 */ 2096 get_seglock: 2097 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2098 if (!seglocked) { 2099 simple_unlock(&vp->v_interlock); 2100 /* 2101 * Take the seglock, because we are going to be writing pages. 2102 */ 2103 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2104 if (error != 0) 2105 return error; 2106 simple_lock(&vp->v_interlock); 2107 } 2108 2109 /* 2110 * VOP_PUTPAGES should not be called while holding the seglock. 2111 * XXXUBC fix lfs_markv, or do this properly. 2112 */ 2113 #ifdef notyet 2114 KASSERT(fs->lfs_seglock == 1); 2115 #endif /* notyet */ 2116 2117 /* 2118 * We assume we're being called with sp->fip pointing at blank space. 2119 * Account for a new FIP in the segment header, and set sp->vp. 2120 * (This should duplicate the setup at the top of lfs_writefile().) 2121 */ 2122 sp = fs->lfs_sp; 2123 if (!seglocked) 2124 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2125 KASSERT(sp->vp == NULL); 2126 sp->vp = vp; 2127 2128 if (!seglocked) { 2129 if (vp->v_flag & VDIROP) 2130 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2131 } 2132 2133 /* 2134 * Loop through genfs_putpages until all pages are gathered. 2135 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2136 * Whenever we lose the interlock we have to rerun check_dirty, as 2137 * well. 2138 */ 2139 again: 2140 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2141 ap->a_flags, 0) < 0) { 2142 simple_unlock(&vp->v_interlock); 2143 sp->vp = NULL; 2144 if (!seglocked) { 2145 lfs_release_finfo(fs); 2146 lfs_segunlock(fs); 2147 } 2148 if (pagedaemon) 2149 return EDEADLK; 2150 /* else seglocked == 0 */ 2151 preempt(1); 2152 simple_lock(&vp->v_interlock); 2153 goto get_seglock; 2154 } 2155 2156 error = genfs_putpages(v); 2157 if (error == EDEADLK || error == EAGAIN) { 2158 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2159 " EDEADLK [2] ino %d off %x (seg %d)\n", 2160 ip->i_number, fs->lfs_offset, 2161 dtosn(fs, fs->lfs_offset))); 2162 /* If nothing to write, short-circuit */ 2163 if (sp->cbpp - sp->bpp > 1) { 2164 /* Write gathered pages */ 2165 lfs_updatemeta(sp); 2166 lfs_release_finfo(fs); 2167 (void) lfs_writeseg(fs, sp); 2168 2169 /* 2170 * Reinitialize brand new FIP and add us to it. 2171 */ 2172 KASSERT(sp->vp == vp); 2173 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2174 } 2175 2176 /* Give the write a chance to complete */ 2177 preempt(1); 2178 2179 /* We've lost the interlock. Start over. */ 2180 if (error == EDEADLK) { 2181 simple_lock(&vp->v_interlock); 2182 goto again; 2183 } 2184 } 2185 2186 KASSERT(sp->vp == vp); 2187 if (!seglocked) { 2188 sp->vp = NULL; 2189 2190 /* Write indirect blocks as well */ 2191 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2192 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2193 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2194 2195 KASSERT(sp->vp == NULL); 2196 sp->vp = vp; 2197 } 2198 2199 /* 2200 * Blocks are now gathered into a segment waiting to be written. 2201 * All that's left to do is update metadata, and write them. 2202 */ 2203 lfs_updatemeta(sp); 2204 KASSERT(sp->vp == vp); 2205 sp->vp = NULL; 2206 2207 if (seglocked) { 2208 /* we're called by lfs_writefile. */ 2209 return error; 2210 } 2211 2212 /* Clean up FIP and send it to disk. */ 2213 lfs_release_finfo(fs); 2214 lfs_writeseg(fs, fs->lfs_sp); 2215 2216 /* 2217 * Remove us from paging queue, since we've now written all our 2218 * pages. 2219 */ 2220 simple_lock(&fs->lfs_interlock); 2221 if (ip->i_flags & IN_PAGING) { 2222 ip->i_flags &= ~IN_PAGING; 2223 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2224 } 2225 simple_unlock(&fs->lfs_interlock); 2226 2227 /* 2228 * XXX - with the malloc/copy writeseg, the pages are freed by now 2229 * even if we don't wait (e.g. if we hold a nested lock). This 2230 * will not be true if we stop using malloc/copy. 2231 */ 2232 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2233 lfs_segunlock(fs); 2234 2235 /* 2236 * Wait for v_numoutput to drop to zero. The seglock should 2237 * take care of this, but there is a slight possibility that 2238 * aiodoned might not have got around to our buffers yet. 2239 */ 2240 if (sync) { 2241 s = splbio(); 2242 simple_lock(&global_v_numoutput_slock); 2243 while (vp->v_numoutput > 0) { 2244 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2245 " num %d\n", ip->i_number, vp->v_numoutput)); 2246 vp->v_flag |= VBWAIT; 2247 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 2248 &global_v_numoutput_slock); 2249 } 2250 simple_unlock(&global_v_numoutput_slock); 2251 splx(s); 2252 } 2253 return error; 2254 } 2255 2256 /* 2257 * Return the last logical file offset that should be written for this file 2258 * if we're doing a write that ends at "size". If writing, we need to know 2259 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2260 * to know about entire blocks. 2261 */ 2262 void 2263 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2264 { 2265 struct inode *ip = VTOI(vp); 2266 struct lfs *fs = ip->i_lfs; 2267 daddr_t olbn, nlbn; 2268 2269 olbn = lblkno(fs, ip->i_size); 2270 nlbn = lblkno(fs, size); 2271 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2272 *eobp = fragroundup(fs, size); 2273 } else { 2274 *eobp = blkroundup(fs, size); 2275 } 2276 } 2277 2278 #ifdef DEBUG 2279 void lfs_dump_vop(void *); 2280 2281 void 2282 lfs_dump_vop(void *v) 2283 { 2284 struct vop_putpages_args /* { 2285 struct vnode *a_vp; 2286 voff_t a_offlo; 2287 voff_t a_offhi; 2288 int a_flags; 2289 } */ *ap = v; 2290 2291 #ifdef DDB 2292 vfs_vnode_print(ap->a_vp, 0, printf); 2293 #endif 2294 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2295 } 2296 #endif 2297 2298 int 2299 lfs_mmap(void *v) 2300 { 2301 struct vop_mmap_args /* { 2302 const struct vnodeop_desc *a_desc; 2303 struct vnode *a_vp; 2304 int a_fflags; 2305 kauth_cred_t a_cred; 2306 struct lwp *a_l; 2307 } */ *ap = v; 2308 2309 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2310 return EOPNOTSUPP; 2311 return ufs_mmap(v); 2312 } 2313
Indexes created Tue Sep 30 20:09:53 GMT 2025