lfs_vnops.c revision 1.200
1 /* $NetBSD: lfs_vnops.c,v 1.200 2007/02/21 23:00:11 thorpej 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 1.200 2007/02/21 23:00:11 thorpej Exp $"); 71 72 #ifdef _KERNEL_OPT 73 #include "opt_compat_netbsd.h" 74 #endif 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/namei.h> 79 #include <sys/resourcevar.h> 80 #include <sys/kernel.h> 81 #include <sys/file.h> 82 #include <sys/stat.h> 83 #include <sys/buf.h> 84 #include <sys/proc.h> 85 #include <sys/mount.h> 86 #include <sys/vnode.h> 87 #include <sys/pool.h> 88 #include <sys/signalvar.h> 89 #include <sys/kauth.h> 90 #include <sys/syslog.h> 91 #include <sys/fstrans.h> 92 93 #include <miscfs/fifofs/fifo.h> 94 #include <miscfs/genfs/genfs.h> 95 #include <miscfs/specfs/specdev.h> 96 97 #include <ufs/ufs/inode.h> 98 #include <ufs/ufs/dir.h> 99 #include <ufs/ufs/ufsmount.h> 100 #include <ufs/ufs/ufs_extern.h> 101 102 #include <uvm/uvm.h> 103 #include <uvm/uvm_pmap.h> 104 #include <uvm/uvm_stat.h> 105 #include <uvm/uvm_pager.h> 106 107 #include <ufs/lfs/lfs.h> 108 #include <ufs/lfs/lfs_extern.h> 109 110 extern pid_t lfs_writer_daemon; 111 112 /* Global vfs data structures for lfs. */ 113 int (**lfs_vnodeop_p)(void *); 114 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 115 { &vop_default_desc, vn_default_error }, 116 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 117 { &vop_create_desc, lfs_create }, /* create */ 118 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 119 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 120 { &vop_open_desc, ufs_open }, /* open */ 121 { &vop_close_desc, lfs_close }, /* close */ 122 { &vop_access_desc, ufs_access }, /* access */ 123 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 124 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 125 { &vop_read_desc, lfs_read }, /* read */ 126 { &vop_write_desc, lfs_write }, /* write */ 127 { &vop_lease_desc, ufs_lease_check }, /* lease */ 128 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 129 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 130 { &vop_poll_desc, ufs_poll }, /* poll */ 131 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 132 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 133 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 134 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 135 { &vop_seek_desc, ufs_seek }, /* seek */ 136 { &vop_remove_desc, lfs_remove }, /* remove */ 137 { &vop_link_desc, lfs_link }, /* link */ 138 { &vop_rename_desc, lfs_rename }, /* rename */ 139 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 140 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 141 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 142 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 143 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 144 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 145 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 146 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 147 { &vop_lock_desc, ufs_lock }, /* lock */ 148 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 149 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 150 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 151 { &vop_print_desc, ufs_print }, /* print */ 152 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 153 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 154 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 155 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 156 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 157 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 158 { NULL, NULL } 159 }; 160 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 161 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 162 163 int (**lfs_specop_p)(void *); 164 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 165 { &vop_default_desc, vn_default_error }, 166 { &vop_lookup_desc, spec_lookup }, /* lookup */ 167 { &vop_create_desc, spec_create }, /* create */ 168 { &vop_mknod_desc, spec_mknod }, /* mknod */ 169 { &vop_open_desc, spec_open }, /* open */ 170 { &vop_close_desc, lfsspec_close }, /* close */ 171 { &vop_access_desc, ufs_access }, /* access */ 172 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 173 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 174 { &vop_read_desc, ufsspec_read }, /* read */ 175 { &vop_write_desc, ufsspec_write }, /* write */ 176 { &vop_lease_desc, spec_lease_check }, /* lease */ 177 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 178 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 179 { &vop_poll_desc, spec_poll }, /* poll */ 180 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 181 { &vop_revoke_desc, spec_revoke }, /* revoke */ 182 { &vop_mmap_desc, spec_mmap }, /* mmap */ 183 { &vop_fsync_desc, spec_fsync }, /* fsync */ 184 { &vop_seek_desc, spec_seek }, /* seek */ 185 { &vop_remove_desc, spec_remove }, /* remove */ 186 { &vop_link_desc, spec_link }, /* link */ 187 { &vop_rename_desc, spec_rename }, /* rename */ 188 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 189 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 190 { &vop_symlink_desc, spec_symlink }, /* symlink */ 191 { &vop_readdir_desc, spec_readdir }, /* readdir */ 192 { &vop_readlink_desc, spec_readlink }, /* readlink */ 193 { &vop_abortop_desc, spec_abortop }, /* abortop */ 194 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 195 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 196 { &vop_lock_desc, ufs_lock }, /* lock */ 197 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 198 { &vop_bmap_desc, spec_bmap }, /* bmap */ 199 { &vop_strategy_desc, spec_strategy }, /* strategy */ 200 { &vop_print_desc, ufs_print }, /* print */ 201 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 202 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 203 { &vop_advlock_desc, spec_advlock }, /* advlock */ 204 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 205 { &vop_getpages_desc, spec_getpages }, /* getpages */ 206 { &vop_putpages_desc, spec_putpages }, /* putpages */ 207 { NULL, NULL } 208 }; 209 const struct vnodeopv_desc lfs_specop_opv_desc = 210 { &lfs_specop_p, lfs_specop_entries }; 211 212 int (**lfs_fifoop_p)(void *); 213 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 214 { &vop_default_desc, vn_default_error }, 215 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 216 { &vop_create_desc, fifo_create }, /* create */ 217 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 218 { &vop_open_desc, fifo_open }, /* open */ 219 { &vop_close_desc, lfsfifo_close }, /* close */ 220 { &vop_access_desc, ufs_access }, /* access */ 221 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 222 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 223 { &vop_read_desc, ufsfifo_read }, /* read */ 224 { &vop_write_desc, ufsfifo_write }, /* write */ 225 { &vop_lease_desc, fifo_lease_check }, /* lease */ 226 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 227 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 228 { &vop_poll_desc, fifo_poll }, /* poll */ 229 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 230 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 231 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 232 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 233 { &vop_seek_desc, fifo_seek }, /* seek */ 234 { &vop_remove_desc, fifo_remove }, /* remove */ 235 { &vop_link_desc, fifo_link }, /* link */ 236 { &vop_rename_desc, fifo_rename }, /* rename */ 237 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 238 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 239 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 240 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 241 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 242 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 243 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 244 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 245 { &vop_lock_desc, ufs_lock }, /* lock */ 246 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 247 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 248 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 249 { &vop_print_desc, ufs_print }, /* print */ 250 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 251 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 252 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 253 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 254 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 255 { NULL, NULL } 256 }; 257 const struct vnodeopv_desc lfs_fifoop_opv_desc = 258 { &lfs_fifoop_p, lfs_fifoop_entries }; 259 260 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int); 261 262 #define LFS_READWRITE 263 #include <ufs/ufs/ufs_readwrite.c> 264 #undef LFS_READWRITE 265 266 /* 267 * Synch an open file. 268 */ 269 /* ARGSUSED */ 270 int 271 lfs_fsync(void *v) 272 { 273 struct vop_fsync_args /* { 274 struct vnode *a_vp; 275 kauth_cred_t a_cred; 276 int a_flags; 277 off_t offlo; 278 off_t offhi; 279 struct lwp *a_l; 280 } */ *ap = v; 281 struct vnode *vp = ap->a_vp; 282 int error, wait; 283 284 /* If we're mounted read-only, don't try to sync. */ 285 if (VTOI(vp)->i_lfs->lfs_ronly) 286 return 0; 287 288 /* 289 * Trickle sync checks for need to do a checkpoint after possible 290 * activity from the pagedaemon. 291 */ 292 if (ap->a_flags & FSYNC_LAZY) { 293 simple_lock(&lfs_subsys_lock); 294 wakeup(&lfs_writer_daemon); 295 simple_unlock(&lfs_subsys_lock); 296 return 0; 297 } 298 299 /* 300 * If a vnode is bring cleaned, flush it out before we try to 301 * reuse it. This prevents the cleaner from writing files twice 302 * in the same partial segment, causing an accounting underflow. 303 */ 304 if (ap->a_flags & FSYNC_RECLAIM && VTOI(vp)->i_flags & IN_CLEANING) { 305 /* printf("avoiding VONWORKLIST panic\n"); */ 306 lfs_vflush(vp); 307 } 308 309 wait = (ap->a_flags & FSYNC_WAIT); 310 simple_lock(&vp->v_interlock); 311 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 312 round_page(ap->a_offhi), 313 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 314 if (error) 315 return error; 316 error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 317 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 318 int l = 0; 319 error = VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE, 320 ap->a_l->l_cred, ap->a_l); 321 } 322 if (wait && !VPISEMPTY(vp)) 323 LFS_SET_UINO(VTOI(vp), IN_MODIFIED); 324 325 return error; 326 } 327 328 /* 329 * Take IN_ADIROP off, then call ufs_inactive. 330 */ 331 int 332 lfs_inactive(void *v) 333 { 334 struct vop_inactive_args /* { 335 struct vnode *a_vp; 336 struct lwp *a_l; 337 } */ *ap = v; 338 339 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 340 341 lfs_unmark_vnode(ap->a_vp); 342 343 /* 344 * The Ifile is only ever inactivated on unmount. 345 * Streamline this process by not giving it more dirty blocks. 346 */ 347 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 348 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 349 VOP_UNLOCK(ap->a_vp, 0); 350 return 0; 351 } 352 353 return ufs_inactive(v); 354 } 355 356 /* 357 * These macros are used to bracket UFS directory ops, so that we can 358 * identify all the pages touched during directory ops which need to 359 * be ordered and flushed atomically, so that they may be recovered. 360 * 361 * Because we have to mark nodes VDIROP in order to prevent 362 * the cache from reclaiming them while a dirop is in progress, we must 363 * also manage the number of nodes so marked (otherwise we can run out). 364 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 365 * is decremented during segment write, when VDIROP is taken off. 366 */ 367 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 368 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 369 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 370 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 371 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 372 static int lfs_set_dirop(struct vnode *, struct vnode *); 373 374 static int 375 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 376 { 377 struct lfs *fs; 378 int error; 379 380 KASSERT(VOP_ISLOCKED(dvp)); 381 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 382 383 fs = VTOI(dvp)->i_lfs; 384 385 ASSERT_NO_SEGLOCK(fs); 386 /* 387 * LFS_NRESERVE calculates direct and indirect blocks as well 388 * as an inode block; an overestimate in most cases. 389 */ 390 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 391 return (error); 392 393 restart: 394 simple_lock(&fs->lfs_interlock); 395 if (fs->lfs_dirops == 0) { 396 simple_unlock(&fs->lfs_interlock); 397 lfs_check(dvp, LFS_UNUSED_LBN, 0); 398 simple_lock(&fs->lfs_interlock); 399 } 400 while (fs->lfs_writer) { 401 error = ltsleep(&fs->lfs_dirops, (PRIBIO + 1) | PCATCH, 402 "lfs_sdirop", 0, &fs->lfs_interlock); 403 if (error == EINTR) { 404 simple_unlock(&fs->lfs_interlock); 405 goto unreserve; 406 } 407 } 408 simple_lock(&lfs_subsys_lock); 409 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 410 wakeup(&lfs_writer_daemon); 411 simple_unlock(&lfs_subsys_lock); 412 simple_unlock(&fs->lfs_interlock); 413 preempt(); 414 goto restart; 415 } 416 417 if (lfs_dirvcount > LFS_MAX_DIROP) { 418 simple_unlock(&fs->lfs_interlock); 419 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 420 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 421 if ((error = ltsleep(&lfs_dirvcount, 422 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 423 &lfs_subsys_lock)) != 0) { 424 goto unreserve; 425 } 426 goto restart; 427 } 428 simple_unlock(&lfs_subsys_lock); 429 430 ++fs->lfs_dirops; 431 fs->lfs_doifile = 1; 432 simple_unlock(&fs->lfs_interlock); 433 434 /* Hold a reference so SET_ENDOP will be happy */ 435 vref(dvp); 436 if (vp) { 437 vref(vp); 438 MARK_VNODE(vp); 439 } 440 441 MARK_VNODE(dvp); 442 return 0; 443 444 unreserve: 445 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 446 return error; 447 } 448 449 /* 450 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 451 * in getnewvnode(), if we have a stacked filesystem mounted on top 452 * of us. 453 * 454 * NB: this means we have to clear the new vnodes on error. Fortunately 455 * SET_ENDOP is there to do that for us. 456 */ 457 static int 458 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 459 { 460 int error; 461 struct lfs *fs; 462 463 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 464 ASSERT_NO_SEGLOCK(fs); 465 if (fs->lfs_ronly) 466 return EROFS; 467 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 468 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 469 dvp, error)); 470 return error; 471 } 472 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 473 if (vpp) { 474 ungetnewvnode(*vpp); 475 *vpp = NULL; 476 } 477 return error; 478 } 479 return 0; 480 } 481 482 #define SET_ENDOP_BASE(fs, dvp, str) \ 483 do { \ 484 simple_lock(&(fs)->lfs_interlock); \ 485 --(fs)->lfs_dirops; \ 486 if (!(fs)->lfs_dirops) { \ 487 if ((fs)->lfs_nadirop) { \ 488 panic("SET_ENDOP: %s: no dirops but " \ 489 " nadirop=%d", (str), \ 490 (fs)->lfs_nadirop); \ 491 } \ 492 wakeup(&(fs)->lfs_writer); \ 493 simple_unlock(&(fs)->lfs_interlock); \ 494 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 495 } else \ 496 simple_unlock(&(fs)->lfs_interlock); \ 497 } while(0) 498 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 499 do { \ 500 UNMARK_VNODE(dvp); \ 501 if (nvpp && *nvpp) \ 502 UNMARK_VNODE(*nvpp); \ 503 /* Check for error return to stem vnode leakage */ \ 504 if (nvpp && *nvpp && !((*nvpp)->v_flag & VDIROP)) \ 505 ungetnewvnode(*(nvpp)); \ 506 SET_ENDOP_BASE((fs), (dvp), (str)); \ 507 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 508 vrele(dvp); \ 509 } while(0) 510 #define SET_ENDOP_CREATE_AP(ap, str) \ 511 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 512 (ap)->a_vpp, (str)) 513 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 514 do { \ 515 UNMARK_VNODE(dvp); \ 516 if (ovp) \ 517 UNMARK_VNODE(ovp); \ 518 SET_ENDOP_BASE((fs), (dvp), (str)); \ 519 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 520 vrele(dvp); \ 521 if (ovp) \ 522 vrele(ovp); \ 523 } while(0) 524 525 void 526 lfs_mark_vnode(struct vnode *vp) 527 { 528 struct inode *ip = VTOI(vp); 529 struct lfs *fs = ip->i_lfs; 530 531 simple_lock(&fs->lfs_interlock); 532 if (!(ip->i_flag & IN_ADIROP)) { 533 if (!(vp->v_flag & VDIROP)) { 534 (void)lfs_vref(vp); 535 simple_lock(&lfs_subsys_lock); 536 ++lfs_dirvcount; 537 ++fs->lfs_dirvcount; 538 simple_unlock(&lfs_subsys_lock); 539 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 540 vp->v_flag |= VDIROP; 541 } 542 ++fs->lfs_nadirop; 543 ip->i_flag |= IN_ADIROP; 544 } else 545 KASSERT(vp->v_flag & VDIROP); 546 simple_unlock(&fs->lfs_interlock); 547 } 548 549 void 550 lfs_unmark_vnode(struct vnode *vp) 551 { 552 struct inode *ip = VTOI(vp); 553 554 if (ip && (ip->i_flag & IN_ADIROP)) { 555 KASSERT(vp->v_flag & VDIROP); 556 simple_lock(&ip->i_lfs->lfs_interlock); 557 --ip->i_lfs->lfs_nadirop; 558 simple_unlock(&ip->i_lfs->lfs_interlock); 559 ip->i_flag &= ~IN_ADIROP; 560 } 561 } 562 563 int 564 lfs_symlink(void *v) 565 { 566 struct vop_symlink_args /* { 567 struct vnode *a_dvp; 568 struct vnode **a_vpp; 569 struct componentname *a_cnp; 570 struct vattr *a_vap; 571 char *a_target; 572 } */ *ap = v; 573 int error; 574 575 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 576 vput(ap->a_dvp); 577 return error; 578 } 579 error = ufs_symlink(ap); 580 SET_ENDOP_CREATE_AP(ap, "symlink"); 581 return (error); 582 } 583 584 int 585 lfs_mknod(void *v) 586 { 587 struct vop_mknod_args /* { 588 struct vnode *a_dvp; 589 struct vnode **a_vpp; 590 struct componentname *a_cnp; 591 struct vattr *a_vap; 592 } */ *ap = v; 593 struct vattr *vap = ap->a_vap; 594 struct vnode **vpp = ap->a_vpp; 595 struct inode *ip; 596 int error; 597 struct mount *mp; 598 ino_t ino; 599 600 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 601 vput(ap->a_dvp); 602 return error; 603 } 604 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 605 ap->a_dvp, vpp, ap->a_cnp); 606 607 /* Either way we're done with the dirop at this point */ 608 SET_ENDOP_CREATE_AP(ap, "mknod"); 609 610 if (error) 611 return (error); 612 613 ip = VTOI(*vpp); 614 mp = (*vpp)->v_mount; 615 ino = ip->i_number; 616 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 617 if (vap->va_rdev != VNOVAL) { 618 /* 619 * Want to be able to use this to make badblock 620 * inodes, so don't truncate the dev number. 621 */ 622 #if 0 623 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 624 UFS_MPNEEDSWAP((*vpp)->v_mount)); 625 #else 626 ip->i_ffs1_rdev = vap->va_rdev; 627 #endif 628 } 629 630 /* 631 * Call fsync to write the vnode so that we don't have to deal with 632 * flushing it when it's marked VDIROP|VXLOCK. 633 * 634 * XXX KS - If we can't flush we also can't call vgone(), so must 635 * return. But, that leaves this vnode in limbo, also not good. 636 * Can this ever happen (barring hardware failure)? 637 */ 638 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 639 curlwp)) != 0) { 640 panic("lfs_mknod: couldn't fsync (ino %llu)", 641 (unsigned long long)ino); 642 /* return (error); */ 643 } 644 /* 645 * Remove vnode so that it will be reloaded by VFS_VGET and 646 * checked to see if it is an alias of an existing entry in 647 * the inode cache. 648 */ 649 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 650 651 VOP_UNLOCK(*vpp, 0); 652 lfs_vunref(*vpp); 653 (*vpp)->v_type = VNON; 654 vgone(*vpp); 655 error = VFS_VGET(mp, ino, vpp); 656 657 if (error != 0) { 658 *vpp = NULL; 659 return (error); 660 } 661 return (0); 662 } 663 664 int 665 lfs_create(void *v) 666 { 667 struct vop_create_args /* { 668 struct vnode *a_dvp; 669 struct vnode **a_vpp; 670 struct componentname *a_cnp; 671 struct vattr *a_vap; 672 } */ *ap = v; 673 int error; 674 675 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 676 vput(ap->a_dvp); 677 return error; 678 } 679 error = ufs_create(ap); 680 SET_ENDOP_CREATE_AP(ap, "create"); 681 return (error); 682 } 683 684 int 685 lfs_mkdir(void *v) 686 { 687 struct vop_mkdir_args /* { 688 struct vnode *a_dvp; 689 struct vnode **a_vpp; 690 struct componentname *a_cnp; 691 struct vattr *a_vap; 692 } */ *ap = v; 693 int error; 694 695 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 696 vput(ap->a_dvp); 697 return error; 698 } 699 error = ufs_mkdir(ap); 700 SET_ENDOP_CREATE_AP(ap, "mkdir"); 701 return (error); 702 } 703 704 int 705 lfs_remove(void *v) 706 { 707 struct vop_remove_args /* { 708 struct vnode *a_dvp; 709 struct vnode *a_vp; 710 struct componentname *a_cnp; 711 } */ *ap = v; 712 struct vnode *dvp, *vp; 713 struct inode *ip; 714 int error; 715 716 dvp = ap->a_dvp; 717 vp = ap->a_vp; 718 ip = VTOI(vp); 719 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 720 if (dvp == vp) 721 vrele(vp); 722 else 723 vput(vp); 724 vput(dvp); 725 return error; 726 } 727 error = ufs_remove(ap); 728 if (ip->i_nlink == 0) 729 lfs_orphan(ip->i_lfs, ip->i_number); 730 SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove"); 731 return (error); 732 } 733 734 int 735 lfs_rmdir(void *v) 736 { 737 struct vop_rmdir_args /* { 738 struct vnodeop_desc *a_desc; 739 struct vnode *a_dvp; 740 struct vnode *a_vp; 741 struct componentname *a_cnp; 742 } */ *ap = v; 743 struct vnode *vp; 744 struct inode *ip; 745 int error; 746 747 vp = ap->a_vp; 748 ip = VTOI(vp); 749 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 750 if (ap->a_dvp == vp) 751 vrele(ap->a_dvp); 752 else 753 vput(ap->a_dvp); 754 vput(vp); 755 return error; 756 } 757 error = ufs_rmdir(ap); 758 if (ip->i_nlink == 0) 759 lfs_orphan(ip->i_lfs, ip->i_number); 760 SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 761 return (error); 762 } 763 764 int 765 lfs_link(void *v) 766 { 767 struct vop_link_args /* { 768 struct vnode *a_dvp; 769 struct vnode *a_vp; 770 struct componentname *a_cnp; 771 } */ *ap = v; 772 int error; 773 struct vnode **vpp = NULL; 774 775 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 776 vput(ap->a_dvp); 777 return error; 778 } 779 error = ufs_link(ap); 780 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 781 return (error); 782 } 783 784 int 785 lfs_rename(void *v) 786 { 787 struct vop_rename_args /* { 788 struct vnode *a_fdvp; 789 struct vnode *a_fvp; 790 struct componentname *a_fcnp; 791 struct vnode *a_tdvp; 792 struct vnode *a_tvp; 793 struct componentname *a_tcnp; 794 } */ *ap = v; 795 struct vnode *tvp, *fvp, *tdvp, *fdvp; 796 struct componentname *tcnp, *fcnp; 797 int error; 798 struct lfs *fs; 799 800 fs = VTOI(ap->a_fdvp)->i_lfs; 801 tvp = ap->a_tvp; 802 tdvp = ap->a_tdvp; 803 tcnp = ap->a_tcnp; 804 fvp = ap->a_fvp; 805 fdvp = ap->a_fdvp; 806 fcnp = ap->a_fcnp; 807 808 /* 809 * Check for cross-device rename. 810 * If it is, we don't want to set dirops, just error out. 811 * (In particular note that MARK_VNODE(tdvp) will DTWT on 812 * a cross-device rename.) 813 * 814 * Copied from ufs_rename. 815 */ 816 if ((fvp->v_mount != tdvp->v_mount) || 817 (tvp && (fvp->v_mount != tvp->v_mount))) { 818 error = EXDEV; 819 goto errout; 820 } 821 822 /* 823 * Check to make sure we're not renaming a vnode onto itself 824 * (deleting a hard link by renaming one name onto another); 825 * if we are we can't recursively call VOP_REMOVE since that 826 * would leave us with an unaccounted-for number of live dirops. 827 * 828 * Inline the relevant section of ufs_rename here, *before* 829 * calling SET_DIROP_REMOVE. 830 */ 831 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 832 (VTOI(tdvp)->i_flags & APPEND))) { 833 error = EPERM; 834 goto errout; 835 } 836 if (fvp == tvp) { 837 if (fvp->v_type == VDIR) { 838 error = EINVAL; 839 goto errout; 840 } 841 842 /* Release destination completely. */ 843 VOP_ABORTOP(tdvp, tcnp); 844 vput(tdvp); 845 vput(tvp); 846 847 /* Delete source. */ 848 vrele(fvp); 849 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 850 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 851 fcnp->cn_nameiop = DELETE; 852 vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY); 853 if ((error = relookup(fdvp, &fvp, fcnp))) { 854 vput(fdvp); 855 return (error); 856 } 857 return (VOP_REMOVE(fdvp, fvp, fcnp)); 858 } 859 860 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 861 goto errout; 862 MARK_VNODE(fdvp); 863 MARK_VNODE(fvp); 864 865 error = ufs_rename(ap); 866 UNMARK_VNODE(fdvp); 867 UNMARK_VNODE(fvp); 868 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 869 return (error); 870 871 errout: 872 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 873 if (tdvp == tvp) 874 vrele(tdvp); 875 else 876 vput(tdvp); 877 if (tvp) 878 vput(tvp); 879 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 880 vrele(fdvp); 881 vrele(fvp); 882 return (error); 883 } 884 885 /* XXX hack to avoid calling ITIMES in getattr */ 886 int 887 lfs_getattr(void *v) 888 { 889 struct vop_getattr_args /* { 890 struct vnode *a_vp; 891 struct vattr *a_vap; 892 kauth_cred_t a_cred; 893 struct lwp *a_l; 894 } */ *ap = v; 895 struct vnode *vp = ap->a_vp; 896 struct inode *ip = VTOI(vp); 897 struct vattr *vap = ap->a_vap; 898 struct lfs *fs = ip->i_lfs; 899 /* 900 * Copy from inode table 901 */ 902 vap->va_fsid = ip->i_dev; 903 vap->va_fileid = ip->i_number; 904 vap->va_mode = ip->i_mode & ~IFMT; 905 vap->va_nlink = ip->i_nlink; 906 vap->va_uid = ip->i_uid; 907 vap->va_gid = ip->i_gid; 908 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 909 vap->va_size = vp->v_size; 910 vap->va_atime.tv_sec = ip->i_ffs1_atime; 911 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 912 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 913 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 914 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 915 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 916 vap->va_flags = ip->i_flags; 917 vap->va_gen = ip->i_gen; 918 /* this doesn't belong here */ 919 if (vp->v_type == VBLK) 920 vap->va_blocksize = BLKDEV_IOSIZE; 921 else if (vp->v_type == VCHR) 922 vap->va_blocksize = MAXBSIZE; 923 else 924 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 925 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 926 vap->va_type = vp->v_type; 927 vap->va_filerev = ip->i_modrev; 928 return (0); 929 } 930 931 /* 932 * Check to make sure the inode blocks won't choke the buffer 933 * cache, then call ufs_setattr as usual. 934 */ 935 int 936 lfs_setattr(void *v) 937 { 938 struct vop_setattr_args /* { 939 struct vnode *a_vp; 940 struct vattr *a_vap; 941 kauth_cred_t a_cred; 942 struct lwp *a_l; 943 } */ *ap = v; 944 struct vnode *vp = ap->a_vp; 945 946 lfs_check(vp, LFS_UNUSED_LBN, 0); 947 return ufs_setattr(v); 948 } 949 950 /* 951 * Release the block we hold on lfs_newseg wrapping. Called on file close, 952 * or explicitly from LFCNWRAPGO. Called with the interlock held. 953 */ 954 static int 955 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor) 956 { 957 if (lockstatus(&fs->lfs_stoplock) != LK_EXCLUSIVE) 958 return EBUSY; 959 960 lockmgr(&fs->lfs_stoplock, LK_RELEASE, &fs->lfs_interlock); 961 962 KASSERT(fs->lfs_nowrap > 0); 963 if (fs->lfs_nowrap <= 0) { 964 return 0; 965 } 966 967 if (--fs->lfs_nowrap == 0) { 968 log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt); 969 wakeup(&fs->lfs_wrappass); 970 lfs_wakeup_cleaner(fs); 971 } 972 if (waitfor) { 973 ltsleep(&fs->lfs_nextseg, PCATCH | PUSER, 974 "segment", 0, &fs->lfs_interlock); 975 } 976 977 return 0; 978 } 979 980 /* 981 * Close called 982 */ 983 /* ARGSUSED */ 984 int 985 lfs_close(void *v) 986 { 987 struct vop_close_args /* { 988 struct vnode *a_vp; 989 int a_fflag; 990 kauth_cred_t a_cred; 991 struct lwp *a_l; 992 } */ *ap = v; 993 struct vnode *vp = ap->a_vp; 994 struct inode *ip = VTOI(vp); 995 struct lfs *fs = ip->i_lfs; 996 997 if ((ip->i_number == ROOTINO || ip->i_number == LFS_IFILE_INUM) && 998 lockstatus(&fs->lfs_stoplock) == LK_EXCLUSIVE) { 999 simple_lock(&fs->lfs_interlock); 1000 log(LOG_NOTICE, "lfs_close: releasing log wrap control\n"); 1001 lfs_wrapgo(fs, ip, 0); 1002 simple_unlock(&fs->lfs_interlock); 1003 } 1004 1005 if (vp == ip->i_lfs->lfs_ivnode && 1006 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 1007 return 0; 1008 1009 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 1010 LFS_ITIMES(ip, NULL, NULL, NULL); 1011 } 1012 return (0); 1013 } 1014 1015 /* 1016 * Close wrapper for special devices. 1017 * 1018 * Update the times on the inode then do device close. 1019 */ 1020 int 1021 lfsspec_close(void *v) 1022 { 1023 struct vop_close_args /* { 1024 struct vnode *a_vp; 1025 int a_fflag; 1026 kauth_cred_t a_cred; 1027 struct lwp *a_l; 1028 } */ *ap = v; 1029 struct vnode *vp; 1030 struct inode *ip; 1031 1032 vp = ap->a_vp; 1033 ip = VTOI(vp); 1034 if (vp->v_usecount > 1) { 1035 LFS_ITIMES(ip, NULL, NULL, NULL); 1036 } 1037 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1038 } 1039 1040 /* 1041 * Close wrapper for fifo's. 1042 * 1043 * Update the times on the inode then do device close. 1044 */ 1045 int 1046 lfsfifo_close(void *v) 1047 { 1048 struct vop_close_args /* { 1049 struct vnode *a_vp; 1050 int a_fflag; 1051 kauth_cred_ a_cred; 1052 struct lwp *a_l; 1053 } */ *ap = v; 1054 struct vnode *vp; 1055 struct inode *ip; 1056 1057 vp = ap->a_vp; 1058 ip = VTOI(vp); 1059 if (ap->a_vp->v_usecount > 1) { 1060 LFS_ITIMES(ip, NULL, NULL, NULL); 1061 } 1062 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1063 } 1064 1065 /* 1066 * Reclaim an inode so that it can be used for other purposes. 1067 */ 1068 1069 int 1070 lfs_reclaim(void *v) 1071 { 1072 struct vop_reclaim_args /* { 1073 struct vnode *a_vp; 1074 struct lwp *a_l; 1075 } */ *ap = v; 1076 struct vnode *vp = ap->a_vp; 1077 struct inode *ip = VTOI(vp); 1078 int error; 1079 1080 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1081 1082 LFS_CLR_UINO(ip, IN_ALLMOD); 1083 if ((error = ufs_reclaim(vp, ap->a_l))) 1084 return (error); 1085 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1086 lfs_deregister_all(vp); 1087 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1088 ip->inode_ext.lfs = NULL; 1089 genfs_node_destroy(vp); 1090 pool_put(&lfs_inode_pool, vp->v_data); 1091 vp->v_data = NULL; 1092 return (0); 1093 } 1094 1095 /* 1096 * Read a block from a storage device. 1097 * In order to avoid reading blocks that are in the process of being 1098 * written by the cleaner---and hence are not mutexed by the normal 1099 * buffer cache / page cache mechanisms---check for collisions before 1100 * reading. 1101 * 1102 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1103 * the active cleaner test. 1104 * 1105 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1106 */ 1107 int 1108 lfs_strategy(void *v) 1109 { 1110 struct vop_strategy_args /* { 1111 struct vnode *a_vp; 1112 struct buf *a_bp; 1113 } */ *ap = v; 1114 struct buf *bp; 1115 struct lfs *fs; 1116 struct vnode *vp; 1117 struct inode *ip; 1118 daddr_t tbn; 1119 int i, sn, error, slept; 1120 1121 bp = ap->a_bp; 1122 vp = ap->a_vp; 1123 ip = VTOI(vp); 1124 fs = ip->i_lfs; 1125 1126 /* lfs uses its strategy routine only for read */ 1127 KASSERT(bp->b_flags & B_READ); 1128 1129 if (vp->v_type == VBLK || vp->v_type == VCHR) 1130 panic("lfs_strategy: spec"); 1131 KASSERT(bp->b_bcount != 0); 1132 if (bp->b_blkno == bp->b_lblkno) { 1133 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1134 NULL); 1135 if (error) { 1136 bp->b_error = error; 1137 bp->b_flags |= B_ERROR; 1138 biodone(bp); 1139 return (error); 1140 } 1141 if ((long)bp->b_blkno == -1) /* no valid data */ 1142 clrbuf(bp); 1143 } 1144 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1145 biodone(bp); 1146 return (0); 1147 } 1148 1149 slept = 1; 1150 simple_lock(&fs->lfs_interlock); 1151 while (slept && fs->lfs_seglock) { 1152 simple_unlock(&fs->lfs_interlock); 1153 /* 1154 * Look through list of intervals. 1155 * There will only be intervals to look through 1156 * if the cleaner holds the seglock. 1157 * Since the cleaner is synchronous, we can trust 1158 * the list of intervals to be current. 1159 */ 1160 tbn = dbtofsb(fs, bp->b_blkno); 1161 sn = dtosn(fs, tbn); 1162 slept = 0; 1163 for (i = 0; i < fs->lfs_cleanind; i++) { 1164 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1165 tbn >= fs->lfs_cleanint[i]) { 1166 DLOG((DLOG_CLEAN, 1167 "lfs_strategy: ino %d lbn %" PRId64 1168 " ind %d sn %d fsb %" PRIx32 1169 " given sn %d fsb %" PRIx64 "\n", 1170 ip->i_number, bp->b_lblkno, i, 1171 dtosn(fs, fs->lfs_cleanint[i]), 1172 fs->lfs_cleanint[i], sn, tbn)); 1173 DLOG((DLOG_CLEAN, 1174 "lfs_strategy: sleeping on ino %d lbn %" 1175 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1176 simple_lock(&fs->lfs_interlock); 1177 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1178 /* Cleaner can't wait for itself */ 1179 ltsleep(&fs->lfs_iocount, 1180 (PRIBIO + 1) | PNORELOCK, 1181 "clean2", 0, 1182 &fs->lfs_interlock); 1183 slept = 1; 1184 break; 1185 } else if (fs->lfs_seglock) { 1186 ltsleep(&fs->lfs_seglock, 1187 (PRIBIO + 1) | PNORELOCK, 1188 "clean1", 0, 1189 &fs->lfs_interlock); 1190 slept = 1; 1191 break; 1192 } 1193 simple_unlock(&fs->lfs_interlock); 1194 } 1195 } 1196 simple_lock(&fs->lfs_interlock); 1197 } 1198 simple_unlock(&fs->lfs_interlock); 1199 1200 vp = ip->i_devvp; 1201 VOP_STRATEGY(vp, bp); 1202 return (0); 1203 } 1204 1205 void 1206 lfs_flush_dirops(struct lfs *fs) 1207 { 1208 struct inode *ip, *nip; 1209 struct vnode *vp; 1210 extern int lfs_dostats; 1211 struct segment *sp; 1212 int waslocked; 1213 1214 ASSERT_MAYBE_SEGLOCK(fs); 1215 KASSERT(fs->lfs_nadirop == 0); 1216 1217 if (fs->lfs_ronly) 1218 return; 1219 1220 simple_lock(&fs->lfs_interlock); 1221 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1222 simple_unlock(&fs->lfs_interlock); 1223 return; 1224 } else 1225 simple_unlock(&fs->lfs_interlock); 1226 1227 if (lfs_dostats) 1228 ++lfs_stats.flush_invoked; 1229 1230 /* 1231 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1232 * Technically this is a checkpoint (the on-disk state is valid) 1233 * even though we are leaving out all the file data. 1234 */ 1235 lfs_imtime(fs); 1236 lfs_seglock(fs, SEGM_CKP); 1237 sp = fs->lfs_sp; 1238 1239 /* 1240 * lfs_writevnodes, optimized to get dirops out of the way. 1241 * Only write dirops, and don't flush files' pages, only 1242 * blocks from the directories. 1243 * 1244 * We don't need to vref these files because they are 1245 * dirops and so hold an extra reference until the 1246 * segunlock clears them of that status. 1247 * 1248 * We don't need to check for IN_ADIROP because we know that 1249 * no dirops are active. 1250 * 1251 */ 1252 simple_lock(&fs->lfs_interlock); 1253 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1254 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1255 simple_unlock(&fs->lfs_interlock); 1256 vp = ITOV(ip); 1257 1258 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1259 1260 /* 1261 * All writes to directories come from dirops; all 1262 * writes to files' direct blocks go through the page 1263 * cache, which we're not touching. Reads to files 1264 * and/or directories will not be affected by writing 1265 * directory blocks inodes and file inodes. So we don't 1266 * really need to lock. If we don't lock, though, 1267 * make sure that we don't clear IN_MODIFIED 1268 * unnecessarily. 1269 */ 1270 if (vp->v_flag & (VXLOCK | VFREEING)) { 1271 simple_lock(&fs->lfs_interlock); 1272 continue; 1273 } 1274 waslocked = VOP_ISLOCKED(vp); 1275 if (vp->v_type != VREG && 1276 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1277 lfs_writefile(fs, sp, vp); 1278 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1279 !(ip->i_flag & IN_ALLMOD)) { 1280 LFS_SET_UINO(ip, IN_MODIFIED); 1281 } 1282 } 1283 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1284 (void) lfs_writeinode(fs, sp, ip); 1285 if (waslocked == LK_EXCLOTHER) 1286 LFS_SET_UINO(ip, IN_MODIFIED); 1287 simple_lock(&fs->lfs_interlock); 1288 } 1289 simple_unlock(&fs->lfs_interlock); 1290 /* We've written all the dirops there are */ 1291 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1292 lfs_finalize_fs_seguse(fs); 1293 (void) lfs_writeseg(fs, sp); 1294 lfs_segunlock(fs); 1295 } 1296 1297 /* 1298 * Flush all vnodes for which the pagedaemon has requested pageouts. 1299 * Skip over any files that are marked VDIROP (since lfs_flush_dirop() 1300 * has just run, this would be an error). If we have to skip a vnode 1301 * for any reason, just skip it; if we have to wait for the cleaner, 1302 * abort. The writer daemon will call us again later. 1303 */ 1304 void 1305 lfs_flush_pchain(struct lfs *fs) 1306 { 1307 struct inode *ip, *nip; 1308 struct vnode *vp; 1309 extern int lfs_dostats; 1310 struct segment *sp; 1311 int error; 1312 1313 ASSERT_NO_SEGLOCK(fs); 1314 1315 if (fs->lfs_ronly) 1316 return; 1317 1318 simple_lock(&fs->lfs_interlock); 1319 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1320 simple_unlock(&fs->lfs_interlock); 1321 return; 1322 } else 1323 simple_unlock(&fs->lfs_interlock); 1324 1325 /* Get dirops out of the way */ 1326 lfs_flush_dirops(fs); 1327 1328 if (lfs_dostats) 1329 ++lfs_stats.flush_invoked; 1330 1331 /* 1332 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1333 */ 1334 lfs_imtime(fs); 1335 lfs_seglock(fs, 0); 1336 sp = fs->lfs_sp; 1337 1338 /* 1339 * lfs_writevnodes, optimized to clear pageout requests. 1340 * Only write non-dirop files that are in the pageout queue. 1341 * We're very conservative about what we write; we want to be 1342 * fast and async. 1343 */ 1344 simple_lock(&fs->lfs_interlock); 1345 top: 1346 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1347 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1348 vp = ITOV(ip); 1349 1350 if (!(ip->i_flags & IN_PAGING)) 1351 goto top; 1352 1353 if (vp->v_flag & (VXLOCK|VDIROP)) 1354 continue; 1355 if (vp->v_type != VREG) 1356 continue; 1357 if (lfs_vref(vp)) 1358 continue; 1359 simple_unlock(&fs->lfs_interlock); 1360 1361 if (VOP_ISLOCKED(vp)) { 1362 lfs_vunref(vp); 1363 simple_lock(&fs->lfs_interlock); 1364 continue; 1365 } 1366 1367 error = lfs_writefile(fs, sp, vp); 1368 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1369 !(ip->i_flag & IN_ALLMOD)) { 1370 LFS_SET_UINO(ip, IN_MODIFIED); 1371 } 1372 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1373 (void) lfs_writeinode(fs, sp, ip); 1374 1375 lfs_vunref(vp); 1376 1377 if (error == EAGAIN) { 1378 lfs_writeseg(fs, sp); 1379 simple_lock(&fs->lfs_interlock); 1380 break; 1381 } 1382 simple_lock(&fs->lfs_interlock); 1383 } 1384 simple_unlock(&fs->lfs_interlock); 1385 (void) lfs_writeseg(fs, sp); 1386 lfs_segunlock(fs); 1387 } 1388 1389 /* 1390 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1391 */ 1392 int 1393 lfs_fcntl(void *v) 1394 { 1395 struct vop_fcntl_args /* { 1396 struct vnode *a_vp; 1397 u_long a_command; 1398 caddr_t a_data; 1399 int a_fflag; 1400 kauth_cred_t a_cred; 1401 struct lwp *a_l; 1402 } */ *ap = v; 1403 struct timeval *tvp; 1404 BLOCK_INFO *blkiov; 1405 CLEANERINFO *cip; 1406 SEGUSE *sup; 1407 int blkcnt, error, oclean; 1408 size_t fh_size; 1409 struct lfs_fcntl_markv blkvp; 1410 struct lwp *l; 1411 fsid_t *fsidp; 1412 struct lfs *fs; 1413 struct buf *bp; 1414 fhandle_t *fhp; 1415 daddr_t off; 1416 1417 /* Only respect LFS fcntls on fs root or Ifile */ 1418 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1419 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1420 return ufs_fcntl(v); 1421 } 1422 1423 /* Avoid locking a draining lock */ 1424 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1425 return ESHUTDOWN; 1426 } 1427 1428 /* LFS control and monitoring fcntls are available only to root */ 1429 l = ap->a_l; 1430 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1431 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1432 NULL)) != 0) 1433 return (error); 1434 1435 fs = VTOI(ap->a_vp)->i_lfs; 1436 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1437 1438 error = 0; 1439 switch (ap->a_command) { 1440 case LFCNSEGWAITALL: 1441 case LFCNSEGWAITALL_COMPAT: 1442 fsidp = NULL; 1443 /* FALLSTHROUGH */ 1444 case LFCNSEGWAIT: 1445 case LFCNSEGWAIT_COMPAT: 1446 tvp = (struct timeval *)ap->a_data; 1447 simple_lock(&fs->lfs_interlock); 1448 ++fs->lfs_sleepers; 1449 simple_unlock(&fs->lfs_interlock); 1450 1451 error = lfs_segwait(fsidp, tvp); 1452 1453 simple_lock(&fs->lfs_interlock); 1454 if (--fs->lfs_sleepers == 0) 1455 wakeup(&fs->lfs_sleepers); 1456 simple_unlock(&fs->lfs_interlock); 1457 return error; 1458 1459 case LFCNBMAPV: 1460 case LFCNMARKV: 1461 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1462 1463 blkcnt = blkvp.blkcnt; 1464 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1465 return (EINVAL); 1466 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1467 if ((error = copyin(blkvp.blkiov, blkiov, 1468 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1469 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1470 return error; 1471 } 1472 1473 simple_lock(&fs->lfs_interlock); 1474 ++fs->lfs_sleepers; 1475 simple_unlock(&fs->lfs_interlock); 1476 if (ap->a_command == LFCNBMAPV) 1477 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1478 else /* LFCNMARKV */ 1479 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1480 if (error == 0) 1481 error = copyout(blkiov, blkvp.blkiov, 1482 blkcnt * sizeof(BLOCK_INFO)); 1483 simple_lock(&fs->lfs_interlock); 1484 if (--fs->lfs_sleepers == 0) 1485 wakeup(&fs->lfs_sleepers); 1486 simple_unlock(&fs->lfs_interlock); 1487 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1488 return error; 1489 1490 case LFCNRECLAIM: 1491 /* 1492 * Flush dirops and write Ifile, allowing empty segments 1493 * to be immediately reclaimed. 1494 */ 1495 lfs_writer_enter(fs, "pndirop"); 1496 off = fs->lfs_offset; 1497 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1498 lfs_flush_dirops(fs); 1499 LFS_CLEANERINFO(cip, fs, bp); 1500 oclean = cip->clean; 1501 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1502 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1503 fs->lfs_sp->seg_flags |= SEGM_PROT; 1504 lfs_segunlock(fs); 1505 lfs_writer_leave(fs); 1506 1507 #ifdef DEBUG 1508 LFS_CLEANERINFO(cip, fs, bp); 1509 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1510 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1511 fs->lfs_offset - off, cip->clean - oclean, 1512 fs->lfs_activesb)); 1513 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1514 #endif 1515 1516 return 0; 1517 1518 #ifdef COMPAT_30 1519 case LFCNIFILEFH_COMPAT: 1520 /* Return the filehandle of the Ifile */ 1521 if ((error = kauth_authorize_generic(l->l_cred, 1522 KAUTH_GENERIC_ISSUSER, NULL)) != 0) 1523 return (error); 1524 fhp = (struct fhandle *)ap->a_data; 1525 fhp->fh_fsid = *fsidp; 1526 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1527 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1528 #endif 1529 1530 case LFCNIFILEFH_COMPAT2: 1531 case LFCNIFILEFH: 1532 /* Return the filehandle of the Ifile */ 1533 fhp = (struct fhandle *)ap->a_data; 1534 fhp->fh_fsid = *fsidp; 1535 fh_size = sizeof(struct lfs_fhandle) - 1536 offsetof(fhandle_t, fh_fid); 1537 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1538 1539 case LFCNREWIND: 1540 /* Move lfs_offset to the lowest-numbered segment */ 1541 return lfs_rewind(fs, *(int *)ap->a_data); 1542 1543 case LFCNINVAL: 1544 /* Mark a segment SEGUSE_INVAL */ 1545 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1546 if (sup->su_nbytes > 0) { 1547 brelse(bp); 1548 lfs_unset_inval_all(fs); 1549 return EBUSY; 1550 } 1551 sup->su_flags |= SEGUSE_INVAL; 1552 VOP_BWRITE(bp); 1553 return 0; 1554 1555 case LFCNRESIZE: 1556 /* Resize the filesystem */ 1557 return lfs_resize_fs(fs, *(int *)ap->a_data); 1558 1559 case LFCNWRAPSTOP: 1560 case LFCNWRAPSTOP_COMPAT: 1561 /* 1562 * Hold lfs_newseg at segment 0; if requested, sleep until 1563 * the filesystem wraps around. To support external agents 1564 * (dump, fsck-based regression test) that need to look at 1565 * a snapshot of the filesystem, without necessarily 1566 * requiring that all fs activity stops. 1567 */ 1568 if (lockstatus(&fs->lfs_stoplock)) 1569 return EALREADY; 1570 1571 simple_lock(&fs->lfs_interlock); 1572 lockmgr(&fs->lfs_stoplock, LK_EXCLUSIVE, &fs->lfs_interlock); 1573 if (fs->lfs_nowrap == 0) 1574 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1575 ++fs->lfs_nowrap; 1576 if (*(int *)ap->a_data == 1 || 1577 ap->a_command == LFCNWRAPSTOP_COMPAT) { 1578 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1579 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1580 "segwrap", 0, &fs->lfs_interlock); 1581 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1582 if (error) { 1583 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1584 } 1585 } 1586 simple_unlock(&fs->lfs_interlock); 1587 return 0; 1588 1589 case LFCNWRAPGO: 1590 case LFCNWRAPGO_COMPAT: 1591 /* 1592 * Having done its work, the agent wakes up the writer. 1593 * If the argument is 1, it sleeps until a new segment 1594 * is selected. 1595 */ 1596 simple_lock(&fs->lfs_interlock); 1597 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1598 (ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1599 *((int *)ap->a_data))); 1600 simple_unlock(&fs->lfs_interlock); 1601 return error; 1602 1603 case LFCNWRAPPASS: 1604 if (lockstatus(&fs->lfs_stoplock) != LK_EXCLUSIVE) 1605 return EALREADY; 1606 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1607 return EALREADY; 1608 simple_lock(&fs->lfs_interlock); 1609 if (fs->lfs_nowrap == 0) { 1610 simple_unlock(&fs->lfs_interlock); 1611 return EBUSY; 1612 } 1613 fs->lfs_wrappass = 1; 1614 wakeup(&fs->lfs_wrappass); 1615 /* Wait for the log to wrap, if asked */ 1616 if (*(int *)ap->a_data) { 1617 lfs_vref(ap->a_vp); 1618 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1619 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1620 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1621 "segwrap", 0, &fs->lfs_interlock); 1622 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1623 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1624 lfs_vunref(ap->a_vp); 1625 } 1626 simple_unlock(&fs->lfs_interlock); 1627 return error; 1628 1629 case LFCNWRAPSTATUS: 1630 simple_lock(&fs->lfs_interlock); 1631 *(int *)ap->a_data = fs->lfs_wrapstatus; 1632 simple_unlock(&fs->lfs_interlock); 1633 return 0; 1634 1635 default: 1636 return ufs_fcntl(v); 1637 } 1638 return 0; 1639 } 1640 1641 int 1642 lfs_getpages(void *v) 1643 { 1644 struct vop_getpages_args /* { 1645 struct vnode *a_vp; 1646 voff_t a_offset; 1647 struct vm_page **a_m; 1648 int *a_count; 1649 int a_centeridx; 1650 vm_prot_t a_access_type; 1651 int a_advice; 1652 int a_flags; 1653 } */ *ap = v; 1654 1655 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1656 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1657 return EPERM; 1658 } 1659 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1660 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1661 } 1662 1663 /* 1664 * we're relying on the fact that genfs_getpages() always read in 1665 * entire filesystem blocks. 1666 */ 1667 return genfs_getpages(v); 1668 } 1669 1670 /* 1671 * Make sure that for all pages in every block in the given range, 1672 * either all are dirty or all are clean. If any of the pages 1673 * we've seen so far are dirty, put the vnode on the paging chain, 1674 * and mark it IN_PAGING. 1675 * 1676 * If checkfirst != 0, don't check all the pages but return at the 1677 * first dirty page. 1678 */ 1679 static int 1680 check_dirty(struct lfs *fs, struct vnode *vp, 1681 off_t startoffset, off_t endoffset, off_t blkeof, 1682 int flags, int checkfirst) 1683 { 1684 int by_list; 1685 struct vm_page *curpg = NULL; /* XXX: gcc */ 1686 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1687 off_t soff = 0; /* XXX: gcc */ 1688 voff_t off; 1689 int i; 1690 int nonexistent; 1691 int any_dirty; /* number of dirty pages */ 1692 int dirty; /* number of dirty pages in a block */ 1693 int tdirty; 1694 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1695 int pagedaemon = (curproc == uvm.pagedaemon_proc); 1696 1697 ASSERT_MAYBE_SEGLOCK(fs); 1698 top: 1699 by_list = (vp->v_uobj.uo_npages <= 1700 ((endoffset - startoffset) >> PAGE_SHIFT) * 1701 UVM_PAGE_HASH_PENALTY); 1702 any_dirty = 0; 1703 1704 if (by_list) { 1705 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1706 } else { 1707 soff = startoffset; 1708 } 1709 while (by_list || soff < MIN(blkeof, endoffset)) { 1710 if (by_list) { 1711 /* 1712 * Find the first page in a block. Skip 1713 * blocks outside our area of interest or beyond 1714 * the end of file. 1715 */ 1716 if (pages_per_block > 1) { 1717 while (curpg && 1718 ((curpg->offset & fs->lfs_bmask) || 1719 curpg->offset >= vp->v_size || 1720 curpg->offset >= endoffset)) 1721 curpg = TAILQ_NEXT(curpg, listq); 1722 } 1723 if (curpg == NULL) 1724 break; 1725 soff = curpg->offset; 1726 } 1727 1728 /* 1729 * Mark all pages in extended range busy; find out if any 1730 * of them are dirty. 1731 */ 1732 nonexistent = dirty = 0; 1733 for (i = 0; i == 0 || i < pages_per_block; i++) { 1734 if (by_list && pages_per_block <= 1) { 1735 pgs[i] = pg = curpg; 1736 } else { 1737 off = soff + (i << PAGE_SHIFT); 1738 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1739 if (pg == NULL) { 1740 ++nonexistent; 1741 continue; 1742 } 1743 } 1744 KASSERT(pg != NULL); 1745 1746 /* 1747 * If we're holding the segment lock, we can deadlock 1748 * against a process that has our page and is waiting 1749 * for the cleaner, while the cleaner waits for the 1750 * segment lock. Just bail in that case. 1751 */ 1752 if ((pg->flags & PG_BUSY) && 1753 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1754 if (by_list && i > 0) 1755 uvm_page_unbusy(pgs, i); 1756 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1757 return -1; 1758 } 1759 1760 while (pg->flags & PG_BUSY) { 1761 pg->flags |= PG_WANTED; 1762 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1763 "lfsput", 0); 1764 simple_lock(&vp->v_interlock); 1765 if (by_list) { 1766 if (i > 0) 1767 uvm_page_unbusy(pgs, i); 1768 goto top; 1769 } 1770 } 1771 pg->flags |= PG_BUSY; 1772 UVM_PAGE_OWN(pg, "lfs_putpages"); 1773 1774 pmap_page_protect(pg, VM_PROT_NONE); 1775 tdirty = (pmap_clear_modify(pg) || 1776 (pg->flags & PG_CLEAN) == 0); 1777 dirty += tdirty; 1778 } 1779 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1780 if (by_list) { 1781 curpg = TAILQ_NEXT(curpg, listq); 1782 } else { 1783 soff += fs->lfs_bsize; 1784 } 1785 continue; 1786 } 1787 1788 any_dirty += dirty; 1789 KASSERT(nonexistent == 0); 1790 1791 /* 1792 * If any are dirty make all dirty; unbusy them, 1793 * but if we were asked to clean, wire them so that 1794 * the pagedaemon doesn't bother us about them while 1795 * they're on their way to disk. 1796 */ 1797 for (i = 0; i == 0 || i < pages_per_block; i++) { 1798 pg = pgs[i]; 1799 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1800 if (dirty) { 1801 pg->flags &= ~PG_CLEAN; 1802 if (flags & PGO_FREE) { 1803 /* 1804 * Wire the page so that 1805 * pdaemon doesn't see it again. 1806 */ 1807 uvm_lock_pageq(); 1808 uvm_pagewire(pg); 1809 uvm_unlock_pageq(); 1810 1811 /* Suspended write flag */ 1812 pg->flags |= PG_DELWRI; 1813 } 1814 } 1815 if (pg->flags & PG_WANTED) 1816 wakeup(pg); 1817 pg->flags &= ~(PG_WANTED|PG_BUSY); 1818 UVM_PAGE_OWN(pg, NULL); 1819 } 1820 1821 if (checkfirst && any_dirty) 1822 break; 1823 1824 if (by_list) { 1825 curpg = TAILQ_NEXT(curpg, listq); 1826 } else { 1827 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1828 } 1829 } 1830 1831 return any_dirty; 1832 } 1833 1834 /* 1835 * lfs_putpages functions like genfs_putpages except that 1836 * 1837 * (1) It needs to bounds-check the incoming requests to ensure that 1838 * they are block-aligned; if they are not, expand the range and 1839 * do the right thing in case, e.g., the requested range is clean 1840 * but the expanded range is dirty. 1841 * 1842 * (2) It needs to explicitly send blocks to be written when it is done. 1843 * VOP_PUTPAGES is not ever called with the seglock held, so 1844 * we simply take the seglock and let lfs_segunlock wait for us. 1845 * XXX Actually we can be called with the seglock held, if we have 1846 * XXX to flush a vnode while lfs_markv is in operation. As of this 1847 * XXX writing we panic in this case. 1848 * 1849 * Assumptions: 1850 * 1851 * (1) The caller does not hold any pages in this vnode busy. If it does, 1852 * there is a danger that when we expand the page range and busy the 1853 * pages we will deadlock. 1854 * 1855 * (2) We are called with vp->v_interlock held; we must return with it 1856 * released. 1857 * 1858 * (3) We don't absolutely have to free pages right away, provided that 1859 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1860 * us a request with PGO_FREE, we take the pages out of the paging 1861 * queue and wake up the writer, which will handle freeing them for us. 1862 * 1863 * We ensure that for any filesystem block, all pages for that 1864 * block are either resident or not, even if those pages are higher 1865 * than EOF; that means that we will be getting requests to free 1866 * "unused" pages above EOF all the time, and should ignore them. 1867 * 1868 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 1869 * into has been set up for us by lfs_writefile. If not, we will 1870 * have to handle allocating and/or freeing an finfo entry. 1871 * 1872 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1873 */ 1874 1875 int 1876 lfs_putpages(void *v) 1877 { 1878 int error; 1879 struct vop_putpages_args /* { 1880 struct vnode *a_vp; 1881 voff_t a_offlo; 1882 voff_t a_offhi; 1883 int a_flags; 1884 } */ *ap = v; 1885 struct vnode *vp; 1886 struct inode *ip; 1887 struct lfs *fs; 1888 struct segment *sp; 1889 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1890 off_t off, max_endoffset; 1891 int s; 1892 bool seglocked, sync, pagedaemon; 1893 struct vm_page *pg; 1894 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1895 1896 vp = ap->a_vp; 1897 ip = VTOI(vp); 1898 fs = ip->i_lfs; 1899 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1900 pagedaemon = (curproc == uvm.pagedaemon_proc); 1901 1902 /* Putpages does nothing for metadata. */ 1903 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1904 simple_unlock(&vp->v_interlock); 1905 return 0; 1906 } 1907 1908 /* 1909 * If there are no pages, don't do anything. 1910 */ 1911 if (vp->v_uobj.uo_npages == 0) { 1912 s = splbio(); 1913 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 1914 (vp->v_flag & VONWORKLST) && 1915 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 1916 vp->v_flag &= ~VWRITEMAPDIRTY; 1917 vn_syncer_remove_from_worklist(vp); 1918 } 1919 splx(s); 1920 simple_unlock(&vp->v_interlock); 1921 1922 /* Remove us from paging queue, if we were on it */ 1923 simple_lock(&fs->lfs_interlock); 1924 if (ip->i_flags & IN_PAGING) { 1925 ip->i_flags &= ~IN_PAGING; 1926 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1927 } 1928 simple_unlock(&fs->lfs_interlock); 1929 return 0; 1930 } 1931 1932 blkeof = blkroundup(fs, ip->i_size); 1933 1934 /* 1935 * Ignore requests to free pages past EOF but in the same block 1936 * as EOF, unless the request is synchronous. (If the request is 1937 * sync, it comes from lfs_truncate.) 1938 * XXXUBC Make these pages look "active" so the pagedaemon won't 1939 * XXXUBC bother us with them again. 1940 */ 1941 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1942 origoffset = ap->a_offlo; 1943 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1944 pg = uvm_pagelookup(&vp->v_uobj, off); 1945 KASSERT(pg != NULL); 1946 while (pg->flags & PG_BUSY) { 1947 pg->flags |= PG_WANTED; 1948 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1949 "lfsput2", 0); 1950 simple_lock(&vp->v_interlock); 1951 } 1952 uvm_lock_pageq(); 1953 uvm_pageactivate(pg); 1954 uvm_unlock_pageq(); 1955 } 1956 ap->a_offlo = blkeof; 1957 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1958 simple_unlock(&vp->v_interlock); 1959 return 0; 1960 } 1961 } 1962 1963 /* 1964 * Extend page range to start and end at block boundaries. 1965 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1966 */ 1967 origoffset = ap->a_offlo; 1968 origendoffset = ap->a_offhi; 1969 startoffset = origoffset & ~(fs->lfs_bmask); 1970 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1971 << fs->lfs_bshift; 1972 1973 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1974 endoffset = max_endoffset; 1975 origendoffset = endoffset; 1976 } else { 1977 origendoffset = round_page(ap->a_offhi); 1978 endoffset = round_page(blkroundup(fs, origendoffset)); 1979 } 1980 1981 KASSERT(startoffset > 0 || endoffset >= startoffset); 1982 if (startoffset == endoffset) { 1983 /* Nothing to do, why were we called? */ 1984 simple_unlock(&vp->v_interlock); 1985 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 1986 PRId64 "\n", startoffset)); 1987 return 0; 1988 } 1989 1990 ap->a_offlo = startoffset; 1991 ap->a_offhi = endoffset; 1992 1993 if (!(ap->a_flags & PGO_CLEANIT)) 1994 return genfs_putpages(v); 1995 1996 /* 1997 * If there are more than one page per block, we don't want 1998 * to get caught locking them backwards; so set PGO_BUSYFAIL 1999 * to avoid deadlocks. 2000 */ 2001 ap->a_flags |= PGO_BUSYFAIL; 2002 2003 do { 2004 int r; 2005 2006 /* If no pages are dirty, we can just use genfs_putpages. */ 2007 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2008 ap->a_flags, 1); 2009 if (r < 0) { 2010 simple_unlock(&vp->v_interlock); 2011 return EDEADLK; 2012 } 2013 if (r > 0) 2014 break; 2015 2016 /* 2017 * Sometimes pages are dirtied between the time that 2018 * we check and the time we try to clean them. 2019 * Instruct lfs_gop_write to return EDEADLK in this case 2020 * so we can write them properly. 2021 */ 2022 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2023 r = genfs_putpages(v); 2024 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2025 if (r != EDEADLK) 2026 return r; 2027 2028 /* Start over. */ 2029 preempt(); 2030 simple_lock(&vp->v_interlock); 2031 } while(1); 2032 2033 /* 2034 * Dirty and asked to clean. 2035 * 2036 * Pagedaemon can't actually write LFS pages; wake up 2037 * the writer to take care of that. The writer will 2038 * notice the pager inode queue and act on that. 2039 */ 2040 if (pagedaemon) { 2041 simple_lock(&fs->lfs_interlock); 2042 if (!(ip->i_flags & IN_PAGING)) { 2043 ip->i_flags |= IN_PAGING; 2044 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2045 } 2046 simple_lock(&lfs_subsys_lock); 2047 wakeup(&lfs_writer_daemon); 2048 simple_unlock(&lfs_subsys_lock); 2049 simple_unlock(&fs->lfs_interlock); 2050 simple_unlock(&vp->v_interlock); 2051 preempt(); 2052 return EWOULDBLOCK; 2053 } 2054 2055 /* 2056 * If this is a file created in a recent dirop, we can't flush its 2057 * inode until the dirop is complete. Drain dirops, then flush the 2058 * filesystem (taking care of any other pending dirops while we're 2059 * at it). 2060 */ 2061 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2062 (vp->v_flag & VDIROP)) { 2063 int locked; 2064 2065 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 2066 locked = (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 2067 simple_unlock(&vp->v_interlock); 2068 lfs_writer_enter(fs, "ppdirop"); 2069 if (locked) 2070 VOP_UNLOCK(vp, 0); 2071 2072 simple_lock(&fs->lfs_interlock); 2073 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2074 simple_unlock(&fs->lfs_interlock); 2075 2076 simple_lock(&vp->v_interlock); 2077 if (locked) { 2078 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2079 simple_lock(&vp->v_interlock); 2080 } 2081 lfs_writer_leave(fs); 2082 2083 /* XXX the flush should have taken care of this one too! */ 2084 } 2085 2086 /* 2087 * This is it. We are going to write some pages. From here on 2088 * down it's all just mechanics. 2089 * 2090 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2091 */ 2092 ap->a_flags &= ~PGO_SYNCIO; 2093 2094 /* 2095 * If we've already got the seglock, flush the node and return. 2096 * The FIP has already been set up for us by lfs_writefile, 2097 * and FIP cleanup and lfs_updatemeta will also be done there, 2098 * unless genfs_putpages returns EDEADLK; then we must flush 2099 * what we have, and correct FIP and segment header accounting. 2100 */ 2101 get_seglock: 2102 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2103 if (!seglocked) { 2104 simple_unlock(&vp->v_interlock); 2105 /* 2106 * Take the seglock, because we are going to be writing pages. 2107 */ 2108 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2109 if (error != 0) 2110 return error; 2111 simple_lock(&vp->v_interlock); 2112 } 2113 2114 /* 2115 * VOP_PUTPAGES should not be called while holding the seglock. 2116 * XXXUBC fix lfs_markv, or do this properly. 2117 */ 2118 #ifdef notyet 2119 KASSERT(fs->lfs_seglock == 1); 2120 #endif /* notyet */ 2121 2122 /* 2123 * We assume we're being called with sp->fip pointing at blank space. 2124 * Account for a new FIP in the segment header, and set sp->vp. 2125 * (This should duplicate the setup at the top of lfs_writefile().) 2126 */ 2127 sp = fs->lfs_sp; 2128 if (!seglocked) 2129 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2130 KASSERT(sp->vp == NULL); 2131 sp->vp = vp; 2132 2133 if (!seglocked) { 2134 if (vp->v_flag & VDIROP) 2135 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2136 } 2137 2138 /* 2139 * Loop through genfs_putpages until all pages are gathered. 2140 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2141 * Whenever we lose the interlock we have to rerun check_dirty, as 2142 * well. 2143 */ 2144 again: 2145 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2146 ap->a_flags, 0) < 0) { 2147 simple_unlock(&vp->v_interlock); 2148 sp->vp = NULL; 2149 if (!seglocked) { 2150 lfs_release_finfo(fs); 2151 lfs_segunlock(fs); 2152 } 2153 if (pagedaemon) 2154 return EDEADLK; 2155 /* else seglocked == 0 */ 2156 preempt(); 2157 simple_lock(&vp->v_interlock); 2158 goto get_seglock; 2159 } 2160 2161 error = genfs_putpages(v); 2162 if (error == EDEADLK || error == EAGAIN) { 2163 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2164 " EDEADLK [2] ino %d off %x (seg %d)\n", 2165 ip->i_number, fs->lfs_offset, 2166 dtosn(fs, fs->lfs_offset))); 2167 /* If nothing to write, short-circuit */ 2168 if (sp->cbpp - sp->bpp > 1) { 2169 /* Write gathered pages */ 2170 lfs_updatemeta(sp); 2171 lfs_release_finfo(fs); 2172 (void) lfs_writeseg(fs, sp); 2173 2174 /* 2175 * Reinitialize brand new FIP and add us to it. 2176 */ 2177 KASSERT(sp->vp == vp); 2178 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2179 } 2180 2181 /* Give the write a chance to complete */ 2182 preempt(); 2183 2184 /* We've lost the interlock. Start over. */ 2185 if (error == EDEADLK) { 2186 simple_lock(&vp->v_interlock); 2187 goto again; 2188 } 2189 } 2190 2191 KASSERT(sp->vp == vp); 2192 if (!seglocked) { 2193 sp->vp = NULL; 2194 2195 /* Write indirect blocks as well */ 2196 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2197 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2198 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2199 2200 KASSERT(sp->vp == NULL); 2201 sp->vp = vp; 2202 } 2203 2204 /* 2205 * Blocks are now gathered into a segment waiting to be written. 2206 * All that's left to do is update metadata, and write them. 2207 */ 2208 lfs_updatemeta(sp); 2209 KASSERT(sp->vp == vp); 2210 sp->vp = NULL; 2211 2212 if (seglocked) { 2213 /* we're called by lfs_writefile. */ 2214 return error; 2215 } 2216 2217 /* Clean up FIP and send it to disk. */ 2218 lfs_release_finfo(fs); 2219 lfs_writeseg(fs, fs->lfs_sp); 2220 2221 /* 2222 * Remove us from paging queue, since we've now written all our 2223 * pages. 2224 */ 2225 simple_lock(&fs->lfs_interlock); 2226 if (ip->i_flags & IN_PAGING) { 2227 ip->i_flags &= ~IN_PAGING; 2228 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2229 } 2230 simple_unlock(&fs->lfs_interlock); 2231 2232 /* 2233 * XXX - with the malloc/copy writeseg, the pages are freed by now 2234 * even if we don't wait (e.g. if we hold a nested lock). This 2235 * will not be true if we stop using malloc/copy. 2236 */ 2237 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2238 lfs_segunlock(fs); 2239 2240 /* 2241 * Wait for v_numoutput to drop to zero. The seglock should 2242 * take care of this, but there is a slight possibility that 2243 * aiodoned might not have got around to our buffers yet. 2244 */ 2245 if (sync) { 2246 s = splbio(); 2247 simple_lock(&global_v_numoutput_slock); 2248 while (vp->v_numoutput > 0) { 2249 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2250 " num %d\n", ip->i_number, vp->v_numoutput)); 2251 vp->v_flag |= VBWAIT; 2252 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 2253 &global_v_numoutput_slock); 2254 } 2255 simple_unlock(&global_v_numoutput_slock); 2256 splx(s); 2257 } 2258 return error; 2259 } 2260 2261 /* 2262 * Return the last logical file offset that should be written for this file 2263 * if we're doing a write that ends at "size". If writing, we need to know 2264 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2265 * to know about entire blocks. 2266 */ 2267 void 2268 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2269 { 2270 struct inode *ip = VTOI(vp); 2271 struct lfs *fs = ip->i_lfs; 2272 daddr_t olbn, nlbn; 2273 2274 olbn = lblkno(fs, ip->i_size); 2275 nlbn = lblkno(fs, size); 2276 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2277 *eobp = fragroundup(fs, size); 2278 } else { 2279 *eobp = blkroundup(fs, size); 2280 } 2281 } 2282 2283 #ifdef DEBUG 2284 void lfs_dump_vop(void *); 2285 2286 void 2287 lfs_dump_vop(void *v) 2288 { 2289 struct vop_putpages_args /* { 2290 struct vnode *a_vp; 2291 voff_t a_offlo; 2292 voff_t a_offhi; 2293 int a_flags; 2294 } */ *ap = v; 2295 2296 #ifdef DDB 2297 vfs_vnode_print(ap->a_vp, 0, printf); 2298 #endif 2299 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2300 } 2301 #endif 2302 2303 int 2304 lfs_mmap(void *v) 2305 { 2306 struct vop_mmap_args /* { 2307 const struct vnodeop_desc *a_desc; 2308 struct vnode *a_vp; 2309 int a_fflags; 2310 kauth_cred_t a_cred; 2311 struct lwp *a_l; 2312 } */ *ap = v; 2313 2314 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2315 return EOPNOTSUPP; 2316 return ufs_mmap(v); 2317 } 2318
Indexes created Tue Sep 30 17:09:57 GMT 2025