lfs_vnops.c revision 1.234
1 /* $NetBSD: lfs_vnops.c,v 1.234 2011/01/05 19:34:27 martin 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 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 /* 32 * Copyright (c) 1986, 1989, 1991, 1993, 1995 33 * The Regents of the University of California. All rights reserved. 34 * 35 * Redistribution and use in source and binary forms, with or without 36 * modification, are permitted provided that the following conditions 37 * are met: 38 * 1. Redistributions of source code must retain the above copyright 39 * notice, this list of conditions and the following disclaimer. 40 * 2. Redistributions in binary form must reproduce the above copyright 41 * notice, this list of conditions and the following disclaimer in the 42 * documentation and/or other materials provided with the distribution. 43 * 3. Neither the name of the University nor the names of its contributors 44 * may be used to endorse or promote products derived from this software 45 * without specific prior written permission. 46 * 47 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 48 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 49 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 50 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 51 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 52 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 53 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 54 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 55 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 56 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 57 * SUCH DAMAGE. 58 * 59 * @(#)lfs_vnops.c 8.13 (Berkeley) 6/10/95 60 */ 61 62 #include <sys/cdefs.h> 63 __KERNEL_RCSID(0, "$NetBSD: lfs_vnops.c,v 1.234 2011/01/05 19:34:27 martin Exp $"); 64 65 #ifdef _KERNEL_OPT 66 #include "opt_compat_netbsd.h" 67 #endif 68 69 #include <sys/param.h> 70 #include <sys/systm.h> 71 #include <sys/namei.h> 72 #include <sys/resourcevar.h> 73 #include <sys/kernel.h> 74 #include <sys/file.h> 75 #include <sys/stat.h> 76 #include <sys/buf.h> 77 #include <sys/proc.h> 78 #include <sys/mount.h> 79 #include <sys/vnode.h> 80 #include <sys/pool.h> 81 #include <sys/signalvar.h> 82 #include <sys/kauth.h> 83 #include <sys/syslog.h> 84 #include <sys/fstrans.h> 85 86 #include <miscfs/fifofs/fifo.h> 87 #include <miscfs/genfs/genfs.h> 88 #include <miscfs/specfs/specdev.h> 89 90 #include <ufs/ufs/inode.h> 91 #include <ufs/ufs/dir.h> 92 #include <ufs/ufs/ufsmount.h> 93 #include <ufs/ufs/ufs_extern.h> 94 95 #include <uvm/uvm.h> 96 #include <uvm/uvm_pmap.h> 97 #include <uvm/uvm_stat.h> 98 #include <uvm/uvm_pager.h> 99 100 #include <ufs/lfs/lfs.h> 101 #include <ufs/lfs/lfs_extern.h> 102 103 extern pid_t lfs_writer_daemon; 104 int lfs_ignore_lazy_sync = 1; 105 106 /* Global vfs data structures for lfs. */ 107 int (**lfs_vnodeop_p)(void *); 108 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 109 { &vop_default_desc, vn_default_error }, 110 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 111 { &vop_create_desc, lfs_create }, /* create */ 112 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 113 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 114 { &vop_open_desc, ufs_open }, /* open */ 115 { &vop_close_desc, lfs_close }, /* close */ 116 { &vop_access_desc, ufs_access }, /* access */ 117 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 118 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 119 { &vop_read_desc, lfs_read }, /* read */ 120 { &vop_write_desc, lfs_write }, /* write */ 121 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 122 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 123 { &vop_poll_desc, ufs_poll }, /* poll */ 124 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 125 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 126 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 127 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 128 { &vop_seek_desc, ufs_seek }, /* seek */ 129 { &vop_remove_desc, lfs_remove }, /* remove */ 130 { &vop_link_desc, lfs_link }, /* link */ 131 { &vop_rename_desc, lfs_rename }, /* rename */ 132 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 133 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 134 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 135 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 136 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 137 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 138 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 139 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 140 { &vop_lock_desc, ufs_lock }, /* lock */ 141 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 142 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 143 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 144 { &vop_print_desc, ufs_print }, /* print */ 145 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 146 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 147 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 148 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 149 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 150 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 151 { NULL, NULL } 152 }; 153 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 154 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 155 156 int (**lfs_specop_p)(void *); 157 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 158 { &vop_default_desc, vn_default_error }, 159 { &vop_lookup_desc, spec_lookup }, /* lookup */ 160 { &vop_create_desc, spec_create }, /* create */ 161 { &vop_mknod_desc, spec_mknod }, /* mknod */ 162 { &vop_open_desc, spec_open }, /* open */ 163 { &vop_close_desc, lfsspec_close }, /* close */ 164 { &vop_access_desc, ufs_access }, /* access */ 165 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 166 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 167 { &vop_read_desc, ufsspec_read }, /* read */ 168 { &vop_write_desc, ufsspec_write }, /* write */ 169 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 170 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 171 { &vop_poll_desc, spec_poll }, /* poll */ 172 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 173 { &vop_revoke_desc, spec_revoke }, /* revoke */ 174 { &vop_mmap_desc, spec_mmap }, /* mmap */ 175 { &vop_fsync_desc, spec_fsync }, /* fsync */ 176 { &vop_seek_desc, spec_seek }, /* seek */ 177 { &vop_remove_desc, spec_remove }, /* remove */ 178 { &vop_link_desc, spec_link }, /* link */ 179 { &vop_rename_desc, spec_rename }, /* rename */ 180 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 181 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 182 { &vop_symlink_desc, spec_symlink }, /* symlink */ 183 { &vop_readdir_desc, spec_readdir }, /* readdir */ 184 { &vop_readlink_desc, spec_readlink }, /* readlink */ 185 { &vop_abortop_desc, spec_abortop }, /* abortop */ 186 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 187 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 188 { &vop_lock_desc, ufs_lock }, /* lock */ 189 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 190 { &vop_bmap_desc, spec_bmap }, /* bmap */ 191 { &vop_strategy_desc, spec_strategy }, /* strategy */ 192 { &vop_print_desc, ufs_print }, /* print */ 193 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 194 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 195 { &vop_advlock_desc, spec_advlock }, /* advlock */ 196 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 197 { &vop_getpages_desc, spec_getpages }, /* getpages */ 198 { &vop_putpages_desc, spec_putpages }, /* putpages */ 199 { NULL, NULL } 200 }; 201 const struct vnodeopv_desc lfs_specop_opv_desc = 202 { &lfs_specop_p, lfs_specop_entries }; 203 204 int (**lfs_fifoop_p)(void *); 205 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 206 { &vop_default_desc, vn_default_error }, 207 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup */ 208 { &vop_create_desc, vn_fifo_bypass }, /* create */ 209 { &vop_mknod_desc, vn_fifo_bypass }, /* mknod */ 210 { &vop_open_desc, vn_fifo_bypass }, /* open */ 211 { &vop_close_desc, lfsfifo_close }, /* close */ 212 { &vop_access_desc, ufs_access }, /* access */ 213 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 214 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 215 { &vop_read_desc, ufsfifo_read }, /* read */ 216 { &vop_write_desc, ufsfifo_write }, /* write */ 217 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */ 218 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 219 { &vop_poll_desc, vn_fifo_bypass }, /* poll */ 220 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */ 221 { &vop_revoke_desc, vn_fifo_bypass }, /* revoke */ 222 { &vop_mmap_desc, vn_fifo_bypass }, /* mmap */ 223 { &vop_fsync_desc, vn_fifo_bypass }, /* fsync */ 224 { &vop_seek_desc, vn_fifo_bypass }, /* seek */ 225 { &vop_remove_desc, vn_fifo_bypass }, /* remove */ 226 { &vop_link_desc, vn_fifo_bypass }, /* link */ 227 { &vop_rename_desc, vn_fifo_bypass }, /* rename */ 228 { &vop_mkdir_desc, vn_fifo_bypass }, /* mkdir */ 229 { &vop_rmdir_desc, vn_fifo_bypass }, /* rmdir */ 230 { &vop_symlink_desc, vn_fifo_bypass }, /* symlink */ 231 { &vop_readdir_desc, vn_fifo_bypass }, /* readdir */ 232 { &vop_readlink_desc, vn_fifo_bypass }, /* readlink */ 233 { &vop_abortop_desc, vn_fifo_bypass }, /* abortop */ 234 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 235 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 236 { &vop_lock_desc, ufs_lock }, /* lock */ 237 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 238 { &vop_bmap_desc, vn_fifo_bypass }, /* bmap */ 239 { &vop_strategy_desc, vn_fifo_bypass }, /* strategy */ 240 { &vop_print_desc, ufs_print }, /* print */ 241 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 242 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */ 243 { &vop_advlock_desc, vn_fifo_bypass }, /* advlock */ 244 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 245 { &vop_putpages_desc, vn_fifo_bypass }, /* putpages */ 246 { NULL, NULL } 247 }; 248 const struct vnodeopv_desc lfs_fifoop_opv_desc = 249 { &lfs_fifoop_p, lfs_fifoop_entries }; 250 251 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int, struct vm_page **); 252 253 #define LFS_READWRITE 254 #include <ufs/ufs/ufs_readwrite.c> 255 #undef LFS_READWRITE 256 257 /* 258 * Synch an open file. 259 */ 260 /* ARGSUSED */ 261 int 262 lfs_fsync(void *v) 263 { 264 struct vop_fsync_args /* { 265 struct vnode *a_vp; 266 kauth_cred_t a_cred; 267 int a_flags; 268 off_t offlo; 269 off_t offhi; 270 } */ *ap = v; 271 struct vnode *vp = ap->a_vp; 272 int error, wait; 273 struct inode *ip = VTOI(vp); 274 struct lfs *fs = ip->i_lfs; 275 276 /* If we're mounted read-only, don't try to sync. */ 277 if (fs->lfs_ronly) 278 return 0; 279 280 /* If a removed vnode is being cleaned, no need to sync here. */ 281 if ((ap->a_flags & FSYNC_RECLAIM) != 0 && ip->i_mode == 0) 282 return 0; 283 284 /* 285 * Trickle sync simply adds this vnode to the pager list, as if 286 * the pagedaemon had requested a pageout. 287 */ 288 if (ap->a_flags & FSYNC_LAZY) { 289 if (lfs_ignore_lazy_sync == 0) { 290 mutex_enter(&lfs_lock); 291 if (!(ip->i_flags & IN_PAGING)) { 292 ip->i_flags |= IN_PAGING; 293 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, 294 i_lfs_pchain); 295 } 296 wakeup(&lfs_writer_daemon); 297 mutex_exit(&lfs_lock); 298 } 299 return 0; 300 } 301 302 /* 303 * If a vnode is bring cleaned, flush it out before we try to 304 * reuse it. This prevents the cleaner from writing files twice 305 * in the same partial segment, causing an accounting underflow. 306 */ 307 if (ap->a_flags & FSYNC_RECLAIM && ip->i_flags & IN_CLEANING) { 308 lfs_vflush(vp); 309 } 310 311 wait = (ap->a_flags & FSYNC_WAIT); 312 do { 313 mutex_enter(&vp->v_interlock); 314 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 315 round_page(ap->a_offhi), 316 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 317 if (error == EAGAIN) { 318 mutex_enter(&lfs_lock); 319 mtsleep(&fs->lfs_avail, PCATCH | PUSER, "lfs_fsync", 320 hz / 100 + 1, &lfs_lock); 321 mutex_exit(&lfs_lock); 322 } 323 } while (error == EAGAIN); 324 if (error) 325 return error; 326 327 if ((ap->a_flags & FSYNC_DATAONLY) == 0) 328 error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 329 330 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 331 int l = 0; 332 error = VOP_IOCTL(ip->i_devvp, DIOCCACHESYNC, &l, FWRITE, 333 curlwp->l_cred); 334 } 335 if (wait && !VPISEMPTY(vp)) 336 LFS_SET_UINO(ip, IN_MODIFIED); 337 338 return error; 339 } 340 341 /* 342 * Take IN_ADIROP off, then call ufs_inactive. 343 */ 344 int 345 lfs_inactive(void *v) 346 { 347 struct vop_inactive_args /* { 348 struct vnode *a_vp; 349 } */ *ap = v; 350 351 lfs_unmark_vnode(ap->a_vp); 352 353 /* 354 * The Ifile is only ever inactivated on unmount. 355 * Streamline this process by not giving it more dirty blocks. 356 */ 357 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 358 mutex_enter(&lfs_lock); 359 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 360 mutex_exit(&lfs_lock); 361 VOP_UNLOCK(ap->a_vp); 362 return 0; 363 } 364 365 return ufs_inactive(v); 366 } 367 368 /* 369 * These macros are used to bracket UFS directory ops, so that we can 370 * identify all the pages touched during directory ops which need to 371 * be ordered and flushed atomically, so that they may be recovered. 372 * 373 * Because we have to mark nodes VU_DIROP in order to prevent 374 * the cache from reclaiming them while a dirop is in progress, we must 375 * also manage the number of nodes so marked (otherwise we can run out). 376 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 377 * is decremented during segment write, when VU_DIROP is taken off. 378 */ 379 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 380 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 381 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 382 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 383 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 384 static int lfs_set_dirop(struct vnode *, struct vnode *); 385 386 static int 387 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 388 { 389 struct lfs *fs; 390 int error; 391 392 KASSERT(VOP_ISLOCKED(dvp)); 393 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 394 395 fs = VTOI(dvp)->i_lfs; 396 397 ASSERT_NO_SEGLOCK(fs); 398 /* 399 * LFS_NRESERVE calculates direct and indirect blocks as well 400 * as an inode block; an overestimate in most cases. 401 */ 402 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 403 return (error); 404 405 restart: 406 mutex_enter(&lfs_lock); 407 if (fs->lfs_dirops == 0) { 408 mutex_exit(&lfs_lock); 409 lfs_check(dvp, LFS_UNUSED_LBN, 0); 410 mutex_enter(&lfs_lock); 411 } 412 while (fs->lfs_writer) { 413 error = mtsleep(&fs->lfs_dirops, (PRIBIO + 1) | PCATCH, 414 "lfs_sdirop", 0, &lfs_lock); 415 if (error == EINTR) { 416 mutex_exit(&lfs_lock); 417 goto unreserve; 418 } 419 } 420 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 421 wakeup(&lfs_writer_daemon); 422 mutex_exit(&lfs_lock); 423 preempt(); 424 goto restart; 425 } 426 427 if (lfs_dirvcount > LFS_MAX_DIROP) { 428 mutex_exit(&lfs_lock); 429 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 430 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 431 if ((error = mtsleep(&lfs_dirvcount, 432 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 433 &lfs_lock)) != 0) { 434 goto unreserve; 435 } 436 goto restart; 437 } 438 439 ++fs->lfs_dirops; 440 fs->lfs_doifile = 1; 441 mutex_exit(&lfs_lock); 442 443 /* Hold a reference so SET_ENDOP will be happy */ 444 vref(dvp); 445 if (vp) { 446 vref(vp); 447 MARK_VNODE(vp); 448 } 449 450 MARK_VNODE(dvp); 451 return 0; 452 453 unreserve: 454 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 455 return error; 456 } 457 458 /* 459 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 460 * in getnewvnode(), if we have a stacked filesystem mounted on top 461 * of us. 462 * 463 * NB: this means we have to clear the new vnodes on error. Fortunately 464 * SET_ENDOP is there to do that for us. 465 */ 466 static int 467 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 468 { 469 int error; 470 struct lfs *fs; 471 472 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 473 ASSERT_NO_SEGLOCK(fs); 474 if (fs->lfs_ronly) 475 return EROFS; 476 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 477 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 478 dvp, error)); 479 return error; 480 } 481 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 482 if (vpp) { 483 ungetnewvnode(*vpp); 484 *vpp = NULL; 485 } 486 return error; 487 } 488 return 0; 489 } 490 491 #define SET_ENDOP_BASE(fs, dvp, str) \ 492 do { \ 493 mutex_enter(&lfs_lock); \ 494 --(fs)->lfs_dirops; \ 495 if (!(fs)->lfs_dirops) { \ 496 if ((fs)->lfs_nadirop) { \ 497 panic("SET_ENDOP: %s: no dirops but " \ 498 " nadirop=%d", (str), \ 499 (fs)->lfs_nadirop); \ 500 } \ 501 wakeup(&(fs)->lfs_writer); \ 502 mutex_exit(&lfs_lock); \ 503 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 504 } else \ 505 mutex_exit(&lfs_lock); \ 506 } while(0) 507 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 508 do { \ 509 UNMARK_VNODE(dvp); \ 510 if (nvpp && *nvpp) \ 511 UNMARK_VNODE(*nvpp); \ 512 /* Check for error return to stem vnode leakage */ \ 513 if (nvpp && *nvpp && !((*nvpp)->v_uflag & VU_DIROP)) \ 514 ungetnewvnode(*(nvpp)); \ 515 SET_ENDOP_BASE((fs), (dvp), (str)); \ 516 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 517 vrele(dvp); \ 518 } while(0) 519 #define SET_ENDOP_CREATE_AP(ap, str) \ 520 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 521 (ap)->a_vpp, (str)) 522 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 523 do { \ 524 UNMARK_VNODE(dvp); \ 525 if (ovp) \ 526 UNMARK_VNODE(ovp); \ 527 SET_ENDOP_BASE((fs), (dvp), (str)); \ 528 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 529 vrele(dvp); \ 530 if (ovp) \ 531 vrele(ovp); \ 532 } while(0) 533 534 void 535 lfs_mark_vnode(struct vnode *vp) 536 { 537 struct inode *ip = VTOI(vp); 538 struct lfs *fs = ip->i_lfs; 539 540 mutex_enter(&lfs_lock); 541 if (!(ip->i_flag & IN_ADIROP)) { 542 if (!(vp->v_uflag & VU_DIROP)) { 543 mutex_enter(&vp->v_interlock); 544 (void)lfs_vref(vp); 545 ++lfs_dirvcount; 546 ++fs->lfs_dirvcount; 547 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 548 vp->v_uflag |= VU_DIROP; 549 } 550 ++fs->lfs_nadirop; 551 ip->i_flag |= IN_ADIROP; 552 } else 553 KASSERT(vp->v_uflag & VU_DIROP); 554 mutex_exit(&lfs_lock); 555 } 556 557 void 558 lfs_unmark_vnode(struct vnode *vp) 559 { 560 struct inode *ip = VTOI(vp); 561 562 if (ip && (ip->i_flag & IN_ADIROP)) { 563 KASSERT(vp->v_uflag & VU_DIROP); 564 mutex_enter(&lfs_lock); 565 --ip->i_lfs->lfs_nadirop; 566 mutex_exit(&lfs_lock); 567 ip->i_flag &= ~IN_ADIROP; 568 } 569 } 570 571 int 572 lfs_symlink(void *v) 573 { 574 struct vop_symlink_args /* { 575 struct vnode *a_dvp; 576 struct vnode **a_vpp; 577 struct componentname *a_cnp; 578 struct vattr *a_vap; 579 char *a_target; 580 } */ *ap = v; 581 int error; 582 583 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 584 vput(ap->a_dvp); 585 return error; 586 } 587 error = ufs_symlink(ap); 588 SET_ENDOP_CREATE_AP(ap, "symlink"); 589 return (error); 590 } 591 592 int 593 lfs_mknod(void *v) 594 { 595 struct vop_mknod_args /* { 596 struct vnode *a_dvp; 597 struct vnode **a_vpp; 598 struct componentname *a_cnp; 599 struct vattr *a_vap; 600 } */ *ap = v; 601 struct vattr *vap = ap->a_vap; 602 struct vnode **vpp = ap->a_vpp; 603 struct inode *ip; 604 int error; 605 struct mount *mp; 606 ino_t ino; 607 608 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 609 vput(ap->a_dvp); 610 return error; 611 } 612 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 613 ap->a_dvp, vpp, ap->a_cnp); 614 615 /* Either way we're done with the dirop at this point */ 616 SET_ENDOP_CREATE_AP(ap, "mknod"); 617 618 if (error) 619 return (error); 620 621 ip = VTOI(*vpp); 622 mp = (*vpp)->v_mount; 623 ino = ip->i_number; 624 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 625 if (vap->va_rdev != VNOVAL) { 626 /* 627 * Want to be able to use this to make badblock 628 * inodes, so don't truncate the dev number. 629 */ 630 #if 0 631 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 632 UFS_MPNEEDSWAP((*vpp)->v_mount)); 633 #else 634 ip->i_ffs1_rdev = vap->va_rdev; 635 #endif 636 } 637 638 /* 639 * Call fsync to write the vnode so that we don't have to deal with 640 * flushing it when it's marked VU_DIROP|VI_XLOCK. 641 * 642 * XXX KS - If we can't flush we also can't call vgone(), so must 643 * return. But, that leaves this vnode in limbo, also not good. 644 * Can this ever happen (barring hardware failure)? 645 */ 646 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0)) != 0) { 647 panic("lfs_mknod: couldn't fsync (ino %llu)", 648 (unsigned long long)ino); 649 /* return (error); */ 650 } 651 /* 652 * Remove vnode so that it will be reloaded by VFS_VGET and 653 * checked to see if it is an alias of an existing entry in 654 * the inode cache. 655 */ 656 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 657 658 VOP_UNLOCK(*vpp); 659 (*vpp)->v_type = VNON; 660 vgone(*vpp); 661 error = VFS_VGET(mp, ino, vpp); 662 663 if (error != 0) { 664 *vpp = NULL; 665 return (error); 666 } 667 return (0); 668 } 669 670 int 671 lfs_create(void *v) 672 { 673 struct vop_create_args /* { 674 struct vnode *a_dvp; 675 struct vnode **a_vpp; 676 struct componentname *a_cnp; 677 struct vattr *a_vap; 678 } */ *ap = v; 679 int error; 680 681 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 682 vput(ap->a_dvp); 683 return error; 684 } 685 error = ufs_create(ap); 686 SET_ENDOP_CREATE_AP(ap, "create"); 687 return (error); 688 } 689 690 int 691 lfs_mkdir(void *v) 692 { 693 struct vop_mkdir_args /* { 694 struct vnode *a_dvp; 695 struct vnode **a_vpp; 696 struct componentname *a_cnp; 697 struct vattr *a_vap; 698 } */ *ap = v; 699 int error; 700 701 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 702 vput(ap->a_dvp); 703 return error; 704 } 705 error = ufs_mkdir(ap); 706 SET_ENDOP_CREATE_AP(ap, "mkdir"); 707 return (error); 708 } 709 710 int 711 lfs_remove(void *v) 712 { 713 struct vop_remove_args /* { 714 struct vnode *a_dvp; 715 struct vnode *a_vp; 716 struct componentname *a_cnp; 717 } */ *ap = v; 718 struct vnode *dvp, *vp; 719 struct inode *ip; 720 int error; 721 722 dvp = ap->a_dvp; 723 vp = ap->a_vp; 724 ip = VTOI(vp); 725 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 726 if (dvp == vp) 727 vrele(vp); 728 else 729 vput(vp); 730 vput(dvp); 731 return error; 732 } 733 error = ufs_remove(ap); 734 if (ip->i_nlink == 0) 735 lfs_orphan(ip->i_lfs, ip->i_number); 736 SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove"); 737 return (error); 738 } 739 740 int 741 lfs_rmdir(void *v) 742 { 743 struct vop_rmdir_args /* { 744 struct vnodeop_desc *a_desc; 745 struct vnode *a_dvp; 746 struct vnode *a_vp; 747 struct componentname *a_cnp; 748 } */ *ap = v; 749 struct vnode *vp; 750 struct inode *ip; 751 int error; 752 753 vp = ap->a_vp; 754 ip = VTOI(vp); 755 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 756 if (ap->a_dvp == vp) 757 vrele(ap->a_dvp); 758 else 759 vput(ap->a_dvp); 760 vput(vp); 761 return error; 762 } 763 error = ufs_rmdir(ap); 764 if (ip->i_nlink == 0) 765 lfs_orphan(ip->i_lfs, ip->i_number); 766 SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 767 return (error); 768 } 769 770 int 771 lfs_link(void *v) 772 { 773 struct vop_link_args /* { 774 struct vnode *a_dvp; 775 struct vnode *a_vp; 776 struct componentname *a_cnp; 777 } */ *ap = v; 778 int error; 779 struct vnode **vpp = NULL; 780 781 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 782 vput(ap->a_dvp); 783 return error; 784 } 785 error = ufs_link(ap); 786 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 787 return (error); 788 } 789 790 int 791 lfs_rename(void *v) 792 { 793 struct vop_rename_args /* { 794 struct vnode *a_fdvp; 795 struct vnode *a_fvp; 796 struct componentname *a_fcnp; 797 struct vnode *a_tdvp; 798 struct vnode *a_tvp; 799 struct componentname *a_tcnp; 800 } */ *ap = v; 801 struct vnode *tvp, *fvp, *tdvp, *fdvp; 802 struct componentname *tcnp, *fcnp; 803 int error; 804 struct lfs *fs; 805 806 fs = VTOI(ap->a_fdvp)->i_lfs; 807 tvp = ap->a_tvp; 808 tdvp = ap->a_tdvp; 809 tcnp = ap->a_tcnp; 810 fvp = ap->a_fvp; 811 fdvp = ap->a_fdvp; 812 fcnp = ap->a_fcnp; 813 814 /* 815 * Check for cross-device rename. 816 * If it is, we don't want to set dirops, just error out. 817 * (In particular note that MARK_VNODE(tdvp) will DTWT on 818 * a cross-device rename.) 819 * 820 * Copied from ufs_rename. 821 */ 822 if ((fvp->v_mount != tdvp->v_mount) || 823 (tvp && (fvp->v_mount != tvp->v_mount))) { 824 error = EXDEV; 825 goto errout; 826 } 827 828 /* 829 * Check to make sure we're not renaming a vnode onto itself 830 * (deleting a hard link by renaming one name onto another); 831 * if we are we can't recursively call VOP_REMOVE since that 832 * would leave us with an unaccounted-for number of live dirops. 833 * 834 * Inline the relevant section of ufs_rename here, *before* 835 * calling SET_DIROP_REMOVE. 836 */ 837 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 838 (VTOI(tdvp)->i_flags & APPEND))) { 839 error = EPERM; 840 goto errout; 841 } 842 if (fvp == tvp) { 843 if (fvp->v_type == VDIR) { 844 error = EINVAL; 845 goto errout; 846 } 847 848 /* Release destination completely. */ 849 VOP_ABORTOP(tdvp, tcnp); 850 vput(tdvp); 851 vput(tvp); 852 853 /* Delete source. */ 854 vrele(fvp); 855 fcnp->cn_flags &= ~(MODMASK); 856 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 857 fcnp->cn_nameiop = DELETE; 858 vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY); 859 if ((error = relookup(fdvp, &fvp, fcnp, 0))) { 860 vput(fdvp); 861 return (error); 862 } 863 return (VOP_REMOVE(fdvp, fvp, fcnp)); 864 } 865 866 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 867 goto errout; 868 MARK_VNODE(fdvp); 869 MARK_VNODE(fvp); 870 871 error = ufs_rename(ap); 872 UNMARK_VNODE(fdvp); 873 UNMARK_VNODE(fvp); 874 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 875 return (error); 876 877 errout: 878 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 879 if (tdvp == tvp) 880 vrele(tdvp); 881 else 882 vput(tdvp); 883 if (tvp) 884 vput(tvp); 885 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 886 vrele(fdvp); 887 vrele(fvp); 888 return (error); 889 } 890 891 /* XXX hack to avoid calling ITIMES in getattr */ 892 int 893 lfs_getattr(void *v) 894 { 895 struct vop_getattr_args /* { 896 struct vnode *a_vp; 897 struct vattr *a_vap; 898 kauth_cred_t a_cred; 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 } */ *ap = v; 948 struct vnode *vp = ap->a_vp; 949 950 lfs_check(vp, LFS_UNUSED_LBN, 0); 951 return ufs_setattr(v); 952 } 953 954 /* 955 * Release the block we hold on lfs_newseg wrapping. Called on file close, 956 * or explicitly from LFCNWRAPGO. Called with the interlock held. 957 */ 958 static int 959 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor) 960 { 961 if (fs->lfs_stoplwp != curlwp) 962 return EBUSY; 963 964 fs->lfs_stoplwp = NULL; 965 cv_signal(&fs->lfs_stopcv); 966 967 KASSERT(fs->lfs_nowrap > 0); 968 if (fs->lfs_nowrap <= 0) { 969 return 0; 970 } 971 972 if (--fs->lfs_nowrap == 0) { 973 log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt); 974 wakeup(&fs->lfs_wrappass); 975 lfs_wakeup_cleaner(fs); 976 } 977 if (waitfor) { 978 mtsleep(&fs->lfs_nextseg, PCATCH | PUSER, "segment", 979 0, &lfs_lock); 980 } 981 982 return 0; 983 } 984 985 /* 986 * Close called 987 */ 988 /* ARGSUSED */ 989 int 990 lfs_close(void *v) 991 { 992 struct vop_close_args /* { 993 struct vnode *a_vp; 994 int a_fflag; 995 kauth_cred_t a_cred; 996 } */ *ap = v; 997 struct vnode *vp = ap->a_vp; 998 struct inode *ip = VTOI(vp); 999 struct lfs *fs = ip->i_lfs; 1000 1001 if ((ip->i_number == ROOTINO || ip->i_number == LFS_IFILE_INUM) && 1002 fs->lfs_stoplwp == curlwp) { 1003 mutex_enter(&lfs_lock); 1004 log(LOG_NOTICE, "lfs_close: releasing log wrap control\n"); 1005 lfs_wrapgo(fs, ip, 0); 1006 mutex_exit(&lfs_lock); 1007 } 1008 1009 if (vp == ip->i_lfs->lfs_ivnode && 1010 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 1011 return 0; 1012 1013 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 1014 LFS_ITIMES(ip, NULL, NULL, NULL); 1015 } 1016 return (0); 1017 } 1018 1019 /* 1020 * Close wrapper for special devices. 1021 * 1022 * Update the times on the inode then do device close. 1023 */ 1024 int 1025 lfsspec_close(void *v) 1026 { 1027 struct vop_close_args /* { 1028 struct vnode *a_vp; 1029 int a_fflag; 1030 kauth_cred_t a_cred; 1031 } */ *ap = v; 1032 struct vnode *vp; 1033 struct inode *ip; 1034 1035 vp = ap->a_vp; 1036 ip = VTOI(vp); 1037 if (vp->v_usecount > 1) { 1038 LFS_ITIMES(ip, NULL, NULL, NULL); 1039 } 1040 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1041 } 1042 1043 /* 1044 * Close wrapper for fifo's. 1045 * 1046 * Update the times on the inode then do device close. 1047 */ 1048 int 1049 lfsfifo_close(void *v) 1050 { 1051 struct vop_close_args /* { 1052 struct vnode *a_vp; 1053 int a_fflag; 1054 kauth_cred_ a_cred; 1055 } */ *ap = v; 1056 struct vnode *vp; 1057 struct inode *ip; 1058 1059 vp = ap->a_vp; 1060 ip = VTOI(vp); 1061 if (ap->a_vp->v_usecount > 1) { 1062 LFS_ITIMES(ip, NULL, NULL, NULL); 1063 } 1064 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1065 } 1066 1067 /* 1068 * Reclaim an inode so that it can be used for other purposes. 1069 */ 1070 1071 int 1072 lfs_reclaim(void *v) 1073 { 1074 struct vop_reclaim_args /* { 1075 struct vnode *a_vp; 1076 } */ *ap = v; 1077 struct vnode *vp = ap->a_vp; 1078 struct inode *ip = VTOI(vp); 1079 struct lfs *fs = ip->i_lfs; 1080 int error; 1081 1082 /* 1083 * The inode must be freed and updated before being removed 1084 * from its hash chain. Other threads trying to gain a hold 1085 * on the inode will be stalled because it is locked (VI_XLOCK). 1086 */ 1087 if (ip->i_nlink <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) 1088 lfs_vfree(vp, ip->i_number, ip->i_omode); 1089 1090 mutex_enter(&lfs_lock); 1091 LFS_CLR_UINO(ip, IN_ALLMOD); 1092 mutex_exit(&lfs_lock); 1093 if ((error = ufs_reclaim(vp))) 1094 return (error); 1095 1096 /* 1097 * Take us off the paging and/or dirop queues if we were on them. 1098 * We shouldn't be on them. 1099 */ 1100 mutex_enter(&lfs_lock); 1101 if (ip->i_flags & IN_PAGING) { 1102 log(LOG_WARNING, "%s: reclaimed vnode is IN_PAGING\n", 1103 fs->lfs_fsmnt); 1104 ip->i_flags &= ~IN_PAGING; 1105 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1106 } 1107 if (vp->v_uflag & VU_DIROP) { 1108 panic("reclaimed vnode is VU_DIROP"); 1109 vp->v_uflag &= ~VU_DIROP; 1110 TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain); 1111 } 1112 mutex_exit(&lfs_lock); 1113 1114 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1115 lfs_deregister_all(vp); 1116 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1117 ip->inode_ext.lfs = NULL; 1118 genfs_node_destroy(vp); 1119 pool_put(&lfs_inode_pool, vp->v_data); 1120 vp->v_data = NULL; 1121 return (0); 1122 } 1123 1124 /* 1125 * Read a block from a storage device. 1126 * In order to avoid reading blocks that are in the process of being 1127 * written by the cleaner---and hence are not mutexed by the normal 1128 * buffer cache / page cache mechanisms---check for collisions before 1129 * reading. 1130 * 1131 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1132 * the active cleaner test. 1133 * 1134 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1135 */ 1136 int 1137 lfs_strategy(void *v) 1138 { 1139 struct vop_strategy_args /* { 1140 struct vnode *a_vp; 1141 struct buf *a_bp; 1142 } */ *ap = v; 1143 struct buf *bp; 1144 struct lfs *fs; 1145 struct vnode *vp; 1146 struct inode *ip; 1147 daddr_t tbn; 1148 int i, sn, error, slept; 1149 1150 bp = ap->a_bp; 1151 vp = ap->a_vp; 1152 ip = VTOI(vp); 1153 fs = ip->i_lfs; 1154 1155 /* lfs uses its strategy routine only for read */ 1156 KASSERT(bp->b_flags & B_READ); 1157 1158 if (vp->v_type == VBLK || vp->v_type == VCHR) 1159 panic("lfs_strategy: spec"); 1160 KASSERT(bp->b_bcount != 0); 1161 if (bp->b_blkno == bp->b_lblkno) { 1162 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1163 NULL); 1164 if (error) { 1165 bp->b_error = error; 1166 bp->b_resid = bp->b_bcount; 1167 biodone(bp); 1168 return (error); 1169 } 1170 if ((long)bp->b_blkno == -1) /* no valid data */ 1171 clrbuf(bp); 1172 } 1173 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1174 bp->b_resid = bp->b_bcount; 1175 biodone(bp); 1176 return (0); 1177 } 1178 1179 slept = 1; 1180 mutex_enter(&lfs_lock); 1181 while (slept && fs->lfs_seglock) { 1182 mutex_exit(&lfs_lock); 1183 /* 1184 * Look through list of intervals. 1185 * There will only be intervals to look through 1186 * if the cleaner holds the seglock. 1187 * Since the cleaner is synchronous, we can trust 1188 * the list of intervals to be current. 1189 */ 1190 tbn = dbtofsb(fs, bp->b_blkno); 1191 sn = dtosn(fs, tbn); 1192 slept = 0; 1193 for (i = 0; i < fs->lfs_cleanind; i++) { 1194 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1195 tbn >= fs->lfs_cleanint[i]) { 1196 DLOG((DLOG_CLEAN, 1197 "lfs_strategy: ino %d lbn %" PRId64 1198 " ind %d sn %d fsb %" PRIx32 1199 " given sn %d fsb %" PRIx64 "\n", 1200 ip->i_number, bp->b_lblkno, i, 1201 dtosn(fs, fs->lfs_cleanint[i]), 1202 fs->lfs_cleanint[i], sn, tbn)); 1203 DLOG((DLOG_CLEAN, 1204 "lfs_strategy: sleeping on ino %d lbn %" 1205 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1206 mutex_enter(&lfs_lock); 1207 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1208 /* Cleaner can't wait for itself */ 1209 mtsleep(&fs->lfs_iocount, 1210 (PRIBIO + 1) | PNORELOCK, 1211 "clean2", 0, 1212 &lfs_lock); 1213 slept = 1; 1214 break; 1215 } else if (fs->lfs_seglock) { 1216 mtsleep(&fs->lfs_seglock, 1217 (PRIBIO + 1) | PNORELOCK, 1218 "clean1", 0, 1219 &lfs_lock); 1220 slept = 1; 1221 break; 1222 } 1223 mutex_exit(&lfs_lock); 1224 } 1225 } 1226 mutex_enter(&lfs_lock); 1227 } 1228 mutex_exit(&lfs_lock); 1229 1230 vp = ip->i_devvp; 1231 VOP_STRATEGY(vp, bp); 1232 return (0); 1233 } 1234 1235 void 1236 lfs_flush_dirops(struct lfs *fs) 1237 { 1238 struct inode *ip, *nip; 1239 struct vnode *vp; 1240 extern int lfs_dostats; 1241 struct segment *sp; 1242 1243 ASSERT_MAYBE_SEGLOCK(fs); 1244 KASSERT(fs->lfs_nadirop == 0); 1245 1246 if (fs->lfs_ronly) 1247 return; 1248 1249 mutex_enter(&lfs_lock); 1250 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1251 mutex_exit(&lfs_lock); 1252 return; 1253 } else 1254 mutex_exit(&lfs_lock); 1255 1256 if (lfs_dostats) 1257 ++lfs_stats.flush_invoked; 1258 1259 /* 1260 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1261 * Technically this is a checkpoint (the on-disk state is valid) 1262 * even though we are leaving out all the file data. 1263 */ 1264 lfs_imtime(fs); 1265 lfs_seglock(fs, SEGM_CKP); 1266 sp = fs->lfs_sp; 1267 1268 /* 1269 * lfs_writevnodes, optimized to get dirops out of the way. 1270 * Only write dirops, and don't flush files' pages, only 1271 * blocks from the directories. 1272 * 1273 * We don't need to vref these files because they are 1274 * dirops and so hold an extra reference until the 1275 * segunlock clears them of that status. 1276 * 1277 * We don't need to check for IN_ADIROP because we know that 1278 * no dirops are active. 1279 * 1280 */ 1281 mutex_enter(&lfs_lock); 1282 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1283 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1284 mutex_exit(&lfs_lock); 1285 vp = ITOV(ip); 1286 1287 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1288 1289 /* 1290 * All writes to directories come from dirops; all 1291 * writes to files' direct blocks go through the page 1292 * cache, which we're not touching. Reads to files 1293 * and/or directories will not be affected by writing 1294 * directory blocks inodes and file inodes. So we don't 1295 * really need to lock. If we don't lock, though, 1296 * make sure that we don't clear IN_MODIFIED 1297 * unnecessarily. 1298 */ 1299 if (vp->v_iflag & VI_XLOCK) { 1300 mutex_enter(&lfs_lock); 1301 continue; 1302 } 1303 /* XXX see below 1304 * waslocked = VOP_ISLOCKED(vp); 1305 */ 1306 if (vp->v_type != VREG && 1307 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1308 lfs_writefile(fs, sp, vp); 1309 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1310 !(ip->i_flag & IN_ALLMOD)) { 1311 mutex_enter(&lfs_lock); 1312 LFS_SET_UINO(ip, IN_MODIFIED); 1313 mutex_exit(&lfs_lock); 1314 } 1315 } 1316 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1317 (void) lfs_writeinode(fs, sp, ip); 1318 mutex_enter(&lfs_lock); 1319 /* 1320 * XXX 1321 * LK_EXCLOTHER is dead -- what is intended here? 1322 * if (waslocked == LK_EXCLOTHER) 1323 * LFS_SET_UINO(ip, IN_MODIFIED); 1324 */ 1325 } 1326 mutex_exit(&lfs_lock); 1327 /* We've written all the dirops there are */ 1328 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1329 lfs_finalize_fs_seguse(fs); 1330 (void) lfs_writeseg(fs, sp); 1331 lfs_segunlock(fs); 1332 } 1333 1334 /* 1335 * Flush all vnodes for which the pagedaemon has requested pageouts. 1336 * Skip over any files that are marked VU_DIROP (since lfs_flush_dirop() 1337 * has just run, this would be an error). If we have to skip a vnode 1338 * for any reason, just skip it; if we have to wait for the cleaner, 1339 * abort. The writer daemon will call us again later. 1340 */ 1341 void 1342 lfs_flush_pchain(struct lfs *fs) 1343 { 1344 struct inode *ip, *nip; 1345 struct vnode *vp; 1346 extern int lfs_dostats; 1347 struct segment *sp; 1348 int error; 1349 1350 ASSERT_NO_SEGLOCK(fs); 1351 1352 if (fs->lfs_ronly) 1353 return; 1354 1355 mutex_enter(&lfs_lock); 1356 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1357 mutex_exit(&lfs_lock); 1358 return; 1359 } else 1360 mutex_exit(&lfs_lock); 1361 1362 /* Get dirops out of the way */ 1363 lfs_flush_dirops(fs); 1364 1365 if (lfs_dostats) 1366 ++lfs_stats.flush_invoked; 1367 1368 /* 1369 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1370 */ 1371 lfs_imtime(fs); 1372 lfs_seglock(fs, 0); 1373 sp = fs->lfs_sp; 1374 1375 /* 1376 * lfs_writevnodes, optimized to clear pageout requests. 1377 * Only write non-dirop files that are in the pageout queue. 1378 * We're very conservative about what we write; we want to be 1379 * fast and async. 1380 */ 1381 mutex_enter(&lfs_lock); 1382 top: 1383 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1384 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1385 vp = ITOV(ip); 1386 1387 if (!(ip->i_flags & IN_PAGING)) 1388 goto top; 1389 1390 mutex_enter(&vp->v_interlock); 1391 if ((vp->v_iflag & VI_XLOCK) || (vp->v_uflag & VU_DIROP) != 0) { 1392 mutex_exit(&vp->v_interlock); 1393 continue; 1394 } 1395 if (vp->v_type != VREG) { 1396 mutex_exit(&vp->v_interlock); 1397 continue; 1398 } 1399 if (lfs_vref(vp)) 1400 continue; 1401 mutex_exit(&lfs_lock); 1402 1403 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_RETRY) != 0) { 1404 lfs_vunref(vp); 1405 mutex_enter(&lfs_lock); 1406 continue; 1407 } 1408 1409 error = lfs_writefile(fs, sp, vp); 1410 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1411 !(ip->i_flag & IN_ALLMOD)) { 1412 mutex_enter(&lfs_lock); 1413 LFS_SET_UINO(ip, IN_MODIFIED); 1414 mutex_exit(&lfs_lock); 1415 } 1416 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1417 (void) lfs_writeinode(fs, sp, ip); 1418 1419 VOP_UNLOCK(vp); 1420 lfs_vunref(vp); 1421 1422 if (error == EAGAIN) { 1423 lfs_writeseg(fs, sp); 1424 mutex_enter(&lfs_lock); 1425 break; 1426 } 1427 mutex_enter(&lfs_lock); 1428 } 1429 mutex_exit(&lfs_lock); 1430 (void) lfs_writeseg(fs, sp); 1431 lfs_segunlock(fs); 1432 } 1433 1434 /* 1435 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1436 */ 1437 int 1438 lfs_fcntl(void *v) 1439 { 1440 struct vop_fcntl_args /* { 1441 struct vnode *a_vp; 1442 u_int a_command; 1443 void * a_data; 1444 int a_fflag; 1445 kauth_cred_t a_cred; 1446 } */ *ap = v; 1447 struct timeval tv; 1448 struct timeval *tvp; 1449 BLOCK_INFO *blkiov; 1450 CLEANERINFO *cip; 1451 SEGUSE *sup; 1452 int blkcnt, error, oclean; 1453 size_t fh_size; 1454 struct lfs_fcntl_markv blkvp; 1455 struct lwp *l; 1456 fsid_t *fsidp; 1457 struct lfs *fs; 1458 struct buf *bp; 1459 fhandle_t *fhp; 1460 daddr_t off; 1461 1462 /* Only respect LFS fcntls on fs root or Ifile */ 1463 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1464 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1465 return ufs_fcntl(v); 1466 } 1467 1468 /* Avoid locking a draining lock */ 1469 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1470 return ESHUTDOWN; 1471 } 1472 1473 /* LFS control and monitoring fcntls are available only to root */ 1474 l = curlwp; 1475 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1476 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1477 NULL)) != 0) 1478 return (error); 1479 1480 fs = VTOI(ap->a_vp)->i_lfs; 1481 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1482 1483 error = 0; 1484 switch ((int)ap->a_command) { 1485 case LFCNSEGWAITALL_COMPAT_50: 1486 case LFCNSEGWAITALL_COMPAT: 1487 fsidp = NULL; 1488 /* FALLSTHROUGH */ 1489 case LFCNSEGWAIT_COMPAT_50: 1490 case LFCNSEGWAIT_COMPAT: 1491 { 1492 struct timeval50 *tvp50 1493 = (struct timeval50 *)ap->a_data; 1494 timeval50_to_timeval(tvp50, &tv); 1495 tvp = &tv; 1496 } 1497 goto segwait_common; 1498 case LFCNSEGWAITALL: 1499 fsidp = NULL; 1500 /* FALLSTHROUGH */ 1501 case LFCNSEGWAIT: 1502 tvp = (struct timeval *)ap->a_data; 1503 segwait_common: 1504 mutex_enter(&lfs_lock); 1505 ++fs->lfs_sleepers; 1506 mutex_exit(&lfs_lock); 1507 1508 error = lfs_segwait(fsidp, tvp); 1509 1510 mutex_enter(&lfs_lock); 1511 if (--fs->lfs_sleepers == 0) 1512 wakeup(&fs->lfs_sleepers); 1513 mutex_exit(&lfs_lock); 1514 return error; 1515 1516 case LFCNBMAPV: 1517 case LFCNMARKV: 1518 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1519 1520 blkcnt = blkvp.blkcnt; 1521 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1522 return (EINVAL); 1523 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1524 if ((error = copyin(blkvp.blkiov, blkiov, 1525 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1526 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1527 return error; 1528 } 1529 1530 mutex_enter(&lfs_lock); 1531 ++fs->lfs_sleepers; 1532 mutex_exit(&lfs_lock); 1533 if (ap->a_command == LFCNBMAPV) 1534 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1535 else /* LFCNMARKV */ 1536 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1537 if (error == 0) 1538 error = copyout(blkiov, blkvp.blkiov, 1539 blkcnt * sizeof(BLOCK_INFO)); 1540 mutex_enter(&lfs_lock); 1541 if (--fs->lfs_sleepers == 0) 1542 wakeup(&fs->lfs_sleepers); 1543 mutex_exit(&lfs_lock); 1544 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1545 return error; 1546 1547 case LFCNRECLAIM: 1548 /* 1549 * Flush dirops and write Ifile, allowing empty segments 1550 * to be immediately reclaimed. 1551 */ 1552 lfs_writer_enter(fs, "pndirop"); 1553 off = fs->lfs_offset; 1554 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1555 lfs_flush_dirops(fs); 1556 LFS_CLEANERINFO(cip, fs, bp); 1557 oclean = cip->clean; 1558 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1559 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1560 fs->lfs_sp->seg_flags |= SEGM_PROT; 1561 lfs_segunlock(fs); 1562 lfs_writer_leave(fs); 1563 1564 #ifdef DEBUG 1565 LFS_CLEANERINFO(cip, fs, bp); 1566 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1567 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1568 fs->lfs_offset - off, cip->clean - oclean, 1569 fs->lfs_activesb)); 1570 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1571 #endif 1572 1573 return 0; 1574 1575 case LFCNIFILEFH_COMPAT: 1576 /* Return the filehandle of the Ifile */ 1577 if ((error = kauth_authorize_system(l->l_cred, 1578 KAUTH_SYSTEM_FILEHANDLE, 0, NULL, NULL, NULL)) != 0) 1579 return (error); 1580 fhp = (struct fhandle *)ap->a_data; 1581 fhp->fh_fsid = *fsidp; 1582 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1583 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1584 1585 case LFCNIFILEFH_COMPAT2: 1586 case LFCNIFILEFH: 1587 /* Return the filehandle of the Ifile */ 1588 fhp = (struct fhandle *)ap->a_data; 1589 fhp->fh_fsid = *fsidp; 1590 fh_size = sizeof(struct lfs_fhandle) - 1591 offsetof(fhandle_t, fh_fid); 1592 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1593 1594 case LFCNREWIND: 1595 /* Move lfs_offset to the lowest-numbered segment */ 1596 return lfs_rewind(fs, *(int *)ap->a_data); 1597 1598 case LFCNINVAL: 1599 /* Mark a segment SEGUSE_INVAL */ 1600 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1601 if (sup->su_nbytes > 0) { 1602 brelse(bp, 0); 1603 lfs_unset_inval_all(fs); 1604 return EBUSY; 1605 } 1606 sup->su_flags |= SEGUSE_INVAL; 1607 VOP_BWRITE(bp); 1608 return 0; 1609 1610 case LFCNRESIZE: 1611 /* Resize the filesystem */ 1612 return lfs_resize_fs(fs, *(int *)ap->a_data); 1613 1614 case LFCNWRAPSTOP: 1615 case LFCNWRAPSTOP_COMPAT: 1616 /* 1617 * Hold lfs_newseg at segment 0; if requested, sleep until 1618 * the filesystem wraps around. To support external agents 1619 * (dump, fsck-based regression test) that need to look at 1620 * a snapshot of the filesystem, without necessarily 1621 * requiring that all fs activity stops. 1622 */ 1623 if (fs->lfs_stoplwp == curlwp) 1624 return EALREADY; 1625 1626 mutex_enter(&lfs_lock); 1627 while (fs->lfs_stoplwp != NULL) 1628 cv_wait(&fs->lfs_stopcv, &lfs_lock); 1629 fs->lfs_stoplwp = curlwp; 1630 if (fs->lfs_nowrap == 0) 1631 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1632 ++fs->lfs_nowrap; 1633 if (*(int *)ap->a_data == 1 1634 || ap->a_command == LFCNWRAPSTOP_COMPAT) { 1635 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1636 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1637 "segwrap", 0, &lfs_lock); 1638 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1639 if (error) { 1640 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1641 } 1642 } 1643 mutex_exit(&lfs_lock); 1644 return 0; 1645 1646 case LFCNWRAPGO: 1647 case LFCNWRAPGO_COMPAT: 1648 /* 1649 * Having done its work, the agent wakes up the writer. 1650 * If the argument is 1, it sleeps until a new segment 1651 * is selected. 1652 */ 1653 mutex_enter(&lfs_lock); 1654 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1655 ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1656 *((int *)ap->a_data)); 1657 mutex_exit(&lfs_lock); 1658 return error; 1659 1660 case LFCNWRAPPASS: 1661 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1662 return EALREADY; 1663 mutex_enter(&lfs_lock); 1664 if (fs->lfs_stoplwp != curlwp) { 1665 mutex_exit(&lfs_lock); 1666 return EALREADY; 1667 } 1668 if (fs->lfs_nowrap == 0) { 1669 mutex_exit(&lfs_lock); 1670 return EBUSY; 1671 } 1672 fs->lfs_wrappass = 1; 1673 wakeup(&fs->lfs_wrappass); 1674 /* Wait for the log to wrap, if asked */ 1675 if (*(int *)ap->a_data) { 1676 mutex_enter(&ap->a_vp->v_interlock); 1677 lfs_vref(ap->a_vp); 1678 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1679 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1680 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1681 "segwrap", 0, &lfs_lock); 1682 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1683 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1684 lfs_vunref(ap->a_vp); 1685 } 1686 mutex_exit(&lfs_lock); 1687 return error; 1688 1689 case LFCNWRAPSTATUS: 1690 mutex_enter(&lfs_lock); 1691 *(int *)ap->a_data = fs->lfs_wrapstatus; 1692 mutex_exit(&lfs_lock); 1693 return 0; 1694 1695 default: 1696 return ufs_fcntl(v); 1697 } 1698 return 0; 1699 } 1700 1701 int 1702 lfs_getpages(void *v) 1703 { 1704 struct vop_getpages_args /* { 1705 struct vnode *a_vp; 1706 voff_t a_offset; 1707 struct vm_page **a_m; 1708 int *a_count; 1709 int a_centeridx; 1710 vm_prot_t a_access_type; 1711 int a_advice; 1712 int a_flags; 1713 } */ *ap = v; 1714 1715 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1716 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1717 return EPERM; 1718 } 1719 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1720 mutex_enter(&lfs_lock); 1721 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1722 mutex_exit(&lfs_lock); 1723 } 1724 1725 /* 1726 * we're relying on the fact that genfs_getpages() always read in 1727 * entire filesystem blocks. 1728 */ 1729 return genfs_getpages(v); 1730 } 1731 1732 /* 1733 * Wait for a page to become unbusy, possibly printing diagnostic messages 1734 * as well. 1735 * 1736 * Called with vp->v_interlock held; return with it held. 1737 */ 1738 static void 1739 wait_for_page(struct vnode *vp, struct vm_page *pg, const char *label) 1740 { 1741 if ((pg->flags & PG_BUSY) == 0) 1742 return; /* Nothing to wait for! */ 1743 1744 #if defined(DEBUG) && defined(UVM_PAGE_TRKOWN) 1745 static struct vm_page *lastpg; 1746 1747 if (label != NULL && pg != lastpg) { 1748 if (pg->owner_tag) { 1749 printf("lfs_putpages[%d.%d]: %s: page %p owner %d.%d [%s]\n", 1750 curproc->p_pid, curlwp->l_lid, label, 1751 pg, pg->owner, pg->lowner, pg->owner_tag); 1752 } else { 1753 printf("lfs_putpages[%d.%d]: %s: page %p unowned?!\n", 1754 curproc->p_pid, curlwp->l_lid, label, pg); 1755 } 1756 } 1757 lastpg = pg; 1758 #endif 1759 1760 pg->flags |= PG_WANTED; 1761 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, "lfsput", 0); 1762 mutex_enter(&vp->v_interlock); 1763 } 1764 1765 /* 1766 * This routine is called by lfs_putpages() when it can't complete the 1767 * write because a page is busy. This means that either (1) someone, 1768 * possibly the pagedaemon, is looking at this page, and will give it up 1769 * presently; or (2) we ourselves are holding the page busy in the 1770 * process of being written (either gathered or actually on its way to 1771 * disk). We don't need to give up the segment lock, but we might need 1772 * to call lfs_writeseg() to expedite the page's journey to disk. 1773 * 1774 * Called with vp->v_interlock held; return with it held. 1775 */ 1776 /* #define BUSYWAIT */ 1777 static void 1778 write_and_wait(struct lfs *fs, struct vnode *vp, struct vm_page *pg, 1779 int seglocked, const char *label) 1780 { 1781 #ifndef BUSYWAIT 1782 struct inode *ip = VTOI(vp); 1783 struct segment *sp = fs->lfs_sp; 1784 int count = 0; 1785 1786 if (pg == NULL) 1787 return; 1788 1789 while (pg->flags & PG_BUSY && 1790 pg->uobject == &vp->v_uobj) { 1791 mutex_exit(&vp->v_interlock); 1792 if (sp->cbpp - sp->bpp > 1) { 1793 /* Write gathered pages */ 1794 lfs_updatemeta(sp); 1795 lfs_release_finfo(fs); 1796 (void) lfs_writeseg(fs, sp); 1797 1798 /* 1799 * Reinitialize FIP 1800 */ 1801 KASSERT(sp->vp == vp); 1802 lfs_acquire_finfo(fs, ip->i_number, 1803 ip->i_gen); 1804 } 1805 ++count; 1806 mutex_enter(&vp->v_interlock); 1807 wait_for_page(vp, pg, label); 1808 } 1809 if (label != NULL && count > 1) 1810 printf("lfs_putpages[%d]: %s: %sn = %d\n", curproc->p_pid, 1811 label, (count > 0 ? "looping, " : ""), count); 1812 #else 1813 preempt(1); 1814 #endif 1815 } 1816 1817 /* 1818 * Make sure that for all pages in every block in the given range, 1819 * either all are dirty or all are clean. If any of the pages 1820 * we've seen so far are dirty, put the vnode on the paging chain, 1821 * and mark it IN_PAGING. 1822 * 1823 * If checkfirst != 0, don't check all the pages but return at the 1824 * first dirty page. 1825 */ 1826 static int 1827 check_dirty(struct lfs *fs, struct vnode *vp, 1828 off_t startoffset, off_t endoffset, off_t blkeof, 1829 int flags, int checkfirst, struct vm_page **pgp) 1830 { 1831 int by_list; 1832 struct vm_page *curpg = NULL; /* XXX: gcc */ 1833 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1834 off_t soff = 0; /* XXX: gcc */ 1835 voff_t off; 1836 int i; 1837 int nonexistent; 1838 int any_dirty; /* number of dirty pages */ 1839 int dirty; /* number of dirty pages in a block */ 1840 int tdirty; 1841 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1842 int pagedaemon = (curlwp == uvm.pagedaemon_lwp); 1843 1844 ASSERT_MAYBE_SEGLOCK(fs); 1845 top: 1846 by_list = (vp->v_uobj.uo_npages <= 1847 ((endoffset - startoffset) >> PAGE_SHIFT) * 1848 UVM_PAGE_TREE_PENALTY); 1849 any_dirty = 0; 1850 1851 if (by_list) { 1852 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1853 } else { 1854 soff = startoffset; 1855 } 1856 while (by_list || soff < MIN(blkeof, endoffset)) { 1857 if (by_list) { 1858 /* 1859 * Find the first page in a block. Skip 1860 * blocks outside our area of interest or beyond 1861 * the end of file. 1862 */ 1863 KASSERT(curpg == NULL 1864 || (curpg->flags & PG_MARKER) == 0); 1865 if (pages_per_block > 1) { 1866 while (curpg && 1867 ((curpg->offset & fs->lfs_bmask) || 1868 curpg->offset >= vp->v_size || 1869 curpg->offset >= endoffset)) { 1870 curpg = TAILQ_NEXT(curpg, listq.queue); 1871 KASSERT(curpg == NULL || 1872 (curpg->flags & PG_MARKER) == 0); 1873 } 1874 } 1875 if (curpg == NULL) 1876 break; 1877 soff = curpg->offset; 1878 } 1879 1880 /* 1881 * Mark all pages in extended range busy; find out if any 1882 * of them are dirty. 1883 */ 1884 nonexistent = dirty = 0; 1885 for (i = 0; i == 0 || i < pages_per_block; i++) { 1886 if (by_list && pages_per_block <= 1) { 1887 pgs[i] = pg = curpg; 1888 } else { 1889 off = soff + (i << PAGE_SHIFT); 1890 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1891 if (pg == NULL) { 1892 ++nonexistent; 1893 continue; 1894 } 1895 } 1896 KASSERT(pg != NULL); 1897 1898 /* 1899 * If we're holding the segment lock, we can deadlock 1900 * against a process that has our page and is waiting 1901 * for the cleaner, while the cleaner waits for the 1902 * segment lock. Just bail in that case. 1903 */ 1904 if ((pg->flags & PG_BUSY) && 1905 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1906 if (i > 0) 1907 uvm_page_unbusy(pgs, i); 1908 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1909 if (pgp) 1910 *pgp = pg; 1911 return -1; 1912 } 1913 1914 while (pg->flags & PG_BUSY) { 1915 wait_for_page(vp, pg, NULL); 1916 if (i > 0) 1917 uvm_page_unbusy(pgs, i); 1918 goto top; 1919 } 1920 pg->flags |= PG_BUSY; 1921 UVM_PAGE_OWN(pg, "lfs_putpages"); 1922 1923 pmap_page_protect(pg, VM_PROT_NONE); 1924 tdirty = (pmap_clear_modify(pg) || 1925 (pg->flags & PG_CLEAN) == 0); 1926 dirty += tdirty; 1927 } 1928 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1929 if (by_list) { 1930 curpg = TAILQ_NEXT(curpg, listq.queue); 1931 } else { 1932 soff += fs->lfs_bsize; 1933 } 1934 continue; 1935 } 1936 1937 any_dirty += dirty; 1938 KASSERT(nonexistent == 0); 1939 1940 /* 1941 * If any are dirty make all dirty; unbusy them, 1942 * but if we were asked to clean, wire them so that 1943 * the pagedaemon doesn't bother us about them while 1944 * they're on their way to disk. 1945 */ 1946 for (i = 0; i == 0 || i < pages_per_block; i++) { 1947 pg = pgs[i]; 1948 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1949 if (dirty) { 1950 pg->flags &= ~PG_CLEAN; 1951 if (flags & PGO_FREE) { 1952 /* 1953 * Wire the page so that 1954 * pdaemon doesn't see it again. 1955 */ 1956 mutex_enter(&uvm_pageqlock); 1957 uvm_pagewire(pg); 1958 mutex_exit(&uvm_pageqlock); 1959 1960 /* Suspended write flag */ 1961 pg->flags |= PG_DELWRI; 1962 } 1963 } 1964 if (pg->flags & PG_WANTED) 1965 wakeup(pg); 1966 pg->flags &= ~(PG_WANTED|PG_BUSY); 1967 UVM_PAGE_OWN(pg, NULL); 1968 } 1969 1970 if (checkfirst && any_dirty) 1971 break; 1972 1973 if (by_list) { 1974 curpg = TAILQ_NEXT(curpg, listq.queue); 1975 } else { 1976 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1977 } 1978 } 1979 1980 return any_dirty; 1981 } 1982 1983 /* 1984 * lfs_putpages functions like genfs_putpages except that 1985 * 1986 * (1) It needs to bounds-check the incoming requests to ensure that 1987 * they are block-aligned; if they are not, expand the range and 1988 * do the right thing in case, e.g., the requested range is clean 1989 * but the expanded range is dirty. 1990 * 1991 * (2) It needs to explicitly send blocks to be written when it is done. 1992 * If VOP_PUTPAGES is called without the seglock held, we simply take 1993 * the seglock and let lfs_segunlock wait for us. 1994 * XXX There might be a bad situation if we have to flush a vnode while 1995 * XXX lfs_markv is in operation. As of this writing we panic in this 1996 * XXX case. 1997 * 1998 * Assumptions: 1999 * 2000 * (1) The caller does not hold any pages in this vnode busy. If it does, 2001 * there is a danger that when we expand the page range and busy the 2002 * pages we will deadlock. 2003 * 2004 * (2) We are called with vp->v_interlock held; we must return with it 2005 * released. 2006 * 2007 * (3) We don't absolutely have to free pages right away, provided that 2008 * the request does not have PGO_SYNCIO. When the pagedaemon gives 2009 * us a request with PGO_FREE, we take the pages out of the paging 2010 * queue and wake up the writer, which will handle freeing them for us. 2011 * 2012 * We ensure that for any filesystem block, all pages for that 2013 * block are either resident or not, even if those pages are higher 2014 * than EOF; that means that we will be getting requests to free 2015 * "unused" pages above EOF all the time, and should ignore them. 2016 * 2017 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 2018 * into has been set up for us by lfs_writefile. If not, we will 2019 * have to handle allocating and/or freeing an finfo entry. 2020 * 2021 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 2022 */ 2023 2024 /* How many times to loop before we should start to worry */ 2025 #define TOOMANY 4 2026 2027 int 2028 lfs_putpages(void *v) 2029 { 2030 int error; 2031 struct vop_putpages_args /* { 2032 struct vnode *a_vp; 2033 voff_t a_offlo; 2034 voff_t a_offhi; 2035 int a_flags; 2036 } */ *ap = v; 2037 struct vnode *vp; 2038 struct inode *ip; 2039 struct lfs *fs; 2040 struct segment *sp; 2041 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 2042 off_t off, max_endoffset; 2043 bool seglocked, sync, pagedaemon; 2044 struct vm_page *pg, *busypg; 2045 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 2046 #ifdef DEBUG 2047 int debug_n_again, debug_n_dirtyclean; 2048 #endif 2049 2050 vp = ap->a_vp; 2051 ip = VTOI(vp); 2052 fs = ip->i_lfs; 2053 sync = (ap->a_flags & PGO_SYNCIO) != 0; 2054 pagedaemon = (curlwp == uvm.pagedaemon_lwp); 2055 2056 /* Putpages does nothing for metadata. */ 2057 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 2058 mutex_exit(&vp->v_interlock); 2059 return 0; 2060 } 2061 2062 /* 2063 * If there are no pages, don't do anything. 2064 */ 2065 if (vp->v_uobj.uo_npages == 0) { 2066 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 2067 (vp->v_iflag & VI_ONWORKLST) && 2068 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 2069 vp->v_iflag &= ~VI_WRMAPDIRTY; 2070 vn_syncer_remove_from_worklist(vp); 2071 } 2072 mutex_exit(&vp->v_interlock); 2073 2074 /* Remove us from paging queue, if we were on it */ 2075 mutex_enter(&lfs_lock); 2076 if (ip->i_flags & IN_PAGING) { 2077 ip->i_flags &= ~IN_PAGING; 2078 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2079 } 2080 mutex_exit(&lfs_lock); 2081 return 0; 2082 } 2083 2084 blkeof = blkroundup(fs, ip->i_size); 2085 2086 /* 2087 * Ignore requests to free pages past EOF but in the same block 2088 * as EOF, unless the request is synchronous. (If the request is 2089 * sync, it comes from lfs_truncate.) 2090 * XXXUBC Make these pages look "active" so the pagedaemon won't 2091 * XXXUBC bother us with them again. 2092 */ 2093 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 2094 origoffset = ap->a_offlo; 2095 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 2096 pg = uvm_pagelookup(&vp->v_uobj, off); 2097 KASSERT(pg != NULL); 2098 while (pg->flags & PG_BUSY) { 2099 pg->flags |= PG_WANTED; 2100 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 2101 "lfsput2", 0); 2102 mutex_enter(&vp->v_interlock); 2103 } 2104 mutex_enter(&uvm_pageqlock); 2105 uvm_pageactivate(pg); 2106 mutex_exit(&uvm_pageqlock); 2107 } 2108 ap->a_offlo = blkeof; 2109 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 2110 mutex_exit(&vp->v_interlock); 2111 return 0; 2112 } 2113 } 2114 2115 /* 2116 * Extend page range to start and end at block boundaries. 2117 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 2118 */ 2119 origoffset = ap->a_offlo; 2120 origendoffset = ap->a_offhi; 2121 startoffset = origoffset & ~(fs->lfs_bmask); 2122 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 2123 << fs->lfs_bshift; 2124 2125 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2126 endoffset = max_endoffset; 2127 origendoffset = endoffset; 2128 } else { 2129 origendoffset = round_page(ap->a_offhi); 2130 endoffset = round_page(blkroundup(fs, origendoffset)); 2131 } 2132 2133 KASSERT(startoffset > 0 || endoffset >= startoffset); 2134 if (startoffset == endoffset) { 2135 /* Nothing to do, why were we called? */ 2136 mutex_exit(&vp->v_interlock); 2137 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 2138 PRId64 "\n", startoffset)); 2139 return 0; 2140 } 2141 2142 ap->a_offlo = startoffset; 2143 ap->a_offhi = endoffset; 2144 2145 /* 2146 * If not cleaning, just send the pages through genfs_putpages 2147 * to be returned to the pool. 2148 */ 2149 if (!(ap->a_flags & PGO_CLEANIT)) 2150 return genfs_putpages(v); 2151 2152 /* Set PGO_BUSYFAIL to avoid deadlocks */ 2153 ap->a_flags |= PGO_BUSYFAIL; 2154 2155 /* 2156 * Likewise, if we are asked to clean but the pages are not 2157 * dirty, we can just free them using genfs_putpages. 2158 */ 2159 #ifdef DEBUG 2160 debug_n_dirtyclean = 0; 2161 #endif 2162 do { 2163 int r; 2164 2165 /* Count the number of dirty pages */ 2166 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2167 ap->a_flags, 1, NULL); 2168 if (r < 0) { 2169 /* Pages are busy with another process */ 2170 mutex_exit(&vp->v_interlock); 2171 return EDEADLK; 2172 } 2173 if (r > 0) /* Some pages are dirty */ 2174 break; 2175 2176 /* 2177 * Sometimes pages are dirtied between the time that 2178 * we check and the time we try to clean them. 2179 * Instruct lfs_gop_write to return EDEADLK in this case 2180 * so we can write them properly. 2181 */ 2182 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2183 r = genfs_do_putpages(vp, startoffset, endoffset, 2184 ap->a_flags & ~PGO_SYNCIO, &busypg); 2185 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2186 if (r != EDEADLK) 2187 return r; 2188 2189 /* One of the pages was busy. Start over. */ 2190 mutex_enter(&vp->v_interlock); 2191 wait_for_page(vp, busypg, "dirtyclean"); 2192 #ifdef DEBUG 2193 ++debug_n_dirtyclean; 2194 #endif 2195 } while(1); 2196 2197 #ifdef DEBUG 2198 if (debug_n_dirtyclean > TOOMANY) 2199 printf("lfs_putpages: dirtyclean: looping, n = %d\n", 2200 debug_n_dirtyclean); 2201 #endif 2202 2203 /* 2204 * Dirty and asked to clean. 2205 * 2206 * Pagedaemon can't actually write LFS pages; wake up 2207 * the writer to take care of that. The writer will 2208 * notice the pager inode queue and act on that. 2209 * 2210 * XXX We must drop the vp->interlock before taking the lfs_lock or we 2211 * get a nasty deadlock with lfs_flush_pchain(). 2212 */ 2213 if (pagedaemon) { 2214 mutex_exit(&vp->v_interlock); 2215 mutex_enter(&lfs_lock); 2216 if (!(ip->i_flags & IN_PAGING)) { 2217 ip->i_flags |= IN_PAGING; 2218 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2219 } 2220 wakeup(&lfs_writer_daemon); 2221 mutex_exit(&lfs_lock); 2222 preempt(); 2223 return EWOULDBLOCK; 2224 } 2225 2226 /* 2227 * If this is a file created in a recent dirop, we can't flush its 2228 * inode until the dirop is complete. Drain dirops, then flush the 2229 * filesystem (taking care of any other pending dirops while we're 2230 * at it). 2231 */ 2232 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2233 (vp->v_uflag & VU_DIROP)) { 2234 int locked; 2235 2236 DLOG((DLOG_PAGE, "lfs_putpages: flushing VU_DIROP\n")); 2237 /* XXX VOP_ISLOCKED() may not be used for lock decisions. */ 2238 locked = (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 2239 mutex_exit(&vp->v_interlock); 2240 lfs_writer_enter(fs, "ppdirop"); 2241 if (locked) 2242 VOP_UNLOCK(vp); /* XXX why? */ 2243 2244 mutex_enter(&lfs_lock); 2245 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2246 mutex_exit(&lfs_lock); 2247 2248 if (locked) 2249 VOP_LOCK(vp, LK_EXCLUSIVE); 2250 mutex_enter(&vp->v_interlock); 2251 lfs_writer_leave(fs); 2252 2253 /* XXX the flush should have taken care of this one too! */ 2254 } 2255 2256 /* 2257 * This is it. We are going to write some pages. From here on 2258 * down it's all just mechanics. 2259 * 2260 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2261 */ 2262 ap->a_flags &= ~PGO_SYNCIO; 2263 2264 /* 2265 * If we've already got the seglock, flush the node and return. 2266 * The FIP has already been set up for us by lfs_writefile, 2267 * and FIP cleanup and lfs_updatemeta will also be done there, 2268 * unless genfs_putpages returns EDEADLK; then we must flush 2269 * what we have, and correct FIP and segment header accounting. 2270 */ 2271 get_seglock: 2272 /* 2273 * If we are not called with the segment locked, lock it. 2274 * Account for a new FIP in the segment header, and set sp->vp. 2275 * (This should duplicate the setup at the top of lfs_writefile().) 2276 */ 2277 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2278 if (!seglocked) { 2279 mutex_exit(&vp->v_interlock); 2280 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2281 if (error != 0) 2282 return error; 2283 mutex_enter(&vp->v_interlock); 2284 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2285 } 2286 sp = fs->lfs_sp; 2287 KASSERT(sp->vp == NULL); 2288 sp->vp = vp; 2289 2290 /* 2291 * Ensure that the partial segment is marked SS_DIROP if this 2292 * vnode is a DIROP. 2293 */ 2294 if (!seglocked && vp->v_uflag & VU_DIROP) 2295 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2296 2297 /* 2298 * Loop over genfs_putpages until all pages are gathered. 2299 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2300 * Whenever we lose the interlock we have to rerun check_dirty, as 2301 * well, since more pages might have been dirtied in our absence. 2302 */ 2303 #ifdef DEBUG 2304 debug_n_again = 0; 2305 #endif 2306 do { 2307 busypg = NULL; 2308 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2309 ap->a_flags, 0, &busypg) < 0) { 2310 mutex_exit(&vp->v_interlock); 2311 2312 mutex_enter(&vp->v_interlock); 2313 write_and_wait(fs, vp, busypg, seglocked, NULL); 2314 if (!seglocked) { 2315 mutex_exit(&vp->v_interlock); 2316 lfs_release_finfo(fs); 2317 lfs_segunlock(fs); 2318 mutex_enter(&vp->v_interlock); 2319 } 2320 sp->vp = NULL; 2321 goto get_seglock; 2322 } 2323 2324 busypg = NULL; 2325 error = genfs_do_putpages(vp, startoffset, endoffset, 2326 ap->a_flags, &busypg); 2327 2328 if (error == EDEADLK || error == EAGAIN) { 2329 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2330 " %d ino %d off %x (seg %d)\n", error, 2331 ip->i_number, fs->lfs_offset, 2332 dtosn(fs, fs->lfs_offset))); 2333 2334 mutex_enter(&vp->v_interlock); 2335 write_and_wait(fs, vp, busypg, seglocked, "again"); 2336 } 2337 #ifdef DEBUG 2338 ++debug_n_again; 2339 #endif 2340 } while (error == EDEADLK); 2341 #ifdef DEBUG 2342 if (debug_n_again > TOOMANY) 2343 printf("lfs_putpages: again: looping, n = %d\n", debug_n_again); 2344 #endif 2345 2346 KASSERT(sp != NULL && sp->vp == vp); 2347 if (!seglocked) { 2348 sp->vp = NULL; 2349 2350 /* Write indirect blocks as well */ 2351 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2352 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2353 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2354 2355 KASSERT(sp->vp == NULL); 2356 sp->vp = vp; 2357 } 2358 2359 /* 2360 * Blocks are now gathered into a segment waiting to be written. 2361 * All that's left to do is update metadata, and write them. 2362 */ 2363 lfs_updatemeta(sp); 2364 KASSERT(sp->vp == vp); 2365 sp->vp = NULL; 2366 2367 /* 2368 * If we were called from lfs_writefile, we don't need to clean up 2369 * the FIP or unlock the segment lock. We're done. 2370 */ 2371 if (seglocked) 2372 return error; 2373 2374 /* Clean up FIP and send it to disk. */ 2375 lfs_release_finfo(fs); 2376 lfs_writeseg(fs, fs->lfs_sp); 2377 2378 /* 2379 * Remove us from paging queue if we wrote all our pages. 2380 */ 2381 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2382 mutex_enter(&lfs_lock); 2383 if (ip->i_flags & IN_PAGING) { 2384 ip->i_flags &= ~IN_PAGING; 2385 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2386 } 2387 mutex_exit(&lfs_lock); 2388 } 2389 2390 /* 2391 * XXX - with the malloc/copy writeseg, the pages are freed by now 2392 * even if we don't wait (e.g. if we hold a nested lock). This 2393 * will not be true if we stop using malloc/copy. 2394 */ 2395 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2396 lfs_segunlock(fs); 2397 2398 /* 2399 * Wait for v_numoutput to drop to zero. The seglock should 2400 * take care of this, but there is a slight possibility that 2401 * aiodoned might not have got around to our buffers yet. 2402 */ 2403 if (sync) { 2404 mutex_enter(&vp->v_interlock); 2405 while (vp->v_numoutput > 0) { 2406 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2407 " num %d\n", ip->i_number, vp->v_numoutput)); 2408 cv_wait(&vp->v_cv, &vp->v_interlock); 2409 } 2410 mutex_exit(&vp->v_interlock); 2411 } 2412 return error; 2413 } 2414 2415 /* 2416 * Return the last logical file offset that should be written for this file 2417 * if we're doing a write that ends at "size". If writing, we need to know 2418 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2419 * to know about entire blocks. 2420 */ 2421 void 2422 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2423 { 2424 struct inode *ip = VTOI(vp); 2425 struct lfs *fs = ip->i_lfs; 2426 daddr_t olbn, nlbn; 2427 2428 olbn = lblkno(fs, ip->i_size); 2429 nlbn = lblkno(fs, size); 2430 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2431 *eobp = fragroundup(fs, size); 2432 } else { 2433 *eobp = blkroundup(fs, size); 2434 } 2435 } 2436 2437 #ifdef DEBUG 2438 void lfs_dump_vop(void *); 2439 2440 void 2441 lfs_dump_vop(void *v) 2442 { 2443 struct vop_putpages_args /* { 2444 struct vnode *a_vp; 2445 voff_t a_offlo; 2446 voff_t a_offhi; 2447 int a_flags; 2448 } */ *ap = v; 2449 2450 #ifdef DDB 2451 vfs_vnode_print(ap->a_vp, 0, printf); 2452 #endif 2453 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2454 } 2455 #endif 2456 2457 int 2458 lfs_mmap(void *v) 2459 { 2460 struct vop_mmap_args /* { 2461 const struct vnodeop_desc *a_desc; 2462 struct vnode *a_vp; 2463 vm_prot_t a_prot; 2464 kauth_cred_t a_cred; 2465 } */ *ap = v; 2466 2467 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2468 return EOPNOTSUPP; 2469 return ufs_mmap(v); 2470 } 2471
Indexes created Tue Sep 30 11:09:46 GMT 2025