lfs_vnops.c revision 1.213.2.1
1 /* $NetBSD: lfs_vnops.c,v 1.213.2.1 2007/12/04 13:03:53 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 1.213.2.1 2007/12/04 13:03:53 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 int lfs_ignore_lazy_sync = 1; 112 113 /* Global vfs data structures for lfs. */ 114 int (**lfs_vnodeop_p)(void *); 115 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 116 { &vop_default_desc, vn_default_error }, 117 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 118 { &vop_create_desc, lfs_create }, /* create */ 119 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 120 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 121 { &vop_open_desc, ufs_open }, /* open */ 122 { &vop_close_desc, lfs_close }, /* close */ 123 { &vop_access_desc, ufs_access }, /* access */ 124 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 125 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 126 { &vop_read_desc, lfs_read }, /* read */ 127 { &vop_write_desc, lfs_write }, /* write */ 128 { &vop_lease_desc, ufs_lease_check }, /* lease */ 129 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 130 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 131 { &vop_poll_desc, ufs_poll }, /* poll */ 132 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 133 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 134 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 135 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 136 { &vop_seek_desc, ufs_seek }, /* seek */ 137 { &vop_remove_desc, lfs_remove }, /* remove */ 138 { &vop_link_desc, lfs_link }, /* link */ 139 { &vop_rename_desc, lfs_rename }, /* rename */ 140 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 141 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 142 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 143 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 144 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 145 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 146 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 147 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 148 { &vop_lock_desc, ufs_lock }, /* lock */ 149 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 150 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 151 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 152 { &vop_print_desc, ufs_print }, /* print */ 153 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 154 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 155 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 156 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 157 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 158 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 159 { NULL, NULL } 160 }; 161 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 162 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 163 164 int (**lfs_specop_p)(void *); 165 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 166 { &vop_default_desc, vn_default_error }, 167 { &vop_lookup_desc, spec_lookup }, /* lookup */ 168 { &vop_create_desc, spec_create }, /* create */ 169 { &vop_mknod_desc, spec_mknod }, /* mknod */ 170 { &vop_open_desc, spec_open }, /* open */ 171 { &vop_close_desc, lfsspec_close }, /* close */ 172 { &vop_access_desc, ufs_access }, /* access */ 173 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 174 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 175 { &vop_read_desc, ufsspec_read }, /* read */ 176 { &vop_write_desc, ufsspec_write }, /* write */ 177 { &vop_lease_desc, spec_lease_check }, /* lease */ 178 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 179 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 180 { &vop_poll_desc, spec_poll }, /* poll */ 181 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 182 { &vop_revoke_desc, spec_revoke }, /* revoke */ 183 { &vop_mmap_desc, spec_mmap }, /* mmap */ 184 { &vop_fsync_desc, spec_fsync }, /* fsync */ 185 { &vop_seek_desc, spec_seek }, /* seek */ 186 { &vop_remove_desc, spec_remove }, /* remove */ 187 { &vop_link_desc, spec_link }, /* link */ 188 { &vop_rename_desc, spec_rename }, /* rename */ 189 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 190 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 191 { &vop_symlink_desc, spec_symlink }, /* symlink */ 192 { &vop_readdir_desc, spec_readdir }, /* readdir */ 193 { &vop_readlink_desc, spec_readlink }, /* readlink */ 194 { &vop_abortop_desc, spec_abortop }, /* abortop */ 195 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 196 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 197 { &vop_lock_desc, ufs_lock }, /* lock */ 198 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 199 { &vop_bmap_desc, spec_bmap }, /* bmap */ 200 { &vop_strategy_desc, spec_strategy }, /* strategy */ 201 { &vop_print_desc, ufs_print }, /* print */ 202 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 203 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 204 { &vop_advlock_desc, spec_advlock }, /* advlock */ 205 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 206 { &vop_getpages_desc, spec_getpages }, /* getpages */ 207 { &vop_putpages_desc, spec_putpages }, /* putpages */ 208 { NULL, NULL } 209 }; 210 const struct vnodeopv_desc lfs_specop_opv_desc = 211 { &lfs_specop_p, lfs_specop_entries }; 212 213 int (**lfs_fifoop_p)(void *); 214 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 215 { &vop_default_desc, vn_default_error }, 216 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 217 { &vop_create_desc, fifo_create }, /* create */ 218 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 219 { &vop_open_desc, fifo_open }, /* open */ 220 { &vop_close_desc, lfsfifo_close }, /* close */ 221 { &vop_access_desc, ufs_access }, /* access */ 222 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 223 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 224 { &vop_read_desc, ufsfifo_read }, /* read */ 225 { &vop_write_desc, ufsfifo_write }, /* write */ 226 { &vop_lease_desc, fifo_lease_check }, /* lease */ 227 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 228 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 229 { &vop_poll_desc, fifo_poll }, /* poll */ 230 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 231 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 232 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 233 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 234 { &vop_seek_desc, fifo_seek }, /* seek */ 235 { &vop_remove_desc, fifo_remove }, /* remove */ 236 { &vop_link_desc, fifo_link }, /* link */ 237 { &vop_rename_desc, fifo_rename }, /* rename */ 238 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 239 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 240 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 241 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 242 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 243 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 244 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 245 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 246 { &vop_lock_desc, ufs_lock }, /* lock */ 247 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 248 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 249 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 250 { &vop_print_desc, ufs_print }, /* print */ 251 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 252 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 253 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 254 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 255 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 256 { NULL, NULL } 257 }; 258 const struct vnodeopv_desc lfs_fifoop_opv_desc = 259 { &lfs_fifoop_p, lfs_fifoop_entries }; 260 261 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int, struct vm_page **); 262 263 #define LFS_READWRITE 264 #include <ufs/ufs/ufs_readwrite.c> 265 #undef LFS_READWRITE 266 267 /* 268 * Synch an open file. 269 */ 270 /* ARGSUSED */ 271 int 272 lfs_fsync(void *v) 273 { 274 struct vop_fsync_args /* { 275 struct vnode *a_vp; 276 kauth_cred_t a_cred; 277 int a_flags; 278 off_t offlo; 279 off_t offhi; 280 } */ *ap = v; 281 struct vnode *vp = ap->a_vp; 282 int error, wait; 283 struct inode *ip = VTOI(vp); 284 struct lfs *fs = ip->i_lfs; 285 286 /* If we're mounted read-only, don't try to sync. */ 287 if (fs->lfs_ronly) 288 return 0; 289 290 /* 291 * Trickle sync simply adds this vnode to the pager list, as if 292 * the pagedaemon had requested a pageout. 293 */ 294 if (ap->a_flags & FSYNC_LAZY) { 295 if (lfs_ignore_lazy_sync == 0) { 296 mutex_enter(&fs->lfs_interlock); 297 if (!(ip->i_flags & IN_PAGING)) { 298 ip->i_flags |= IN_PAGING; 299 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, 300 i_lfs_pchain); 301 } 302 mutex_exit(&fs->lfs_interlock); 303 mutex_enter(&lfs_subsys_lock); 304 wakeup(&lfs_writer_daemon); 305 mutex_exit(&lfs_subsys_lock); 306 } 307 return 0; 308 } 309 310 /* 311 * If a vnode is bring cleaned, flush it out before we try to 312 * reuse it. This prevents the cleaner from writing files twice 313 * in the same partial segment, causing an accounting underflow. 314 */ 315 if (ap->a_flags & FSYNC_RECLAIM && ip->i_flags & IN_CLEANING) { 316 lfs_vflush(vp); 317 } 318 319 wait = (ap->a_flags & FSYNC_WAIT); 320 do { 321 mutex_enter(&vp->v_interlock); 322 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 323 round_page(ap->a_offhi), 324 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 325 if (error == EAGAIN) { 326 mutex_enter(&fs->lfs_interlock); 327 mtsleep(&fs->lfs_avail, PCATCH | PUSER, "lfs_fsync", 328 hz / 100 + 1, &fs->lfs_interlock); 329 mutex_exit(&fs->lfs_interlock); 330 } 331 } while (error == EAGAIN); 332 if (error) 333 return error; 334 335 if ((ap->a_flags & FSYNC_DATAONLY) == 0) 336 error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 337 338 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 339 int l = 0; 340 error = VOP_IOCTL(ip->i_devvp, DIOCCACHESYNC, &l, FWRITE, 341 curlwp->l_cred); 342 } 343 if (wait && !VPISEMPTY(vp)) 344 LFS_SET_UINO(ip, IN_MODIFIED); 345 346 return error; 347 } 348 349 /* 350 * Take IN_ADIROP off, then call ufs_inactive. 351 */ 352 int 353 lfs_inactive(void *v) 354 { 355 struct vop_inactive_args /* { 356 struct vnode *a_vp; 357 } */ *ap = v; 358 359 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 360 361 lfs_unmark_vnode(ap->a_vp); 362 363 /* 364 * The Ifile is only ever inactivated on unmount. 365 * Streamline this process by not giving it more dirty blocks. 366 */ 367 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 368 mutex_enter(&VTOI(ap->a_vp)->i_lfs->lfs_interlock); 369 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 370 mutex_exit(&VTOI(ap->a_vp)->i_lfs->lfs_interlock); 371 VOP_UNLOCK(ap->a_vp, 0); 372 return 0; 373 } 374 375 return ufs_inactive(v); 376 } 377 378 /* 379 * These macros are used to bracket UFS directory ops, so that we can 380 * identify all the pages touched during directory ops which need to 381 * be ordered and flushed atomically, so that they may be recovered. 382 * 383 * Because we have to mark nodes VDIROP in order to prevent 384 * the cache from reclaiming them while a dirop is in progress, we must 385 * also manage the number of nodes so marked (otherwise we can run out). 386 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 387 * is decremented during segment write, when VDIROP is taken off. 388 */ 389 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 390 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 391 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 392 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 393 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 394 static int lfs_set_dirop(struct vnode *, struct vnode *); 395 396 static int 397 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 398 { 399 struct lfs *fs; 400 int error; 401 402 KASSERT(VOP_ISLOCKED(dvp)); 403 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 404 405 fs = VTOI(dvp)->i_lfs; 406 407 ASSERT_NO_SEGLOCK(fs); 408 /* 409 * LFS_NRESERVE calculates direct and indirect blocks as well 410 * as an inode block; an overestimate in most cases. 411 */ 412 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 413 return (error); 414 415 restart: 416 mutex_enter(&fs->lfs_interlock); 417 if (fs->lfs_dirops == 0) { 418 mutex_exit(&fs->lfs_interlock); 419 lfs_check(dvp, LFS_UNUSED_LBN, 0); 420 mutex_enter(&fs->lfs_interlock); 421 } 422 while (fs->lfs_writer) { 423 error = mtsleep(&fs->lfs_dirops, (PRIBIO + 1) | PCATCH, 424 "lfs_sdirop", 0, &fs->lfs_interlock); 425 if (error == EINTR) { 426 mutex_exit(&fs->lfs_interlock); 427 goto unreserve; 428 } 429 } 430 mutex_enter(&lfs_subsys_lock); 431 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 432 wakeup(&lfs_writer_daemon); 433 mutex_exit(&lfs_subsys_lock); 434 mutex_exit(&fs->lfs_interlock); 435 preempt(); 436 goto restart; 437 } 438 439 if (lfs_dirvcount > LFS_MAX_DIROP) { 440 mutex_exit(&fs->lfs_interlock); 441 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 442 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 443 if ((error = mtsleep(&lfs_dirvcount, 444 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 445 &lfs_subsys_lock)) != 0) { 446 goto unreserve; 447 } 448 goto restart; 449 } 450 mutex_exit(&lfs_subsys_lock); 451 452 ++fs->lfs_dirops; 453 fs->lfs_doifile = 1; 454 mutex_exit(&fs->lfs_interlock); 455 456 /* Hold a reference so SET_ENDOP will be happy */ 457 vref(dvp); 458 if (vp) { 459 vref(vp); 460 MARK_VNODE(vp); 461 } 462 463 MARK_VNODE(dvp); 464 return 0; 465 466 unreserve: 467 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 468 return error; 469 } 470 471 /* 472 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 473 * in getnewvnode(), if we have a stacked filesystem mounted on top 474 * of us. 475 * 476 * NB: this means we have to clear the new vnodes on error. Fortunately 477 * SET_ENDOP is there to do that for us. 478 */ 479 static int 480 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 481 { 482 int error; 483 struct lfs *fs; 484 485 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 486 ASSERT_NO_SEGLOCK(fs); 487 if (fs->lfs_ronly) 488 return EROFS; 489 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 490 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 491 dvp, error)); 492 return error; 493 } 494 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 495 if (vpp) { 496 ungetnewvnode(*vpp); 497 *vpp = NULL; 498 } 499 return error; 500 } 501 return 0; 502 } 503 504 #define SET_ENDOP_BASE(fs, dvp, str) \ 505 do { \ 506 mutex_enter(&(fs)->lfs_interlock); \ 507 --(fs)->lfs_dirops; \ 508 if (!(fs)->lfs_dirops) { \ 509 if ((fs)->lfs_nadirop) { \ 510 panic("SET_ENDOP: %s: no dirops but " \ 511 " nadirop=%d", (str), \ 512 (fs)->lfs_nadirop); \ 513 } \ 514 wakeup(&(fs)->lfs_writer); \ 515 mutex_exit(&(fs)->lfs_interlock); \ 516 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 517 } else \ 518 mutex_exit(&(fs)->lfs_interlock); \ 519 } while(0) 520 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 521 do { \ 522 UNMARK_VNODE(dvp); \ 523 if (nvpp && *nvpp) \ 524 UNMARK_VNODE(*nvpp); \ 525 /* Check for error return to stem vnode leakage */ \ 526 if (nvpp && *nvpp && !((*nvpp)->v_uflag & VU_DIROP)) \ 527 ungetnewvnode(*(nvpp)); \ 528 SET_ENDOP_BASE((fs), (dvp), (str)); \ 529 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 530 vrele(dvp); \ 531 } while(0) 532 #define SET_ENDOP_CREATE_AP(ap, str) \ 533 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 534 (ap)->a_vpp, (str)) 535 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 536 do { \ 537 UNMARK_VNODE(dvp); \ 538 if (ovp) \ 539 UNMARK_VNODE(ovp); \ 540 SET_ENDOP_BASE((fs), (dvp), (str)); \ 541 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 542 vrele(dvp); \ 543 if (ovp) \ 544 vrele(ovp); \ 545 } while(0) 546 547 void 548 lfs_mark_vnode(struct vnode *vp) 549 { 550 struct inode *ip = VTOI(vp); 551 struct lfs *fs = ip->i_lfs; 552 553 mutex_enter(&fs->lfs_interlock); 554 if (!(ip->i_flag & IN_ADIROP)) { 555 if (!(vp->v_uflag & VU_DIROP)) { 556 (void)lfs_vref(vp); 557 mutex_enter(&lfs_subsys_lock); 558 ++lfs_dirvcount; 559 ++fs->lfs_dirvcount; 560 mutex_exit(&lfs_subsys_lock); 561 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 562 vp->v_uflag |= VU_DIROP; 563 } 564 ++fs->lfs_nadirop; 565 ip->i_flag |= IN_ADIROP; 566 } else 567 KASSERT(vp->v_uflag & VU_DIROP); 568 mutex_exit(&fs->lfs_interlock); 569 } 570 571 void 572 lfs_unmark_vnode(struct vnode *vp) 573 { 574 struct inode *ip = VTOI(vp); 575 576 if (ip && (ip->i_flag & IN_ADIROP)) { 577 KASSERT(vp->v_uflag & VU_DIROP); 578 mutex_enter(&ip->i_lfs->lfs_interlock); 579 --ip->i_lfs->lfs_nadirop; 580 mutex_exit(&ip->i_lfs->lfs_interlock); 581 ip->i_flag &= ~IN_ADIROP; 582 } 583 } 584 585 int 586 lfs_symlink(void *v) 587 { 588 struct vop_symlink_args /* { 589 struct vnode *a_dvp; 590 struct vnode **a_vpp; 591 struct componentname *a_cnp; 592 struct vattr *a_vap; 593 char *a_target; 594 } */ *ap = v; 595 int error; 596 597 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 598 vput(ap->a_dvp); 599 return error; 600 } 601 error = ufs_symlink(ap); 602 SET_ENDOP_CREATE_AP(ap, "symlink"); 603 return (error); 604 } 605 606 int 607 lfs_mknod(void *v) 608 { 609 struct vop_mknod_args /* { 610 struct vnode *a_dvp; 611 struct vnode **a_vpp; 612 struct componentname *a_cnp; 613 struct vattr *a_vap; 614 } */ *ap = v; 615 struct vattr *vap = ap->a_vap; 616 struct vnode **vpp = ap->a_vpp; 617 struct inode *ip; 618 int error; 619 struct mount *mp; 620 ino_t ino; 621 622 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 623 vput(ap->a_dvp); 624 return error; 625 } 626 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 627 ap->a_dvp, vpp, ap->a_cnp); 628 629 /* Either way we're done with the dirop at this point */ 630 SET_ENDOP_CREATE_AP(ap, "mknod"); 631 632 if (error) 633 return (error); 634 635 ip = VTOI(*vpp); 636 mp = (*vpp)->v_mount; 637 ino = ip->i_number; 638 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 639 if (vap->va_rdev != VNOVAL) { 640 /* 641 * Want to be able to use this to make badblock 642 * inodes, so don't truncate the dev number. 643 */ 644 #if 0 645 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 646 UFS_MPNEEDSWAP((*vpp)->v_mount)); 647 #else 648 ip->i_ffs1_rdev = vap->va_rdev; 649 #endif 650 } 651 652 /* 653 * Call fsync to write the vnode so that we don't have to deal with 654 * flushing it when it's marked VDIROP|VXLOCK. 655 * 656 * XXX KS - If we can't flush we also can't call vgone(), so must 657 * return. But, that leaves this vnode in limbo, also not good. 658 * Can this ever happen (barring hardware failure)? 659 */ 660 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0)) != 0) { 661 panic("lfs_mknod: couldn't fsync (ino %llu)", 662 (unsigned long long)ino); 663 /* return (error); */ 664 } 665 /* 666 * Remove vnode so that it will be reloaded by VFS_VGET and 667 * checked to see if it is an alias of an existing entry in 668 * the inode cache. 669 */ 670 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 671 672 VOP_UNLOCK(*vpp, 0); 673 lfs_vunref(*vpp); /* XXXAD */ 674 (*vpp)->v_type = VNON; 675 vgone(*vpp); 676 error = VFS_VGET(mp, ino, vpp); 677 678 if (error != 0) { 679 *vpp = NULL; 680 return (error); 681 } 682 return (0); 683 } 684 685 int 686 lfs_create(void *v) 687 { 688 struct vop_create_args /* { 689 struct vnode *a_dvp; 690 struct vnode **a_vpp; 691 struct componentname *a_cnp; 692 struct vattr *a_vap; 693 } */ *ap = v; 694 int error; 695 696 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 697 vput(ap->a_dvp); 698 return error; 699 } 700 error = ufs_create(ap); 701 SET_ENDOP_CREATE_AP(ap, "create"); 702 return (error); 703 } 704 705 int 706 lfs_mkdir(void *v) 707 { 708 struct vop_mkdir_args /* { 709 struct vnode *a_dvp; 710 struct vnode **a_vpp; 711 struct componentname *a_cnp; 712 struct vattr *a_vap; 713 } */ *ap = v; 714 int error; 715 716 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 717 vput(ap->a_dvp); 718 return error; 719 } 720 error = ufs_mkdir(ap); 721 SET_ENDOP_CREATE_AP(ap, "mkdir"); 722 return (error); 723 } 724 725 int 726 lfs_remove(void *v) 727 { 728 struct vop_remove_args /* { 729 struct vnode *a_dvp; 730 struct vnode *a_vp; 731 struct componentname *a_cnp; 732 } */ *ap = v; 733 struct vnode *dvp, *vp; 734 struct inode *ip; 735 int error; 736 737 dvp = ap->a_dvp; 738 vp = ap->a_vp; 739 ip = VTOI(vp); 740 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 741 if (dvp == vp) 742 vrele(vp); 743 else 744 vput(vp); 745 vput(dvp); 746 return error; 747 } 748 error = ufs_remove(ap); 749 if (ip->i_nlink == 0) 750 lfs_orphan(ip->i_lfs, ip->i_number); 751 SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove"); 752 return (error); 753 } 754 755 int 756 lfs_rmdir(void *v) 757 { 758 struct vop_rmdir_args /* { 759 struct vnodeop_desc *a_desc; 760 struct vnode *a_dvp; 761 struct vnode *a_vp; 762 struct componentname *a_cnp; 763 } */ *ap = v; 764 struct vnode *vp; 765 struct inode *ip; 766 int error; 767 768 vp = ap->a_vp; 769 ip = VTOI(vp); 770 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 771 if (ap->a_dvp == vp) 772 vrele(ap->a_dvp); 773 else 774 vput(ap->a_dvp); 775 vput(vp); 776 return error; 777 } 778 error = ufs_rmdir(ap); 779 if (ip->i_nlink == 0) 780 lfs_orphan(ip->i_lfs, ip->i_number); 781 SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 782 return (error); 783 } 784 785 int 786 lfs_link(void *v) 787 { 788 struct vop_link_args /* { 789 struct vnode *a_dvp; 790 struct vnode *a_vp; 791 struct componentname *a_cnp; 792 } */ *ap = v; 793 int error; 794 struct vnode **vpp = NULL; 795 796 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 797 vput(ap->a_dvp); 798 return error; 799 } 800 error = ufs_link(ap); 801 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 802 return (error); 803 } 804 805 int 806 lfs_rename(void *v) 807 { 808 struct vop_rename_args /* { 809 struct vnode *a_fdvp; 810 struct vnode *a_fvp; 811 struct componentname *a_fcnp; 812 struct vnode *a_tdvp; 813 struct vnode *a_tvp; 814 struct componentname *a_tcnp; 815 } */ *ap = v; 816 struct vnode *tvp, *fvp, *tdvp, *fdvp; 817 struct componentname *tcnp, *fcnp; 818 int error; 819 struct lfs *fs; 820 821 fs = VTOI(ap->a_fdvp)->i_lfs; 822 tvp = ap->a_tvp; 823 tdvp = ap->a_tdvp; 824 tcnp = ap->a_tcnp; 825 fvp = ap->a_fvp; 826 fdvp = ap->a_fdvp; 827 fcnp = ap->a_fcnp; 828 829 /* 830 * Check for cross-device rename. 831 * If it is, we don't want to set dirops, just error out. 832 * (In particular note that MARK_VNODE(tdvp) will DTWT on 833 * a cross-device rename.) 834 * 835 * Copied from ufs_rename. 836 */ 837 if ((fvp->v_mount != tdvp->v_mount) || 838 (tvp && (fvp->v_mount != tvp->v_mount))) { 839 error = EXDEV; 840 goto errout; 841 } 842 843 /* 844 * Check to make sure we're not renaming a vnode onto itself 845 * (deleting a hard link by renaming one name onto another); 846 * if we are we can't recursively call VOP_REMOVE since that 847 * would leave us with an unaccounted-for number of live dirops. 848 * 849 * Inline the relevant section of ufs_rename here, *before* 850 * calling SET_DIROP_REMOVE. 851 */ 852 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 853 (VTOI(tdvp)->i_flags & APPEND))) { 854 error = EPERM; 855 goto errout; 856 } 857 if (fvp == tvp) { 858 if (fvp->v_type == VDIR) { 859 error = EINVAL; 860 goto errout; 861 } 862 863 /* Release destination completely. */ 864 VOP_ABORTOP(tdvp, tcnp); 865 vput(tdvp); 866 vput(tvp); 867 868 /* Delete source. */ 869 vrele(fvp); 870 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 871 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 872 fcnp->cn_nameiop = DELETE; 873 vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY); 874 if ((error = relookup(fdvp, &fvp, fcnp))) { 875 vput(fdvp); 876 return (error); 877 } 878 return (VOP_REMOVE(fdvp, fvp, fcnp)); 879 } 880 881 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 882 goto errout; 883 MARK_VNODE(fdvp); 884 MARK_VNODE(fvp); 885 886 error = ufs_rename(ap); 887 UNMARK_VNODE(fdvp); 888 UNMARK_VNODE(fvp); 889 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 890 return (error); 891 892 errout: 893 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 894 if (tdvp == tvp) 895 vrele(tdvp); 896 else 897 vput(tdvp); 898 if (tvp) 899 vput(tvp); 900 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 901 vrele(fdvp); 902 vrele(fvp); 903 return (error); 904 } 905 906 /* XXX hack to avoid calling ITIMES in getattr */ 907 int 908 lfs_getattr(void *v) 909 { 910 struct vop_getattr_args /* { 911 struct vnode *a_vp; 912 struct vattr *a_vap; 913 kauth_cred_t a_cred; 914 } */ *ap = v; 915 struct vnode *vp = ap->a_vp; 916 struct inode *ip = VTOI(vp); 917 struct vattr *vap = ap->a_vap; 918 struct lfs *fs = ip->i_lfs; 919 /* 920 * Copy from inode table 921 */ 922 vap->va_fsid = ip->i_dev; 923 vap->va_fileid = ip->i_number; 924 vap->va_mode = ip->i_mode & ~IFMT; 925 vap->va_nlink = ip->i_nlink; 926 vap->va_uid = ip->i_uid; 927 vap->va_gid = ip->i_gid; 928 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 929 vap->va_size = vp->v_size; 930 vap->va_atime.tv_sec = ip->i_ffs1_atime; 931 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 932 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 933 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 934 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 935 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 936 vap->va_flags = ip->i_flags; 937 vap->va_gen = ip->i_gen; 938 /* this doesn't belong here */ 939 if (vp->v_type == VBLK) 940 vap->va_blocksize = BLKDEV_IOSIZE; 941 else if (vp->v_type == VCHR) 942 vap->va_blocksize = MAXBSIZE; 943 else 944 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 945 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 946 vap->va_type = vp->v_type; 947 vap->va_filerev = ip->i_modrev; 948 return (0); 949 } 950 951 /* 952 * Check to make sure the inode blocks won't choke the buffer 953 * cache, then call ufs_setattr as usual. 954 */ 955 int 956 lfs_setattr(void *v) 957 { 958 struct vop_setattr_args /* { 959 struct vnode *a_vp; 960 struct vattr *a_vap; 961 kauth_cred_t a_cred; 962 } */ *ap = v; 963 struct vnode *vp = ap->a_vp; 964 965 lfs_check(vp, LFS_UNUSED_LBN, 0); 966 return ufs_setattr(v); 967 } 968 969 /* 970 * Release the block we hold on lfs_newseg wrapping. Called on file close, 971 * or explicitly from LFCNWRAPGO. Called with the interlock held. 972 */ 973 static int 974 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor) 975 { 976 if (fs->lfs_stoplwp != curlwp) 977 return EBUSY; 978 979 fs->lfs_stoplwp = NULL; 980 cv_signal(&fs->lfs_stopcv); 981 982 KASSERT(fs->lfs_nowrap > 0); 983 if (fs->lfs_nowrap <= 0) { 984 return 0; 985 } 986 987 if (--fs->lfs_nowrap == 0) { 988 log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt); 989 wakeup(&fs->lfs_wrappass); 990 lfs_wakeup_cleaner(fs); 991 } 992 if (waitfor) { 993 mtsleep(&fs->lfs_nextseg, PCATCH | PUSER, 994 "segment", 0, &fs->lfs_interlock); 995 } 996 997 return 0; 998 } 999 1000 /* 1001 * Close called 1002 */ 1003 /* ARGSUSED */ 1004 int 1005 lfs_close(void *v) 1006 { 1007 struct vop_close_args /* { 1008 struct vnode *a_vp; 1009 int a_fflag; 1010 kauth_cred_t a_cred; 1011 } */ *ap = v; 1012 struct vnode *vp = ap->a_vp; 1013 struct inode *ip = VTOI(vp); 1014 struct lfs *fs = ip->i_lfs; 1015 1016 if ((ip->i_number == ROOTINO || ip->i_number == LFS_IFILE_INUM) && 1017 fs->lfs_stoplwp == curlwp) { 1018 mutex_enter(&fs->lfs_interlock); 1019 log(LOG_NOTICE, "lfs_close: releasing log wrap control\n"); 1020 lfs_wrapgo(fs, ip, 0); 1021 mutex_exit(&fs->lfs_interlock); 1022 } 1023 1024 if (vp == ip->i_lfs->lfs_ivnode && 1025 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 1026 return 0; 1027 1028 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 1029 LFS_ITIMES(ip, NULL, NULL, NULL); 1030 } 1031 return (0); 1032 } 1033 1034 /* 1035 * Close wrapper for special devices. 1036 * 1037 * Update the times on the inode then do device close. 1038 */ 1039 int 1040 lfsspec_close(void *v) 1041 { 1042 struct vop_close_args /* { 1043 struct vnode *a_vp; 1044 int a_fflag; 1045 kauth_cred_t a_cred; 1046 } */ *ap = v; 1047 struct vnode *vp; 1048 struct inode *ip; 1049 1050 vp = ap->a_vp; 1051 ip = VTOI(vp); 1052 if (vp->v_usecount > 1) { 1053 LFS_ITIMES(ip, NULL, NULL, NULL); 1054 } 1055 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1056 } 1057 1058 /* 1059 * Close wrapper for fifo's. 1060 * 1061 * Update the times on the inode then do device close. 1062 */ 1063 int 1064 lfsfifo_close(void *v) 1065 { 1066 struct vop_close_args /* { 1067 struct vnode *a_vp; 1068 int a_fflag; 1069 kauth_cred_ a_cred; 1070 } */ *ap = v; 1071 struct vnode *vp; 1072 struct inode *ip; 1073 1074 vp = ap->a_vp; 1075 ip = VTOI(vp); 1076 if (ap->a_vp->v_usecount > 1) { 1077 LFS_ITIMES(ip, NULL, NULL, NULL); 1078 } 1079 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1080 } 1081 1082 /* 1083 * Reclaim an inode so that it can be used for other purposes. 1084 */ 1085 1086 int 1087 lfs_reclaim(void *v) 1088 { 1089 struct vop_reclaim_args /* { 1090 struct vnode *a_vp; 1091 } */ *ap = v; 1092 struct vnode *vp = ap->a_vp; 1093 struct inode *ip = VTOI(vp); 1094 struct lfs *fs = ip->i_lfs; 1095 int error; 1096 1097 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1098 1099 mutex_enter(&fs->lfs_interlock); 1100 LFS_CLR_UINO(ip, IN_ALLMOD); 1101 mutex_exit(&fs->lfs_interlock); 1102 if ((error = ufs_reclaim(vp))) 1103 return (error); 1104 1105 /* 1106 * Take us off the paging and/or dirop queues if we were on them. 1107 * We shouldn't be on them. 1108 */ 1109 mutex_enter(&fs->lfs_interlock); 1110 if (ip->i_flags & IN_PAGING) { 1111 log(LOG_WARNING, "%s: reclaimed vnode is IN_PAGING\n", 1112 fs->lfs_fsmnt); 1113 ip->i_flags &= ~IN_PAGING; 1114 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1115 } 1116 if (vp->v_uflag & VU_DIROP) { 1117 panic("reclaimed vnode is VDIROP"); 1118 vp->v_uflag &= ~VU_DIROP; 1119 TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain); 1120 } 1121 mutex_exit(&fs->lfs_interlock); 1122 1123 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1124 lfs_deregister_all(vp); 1125 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1126 ip->inode_ext.lfs = NULL; 1127 genfs_node_destroy(vp); 1128 pool_put(&lfs_inode_pool, vp->v_data); 1129 vp->v_data = NULL; 1130 return (0); 1131 } 1132 1133 /* 1134 * Read a block from a storage device. 1135 * In order to avoid reading blocks that are in the process of being 1136 * written by the cleaner---and hence are not mutexed by the normal 1137 * buffer cache / page cache mechanisms---check for collisions before 1138 * reading. 1139 * 1140 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1141 * the active cleaner test. 1142 * 1143 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1144 */ 1145 int 1146 lfs_strategy(void *v) 1147 { 1148 struct vop_strategy_args /* { 1149 struct vnode *a_vp; 1150 struct buf *a_bp; 1151 } */ *ap = v; 1152 struct buf *bp; 1153 struct lfs *fs; 1154 struct vnode *vp; 1155 struct inode *ip; 1156 daddr_t tbn; 1157 int i, sn, error, slept; 1158 1159 bp = ap->a_bp; 1160 vp = ap->a_vp; 1161 ip = VTOI(vp); 1162 fs = ip->i_lfs; 1163 1164 /* lfs uses its strategy routine only for read */ 1165 KASSERT(bp->b_flags & B_READ); 1166 1167 if (vp->v_type == VBLK || vp->v_type == VCHR) 1168 panic("lfs_strategy: spec"); 1169 KASSERT(bp->b_bcount != 0); 1170 if (bp->b_blkno == bp->b_lblkno) { 1171 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1172 NULL); 1173 if (error) { 1174 bp->b_error = error; 1175 bp->b_resid = bp->b_bcount; 1176 biodone(bp); 1177 return (error); 1178 } 1179 if ((long)bp->b_blkno == -1) /* no valid data */ 1180 clrbuf(bp); 1181 } 1182 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1183 bp->b_resid = bp->b_bcount; 1184 biodone(bp); 1185 return (0); 1186 } 1187 1188 slept = 1; 1189 mutex_enter(&fs->lfs_interlock); 1190 while (slept && fs->lfs_seglock) { 1191 mutex_exit(&fs->lfs_interlock); 1192 /* 1193 * Look through list of intervals. 1194 * There will only be intervals to look through 1195 * if the cleaner holds the seglock. 1196 * Since the cleaner is synchronous, we can trust 1197 * the list of intervals to be current. 1198 */ 1199 tbn = dbtofsb(fs, bp->b_blkno); 1200 sn = dtosn(fs, tbn); 1201 slept = 0; 1202 for (i = 0; i < fs->lfs_cleanind; i++) { 1203 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1204 tbn >= fs->lfs_cleanint[i]) { 1205 DLOG((DLOG_CLEAN, 1206 "lfs_strategy: ino %d lbn %" PRId64 1207 " ind %d sn %d fsb %" PRIx32 1208 " given sn %d fsb %" PRIx64 "\n", 1209 ip->i_number, bp->b_lblkno, i, 1210 dtosn(fs, fs->lfs_cleanint[i]), 1211 fs->lfs_cleanint[i], sn, tbn)); 1212 DLOG((DLOG_CLEAN, 1213 "lfs_strategy: sleeping on ino %d lbn %" 1214 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1215 mutex_enter(&fs->lfs_interlock); 1216 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1217 /* Cleaner can't wait for itself */ 1218 mtsleep(&fs->lfs_iocount, 1219 (PRIBIO + 1) | PNORELOCK, 1220 "clean2", 0, 1221 &fs->lfs_interlock); 1222 slept = 1; 1223 break; 1224 } else if (fs->lfs_seglock) { 1225 mtsleep(&fs->lfs_seglock, 1226 (PRIBIO + 1) | PNORELOCK, 1227 "clean1", 0, 1228 &fs->lfs_interlock); 1229 slept = 1; 1230 break; 1231 } 1232 mutex_exit(&fs->lfs_interlock); 1233 } 1234 } 1235 mutex_enter(&fs->lfs_interlock); 1236 } 1237 mutex_exit(&fs->lfs_interlock); 1238 1239 vp = ip->i_devvp; 1240 VOP_STRATEGY(vp, bp); 1241 return (0); 1242 } 1243 1244 void 1245 lfs_flush_dirops(struct lfs *fs) 1246 { 1247 struct inode *ip, *nip; 1248 struct vnode *vp; 1249 extern int lfs_dostats; 1250 struct segment *sp; 1251 int waslocked; 1252 1253 ASSERT_MAYBE_SEGLOCK(fs); 1254 KASSERT(fs->lfs_nadirop == 0); 1255 1256 if (fs->lfs_ronly) 1257 return; 1258 1259 mutex_enter(&fs->lfs_interlock); 1260 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1261 mutex_exit(&fs->lfs_interlock); 1262 return; 1263 } else 1264 mutex_exit(&fs->lfs_interlock); 1265 1266 if (lfs_dostats) 1267 ++lfs_stats.flush_invoked; 1268 1269 /* 1270 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1271 * Technically this is a checkpoint (the on-disk state is valid) 1272 * even though we are leaving out all the file data. 1273 */ 1274 lfs_imtime(fs); 1275 lfs_seglock(fs, SEGM_CKP); 1276 sp = fs->lfs_sp; 1277 1278 /* 1279 * lfs_writevnodes, optimized to get dirops out of the way. 1280 * Only write dirops, and don't flush files' pages, only 1281 * blocks from the directories. 1282 * 1283 * We don't need to vref these files because they are 1284 * dirops and so hold an extra reference until the 1285 * segunlock clears them of that status. 1286 * 1287 * We don't need to check for IN_ADIROP because we know that 1288 * no dirops are active. 1289 * 1290 */ 1291 mutex_enter(&fs->lfs_interlock); 1292 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1293 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1294 mutex_exit(&fs->lfs_interlock); 1295 vp = ITOV(ip); 1296 1297 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1298 1299 /* 1300 * All writes to directories come from dirops; all 1301 * writes to files' direct blocks go through the page 1302 * cache, which we're not touching. Reads to files 1303 * and/or directories will not be affected by writing 1304 * directory blocks inodes and file inodes. So we don't 1305 * really need to lock. If we don't lock, though, 1306 * make sure that we don't clear IN_MODIFIED 1307 * unnecessarily. 1308 */ 1309 if (vp->v_iflag & VI_XLOCK) { 1310 mutex_enter(&fs->lfs_interlock); 1311 continue; 1312 } 1313 waslocked = VOP_ISLOCKED(vp); 1314 if (vp->v_type != VREG && 1315 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1316 lfs_writefile(fs, sp, vp); 1317 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1318 !(ip->i_flag & IN_ALLMOD)) { 1319 mutex_enter(&fs->lfs_interlock); 1320 LFS_SET_UINO(ip, IN_MODIFIED); 1321 mutex_exit(&fs->lfs_interlock); 1322 } 1323 } 1324 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1325 (void) lfs_writeinode(fs, sp, ip); 1326 mutex_enter(&fs->lfs_interlock); 1327 if (waslocked == LK_EXCLOTHER) 1328 LFS_SET_UINO(ip, IN_MODIFIED); 1329 } 1330 mutex_exit(&fs->lfs_interlock); 1331 /* We've written all the dirops there are */ 1332 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1333 lfs_finalize_fs_seguse(fs); 1334 (void) lfs_writeseg(fs, sp); 1335 lfs_segunlock(fs); 1336 } 1337 1338 /* 1339 * Flush all vnodes for which the pagedaemon has requested pageouts. 1340 * Skip over any files that are marked VDIROP (since lfs_flush_dirop() 1341 * has just run, this would be an error). If we have to skip a vnode 1342 * for any reason, just skip it; if we have to wait for the cleaner, 1343 * abort. The writer daemon will call us again later. 1344 */ 1345 void 1346 lfs_flush_pchain(struct lfs *fs) 1347 { 1348 struct inode *ip, *nip; 1349 struct vnode *vp; 1350 extern int lfs_dostats; 1351 struct segment *sp; 1352 int error; 1353 1354 ASSERT_NO_SEGLOCK(fs); 1355 1356 if (fs->lfs_ronly) 1357 return; 1358 1359 mutex_enter(&fs->lfs_interlock); 1360 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1361 mutex_exit(&fs->lfs_interlock); 1362 return; 1363 } else 1364 mutex_exit(&fs->lfs_interlock); 1365 1366 /* Get dirops out of the way */ 1367 lfs_flush_dirops(fs); 1368 1369 if (lfs_dostats) 1370 ++lfs_stats.flush_invoked; 1371 1372 /* 1373 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1374 */ 1375 lfs_imtime(fs); 1376 lfs_seglock(fs, 0); 1377 sp = fs->lfs_sp; 1378 1379 /* 1380 * lfs_writevnodes, optimized to clear pageout requests. 1381 * Only write non-dirop files that are in the pageout queue. 1382 * We're very conservative about what we write; we want to be 1383 * fast and async. 1384 */ 1385 mutex_enter(&fs->lfs_interlock); 1386 top: 1387 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1388 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1389 vp = ITOV(ip); 1390 1391 if (!(ip->i_flags & IN_PAGING)) 1392 goto top; 1393 1394 if ((vp->v_iflag & VI_XLOCK) || (vp->v_uflag & VU_DIROP) != 0) 1395 continue; 1396 if (vp->v_type != VREG) 1397 continue; 1398 if (lfs_vref(vp)) 1399 continue; 1400 mutex_exit(&fs->lfs_interlock); 1401 1402 if (VOP_ISLOCKED(vp)) { 1403 lfs_vunref(vp); 1404 mutex_enter(&fs->lfs_interlock); 1405 continue; 1406 } 1407 1408 error = lfs_writefile(fs, sp, vp); 1409 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1410 !(ip->i_flag & IN_ALLMOD)) { 1411 mutex_enter(&fs->lfs_interlock); 1412 LFS_SET_UINO(ip, IN_MODIFIED); 1413 mutex_exit(&fs->lfs_interlock); 1414 } 1415 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1416 (void) lfs_writeinode(fs, sp, ip); 1417 1418 lfs_vunref(vp); 1419 1420 if (error == EAGAIN) { 1421 lfs_writeseg(fs, sp); 1422 mutex_enter(&fs->lfs_interlock); 1423 break; 1424 } 1425 mutex_enter(&fs->lfs_interlock); 1426 } 1427 mutex_exit(&fs->lfs_interlock); 1428 (void) lfs_writeseg(fs, sp); 1429 lfs_segunlock(fs); 1430 } 1431 1432 /* 1433 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1434 */ 1435 int 1436 lfs_fcntl(void *v) 1437 { 1438 struct vop_fcntl_args /* { 1439 struct vnode *a_vp; 1440 u_long a_command; 1441 void * a_data; 1442 int a_fflag; 1443 kauth_cred_t a_cred; 1444 } */ *ap = v; 1445 struct timeval *tvp; 1446 BLOCK_INFO *blkiov; 1447 CLEANERINFO *cip; 1448 SEGUSE *sup; 1449 int blkcnt, error, oclean; 1450 size_t fh_size; 1451 struct lfs_fcntl_markv blkvp; 1452 struct lwp *l; 1453 fsid_t *fsidp; 1454 struct lfs *fs; 1455 struct buf *bp; 1456 fhandle_t *fhp; 1457 daddr_t off; 1458 1459 /* Only respect LFS fcntls on fs root or Ifile */ 1460 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1461 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1462 return ufs_fcntl(v); 1463 } 1464 1465 /* Avoid locking a draining lock */ 1466 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1467 return ESHUTDOWN; 1468 } 1469 1470 /* LFS control and monitoring fcntls are available only to root */ 1471 l = curlwp; 1472 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1473 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1474 NULL)) != 0) 1475 return (error); 1476 1477 fs = VTOI(ap->a_vp)->i_lfs; 1478 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1479 1480 error = 0; 1481 switch (ap->a_command) { 1482 case LFCNSEGWAITALL: 1483 case LFCNSEGWAITALL_COMPAT: 1484 fsidp = NULL; 1485 /* FALLSTHROUGH */ 1486 case LFCNSEGWAIT: 1487 case LFCNSEGWAIT_COMPAT: 1488 tvp = (struct timeval *)ap->a_data; 1489 mutex_enter(&fs->lfs_interlock); 1490 ++fs->lfs_sleepers; 1491 mutex_exit(&fs->lfs_interlock); 1492 1493 error = lfs_segwait(fsidp, tvp); 1494 1495 mutex_enter(&fs->lfs_interlock); 1496 if (--fs->lfs_sleepers == 0) 1497 wakeup(&fs->lfs_sleepers); 1498 mutex_exit(&fs->lfs_interlock); 1499 return error; 1500 1501 case LFCNBMAPV: 1502 case LFCNMARKV: 1503 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1504 1505 blkcnt = blkvp.blkcnt; 1506 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1507 return (EINVAL); 1508 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1509 if ((error = copyin(blkvp.blkiov, blkiov, 1510 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1511 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1512 return error; 1513 } 1514 1515 mutex_enter(&fs->lfs_interlock); 1516 ++fs->lfs_sleepers; 1517 mutex_exit(&fs->lfs_interlock); 1518 if (ap->a_command == LFCNBMAPV) 1519 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1520 else /* LFCNMARKV */ 1521 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1522 if (error == 0) 1523 error = copyout(blkiov, blkvp.blkiov, 1524 blkcnt * sizeof(BLOCK_INFO)); 1525 mutex_enter(&fs->lfs_interlock); 1526 if (--fs->lfs_sleepers == 0) 1527 wakeup(&fs->lfs_sleepers); 1528 mutex_exit(&fs->lfs_interlock); 1529 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1530 return error; 1531 1532 case LFCNRECLAIM: 1533 /* 1534 * Flush dirops and write Ifile, allowing empty segments 1535 * to be immediately reclaimed. 1536 */ 1537 lfs_writer_enter(fs, "pndirop"); 1538 off = fs->lfs_offset; 1539 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1540 lfs_flush_dirops(fs); 1541 LFS_CLEANERINFO(cip, fs, bp); 1542 oclean = cip->clean; 1543 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1544 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1545 fs->lfs_sp->seg_flags |= SEGM_PROT; 1546 lfs_segunlock(fs); 1547 lfs_writer_leave(fs); 1548 1549 #ifdef DEBUG 1550 LFS_CLEANERINFO(cip, fs, bp); 1551 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1552 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1553 fs->lfs_offset - off, cip->clean - oclean, 1554 fs->lfs_activesb)); 1555 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1556 #endif 1557 1558 return 0; 1559 1560 #ifdef COMPAT_30 1561 case LFCNIFILEFH_COMPAT: 1562 /* Return the filehandle of the Ifile */ 1563 if ((error = kauth_authorize_generic(l->l_cred, 1564 KAUTH_GENERIC_ISSUSER, NULL)) != 0) 1565 return (error); 1566 fhp = (struct fhandle *)ap->a_data; 1567 fhp->fh_fsid = *fsidp; 1568 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1569 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1570 #endif 1571 1572 case LFCNIFILEFH_COMPAT2: 1573 case LFCNIFILEFH: 1574 /* Return the filehandle of the Ifile */ 1575 fhp = (struct fhandle *)ap->a_data; 1576 fhp->fh_fsid = *fsidp; 1577 fh_size = sizeof(struct lfs_fhandle) - 1578 offsetof(fhandle_t, fh_fid); 1579 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1580 1581 case LFCNREWIND: 1582 /* Move lfs_offset to the lowest-numbered segment */ 1583 return lfs_rewind(fs, *(int *)ap->a_data); 1584 1585 case LFCNINVAL: 1586 /* Mark a segment SEGUSE_INVAL */ 1587 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1588 if (sup->su_nbytes > 0) { 1589 brelse(bp, 0); 1590 lfs_unset_inval_all(fs); 1591 return EBUSY; 1592 } 1593 sup->su_flags |= SEGUSE_INVAL; 1594 VOP_BWRITE(bp); 1595 return 0; 1596 1597 case LFCNRESIZE: 1598 /* Resize the filesystem */ 1599 return lfs_resize_fs(fs, *(int *)ap->a_data); 1600 1601 case LFCNWRAPSTOP: 1602 case LFCNWRAPSTOP_COMPAT: 1603 /* 1604 * Hold lfs_newseg at segment 0; if requested, sleep until 1605 * the filesystem wraps around. To support external agents 1606 * (dump, fsck-based regression test) that need to look at 1607 * a snapshot of the filesystem, without necessarily 1608 * requiring that all fs activity stops. 1609 */ 1610 if (fs->lfs_stoplwp == curlwp) 1611 return EALREADY; 1612 1613 mutex_enter(&fs->lfs_interlock); 1614 while (fs->lfs_stoplwp != NULL) 1615 cv_wait(&fs->lfs_stopcv, &fs->lfs_interlock); 1616 fs->lfs_stoplwp = curlwp; 1617 if (fs->lfs_nowrap == 0) 1618 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1619 ++fs->lfs_nowrap; 1620 if (*(int *)ap->a_data == 1 || 1621 ap->a_command == LFCNWRAPSTOP_COMPAT) { 1622 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1623 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1624 "segwrap", 0, &fs->lfs_interlock); 1625 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1626 if (error) { 1627 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1628 } 1629 } 1630 mutex_exit(&fs->lfs_interlock); 1631 return 0; 1632 1633 case LFCNWRAPGO: 1634 case LFCNWRAPGO_COMPAT: 1635 /* 1636 * Having done its work, the agent wakes up the writer. 1637 * If the argument is 1, it sleeps until a new segment 1638 * is selected. 1639 */ 1640 mutex_enter(&fs->lfs_interlock); 1641 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1642 (ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1643 *((int *)ap->a_data))); 1644 mutex_exit(&fs->lfs_interlock); 1645 return error; 1646 1647 case LFCNWRAPPASS: 1648 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1649 return EALREADY; 1650 mutex_enter(&fs->lfs_interlock); 1651 if (fs->lfs_stoplwp != curlwp) { 1652 mutex_exit(&fs->lfs_interlock); 1653 return EALREADY; 1654 } 1655 if (fs->lfs_nowrap == 0) { 1656 mutex_exit(&fs->lfs_interlock); 1657 return EBUSY; 1658 } 1659 fs->lfs_wrappass = 1; 1660 wakeup(&fs->lfs_wrappass); 1661 /* Wait for the log to wrap, if asked */ 1662 if (*(int *)ap->a_data) { 1663 lfs_vref(ap->a_vp); 1664 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1665 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1666 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1667 "segwrap", 0, &fs->lfs_interlock); 1668 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1669 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1670 lfs_vunref(ap->a_vp); 1671 } 1672 mutex_exit(&fs->lfs_interlock); 1673 return error; 1674 1675 case LFCNWRAPSTATUS: 1676 mutex_enter(&fs->lfs_interlock); 1677 *(int *)ap->a_data = fs->lfs_wrapstatus; 1678 mutex_exit(&fs->lfs_interlock); 1679 return 0; 1680 1681 default: 1682 return ufs_fcntl(v); 1683 } 1684 return 0; 1685 } 1686 1687 int 1688 lfs_getpages(void *v) 1689 { 1690 struct vop_getpages_args /* { 1691 struct vnode *a_vp; 1692 voff_t a_offset; 1693 struct vm_page **a_m; 1694 int *a_count; 1695 int a_centeridx; 1696 vm_prot_t a_access_type; 1697 int a_advice; 1698 int a_flags; 1699 } */ *ap = v; 1700 1701 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1702 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1703 return EPERM; 1704 } 1705 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1706 mutex_enter(&VTOI(ap->a_vp)->i_lfs->lfs_interlock); 1707 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1708 mutex_exit(&VTOI(ap->a_vp)->i_lfs->lfs_interlock); 1709 } 1710 1711 /* 1712 * we're relying on the fact that genfs_getpages() always read in 1713 * entire filesystem blocks. 1714 */ 1715 return genfs_getpages(v); 1716 } 1717 1718 /* 1719 * Wait for a page to become unbusy, possibly printing diagnostic messages 1720 * as well. 1721 * 1722 * Called with vp->v_interlock held; return with it held. 1723 */ 1724 static void 1725 wait_for_page(struct vnode *vp, struct vm_page *pg, const char *label) 1726 { 1727 if ((pg->flags & PG_BUSY) == 0) 1728 return; /* Nothing to wait for! */ 1729 1730 #if defined(DEBUG) && defined(UVM_PAGE_TRKOWN) 1731 static struct vm_page *lastpg; 1732 1733 if (label != NULL && pg != lastpg) { 1734 if (pg->owner_tag) { 1735 printf("lfs_putpages[%d.%d]: %s: page %p owner %d.%d [%s]\n", 1736 curproc->p_pid, curlwp->l_lid, label, 1737 pg, pg->owner, pg->lowner, pg->owner_tag); 1738 } else { 1739 printf("lfs_putpages[%d.%d]: %s: page %p unowned?!\n", 1740 curproc->p_pid, curlwp->l_lid, label, pg); 1741 } 1742 } 1743 lastpg = pg; 1744 #endif 1745 1746 pg->flags |= PG_WANTED; 1747 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, "lfsput", 0); 1748 mutex_enter(&vp->v_interlock); 1749 } 1750 1751 /* 1752 * This routine is called by lfs_putpages() when it can't complete the 1753 * write because a page is busy. This means that either (1) someone, 1754 * possibly the pagedaemon, is looking at this page, and will give it up 1755 * presently; or (2) we ourselves are holding the page busy in the 1756 * process of being written (either gathered or actually on its way to 1757 * disk). We don't need to give up the segment lock, but we might need 1758 * to call lfs_writeseg() to expedite the page's journey to disk. 1759 * 1760 * Called with vp->v_interlock held; return with it held. 1761 */ 1762 /* #define BUSYWAIT */ 1763 static void 1764 write_and_wait(struct lfs *fs, struct vnode *vp, struct vm_page *pg, 1765 int seglocked, const char *label) 1766 { 1767 #ifndef BUSYWAIT 1768 struct inode *ip = VTOI(vp); 1769 struct segment *sp = fs->lfs_sp; 1770 int count = 0; 1771 1772 if (pg == NULL) 1773 return; 1774 1775 while (pg->flags & PG_BUSY) { 1776 mutex_exit(&vp->v_interlock); 1777 if (sp->cbpp - sp->bpp > 1) { 1778 /* Write gathered pages */ 1779 lfs_updatemeta(sp); 1780 lfs_release_finfo(fs); 1781 (void) lfs_writeseg(fs, sp); 1782 1783 /* 1784 * Reinitialize FIP 1785 */ 1786 KASSERT(sp->vp == vp); 1787 lfs_acquire_finfo(fs, ip->i_number, 1788 ip->i_gen); 1789 } 1790 ++count; 1791 mutex_enter(&vp->v_interlock); 1792 wait_for_page(vp, pg, label); 1793 } 1794 if (label != NULL && count > 1) 1795 printf("lfs_putpages[%d]: %s: %sn = %d\n", curproc->p_pid, 1796 label, (count > 0 ? "looping, " : ""), count); 1797 #else 1798 preempt(1); 1799 #endif 1800 } 1801 1802 /* 1803 * Make sure that for all pages in every block in the given range, 1804 * either all are dirty or all are clean. If any of the pages 1805 * we've seen so far are dirty, put the vnode on the paging chain, 1806 * and mark it IN_PAGING. 1807 * 1808 * If checkfirst != 0, don't check all the pages but return at the 1809 * first dirty page. 1810 */ 1811 static int 1812 check_dirty(struct lfs *fs, struct vnode *vp, 1813 off_t startoffset, off_t endoffset, off_t blkeof, 1814 int flags, int checkfirst, struct vm_page **pgp) 1815 { 1816 int by_list; 1817 struct vm_page *curpg = NULL; /* XXX: gcc */ 1818 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1819 off_t soff = 0; /* XXX: gcc */ 1820 voff_t off; 1821 int i; 1822 int nonexistent; 1823 int any_dirty; /* number of dirty pages */ 1824 int dirty; /* number of dirty pages in a block */ 1825 int tdirty; 1826 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1827 int pagedaemon = (curlwp == uvm.pagedaemon_lwp); 1828 1829 ASSERT_MAYBE_SEGLOCK(fs); 1830 top: 1831 by_list = (vp->v_uobj.uo_npages <= 1832 ((endoffset - startoffset) >> PAGE_SHIFT) * 1833 UVM_PAGE_HASH_PENALTY); 1834 any_dirty = 0; 1835 1836 if (by_list) { 1837 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1838 } else { 1839 soff = startoffset; 1840 } 1841 while (by_list || soff < MIN(blkeof, endoffset)) { 1842 if (by_list) { 1843 /* 1844 * Find the first page in a block. Skip 1845 * blocks outside our area of interest or beyond 1846 * the end of file. 1847 */ 1848 if (pages_per_block > 1) { 1849 while (curpg && 1850 ((curpg->offset & fs->lfs_bmask) || 1851 curpg->offset >= vp->v_size || 1852 curpg->offset >= endoffset)) 1853 curpg = TAILQ_NEXT(curpg, listq); 1854 } 1855 if (curpg == NULL) 1856 break; 1857 soff = curpg->offset; 1858 } 1859 1860 /* 1861 * Mark all pages in extended range busy; find out if any 1862 * of them are dirty. 1863 */ 1864 nonexistent = dirty = 0; 1865 for (i = 0; i == 0 || i < pages_per_block; i++) { 1866 if (by_list && pages_per_block <= 1) { 1867 pgs[i] = pg = curpg; 1868 } else { 1869 off = soff + (i << PAGE_SHIFT); 1870 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1871 if (pg == NULL) { 1872 ++nonexistent; 1873 continue; 1874 } 1875 } 1876 KASSERT(pg != NULL); 1877 1878 /* 1879 * If we're holding the segment lock, we can deadlock 1880 * against a process that has our page and is waiting 1881 * for the cleaner, while the cleaner waits for the 1882 * segment lock. Just bail in that case. 1883 */ 1884 if ((pg->flags & PG_BUSY) && 1885 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1886 if (i > 0) 1887 uvm_page_unbusy(pgs, i); 1888 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1889 if (pgp) 1890 *pgp = pg; 1891 return -1; 1892 } 1893 1894 while (pg->flags & PG_BUSY) { 1895 wait_for_page(vp, pg, NULL); 1896 if (i > 0) 1897 uvm_page_unbusy(pgs, i); 1898 goto top; 1899 } 1900 pg->flags |= PG_BUSY; 1901 UVM_PAGE_OWN(pg, "lfs_putpages"); 1902 1903 pmap_page_protect(pg, VM_PROT_NONE); 1904 tdirty = (pmap_clear_modify(pg) || 1905 (pg->flags & PG_CLEAN) == 0); 1906 dirty += tdirty; 1907 } 1908 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1909 if (by_list) { 1910 curpg = TAILQ_NEXT(curpg, listq); 1911 } else { 1912 soff += fs->lfs_bsize; 1913 } 1914 continue; 1915 } 1916 1917 any_dirty += dirty; 1918 KASSERT(nonexistent == 0); 1919 1920 /* 1921 * If any are dirty make all dirty; unbusy them, 1922 * but if we were asked to clean, wire them so that 1923 * the pagedaemon doesn't bother us about them while 1924 * they're on their way to disk. 1925 */ 1926 for (i = 0; i == 0 || i < pages_per_block; i++) { 1927 pg = pgs[i]; 1928 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1929 if (dirty) { 1930 pg->flags &= ~PG_CLEAN; 1931 if (flags & PGO_FREE) { 1932 /* 1933 * Wire the page so that 1934 * pdaemon doesn't see it again. 1935 */ 1936 mutex_enter(&uvm_pageqlock); 1937 uvm_pagewire(pg); 1938 mutex_exit(&uvm_pageqlock); 1939 1940 /* Suspended write flag */ 1941 pg->flags |= PG_DELWRI; 1942 } 1943 } 1944 if (pg->flags & PG_WANTED) 1945 wakeup(pg); 1946 pg->flags &= ~(PG_WANTED|PG_BUSY); 1947 UVM_PAGE_OWN(pg, NULL); 1948 } 1949 1950 if (checkfirst && any_dirty) 1951 break; 1952 1953 if (by_list) { 1954 curpg = TAILQ_NEXT(curpg, listq); 1955 } else { 1956 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1957 } 1958 } 1959 1960 return any_dirty; 1961 } 1962 1963 /* 1964 * lfs_putpages functions like genfs_putpages except that 1965 * 1966 * (1) It needs to bounds-check the incoming requests to ensure that 1967 * they are block-aligned; if they are not, expand the range and 1968 * do the right thing in case, e.g., the requested range is clean 1969 * but the expanded range is dirty. 1970 * 1971 * (2) It needs to explicitly send blocks to be written when it is done. 1972 * If VOP_PUTPAGES is called without the seglock held, we simply take 1973 * the seglock and let lfs_segunlock wait for us. 1974 * XXX There might be a bad situation if we have to flush a vnode while 1975 * XXX lfs_markv is in operation. As of this writing we panic in this 1976 * XXX case. 1977 * 1978 * Assumptions: 1979 * 1980 * (1) The caller does not hold any pages in this vnode busy. If it does, 1981 * there is a danger that when we expand the page range and busy the 1982 * pages we will deadlock. 1983 * 1984 * (2) We are called with vp->v_interlock held; we must return with it 1985 * released. 1986 * 1987 * (3) We don't absolutely have to free pages right away, provided that 1988 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1989 * us a request with PGO_FREE, we take the pages out of the paging 1990 * queue and wake up the writer, which will handle freeing them for us. 1991 * 1992 * We ensure that for any filesystem block, all pages for that 1993 * block are either resident or not, even if those pages are higher 1994 * than EOF; that means that we will be getting requests to free 1995 * "unused" pages above EOF all the time, and should ignore them. 1996 * 1997 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 1998 * into has been set up for us by lfs_writefile. If not, we will 1999 * have to handle allocating and/or freeing an finfo entry. 2000 * 2001 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 2002 */ 2003 2004 /* How many times to loop before we should start to worry */ 2005 #define TOOMANY 4 2006 2007 int 2008 lfs_putpages(void *v) 2009 { 2010 int error; 2011 struct vop_putpages_args /* { 2012 struct vnode *a_vp; 2013 voff_t a_offlo; 2014 voff_t a_offhi; 2015 int a_flags; 2016 } */ *ap = v; 2017 struct vnode *vp; 2018 struct inode *ip; 2019 struct lfs *fs; 2020 struct segment *sp; 2021 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 2022 off_t off, max_endoffset; 2023 bool seglocked, sync, pagedaemon; 2024 struct vm_page *pg, *busypg; 2025 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 2026 #ifdef DEBUG 2027 int debug_n_again, debug_n_dirtyclean; 2028 #endif 2029 2030 vp = ap->a_vp; 2031 ip = VTOI(vp); 2032 fs = ip->i_lfs; 2033 sync = (ap->a_flags & PGO_SYNCIO) != 0; 2034 pagedaemon = (curlwp == uvm.pagedaemon_lwp); 2035 2036 /* Putpages does nothing for metadata. */ 2037 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 2038 mutex_exit(&vp->v_interlock); 2039 return 0; 2040 } 2041 2042 /* 2043 * If there are no pages, don't do anything. 2044 */ 2045 if (vp->v_uobj.uo_npages == 0) { 2046 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 2047 (vp->v_iflag & VI_ONWORKLST) && 2048 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 2049 vp->v_iflag &= ~VI_WRMAPDIRTY; 2050 vn_syncer_remove_from_worklist(vp); 2051 } 2052 mutex_exit(&vp->v_interlock); 2053 2054 /* Remove us from paging queue, if we were on it */ 2055 mutex_enter(&fs->lfs_interlock); 2056 if (ip->i_flags & IN_PAGING) { 2057 ip->i_flags &= ~IN_PAGING; 2058 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2059 } 2060 mutex_exit(&fs->lfs_interlock); 2061 return 0; 2062 } 2063 2064 blkeof = blkroundup(fs, ip->i_size); 2065 2066 /* 2067 * Ignore requests to free pages past EOF but in the same block 2068 * as EOF, unless the request is synchronous. (If the request is 2069 * sync, it comes from lfs_truncate.) 2070 * XXXUBC Make these pages look "active" so the pagedaemon won't 2071 * XXXUBC bother us with them again. 2072 */ 2073 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 2074 origoffset = ap->a_offlo; 2075 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 2076 pg = uvm_pagelookup(&vp->v_uobj, off); 2077 KASSERT(pg != NULL); 2078 while (pg->flags & PG_BUSY) { 2079 pg->flags |= PG_WANTED; 2080 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 2081 "lfsput2", 0); 2082 mutex_enter(&vp->v_interlock); 2083 } 2084 mutex_enter(&uvm_pageqlock); 2085 uvm_pageactivate(pg); 2086 mutex_exit(&uvm_pageqlock); 2087 } 2088 ap->a_offlo = blkeof; 2089 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 2090 mutex_exit(&vp->v_interlock); 2091 return 0; 2092 } 2093 } 2094 2095 /* 2096 * Extend page range to start and end at block boundaries. 2097 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 2098 */ 2099 origoffset = ap->a_offlo; 2100 origendoffset = ap->a_offhi; 2101 startoffset = origoffset & ~(fs->lfs_bmask); 2102 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 2103 << fs->lfs_bshift; 2104 2105 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2106 endoffset = max_endoffset; 2107 origendoffset = endoffset; 2108 } else { 2109 origendoffset = round_page(ap->a_offhi); 2110 endoffset = round_page(blkroundup(fs, origendoffset)); 2111 } 2112 2113 KASSERT(startoffset > 0 || endoffset >= startoffset); 2114 if (startoffset == endoffset) { 2115 /* Nothing to do, why were we called? */ 2116 mutex_exit(&vp->v_interlock); 2117 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 2118 PRId64 "\n", startoffset)); 2119 return 0; 2120 } 2121 2122 ap->a_offlo = startoffset; 2123 ap->a_offhi = endoffset; 2124 2125 /* 2126 * If not cleaning, just send the pages through genfs_putpages 2127 * to be returned to the pool. 2128 */ 2129 if (!(ap->a_flags & PGO_CLEANIT)) 2130 return genfs_putpages(v); 2131 2132 /* Set PGO_BUSYFAIL to avoid deadlocks */ 2133 ap->a_flags |= PGO_BUSYFAIL; 2134 2135 /* 2136 * Likewise, if we are asked to clean but the pages are not 2137 * dirty, we can just free them using genfs_putpages. 2138 */ 2139 #ifdef DEBUG 2140 debug_n_dirtyclean = 0; 2141 #endif 2142 do { 2143 int r; 2144 2145 /* Count the number of dirty pages */ 2146 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2147 ap->a_flags, 1, NULL); 2148 if (r < 0) { 2149 /* Pages are busy with another process */ 2150 mutex_exit(&vp->v_interlock); 2151 return EDEADLK; 2152 } 2153 if (r > 0) /* Some pages are dirty */ 2154 break; 2155 2156 /* 2157 * Sometimes pages are dirtied between the time that 2158 * we check and the time we try to clean them. 2159 * Instruct lfs_gop_write to return EDEADLK in this case 2160 * so we can write them properly. 2161 */ 2162 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2163 r = genfs_do_putpages(vp, startoffset, endoffset, 2164 ap->a_flags, &busypg); 2165 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2166 if (r != EDEADLK) 2167 return r; 2168 2169 /* One of the pages was busy. Start over. */ 2170 mutex_enter(&vp->v_interlock); 2171 wait_for_page(vp, busypg, "dirtyclean"); 2172 #ifdef DEBUG 2173 ++debug_n_dirtyclean; 2174 #endif 2175 } while(1); 2176 2177 #ifdef DEBUG 2178 if (debug_n_dirtyclean > TOOMANY) 2179 printf("lfs_putpages: dirtyclean: looping, n = %d\n", 2180 debug_n_dirtyclean); 2181 #endif 2182 2183 /* 2184 * Dirty and asked to clean. 2185 * 2186 * Pagedaemon can't actually write LFS pages; wake up 2187 * the writer to take care of that. The writer will 2188 * notice the pager inode queue and act on that. 2189 */ 2190 if (pagedaemon) { 2191 mutex_enter(&fs->lfs_interlock); 2192 if (!(ip->i_flags & IN_PAGING)) { 2193 ip->i_flags |= IN_PAGING; 2194 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2195 } 2196 mutex_enter(&lfs_subsys_lock); 2197 wakeup(&lfs_writer_daemon); 2198 mutex_exit(&lfs_subsys_lock); 2199 mutex_exit(&fs->lfs_interlock); 2200 mutex_exit(&vp->v_interlock); 2201 preempt(); 2202 return EWOULDBLOCK; 2203 } 2204 2205 /* 2206 * If this is a file created in a recent dirop, we can't flush its 2207 * inode until the dirop is complete. Drain dirops, then flush the 2208 * filesystem (taking care of any other pending dirops while we're 2209 * at it). 2210 */ 2211 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2212 (vp->v_uflag & VU_DIROP)) { 2213 int locked; 2214 2215 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 2216 locked = (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 2217 mutex_exit(&vp->v_interlock); 2218 lfs_writer_enter(fs, "ppdirop"); 2219 if (locked) 2220 VOP_UNLOCK(vp, 0); /* XXX why? */ 2221 2222 mutex_enter(&fs->lfs_interlock); 2223 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2224 mutex_exit(&fs->lfs_interlock); 2225 2226 mutex_enter(&vp->v_interlock); 2227 if (locked) { 2228 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2229 mutex_enter(&vp->v_interlock); 2230 } 2231 lfs_writer_leave(fs); 2232 2233 /* XXX the flush should have taken care of this one too! */ 2234 } 2235 2236 /* 2237 * This is it. We are going to write some pages. From here on 2238 * down it's all just mechanics. 2239 * 2240 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2241 */ 2242 ap->a_flags &= ~PGO_SYNCIO; 2243 2244 /* 2245 * If we've already got the seglock, flush the node and return. 2246 * The FIP has already been set up for us by lfs_writefile, 2247 * and FIP cleanup and lfs_updatemeta will also be done there, 2248 * unless genfs_putpages returns EDEADLK; then we must flush 2249 * what we have, and correct FIP and segment header accounting. 2250 */ 2251 get_seglock: 2252 /* 2253 * If we are not called with the segment locked, lock it. 2254 * Account for a new FIP in the segment header, and set sp->vp. 2255 * (This should duplicate the setup at the top of lfs_writefile().) 2256 */ 2257 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2258 if (!seglocked) { 2259 mutex_exit(&vp->v_interlock); 2260 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2261 if (error != 0) 2262 return error; 2263 mutex_enter(&vp->v_interlock); 2264 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2265 } 2266 sp = fs->lfs_sp; 2267 KASSERT(sp->vp == NULL); 2268 sp->vp = vp; 2269 2270 /* 2271 * Ensure that the partial segment is marked SS_DIROP if this 2272 * vnode is a DIROP. 2273 */ 2274 if (!seglocked && vp->v_uflag & VU_DIROP) 2275 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2276 2277 /* 2278 * Loop over genfs_putpages until all pages are gathered. 2279 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2280 * Whenever we lose the interlock we have to rerun check_dirty, as 2281 * well, since more pages might have been dirtied in our absence. 2282 */ 2283 #ifdef DEBUG 2284 debug_n_again = 0; 2285 #endif 2286 do { 2287 busypg = NULL; 2288 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2289 ap->a_flags, 0, &busypg) < 0) { 2290 mutex_exit(&vp->v_interlock); 2291 2292 mutex_enter(&vp->v_interlock); 2293 write_and_wait(fs, vp, busypg, seglocked, NULL); 2294 if (!seglocked) { 2295 lfs_release_finfo(fs); 2296 lfs_segunlock(fs); 2297 } 2298 sp->vp = NULL; 2299 goto get_seglock; 2300 } 2301 2302 busypg = NULL; 2303 error = genfs_do_putpages(vp, startoffset, endoffset, 2304 ap->a_flags, &busypg); 2305 2306 if (error == EDEADLK || error == EAGAIN) { 2307 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2308 " %d ino %d off %x (seg %d)\n", error, 2309 ip->i_number, fs->lfs_offset, 2310 dtosn(fs, fs->lfs_offset))); 2311 2312 mutex_enter(&vp->v_interlock); 2313 write_and_wait(fs, vp, busypg, seglocked, "again"); 2314 } 2315 #ifdef DEBUG 2316 ++debug_n_again; 2317 #endif 2318 } while (error == EDEADLK); 2319 #ifdef DEBUG 2320 if (debug_n_again > TOOMANY) 2321 printf("lfs_putpages: again: looping, n = %d\n", debug_n_again); 2322 #endif 2323 2324 KASSERT(sp != NULL && sp->vp == vp); 2325 if (!seglocked) { 2326 sp->vp = NULL; 2327 2328 /* Write indirect blocks as well */ 2329 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2330 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2331 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2332 2333 KASSERT(sp->vp == NULL); 2334 sp->vp = vp; 2335 } 2336 2337 /* 2338 * Blocks are now gathered into a segment waiting to be written. 2339 * All that's left to do is update metadata, and write them. 2340 */ 2341 lfs_updatemeta(sp); 2342 KASSERT(sp->vp == vp); 2343 sp->vp = NULL; 2344 2345 /* 2346 * If we were called from lfs_writefile, we don't need to clean up 2347 * the FIP or unlock the segment lock. We're done. 2348 */ 2349 if (seglocked) 2350 return error; 2351 2352 /* Clean up FIP and send it to disk. */ 2353 lfs_release_finfo(fs); 2354 lfs_writeseg(fs, fs->lfs_sp); 2355 2356 /* 2357 * Remove us from paging queue if we wrote all our pages. 2358 */ 2359 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2360 mutex_enter(&fs->lfs_interlock); 2361 if (ip->i_flags & IN_PAGING) { 2362 ip->i_flags &= ~IN_PAGING; 2363 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2364 } 2365 mutex_exit(&fs->lfs_interlock); 2366 } 2367 2368 /* 2369 * XXX - with the malloc/copy writeseg, the pages are freed by now 2370 * even if we don't wait (e.g. if we hold a nested lock). This 2371 * will not be true if we stop using malloc/copy. 2372 */ 2373 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2374 lfs_segunlock(fs); 2375 2376 /* 2377 * Wait for v_numoutput to drop to zero. The seglock should 2378 * take care of this, but there is a slight possibility that 2379 * aiodoned might not have got around to our buffers yet. 2380 */ 2381 if (sync) { 2382 mutex_enter(&vp->v_interlock); 2383 while (vp->v_numoutput > 0) { 2384 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2385 " num %d\n", ip->i_number, vp->v_numoutput)); 2386 cv_wait(&vp->v_cv, &vp->v_interlock); 2387 } 2388 mutex_exit(&vp->v_interlock); 2389 } 2390 return error; 2391 } 2392 2393 /* 2394 * Return the last logical file offset that should be written for this file 2395 * if we're doing a write that ends at "size". If writing, we need to know 2396 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2397 * to know about entire blocks. 2398 */ 2399 void 2400 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2401 { 2402 struct inode *ip = VTOI(vp); 2403 struct lfs *fs = ip->i_lfs; 2404 daddr_t olbn, nlbn; 2405 2406 olbn = lblkno(fs, ip->i_size); 2407 nlbn = lblkno(fs, size); 2408 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2409 *eobp = fragroundup(fs, size); 2410 } else { 2411 *eobp = blkroundup(fs, size); 2412 } 2413 } 2414 2415 #ifdef DEBUG 2416 void lfs_dump_vop(void *); 2417 2418 void 2419 lfs_dump_vop(void *v) 2420 { 2421 struct vop_putpages_args /* { 2422 struct vnode *a_vp; 2423 voff_t a_offlo; 2424 voff_t a_offhi; 2425 int a_flags; 2426 } */ *ap = v; 2427 2428 #ifdef DDB 2429 vfs_vnode_print(ap->a_vp, 0, printf); 2430 #endif 2431 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2432 } 2433 #endif 2434 2435 int 2436 lfs_mmap(void *v) 2437 { 2438 struct vop_mmap_args /* { 2439 const struct vnodeop_desc *a_desc; 2440 struct vnode *a_vp; 2441 vm_prot_t a_prot; 2442 kauth_cred_t a_cred; 2443 } */ *ap = v; 2444 2445 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2446 return EOPNOTSUPP; 2447 return ufs_mmap(v); 2448 } 2449
Indexes created Mon Oct 20 10:10:01 GMT 2025