lfs_vnops.c revision 1.195
1 /* $NetBSD: lfs_vnops.c,v 1.195 2007/01/03 02:42:23 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 1.195 2007/01/03 02:42:23 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 error = ltsleep(&fs->lfs_dirops, (PRIBIO + 1) | PCATCH, 401 "lfs_sdirop", 0, &fs->lfs_interlock); 402 if (error == EINTR) { 403 simple_unlock(&fs->lfs_interlock); 404 goto unreserve; 405 } 406 } 407 simple_lock(&lfs_subsys_lock); 408 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 409 wakeup(&lfs_writer_daemon); 410 simple_unlock(&lfs_subsys_lock); 411 simple_unlock(&fs->lfs_interlock); 412 preempt(1); 413 goto restart; 414 } 415 416 if (lfs_dirvcount > LFS_MAX_DIROP) { 417 simple_unlock(&fs->lfs_interlock); 418 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 419 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 420 if ((error = ltsleep(&lfs_dirvcount, 421 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 422 &lfs_subsys_lock)) != 0) { 423 goto unreserve; 424 } 425 goto restart; 426 } 427 simple_unlock(&lfs_subsys_lock); 428 429 ++fs->lfs_dirops; 430 fs->lfs_doifile = 1; 431 simple_unlock(&fs->lfs_interlock); 432 433 /* Hold a reference so SET_ENDOP will be happy */ 434 vref(dvp); 435 if (vp) { 436 vref(vp); 437 MARK_VNODE(vp); 438 } 439 440 MARK_VNODE(dvp); 441 return 0; 442 443 unreserve: 444 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 445 return error; 446 } 447 448 /* 449 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 450 * in getnewvnode(), if we have a stacked filesystem mounted on top 451 * of us. 452 * 453 * NB: this means we have to clear the new vnodes on error. Fortunately 454 * SET_ENDOP is there to do that for us. 455 */ 456 static int 457 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 458 { 459 int error; 460 struct lfs *fs; 461 462 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 463 ASSERT_NO_SEGLOCK(fs); 464 if (fs->lfs_ronly) 465 return EROFS; 466 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 467 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 468 dvp, error)); 469 return error; 470 } 471 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 472 if (vpp) { 473 ungetnewvnode(*vpp); 474 *vpp = NULL; 475 } 476 return error; 477 } 478 return 0; 479 } 480 481 #define SET_ENDOP_BASE(fs, dvp, str) \ 482 do { \ 483 simple_lock(&(fs)->lfs_interlock); \ 484 --(fs)->lfs_dirops; \ 485 if (!(fs)->lfs_dirops) { \ 486 if ((fs)->lfs_nadirop) { \ 487 panic("SET_ENDOP: %s: no dirops but " \ 488 " nadirop=%d", (str), \ 489 (fs)->lfs_nadirop); \ 490 } \ 491 wakeup(&(fs)->lfs_writer); \ 492 simple_unlock(&(fs)->lfs_interlock); \ 493 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 494 } else \ 495 simple_unlock(&(fs)->lfs_interlock); \ 496 } while(0) 497 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 498 do { \ 499 UNMARK_VNODE(dvp); \ 500 if (nvpp && *nvpp) \ 501 UNMARK_VNODE(*nvpp); \ 502 /* Check for error return to stem vnode leakage */ \ 503 if (nvpp && *nvpp && !((*nvpp)->v_flag & VDIROP)) \ 504 ungetnewvnode(*(nvpp)); \ 505 SET_ENDOP_BASE((fs), (dvp), (str)); \ 506 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 507 vrele(dvp); \ 508 } while(0) 509 #define SET_ENDOP_CREATE_AP(ap, str) \ 510 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 511 (ap)->a_vpp, (str)) 512 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 513 do { \ 514 UNMARK_VNODE(dvp); \ 515 if (ovp) \ 516 UNMARK_VNODE(ovp); \ 517 SET_ENDOP_BASE((fs), (dvp), (str)); \ 518 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 519 vrele(dvp); \ 520 if (ovp) \ 521 vrele(ovp); \ 522 } while(0) 523 524 void 525 lfs_mark_vnode(struct vnode *vp) 526 { 527 struct inode *ip = VTOI(vp); 528 struct lfs *fs = ip->i_lfs; 529 530 simple_lock(&fs->lfs_interlock); 531 if (!(ip->i_flag & IN_ADIROP)) { 532 if (!(vp->v_flag & VDIROP)) { 533 (void)lfs_vref(vp); 534 simple_lock(&lfs_subsys_lock); 535 ++lfs_dirvcount; 536 ++fs->lfs_dirvcount; 537 simple_unlock(&lfs_subsys_lock); 538 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 539 vp->v_flag |= VDIROP; 540 } 541 ++fs->lfs_nadirop; 542 ip->i_flag |= IN_ADIROP; 543 } else 544 KASSERT(vp->v_flag & VDIROP); 545 simple_unlock(&fs->lfs_interlock); 546 } 547 548 void 549 lfs_unmark_vnode(struct vnode *vp) 550 { 551 struct inode *ip = VTOI(vp); 552 553 if (ip && (ip->i_flag & IN_ADIROP)) { 554 KASSERT(vp->v_flag & VDIROP); 555 simple_lock(&ip->i_lfs->lfs_interlock); 556 --ip->i_lfs->lfs_nadirop; 557 simple_unlock(&ip->i_lfs->lfs_interlock); 558 ip->i_flag &= ~IN_ADIROP; 559 } 560 } 561 562 int 563 lfs_symlink(void *v) 564 { 565 struct vop_symlink_args /* { 566 struct vnode *a_dvp; 567 struct vnode **a_vpp; 568 struct componentname *a_cnp; 569 struct vattr *a_vap; 570 char *a_target; 571 } */ *ap = v; 572 int error; 573 574 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 575 vput(ap->a_dvp); 576 return error; 577 } 578 error = ufs_symlink(ap); 579 SET_ENDOP_CREATE_AP(ap, "symlink"); 580 return (error); 581 } 582 583 int 584 lfs_mknod(void *v) 585 { 586 struct vop_mknod_args /* { 587 struct vnode *a_dvp; 588 struct vnode **a_vpp; 589 struct componentname *a_cnp; 590 struct vattr *a_vap; 591 } */ *ap = v; 592 struct vattr *vap = ap->a_vap; 593 struct vnode **vpp = ap->a_vpp; 594 struct inode *ip; 595 int error; 596 struct mount *mp; 597 ino_t ino; 598 599 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 600 vput(ap->a_dvp); 601 return error; 602 } 603 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 604 ap->a_dvp, vpp, ap->a_cnp); 605 606 /* Either way we're done with the dirop at this point */ 607 SET_ENDOP_CREATE_AP(ap, "mknod"); 608 609 if (error) 610 return (error); 611 612 ip = VTOI(*vpp); 613 mp = (*vpp)->v_mount; 614 ino = ip->i_number; 615 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 616 if (vap->va_rdev != VNOVAL) { 617 /* 618 * Want to be able to use this to make badblock 619 * inodes, so don't truncate the dev number. 620 */ 621 #if 0 622 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 623 UFS_MPNEEDSWAP((*vpp)->v_mount)); 624 #else 625 ip->i_ffs1_rdev = vap->va_rdev; 626 #endif 627 } 628 629 /* 630 * Call fsync to write the vnode so that we don't have to deal with 631 * flushing it when it's marked VDIROP|VXLOCK. 632 * 633 * XXX KS - If we can't flush we also can't call vgone(), so must 634 * return. But, that leaves this vnode in limbo, also not good. 635 * Can this ever happen (barring hardware failure)? 636 */ 637 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 638 curlwp)) != 0) { 639 panic("lfs_mknod: couldn't fsync (ino %llu)", 640 (unsigned long long)ino); 641 /* return (error); */ 642 } 643 /* 644 * Remove vnode so that it will be reloaded by VFS_VGET and 645 * checked to see if it is an alias of an existing entry in 646 * the inode cache. 647 */ 648 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 649 650 VOP_UNLOCK(*vpp, 0); 651 lfs_vunref(*vpp); 652 (*vpp)->v_type = VNON; 653 vgone(*vpp); 654 error = VFS_VGET(mp, ino, vpp); 655 656 if (error != 0) { 657 *vpp = NULL; 658 return (error); 659 } 660 return (0); 661 } 662 663 int 664 lfs_create(void *v) 665 { 666 struct vop_create_args /* { 667 struct vnode *a_dvp; 668 struct vnode **a_vpp; 669 struct componentname *a_cnp; 670 struct vattr *a_vap; 671 } */ *ap = v; 672 int error; 673 674 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 675 vput(ap->a_dvp); 676 return error; 677 } 678 error = ufs_create(ap); 679 SET_ENDOP_CREATE_AP(ap, "create"); 680 return (error); 681 } 682 683 int 684 lfs_mkdir(void *v) 685 { 686 struct vop_mkdir_args /* { 687 struct vnode *a_dvp; 688 struct vnode **a_vpp; 689 struct componentname *a_cnp; 690 struct vattr *a_vap; 691 } */ *ap = v; 692 int error; 693 694 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 695 vput(ap->a_dvp); 696 return error; 697 } 698 error = ufs_mkdir(ap); 699 SET_ENDOP_CREATE_AP(ap, "mkdir"); 700 return (error); 701 } 702 703 int 704 lfs_remove(void *v) 705 { 706 struct vop_remove_args /* { 707 struct vnode *a_dvp; 708 struct vnode *a_vp; 709 struct componentname *a_cnp; 710 } */ *ap = v; 711 struct vnode *dvp, *vp; 712 struct inode *ip; 713 int error; 714 715 dvp = ap->a_dvp; 716 vp = ap->a_vp; 717 ip = VTOI(vp); 718 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 719 if (dvp == vp) 720 vrele(vp); 721 else 722 vput(vp); 723 vput(dvp); 724 return error; 725 } 726 error = ufs_remove(ap); 727 if (ip->i_nlink == 0) 728 lfs_orphan(ip->i_lfs, ip->i_number); 729 SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove"); 730 return (error); 731 } 732 733 int 734 lfs_rmdir(void *v) 735 { 736 struct vop_rmdir_args /* { 737 struct vnodeop_desc *a_desc; 738 struct vnode *a_dvp; 739 struct vnode *a_vp; 740 struct componentname *a_cnp; 741 } */ *ap = v; 742 struct vnode *vp; 743 struct inode *ip; 744 int error; 745 746 vp = ap->a_vp; 747 ip = VTOI(vp); 748 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 749 if (ap->a_dvp == vp) 750 vrele(ap->a_dvp); 751 else 752 vput(ap->a_dvp); 753 vput(vp); 754 return error; 755 } 756 error = ufs_rmdir(ap); 757 if (ip->i_nlink == 0) 758 lfs_orphan(ip->i_lfs, ip->i_number); 759 SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 760 return (error); 761 } 762 763 int 764 lfs_link(void *v) 765 { 766 struct vop_link_args /* { 767 struct vnode *a_dvp; 768 struct vnode *a_vp; 769 struct componentname *a_cnp; 770 } */ *ap = v; 771 int error; 772 struct vnode **vpp = NULL; 773 774 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 775 vput(ap->a_dvp); 776 return error; 777 } 778 error = ufs_link(ap); 779 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 780 return (error); 781 } 782 783 int 784 lfs_rename(void *v) 785 { 786 struct vop_rename_args /* { 787 struct vnode *a_fdvp; 788 struct vnode *a_fvp; 789 struct componentname *a_fcnp; 790 struct vnode *a_tdvp; 791 struct vnode *a_tvp; 792 struct componentname *a_tcnp; 793 } */ *ap = v; 794 struct vnode *tvp, *fvp, *tdvp, *fdvp; 795 struct componentname *tcnp, *fcnp; 796 int error; 797 struct lfs *fs; 798 799 fs = VTOI(ap->a_fdvp)->i_lfs; 800 tvp = ap->a_tvp; 801 tdvp = ap->a_tdvp; 802 tcnp = ap->a_tcnp; 803 fvp = ap->a_fvp; 804 fdvp = ap->a_fdvp; 805 fcnp = ap->a_fcnp; 806 807 /* 808 * Check for cross-device rename. 809 * If it is, we don't want to set dirops, just error out. 810 * (In particular note that MARK_VNODE(tdvp) will DTWT on 811 * a cross-device rename.) 812 * 813 * Copied from ufs_rename. 814 */ 815 if ((fvp->v_mount != tdvp->v_mount) || 816 (tvp && (fvp->v_mount != tvp->v_mount))) { 817 error = EXDEV; 818 goto errout; 819 } 820 821 /* 822 * Check to make sure we're not renaming a vnode onto itself 823 * (deleting a hard link by renaming one name onto another); 824 * if we are we can't recursively call VOP_REMOVE since that 825 * would leave us with an unaccounted-for number of live dirops. 826 * 827 * Inline the relevant section of ufs_rename here, *before* 828 * calling SET_DIROP_REMOVE. 829 */ 830 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 831 (VTOI(tdvp)->i_flags & APPEND))) { 832 error = EPERM; 833 goto errout; 834 } 835 if (fvp == tvp) { 836 if (fvp->v_type == VDIR) { 837 error = EINVAL; 838 goto errout; 839 } 840 841 /* Release destination completely. */ 842 VOP_ABORTOP(tdvp, tcnp); 843 vput(tdvp); 844 vput(tvp); 845 846 /* Delete source. */ 847 vrele(fvp); 848 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 849 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 850 fcnp->cn_nameiop = DELETE; 851 vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY); 852 if ((error = relookup(fdvp, &fvp, fcnp))) { 853 vput(fdvp); 854 return (error); 855 } 856 return (VOP_REMOVE(fdvp, fvp, fcnp)); 857 } 858 859 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 860 goto errout; 861 MARK_VNODE(fdvp); 862 MARK_VNODE(fvp); 863 864 error = ufs_rename(ap); 865 UNMARK_VNODE(fdvp); 866 UNMARK_VNODE(fvp); 867 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 868 return (error); 869 870 errout: 871 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 872 if (tdvp == tvp) 873 vrele(tdvp); 874 else 875 vput(tdvp); 876 if (tvp) 877 vput(tvp); 878 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 879 vrele(fdvp); 880 vrele(fvp); 881 return (error); 882 } 883 884 /* XXX hack to avoid calling ITIMES in getattr */ 885 int 886 lfs_getattr(void *v) 887 { 888 struct vop_getattr_args /* { 889 struct vnode *a_vp; 890 struct vattr *a_vap; 891 kauth_cred_t a_cred; 892 struct lwp *a_l; 893 } */ *ap = v; 894 struct vnode *vp = ap->a_vp; 895 struct inode *ip = VTOI(vp); 896 struct vattr *vap = ap->a_vap; 897 struct lfs *fs = ip->i_lfs; 898 /* 899 * Copy from inode table 900 */ 901 vap->va_fsid = ip->i_dev; 902 vap->va_fileid = ip->i_number; 903 vap->va_mode = ip->i_mode & ~IFMT; 904 vap->va_nlink = ip->i_nlink; 905 vap->va_uid = ip->i_uid; 906 vap->va_gid = ip->i_gid; 907 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 908 vap->va_size = vp->v_size; 909 vap->va_atime.tv_sec = ip->i_ffs1_atime; 910 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 911 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 912 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 913 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 914 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 915 vap->va_flags = ip->i_flags; 916 vap->va_gen = ip->i_gen; 917 /* this doesn't belong here */ 918 if (vp->v_type == VBLK) 919 vap->va_blocksize = BLKDEV_IOSIZE; 920 else if (vp->v_type == VCHR) 921 vap->va_blocksize = MAXBSIZE; 922 else 923 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 924 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 925 vap->va_type = vp->v_type; 926 vap->va_filerev = ip->i_modrev; 927 return (0); 928 } 929 930 /* 931 * Check to make sure the inode blocks won't choke the buffer 932 * cache, then call ufs_setattr as usual. 933 */ 934 int 935 lfs_setattr(void *v) 936 { 937 struct vop_setattr_args /* { 938 struct vnode *a_vp; 939 struct vattr *a_vap; 940 kauth_cred_t a_cred; 941 struct lwp *a_l; 942 } */ *ap = v; 943 struct vnode *vp = ap->a_vp; 944 945 lfs_check(vp, LFS_UNUSED_LBN, 0); 946 return ufs_setattr(v); 947 } 948 949 /* 950 * Release the block we hold on lfs_newseg wrapping. Called on file close, 951 * or explicitly from LFCNWRAPGO. Called with the interlock held. 952 */ 953 static int 954 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor) 955 { 956 if (lockstatus(&fs->lfs_stoplock) != LK_EXCLUSIVE) 957 return EBUSY; 958 959 lockmgr(&fs->lfs_stoplock, LK_RELEASE, &fs->lfs_interlock); 960 961 KASSERT(fs->lfs_nowrap > 0); 962 if (fs->lfs_nowrap <= 0) { 963 return 0; 964 } 965 966 if (--fs->lfs_nowrap == 0) { 967 log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt); 968 wakeup(&fs->lfs_wrappass); 969 lfs_wakeup_cleaner(fs); 970 } 971 if (waitfor) { 972 ltsleep(&fs->lfs_nextseg, PCATCH | PUSER, 973 "segment", 0, &fs->lfs_interlock); 974 } 975 976 return 0; 977 } 978 979 /* 980 * Close called 981 */ 982 /* ARGSUSED */ 983 int 984 lfs_close(void *v) 985 { 986 struct vop_close_args /* { 987 struct vnode *a_vp; 988 int a_fflag; 989 kauth_cred_t a_cred; 990 struct lwp *a_l; 991 } */ *ap = v; 992 struct vnode *vp = ap->a_vp; 993 struct inode *ip = VTOI(vp); 994 struct lfs *fs = ip->i_lfs; 995 996 if ((ip->i_number == ROOTINO || ip->i_number == LFS_IFILE_INUM) && 997 lockstatus(&fs->lfs_stoplock) == LK_EXCLUSIVE) { 998 simple_lock(&fs->lfs_interlock); 999 log(LOG_NOTICE, "lfs_close: releasing log wrap control\n"); 1000 lfs_wrapgo(fs, ip, 0); 1001 simple_unlock(&fs->lfs_interlock); 1002 } 1003 1004 if (vp == ip->i_lfs->lfs_ivnode && 1005 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 1006 return 0; 1007 1008 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 1009 LFS_ITIMES(ip, NULL, NULL, NULL); 1010 } 1011 return (0); 1012 } 1013 1014 /* 1015 * Close wrapper for special devices. 1016 * 1017 * Update the times on the inode then do device close. 1018 */ 1019 int 1020 lfsspec_close(void *v) 1021 { 1022 struct vop_close_args /* { 1023 struct vnode *a_vp; 1024 int a_fflag; 1025 kauth_cred_t a_cred; 1026 struct lwp *a_l; 1027 } */ *ap = v; 1028 struct vnode *vp; 1029 struct inode *ip; 1030 1031 vp = ap->a_vp; 1032 ip = VTOI(vp); 1033 if (vp->v_usecount > 1) { 1034 LFS_ITIMES(ip, NULL, NULL, NULL); 1035 } 1036 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1037 } 1038 1039 /* 1040 * Close wrapper for fifo's. 1041 * 1042 * Update the times on the inode then do device close. 1043 */ 1044 int 1045 lfsfifo_close(void *v) 1046 { 1047 struct vop_close_args /* { 1048 struct vnode *a_vp; 1049 int a_fflag; 1050 kauth_cred_ a_cred; 1051 struct lwp *a_l; 1052 } */ *ap = v; 1053 struct vnode *vp; 1054 struct inode *ip; 1055 1056 vp = ap->a_vp; 1057 ip = VTOI(vp); 1058 if (ap->a_vp->v_usecount > 1) { 1059 LFS_ITIMES(ip, NULL, NULL, NULL); 1060 } 1061 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1062 } 1063 1064 /* 1065 * Reclaim an inode so that it can be used for other purposes. 1066 */ 1067 1068 int 1069 lfs_reclaim(void *v) 1070 { 1071 struct vop_reclaim_args /* { 1072 struct vnode *a_vp; 1073 struct lwp *a_l; 1074 } */ *ap = v; 1075 struct vnode *vp = ap->a_vp; 1076 struct inode *ip = VTOI(vp); 1077 int error; 1078 1079 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1080 1081 LFS_CLR_UINO(ip, IN_ALLMOD); 1082 if ((error = ufs_reclaim(vp, ap->a_l))) 1083 return (error); 1084 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1085 lfs_deregister_all(vp); 1086 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1087 ip->inode_ext.lfs = NULL; 1088 pool_put(&lfs_inode_pool, vp->v_data); 1089 vp->v_data = NULL; 1090 return (0); 1091 } 1092 1093 /* 1094 * Read a block from a storage device. 1095 * In order to avoid reading blocks that are in the process of being 1096 * written by the cleaner---and hence are not mutexed by the normal 1097 * buffer cache / page cache mechanisms---check for collisions before 1098 * reading. 1099 * 1100 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1101 * the active cleaner test. 1102 * 1103 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1104 */ 1105 int 1106 lfs_strategy(void *v) 1107 { 1108 struct vop_strategy_args /* { 1109 struct vnode *a_vp; 1110 struct buf *a_bp; 1111 } */ *ap = v; 1112 struct buf *bp; 1113 struct lfs *fs; 1114 struct vnode *vp; 1115 struct inode *ip; 1116 daddr_t tbn; 1117 int i, sn, error, slept; 1118 1119 bp = ap->a_bp; 1120 vp = ap->a_vp; 1121 ip = VTOI(vp); 1122 fs = ip->i_lfs; 1123 1124 /* lfs uses its strategy routine only for read */ 1125 KASSERT(bp->b_flags & B_READ); 1126 1127 if (vp->v_type == VBLK || vp->v_type == VCHR) 1128 panic("lfs_strategy: spec"); 1129 KASSERT(bp->b_bcount != 0); 1130 if (bp->b_blkno == bp->b_lblkno) { 1131 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1132 NULL); 1133 if (error) { 1134 bp->b_error = error; 1135 bp->b_flags |= B_ERROR; 1136 biodone(bp); 1137 return (error); 1138 } 1139 if ((long)bp->b_blkno == -1) /* no valid data */ 1140 clrbuf(bp); 1141 } 1142 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1143 biodone(bp); 1144 return (0); 1145 } 1146 1147 slept = 1; 1148 simple_lock(&fs->lfs_interlock); 1149 while (slept && fs->lfs_seglock) { 1150 simple_unlock(&fs->lfs_interlock); 1151 /* 1152 * Look through list of intervals. 1153 * There will only be intervals to look through 1154 * if the cleaner holds the seglock. 1155 * Since the cleaner is synchronous, we can trust 1156 * the list of intervals to be current. 1157 */ 1158 tbn = dbtofsb(fs, bp->b_blkno); 1159 sn = dtosn(fs, tbn); 1160 slept = 0; 1161 for (i = 0; i < fs->lfs_cleanind; i++) { 1162 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1163 tbn >= fs->lfs_cleanint[i]) { 1164 DLOG((DLOG_CLEAN, 1165 "lfs_strategy: ino %d lbn %" PRId64 1166 " ind %d sn %d fsb %" PRIx32 1167 " given sn %d fsb %" PRIx64 "\n", 1168 ip->i_number, bp->b_lblkno, i, 1169 dtosn(fs, fs->lfs_cleanint[i]), 1170 fs->lfs_cleanint[i], sn, tbn)); 1171 DLOG((DLOG_CLEAN, 1172 "lfs_strategy: sleeping on ino %d lbn %" 1173 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1174 simple_lock(&fs->lfs_interlock); 1175 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1176 /* Cleaner can't wait for itself */ 1177 ltsleep(&fs->lfs_iocount, 1178 (PRIBIO + 1) | PNORELOCK, 1179 "clean2", 0, 1180 &fs->lfs_interlock); 1181 slept = 1; 1182 break; 1183 } else if (fs->lfs_seglock) { 1184 ltsleep(&fs->lfs_seglock, 1185 (PRIBIO + 1) | PNORELOCK, 1186 "clean1", 0, 1187 &fs->lfs_interlock); 1188 slept = 1; 1189 break; 1190 } 1191 simple_unlock(&fs->lfs_interlock); 1192 } 1193 } 1194 simple_lock(&fs->lfs_interlock); 1195 } 1196 simple_unlock(&fs->lfs_interlock); 1197 1198 vp = ip->i_devvp; 1199 VOP_STRATEGY(vp, bp); 1200 return (0); 1201 } 1202 1203 void 1204 lfs_flush_dirops(struct lfs *fs) 1205 { 1206 struct inode *ip, *nip; 1207 struct vnode *vp; 1208 extern int lfs_dostats; 1209 struct segment *sp; 1210 int waslocked; 1211 1212 ASSERT_MAYBE_SEGLOCK(fs); 1213 KASSERT(fs->lfs_nadirop == 0); 1214 1215 if (fs->lfs_ronly) 1216 return; 1217 1218 simple_lock(&fs->lfs_interlock); 1219 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1220 simple_unlock(&fs->lfs_interlock); 1221 return; 1222 } else 1223 simple_unlock(&fs->lfs_interlock); 1224 1225 if (lfs_dostats) 1226 ++lfs_stats.flush_invoked; 1227 1228 /* 1229 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1230 * Technically this is a checkpoint (the on-disk state is valid) 1231 * even though we are leaving out all the file data. 1232 */ 1233 lfs_imtime(fs); 1234 lfs_seglock(fs, SEGM_CKP); 1235 sp = fs->lfs_sp; 1236 1237 /* 1238 * lfs_writevnodes, optimized to get dirops out of the way. 1239 * Only write dirops, and don't flush files' pages, only 1240 * blocks from the directories. 1241 * 1242 * We don't need to vref these files because they are 1243 * dirops and so hold an extra reference until the 1244 * segunlock clears them of that status. 1245 * 1246 * We don't need to check for IN_ADIROP because we know that 1247 * no dirops are active. 1248 * 1249 */ 1250 simple_lock(&fs->lfs_interlock); 1251 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1252 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1253 simple_unlock(&fs->lfs_interlock); 1254 vp = ITOV(ip); 1255 1256 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1257 1258 /* 1259 * All writes to directories come from dirops; all 1260 * writes to files' direct blocks go through the page 1261 * cache, which we're not touching. Reads to files 1262 * and/or directories will not be affected by writing 1263 * directory blocks inodes and file inodes. So we don't 1264 * really need to lock. If we don't lock, though, 1265 * make sure that we don't clear IN_MODIFIED 1266 * unnecessarily. 1267 */ 1268 if (vp->v_flag & (VXLOCK | VFREEING)) { 1269 simple_lock(&fs->lfs_interlock); 1270 continue; 1271 } 1272 waslocked = VOP_ISLOCKED(vp); 1273 if (vp->v_type != VREG && 1274 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1275 lfs_writefile(fs, sp, vp); 1276 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1277 !(ip->i_flag & IN_ALLMOD)) { 1278 LFS_SET_UINO(ip, IN_MODIFIED); 1279 } 1280 } 1281 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1282 (void) lfs_writeinode(fs, sp, ip); 1283 if (waslocked == LK_EXCLOTHER) 1284 LFS_SET_UINO(ip, IN_MODIFIED); 1285 simple_lock(&fs->lfs_interlock); 1286 } 1287 simple_unlock(&fs->lfs_interlock); 1288 /* We've written all the dirops there are */ 1289 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1290 lfs_finalize_fs_seguse(fs); 1291 (void) lfs_writeseg(fs, sp); 1292 lfs_segunlock(fs); 1293 } 1294 1295 /* 1296 * Flush all vnodes for which the pagedaemon has requested pageouts. 1297 * Skip over any files that are marked VDIROP (since lfs_flush_dirop() 1298 * has just run, this would be an error). If we have to skip a vnode 1299 * for any reason, just skip it; if we have to wait for the cleaner, 1300 * abort. The writer daemon will call us again later. 1301 */ 1302 void 1303 lfs_flush_pchain(struct lfs *fs) 1304 { 1305 struct inode *ip, *nip; 1306 struct vnode *vp; 1307 extern int lfs_dostats; 1308 struct segment *sp; 1309 int error; 1310 1311 ASSERT_NO_SEGLOCK(fs); 1312 1313 if (fs->lfs_ronly) 1314 return; 1315 1316 simple_lock(&fs->lfs_interlock); 1317 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1318 simple_unlock(&fs->lfs_interlock); 1319 return; 1320 } else 1321 simple_unlock(&fs->lfs_interlock); 1322 1323 /* Get dirops out of the way */ 1324 lfs_flush_dirops(fs); 1325 1326 if (lfs_dostats) 1327 ++lfs_stats.flush_invoked; 1328 1329 /* 1330 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1331 */ 1332 lfs_imtime(fs); 1333 lfs_seglock(fs, 0); 1334 sp = fs->lfs_sp; 1335 1336 /* 1337 * lfs_writevnodes, optimized to clear pageout requests. 1338 * Only write non-dirop files that are in the pageout queue. 1339 * We're very conservative about what we write; we want to be 1340 * fast and async. 1341 */ 1342 simple_lock(&fs->lfs_interlock); 1343 top: 1344 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1345 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1346 vp = ITOV(ip); 1347 1348 if (!(ip->i_flags & IN_PAGING)) 1349 goto top; 1350 1351 if (vp->v_flag & (VXLOCK|VDIROP)) 1352 continue; 1353 if (vp->v_type != VREG) 1354 continue; 1355 if (lfs_vref(vp)) 1356 continue; 1357 simple_unlock(&fs->lfs_interlock); 1358 1359 if (VOP_ISLOCKED(vp)) { 1360 lfs_vunref(vp); 1361 simple_lock(&fs->lfs_interlock); 1362 continue; 1363 } 1364 1365 error = lfs_writefile(fs, sp, vp); 1366 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1367 !(ip->i_flag & IN_ALLMOD)) { 1368 LFS_SET_UINO(ip, IN_MODIFIED); 1369 } 1370 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1371 (void) lfs_writeinode(fs, sp, ip); 1372 1373 lfs_vunref(vp); 1374 1375 if (error == EAGAIN) { 1376 lfs_writeseg(fs, sp); 1377 simple_lock(&fs->lfs_interlock); 1378 break; 1379 } 1380 simple_lock(&fs->lfs_interlock); 1381 } 1382 simple_unlock(&fs->lfs_interlock); 1383 (void) lfs_writeseg(fs, sp); 1384 lfs_segunlock(fs); 1385 } 1386 1387 /* 1388 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1389 */ 1390 int 1391 lfs_fcntl(void *v) 1392 { 1393 struct vop_fcntl_args /* { 1394 struct vnode *a_vp; 1395 u_long a_command; 1396 caddr_t a_data; 1397 int a_fflag; 1398 kauth_cred_t a_cred; 1399 struct lwp *a_l; 1400 } */ *ap = v; 1401 struct timeval *tvp; 1402 BLOCK_INFO *blkiov; 1403 CLEANERINFO *cip; 1404 SEGUSE *sup; 1405 int blkcnt, error, oclean; 1406 size_t fh_size; 1407 struct lfs_fcntl_markv blkvp; 1408 struct lwp *l; 1409 fsid_t *fsidp; 1410 struct lfs *fs; 1411 struct buf *bp; 1412 fhandle_t *fhp; 1413 daddr_t off; 1414 1415 /* Only respect LFS fcntls on fs root or Ifile */ 1416 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1417 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1418 return ufs_fcntl(v); 1419 } 1420 1421 /* Avoid locking a draining lock */ 1422 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1423 return ESHUTDOWN; 1424 } 1425 1426 /* LFS control and monitoring fcntls are available only to root */ 1427 l = ap->a_l; 1428 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1429 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1430 &l->l_acflag)) != 0) 1431 return (error); 1432 1433 fs = VTOI(ap->a_vp)->i_lfs; 1434 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1435 1436 error = 0; 1437 switch (ap->a_command) { 1438 case LFCNSEGWAITALL: 1439 case LFCNSEGWAITALL_COMPAT: 1440 fsidp = NULL; 1441 /* FALLSTHROUGH */ 1442 case LFCNSEGWAIT: 1443 case LFCNSEGWAIT_COMPAT: 1444 tvp = (struct timeval *)ap->a_data; 1445 simple_lock(&fs->lfs_interlock); 1446 ++fs->lfs_sleepers; 1447 simple_unlock(&fs->lfs_interlock); 1448 1449 error = lfs_segwait(fsidp, tvp); 1450 1451 simple_lock(&fs->lfs_interlock); 1452 if (--fs->lfs_sleepers == 0) 1453 wakeup(&fs->lfs_sleepers); 1454 simple_unlock(&fs->lfs_interlock); 1455 return error; 1456 1457 case LFCNBMAPV: 1458 case LFCNMARKV: 1459 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1460 1461 blkcnt = blkvp.blkcnt; 1462 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1463 return (EINVAL); 1464 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1465 if ((error = copyin(blkvp.blkiov, blkiov, 1466 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1467 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1468 return error; 1469 } 1470 1471 simple_lock(&fs->lfs_interlock); 1472 ++fs->lfs_sleepers; 1473 simple_unlock(&fs->lfs_interlock); 1474 if (ap->a_command == LFCNBMAPV) 1475 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1476 else /* LFCNMARKV */ 1477 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1478 if (error == 0) 1479 error = copyout(blkiov, blkvp.blkiov, 1480 blkcnt * sizeof(BLOCK_INFO)); 1481 simple_lock(&fs->lfs_interlock); 1482 if (--fs->lfs_sleepers == 0) 1483 wakeup(&fs->lfs_sleepers); 1484 simple_unlock(&fs->lfs_interlock); 1485 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1486 return error; 1487 1488 case LFCNRECLAIM: 1489 /* 1490 * Flush dirops and write Ifile, allowing empty segments 1491 * to be immediately reclaimed. 1492 */ 1493 lfs_writer_enter(fs, "pndirop"); 1494 off = fs->lfs_offset; 1495 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1496 lfs_flush_dirops(fs); 1497 LFS_CLEANERINFO(cip, fs, bp); 1498 oclean = cip->clean; 1499 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1500 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1501 fs->lfs_sp->seg_flags |= SEGM_PROT; 1502 lfs_segunlock(fs); 1503 lfs_writer_leave(fs); 1504 1505 #ifdef DEBUG 1506 LFS_CLEANERINFO(cip, fs, bp); 1507 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1508 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1509 fs->lfs_offset - off, cip->clean - oclean, 1510 fs->lfs_activesb)); 1511 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1512 #endif 1513 1514 return 0; 1515 1516 #ifdef COMPAT_30 1517 case LFCNIFILEFH_COMPAT: 1518 /* Return the filehandle of the Ifile */ 1519 if ((error = kauth_authorize_generic(l->l_cred, 1520 KAUTH_GENERIC_ISSUSER, &l->l_acflag)) != 0) 1521 return (error); 1522 fhp = (struct fhandle *)ap->a_data; 1523 fhp->fh_fsid = *fsidp; 1524 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1525 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1526 #endif 1527 1528 case LFCNIFILEFH_COMPAT2: 1529 case LFCNIFILEFH: 1530 /* Return the filehandle of the Ifile */ 1531 fhp = (struct fhandle *)ap->a_data; 1532 fhp->fh_fsid = *fsidp; 1533 fh_size = sizeof(struct lfs_fhandle) - 1534 offsetof(fhandle_t, fh_fid); 1535 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1536 1537 case LFCNREWIND: 1538 /* Move lfs_offset to the lowest-numbered segment */ 1539 return lfs_rewind(fs, *(int *)ap->a_data); 1540 1541 case LFCNINVAL: 1542 /* Mark a segment SEGUSE_INVAL */ 1543 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1544 if (sup->su_nbytes > 0) { 1545 brelse(bp); 1546 lfs_unset_inval_all(fs); 1547 return EBUSY; 1548 } 1549 sup->su_flags |= SEGUSE_INVAL; 1550 VOP_BWRITE(bp); 1551 return 0; 1552 1553 case LFCNRESIZE: 1554 /* Resize the filesystem */ 1555 return lfs_resize_fs(fs, *(int *)ap->a_data); 1556 1557 case LFCNWRAPSTOP: 1558 case LFCNWRAPSTOP_COMPAT: 1559 /* 1560 * Hold lfs_newseg at segment 0; if requested, sleep until 1561 * the filesystem wraps around. To support external agents 1562 * (dump, fsck-based regression test) that need to look at 1563 * a snapshot of the filesystem, without necessarily 1564 * requiring that all fs activity stops. 1565 */ 1566 if (lockstatus(&fs->lfs_stoplock)) 1567 return EALREADY; 1568 1569 simple_lock(&fs->lfs_interlock); 1570 lockmgr(&fs->lfs_stoplock, LK_EXCLUSIVE, &fs->lfs_interlock); 1571 if (fs->lfs_nowrap == 0) 1572 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1573 ++fs->lfs_nowrap; 1574 if (*(int *)ap->a_data == 1 || 1575 ap->a_command == LFCNWRAPSTOP_COMPAT) { 1576 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1577 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1578 "segwrap", 0, &fs->lfs_interlock); 1579 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1580 if (error) { 1581 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1582 } 1583 } 1584 simple_unlock(&fs->lfs_interlock); 1585 return 0; 1586 1587 case LFCNWRAPGO: 1588 case LFCNWRAPGO_COMPAT: 1589 /* 1590 * Having done its work, the agent wakes up the writer. 1591 * If the argument is 1, it sleeps until a new segment 1592 * is selected. 1593 */ 1594 simple_lock(&fs->lfs_interlock); 1595 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1596 (ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1597 *((int *)ap->a_data))); 1598 simple_unlock(&fs->lfs_interlock); 1599 return error; 1600 1601 case LFCNWRAPPASS: 1602 if (lockstatus(&fs->lfs_stoplock) != LK_EXCLUSIVE) 1603 return EALREADY; 1604 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1605 return EALREADY; 1606 simple_lock(&fs->lfs_interlock); 1607 if (fs->lfs_nowrap == 0) { 1608 simple_unlock(&fs->lfs_interlock); 1609 return EBUSY; 1610 } 1611 fs->lfs_wrappass = 1; 1612 wakeup(&fs->lfs_wrappass); 1613 /* Wait for the log to wrap, if asked */ 1614 if (*(int *)ap->a_data) { 1615 lfs_vref(ap->a_vp); 1616 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1617 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1618 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1619 "segwrap", 0, &fs->lfs_interlock); 1620 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1621 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1622 lfs_vunref(ap->a_vp); 1623 } 1624 simple_unlock(&fs->lfs_interlock); 1625 return error; 1626 1627 case LFCNWRAPSTATUS: 1628 simple_lock(&fs->lfs_interlock); 1629 *(int *)ap->a_data = fs->lfs_wrapstatus; 1630 simple_unlock(&fs->lfs_interlock); 1631 return 0; 1632 1633 default: 1634 return ufs_fcntl(v); 1635 } 1636 return 0; 1637 } 1638 1639 int 1640 lfs_getpages(void *v) 1641 { 1642 struct vop_getpages_args /* { 1643 struct vnode *a_vp; 1644 voff_t a_offset; 1645 struct vm_page **a_m; 1646 int *a_count; 1647 int a_centeridx; 1648 vm_prot_t a_access_type; 1649 int a_advice; 1650 int a_flags; 1651 } */ *ap = v; 1652 1653 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1654 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1655 return EPERM; 1656 } 1657 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1658 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1659 } 1660 1661 /* 1662 * we're relying on the fact that genfs_getpages() always read in 1663 * entire filesystem blocks. 1664 */ 1665 return genfs_getpages(v); 1666 } 1667 1668 /* 1669 * Make sure that for all pages in every block in the given range, 1670 * either all are dirty or all are clean. If any of the pages 1671 * we've seen so far are dirty, put the vnode on the paging chain, 1672 * and mark it IN_PAGING. 1673 * 1674 * If checkfirst != 0, don't check all the pages but return at the 1675 * first dirty page. 1676 */ 1677 static int 1678 check_dirty(struct lfs *fs, struct vnode *vp, 1679 off_t startoffset, off_t endoffset, off_t blkeof, 1680 int flags, int checkfirst) 1681 { 1682 int by_list; 1683 struct vm_page *curpg = NULL; /* XXX: gcc */ 1684 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1685 off_t soff = 0; /* XXX: gcc */ 1686 voff_t off; 1687 int i; 1688 int nonexistent; 1689 int any_dirty; /* number of dirty pages */ 1690 int dirty; /* number of dirty pages in a block */ 1691 int tdirty; 1692 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1693 int pagedaemon = (curproc == uvm.pagedaemon_proc); 1694 1695 ASSERT_MAYBE_SEGLOCK(fs); 1696 top: 1697 by_list = (vp->v_uobj.uo_npages <= 1698 ((endoffset - startoffset) >> PAGE_SHIFT) * 1699 UVM_PAGE_HASH_PENALTY); 1700 any_dirty = 0; 1701 1702 if (by_list) { 1703 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1704 } else { 1705 soff = startoffset; 1706 } 1707 while (by_list || soff < MIN(blkeof, endoffset)) { 1708 if (by_list) { 1709 /* 1710 * Find the first page in a block. Skip 1711 * blocks outside our area of interest or beyond 1712 * the end of file. 1713 */ 1714 if (pages_per_block > 1) { 1715 while (curpg && 1716 ((curpg->offset & fs->lfs_bmask) || 1717 curpg->offset >= vp->v_size || 1718 curpg->offset >= endoffset)) 1719 curpg = TAILQ_NEXT(curpg, listq); 1720 } 1721 if (curpg == NULL) 1722 break; 1723 soff = curpg->offset; 1724 } 1725 1726 /* 1727 * Mark all pages in extended range busy; find out if any 1728 * of them are dirty. 1729 */ 1730 nonexistent = dirty = 0; 1731 for (i = 0; i == 0 || i < pages_per_block; i++) { 1732 if (by_list && pages_per_block <= 1) { 1733 pgs[i] = pg = curpg; 1734 } else { 1735 off = soff + (i << PAGE_SHIFT); 1736 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1737 if (pg == NULL) { 1738 ++nonexistent; 1739 continue; 1740 } 1741 } 1742 KASSERT(pg != NULL); 1743 1744 /* 1745 * If we're holding the segment lock, we can deadlock 1746 * against a process that has our page and is waiting 1747 * for the cleaner, while the cleaner waits for the 1748 * segment lock. Just bail in that case. 1749 */ 1750 if ((pg->flags & PG_BUSY) && 1751 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1752 if (by_list && i > 0) 1753 uvm_page_unbusy(pgs, i); 1754 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1755 return -1; 1756 } 1757 1758 while (pg->flags & PG_BUSY) { 1759 pg->flags |= PG_WANTED; 1760 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1761 "lfsput", 0); 1762 simple_lock(&vp->v_interlock); 1763 if (by_list) { 1764 if (i > 0) 1765 uvm_page_unbusy(pgs, i); 1766 goto top; 1767 } 1768 } 1769 pg->flags |= PG_BUSY; 1770 UVM_PAGE_OWN(pg, "lfs_putpages"); 1771 1772 pmap_page_protect(pg, VM_PROT_NONE); 1773 tdirty = (pmap_clear_modify(pg) || 1774 (pg->flags & PG_CLEAN) == 0); 1775 dirty += tdirty; 1776 } 1777 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1778 if (by_list) { 1779 curpg = TAILQ_NEXT(curpg, listq); 1780 } else { 1781 soff += fs->lfs_bsize; 1782 } 1783 continue; 1784 } 1785 1786 any_dirty += dirty; 1787 KASSERT(nonexistent == 0); 1788 1789 /* 1790 * If any are dirty make all dirty; unbusy them, 1791 * but if we were asked to clean, wire them so that 1792 * the pagedaemon doesn't bother us about them while 1793 * they're on their way to disk. 1794 */ 1795 for (i = 0; i == 0 || i < pages_per_block; i++) { 1796 pg = pgs[i]; 1797 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1798 if (dirty) { 1799 pg->flags &= ~PG_CLEAN; 1800 if (flags & PGO_FREE) { 1801 /* 1802 * Wire the page so that 1803 * pdaemon doesn't see it again. 1804 */ 1805 uvm_lock_pageq(); 1806 uvm_pagewire(pg); 1807 uvm_unlock_pageq(); 1808 1809 /* Suspended write flag */ 1810 pg->flags |= PG_DELWRI; 1811 } 1812 } 1813 if (pg->flags & PG_WANTED) 1814 wakeup(pg); 1815 pg->flags &= ~(PG_WANTED|PG_BUSY); 1816 UVM_PAGE_OWN(pg, NULL); 1817 } 1818 1819 if (checkfirst && any_dirty) 1820 break; 1821 1822 if (by_list) { 1823 curpg = TAILQ_NEXT(curpg, listq); 1824 } else { 1825 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1826 } 1827 } 1828 1829 return any_dirty; 1830 } 1831 1832 /* 1833 * lfs_putpages functions like genfs_putpages except that 1834 * 1835 * (1) It needs to bounds-check the incoming requests to ensure that 1836 * they are block-aligned; if they are not, expand the range and 1837 * do the right thing in case, e.g., the requested range is clean 1838 * but the expanded range is dirty. 1839 * 1840 * (2) It needs to explicitly send blocks to be written when it is done. 1841 * VOP_PUTPAGES is not ever called with the seglock held, so 1842 * we simply take the seglock and let lfs_segunlock wait for us. 1843 * XXX Actually we can be called with the seglock held, if we have 1844 * XXX to flush a vnode while lfs_markv is in operation. As of this 1845 * XXX writing we panic in this case. 1846 * 1847 * Assumptions: 1848 * 1849 * (1) The caller does not hold any pages in this vnode busy. If it does, 1850 * there is a danger that when we expand the page range and busy the 1851 * pages we will deadlock. 1852 * 1853 * (2) We are called with vp->v_interlock held; we must return with it 1854 * released. 1855 * 1856 * (3) We don't absolutely have to free pages right away, provided that 1857 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1858 * us a request with PGO_FREE, we take the pages out of the paging 1859 * queue and wake up the writer, which will handle freeing them for us. 1860 * 1861 * We ensure that for any filesystem block, all pages for that 1862 * block are either resident or not, even if those pages are higher 1863 * than EOF; that means that we will be getting requests to free 1864 * "unused" pages above EOF all the time, and should ignore them. 1865 * 1866 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 1867 * into has been set up for us by lfs_writefile. If not, we will 1868 * have to handle allocating and/or freeing an finfo entry. 1869 * 1870 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1871 */ 1872 1873 int 1874 lfs_putpages(void *v) 1875 { 1876 int error; 1877 struct vop_putpages_args /* { 1878 struct vnode *a_vp; 1879 voff_t a_offlo; 1880 voff_t a_offhi; 1881 int a_flags; 1882 } */ *ap = v; 1883 struct vnode *vp; 1884 struct inode *ip; 1885 struct lfs *fs; 1886 struct segment *sp; 1887 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1888 off_t off, max_endoffset; 1889 int s; 1890 boolean_t seglocked, sync, pagedaemon; 1891 struct vm_page *pg; 1892 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1893 1894 vp = ap->a_vp; 1895 ip = VTOI(vp); 1896 fs = ip->i_lfs; 1897 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1898 pagedaemon = (curproc == uvm.pagedaemon_proc); 1899 1900 /* Putpages does nothing for metadata. */ 1901 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1902 simple_unlock(&vp->v_interlock); 1903 return 0; 1904 } 1905 1906 /* 1907 * If there are no pages, don't do anything. 1908 */ 1909 if (vp->v_uobj.uo_npages == 0) { 1910 s = splbio(); 1911 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 1912 (vp->v_flag & VONWORKLST) && 1913 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 1914 vp->v_flag &= ~VWRITEMAPDIRTY; 1915 vn_syncer_remove_from_worklist(vp); 1916 } 1917 splx(s); 1918 simple_unlock(&vp->v_interlock); 1919 1920 /* Remove us from paging queue, if we were on it */ 1921 simple_lock(&fs->lfs_interlock); 1922 if (ip->i_flags & IN_PAGING) { 1923 ip->i_flags &= ~IN_PAGING; 1924 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1925 } 1926 simple_unlock(&fs->lfs_interlock); 1927 return 0; 1928 } 1929 1930 blkeof = blkroundup(fs, ip->i_size); 1931 1932 /* 1933 * Ignore requests to free pages past EOF but in the same block 1934 * as EOF, unless the request is synchronous. (If the request is 1935 * sync, it comes from lfs_truncate.) 1936 * XXXUBC Make these pages look "active" so the pagedaemon won't 1937 * XXXUBC bother us with them again. 1938 */ 1939 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1940 origoffset = ap->a_offlo; 1941 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1942 pg = uvm_pagelookup(&vp->v_uobj, off); 1943 KASSERT(pg != NULL); 1944 while (pg->flags & PG_BUSY) { 1945 pg->flags |= PG_WANTED; 1946 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1947 "lfsput2", 0); 1948 simple_lock(&vp->v_interlock); 1949 } 1950 uvm_lock_pageq(); 1951 uvm_pageactivate(pg); 1952 uvm_unlock_pageq(); 1953 } 1954 ap->a_offlo = blkeof; 1955 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1956 simple_unlock(&vp->v_interlock); 1957 return 0; 1958 } 1959 } 1960 1961 /* 1962 * Extend page range to start and end at block boundaries. 1963 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1964 */ 1965 origoffset = ap->a_offlo; 1966 origendoffset = ap->a_offhi; 1967 startoffset = origoffset & ~(fs->lfs_bmask); 1968 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1969 << fs->lfs_bshift; 1970 1971 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1972 endoffset = max_endoffset; 1973 origendoffset = endoffset; 1974 } else { 1975 origendoffset = round_page(ap->a_offhi); 1976 endoffset = round_page(blkroundup(fs, origendoffset)); 1977 } 1978 1979 KASSERT(startoffset > 0 || endoffset >= startoffset); 1980 if (startoffset == endoffset) { 1981 /* Nothing to do, why were we called? */ 1982 simple_unlock(&vp->v_interlock); 1983 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 1984 PRId64 "\n", startoffset)); 1985 return 0; 1986 } 1987 1988 ap->a_offlo = startoffset; 1989 ap->a_offhi = endoffset; 1990 1991 if (!(ap->a_flags & PGO_CLEANIT)) 1992 return genfs_putpages(v); 1993 1994 /* 1995 * If there are more than one page per block, we don't want 1996 * to get caught locking them backwards; so set PGO_BUSYFAIL 1997 * to avoid deadlocks. 1998 */ 1999 ap->a_flags |= PGO_BUSYFAIL; 2000 2001 do { 2002 int r; 2003 2004 /* If no pages are dirty, we can just use genfs_putpages. */ 2005 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2006 ap->a_flags, 1); 2007 if (r < 0) { 2008 simple_unlock(&vp->v_interlock); 2009 return EDEADLK; 2010 } 2011 if (r > 0) 2012 break; 2013 2014 /* 2015 * Sometimes pages are dirtied between the time that 2016 * we check and the time we try to clean them. 2017 * Instruct lfs_gop_write to return EDEADLK in this case 2018 * so we can write them properly. 2019 */ 2020 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2021 r = genfs_putpages(v); 2022 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2023 if (r != EDEADLK) 2024 return r; 2025 2026 /* Start over. */ 2027 preempt(1); 2028 simple_lock(&vp->v_interlock); 2029 } while(1); 2030 2031 /* 2032 * Dirty and asked to clean. 2033 * 2034 * Pagedaemon can't actually write LFS pages; wake up 2035 * the writer to take care of that. The writer will 2036 * notice the pager inode queue and act on that. 2037 */ 2038 if (pagedaemon) { 2039 simple_lock(&fs->lfs_interlock); 2040 if (!(ip->i_flags & IN_PAGING)) { 2041 ip->i_flags |= IN_PAGING; 2042 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2043 } 2044 simple_lock(&lfs_subsys_lock); 2045 wakeup(&lfs_writer_daemon); 2046 simple_unlock(&lfs_subsys_lock); 2047 simple_unlock(&fs->lfs_interlock); 2048 simple_unlock(&vp->v_interlock); 2049 preempt(1); 2050 return EWOULDBLOCK; 2051 } 2052 2053 /* 2054 * If this is a file created in a recent dirop, we can't flush its 2055 * inode until the dirop is complete. Drain dirops, then flush the 2056 * filesystem (taking care of any other pending dirops while we're 2057 * at it). 2058 */ 2059 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2060 (vp->v_flag & VDIROP)) { 2061 int locked; 2062 2063 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 2064 locked = (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 2065 simple_unlock(&vp->v_interlock); 2066 lfs_writer_enter(fs, "ppdirop"); 2067 if (locked) 2068 VOP_UNLOCK(vp, 0); 2069 2070 simple_lock(&fs->lfs_interlock); 2071 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2072 simple_unlock(&fs->lfs_interlock); 2073 2074 simple_lock(&vp->v_interlock); 2075 if (locked) { 2076 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2077 simple_lock(&vp->v_interlock); 2078 } 2079 lfs_writer_leave(fs); 2080 2081 /* XXX the flush should have taken care of this one too! */ 2082 } 2083 2084 /* 2085 * This is it. We are going to write some pages. From here on 2086 * down it's all just mechanics. 2087 * 2088 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2089 */ 2090 ap->a_flags &= ~PGO_SYNCIO; 2091 2092 /* 2093 * If we've already got the seglock, flush the node and return. 2094 * The FIP has already been set up for us by lfs_writefile, 2095 * and FIP cleanup and lfs_updatemeta will also be done there, 2096 * unless genfs_putpages returns EDEADLK; then we must flush 2097 * what we have, and correct FIP and segment header accounting. 2098 */ 2099 get_seglock: 2100 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2101 if (!seglocked) { 2102 simple_unlock(&vp->v_interlock); 2103 /* 2104 * Take the seglock, because we are going to be writing pages. 2105 */ 2106 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2107 if (error != 0) 2108 return error; 2109 simple_lock(&vp->v_interlock); 2110 } 2111 2112 /* 2113 * VOP_PUTPAGES should not be called while holding the seglock. 2114 * XXXUBC fix lfs_markv, or do this properly. 2115 */ 2116 #ifdef notyet 2117 KASSERT(fs->lfs_seglock == 1); 2118 #endif /* notyet */ 2119 2120 /* 2121 * We assume we're being called with sp->fip pointing at blank space. 2122 * Account for a new FIP in the segment header, and set sp->vp. 2123 * (This should duplicate the setup at the top of lfs_writefile().) 2124 */ 2125 sp = fs->lfs_sp; 2126 if (!seglocked) 2127 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2128 KASSERT(sp->vp == NULL); 2129 sp->vp = vp; 2130 2131 if (!seglocked) { 2132 if (vp->v_flag & VDIROP) 2133 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2134 } 2135 2136 /* 2137 * Loop through genfs_putpages until all pages are gathered. 2138 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2139 * Whenever we lose the interlock we have to rerun check_dirty, as 2140 * well. 2141 */ 2142 again: 2143 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2144 ap->a_flags, 0) < 0) { 2145 simple_unlock(&vp->v_interlock); 2146 sp->vp = NULL; 2147 if (!seglocked) { 2148 lfs_release_finfo(fs); 2149 lfs_segunlock(fs); 2150 } 2151 if (pagedaemon) 2152 return EDEADLK; 2153 /* else seglocked == 0 */ 2154 preempt(1); 2155 simple_lock(&vp->v_interlock); 2156 goto get_seglock; 2157 } 2158 2159 error = genfs_putpages(v); 2160 if (error == EDEADLK || error == EAGAIN) { 2161 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2162 " EDEADLK [2] ino %d off %x (seg %d)\n", 2163 ip->i_number, fs->lfs_offset, 2164 dtosn(fs, fs->lfs_offset))); 2165 /* If nothing to write, short-circuit */ 2166 if (sp->cbpp - sp->bpp > 1) { 2167 /* Write gathered pages */ 2168 lfs_updatemeta(sp); 2169 lfs_release_finfo(fs); 2170 (void) lfs_writeseg(fs, sp); 2171 2172 /* 2173 * Reinitialize brand new FIP and add us to it. 2174 */ 2175 KASSERT(sp->vp == vp); 2176 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2177 } 2178 2179 /* Give the write a chance to complete */ 2180 preempt(1); 2181 2182 /* We've lost the interlock. Start over. */ 2183 if (error == EDEADLK) { 2184 simple_lock(&vp->v_interlock); 2185 goto again; 2186 } 2187 } 2188 2189 KASSERT(sp->vp == vp); 2190 if (!seglocked) { 2191 sp->vp = NULL; 2192 2193 /* Write indirect blocks as well */ 2194 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2195 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2196 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2197 2198 KASSERT(sp->vp == NULL); 2199 sp->vp = vp; 2200 } 2201 2202 /* 2203 * Blocks are now gathered into a segment waiting to be written. 2204 * All that's left to do is update metadata, and write them. 2205 */ 2206 lfs_updatemeta(sp); 2207 KASSERT(sp->vp == vp); 2208 sp->vp = NULL; 2209 2210 if (seglocked) { 2211 /* we're called by lfs_writefile. */ 2212 return error; 2213 } 2214 2215 /* Clean up FIP and send it to disk. */ 2216 lfs_release_finfo(fs); 2217 lfs_writeseg(fs, fs->lfs_sp); 2218 2219 /* 2220 * Remove us from paging queue, since we've now written all our 2221 * pages. 2222 */ 2223 simple_lock(&fs->lfs_interlock); 2224 if (ip->i_flags & IN_PAGING) { 2225 ip->i_flags &= ~IN_PAGING; 2226 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2227 } 2228 simple_unlock(&fs->lfs_interlock); 2229 2230 /* 2231 * XXX - with the malloc/copy writeseg, the pages are freed by now 2232 * even if we don't wait (e.g. if we hold a nested lock). This 2233 * will not be true if we stop using malloc/copy. 2234 */ 2235 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2236 lfs_segunlock(fs); 2237 2238 /* 2239 * Wait for v_numoutput to drop to zero. The seglock should 2240 * take care of this, but there is a slight possibility that 2241 * aiodoned might not have got around to our buffers yet. 2242 */ 2243 if (sync) { 2244 s = splbio(); 2245 simple_lock(&global_v_numoutput_slock); 2246 while (vp->v_numoutput > 0) { 2247 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2248 " num %d\n", ip->i_number, vp->v_numoutput)); 2249 vp->v_flag |= VBWAIT; 2250 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 2251 &global_v_numoutput_slock); 2252 } 2253 simple_unlock(&global_v_numoutput_slock); 2254 splx(s); 2255 } 2256 return error; 2257 } 2258 2259 /* 2260 * Return the last logical file offset that should be written for this file 2261 * if we're doing a write that ends at "size". If writing, we need to know 2262 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2263 * to know about entire blocks. 2264 */ 2265 void 2266 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2267 { 2268 struct inode *ip = VTOI(vp); 2269 struct lfs *fs = ip->i_lfs; 2270 daddr_t olbn, nlbn; 2271 2272 olbn = lblkno(fs, ip->i_size); 2273 nlbn = lblkno(fs, size); 2274 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2275 *eobp = fragroundup(fs, size); 2276 } else { 2277 *eobp = blkroundup(fs, size); 2278 } 2279 } 2280 2281 #ifdef DEBUG 2282 void lfs_dump_vop(void *); 2283 2284 void 2285 lfs_dump_vop(void *v) 2286 { 2287 struct vop_putpages_args /* { 2288 struct vnode *a_vp; 2289 voff_t a_offlo; 2290 voff_t a_offhi; 2291 int a_flags; 2292 } */ *ap = v; 2293 2294 #ifdef DDB 2295 vfs_vnode_print(ap->a_vp, 0, printf); 2296 #endif 2297 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2298 } 2299 #endif 2300 2301 int 2302 lfs_mmap(void *v) 2303 { 2304 struct vop_mmap_args /* { 2305 const struct vnodeop_desc *a_desc; 2306 struct vnode *a_vp; 2307 int a_fflags; 2308 kauth_cred_t a_cred; 2309 struct lwp *a_l; 2310 } */ *ap = v; 2311 2312 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2313 return EOPNOTSUPP; 2314 return ufs_mmap(v); 2315 } 2316
Indexes created Sat Sep 27 14:09:57 GMT 2025