lfs_vnops.c revision 1.201.2.5
1 /* $NetBSD: lfs_vnops.c,v 1.201.2.5 2007/05/13 17:36:45 ad 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 1.201.2.5 2007/05/13 17:36:45 ad Exp $"); 71 72 #ifdef _KERNEL_OPT 73 #include "opt_compat_netbsd.h" 74 #endif 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/namei.h> 79 #include <sys/resourcevar.h> 80 #include <sys/kernel.h> 81 #include <sys/file.h> 82 #include <sys/stat.h> 83 #include <sys/buf.h> 84 #include <sys/proc.h> 85 #include <sys/mount.h> 86 #include <sys/vnode.h> 87 #include <sys/pool.h> 88 #include <sys/signalvar.h> 89 #include <sys/kauth.h> 90 #include <sys/syslog.h> 91 #include <sys/fstrans.h> 92 93 #include <miscfs/fifofs/fifo.h> 94 #include <miscfs/genfs/genfs.h> 95 #include <miscfs/specfs/specdev.h> 96 97 #include <ufs/ufs/inode.h> 98 #include <ufs/ufs/dir.h> 99 #include <ufs/ufs/ufsmount.h> 100 #include <ufs/ufs/ufs_extern.h> 101 102 #include <uvm/uvm.h> 103 #include <uvm/uvm_pmap.h> 104 #include <uvm/uvm_stat.h> 105 #include <uvm/uvm_pager.h> 106 107 #include <ufs/lfs/lfs.h> 108 #include <ufs/lfs/lfs_extern.h> 109 110 extern pid_t lfs_writer_daemon; 111 112 /* Global vfs data structures for lfs. */ 113 int (**lfs_vnodeop_p)(void *); 114 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 115 { &vop_default_desc, vn_default_error }, 116 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 117 { &vop_create_desc, lfs_create }, /* create */ 118 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 119 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 120 { &vop_open_desc, ufs_open }, /* open */ 121 { &vop_close_desc, lfs_close }, /* close */ 122 { &vop_access_desc, ufs_access }, /* access */ 123 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 124 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 125 { &vop_read_desc, lfs_read }, /* read */ 126 { &vop_write_desc, lfs_write }, /* write */ 127 { &vop_lease_desc, ufs_lease_check }, /* lease */ 128 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 129 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 130 { &vop_poll_desc, ufs_poll }, /* poll */ 131 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 132 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 133 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 134 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 135 { &vop_seek_desc, ufs_seek }, /* seek */ 136 { &vop_remove_desc, lfs_remove }, /* remove */ 137 { &vop_link_desc, lfs_link }, /* link */ 138 { &vop_rename_desc, lfs_rename }, /* rename */ 139 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 140 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 141 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 142 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 143 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 144 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 145 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 146 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 147 { &vop_lock_desc, ufs_lock }, /* lock */ 148 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 149 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 150 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 151 { &vop_print_desc, ufs_print }, /* print */ 152 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 153 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 154 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 155 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 156 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 157 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 158 { NULL, NULL } 159 }; 160 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 161 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 162 163 int (**lfs_specop_p)(void *); 164 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 165 { &vop_default_desc, vn_default_error }, 166 { &vop_lookup_desc, spec_lookup }, /* lookup */ 167 { &vop_create_desc, spec_create }, /* create */ 168 { &vop_mknod_desc, spec_mknod }, /* mknod */ 169 { &vop_open_desc, spec_open }, /* open */ 170 { &vop_close_desc, lfsspec_close }, /* close */ 171 { &vop_access_desc, ufs_access }, /* access */ 172 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 173 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 174 { &vop_read_desc, ufsspec_read }, /* read */ 175 { &vop_write_desc, ufsspec_write }, /* write */ 176 { &vop_lease_desc, spec_lease_check }, /* lease */ 177 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 178 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 179 { &vop_poll_desc, spec_poll }, /* poll */ 180 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 181 { &vop_revoke_desc, spec_revoke }, /* revoke */ 182 { &vop_mmap_desc, spec_mmap }, /* mmap */ 183 { &vop_fsync_desc, spec_fsync }, /* fsync */ 184 { &vop_seek_desc, spec_seek }, /* seek */ 185 { &vop_remove_desc, spec_remove }, /* remove */ 186 { &vop_link_desc, spec_link }, /* link */ 187 { &vop_rename_desc, spec_rename }, /* rename */ 188 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 189 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 190 { &vop_symlink_desc, spec_symlink }, /* symlink */ 191 { &vop_readdir_desc, spec_readdir }, /* readdir */ 192 { &vop_readlink_desc, spec_readlink }, /* readlink */ 193 { &vop_abortop_desc, spec_abortop }, /* abortop */ 194 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 195 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 196 { &vop_lock_desc, ufs_lock }, /* lock */ 197 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 198 { &vop_bmap_desc, spec_bmap }, /* bmap */ 199 { &vop_strategy_desc, spec_strategy }, /* strategy */ 200 { &vop_print_desc, ufs_print }, /* print */ 201 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 202 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 203 { &vop_advlock_desc, spec_advlock }, /* advlock */ 204 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 205 { &vop_getpages_desc, spec_getpages }, /* getpages */ 206 { &vop_putpages_desc, spec_putpages }, /* putpages */ 207 { NULL, NULL } 208 }; 209 const struct vnodeopv_desc lfs_specop_opv_desc = 210 { &lfs_specop_p, lfs_specop_entries }; 211 212 int (**lfs_fifoop_p)(void *); 213 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 214 { &vop_default_desc, vn_default_error }, 215 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 216 { &vop_create_desc, fifo_create }, /* create */ 217 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 218 { &vop_open_desc, fifo_open }, /* open */ 219 { &vop_close_desc, lfsfifo_close }, /* close */ 220 { &vop_access_desc, ufs_access }, /* access */ 221 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 222 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 223 { &vop_read_desc, ufsfifo_read }, /* read */ 224 { &vop_write_desc, ufsfifo_write }, /* write */ 225 { &vop_lease_desc, fifo_lease_check }, /* lease */ 226 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 227 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 228 { &vop_poll_desc, fifo_poll }, /* poll */ 229 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 230 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 231 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 232 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 233 { &vop_seek_desc, fifo_seek }, /* seek */ 234 { &vop_remove_desc, fifo_remove }, /* remove */ 235 { &vop_link_desc, fifo_link }, /* link */ 236 { &vop_rename_desc, fifo_rename }, /* rename */ 237 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 238 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 239 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 240 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 241 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 242 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 243 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 244 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 245 { &vop_lock_desc, ufs_lock }, /* lock */ 246 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 247 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 248 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 249 { &vop_print_desc, ufs_print }, /* print */ 250 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 251 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 252 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 253 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 254 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 255 { NULL, NULL } 256 }; 257 const struct vnodeopv_desc lfs_fifoop_opv_desc = 258 { &lfs_fifoop_p, lfs_fifoop_entries }; 259 260 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int); 261 262 #define LFS_READWRITE 263 #include <ufs/ufs/ufs_readwrite.c> 264 #undef LFS_READWRITE 265 266 /* 267 * Synch an open file. 268 */ 269 /* ARGSUSED */ 270 int 271 lfs_fsync(void *v) 272 { 273 struct vop_fsync_args /* { 274 struct vnode *a_vp; 275 kauth_cred_t a_cred; 276 int a_flags; 277 off_t offlo; 278 off_t offhi; 279 struct lwp *a_l; 280 } */ *ap = v; 281 struct vnode *vp = ap->a_vp; 282 int error, wait; 283 284 /* If we're mounted read-only, don't try to sync. */ 285 if (VTOI(vp)->i_lfs->lfs_ronly) 286 return 0; 287 288 /* 289 * Trickle sync checks for need to do a checkpoint after possible 290 * activity from the pagedaemon. 291 */ 292 if (ap->a_flags & FSYNC_LAZY) { 293 mutex_enter(&lfs_subsys_lock); 294 wakeup(&lfs_writer_daemon); 295 mutex_exit(&lfs_subsys_lock); 296 return 0; 297 } 298 299 /* 300 * If a vnode is bring cleaned, flush it out before we try to 301 * reuse it. This prevents the cleaner from writing files twice 302 * in the same partial segment, causing an accounting underflow. 303 */ 304 if (ap->a_flags & FSYNC_RECLAIM && VTOI(vp)->i_flags & IN_CLEANING) { 305 /* printf("avoiding VONWORKLIST panic\n"); */ 306 lfs_vflush(vp); 307 } 308 309 wait = (ap->a_flags & FSYNC_WAIT); 310 mutex_enter(&vp->v_interlock); 311 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 312 round_page(ap->a_offhi), 313 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 314 if (error) 315 return error; 316 error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 317 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 318 int l = 0; 319 error = VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE, 320 ap->a_l->l_cred, ap->a_l); 321 } 322 if (wait && !VPISEMPTY(vp)) { 323 mutex_enter(&VTOI(vp)->i_lfs->lfs_interlock); 324 LFS_SET_UINO(VTOI(vp), IN_MODIFIED); 325 mutex_exit(&VTOI(vp)->i_lfs->lfs_interlock); 326 } 327 328 return error; 329 } 330 331 /* 332 * Take IN_ADIROP off, then call ufs_inactive. 333 */ 334 int 335 lfs_inactive(void *v) 336 { 337 struct vop_inactive_args /* { 338 struct vnode *a_vp; 339 struct lwp *a_l; 340 } */ *ap = v; 341 342 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 343 344 lfs_unmark_vnode(ap->a_vp); 345 346 /* 347 * The Ifile is only ever inactivated on unmount. 348 * Streamline this process by not giving it more dirty blocks. 349 */ 350 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 351 mutex_enter(&VTOI(ap->a_vp)->i_lfs->lfs_interlock); 352 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 353 mutex_exit(&VTOI(ap->a_vp)->i_lfs->lfs_interlock); 354 VOP_UNLOCK(ap->a_vp, 0); 355 return 0; 356 } 357 358 return ufs_inactive(v); 359 } 360 361 /* 362 * These macros are used to bracket UFS directory ops, so that we can 363 * identify all the pages touched during directory ops which need to 364 * be ordered and flushed atomically, so that they may be recovered. 365 * 366 * Because we have to mark nodes VDIROP in order to prevent 367 * the cache from reclaiming them while a dirop is in progress, we must 368 * also manage the number of nodes so marked (otherwise we can run out). 369 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 370 * is decremented during segment write, when VDIROP is taken off. 371 */ 372 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 373 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 374 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 375 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 376 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 377 static int lfs_set_dirop(struct vnode *, struct vnode *); 378 379 static int 380 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 381 { 382 struct lfs *fs; 383 int error; 384 385 KASSERT(VOP_ISLOCKED(dvp)); 386 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 387 388 fs = VTOI(dvp)->i_lfs; 389 390 ASSERT_NO_SEGLOCK(fs); 391 /* 392 * LFS_NRESERVE calculates direct and indirect blocks as well 393 * as an inode block; an overestimate in most cases. 394 */ 395 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 396 return (error); 397 398 restart: 399 mutex_enter(&fs->lfs_interlock); 400 if (fs->lfs_dirops == 0) { 401 mutex_exit(&fs->lfs_interlock); 402 lfs_check(dvp, LFS_UNUSED_LBN, 0); 403 mutex_enter(&fs->lfs_interlock); 404 } 405 while (fs->lfs_writer) { 406 error = mtsleep(&fs->lfs_dirops, (PRIBIO + 1) | PCATCH, 407 "lfs_sdirop", 0, &fs->lfs_interlock); 408 if (error == EINTR) { 409 mutex_exit(&fs->lfs_interlock); 410 goto unreserve; 411 } 412 } 413 mutex_enter(&lfs_subsys_lock); 414 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 415 wakeup(&lfs_writer_daemon); 416 mutex_exit(&lfs_subsys_lock); 417 mutex_exit(&fs->lfs_interlock); 418 preempt(); 419 goto restart; 420 } 421 422 if (lfs_dirvcount > LFS_MAX_DIROP) { 423 mutex_exit(&fs->lfs_interlock); 424 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 425 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 426 if ((error = mtsleep(&lfs_dirvcount, 427 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 428 &lfs_subsys_lock)) != 0) { 429 goto unreserve; 430 } 431 goto restart; 432 } 433 mutex_exit(&lfs_subsys_lock); 434 435 ++fs->lfs_dirops; 436 fs->lfs_doifile = 1; 437 mutex_exit(&fs->lfs_interlock); 438 439 /* Hold a reference so SET_ENDOP will be happy */ 440 vref(dvp); 441 if (vp) { 442 vref(vp); 443 MARK_VNODE(vp); 444 } 445 446 MARK_VNODE(dvp); 447 return 0; 448 449 unreserve: 450 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 451 return error; 452 } 453 454 /* 455 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 456 * in getnewvnode(), if we have a stacked filesystem mounted on top 457 * of us. 458 * 459 * NB: this means we have to clear the new vnodes on error. Fortunately 460 * SET_ENDOP is there to do that for us. 461 */ 462 static int 463 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 464 { 465 int error; 466 struct lfs *fs; 467 468 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 469 ASSERT_NO_SEGLOCK(fs); 470 if (fs->lfs_ronly) 471 return EROFS; 472 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 473 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 474 dvp, error)); 475 return error; 476 } 477 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 478 if (vpp) { 479 ungetnewvnode(*vpp); 480 *vpp = NULL; 481 } 482 return error; 483 } 484 return 0; 485 } 486 487 #define SET_ENDOP_BASE(fs, dvp, str) \ 488 do { \ 489 mutex_enter(&(fs)->lfs_interlock); \ 490 --(fs)->lfs_dirops; \ 491 if (!(fs)->lfs_dirops) { \ 492 if ((fs)->lfs_nadirop) { \ 493 panic("SET_ENDOP: %s: no dirops but " \ 494 " nadirop=%d", (str), \ 495 (fs)->lfs_nadirop); \ 496 } \ 497 wakeup(&(fs)->lfs_writer); \ 498 mutex_exit(&(fs)->lfs_interlock); \ 499 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 500 } else \ 501 mutex_exit(&(fs)->lfs_interlock); \ 502 } while(0) 503 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 504 do { \ 505 UNMARK_VNODE(dvp); \ 506 if (nvpp && *nvpp) \ 507 UNMARK_VNODE(*nvpp); \ 508 /* Check for error return to stem vnode leakage */ \ 509 if (nvpp && *nvpp && !((*nvpp)->v_flag & VDIROP)) \ 510 ungetnewvnode(*(nvpp)); \ 511 SET_ENDOP_BASE((fs), (dvp), (str)); \ 512 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 513 vrele(dvp); \ 514 } while(0) 515 #define SET_ENDOP_CREATE_AP(ap, str) \ 516 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 517 (ap)->a_vpp, (str)) 518 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 519 do { \ 520 UNMARK_VNODE(dvp); \ 521 if (ovp) \ 522 UNMARK_VNODE(ovp); \ 523 SET_ENDOP_BASE((fs), (dvp), (str)); \ 524 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 525 vrele(dvp); \ 526 if (ovp) \ 527 vrele(ovp); \ 528 } while(0) 529 530 void 531 lfs_mark_vnode(struct vnode *vp) 532 { 533 struct inode *ip = VTOI(vp); 534 struct lfs *fs = ip->i_lfs; 535 536 mutex_enter(&fs->lfs_interlock); 537 if (!(ip->i_flag & IN_ADIROP)) { 538 if (!(vp->v_flag & VDIROP)) { 539 (void)lfs_vref(vp); 540 mutex_enter(&lfs_subsys_lock); 541 ++lfs_dirvcount; 542 ++fs->lfs_dirvcount; 543 mutex_exit(&lfs_subsys_lock); 544 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 545 vp->v_flag |= VDIROP; 546 } 547 ++fs->lfs_nadirop; 548 ip->i_flag |= IN_ADIROP; 549 } else 550 KASSERT(vp->v_flag & VDIROP); 551 mutex_exit(&fs->lfs_interlock); 552 } 553 554 void 555 lfs_unmark_vnode(struct vnode *vp) 556 { 557 struct inode *ip = VTOI(vp); 558 559 if (ip && (ip->i_flag & IN_ADIROP)) { 560 KASSERT(vp->v_flag & VDIROP); 561 mutex_enter(&ip->i_lfs->lfs_interlock); 562 --ip->i_lfs->lfs_nadirop; 563 mutex_exit(&ip->i_lfs->lfs_interlock); 564 ip->i_flag &= ~IN_ADIROP; 565 } 566 } 567 568 int 569 lfs_symlink(void *v) 570 { 571 struct vop_symlink_args /* { 572 struct vnode *a_dvp; 573 struct vnode **a_vpp; 574 struct componentname *a_cnp; 575 struct vattr *a_vap; 576 char *a_target; 577 } */ *ap = v; 578 int error; 579 580 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 581 vput(ap->a_dvp); 582 return error; 583 } 584 error = ufs_symlink(ap); 585 SET_ENDOP_CREATE_AP(ap, "symlink"); 586 return (error); 587 } 588 589 int 590 lfs_mknod(void *v) 591 { 592 struct vop_mknod_args /* { 593 struct vnode *a_dvp; 594 struct vnode **a_vpp; 595 struct componentname *a_cnp; 596 struct vattr *a_vap; 597 } */ *ap = v; 598 struct vattr *vap = ap->a_vap; 599 struct vnode **vpp = ap->a_vpp; 600 struct inode *ip; 601 int error; 602 struct mount *mp; 603 ino_t ino; 604 605 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 606 vput(ap->a_dvp); 607 return error; 608 } 609 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 610 ap->a_dvp, vpp, ap->a_cnp); 611 612 /* Either way we're done with the dirop at this point */ 613 SET_ENDOP_CREATE_AP(ap, "mknod"); 614 615 if (error) 616 return (error); 617 618 ip = VTOI(*vpp); 619 mp = (*vpp)->v_mount; 620 ino = ip->i_number; 621 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 622 if (vap->va_rdev != VNOVAL) { 623 /* 624 * Want to be able to use this to make badblock 625 * inodes, so don't truncate the dev number. 626 */ 627 #if 0 628 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 629 UFS_MPNEEDSWAP((*vpp)->v_mount)); 630 #else 631 ip->i_ffs1_rdev = vap->va_rdev; 632 #endif 633 } 634 635 /* 636 * Call fsync to write the vnode so that we don't have to deal with 637 * flushing it when it's marked VDIROP|VXLOCK. 638 * 639 * XXX KS - If we can't flush we also can't call vgone(), so must 640 * return. But, that leaves this vnode in limbo, also not good. 641 * Can this ever happen (barring hardware failure)? 642 */ 643 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 644 curlwp)) != 0) { 645 panic("lfs_mknod: couldn't fsync (ino %llu)", 646 (unsigned long long)ino); 647 /* return (error); */ 648 } 649 /* 650 * Remove vnode so that it will be reloaded by VFS_VGET and 651 * checked to see if it is an alias of an existing entry in 652 * the inode cache. 653 */ 654 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 655 656 VOP_UNLOCK(*vpp, 0); 657 lfs_vunref(*vpp); 658 (*vpp)->v_type = VNON; 659 vgone(*vpp); 660 error = VFS_VGET(mp, ino, vpp); 661 662 if (error != 0) { 663 *vpp = NULL; 664 return (error); 665 } 666 return (0); 667 } 668 669 int 670 lfs_create(void *v) 671 { 672 struct vop_create_args /* { 673 struct vnode *a_dvp; 674 struct vnode **a_vpp; 675 struct componentname *a_cnp; 676 struct vattr *a_vap; 677 } */ *ap = v; 678 int error; 679 680 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 681 vput(ap->a_dvp); 682 return error; 683 } 684 error = ufs_create(ap); 685 SET_ENDOP_CREATE_AP(ap, "create"); 686 return (error); 687 } 688 689 int 690 lfs_mkdir(void *v) 691 { 692 struct vop_mkdir_args /* { 693 struct vnode *a_dvp; 694 struct vnode **a_vpp; 695 struct componentname *a_cnp; 696 struct vattr *a_vap; 697 } */ *ap = v; 698 int error; 699 700 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 701 vput(ap->a_dvp); 702 return error; 703 } 704 error = ufs_mkdir(ap); 705 SET_ENDOP_CREATE_AP(ap, "mkdir"); 706 return (error); 707 } 708 709 int 710 lfs_remove(void *v) 711 { 712 struct vop_remove_args /* { 713 struct vnode *a_dvp; 714 struct vnode *a_vp; 715 struct componentname *a_cnp; 716 } */ *ap = v; 717 struct vnode *dvp, *vp; 718 struct inode *ip; 719 int error; 720 721 dvp = ap->a_dvp; 722 vp = ap->a_vp; 723 ip = VTOI(vp); 724 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 725 if (dvp == vp) 726 vrele(vp); 727 else 728 vput(vp); 729 vput(dvp); 730 return error; 731 } 732 error = ufs_remove(ap); 733 if (ip->i_nlink == 0) 734 lfs_orphan(ip->i_lfs, ip->i_number); 735 SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove"); 736 return (error); 737 } 738 739 int 740 lfs_rmdir(void *v) 741 { 742 struct vop_rmdir_args /* { 743 struct vnodeop_desc *a_desc; 744 struct vnode *a_dvp; 745 struct vnode *a_vp; 746 struct componentname *a_cnp; 747 } */ *ap = v; 748 struct vnode *vp; 749 struct inode *ip; 750 int error; 751 752 vp = ap->a_vp; 753 ip = VTOI(vp); 754 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 755 if (ap->a_dvp == vp) 756 vrele(ap->a_dvp); 757 else 758 vput(ap->a_dvp); 759 vput(vp); 760 return error; 761 } 762 error = ufs_rmdir(ap); 763 if (ip->i_nlink == 0) 764 lfs_orphan(ip->i_lfs, ip->i_number); 765 SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 766 return (error); 767 } 768 769 int 770 lfs_link(void *v) 771 { 772 struct vop_link_args /* { 773 struct vnode *a_dvp; 774 struct vnode *a_vp; 775 struct componentname *a_cnp; 776 } */ *ap = v; 777 int error; 778 struct vnode **vpp = NULL; 779 780 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 781 vput(ap->a_dvp); 782 return error; 783 } 784 error = ufs_link(ap); 785 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 786 return (error); 787 } 788 789 int 790 lfs_rename(void *v) 791 { 792 struct vop_rename_args /* { 793 struct vnode *a_fdvp; 794 struct vnode *a_fvp; 795 struct componentname *a_fcnp; 796 struct vnode *a_tdvp; 797 struct vnode *a_tvp; 798 struct componentname *a_tcnp; 799 } */ *ap = v; 800 struct vnode *tvp, *fvp, *tdvp, *fdvp; 801 struct componentname *tcnp, *fcnp; 802 int error; 803 struct lfs *fs; 804 805 fs = VTOI(ap->a_fdvp)->i_lfs; 806 tvp = ap->a_tvp; 807 tdvp = ap->a_tdvp; 808 tcnp = ap->a_tcnp; 809 fvp = ap->a_fvp; 810 fdvp = ap->a_fdvp; 811 fcnp = ap->a_fcnp; 812 813 /* 814 * Check for cross-device rename. 815 * If it is, we don't want to set dirops, just error out. 816 * (In particular note that MARK_VNODE(tdvp) will DTWT on 817 * a cross-device rename.) 818 * 819 * Copied from ufs_rename. 820 */ 821 if ((fvp->v_mount != tdvp->v_mount) || 822 (tvp && (fvp->v_mount != tvp->v_mount))) { 823 error = EXDEV; 824 goto errout; 825 } 826 827 /* 828 * Check to make sure we're not renaming a vnode onto itself 829 * (deleting a hard link by renaming one name onto another); 830 * if we are we can't recursively call VOP_REMOVE since that 831 * would leave us with an unaccounted-for number of live dirops. 832 * 833 * Inline the relevant section of ufs_rename here, *before* 834 * calling SET_DIROP_REMOVE. 835 */ 836 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 837 (VTOI(tdvp)->i_flags & APPEND))) { 838 error = EPERM; 839 goto errout; 840 } 841 if (fvp == tvp) { 842 if (fvp->v_type == VDIR) { 843 error = EINVAL; 844 goto errout; 845 } 846 847 /* Release destination completely. */ 848 VOP_ABORTOP(tdvp, tcnp); 849 vput(tdvp); 850 vput(tvp); 851 852 /* Delete source. */ 853 vrele(fvp); 854 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 855 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 856 fcnp->cn_nameiop = DELETE; 857 vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY); 858 if ((error = relookup(fdvp, &fvp, fcnp))) { 859 vput(fdvp); 860 return (error); 861 } 862 return (VOP_REMOVE(fdvp, fvp, fcnp)); 863 } 864 865 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 866 goto errout; 867 MARK_VNODE(fdvp); 868 MARK_VNODE(fvp); 869 870 error = ufs_rename(ap); 871 UNMARK_VNODE(fdvp); 872 UNMARK_VNODE(fvp); 873 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 874 return (error); 875 876 errout: 877 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 878 if (tdvp == tvp) 879 vrele(tdvp); 880 else 881 vput(tdvp); 882 if (tvp) 883 vput(tvp); 884 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 885 vrele(fdvp); 886 vrele(fvp); 887 return (error); 888 } 889 890 /* XXX hack to avoid calling ITIMES in getattr */ 891 int 892 lfs_getattr(void *v) 893 { 894 struct vop_getattr_args /* { 895 struct vnode *a_vp; 896 struct vattr *a_vap; 897 kauth_cred_t a_cred; 898 struct lwp *a_l; 899 } */ *ap = v; 900 struct vnode *vp = ap->a_vp; 901 struct inode *ip = VTOI(vp); 902 struct vattr *vap = ap->a_vap; 903 struct lfs *fs = ip->i_lfs; 904 /* 905 * Copy from inode table 906 */ 907 vap->va_fsid = ip->i_dev; 908 vap->va_fileid = ip->i_number; 909 vap->va_mode = ip->i_mode & ~IFMT; 910 vap->va_nlink = ip->i_nlink; 911 vap->va_uid = ip->i_uid; 912 vap->va_gid = ip->i_gid; 913 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 914 vap->va_size = vp->v_size; 915 vap->va_atime.tv_sec = ip->i_ffs1_atime; 916 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 917 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 918 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 919 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 920 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 921 vap->va_flags = ip->i_flags; 922 vap->va_gen = ip->i_gen; 923 /* this doesn't belong here */ 924 if (vp->v_type == VBLK) 925 vap->va_blocksize = BLKDEV_IOSIZE; 926 else if (vp->v_type == VCHR) 927 vap->va_blocksize = MAXBSIZE; 928 else 929 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 930 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 931 vap->va_type = vp->v_type; 932 vap->va_filerev = ip->i_modrev; 933 return (0); 934 } 935 936 /* 937 * Check to make sure the inode blocks won't choke the buffer 938 * cache, then call ufs_setattr as usual. 939 */ 940 int 941 lfs_setattr(void *v) 942 { 943 struct vop_setattr_args /* { 944 struct vnode *a_vp; 945 struct vattr *a_vap; 946 kauth_cred_t a_cred; 947 struct lwp *a_l; 948 } */ *ap = v; 949 struct vnode *vp = ap->a_vp; 950 951 lfs_check(vp, LFS_UNUSED_LBN, 0); 952 return ufs_setattr(v); 953 } 954 955 /* 956 * Release the block we hold on lfs_newseg wrapping. Called on file close, 957 * or explicitly from LFCNWRAPGO. Called with the interlock held. 958 */ 959 static int 960 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor) 961 { 962 if (fs->lfs_stoplwp != curlwp) 963 return EBUSY; 964 965 fs->lfs_stoplwp = NULL; 966 cv_signal(&fs->lfs_stopcv); 967 968 KASSERT(fs->lfs_nowrap > 0); 969 if (fs->lfs_nowrap <= 0) { 970 return 0; 971 } 972 973 if (--fs->lfs_nowrap == 0) { 974 log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt); 975 wakeup(&fs->lfs_wrappass); 976 lfs_wakeup_cleaner(fs); 977 } 978 if (waitfor) { 979 mtsleep(&fs->lfs_nextseg, PCATCH | PUSER, 980 "segment", 0, &fs->lfs_interlock); 981 } 982 983 return 0; 984 } 985 986 /* 987 * Close called 988 */ 989 /* ARGSUSED */ 990 int 991 lfs_close(void *v) 992 { 993 struct vop_close_args /* { 994 struct vnode *a_vp; 995 int a_fflag; 996 kauth_cred_t a_cred; 997 struct lwp *a_l; 998 } */ *ap = v; 999 struct vnode *vp = ap->a_vp; 1000 struct inode *ip = VTOI(vp); 1001 struct lfs *fs = ip->i_lfs; 1002 1003 if ((ip->i_number == ROOTINO || ip->i_number == LFS_IFILE_INUM) && 1004 fs->lfs_stoplwp == curlwp) { 1005 mutex_enter(&fs->lfs_interlock); 1006 log(LOG_NOTICE, "lfs_close: releasing log wrap control\n"); 1007 lfs_wrapgo(fs, ip, 0); 1008 mutex_exit(&fs->lfs_interlock); 1009 } 1010 1011 if (vp == ip->i_lfs->lfs_ivnode && 1012 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 1013 return 0; 1014 1015 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 1016 LFS_ITIMES(ip, NULL, NULL, NULL); 1017 } 1018 return (0); 1019 } 1020 1021 /* 1022 * Close wrapper for special devices. 1023 * 1024 * Update the times on the inode then do device close. 1025 */ 1026 int 1027 lfsspec_close(void *v) 1028 { 1029 struct vop_close_args /* { 1030 struct vnode *a_vp; 1031 int a_fflag; 1032 kauth_cred_t a_cred; 1033 struct lwp *a_l; 1034 } */ *ap = v; 1035 struct vnode *vp; 1036 struct inode *ip; 1037 1038 vp = ap->a_vp; 1039 ip = VTOI(vp); 1040 if (vp->v_usecount > 1) { 1041 LFS_ITIMES(ip, NULL, NULL, NULL); 1042 } 1043 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1044 } 1045 1046 /* 1047 * Close wrapper for fifo's. 1048 * 1049 * Update the times on the inode then do device close. 1050 */ 1051 int 1052 lfsfifo_close(void *v) 1053 { 1054 struct vop_close_args /* { 1055 struct vnode *a_vp; 1056 int a_fflag; 1057 kauth_cred_ a_cred; 1058 struct lwp *a_l; 1059 } */ *ap = v; 1060 struct vnode *vp; 1061 struct inode *ip; 1062 1063 vp = ap->a_vp; 1064 ip = VTOI(vp); 1065 if (ap->a_vp->v_usecount > 1) { 1066 LFS_ITIMES(ip, NULL, NULL, NULL); 1067 } 1068 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1069 } 1070 1071 /* 1072 * Reclaim an inode so that it can be used for other purposes. 1073 */ 1074 1075 int 1076 lfs_reclaim(void *v) 1077 { 1078 struct vop_reclaim_args /* { 1079 struct vnode *a_vp; 1080 struct lwp *a_l; 1081 } */ *ap = v; 1082 struct vnode *vp = ap->a_vp; 1083 struct inode *ip = VTOI(vp); 1084 int error; 1085 1086 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1087 1088 mutex_enter(&ip->i_lfs->lfs_interlock); 1089 LFS_CLR_UINO(ip, IN_ALLMOD); 1090 mutex_exit(&ip->i_lfs->lfs_interlock); 1091 if ((error = ufs_reclaim(vp, ap->a_l))) 1092 return (error); 1093 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1094 lfs_deregister_all(vp); 1095 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1096 ip->inode_ext.lfs = NULL; 1097 genfs_node_destroy(vp); 1098 pool_put(&lfs_inode_pool, vp->v_data); 1099 vp->v_data = NULL; 1100 return (0); 1101 } 1102 1103 /* 1104 * Read a block from a storage device. 1105 * In order to avoid reading blocks that are in the process of being 1106 * written by the cleaner---and hence are not mutexed by the normal 1107 * buffer cache / page cache mechanisms---check for collisions before 1108 * reading. 1109 * 1110 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1111 * the active cleaner test. 1112 * 1113 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1114 */ 1115 int 1116 lfs_strategy(void *v) 1117 { 1118 struct vop_strategy_args /* { 1119 struct vnode *a_vp; 1120 struct buf *a_bp; 1121 } */ *ap = v; 1122 struct buf *bp; 1123 struct lfs *fs; 1124 struct vnode *vp; 1125 struct inode *ip; 1126 daddr_t tbn; 1127 int i, sn, error, slept; 1128 1129 bp = ap->a_bp; 1130 vp = ap->a_vp; 1131 ip = VTOI(vp); 1132 fs = ip->i_lfs; 1133 1134 /* lfs uses its strategy routine only for read */ 1135 KASSERT(bp->b_flags & B_READ); 1136 1137 if (vp->v_type == VBLK || vp->v_type == VCHR) 1138 panic("lfs_strategy: spec"); 1139 KASSERT(bp->b_bcount != 0); 1140 if (bp->b_blkno == bp->b_lblkno) { 1141 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1142 NULL); 1143 if (error) { 1144 biodone(bp, error, 0); 1145 return (error); 1146 } 1147 if ((long)bp->b_blkno == -1) /* no valid data */ 1148 clrbuf(bp); 1149 } 1150 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1151 biodone(bp, 0, bp->b_bcount); 1152 return (0); 1153 } 1154 1155 slept = 1; 1156 mutex_enter(&fs->lfs_interlock); 1157 while (slept && fs->lfs_seglock) { 1158 mutex_exit(&fs->lfs_interlock); 1159 /* 1160 * Look through list of intervals. 1161 * There will only be intervals to look through 1162 * if the cleaner holds the seglock. 1163 * Since the cleaner is synchronous, we can trust 1164 * the list of intervals to be current. 1165 */ 1166 tbn = dbtofsb(fs, bp->b_blkno); 1167 sn = dtosn(fs, tbn); 1168 slept = 0; 1169 for (i = 0; i < fs->lfs_cleanind; i++) { 1170 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1171 tbn >= fs->lfs_cleanint[i]) { 1172 DLOG((DLOG_CLEAN, 1173 "lfs_strategy: ino %d lbn %" PRId64 1174 " ind %d sn %d fsb %" PRIx32 1175 " given sn %d fsb %" PRIx64 "\n", 1176 ip->i_number, bp->b_lblkno, i, 1177 dtosn(fs, fs->lfs_cleanint[i]), 1178 fs->lfs_cleanint[i], sn, tbn)); 1179 DLOG((DLOG_CLEAN, 1180 "lfs_strategy: sleeping on ino %d lbn %" 1181 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1182 mutex_enter(&fs->lfs_interlock); 1183 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1184 /* Cleaner can't wait for itself */ 1185 mtsleep(&fs->lfs_iocount, 1186 (PRIBIO + 1) | PNORELOCK, 1187 "clean2", 0, 1188 &fs->lfs_interlock); 1189 slept = 1; 1190 break; 1191 } else if (fs->lfs_seglock) { 1192 mtsleep(&fs->lfs_seglock, 1193 (PRIBIO + 1) | PNORELOCK, 1194 "clean1", 0, 1195 &fs->lfs_interlock); 1196 slept = 1; 1197 break; 1198 } 1199 mutex_exit(&fs->lfs_interlock); 1200 } 1201 } 1202 mutex_enter(&fs->lfs_interlock); 1203 } 1204 mutex_exit(&fs->lfs_interlock); 1205 1206 vp = ip->i_devvp; 1207 VOP_STRATEGY(vp, bp); 1208 return (0); 1209 } 1210 1211 void 1212 lfs_flush_dirops(struct lfs *fs) 1213 { 1214 struct inode *ip, *nip; 1215 struct vnode *vp; 1216 extern int lfs_dostats; 1217 struct segment *sp; 1218 int waslocked; 1219 1220 ASSERT_MAYBE_SEGLOCK(fs); 1221 KASSERT(fs->lfs_nadirop == 0); 1222 1223 if (fs->lfs_ronly) 1224 return; 1225 1226 mutex_enter(&fs->lfs_interlock); 1227 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1228 mutex_exit(&fs->lfs_interlock); 1229 return; 1230 } else 1231 mutex_exit(&fs->lfs_interlock); 1232 1233 if (lfs_dostats) 1234 ++lfs_stats.flush_invoked; 1235 1236 /* 1237 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1238 * Technically this is a checkpoint (the on-disk state is valid) 1239 * even though we are leaving out all the file data. 1240 */ 1241 lfs_imtime(fs); 1242 lfs_seglock(fs, SEGM_CKP); 1243 sp = fs->lfs_sp; 1244 1245 /* 1246 * lfs_writevnodes, optimized to get dirops out of the way. 1247 * Only write dirops, and don't flush files' pages, only 1248 * blocks from the directories. 1249 * 1250 * We don't need to vref these files because they are 1251 * dirops and so hold an extra reference until the 1252 * segunlock clears them of that status. 1253 * 1254 * We don't need to check for IN_ADIROP because we know that 1255 * no dirops are active. 1256 * 1257 */ 1258 mutex_enter(&fs->lfs_interlock); 1259 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1260 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1261 mutex_exit(&fs->lfs_interlock); 1262 vp = ITOV(ip); 1263 1264 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1265 1266 /* 1267 * All writes to directories come from dirops; all 1268 * writes to files' direct blocks go through the page 1269 * cache, which we're not touching. Reads to files 1270 * and/or directories will not be affected by writing 1271 * directory blocks inodes and file inodes. So we don't 1272 * really need to lock. If we don't lock, though, 1273 * make sure that we don't clear IN_MODIFIED 1274 * unnecessarily. 1275 */ 1276 if (vp->v_flag & (VXLOCK | VFREEING)) { 1277 mutex_enter(&fs->lfs_interlock); 1278 continue; 1279 } 1280 waslocked = VOP_ISLOCKED(vp); 1281 if (vp->v_type != VREG && 1282 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1283 lfs_writefile(fs, sp, vp); 1284 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1285 !(ip->i_flag & IN_ALLMOD)) { 1286 mutex_enter(&fs->lfs_interlock); 1287 LFS_SET_UINO(ip, IN_MODIFIED); 1288 mutex_exit(&fs->lfs_interlock); 1289 } 1290 } 1291 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1292 (void) lfs_writeinode(fs, sp, ip); 1293 mutex_enter(&fs->lfs_interlock); 1294 if (waslocked == LK_EXCLOTHER) 1295 LFS_SET_UINO(ip, IN_MODIFIED); 1296 } 1297 mutex_exit(&fs->lfs_interlock); 1298 /* We've written all the dirops there are */ 1299 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1300 lfs_finalize_fs_seguse(fs); 1301 (void) lfs_writeseg(fs, sp); 1302 lfs_segunlock(fs); 1303 } 1304 1305 /* 1306 * Flush all vnodes for which the pagedaemon has requested pageouts. 1307 * Skip over any files that are marked VDIROP (since lfs_flush_dirop() 1308 * has just run, this would be an error). If we have to skip a vnode 1309 * for any reason, just skip it; if we have to wait for the cleaner, 1310 * abort. The writer daemon will call us again later. 1311 */ 1312 void 1313 lfs_flush_pchain(struct lfs *fs) 1314 { 1315 struct inode *ip, *nip; 1316 struct vnode *vp; 1317 extern int lfs_dostats; 1318 struct segment *sp; 1319 int error; 1320 1321 ASSERT_NO_SEGLOCK(fs); 1322 1323 if (fs->lfs_ronly) 1324 return; 1325 1326 mutex_enter(&fs->lfs_interlock); 1327 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1328 mutex_exit(&fs->lfs_interlock); 1329 return; 1330 } else 1331 mutex_exit(&fs->lfs_interlock); 1332 1333 /* Get dirops out of the way */ 1334 lfs_flush_dirops(fs); 1335 1336 if (lfs_dostats) 1337 ++lfs_stats.flush_invoked; 1338 1339 /* 1340 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1341 */ 1342 lfs_imtime(fs); 1343 lfs_seglock(fs, 0); 1344 sp = fs->lfs_sp; 1345 1346 /* 1347 * lfs_writevnodes, optimized to clear pageout requests. 1348 * Only write non-dirop files that are in the pageout queue. 1349 * We're very conservative about what we write; we want to be 1350 * fast and async. 1351 */ 1352 mutex_enter(&fs->lfs_interlock); 1353 top: 1354 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1355 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1356 vp = ITOV(ip); 1357 1358 if (!(ip->i_flags & IN_PAGING)) 1359 goto top; 1360 1361 if (vp->v_flag & (VXLOCK|VDIROP)) 1362 continue; 1363 if (vp->v_type != VREG) 1364 continue; 1365 if (lfs_vref(vp)) 1366 continue; 1367 mutex_exit(&fs->lfs_interlock); 1368 1369 if (VOP_ISLOCKED(vp)) { 1370 lfs_vunref(vp); 1371 mutex_enter(&fs->lfs_interlock); 1372 continue; 1373 } 1374 1375 error = lfs_writefile(fs, sp, vp); 1376 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1377 !(ip->i_flag & IN_ALLMOD)) { 1378 mutex_enter(&fs->lfs_interlock); 1379 LFS_SET_UINO(ip, IN_MODIFIED); 1380 mutex_exit(&fs->lfs_interlock); 1381 } 1382 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1383 (void) lfs_writeinode(fs, sp, ip); 1384 1385 lfs_vunref(vp); 1386 1387 if (error == EAGAIN) { 1388 lfs_writeseg(fs, sp); 1389 mutex_enter(&fs->lfs_interlock); 1390 break; 1391 } 1392 mutex_enter(&fs->lfs_interlock); 1393 } 1394 mutex_exit(&fs->lfs_interlock); 1395 (void) lfs_writeseg(fs, sp); 1396 lfs_segunlock(fs); 1397 } 1398 1399 /* 1400 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1401 */ 1402 int 1403 lfs_fcntl(void *v) 1404 { 1405 struct vop_fcntl_args /* { 1406 struct vnode *a_vp; 1407 u_long a_command; 1408 void * a_data; 1409 int a_fflag; 1410 kauth_cred_t a_cred; 1411 struct lwp *a_l; 1412 } */ *ap = v; 1413 struct timeval *tvp; 1414 BLOCK_INFO *blkiov; 1415 CLEANERINFO *cip; 1416 SEGUSE *sup; 1417 int blkcnt, error, oclean; 1418 size_t fh_size; 1419 struct lfs_fcntl_markv blkvp; 1420 struct lwp *l; 1421 fsid_t *fsidp; 1422 struct lfs *fs; 1423 struct buf *bp; 1424 fhandle_t *fhp; 1425 daddr_t off; 1426 1427 /* Only respect LFS fcntls on fs root or Ifile */ 1428 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1429 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1430 return ufs_fcntl(v); 1431 } 1432 1433 /* Avoid locking a draining lock */ 1434 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1435 return ESHUTDOWN; 1436 } 1437 1438 /* LFS control and monitoring fcntls are available only to root */ 1439 l = ap->a_l; 1440 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1441 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1442 NULL)) != 0) 1443 return (error); 1444 1445 fs = VTOI(ap->a_vp)->i_lfs; 1446 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1447 1448 error = 0; 1449 switch (ap->a_command) { 1450 case LFCNSEGWAITALL: 1451 case LFCNSEGWAITALL_COMPAT: 1452 fsidp = NULL; 1453 /* FALLSTHROUGH */ 1454 case LFCNSEGWAIT: 1455 case LFCNSEGWAIT_COMPAT: 1456 tvp = (struct timeval *)ap->a_data; 1457 mutex_enter(&fs->lfs_interlock); 1458 ++fs->lfs_sleepers; 1459 mutex_exit(&fs->lfs_interlock); 1460 1461 error = lfs_segwait(fsidp, tvp); 1462 1463 mutex_enter(&fs->lfs_interlock); 1464 if (--fs->lfs_sleepers == 0) 1465 wakeup(&fs->lfs_sleepers); 1466 mutex_exit(&fs->lfs_interlock); 1467 return error; 1468 1469 case LFCNBMAPV: 1470 case LFCNMARKV: 1471 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1472 1473 blkcnt = blkvp.blkcnt; 1474 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1475 return (EINVAL); 1476 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1477 if ((error = copyin(blkvp.blkiov, blkiov, 1478 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1479 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1480 return error; 1481 } 1482 1483 mutex_enter(&fs->lfs_interlock); 1484 ++fs->lfs_sleepers; 1485 mutex_exit(&fs->lfs_interlock); 1486 if (ap->a_command == LFCNBMAPV) 1487 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1488 else /* LFCNMARKV */ 1489 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1490 if (error == 0) 1491 error = copyout(blkiov, blkvp.blkiov, 1492 blkcnt * sizeof(BLOCK_INFO)); 1493 mutex_enter(&fs->lfs_interlock); 1494 if (--fs->lfs_sleepers == 0) 1495 wakeup(&fs->lfs_sleepers); 1496 mutex_exit(&fs->lfs_interlock); 1497 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1498 return error; 1499 1500 case LFCNRECLAIM: 1501 /* 1502 * Flush dirops and write Ifile, allowing empty segments 1503 * to be immediately reclaimed. 1504 */ 1505 lfs_writer_enter(fs, "pndirop"); 1506 off = fs->lfs_offset; 1507 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1508 lfs_flush_dirops(fs); 1509 LFS_CLEANERINFO(cip, fs, bp); 1510 oclean = cip->clean; 1511 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1512 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1513 fs->lfs_sp->seg_flags |= SEGM_PROT; 1514 lfs_segunlock(fs); 1515 lfs_writer_leave(fs); 1516 1517 #ifdef DEBUG 1518 LFS_CLEANERINFO(cip, fs, bp); 1519 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1520 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1521 fs->lfs_offset - off, cip->clean - oclean, 1522 fs->lfs_activesb)); 1523 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1524 #endif 1525 1526 return 0; 1527 1528 #ifdef COMPAT_30 1529 case LFCNIFILEFH_COMPAT: 1530 /* Return the filehandle of the Ifile */ 1531 if ((error = kauth_authorize_generic(l->l_cred, 1532 KAUTH_GENERIC_ISSUSER, NULL)) != 0) 1533 return (error); 1534 fhp = (struct fhandle *)ap->a_data; 1535 fhp->fh_fsid = *fsidp; 1536 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1537 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1538 #endif 1539 1540 case LFCNIFILEFH_COMPAT2: 1541 case LFCNIFILEFH: 1542 /* Return the filehandle of the Ifile */ 1543 fhp = (struct fhandle *)ap->a_data; 1544 fhp->fh_fsid = *fsidp; 1545 fh_size = sizeof(struct lfs_fhandle) - 1546 offsetof(fhandle_t, fh_fid); 1547 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1548 1549 case LFCNREWIND: 1550 /* Move lfs_offset to the lowest-numbered segment */ 1551 return lfs_rewind(fs, *(int *)ap->a_data); 1552 1553 case LFCNINVAL: 1554 /* Mark a segment SEGUSE_INVAL */ 1555 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1556 if (sup->su_nbytes > 0) { 1557 brelse(bp, 0); 1558 lfs_unset_inval_all(fs); 1559 return EBUSY; 1560 } 1561 sup->su_flags |= SEGUSE_INVAL; 1562 VOP_BWRITE(bp); 1563 return 0; 1564 1565 case LFCNRESIZE: 1566 /* Resize the filesystem */ 1567 return lfs_resize_fs(fs, *(int *)ap->a_data); 1568 1569 case LFCNWRAPSTOP: 1570 case LFCNWRAPSTOP_COMPAT: 1571 /* 1572 * Hold lfs_newseg at segment 0; if requested, sleep until 1573 * the filesystem wraps around. To support external agents 1574 * (dump, fsck-based regression test) that need to look at 1575 * a snapshot of the filesystem, without necessarily 1576 * requiring that all fs activity stops. 1577 */ 1578 if (fs->lfs_stoplwp == curlwp) 1579 return EALREADY; 1580 1581 mutex_enter(&fs->lfs_interlock); 1582 while (fs->lfs_stoplwp != NULL) 1583 cv_wait(&fs->lfs_stopcv, &fs->lfs_interlock); 1584 fs->lfs_stoplwp = curlwp; 1585 if (fs->lfs_nowrap == 0) 1586 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1587 ++fs->lfs_nowrap; 1588 if (*(int *)ap->a_data == 1 || 1589 ap->a_command == LFCNWRAPSTOP_COMPAT) { 1590 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1591 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1592 "segwrap", 0, &fs->lfs_interlock); 1593 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1594 if (error) { 1595 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1596 } 1597 } 1598 mutex_exit(&fs->lfs_interlock); 1599 return 0; 1600 1601 case LFCNWRAPGO: 1602 case LFCNWRAPGO_COMPAT: 1603 /* 1604 * Having done its work, the agent wakes up the writer. 1605 * If the argument is 1, it sleeps until a new segment 1606 * is selected. 1607 */ 1608 mutex_enter(&fs->lfs_interlock); 1609 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1610 (ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1611 *((int *)ap->a_data))); 1612 mutex_exit(&fs->lfs_interlock); 1613 return error; 1614 1615 case LFCNWRAPPASS: 1616 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1617 return EALREADY; 1618 mutex_enter(&fs->lfs_interlock); 1619 if (fs->lfs_stoplwp != curlwp) { 1620 mutex_exit(&fs->lfs_interlock); 1621 return EALREADY; 1622 } 1623 if (fs->lfs_nowrap == 0) { 1624 mutex_exit(&fs->lfs_interlock); 1625 return EBUSY; 1626 } 1627 fs->lfs_wrappass = 1; 1628 wakeup(&fs->lfs_wrappass); 1629 /* Wait for the log to wrap, if asked */ 1630 if (*(int *)ap->a_data) { 1631 lfs_vref(ap->a_vp); 1632 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1633 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1634 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1635 "segwrap", 0, &fs->lfs_interlock); 1636 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1637 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1638 lfs_vunref(ap->a_vp); 1639 } 1640 mutex_exit(&fs->lfs_interlock); 1641 return error; 1642 1643 case LFCNWRAPSTATUS: 1644 mutex_enter(&fs->lfs_interlock); 1645 *(int *)ap->a_data = fs->lfs_wrapstatus; 1646 mutex_exit(&fs->lfs_interlock); 1647 return 0; 1648 1649 default: 1650 return ufs_fcntl(v); 1651 } 1652 return 0; 1653 } 1654 1655 int 1656 lfs_getpages(void *v) 1657 { 1658 struct vop_getpages_args /* { 1659 struct vnode *a_vp; 1660 voff_t a_offset; 1661 struct vm_page **a_m; 1662 int *a_count; 1663 int a_centeridx; 1664 vm_prot_t a_access_type; 1665 int a_advice; 1666 int a_flags; 1667 } */ *ap = v; 1668 1669 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1670 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1671 return EPERM; 1672 } 1673 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1674 mutex_enter(&VTOI(ap->a_vp)->i_lfs->lfs_interlock); 1675 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1676 mutex_exit(&VTOI(ap->a_vp)->i_lfs->lfs_interlock); 1677 } 1678 1679 /* 1680 * we're relying on the fact that genfs_getpages() always read in 1681 * entire filesystem blocks. 1682 */ 1683 return genfs_getpages(v); 1684 } 1685 1686 /* 1687 * Make sure that for all pages in every block in the given range, 1688 * either all are dirty or all are clean. If any of the pages 1689 * we've seen so far are dirty, put the vnode on the paging chain, 1690 * and mark it IN_PAGING. 1691 * 1692 * If checkfirst != 0, don't check all the pages but return at the 1693 * first dirty page. 1694 */ 1695 static int 1696 check_dirty(struct lfs *fs, struct vnode *vp, 1697 off_t startoffset, off_t endoffset, off_t blkeof, 1698 int flags, int checkfirst) 1699 { 1700 int by_list; 1701 struct vm_page *curpg = NULL; /* XXX: gcc */ 1702 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1703 off_t soff = 0; /* XXX: gcc */ 1704 voff_t off; 1705 int i; 1706 int nonexistent; 1707 int any_dirty; /* number of dirty pages */ 1708 int dirty; /* number of dirty pages in a block */ 1709 int tdirty; 1710 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1711 int pagedaemon = (curlwp == uvm.pagedaemon_lwp); 1712 1713 ASSERT_MAYBE_SEGLOCK(fs); 1714 top: 1715 by_list = (vp->v_uobj.uo_npages <= 1716 ((endoffset - startoffset) >> PAGE_SHIFT) * 1717 UVM_PAGE_HASH_PENALTY); 1718 any_dirty = 0; 1719 1720 if (by_list) { 1721 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1722 } else { 1723 soff = startoffset; 1724 } 1725 while (by_list || soff < MIN(blkeof, endoffset)) { 1726 if (by_list) { 1727 /* 1728 * Find the first page in a block. Skip 1729 * blocks outside our area of interest or beyond 1730 * the end of file. 1731 */ 1732 if (pages_per_block > 1) { 1733 while (curpg && 1734 ((curpg->offset & fs->lfs_bmask) || 1735 curpg->offset >= vp->v_size || 1736 curpg->offset >= endoffset)) 1737 curpg = TAILQ_NEXT(curpg, listq); 1738 } 1739 if (curpg == NULL) 1740 break; 1741 soff = curpg->offset; 1742 } 1743 1744 /* 1745 * Mark all pages in extended range busy; find out if any 1746 * of them are dirty. 1747 */ 1748 nonexistent = dirty = 0; 1749 for (i = 0; i == 0 || i < pages_per_block; i++) { 1750 if (by_list && pages_per_block <= 1) { 1751 pgs[i] = pg = curpg; 1752 } else { 1753 off = soff + (i << PAGE_SHIFT); 1754 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1755 if (pg == NULL) { 1756 ++nonexistent; 1757 continue; 1758 } 1759 } 1760 KASSERT(pg != NULL); 1761 1762 /* 1763 * If we're holding the segment lock, we can deadlock 1764 * against a process that has our page and is waiting 1765 * for the cleaner, while the cleaner waits for the 1766 * segment lock. Just bail in that case. 1767 */ 1768 if ((pg->flags & PG_BUSY) && 1769 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1770 if (by_list && i > 0) 1771 uvm_page_unbusy(pgs, i); 1772 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1773 return -1; 1774 } 1775 1776 while (pg->flags & PG_BUSY) { 1777 pg->flags |= PG_WANTED; 1778 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1779 "lfsput", 0); 1780 mutex_enter(&vp->v_interlock); 1781 if (by_list) { 1782 if (i > 0) 1783 uvm_page_unbusy(pgs, i); 1784 goto top; 1785 } 1786 } 1787 pg->flags |= PG_BUSY; 1788 UVM_PAGE_OWN(pg, "lfs_putpages"); 1789 1790 pmap_page_protect(pg, VM_PROT_NONE); 1791 tdirty = (pmap_clear_modify(pg) || 1792 (pg->flags & PG_CLEAN) == 0); 1793 dirty += tdirty; 1794 } 1795 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1796 if (by_list) { 1797 curpg = TAILQ_NEXT(curpg, listq); 1798 } else { 1799 soff += fs->lfs_bsize; 1800 } 1801 continue; 1802 } 1803 1804 any_dirty += dirty; 1805 KASSERT(nonexistent == 0); 1806 1807 /* 1808 * If any are dirty make all dirty; unbusy them, 1809 * but if we were asked to clean, wire them so that 1810 * the pagedaemon doesn't bother us about them while 1811 * they're on their way to disk. 1812 */ 1813 for (i = 0; i == 0 || i < pages_per_block; i++) { 1814 pg = pgs[i]; 1815 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1816 if (dirty) { 1817 pg->flags &= ~PG_CLEAN; 1818 if (flags & PGO_FREE) { 1819 /* 1820 * Wire the page so that 1821 * pdaemon doesn't see it again. 1822 */ 1823 mutex_enter(&uvm_pageqlock); 1824 uvm_pagewire(pg); 1825 mutex_exit(&uvm_pageqlock); 1826 1827 /* Suspended write flag */ 1828 pg->flags |= PG_DELWRI; 1829 } 1830 } 1831 if (pg->flags & PG_WANTED) 1832 wakeup(pg); 1833 pg->flags &= ~(PG_WANTED|PG_BUSY); 1834 UVM_PAGE_OWN(pg, NULL); 1835 } 1836 1837 if (checkfirst && any_dirty) 1838 break; 1839 1840 if (by_list) { 1841 curpg = TAILQ_NEXT(curpg, listq); 1842 } else { 1843 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1844 } 1845 } 1846 1847 return any_dirty; 1848 } 1849 1850 /* 1851 * lfs_putpages functions like genfs_putpages except that 1852 * 1853 * (1) It needs to bounds-check the incoming requests to ensure that 1854 * they are block-aligned; if they are not, expand the range and 1855 * do the right thing in case, e.g., the requested range is clean 1856 * but the expanded range is dirty. 1857 * 1858 * (2) It needs to explicitly send blocks to be written when it is done. 1859 * If VOP_PUTPAGES is called without the seglock held, we simply take 1860 * the seglock and let lfs_segunlock wait for us. 1861 * XXX There might be a bad situation if we have to flush a vnode while 1862 * XXX lfs_markv is in operation. As of this writing we panic in this 1863 * XXX case. 1864 * 1865 * Assumptions: 1866 * 1867 * (1) The caller does not hold any pages in this vnode busy. If it does, 1868 * there is a danger that when we expand the page range and busy the 1869 * pages we will deadlock. 1870 * 1871 * (2) We are called with vp->v_interlock held; we must return with it 1872 * released. 1873 * 1874 * (3) We don't absolutely have to free pages right away, provided that 1875 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1876 * us a request with PGO_FREE, we take the pages out of the paging 1877 * queue and wake up the writer, which will handle freeing them for us. 1878 * 1879 * We ensure that for any filesystem block, all pages for that 1880 * block are either resident or not, even if those pages are higher 1881 * than EOF; that means that we will be getting requests to free 1882 * "unused" pages above EOF all the time, and should ignore them. 1883 * 1884 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 1885 * into has been set up for us by lfs_writefile. If not, we will 1886 * have to handle allocating and/or freeing an finfo entry. 1887 * 1888 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1889 */ 1890 1891 int 1892 lfs_putpages(void *v) 1893 { 1894 int error; 1895 struct vop_putpages_args /* { 1896 struct vnode *a_vp; 1897 voff_t a_offlo; 1898 voff_t a_offhi; 1899 int a_flags; 1900 } */ *ap = v; 1901 struct vnode *vp; 1902 struct inode *ip; 1903 struct lfs *fs; 1904 struct segment *sp; 1905 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1906 off_t off, max_endoffset; 1907 int s; 1908 bool seglocked, sync, pagedaemon; 1909 struct vm_page *pg; 1910 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1911 1912 vp = ap->a_vp; 1913 ip = VTOI(vp); 1914 fs = ip->i_lfs; 1915 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1916 pagedaemon = (curlwp == uvm.pagedaemon_lwp); 1917 1918 /* Putpages does nothing for metadata. */ 1919 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1920 mutex_exit(&vp->v_interlock); 1921 return 0; 1922 } 1923 1924 /* 1925 * If there are no pages, don't do anything. 1926 */ 1927 if (vp->v_uobj.uo_npages == 0) { 1928 s = splbio(); 1929 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 1930 (vp->v_flag & VONWORKLST) && 1931 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 1932 vp->v_flag &= ~VWRITEMAPDIRTY; 1933 vn_syncer_remove_from_worklist(vp); 1934 } 1935 splx(s); 1936 mutex_exit(&vp->v_interlock); 1937 1938 /* Remove us from paging queue, if we were on it */ 1939 mutex_enter(&fs->lfs_interlock); 1940 if (ip->i_flags & IN_PAGING) { 1941 ip->i_flags &= ~IN_PAGING; 1942 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1943 } 1944 mutex_exit(&fs->lfs_interlock); 1945 return 0; 1946 } 1947 1948 blkeof = blkroundup(fs, ip->i_size); 1949 1950 /* 1951 * Ignore requests to free pages past EOF but in the same block 1952 * as EOF, unless the request is synchronous. (If the request is 1953 * sync, it comes from lfs_truncate.) 1954 * XXXUBC Make these pages look "active" so the pagedaemon won't 1955 * XXXUBC bother us with them again. 1956 */ 1957 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1958 origoffset = ap->a_offlo; 1959 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1960 pg = uvm_pagelookup(&vp->v_uobj, off); 1961 KASSERT(pg != NULL); 1962 while (pg->flags & PG_BUSY) { 1963 pg->flags |= PG_WANTED; 1964 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1965 "lfsput2", 0); 1966 mutex_enter(&vp->v_interlock); 1967 } 1968 mutex_enter(&uvm_pageqlock); 1969 uvm_pageactivate(pg); 1970 mutex_exit(&uvm_pageqlock); 1971 } 1972 ap->a_offlo = blkeof; 1973 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1974 mutex_exit(&vp->v_interlock); 1975 return 0; 1976 } 1977 } 1978 1979 /* 1980 * Extend page range to start and end at block boundaries. 1981 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1982 */ 1983 origoffset = ap->a_offlo; 1984 origendoffset = ap->a_offhi; 1985 startoffset = origoffset & ~(fs->lfs_bmask); 1986 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1987 << fs->lfs_bshift; 1988 1989 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1990 endoffset = max_endoffset; 1991 origendoffset = endoffset; 1992 } else { 1993 origendoffset = round_page(ap->a_offhi); 1994 endoffset = round_page(blkroundup(fs, origendoffset)); 1995 } 1996 1997 KASSERT(startoffset > 0 || endoffset >= startoffset); 1998 if (startoffset == endoffset) { 1999 /* Nothing to do, why were we called? */ 2000 mutex_exit(&vp->v_interlock); 2001 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 2002 PRId64 "\n", startoffset)); 2003 return 0; 2004 } 2005 2006 ap->a_offlo = startoffset; 2007 ap->a_offhi = endoffset; 2008 2009 if (!(ap->a_flags & PGO_CLEANIT)) 2010 return genfs_putpages(v); 2011 2012 /* 2013 * If there are more than one page per block, we don't want 2014 * to get caught locking them backwards; so set PGO_BUSYFAIL 2015 * to avoid deadlocks. 2016 */ 2017 ap->a_flags |= PGO_BUSYFAIL; 2018 2019 do { 2020 int r; 2021 2022 /* If no pages are dirty, we can just use genfs_putpages. */ 2023 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2024 ap->a_flags, 1); 2025 if (r < 0) { 2026 mutex_exit(&vp->v_interlock); 2027 return EDEADLK; 2028 } 2029 if (r > 0) 2030 break; 2031 2032 /* 2033 * Sometimes pages are dirtied between the time that 2034 * we check and the time we try to clean them. 2035 * Instruct lfs_gop_write to return EDEADLK in this case 2036 * so we can write them properly. 2037 */ 2038 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2039 r = genfs_putpages(v); 2040 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2041 if (r != EDEADLK) 2042 return r; 2043 2044 /* Start over. */ 2045 preempt(); 2046 mutex_enter(&vp->v_interlock); 2047 } while(1); 2048 2049 /* 2050 * Dirty and asked to clean. 2051 * 2052 * Pagedaemon can't actually write LFS pages; wake up 2053 * the writer to take care of that. The writer will 2054 * notice the pager inode queue and act on that. 2055 */ 2056 if (pagedaemon) { 2057 mutex_enter(&fs->lfs_interlock); 2058 if (!(ip->i_flags & IN_PAGING)) { 2059 ip->i_flags |= IN_PAGING; 2060 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2061 } 2062 mutex_enter(&lfs_subsys_lock); 2063 wakeup(&lfs_writer_daemon); 2064 mutex_exit(&lfs_subsys_lock); 2065 mutex_exit(&fs->lfs_interlock); 2066 mutex_exit(&vp->v_interlock); 2067 preempt(); 2068 return EWOULDBLOCK; 2069 } 2070 2071 /* 2072 * If this is a file created in a recent dirop, we can't flush its 2073 * inode until the dirop is complete. Drain dirops, then flush the 2074 * filesystem (taking care of any other pending dirops while we're 2075 * at it). 2076 */ 2077 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2078 (vp->v_flag & VDIROP)) { 2079 int locked; 2080 2081 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 2082 locked = (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 2083 mutex_exit(&vp->v_interlock); 2084 lfs_writer_enter(fs, "ppdirop"); 2085 if (locked) 2086 VOP_UNLOCK(vp, 0); 2087 2088 mutex_enter(&fs->lfs_interlock); 2089 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2090 mutex_exit(&fs->lfs_interlock); 2091 2092 mutex_enter(&vp->v_interlock); 2093 if (locked) { 2094 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2095 mutex_enter(&vp->v_interlock); 2096 } 2097 lfs_writer_leave(fs); 2098 2099 /* XXX the flush should have taken care of this one too! */ 2100 } 2101 2102 /* 2103 * This is it. We are going to write some pages. From here on 2104 * down it's all just mechanics. 2105 * 2106 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2107 */ 2108 ap->a_flags &= ~PGO_SYNCIO; 2109 2110 /* 2111 * If we've already got the seglock, flush the node and return. 2112 * The FIP has already been set up for us by lfs_writefile, 2113 * and FIP cleanup and lfs_updatemeta will also be done there, 2114 * unless genfs_putpages returns EDEADLK; then we must flush 2115 * what we have, and correct FIP and segment header accounting. 2116 */ 2117 get_seglock: 2118 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2119 if (!seglocked) { 2120 mutex_exit(&vp->v_interlock); 2121 /* 2122 * Take the seglock, because we are going to be writing pages. 2123 */ 2124 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2125 if (error != 0) 2126 return error; 2127 mutex_enter(&vp->v_interlock); 2128 } 2129 2130 /* 2131 * VOP_PUTPAGES should not be called while holding the seglock. 2132 * XXXUBC fix lfs_markv, or do this properly. 2133 */ 2134 #ifdef notyet 2135 KASSERT(fs->lfs_seglock == 1); 2136 #endif /* notyet */ 2137 2138 /* 2139 * We assume we're being called with sp->fip pointing at blank space. 2140 * Account for a new FIP in the segment header, and set sp->vp. 2141 * (This should duplicate the setup at the top of lfs_writefile().) 2142 */ 2143 sp = fs->lfs_sp; 2144 if (!seglocked) 2145 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2146 KASSERT(sp->vp == NULL); 2147 sp->vp = vp; 2148 2149 if (!seglocked) { 2150 if (vp->v_flag & VDIROP) 2151 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2152 } 2153 2154 /* 2155 * Loop through genfs_putpages until all pages are gathered. 2156 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2157 * Whenever we lose the interlock we have to rerun check_dirty, as 2158 * well. 2159 */ 2160 again: 2161 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2162 ap->a_flags, 0) < 0) { 2163 mutex_exit(&vp->v_interlock); 2164 sp->vp = NULL; 2165 if (!seglocked) { 2166 lfs_release_finfo(fs); 2167 lfs_segunlock(fs); 2168 } 2169 if (pagedaemon) 2170 return EDEADLK; 2171 /* else seglocked == 0 */ 2172 preempt(); 2173 mutex_enter(&vp->v_interlock); 2174 goto get_seglock; 2175 } 2176 2177 error = genfs_putpages(v); 2178 if (error == EDEADLK || error == EAGAIN) { 2179 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2180 " EDEADLK [2] ino %d off %x (seg %d)\n", 2181 ip->i_number, fs->lfs_offset, 2182 dtosn(fs, fs->lfs_offset))); 2183 /* If nothing to write, short-circuit */ 2184 if (sp->cbpp - sp->bpp > 1) { 2185 /* Write gathered pages */ 2186 lfs_updatemeta(sp); 2187 lfs_release_finfo(fs); 2188 (void) lfs_writeseg(fs, sp); 2189 2190 /* 2191 * Reinitialize brand new FIP and add us to it. 2192 */ 2193 KASSERT(sp->vp == vp); 2194 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2195 } 2196 2197 /* Give the write a chance to complete */ 2198 preempt(); 2199 2200 /* We've lost the interlock. Start over. */ 2201 if (error == EDEADLK) { 2202 mutex_enter(&vp->v_interlock); 2203 goto again; 2204 } 2205 } 2206 2207 KASSERT(sp->vp == vp); 2208 if (!seglocked) { 2209 sp->vp = NULL; 2210 2211 /* Write indirect blocks as well */ 2212 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2213 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2214 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2215 2216 KASSERT(sp->vp == NULL); 2217 sp->vp = vp; 2218 } 2219 2220 /* 2221 * Blocks are now gathered into a segment waiting to be written. 2222 * All that's left to do is update metadata, and write them. 2223 */ 2224 lfs_updatemeta(sp); 2225 KASSERT(sp->vp == vp); 2226 sp->vp = NULL; 2227 2228 if (seglocked) { 2229 /* we're called by lfs_writefile. */ 2230 return error; 2231 } 2232 2233 /* Clean up FIP and send it to disk. */ 2234 lfs_release_finfo(fs); 2235 lfs_writeseg(fs, fs->lfs_sp); 2236 2237 /* 2238 * Remove us from paging queue, since we've now written all our 2239 * pages. 2240 */ 2241 mutex_enter(&fs->lfs_interlock); 2242 if (ip->i_flags & IN_PAGING) { 2243 ip->i_flags &= ~IN_PAGING; 2244 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2245 } 2246 mutex_exit(&fs->lfs_interlock); 2247 2248 /* 2249 * XXX - with the malloc/copy writeseg, the pages are freed by now 2250 * even if we don't wait (e.g. if we hold a nested lock). This 2251 * will not be true if we stop using malloc/copy. 2252 */ 2253 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2254 lfs_segunlock(fs); 2255 2256 /* 2257 * Wait for v_numoutput to drop to zero. The seglock should 2258 * take care of this, but there is a slight possibility that 2259 * aiodoned might not have got around to our buffers yet. 2260 */ 2261 if (sync) { 2262 mutex_enter(&global_v_numoutput_lock); 2263 while (vp->v_numoutput > 0) { 2264 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2265 " num %d\n", ip->i_number, vp->v_numoutput)); 2266 cv_wait(&vp->v_outputcv, &global_v_numoutput_lock); 2267 } 2268 mutex_exit(&global_v_numoutput_lock); 2269 } 2270 return error; 2271 } 2272 2273 /* 2274 * Return the last logical file offset that should be written for this file 2275 * if we're doing a write that ends at "size". If writing, we need to know 2276 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2277 * to know about entire blocks. 2278 */ 2279 void 2280 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2281 { 2282 struct inode *ip = VTOI(vp); 2283 struct lfs *fs = ip->i_lfs; 2284 daddr_t olbn, nlbn; 2285 2286 olbn = lblkno(fs, ip->i_size); 2287 nlbn = lblkno(fs, size); 2288 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2289 *eobp = fragroundup(fs, size); 2290 } else { 2291 *eobp = blkroundup(fs, size); 2292 } 2293 } 2294 2295 #ifdef DEBUG 2296 void lfs_dump_vop(void *); 2297 2298 void 2299 lfs_dump_vop(void *v) 2300 { 2301 struct vop_putpages_args /* { 2302 struct vnode *a_vp; 2303 voff_t a_offlo; 2304 voff_t a_offhi; 2305 int a_flags; 2306 } */ *ap = v; 2307 2308 #ifdef DDB 2309 vfs_vnode_print(ap->a_vp, 0, printf); 2310 #endif 2311 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2312 } 2313 #endif 2314 2315 int 2316 lfs_mmap(void *v) 2317 { 2318 struct vop_mmap_args /* { 2319 const struct vnodeop_desc *a_desc; 2320 struct vnode *a_vp; 2321 int a_fflags; 2322 kauth_cred_t a_cred; 2323 struct lwp *a_l; 2324 } */ *ap = v; 2325 2326 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2327 return EOPNOTSUPP; 2328 return ufs_mmap(v); 2329 } 2330
Indexes created Mon Oct 20 10:10:01 GMT 2025