lfs_vnops.c revision 1.190
1 /* $NetBSD: lfs_vnops.c,v 1.190 2006/09/28 23:08:23 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 1.190 2006/09/28 23:08:23 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 vrele(ap->a_dvp); 750 if (ap->a_vp != ap->a_dvp) 751 VOP_UNLOCK(ap->a_dvp, 0); 752 vput(vp); 753 return error; 754 } 755 error = ufs_rmdir(ap); 756 if (ip->i_nlink == 0) 757 lfs_orphan(ip->i_lfs, ip->i_number); 758 SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 759 return (error); 760 } 761 762 int 763 lfs_link(void *v) 764 { 765 struct vop_link_args /* { 766 struct vnode *a_dvp; 767 struct vnode *a_vp; 768 struct componentname *a_cnp; 769 } */ *ap = v; 770 int error; 771 struct vnode **vpp = NULL; 772 773 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 774 vput(ap->a_dvp); 775 return error; 776 } 777 error = ufs_link(ap); 778 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 779 return (error); 780 } 781 782 int 783 lfs_rename(void *v) 784 { 785 struct vop_rename_args /* { 786 struct vnode *a_fdvp; 787 struct vnode *a_fvp; 788 struct componentname *a_fcnp; 789 struct vnode *a_tdvp; 790 struct vnode *a_tvp; 791 struct componentname *a_tcnp; 792 } */ *ap = v; 793 struct vnode *tvp, *fvp, *tdvp, *fdvp; 794 struct componentname *tcnp, *fcnp; 795 int error; 796 struct lfs *fs; 797 798 fs = VTOI(ap->a_fdvp)->i_lfs; 799 tvp = ap->a_tvp; 800 tdvp = ap->a_tdvp; 801 tcnp = ap->a_tcnp; 802 fvp = ap->a_fvp; 803 fdvp = ap->a_fdvp; 804 fcnp = ap->a_fcnp; 805 806 /* 807 * Check for cross-device rename. 808 * If it is, we don't want to set dirops, just error out. 809 * (In particular note that MARK_VNODE(tdvp) will DTWT on 810 * a cross-device rename.) 811 * 812 * Copied from ufs_rename. 813 */ 814 if ((fvp->v_mount != tdvp->v_mount) || 815 (tvp && (fvp->v_mount != tvp->v_mount))) { 816 error = EXDEV; 817 goto errout; 818 } 819 820 /* 821 * Check to make sure we're not renaming a vnode onto itself 822 * (deleting a hard link by renaming one name onto another); 823 * if we are we can't recursively call VOP_REMOVE since that 824 * would leave us with an unaccounted-for number of live dirops. 825 * 826 * Inline the relevant section of ufs_rename here, *before* 827 * calling SET_DIROP_REMOVE. 828 */ 829 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 830 (VTOI(tdvp)->i_flags & APPEND))) { 831 error = EPERM; 832 goto errout; 833 } 834 if (fvp == tvp) { 835 if (fvp->v_type == VDIR) { 836 error = EINVAL; 837 goto errout; 838 } 839 840 /* Release destination completely. */ 841 VOP_ABORTOP(tdvp, tcnp); 842 vput(tdvp); 843 vput(tvp); 844 845 /* Delete source. */ 846 vrele(fvp); 847 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 848 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 849 fcnp->cn_nameiop = DELETE; 850 if ((error = relookup(fdvp, &fvp, fcnp))){ 851 /* relookup blew away fdvp */ 852 return (error); 853 } 854 return (VOP_REMOVE(fdvp, fvp, fcnp)); 855 } 856 857 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 858 goto errout; 859 MARK_VNODE(fdvp); 860 MARK_VNODE(fvp); 861 862 error = ufs_rename(ap); 863 UNMARK_VNODE(fdvp); 864 UNMARK_VNODE(fvp); 865 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 866 return (error); 867 868 errout: 869 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 870 if (tdvp == tvp) 871 vrele(tdvp); 872 else 873 vput(tdvp); 874 if (tvp) 875 vput(tvp); 876 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 877 vrele(fdvp); 878 vrele(fvp); 879 return (error); 880 } 881 882 /* XXX hack to avoid calling ITIMES in getattr */ 883 int 884 lfs_getattr(void *v) 885 { 886 struct vop_getattr_args /* { 887 struct vnode *a_vp; 888 struct vattr *a_vap; 889 kauth_cred_t a_cred; 890 struct lwp *a_l; 891 } */ *ap = v; 892 struct vnode *vp = ap->a_vp; 893 struct inode *ip = VTOI(vp); 894 struct vattr *vap = ap->a_vap; 895 struct lfs *fs = ip->i_lfs; 896 /* 897 * Copy from inode table 898 */ 899 vap->va_fsid = ip->i_dev; 900 vap->va_fileid = ip->i_number; 901 vap->va_mode = ip->i_mode & ~IFMT; 902 vap->va_nlink = ip->i_nlink; 903 vap->va_uid = ip->i_uid; 904 vap->va_gid = ip->i_gid; 905 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 906 vap->va_size = vp->v_size; 907 vap->va_atime.tv_sec = ip->i_ffs1_atime; 908 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 909 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 910 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 911 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 912 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 913 vap->va_flags = ip->i_flags; 914 vap->va_gen = ip->i_gen; 915 /* this doesn't belong here */ 916 if (vp->v_type == VBLK) 917 vap->va_blocksize = BLKDEV_IOSIZE; 918 else if (vp->v_type == VCHR) 919 vap->va_blocksize = MAXBSIZE; 920 else 921 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 922 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 923 vap->va_type = vp->v_type; 924 vap->va_filerev = ip->i_modrev; 925 return (0); 926 } 927 928 /* 929 * Check to make sure the inode blocks won't choke the buffer 930 * cache, then call ufs_setattr as usual. 931 */ 932 int 933 lfs_setattr(void *v) 934 { 935 struct vop_setattr_args /* { 936 struct vnode *a_vp; 937 struct vattr *a_vap; 938 kauth_cred_t a_cred; 939 struct lwp *a_l; 940 } */ *ap = v; 941 struct vnode *vp = ap->a_vp; 942 943 lfs_check(vp, LFS_UNUSED_LBN, 0); 944 return ufs_setattr(v); 945 } 946 947 /* 948 * Release the block we hold on lfs_newseg wrapping. Called on file close, 949 * or explicitly from LFCNWRAPGO. Called with the interlock held. 950 */ 951 static int 952 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor) 953 { 954 if (lockstatus(&fs->lfs_stoplock) != LK_EXCLUSIVE) 955 return EBUSY; 956 957 lockmgr(&fs->lfs_stoplock, LK_RELEASE, &fs->lfs_interlock); 958 959 KASSERT(fs->lfs_nowrap > 0); 960 if (fs->lfs_nowrap <= 0) { 961 return 0; 962 } 963 964 if (--fs->lfs_nowrap == 0) { 965 log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt); 966 wakeup(&fs->lfs_wrappass); 967 lfs_wakeup_cleaner(fs); 968 } 969 if (waitfor) { 970 ltsleep(&fs->lfs_nextseg, PCATCH | PUSER, 971 "segment", 0, &fs->lfs_interlock); 972 } 973 974 return 0; 975 } 976 977 /* 978 * Close called 979 */ 980 /* ARGSUSED */ 981 int 982 lfs_close(void *v) 983 { 984 struct vop_close_args /* { 985 struct vnode *a_vp; 986 int a_fflag; 987 kauth_cred_t a_cred; 988 struct lwp *a_l; 989 } */ *ap = v; 990 struct vnode *vp = ap->a_vp; 991 struct inode *ip = VTOI(vp); 992 struct lfs *fs = ip->i_lfs; 993 994 if ((ip->i_number == ROOTINO || ip->i_number == LFS_IFILE_INUM) && 995 lockstatus(&fs->lfs_stoplock) == LK_EXCLUSIVE) { 996 simple_lock(&fs->lfs_interlock); 997 log(LOG_NOTICE, "lfs_close: releasing log wrap control\n"); 998 lfs_wrapgo(fs, ip, 0); 999 simple_unlock(&fs->lfs_interlock); 1000 } 1001 1002 if (vp == ip->i_lfs->lfs_ivnode && 1003 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 1004 return 0; 1005 1006 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 1007 LFS_ITIMES(ip, NULL, NULL, NULL); 1008 } 1009 return (0); 1010 } 1011 1012 /* 1013 * Close wrapper for special devices. 1014 * 1015 * Update the times on the inode then do device close. 1016 */ 1017 int 1018 lfsspec_close(void *v) 1019 { 1020 struct vop_close_args /* { 1021 struct vnode *a_vp; 1022 int a_fflag; 1023 kauth_cred_t a_cred; 1024 struct lwp *a_l; 1025 } */ *ap = v; 1026 struct vnode *vp; 1027 struct inode *ip; 1028 1029 vp = ap->a_vp; 1030 ip = VTOI(vp); 1031 if (vp->v_usecount > 1) { 1032 LFS_ITIMES(ip, NULL, NULL, NULL); 1033 } 1034 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1035 } 1036 1037 /* 1038 * Close wrapper for fifo's. 1039 * 1040 * Update the times on the inode then do device close. 1041 */ 1042 int 1043 lfsfifo_close(void *v) 1044 { 1045 struct vop_close_args /* { 1046 struct vnode *a_vp; 1047 int a_fflag; 1048 kauth_cred_ a_cred; 1049 struct lwp *a_l; 1050 } */ *ap = v; 1051 struct vnode *vp; 1052 struct inode *ip; 1053 1054 vp = ap->a_vp; 1055 ip = VTOI(vp); 1056 if (ap->a_vp->v_usecount > 1) { 1057 LFS_ITIMES(ip, NULL, NULL, NULL); 1058 } 1059 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1060 } 1061 1062 /* 1063 * Reclaim an inode so that it can be used for other purposes. 1064 */ 1065 1066 int 1067 lfs_reclaim(void *v) 1068 { 1069 struct vop_reclaim_args /* { 1070 struct vnode *a_vp; 1071 struct lwp *a_l; 1072 } */ *ap = v; 1073 struct vnode *vp = ap->a_vp; 1074 struct inode *ip = VTOI(vp); 1075 int error; 1076 1077 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1078 1079 LFS_CLR_UINO(ip, IN_ALLMOD); 1080 if ((error = ufs_reclaim(vp, ap->a_l))) 1081 return (error); 1082 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1083 lfs_deregister_all(vp); 1084 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1085 ip->inode_ext.lfs = NULL; 1086 pool_put(&lfs_inode_pool, vp->v_data); 1087 vp->v_data = NULL; 1088 return (0); 1089 } 1090 1091 /* 1092 * Read a block from a storage device. 1093 * In order to avoid reading blocks that are in the process of being 1094 * written by the cleaner---and hence are not mutexed by the normal 1095 * buffer cache / page cache mechanisms---check for collisions before 1096 * reading. 1097 * 1098 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1099 * the active cleaner test. 1100 * 1101 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1102 */ 1103 int 1104 lfs_strategy(void *v) 1105 { 1106 struct vop_strategy_args /* { 1107 struct vnode *a_vp; 1108 struct buf *a_bp; 1109 } */ *ap = v; 1110 struct buf *bp; 1111 struct lfs *fs; 1112 struct vnode *vp; 1113 struct inode *ip; 1114 daddr_t tbn; 1115 int i, sn, error, slept; 1116 1117 bp = ap->a_bp; 1118 vp = ap->a_vp; 1119 ip = VTOI(vp); 1120 fs = ip->i_lfs; 1121 1122 /* lfs uses its strategy routine only for read */ 1123 KASSERT(bp->b_flags & B_READ); 1124 1125 if (vp->v_type == VBLK || vp->v_type == VCHR) 1126 panic("lfs_strategy: spec"); 1127 KASSERT(bp->b_bcount != 0); 1128 if (bp->b_blkno == bp->b_lblkno) { 1129 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1130 NULL); 1131 if (error) { 1132 bp->b_error = error; 1133 bp->b_flags |= B_ERROR; 1134 biodone(bp); 1135 return (error); 1136 } 1137 if ((long)bp->b_blkno == -1) /* no valid data */ 1138 clrbuf(bp); 1139 } 1140 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1141 biodone(bp); 1142 return (0); 1143 } 1144 1145 slept = 1; 1146 simple_lock(&fs->lfs_interlock); 1147 while (slept && fs->lfs_seglock) { 1148 simple_unlock(&fs->lfs_interlock); 1149 /* 1150 * Look through list of intervals. 1151 * There will only be intervals to look through 1152 * if the cleaner holds the seglock. 1153 * Since the cleaner is synchronous, we can trust 1154 * the list of intervals to be current. 1155 */ 1156 tbn = dbtofsb(fs, bp->b_blkno); 1157 sn = dtosn(fs, tbn); 1158 slept = 0; 1159 for (i = 0; i < fs->lfs_cleanind; i++) { 1160 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1161 tbn >= fs->lfs_cleanint[i]) { 1162 DLOG((DLOG_CLEAN, 1163 "lfs_strategy: ino %d lbn %" PRId64 1164 " ind %d sn %d fsb %" PRIx32 1165 " given sn %d fsb %" PRIx64 "\n", 1166 ip->i_number, bp->b_lblkno, i, 1167 dtosn(fs, fs->lfs_cleanint[i]), 1168 fs->lfs_cleanint[i], sn, tbn)); 1169 DLOG((DLOG_CLEAN, 1170 "lfs_strategy: sleeping on ino %d lbn %" 1171 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1172 simple_lock(&fs->lfs_interlock); 1173 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1174 /* Cleaner can't wait for itself */ 1175 ltsleep(&fs->lfs_iocount, 1176 (PRIBIO + 1) | PNORELOCK, 1177 "clean2", 0, 1178 &fs->lfs_interlock); 1179 slept = 1; 1180 break; 1181 } else if (fs->lfs_seglock) { 1182 ltsleep(&fs->lfs_seglock, 1183 (PRIBIO + 1) | PNORELOCK, 1184 "clean1", 0, 1185 &fs->lfs_interlock); 1186 slept = 1; 1187 break; 1188 } 1189 simple_unlock(&fs->lfs_interlock); 1190 } 1191 } 1192 simple_lock(&fs->lfs_interlock); 1193 } 1194 simple_unlock(&fs->lfs_interlock); 1195 1196 vp = ip->i_devvp; 1197 VOP_STRATEGY(vp, bp); 1198 return (0); 1199 } 1200 1201 void 1202 lfs_flush_dirops(struct lfs *fs) 1203 { 1204 struct inode *ip, *nip; 1205 struct vnode *vp; 1206 extern int lfs_dostats; 1207 struct segment *sp; 1208 int waslocked; 1209 1210 ASSERT_MAYBE_SEGLOCK(fs); 1211 KASSERT(fs->lfs_nadirop == 0); 1212 1213 if (fs->lfs_ronly) 1214 return; 1215 1216 simple_lock(&fs->lfs_interlock); 1217 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1218 simple_unlock(&fs->lfs_interlock); 1219 return; 1220 } else 1221 simple_unlock(&fs->lfs_interlock); 1222 1223 if (lfs_dostats) 1224 ++lfs_stats.flush_invoked; 1225 1226 /* 1227 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1228 * Technically this is a checkpoint (the on-disk state is valid) 1229 * even though we are leaving out all the file data. 1230 */ 1231 lfs_imtime(fs); 1232 lfs_seglock(fs, SEGM_CKP); 1233 sp = fs->lfs_sp; 1234 1235 /* 1236 * lfs_writevnodes, optimized to get dirops out of the way. 1237 * Only write dirops, and don't flush files' pages, only 1238 * blocks from the directories. 1239 * 1240 * We don't need to vref these files because they are 1241 * dirops and so hold an extra reference until the 1242 * segunlock clears them of that status. 1243 * 1244 * We don't need to check for IN_ADIROP because we know that 1245 * no dirops are active. 1246 * 1247 */ 1248 simple_lock(&fs->lfs_interlock); 1249 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1250 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1251 simple_unlock(&fs->lfs_interlock); 1252 vp = ITOV(ip); 1253 1254 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1255 1256 /* 1257 * All writes to directories come from dirops; all 1258 * writes to files' direct blocks go through the page 1259 * cache, which we're not touching. Reads to files 1260 * and/or directories will not be affected by writing 1261 * directory blocks inodes and file inodes. So we don't 1262 * really need to lock. If we don't lock, though, 1263 * make sure that we don't clear IN_MODIFIED 1264 * unnecessarily. 1265 */ 1266 if (vp->v_flag & (VXLOCK | VFREEING)) { 1267 simple_lock(&fs->lfs_interlock); 1268 continue; 1269 } 1270 waslocked = VOP_ISLOCKED(vp); 1271 if (vp->v_type != VREG && 1272 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1273 lfs_writefile(fs, sp, vp); 1274 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1275 !(ip->i_flag & IN_ALLMOD)) { 1276 LFS_SET_UINO(ip, IN_MODIFIED); 1277 } 1278 } 1279 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1280 (void) lfs_writeinode(fs, sp, ip); 1281 if (waslocked == LK_EXCLOTHER) 1282 LFS_SET_UINO(ip, IN_MODIFIED); 1283 simple_lock(&fs->lfs_interlock); 1284 } 1285 simple_unlock(&fs->lfs_interlock); 1286 /* We've written all the dirops there are */ 1287 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1288 lfs_finalize_fs_seguse(fs); 1289 (void) lfs_writeseg(fs, sp); 1290 lfs_segunlock(fs); 1291 } 1292 1293 /* 1294 * Flush all vnodes for which the pagedaemon has requested pageouts. 1295 * Skip over any files that are marked VDIROP (since lfs_flush_dirop() 1296 * has just run, this would be an error). If we have to skip a vnode 1297 * for any reason, just skip it; if we have to wait for the cleaner, 1298 * abort. The writer daemon will call us again later. 1299 */ 1300 void 1301 lfs_flush_pchain(struct lfs *fs) 1302 { 1303 struct inode *ip, *nip; 1304 struct vnode *vp; 1305 extern int lfs_dostats; 1306 struct segment *sp; 1307 int error; 1308 1309 ASSERT_NO_SEGLOCK(fs); 1310 1311 if (fs->lfs_ronly) 1312 return; 1313 1314 simple_lock(&fs->lfs_interlock); 1315 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1316 simple_unlock(&fs->lfs_interlock); 1317 return; 1318 } else 1319 simple_unlock(&fs->lfs_interlock); 1320 1321 /* Get dirops out of the way */ 1322 lfs_flush_dirops(fs); 1323 1324 if (lfs_dostats) 1325 ++lfs_stats.flush_invoked; 1326 1327 /* 1328 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1329 */ 1330 lfs_imtime(fs); 1331 lfs_seglock(fs, 0); 1332 sp = fs->lfs_sp; 1333 1334 /* 1335 * lfs_writevnodes, optimized to clear pageout requests. 1336 * Only write non-dirop files that are in the pageout queue. 1337 * We're very conservative about what we write; we want to be 1338 * fast and async. 1339 */ 1340 simple_lock(&fs->lfs_interlock); 1341 top: 1342 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1343 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1344 vp = ITOV(ip); 1345 1346 if (!(ip->i_flags & IN_PAGING)) 1347 goto top; 1348 1349 if (vp->v_flag & (VXLOCK|VDIROP)) 1350 continue; 1351 if (vp->v_type != VREG) 1352 continue; 1353 if (lfs_vref(vp)) 1354 continue; 1355 simple_unlock(&fs->lfs_interlock); 1356 1357 if (VOP_ISLOCKED(vp)) { 1358 lfs_vunref(vp); 1359 simple_lock(&fs->lfs_interlock); 1360 continue; 1361 } 1362 1363 error = lfs_writefile(fs, sp, vp); 1364 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1365 !(ip->i_flag & IN_ALLMOD)) { 1366 LFS_SET_UINO(ip, IN_MODIFIED); 1367 } 1368 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1369 (void) lfs_writeinode(fs, sp, ip); 1370 1371 lfs_vunref(vp); 1372 1373 if (error == EAGAIN) { 1374 lfs_writeseg(fs, sp); 1375 simple_lock(&fs->lfs_interlock); 1376 break; 1377 } 1378 simple_lock(&fs->lfs_interlock); 1379 } 1380 simple_unlock(&fs->lfs_interlock); 1381 (void) lfs_writeseg(fs, sp); 1382 lfs_segunlock(fs); 1383 } 1384 1385 /* 1386 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1387 */ 1388 int 1389 lfs_fcntl(void *v) 1390 { 1391 struct vop_fcntl_args /* { 1392 struct vnode *a_vp; 1393 u_long a_command; 1394 caddr_t a_data; 1395 int a_fflag; 1396 kauth_cred_t a_cred; 1397 struct lwp *a_l; 1398 } */ *ap = v; 1399 struct timeval *tvp; 1400 BLOCK_INFO *blkiov; 1401 CLEANERINFO *cip; 1402 SEGUSE *sup; 1403 int blkcnt, error, oclean; 1404 size_t fh_size; 1405 struct lfs_fcntl_markv blkvp; 1406 struct lwp *l; 1407 fsid_t *fsidp; 1408 struct lfs *fs; 1409 struct buf *bp; 1410 fhandle_t *fhp; 1411 daddr_t off; 1412 1413 /* Only respect LFS fcntls on fs root or Ifile */ 1414 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1415 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1416 return ufs_fcntl(v); 1417 } 1418 1419 /* Avoid locking a draining lock */ 1420 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1421 return ESHUTDOWN; 1422 } 1423 1424 /* LFS control and monitoring fcntls are available only to root */ 1425 l = ap->a_l; 1426 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1427 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1428 &l->l_acflag)) != 0) 1429 return (error); 1430 1431 fs = VTOI(ap->a_vp)->i_lfs; 1432 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1433 1434 error = 0; 1435 switch (ap->a_command) { 1436 case LFCNSEGWAITALL: 1437 case LFCNSEGWAITALL_COMPAT: 1438 fsidp = NULL; 1439 /* FALLSTHROUGH */ 1440 case LFCNSEGWAIT: 1441 case LFCNSEGWAIT_COMPAT: 1442 tvp = (struct timeval *)ap->a_data; 1443 simple_lock(&fs->lfs_interlock); 1444 ++fs->lfs_sleepers; 1445 simple_unlock(&fs->lfs_interlock); 1446 1447 error = lfs_segwait(fsidp, tvp); 1448 1449 simple_lock(&fs->lfs_interlock); 1450 if (--fs->lfs_sleepers == 0) 1451 wakeup(&fs->lfs_sleepers); 1452 simple_unlock(&fs->lfs_interlock); 1453 return error; 1454 1455 case LFCNBMAPV: 1456 case LFCNMARKV: 1457 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1458 1459 blkcnt = blkvp.blkcnt; 1460 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1461 return (EINVAL); 1462 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1463 if ((error = copyin(blkvp.blkiov, blkiov, 1464 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1465 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1466 return error; 1467 } 1468 1469 simple_lock(&fs->lfs_interlock); 1470 ++fs->lfs_sleepers; 1471 simple_unlock(&fs->lfs_interlock); 1472 if (ap->a_command == LFCNBMAPV) 1473 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1474 else /* LFCNMARKV */ 1475 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1476 if (error == 0) 1477 error = copyout(blkiov, blkvp.blkiov, 1478 blkcnt * sizeof(BLOCK_INFO)); 1479 simple_lock(&fs->lfs_interlock); 1480 if (--fs->lfs_sleepers == 0) 1481 wakeup(&fs->lfs_sleepers); 1482 simple_unlock(&fs->lfs_interlock); 1483 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1484 return error; 1485 1486 case LFCNRECLAIM: 1487 /* 1488 * Flush dirops and write Ifile, allowing empty segments 1489 * to be immediately reclaimed. 1490 */ 1491 lfs_writer_enter(fs, "pndirop"); 1492 off = fs->lfs_offset; 1493 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1494 lfs_flush_dirops(fs); 1495 LFS_CLEANERINFO(cip, fs, bp); 1496 oclean = cip->clean; 1497 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1498 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1499 fs->lfs_sp->seg_flags |= SEGM_PROT; 1500 lfs_segunlock(fs); 1501 lfs_writer_leave(fs); 1502 1503 #ifdef DEBUG 1504 LFS_CLEANERINFO(cip, fs, bp); 1505 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1506 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1507 fs->lfs_offset - off, cip->clean - oclean, 1508 fs->lfs_activesb)); 1509 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1510 #endif 1511 1512 return 0; 1513 1514 #ifdef COMPAT_30 1515 case LFCNIFILEFH_COMPAT: 1516 /* Return the filehandle of the Ifile */ 1517 if ((error = kauth_authorize_generic(l->l_cred, 1518 KAUTH_GENERIC_ISSUSER, &l->l_acflag)) != 0) 1519 return (error); 1520 fhp = (struct fhandle *)ap->a_data; 1521 fhp->fh_fsid = *fsidp; 1522 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1523 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1524 #endif 1525 1526 case LFCNIFILEFH_COMPAT2: 1527 case LFCNIFILEFH: 1528 /* Return the filehandle of the Ifile */ 1529 fhp = (struct fhandle *)ap->a_data; 1530 fhp->fh_fsid = *fsidp; 1531 fh_size = sizeof(struct lfs_fhandle) - 1532 offsetof(fhandle_t, fh_fid); 1533 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1534 1535 case LFCNREWIND: 1536 /* Move lfs_offset to the lowest-numbered segment */ 1537 return lfs_rewind(fs, *(int *)ap->a_data); 1538 1539 case LFCNINVAL: 1540 /* Mark a segment SEGUSE_INVAL */ 1541 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1542 if (sup->su_nbytes > 0) { 1543 brelse(bp); 1544 lfs_unset_inval_all(fs); 1545 return EBUSY; 1546 } 1547 sup->su_flags |= SEGUSE_INVAL; 1548 VOP_BWRITE(bp); 1549 return 0; 1550 1551 case LFCNRESIZE: 1552 /* Resize the filesystem */ 1553 return lfs_resize_fs(fs, *(int *)ap->a_data); 1554 1555 case LFCNWRAPSTOP: 1556 case LFCNWRAPSTOP_COMPAT: 1557 /* 1558 * Hold lfs_newseg at segment 0; if requested, sleep until 1559 * the filesystem wraps around. To support external agents 1560 * (dump, fsck-based regression test) that need to look at 1561 * a snapshot of the filesystem, without necessarily 1562 * requiring that all fs activity stops. 1563 */ 1564 if (lockstatus(&fs->lfs_stoplock)) 1565 return EALREADY; 1566 1567 simple_lock(&fs->lfs_interlock); 1568 lockmgr(&fs->lfs_stoplock, LK_EXCLUSIVE, &fs->lfs_interlock); 1569 if (fs->lfs_nowrap == 0) 1570 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1571 ++fs->lfs_nowrap; 1572 if (*(int *)ap->a_data == 1 || 1573 ap->a_command == LFCNWRAPSTOP_COMPAT) { 1574 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1575 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1576 "segwrap", 0, &fs->lfs_interlock); 1577 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1578 if (error) { 1579 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1580 } 1581 } 1582 simple_unlock(&fs->lfs_interlock); 1583 return 0; 1584 1585 case LFCNWRAPGO: 1586 case LFCNWRAPGO_COMPAT: 1587 /* 1588 * Having done its work, the agent wakes up the writer. 1589 * If the argument is 1, it sleeps until a new segment 1590 * is selected. 1591 */ 1592 simple_lock(&fs->lfs_interlock); 1593 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1594 (ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1595 *((int *)ap->a_data))); 1596 simple_unlock(&fs->lfs_interlock); 1597 return error; 1598 1599 case LFCNWRAPPASS: 1600 if (lockstatus(&fs->lfs_stoplock) != LK_EXCLUSIVE) 1601 return EALREADY; 1602 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1603 return EALREADY; 1604 simple_lock(&fs->lfs_interlock); 1605 if (fs->lfs_nowrap == 0) { 1606 simple_unlock(&fs->lfs_interlock); 1607 return EBUSY; 1608 } 1609 fs->lfs_wrappass = 1; 1610 wakeup(&fs->lfs_wrappass); 1611 /* Wait for the log to wrap, if asked */ 1612 if (*(int *)ap->a_data) { 1613 lfs_vref(ap->a_vp); 1614 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1615 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1616 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1617 "segwrap", 0, &fs->lfs_interlock); 1618 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1619 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1620 lfs_vunref(ap->a_vp); 1621 } 1622 simple_unlock(&fs->lfs_interlock); 1623 return error; 1624 1625 case LFCNWRAPSTATUS: 1626 simple_lock(&fs->lfs_interlock); 1627 *(int *)ap->a_data = fs->lfs_wrapstatus; 1628 simple_unlock(&fs->lfs_interlock); 1629 return 0; 1630 1631 default: 1632 return ufs_fcntl(v); 1633 } 1634 return 0; 1635 } 1636 1637 int 1638 lfs_getpages(void *v) 1639 { 1640 struct vop_getpages_args /* { 1641 struct vnode *a_vp; 1642 voff_t a_offset; 1643 struct vm_page **a_m; 1644 int *a_count; 1645 int a_centeridx; 1646 vm_prot_t a_access_type; 1647 int a_advice; 1648 int a_flags; 1649 } */ *ap = v; 1650 1651 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1652 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1653 return EPERM; 1654 } 1655 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1656 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1657 } 1658 1659 /* 1660 * we're relying on the fact that genfs_getpages() always read in 1661 * entire filesystem blocks. 1662 */ 1663 return genfs_getpages(v); 1664 } 1665 1666 /* 1667 * Make sure that for all pages in every block in the given range, 1668 * either all are dirty or all are clean. If any of the pages 1669 * we've seen so far are dirty, put the vnode on the paging chain, 1670 * and mark it IN_PAGING. 1671 * 1672 * If checkfirst != 0, don't check all the pages but return at the 1673 * first dirty page. 1674 */ 1675 static int 1676 check_dirty(struct lfs *fs, struct vnode *vp, 1677 off_t startoffset, off_t endoffset, off_t blkeof, 1678 int flags, int checkfirst) 1679 { 1680 int by_list; 1681 struct vm_page *curpg = NULL; /* XXX: gcc */ 1682 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1683 off_t soff = 0; /* XXX: gcc */ 1684 voff_t off; 1685 int i; 1686 int nonexistent; 1687 int any_dirty; /* number of dirty pages */ 1688 int dirty; /* number of dirty pages in a block */ 1689 int tdirty; 1690 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1691 int pagedaemon = (curproc == uvm.pagedaemon_proc); 1692 1693 ASSERT_MAYBE_SEGLOCK(fs); 1694 top: 1695 by_list = (vp->v_uobj.uo_npages <= 1696 ((endoffset - startoffset) >> PAGE_SHIFT) * 1697 UVM_PAGE_HASH_PENALTY); 1698 any_dirty = 0; 1699 1700 if (by_list) { 1701 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1702 } else { 1703 soff = startoffset; 1704 } 1705 while (by_list || soff < MIN(blkeof, endoffset)) { 1706 if (by_list) { 1707 /* 1708 * Find the first page in a block. Skip 1709 * blocks outside our area of interest or beyond 1710 * the end of file. 1711 */ 1712 if (pages_per_block > 1) { 1713 while (curpg && 1714 ((curpg->offset & fs->lfs_bmask) || 1715 curpg->offset >= vp->v_size || 1716 curpg->offset >= endoffset)) 1717 curpg = TAILQ_NEXT(curpg, listq); 1718 } 1719 if (curpg == NULL) 1720 break; 1721 soff = curpg->offset; 1722 } 1723 1724 /* 1725 * Mark all pages in extended range busy; find out if any 1726 * of them are dirty. 1727 */ 1728 nonexistent = dirty = 0; 1729 for (i = 0; i == 0 || i < pages_per_block; i++) { 1730 if (by_list && pages_per_block <= 1) { 1731 pgs[i] = pg = curpg; 1732 } else { 1733 off = soff + (i << PAGE_SHIFT); 1734 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1735 if (pg == NULL) { 1736 ++nonexistent; 1737 continue; 1738 } 1739 } 1740 KASSERT(pg != NULL); 1741 1742 /* 1743 * If we're holding the segment lock, we can deadlock 1744 * against a process that has our page and is waiting 1745 * for the cleaner, while the cleaner waits for the 1746 * segment lock. Just bail in that case. 1747 */ 1748 if ((pg->flags & PG_BUSY) && 1749 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1750 if (by_list && i > 0) 1751 uvm_page_unbusy(pgs, i); 1752 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1753 return -1; 1754 } 1755 1756 while (pg->flags & PG_BUSY) { 1757 pg->flags |= PG_WANTED; 1758 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1759 "lfsput", 0); 1760 simple_lock(&vp->v_interlock); 1761 if (by_list) { 1762 if (i > 0) 1763 uvm_page_unbusy(pgs, i); 1764 goto top; 1765 } 1766 } 1767 pg->flags |= PG_BUSY; 1768 UVM_PAGE_OWN(pg, "lfs_putpages"); 1769 1770 pmap_page_protect(pg, VM_PROT_NONE); 1771 tdirty = (pmap_clear_modify(pg) || 1772 (pg->flags & PG_CLEAN) == 0); 1773 dirty += tdirty; 1774 } 1775 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1776 if (by_list) { 1777 curpg = TAILQ_NEXT(curpg, listq); 1778 } else { 1779 soff += fs->lfs_bsize; 1780 } 1781 continue; 1782 } 1783 1784 any_dirty += dirty; 1785 KASSERT(nonexistent == 0); 1786 1787 /* 1788 * If any are dirty make all dirty; unbusy them, 1789 * but if we were asked to clean, wire them so that 1790 * the pagedaemon doesn't bother us about them while 1791 * they're on their way to disk. 1792 */ 1793 for (i = 0; i == 0 || i < pages_per_block; i++) { 1794 pg = pgs[i]; 1795 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1796 if (dirty) { 1797 pg->flags &= ~PG_CLEAN; 1798 if (flags & PGO_FREE) { 1799 /* 1800 * Wire the page so that 1801 * pdaemon doesn't see it again. 1802 */ 1803 uvm_lock_pageq(); 1804 uvm_pagewire(pg); 1805 uvm_unlock_pageq(); 1806 1807 /* Suspended write flag */ 1808 pg->flags |= PG_DELWRI; 1809 } 1810 } 1811 if (pg->flags & PG_WANTED) 1812 wakeup(pg); 1813 pg->flags &= ~(PG_WANTED|PG_BUSY); 1814 UVM_PAGE_OWN(pg, NULL); 1815 } 1816 1817 if (checkfirst && any_dirty) 1818 break; 1819 1820 if (by_list) { 1821 curpg = TAILQ_NEXT(curpg, listq); 1822 } else { 1823 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1824 } 1825 } 1826 1827 return any_dirty; 1828 } 1829 1830 /* 1831 * lfs_putpages functions like genfs_putpages except that 1832 * 1833 * (1) It needs to bounds-check the incoming requests to ensure that 1834 * they are block-aligned; if they are not, expand the range and 1835 * do the right thing in case, e.g., the requested range is clean 1836 * but the expanded range is dirty. 1837 * 1838 * (2) It needs to explicitly send blocks to be written when it is done. 1839 * VOP_PUTPAGES is not ever called with the seglock held, so 1840 * we simply take the seglock and let lfs_segunlock wait for us. 1841 * XXX Actually we can be called with the seglock held, if we have 1842 * XXX to flush a vnode while lfs_markv is in operation. As of this 1843 * XXX writing we panic in this case. 1844 * 1845 * Assumptions: 1846 * 1847 * (1) The caller does not hold any pages in this vnode busy. If it does, 1848 * there is a danger that when we expand the page range and busy the 1849 * pages we will deadlock. 1850 * 1851 * (2) We are called with vp->v_interlock held; we must return with it 1852 * released. 1853 * 1854 * (3) We don't absolutely have to free pages right away, provided that 1855 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1856 * us a request with PGO_FREE, we take the pages out of the paging 1857 * queue and wake up the writer, which will handle freeing them for us. 1858 * 1859 * We ensure that for any filesystem block, all pages for that 1860 * block are either resident or not, even if those pages are higher 1861 * than EOF; that means that we will be getting requests to free 1862 * "unused" pages above EOF all the time, and should ignore them. 1863 * 1864 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 1865 * into has been set up for us by lfs_writefile. If not, we will 1866 * have to handle allocating and/or freeing an finfo entry. 1867 * 1868 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1869 */ 1870 1871 int 1872 lfs_putpages(void *v) 1873 { 1874 int error; 1875 struct vop_putpages_args /* { 1876 struct vnode *a_vp; 1877 voff_t a_offlo; 1878 voff_t a_offhi; 1879 int a_flags; 1880 } */ *ap = v; 1881 struct vnode *vp; 1882 struct inode *ip; 1883 struct lfs *fs; 1884 struct segment *sp; 1885 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1886 off_t off, max_endoffset; 1887 int s; 1888 boolean_t seglocked, sync, pagedaemon; 1889 struct vm_page *pg; 1890 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1891 1892 vp = ap->a_vp; 1893 ip = VTOI(vp); 1894 fs = ip->i_lfs; 1895 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1896 pagedaemon = (curproc == uvm.pagedaemon_proc); 1897 1898 /* Putpages does nothing for metadata. */ 1899 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1900 simple_unlock(&vp->v_interlock); 1901 return 0; 1902 } 1903 1904 /* 1905 * If there are no pages, don't do anything. 1906 */ 1907 if (vp->v_uobj.uo_npages == 0) { 1908 s = splbio(); 1909 if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL && 1910 (vp->v_flag & VONWORKLST)) { 1911 vp->v_flag &= ~VONWORKLST; 1912 LIST_REMOVE(vp, v_synclist); 1913 } 1914 splx(s); 1915 simple_unlock(&vp->v_interlock); 1916 1917 /* Remove us from paging queue, if we were on it */ 1918 simple_lock(&fs->lfs_interlock); 1919 if (ip->i_flags & IN_PAGING) { 1920 ip->i_flags &= ~IN_PAGING; 1921 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1922 } 1923 simple_unlock(&fs->lfs_interlock); 1924 return 0; 1925 } 1926 1927 blkeof = blkroundup(fs, ip->i_size); 1928 1929 /* 1930 * Ignore requests to free pages past EOF but in the same block 1931 * as EOF, unless the request is synchronous. (If the request is 1932 * sync, it comes from lfs_truncate.) 1933 * XXXUBC Make these pages look "active" so the pagedaemon won't 1934 * XXXUBC bother us with them again. 1935 */ 1936 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1937 origoffset = ap->a_offlo; 1938 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1939 pg = uvm_pagelookup(&vp->v_uobj, off); 1940 KASSERT(pg != NULL); 1941 while (pg->flags & PG_BUSY) { 1942 pg->flags |= PG_WANTED; 1943 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1944 "lfsput2", 0); 1945 simple_lock(&vp->v_interlock); 1946 } 1947 uvm_lock_pageq(); 1948 uvm_pageactivate(pg); 1949 uvm_unlock_pageq(); 1950 } 1951 ap->a_offlo = blkeof; 1952 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1953 simple_unlock(&vp->v_interlock); 1954 return 0; 1955 } 1956 } 1957 1958 /* 1959 * Extend page range to start and end at block boundaries. 1960 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1961 */ 1962 origoffset = ap->a_offlo; 1963 origendoffset = ap->a_offhi; 1964 startoffset = origoffset & ~(fs->lfs_bmask); 1965 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1966 << fs->lfs_bshift; 1967 1968 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1969 endoffset = max_endoffset; 1970 origendoffset = endoffset; 1971 } else { 1972 origendoffset = round_page(ap->a_offhi); 1973 endoffset = round_page(blkroundup(fs, origendoffset)); 1974 } 1975 1976 KASSERT(startoffset > 0 || endoffset >= startoffset); 1977 if (startoffset == endoffset) { 1978 /* Nothing to do, why were we called? */ 1979 simple_unlock(&vp->v_interlock); 1980 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 1981 PRId64 "\n", startoffset)); 1982 return 0; 1983 } 1984 1985 ap->a_offlo = startoffset; 1986 ap->a_offhi = endoffset; 1987 1988 if (!(ap->a_flags & PGO_CLEANIT)) 1989 return genfs_putpages(v); 1990 1991 /* 1992 * If there are more than one page per block, we don't want 1993 * to get caught locking them backwards; so set PGO_BUSYFAIL 1994 * to avoid deadlocks. 1995 */ 1996 ap->a_flags |= PGO_BUSYFAIL; 1997 1998 do { 1999 int r; 2000 2001 /* If no pages are dirty, we can just use genfs_putpages. */ 2002 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2003 ap->a_flags, 1); 2004 if (r < 0) { 2005 simple_unlock(&vp->v_interlock); 2006 return EDEADLK; 2007 } 2008 if (r > 0) 2009 break; 2010 2011 /* 2012 * Sometimes pages are dirtied between the time that 2013 * we check and the time we try to clean them. 2014 * Instruct lfs_gop_write to return EDEADLK in this case 2015 * so we can write them properly. 2016 */ 2017 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2018 r = genfs_putpages(v); 2019 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2020 if (r != EDEADLK) 2021 return r; 2022 2023 /* Start over. */ 2024 preempt(1); 2025 simple_lock(&vp->v_interlock); 2026 } while(1); 2027 2028 /* 2029 * Dirty and asked to clean. 2030 * 2031 * Pagedaemon can't actually write LFS pages; wake up 2032 * the writer to take care of that. The writer will 2033 * notice the pager inode queue and act on that. 2034 */ 2035 if (pagedaemon) { 2036 simple_lock(&fs->lfs_interlock); 2037 if (!(ip->i_flags & IN_PAGING)) { 2038 ip->i_flags |= IN_PAGING; 2039 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2040 } 2041 simple_lock(&lfs_subsys_lock); 2042 wakeup(&lfs_writer_daemon); 2043 simple_unlock(&lfs_subsys_lock); 2044 simple_unlock(&fs->lfs_interlock); 2045 simple_unlock(&vp->v_interlock); 2046 preempt(1); 2047 return EWOULDBLOCK; 2048 } 2049 2050 /* 2051 * If this is a file created in a recent dirop, we can't flush its 2052 * inode until the dirop is complete. Drain dirops, then flush the 2053 * filesystem (taking care of any other pending dirops while we're 2054 * at it). 2055 */ 2056 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2057 (vp->v_flag & VDIROP)) { 2058 int locked; 2059 2060 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 2061 locked = (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 2062 simple_unlock(&vp->v_interlock); 2063 lfs_writer_enter(fs, "ppdirop"); 2064 if (locked) 2065 VOP_UNLOCK(vp, 0); 2066 2067 simple_lock(&fs->lfs_interlock); 2068 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2069 simple_unlock(&fs->lfs_interlock); 2070 2071 simple_lock(&vp->v_interlock); 2072 if (locked) { 2073 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2074 simple_lock(&vp->v_interlock); 2075 } 2076 lfs_writer_leave(fs); 2077 2078 /* XXX the flush should have taken care of this one too! */ 2079 } 2080 2081 /* 2082 * This is it. We are going to write some pages. From here on 2083 * down it's all just mechanics. 2084 * 2085 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2086 */ 2087 ap->a_flags &= ~PGO_SYNCIO; 2088 2089 /* 2090 * If we've already got the seglock, flush the node and return. 2091 * The FIP has already been set up for us by lfs_writefile, 2092 * and FIP cleanup and lfs_updatemeta will also be done there, 2093 * unless genfs_putpages returns EDEADLK; then we must flush 2094 * what we have, and correct FIP and segment header accounting. 2095 */ 2096 get_seglock: 2097 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2098 if (!seglocked) { 2099 simple_unlock(&vp->v_interlock); 2100 /* 2101 * Take the seglock, because we are going to be writing pages. 2102 */ 2103 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2104 if (error != 0) 2105 return error; 2106 simple_lock(&vp->v_interlock); 2107 } 2108 2109 /* 2110 * VOP_PUTPAGES should not be called while holding the seglock. 2111 * XXXUBC fix lfs_markv, or do this properly. 2112 */ 2113 #ifdef notyet 2114 KASSERT(fs->lfs_seglock == 1); 2115 #endif /* notyet */ 2116 2117 /* 2118 * We assume we're being called with sp->fip pointing at blank space. 2119 * Account for a new FIP in the segment header, and set sp->vp. 2120 * (This should duplicate the setup at the top of lfs_writefile().) 2121 */ 2122 sp = fs->lfs_sp; 2123 if (!seglocked) 2124 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2125 KASSERT(sp->vp == NULL); 2126 sp->vp = vp; 2127 2128 if (!seglocked) { 2129 if (vp->v_flag & VDIROP) 2130 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2131 } 2132 2133 /* 2134 * Loop through genfs_putpages until all pages are gathered. 2135 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2136 * Whenever we lose the interlock we have to rerun check_dirty, as 2137 * well. 2138 */ 2139 again: 2140 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2141 ap->a_flags, 0) < 0) { 2142 simple_unlock(&vp->v_interlock); 2143 sp->vp = NULL; 2144 if (!seglocked) { 2145 lfs_release_finfo(fs); 2146 lfs_segunlock(fs); 2147 } 2148 if (pagedaemon) 2149 return EDEADLK; 2150 /* else seglocked == 0 */ 2151 preempt(1); 2152 simple_lock(&vp->v_interlock); 2153 goto get_seglock; 2154 } 2155 2156 error = genfs_putpages(v); 2157 if (error == EDEADLK || error == EAGAIN) { 2158 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2159 " EDEADLK [2] ino %d off %x (seg %d)\n", 2160 ip->i_number, fs->lfs_offset, 2161 dtosn(fs, fs->lfs_offset))); 2162 /* If nothing to write, short-circuit */ 2163 if (sp->cbpp - sp->bpp > 1) { 2164 /* Write gathered pages */ 2165 lfs_updatemeta(sp); 2166 lfs_release_finfo(fs); 2167 (void) lfs_writeseg(fs, sp); 2168 2169 /* 2170 * Reinitialize brand new FIP and add us to it. 2171 */ 2172 KASSERT(sp->vp == vp); 2173 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2174 } 2175 2176 /* Give the write a chance to complete */ 2177 preempt(1); 2178 2179 /* We've lost the interlock. Start over. */ 2180 if (error == EDEADLK) { 2181 simple_lock(&vp->v_interlock); 2182 goto again; 2183 } 2184 } 2185 2186 KASSERT(sp->vp == vp); 2187 if (!seglocked) { 2188 sp->vp = NULL; 2189 2190 /* Write indirect blocks as well */ 2191 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2192 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2193 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2194 2195 KASSERT(sp->vp == NULL); 2196 sp->vp = vp; 2197 } 2198 2199 /* 2200 * Blocks are now gathered into a segment waiting to be written. 2201 * All that's left to do is update metadata, and write them. 2202 */ 2203 lfs_updatemeta(sp); 2204 KASSERT(sp->vp == vp); 2205 sp->vp = NULL; 2206 2207 if (seglocked) { 2208 /* we're called by lfs_writefile. */ 2209 return error; 2210 } 2211 2212 /* Clean up FIP and send it to disk. */ 2213 lfs_release_finfo(fs); 2214 lfs_writeseg(fs, fs->lfs_sp); 2215 2216 /* 2217 * Remove us from paging queue, since we've now written all our 2218 * pages. 2219 */ 2220 simple_lock(&fs->lfs_interlock); 2221 if (ip->i_flags & IN_PAGING) { 2222 ip->i_flags &= ~IN_PAGING; 2223 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2224 } 2225 simple_unlock(&fs->lfs_interlock); 2226 2227 /* 2228 * XXX - with the malloc/copy writeseg, the pages are freed by now 2229 * even if we don't wait (e.g. if we hold a nested lock). This 2230 * will not be true if we stop using malloc/copy. 2231 */ 2232 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2233 lfs_segunlock(fs); 2234 2235 /* 2236 * Wait for v_numoutput to drop to zero. The seglock should 2237 * take care of this, but there is a slight possibility that 2238 * aiodoned might not have got around to our buffers yet. 2239 */ 2240 if (sync) { 2241 s = splbio(); 2242 simple_lock(&global_v_numoutput_slock); 2243 while (vp->v_numoutput > 0) { 2244 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2245 " num %d\n", ip->i_number, vp->v_numoutput)); 2246 vp->v_flag |= VBWAIT; 2247 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 2248 &global_v_numoutput_slock); 2249 } 2250 simple_unlock(&global_v_numoutput_slock); 2251 splx(s); 2252 } 2253 return error; 2254 } 2255 2256 /* 2257 * Return the last logical file offset that should be written for this file 2258 * if we're doing a write that ends at "size". If writing, we need to know 2259 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2260 * to know about entire blocks. 2261 */ 2262 void 2263 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2264 { 2265 struct inode *ip = VTOI(vp); 2266 struct lfs *fs = ip->i_lfs; 2267 daddr_t olbn, nlbn; 2268 2269 olbn = lblkno(fs, ip->i_size); 2270 nlbn = lblkno(fs, size); 2271 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2272 *eobp = fragroundup(fs, size); 2273 } else { 2274 *eobp = blkroundup(fs, size); 2275 } 2276 } 2277 2278 #ifdef DEBUG 2279 void lfs_dump_vop(void *); 2280 2281 void 2282 lfs_dump_vop(void *v) 2283 { 2284 struct vop_putpages_args /* { 2285 struct vnode *a_vp; 2286 voff_t a_offlo; 2287 voff_t a_offhi; 2288 int a_flags; 2289 } */ *ap = v; 2290 2291 #ifdef DDB 2292 vfs_vnode_print(ap->a_vp, 0, printf); 2293 #endif 2294 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2295 } 2296 #endif 2297 2298 int 2299 lfs_mmap(void *v) 2300 { 2301 struct vop_mmap_args /* { 2302 const struct vnodeop_desc *a_desc; 2303 struct vnode *a_vp; 2304 int a_fflags; 2305 kauth_cred_t a_cred; 2306 struct lwp *a_l; 2307 } */ *ap = v; 2308 2309 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2310 return EOPNOTSUPP; 2311 return ufs_mmap(v); 2312 } 2313
Indexes created Tue Sep 30 17:09:57 GMT 2025