lfs_vnops.c revision 1.237
1 /* $NetBSD: lfs_vnops.c,v 1.237 2011/07/12 16:59:48 dholland 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 1.237 2011/07/12 16:59:48 dholland 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 == NULL) { 477 return lfs_set_dirop(dvp, NULL); 478 } 479 error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, NULL, vpp); 480 if (error) { 481 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 482 dvp, error)); 483 return error; 484 } 485 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 486 ungetnewvnode(*vpp); 487 *vpp = NULL; 488 return error; 489 } 490 return 0; 491 } 492 493 #define SET_ENDOP_BASE(fs, dvp, str) \ 494 do { \ 495 mutex_enter(&lfs_lock); \ 496 --(fs)->lfs_dirops; \ 497 if (!(fs)->lfs_dirops) { \ 498 if ((fs)->lfs_nadirop) { \ 499 panic("SET_ENDOP: %s: no dirops but " \ 500 " nadirop=%d", (str), \ 501 (fs)->lfs_nadirop); \ 502 } \ 503 wakeup(&(fs)->lfs_writer); \ 504 mutex_exit(&lfs_lock); \ 505 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 506 } else \ 507 mutex_exit(&lfs_lock); \ 508 } while(0) 509 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 510 do { \ 511 UNMARK_VNODE(dvp); \ 512 if (nvpp && *nvpp) \ 513 UNMARK_VNODE(*nvpp); \ 514 /* Check for error return to stem vnode leakage */ \ 515 if (nvpp && *nvpp && !((*nvpp)->v_uflag & VU_DIROP)) \ 516 ungetnewvnode(*(nvpp)); \ 517 SET_ENDOP_BASE((fs), (dvp), (str)); \ 518 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 519 vrele(dvp); \ 520 } while(0) 521 #define SET_ENDOP_CREATE_AP(ap, str) \ 522 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 523 (ap)->a_vpp, (str)) 524 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 525 do { \ 526 UNMARK_VNODE(dvp); \ 527 if (ovp) \ 528 UNMARK_VNODE(ovp); \ 529 SET_ENDOP_BASE((fs), (dvp), (str)); \ 530 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 531 vrele(dvp); \ 532 if (ovp) \ 533 vrele(ovp); \ 534 } while(0) 535 536 void 537 lfs_mark_vnode(struct vnode *vp) 538 { 539 struct inode *ip = VTOI(vp); 540 struct lfs *fs = ip->i_lfs; 541 542 mutex_enter(&lfs_lock); 543 if (!(ip->i_flag & IN_ADIROP)) { 544 if (!(vp->v_uflag & VU_DIROP)) { 545 mutex_enter(vp->v_interlock); 546 (void)lfs_vref(vp); 547 ++lfs_dirvcount; 548 ++fs->lfs_dirvcount; 549 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 550 vp->v_uflag |= VU_DIROP; 551 } 552 ++fs->lfs_nadirop; 553 ip->i_flag |= IN_ADIROP; 554 } else 555 KASSERT(vp->v_uflag & VU_DIROP); 556 mutex_exit(&lfs_lock); 557 } 558 559 void 560 lfs_unmark_vnode(struct vnode *vp) 561 { 562 struct inode *ip = VTOI(vp); 563 564 if (ip && (ip->i_flag & IN_ADIROP)) { 565 KASSERT(vp->v_uflag & VU_DIROP); 566 mutex_enter(&lfs_lock); 567 --ip->i_lfs->lfs_nadirop; 568 mutex_exit(&lfs_lock); 569 ip->i_flag &= ~IN_ADIROP; 570 } 571 } 572 573 int 574 lfs_symlink(void *v) 575 { 576 struct vop_symlink_args /* { 577 struct vnode *a_dvp; 578 struct vnode **a_vpp; 579 struct componentname *a_cnp; 580 struct vattr *a_vap; 581 char *a_target; 582 } */ *ap = v; 583 int error; 584 585 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 586 vput(ap->a_dvp); 587 return error; 588 } 589 error = ufs_symlink(ap); 590 SET_ENDOP_CREATE_AP(ap, "symlink"); 591 return (error); 592 } 593 594 int 595 lfs_mknod(void *v) 596 { 597 struct vop_mknod_args /* { 598 struct vnode *a_dvp; 599 struct vnode **a_vpp; 600 struct componentname *a_cnp; 601 struct vattr *a_vap; 602 } */ *ap = v; 603 struct vattr *vap = ap->a_vap; 604 struct vnode **vpp = ap->a_vpp; 605 struct inode *ip; 606 int error; 607 struct mount *mp; 608 ino_t ino; 609 struct ufs_lookup_results *ulr; 610 611 /* XXX should handle this material another way */ 612 ulr = &VTOI(ap->a_dvp)->i_crap; 613 UFS_CHECK_CRAPCOUNTER(VTOI(ap->a_dvp)); 614 615 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 616 vput(ap->a_dvp); 617 return error; 618 } 619 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 620 ap->a_dvp, ulr, vpp, ap->a_cnp); 621 622 /* Either way we're done with the dirop at this point */ 623 SET_ENDOP_CREATE_AP(ap, "mknod"); 624 625 if (error) 626 return (error); 627 628 ip = VTOI(*vpp); 629 mp = (*vpp)->v_mount; 630 ino = ip->i_number; 631 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 632 if (vap->va_rdev != VNOVAL) { 633 /* 634 * Want to be able to use this to make badblock 635 * inodes, so don't truncate the dev number. 636 */ 637 #if 0 638 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 639 UFS_MPNEEDSWAP((*vpp)->v_mount)); 640 #else 641 ip->i_ffs1_rdev = vap->va_rdev; 642 #endif 643 } 644 645 /* 646 * Call fsync to write the vnode so that we don't have to deal with 647 * flushing it when it's marked VU_DIROP|VI_XLOCK. 648 * 649 * XXX KS - If we can't flush we also can't call vgone(), so must 650 * return. But, that leaves this vnode in limbo, also not good. 651 * Can this ever happen (barring hardware failure)? 652 */ 653 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0)) != 0) { 654 panic("lfs_mknod: couldn't fsync (ino %llu)", 655 (unsigned long long)ino); 656 /* return (error); */ 657 } 658 /* 659 * Remove vnode so that it will be reloaded by VFS_VGET and 660 * checked to see if it is an alias of an existing entry in 661 * the inode cache. 662 */ 663 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 664 665 VOP_UNLOCK(*vpp); 666 (*vpp)->v_type = VNON; 667 vgone(*vpp); 668 error = VFS_VGET(mp, ino, vpp); 669 670 if (error != 0) { 671 *vpp = NULL; 672 return (error); 673 } 674 return (0); 675 } 676 677 int 678 lfs_create(void *v) 679 { 680 struct vop_create_args /* { 681 struct vnode *a_dvp; 682 struct vnode **a_vpp; 683 struct componentname *a_cnp; 684 struct vattr *a_vap; 685 } */ *ap = v; 686 int error; 687 688 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 689 vput(ap->a_dvp); 690 return error; 691 } 692 error = ufs_create(ap); 693 SET_ENDOP_CREATE_AP(ap, "create"); 694 return (error); 695 } 696 697 int 698 lfs_mkdir(void *v) 699 { 700 struct vop_mkdir_args /* { 701 struct vnode *a_dvp; 702 struct vnode **a_vpp; 703 struct componentname *a_cnp; 704 struct vattr *a_vap; 705 } */ *ap = v; 706 int error; 707 708 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 709 vput(ap->a_dvp); 710 return error; 711 } 712 error = ufs_mkdir(ap); 713 SET_ENDOP_CREATE_AP(ap, "mkdir"); 714 return (error); 715 } 716 717 int 718 lfs_remove(void *v) 719 { 720 struct vop_remove_args /* { 721 struct vnode *a_dvp; 722 struct vnode *a_vp; 723 struct componentname *a_cnp; 724 } */ *ap = v; 725 struct vnode *dvp, *vp; 726 struct inode *ip; 727 int error; 728 729 dvp = ap->a_dvp; 730 vp = ap->a_vp; 731 ip = VTOI(vp); 732 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 733 if (dvp == vp) 734 vrele(vp); 735 else 736 vput(vp); 737 vput(dvp); 738 return error; 739 } 740 error = ufs_remove(ap); 741 if (ip->i_nlink == 0) 742 lfs_orphan(ip->i_lfs, ip->i_number); 743 SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove"); 744 return (error); 745 } 746 747 int 748 lfs_rmdir(void *v) 749 { 750 struct vop_rmdir_args /* { 751 struct vnodeop_desc *a_desc; 752 struct vnode *a_dvp; 753 struct vnode *a_vp; 754 struct componentname *a_cnp; 755 } */ *ap = v; 756 struct vnode *vp; 757 struct inode *ip; 758 int error; 759 760 vp = ap->a_vp; 761 ip = VTOI(vp); 762 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 763 if (ap->a_dvp == vp) 764 vrele(ap->a_dvp); 765 else 766 vput(ap->a_dvp); 767 vput(vp); 768 return error; 769 } 770 error = ufs_rmdir(ap); 771 if (ip->i_nlink == 0) 772 lfs_orphan(ip->i_lfs, ip->i_number); 773 SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 774 return (error); 775 } 776 777 int 778 lfs_link(void *v) 779 { 780 struct vop_link_args /* { 781 struct vnode *a_dvp; 782 struct vnode *a_vp; 783 struct componentname *a_cnp; 784 } */ *ap = v; 785 int error; 786 struct vnode **vpp = NULL; 787 788 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 789 vput(ap->a_dvp); 790 return error; 791 } 792 error = ufs_link(ap); 793 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 794 return (error); 795 } 796 797 int 798 lfs_rename(void *v) 799 { 800 struct vop_rename_args /* { 801 struct vnode *a_fdvp; 802 struct vnode *a_fvp; 803 struct componentname *a_fcnp; 804 struct vnode *a_tdvp; 805 struct vnode *a_tvp; 806 struct componentname *a_tcnp; 807 } */ *ap = v; 808 struct vnode *tvp, *fvp, *tdvp, *fdvp; 809 struct componentname *tcnp, *fcnp; 810 int error; 811 struct lfs *fs; 812 813 fs = VTOI(ap->a_fdvp)->i_lfs; 814 tvp = ap->a_tvp; 815 tdvp = ap->a_tdvp; 816 tcnp = ap->a_tcnp; 817 fvp = ap->a_fvp; 818 fdvp = ap->a_fdvp; 819 fcnp = ap->a_fcnp; 820 821 /* 822 * Check for cross-device rename. 823 * If it is, we don't want to set dirops, just error out. 824 * (In particular note that MARK_VNODE(tdvp) will DTWT on 825 * a cross-device rename.) 826 * 827 * Copied from ufs_rename. 828 */ 829 if ((fvp->v_mount != tdvp->v_mount) || 830 (tvp && (fvp->v_mount != tvp->v_mount))) { 831 error = EXDEV; 832 goto errout; 833 } 834 835 /* 836 * Check to make sure we're not renaming a vnode onto itself 837 * (deleting a hard link by renaming one name onto another); 838 * if we are we can't recursively call VOP_REMOVE since that 839 * would leave us with an unaccounted-for number of live dirops. 840 * 841 * Inline the relevant section of ufs_rename here, *before* 842 * calling SET_DIROP_REMOVE. 843 */ 844 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 845 (VTOI(tdvp)->i_flags & APPEND))) { 846 error = EPERM; 847 goto errout; 848 } 849 if (fvp == tvp) { 850 if (fvp->v_type == VDIR) { 851 error = EINVAL; 852 goto errout; 853 } 854 855 /* Release destination completely. */ 856 VOP_ABORTOP(tdvp, tcnp); 857 vput(tdvp); 858 vput(tvp); 859 860 /* Delete source. */ 861 vrele(fvp); 862 fcnp->cn_flags &= ~(MODMASK); 863 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 864 fcnp->cn_nameiop = DELETE; 865 vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY); 866 if ((error = relookup(fdvp, &fvp, fcnp, 0))) { 867 vput(fdvp); 868 return (error); 869 } 870 return (VOP_REMOVE(fdvp, fvp, fcnp)); 871 } 872 873 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 874 goto errout; 875 MARK_VNODE(fdvp); 876 MARK_VNODE(fvp); 877 878 error = ufs_rename(ap); 879 UNMARK_VNODE(fdvp); 880 UNMARK_VNODE(fvp); 881 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 882 return (error); 883 884 errout: 885 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 886 if (tdvp == tvp) 887 vrele(tdvp); 888 else 889 vput(tdvp); 890 if (tvp) 891 vput(tvp); 892 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 893 vrele(fdvp); 894 vrele(fvp); 895 return (error); 896 } 897 898 /* XXX hack to avoid calling ITIMES in getattr */ 899 int 900 lfs_getattr(void *v) 901 { 902 struct vop_getattr_args /* { 903 struct vnode *a_vp; 904 struct vattr *a_vap; 905 kauth_cred_t a_cred; 906 } */ *ap = v; 907 struct vnode *vp = ap->a_vp; 908 struct inode *ip = VTOI(vp); 909 struct vattr *vap = ap->a_vap; 910 struct lfs *fs = ip->i_lfs; 911 /* 912 * Copy from inode table 913 */ 914 vap->va_fsid = ip->i_dev; 915 vap->va_fileid = ip->i_number; 916 vap->va_mode = ip->i_mode & ~IFMT; 917 vap->va_nlink = ip->i_nlink; 918 vap->va_uid = ip->i_uid; 919 vap->va_gid = ip->i_gid; 920 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 921 vap->va_size = vp->v_size; 922 vap->va_atime.tv_sec = ip->i_ffs1_atime; 923 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 924 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 925 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 926 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 927 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 928 vap->va_flags = ip->i_flags; 929 vap->va_gen = ip->i_gen; 930 /* this doesn't belong here */ 931 if (vp->v_type == VBLK) 932 vap->va_blocksize = BLKDEV_IOSIZE; 933 else if (vp->v_type == VCHR) 934 vap->va_blocksize = MAXBSIZE; 935 else 936 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 937 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 938 vap->va_type = vp->v_type; 939 vap->va_filerev = ip->i_modrev; 940 return (0); 941 } 942 943 /* 944 * Check to make sure the inode blocks won't choke the buffer 945 * cache, then call ufs_setattr as usual. 946 */ 947 int 948 lfs_setattr(void *v) 949 { 950 struct vop_setattr_args /* { 951 struct vnode *a_vp; 952 struct vattr *a_vap; 953 kauth_cred_t a_cred; 954 } */ *ap = v; 955 struct vnode *vp = ap->a_vp; 956 957 lfs_check(vp, LFS_UNUSED_LBN, 0); 958 return ufs_setattr(v); 959 } 960 961 /* 962 * Release the block we hold on lfs_newseg wrapping. Called on file close, 963 * or explicitly from LFCNWRAPGO. Called with the interlock held. 964 */ 965 static int 966 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor) 967 { 968 if (fs->lfs_stoplwp != curlwp) 969 return EBUSY; 970 971 fs->lfs_stoplwp = NULL; 972 cv_signal(&fs->lfs_stopcv); 973 974 KASSERT(fs->lfs_nowrap > 0); 975 if (fs->lfs_nowrap <= 0) { 976 return 0; 977 } 978 979 if (--fs->lfs_nowrap == 0) { 980 log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt); 981 wakeup(&fs->lfs_wrappass); 982 lfs_wakeup_cleaner(fs); 983 } 984 if (waitfor) { 985 mtsleep(&fs->lfs_nextseg, PCATCH | PUSER, "segment", 986 0, &lfs_lock); 987 } 988 989 return 0; 990 } 991 992 /* 993 * Close called 994 */ 995 /* ARGSUSED */ 996 int 997 lfs_close(void *v) 998 { 999 struct vop_close_args /* { 1000 struct vnode *a_vp; 1001 int a_fflag; 1002 kauth_cred_t a_cred; 1003 } */ *ap = v; 1004 struct vnode *vp = ap->a_vp; 1005 struct inode *ip = VTOI(vp); 1006 struct lfs *fs = ip->i_lfs; 1007 1008 if ((ip->i_number == ROOTINO || ip->i_number == LFS_IFILE_INUM) && 1009 fs->lfs_stoplwp == curlwp) { 1010 mutex_enter(&lfs_lock); 1011 log(LOG_NOTICE, "lfs_close: releasing log wrap control\n"); 1012 lfs_wrapgo(fs, ip, 0); 1013 mutex_exit(&lfs_lock); 1014 } 1015 1016 if (vp == ip->i_lfs->lfs_ivnode && 1017 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 1018 return 0; 1019 1020 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 1021 LFS_ITIMES(ip, NULL, NULL, NULL); 1022 } 1023 return (0); 1024 } 1025 1026 /* 1027 * Close wrapper for special devices. 1028 * 1029 * Update the times on the inode then do device close. 1030 */ 1031 int 1032 lfsspec_close(void *v) 1033 { 1034 struct vop_close_args /* { 1035 struct vnode *a_vp; 1036 int a_fflag; 1037 kauth_cred_t a_cred; 1038 } */ *ap = v; 1039 struct vnode *vp; 1040 struct inode *ip; 1041 1042 vp = ap->a_vp; 1043 ip = VTOI(vp); 1044 if (vp->v_usecount > 1) { 1045 LFS_ITIMES(ip, NULL, NULL, NULL); 1046 } 1047 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1048 } 1049 1050 /* 1051 * Close wrapper for fifo's. 1052 * 1053 * Update the times on the inode then do device close. 1054 */ 1055 int 1056 lfsfifo_close(void *v) 1057 { 1058 struct vop_close_args /* { 1059 struct vnode *a_vp; 1060 int a_fflag; 1061 kauth_cred_ a_cred; 1062 } */ *ap = v; 1063 struct vnode *vp; 1064 struct inode *ip; 1065 1066 vp = ap->a_vp; 1067 ip = VTOI(vp); 1068 if (ap->a_vp->v_usecount > 1) { 1069 LFS_ITIMES(ip, NULL, NULL, NULL); 1070 } 1071 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1072 } 1073 1074 /* 1075 * Reclaim an inode so that it can be used for other purposes. 1076 */ 1077 1078 int 1079 lfs_reclaim(void *v) 1080 { 1081 struct vop_reclaim_args /* { 1082 struct vnode *a_vp; 1083 } */ *ap = v; 1084 struct vnode *vp = ap->a_vp; 1085 struct inode *ip = VTOI(vp); 1086 struct lfs *fs = ip->i_lfs; 1087 int error; 1088 1089 /* 1090 * The inode must be freed and updated before being removed 1091 * from its hash chain. Other threads trying to gain a hold 1092 * on the inode will be stalled because it is locked (VI_XLOCK). 1093 */ 1094 if (ip->i_nlink <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) 1095 lfs_vfree(vp, ip->i_number, ip->i_omode); 1096 1097 mutex_enter(&lfs_lock); 1098 LFS_CLR_UINO(ip, IN_ALLMOD); 1099 mutex_exit(&lfs_lock); 1100 if ((error = ufs_reclaim(vp))) 1101 return (error); 1102 1103 /* 1104 * Take us off the paging and/or dirop queues if we were on them. 1105 * We shouldn't be on them. 1106 */ 1107 mutex_enter(&lfs_lock); 1108 if (ip->i_flags & IN_PAGING) { 1109 log(LOG_WARNING, "%s: reclaimed vnode is IN_PAGING\n", 1110 fs->lfs_fsmnt); 1111 ip->i_flags &= ~IN_PAGING; 1112 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1113 } 1114 if (vp->v_uflag & VU_DIROP) { 1115 panic("reclaimed vnode is VU_DIROP"); 1116 vp->v_uflag &= ~VU_DIROP; 1117 TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain); 1118 } 1119 mutex_exit(&lfs_lock); 1120 1121 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1122 lfs_deregister_all(vp); 1123 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1124 ip->inode_ext.lfs = NULL; 1125 genfs_node_destroy(vp); 1126 pool_put(&lfs_inode_pool, vp->v_data); 1127 vp->v_data = NULL; 1128 return (0); 1129 } 1130 1131 /* 1132 * Read a block from a storage device. 1133 * In order to avoid reading blocks that are in the process of being 1134 * written by the cleaner---and hence are not mutexed by the normal 1135 * buffer cache / page cache mechanisms---check for collisions before 1136 * reading. 1137 * 1138 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1139 * the active cleaner test. 1140 * 1141 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1142 */ 1143 int 1144 lfs_strategy(void *v) 1145 { 1146 struct vop_strategy_args /* { 1147 struct vnode *a_vp; 1148 struct buf *a_bp; 1149 } */ *ap = v; 1150 struct buf *bp; 1151 struct lfs *fs; 1152 struct vnode *vp; 1153 struct inode *ip; 1154 daddr_t tbn; 1155 int i, sn, error, slept; 1156 1157 bp = ap->a_bp; 1158 vp = ap->a_vp; 1159 ip = VTOI(vp); 1160 fs = ip->i_lfs; 1161 1162 /* lfs uses its strategy routine only for read */ 1163 KASSERT(bp->b_flags & B_READ); 1164 1165 if (vp->v_type == VBLK || vp->v_type == VCHR) 1166 panic("lfs_strategy: spec"); 1167 KASSERT(bp->b_bcount != 0); 1168 if (bp->b_blkno == bp->b_lblkno) { 1169 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1170 NULL); 1171 if (error) { 1172 bp->b_error = error; 1173 bp->b_resid = bp->b_bcount; 1174 biodone(bp); 1175 return (error); 1176 } 1177 if ((long)bp->b_blkno == -1) /* no valid data */ 1178 clrbuf(bp); 1179 } 1180 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1181 bp->b_resid = bp->b_bcount; 1182 biodone(bp); 1183 return (0); 1184 } 1185 1186 slept = 1; 1187 mutex_enter(&lfs_lock); 1188 while (slept && fs->lfs_seglock) { 1189 mutex_exit(&lfs_lock); 1190 /* 1191 * Look through list of intervals. 1192 * There will only be intervals to look through 1193 * if the cleaner holds the seglock. 1194 * Since the cleaner is synchronous, we can trust 1195 * the list of intervals to be current. 1196 */ 1197 tbn = dbtofsb(fs, bp->b_blkno); 1198 sn = dtosn(fs, tbn); 1199 slept = 0; 1200 for (i = 0; i < fs->lfs_cleanind; i++) { 1201 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1202 tbn >= fs->lfs_cleanint[i]) { 1203 DLOG((DLOG_CLEAN, 1204 "lfs_strategy: ino %d lbn %" PRId64 1205 " ind %d sn %d fsb %" PRIx32 1206 " given sn %d fsb %" PRIx64 "\n", 1207 ip->i_number, bp->b_lblkno, i, 1208 dtosn(fs, fs->lfs_cleanint[i]), 1209 fs->lfs_cleanint[i], sn, tbn)); 1210 DLOG((DLOG_CLEAN, 1211 "lfs_strategy: sleeping on ino %d lbn %" 1212 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1213 mutex_enter(&lfs_lock); 1214 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1215 /* Cleaner can't wait for itself */ 1216 mtsleep(&fs->lfs_iocount, 1217 (PRIBIO + 1) | PNORELOCK, 1218 "clean2", 0, 1219 &lfs_lock); 1220 slept = 1; 1221 break; 1222 } else if (fs->lfs_seglock) { 1223 mtsleep(&fs->lfs_seglock, 1224 (PRIBIO + 1) | PNORELOCK, 1225 "clean1", 0, 1226 &lfs_lock); 1227 slept = 1; 1228 break; 1229 } 1230 mutex_exit(&lfs_lock); 1231 } 1232 } 1233 mutex_enter(&lfs_lock); 1234 } 1235 mutex_exit(&lfs_lock); 1236 1237 vp = ip->i_devvp; 1238 VOP_STRATEGY(vp, bp); 1239 return (0); 1240 } 1241 1242 void 1243 lfs_flush_dirops(struct lfs *fs) 1244 { 1245 struct inode *ip, *nip; 1246 struct vnode *vp; 1247 extern int lfs_dostats; 1248 struct segment *sp; 1249 1250 ASSERT_MAYBE_SEGLOCK(fs); 1251 KASSERT(fs->lfs_nadirop == 0); 1252 1253 if (fs->lfs_ronly) 1254 return; 1255 1256 mutex_enter(&lfs_lock); 1257 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1258 mutex_exit(&lfs_lock); 1259 return; 1260 } else 1261 mutex_exit(&lfs_lock); 1262 1263 if (lfs_dostats) 1264 ++lfs_stats.flush_invoked; 1265 1266 /* 1267 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1268 * Technically this is a checkpoint (the on-disk state is valid) 1269 * even though we are leaving out all the file data. 1270 */ 1271 lfs_imtime(fs); 1272 lfs_seglock(fs, SEGM_CKP); 1273 sp = fs->lfs_sp; 1274 1275 /* 1276 * lfs_writevnodes, optimized to get dirops out of the way. 1277 * Only write dirops, and don't flush files' pages, only 1278 * blocks from the directories. 1279 * 1280 * We don't need to vref these files because they are 1281 * dirops and so hold an extra reference until the 1282 * segunlock clears them of that status. 1283 * 1284 * We don't need to check for IN_ADIROP because we know that 1285 * no dirops are active. 1286 * 1287 */ 1288 mutex_enter(&lfs_lock); 1289 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1290 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1291 mutex_exit(&lfs_lock); 1292 vp = ITOV(ip); 1293 1294 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1295 1296 /* 1297 * All writes to directories come from dirops; all 1298 * writes to files' direct blocks go through the page 1299 * cache, which we're not touching. Reads to files 1300 * and/or directories will not be affected by writing 1301 * directory blocks inodes and file inodes. So we don't 1302 * really need to lock. If we don't lock, though, 1303 * make sure that we don't clear IN_MODIFIED 1304 * unnecessarily. 1305 */ 1306 if (vp->v_iflag & VI_XLOCK) { 1307 mutex_enter(&lfs_lock); 1308 continue; 1309 } 1310 /* XXX see below 1311 * waslocked = VOP_ISLOCKED(vp); 1312 */ 1313 if (vp->v_type != VREG && 1314 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1315 lfs_writefile(fs, sp, vp); 1316 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1317 !(ip->i_flag & IN_ALLMOD)) { 1318 mutex_enter(&lfs_lock); 1319 LFS_SET_UINO(ip, IN_MODIFIED); 1320 mutex_exit(&lfs_lock); 1321 } 1322 } 1323 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1324 (void) lfs_writeinode(fs, sp, ip); 1325 mutex_enter(&lfs_lock); 1326 /* 1327 * XXX 1328 * LK_EXCLOTHER is dead -- what is intended here? 1329 * if (waslocked == LK_EXCLOTHER) 1330 * LFS_SET_UINO(ip, IN_MODIFIED); 1331 */ 1332 } 1333 mutex_exit(&lfs_lock); 1334 /* We've written all the dirops there are */ 1335 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1336 lfs_finalize_fs_seguse(fs); 1337 (void) lfs_writeseg(fs, sp); 1338 lfs_segunlock(fs); 1339 } 1340 1341 /* 1342 * Flush all vnodes for which the pagedaemon has requested pageouts. 1343 * Skip over any files that are marked VU_DIROP (since lfs_flush_dirop() 1344 * has just run, this would be an error). If we have to skip a vnode 1345 * for any reason, just skip it; if we have to wait for the cleaner, 1346 * abort. The writer daemon will call us again later. 1347 */ 1348 void 1349 lfs_flush_pchain(struct lfs *fs) 1350 { 1351 struct inode *ip, *nip; 1352 struct vnode *vp; 1353 extern int lfs_dostats; 1354 struct segment *sp; 1355 int error; 1356 1357 ASSERT_NO_SEGLOCK(fs); 1358 1359 if (fs->lfs_ronly) 1360 return; 1361 1362 mutex_enter(&lfs_lock); 1363 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1364 mutex_exit(&lfs_lock); 1365 return; 1366 } else 1367 mutex_exit(&lfs_lock); 1368 1369 /* Get dirops out of the way */ 1370 lfs_flush_dirops(fs); 1371 1372 if (lfs_dostats) 1373 ++lfs_stats.flush_invoked; 1374 1375 /* 1376 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1377 */ 1378 lfs_imtime(fs); 1379 lfs_seglock(fs, 0); 1380 sp = fs->lfs_sp; 1381 1382 /* 1383 * lfs_writevnodes, optimized to clear pageout requests. 1384 * Only write non-dirop files that are in the pageout queue. 1385 * We're very conservative about what we write; we want to be 1386 * fast and async. 1387 */ 1388 mutex_enter(&lfs_lock); 1389 top: 1390 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1391 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1392 vp = ITOV(ip); 1393 1394 if (!(ip->i_flags & IN_PAGING)) 1395 goto top; 1396 1397 mutex_enter(vp->v_interlock); 1398 if ((vp->v_iflag & VI_XLOCK) || (vp->v_uflag & VU_DIROP) != 0) { 1399 mutex_exit(vp->v_interlock); 1400 continue; 1401 } 1402 if (vp->v_type != VREG) { 1403 mutex_exit(vp->v_interlock); 1404 continue; 1405 } 1406 if (lfs_vref(vp)) 1407 continue; 1408 mutex_exit(&lfs_lock); 1409 1410 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_RETRY) != 0) { 1411 lfs_vunref(vp); 1412 mutex_enter(&lfs_lock); 1413 continue; 1414 } 1415 1416 error = lfs_writefile(fs, sp, vp); 1417 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1418 !(ip->i_flag & IN_ALLMOD)) { 1419 mutex_enter(&lfs_lock); 1420 LFS_SET_UINO(ip, IN_MODIFIED); 1421 mutex_exit(&lfs_lock); 1422 } 1423 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1424 (void) lfs_writeinode(fs, sp, ip); 1425 1426 VOP_UNLOCK(vp); 1427 lfs_vunref(vp); 1428 1429 if (error == EAGAIN) { 1430 lfs_writeseg(fs, sp); 1431 mutex_enter(&lfs_lock); 1432 break; 1433 } 1434 mutex_enter(&lfs_lock); 1435 } 1436 mutex_exit(&lfs_lock); 1437 (void) lfs_writeseg(fs, sp); 1438 lfs_segunlock(fs); 1439 } 1440 1441 /* 1442 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1443 */ 1444 int 1445 lfs_fcntl(void *v) 1446 { 1447 struct vop_fcntl_args /* { 1448 struct vnode *a_vp; 1449 u_int a_command; 1450 void * a_data; 1451 int a_fflag; 1452 kauth_cred_t a_cred; 1453 } */ *ap = v; 1454 struct timeval tv; 1455 struct timeval *tvp; 1456 BLOCK_INFO *blkiov; 1457 CLEANERINFO *cip; 1458 SEGUSE *sup; 1459 int blkcnt, error, oclean; 1460 size_t fh_size; 1461 struct lfs_fcntl_markv blkvp; 1462 struct lwp *l; 1463 fsid_t *fsidp; 1464 struct lfs *fs; 1465 struct buf *bp; 1466 fhandle_t *fhp; 1467 daddr_t off; 1468 1469 /* Only respect LFS fcntls on fs root or Ifile */ 1470 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1471 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1472 return ufs_fcntl(v); 1473 } 1474 1475 /* Avoid locking a draining lock */ 1476 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1477 return ESHUTDOWN; 1478 } 1479 1480 /* LFS control and monitoring fcntls are available only to root */ 1481 l = curlwp; 1482 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1483 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1484 NULL)) != 0) 1485 return (error); 1486 1487 fs = VTOI(ap->a_vp)->i_lfs; 1488 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1489 1490 error = 0; 1491 switch ((int)ap->a_command) { 1492 case LFCNSEGWAITALL_COMPAT_50: 1493 case LFCNSEGWAITALL_COMPAT: 1494 fsidp = NULL; 1495 /* FALLSTHROUGH */ 1496 case LFCNSEGWAIT_COMPAT_50: 1497 case LFCNSEGWAIT_COMPAT: 1498 { 1499 struct timeval50 *tvp50 1500 = (struct timeval50 *)ap->a_data; 1501 timeval50_to_timeval(tvp50, &tv); 1502 tvp = &tv; 1503 } 1504 goto segwait_common; 1505 case LFCNSEGWAITALL: 1506 fsidp = NULL; 1507 /* FALLSTHROUGH */ 1508 case LFCNSEGWAIT: 1509 tvp = (struct timeval *)ap->a_data; 1510 segwait_common: 1511 mutex_enter(&lfs_lock); 1512 ++fs->lfs_sleepers; 1513 mutex_exit(&lfs_lock); 1514 1515 error = lfs_segwait(fsidp, tvp); 1516 1517 mutex_enter(&lfs_lock); 1518 if (--fs->lfs_sleepers == 0) 1519 wakeup(&fs->lfs_sleepers); 1520 mutex_exit(&lfs_lock); 1521 return error; 1522 1523 case LFCNBMAPV: 1524 case LFCNMARKV: 1525 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1526 1527 blkcnt = blkvp.blkcnt; 1528 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1529 return (EINVAL); 1530 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1531 if ((error = copyin(blkvp.blkiov, blkiov, 1532 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1533 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1534 return error; 1535 } 1536 1537 mutex_enter(&lfs_lock); 1538 ++fs->lfs_sleepers; 1539 mutex_exit(&lfs_lock); 1540 if (ap->a_command == LFCNBMAPV) 1541 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1542 else /* LFCNMARKV */ 1543 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1544 if (error == 0) 1545 error = copyout(blkiov, blkvp.blkiov, 1546 blkcnt * sizeof(BLOCK_INFO)); 1547 mutex_enter(&lfs_lock); 1548 if (--fs->lfs_sleepers == 0) 1549 wakeup(&fs->lfs_sleepers); 1550 mutex_exit(&lfs_lock); 1551 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1552 return error; 1553 1554 case LFCNRECLAIM: 1555 /* 1556 * Flush dirops and write Ifile, allowing empty segments 1557 * to be immediately reclaimed. 1558 */ 1559 lfs_writer_enter(fs, "pndirop"); 1560 off = fs->lfs_offset; 1561 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1562 lfs_flush_dirops(fs); 1563 LFS_CLEANERINFO(cip, fs, bp); 1564 oclean = cip->clean; 1565 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1566 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1567 fs->lfs_sp->seg_flags |= SEGM_PROT; 1568 lfs_segunlock(fs); 1569 lfs_writer_leave(fs); 1570 1571 #ifdef DEBUG 1572 LFS_CLEANERINFO(cip, fs, bp); 1573 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1574 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1575 fs->lfs_offset - off, cip->clean - oclean, 1576 fs->lfs_activesb)); 1577 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1578 #endif 1579 1580 return 0; 1581 1582 case LFCNIFILEFH_COMPAT: 1583 /* Return the filehandle of the Ifile */ 1584 if ((error = kauth_authorize_system(l->l_cred, 1585 KAUTH_SYSTEM_FILEHANDLE, 0, NULL, NULL, NULL)) != 0) 1586 return (error); 1587 fhp = (struct fhandle *)ap->a_data; 1588 fhp->fh_fsid = *fsidp; 1589 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1590 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1591 1592 case LFCNIFILEFH_COMPAT2: 1593 case LFCNIFILEFH: 1594 /* Return the filehandle of the Ifile */ 1595 fhp = (struct fhandle *)ap->a_data; 1596 fhp->fh_fsid = *fsidp; 1597 fh_size = sizeof(struct lfs_fhandle) - 1598 offsetof(fhandle_t, fh_fid); 1599 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1600 1601 case LFCNREWIND: 1602 /* Move lfs_offset to the lowest-numbered segment */ 1603 return lfs_rewind(fs, *(int *)ap->a_data); 1604 1605 case LFCNINVAL: 1606 /* Mark a segment SEGUSE_INVAL */ 1607 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1608 if (sup->su_nbytes > 0) { 1609 brelse(bp, 0); 1610 lfs_unset_inval_all(fs); 1611 return EBUSY; 1612 } 1613 sup->su_flags |= SEGUSE_INVAL; 1614 VOP_BWRITE(bp->b_vp, bp); 1615 return 0; 1616 1617 case LFCNRESIZE: 1618 /* Resize the filesystem */ 1619 return lfs_resize_fs(fs, *(int *)ap->a_data); 1620 1621 case LFCNWRAPSTOP: 1622 case LFCNWRAPSTOP_COMPAT: 1623 /* 1624 * Hold lfs_newseg at segment 0; if requested, sleep until 1625 * the filesystem wraps around. To support external agents 1626 * (dump, fsck-based regression test) that need to look at 1627 * a snapshot of the filesystem, without necessarily 1628 * requiring that all fs activity stops. 1629 */ 1630 if (fs->lfs_stoplwp == curlwp) 1631 return EALREADY; 1632 1633 mutex_enter(&lfs_lock); 1634 while (fs->lfs_stoplwp != NULL) 1635 cv_wait(&fs->lfs_stopcv, &lfs_lock); 1636 fs->lfs_stoplwp = curlwp; 1637 if (fs->lfs_nowrap == 0) 1638 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1639 ++fs->lfs_nowrap; 1640 if (*(int *)ap->a_data == 1 1641 || ap->a_command == LFCNWRAPSTOP_COMPAT) { 1642 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1643 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1644 "segwrap", 0, &lfs_lock); 1645 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1646 if (error) { 1647 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1648 } 1649 } 1650 mutex_exit(&lfs_lock); 1651 return 0; 1652 1653 case LFCNWRAPGO: 1654 case LFCNWRAPGO_COMPAT: 1655 /* 1656 * Having done its work, the agent wakes up the writer. 1657 * If the argument is 1, it sleeps until a new segment 1658 * is selected. 1659 */ 1660 mutex_enter(&lfs_lock); 1661 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1662 ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1663 *((int *)ap->a_data)); 1664 mutex_exit(&lfs_lock); 1665 return error; 1666 1667 case LFCNWRAPPASS: 1668 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1669 return EALREADY; 1670 mutex_enter(&lfs_lock); 1671 if (fs->lfs_stoplwp != curlwp) { 1672 mutex_exit(&lfs_lock); 1673 return EALREADY; 1674 } 1675 if (fs->lfs_nowrap == 0) { 1676 mutex_exit(&lfs_lock); 1677 return EBUSY; 1678 } 1679 fs->lfs_wrappass = 1; 1680 wakeup(&fs->lfs_wrappass); 1681 /* Wait for the log to wrap, if asked */ 1682 if (*(int *)ap->a_data) { 1683 mutex_enter(ap->a_vp->v_interlock); 1684 lfs_vref(ap->a_vp); 1685 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1686 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1687 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1688 "segwrap", 0, &lfs_lock); 1689 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1690 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1691 lfs_vunref(ap->a_vp); 1692 } 1693 mutex_exit(&lfs_lock); 1694 return error; 1695 1696 case LFCNWRAPSTATUS: 1697 mutex_enter(&lfs_lock); 1698 *(int *)ap->a_data = fs->lfs_wrapstatus; 1699 mutex_exit(&lfs_lock); 1700 return 0; 1701 1702 default: 1703 return ufs_fcntl(v); 1704 } 1705 return 0; 1706 } 1707 1708 int 1709 lfs_getpages(void *v) 1710 { 1711 struct vop_getpages_args /* { 1712 struct vnode *a_vp; 1713 voff_t a_offset; 1714 struct vm_page **a_m; 1715 int *a_count; 1716 int a_centeridx; 1717 vm_prot_t a_access_type; 1718 int a_advice; 1719 int a_flags; 1720 } */ *ap = v; 1721 1722 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1723 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1724 return EPERM; 1725 } 1726 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1727 mutex_enter(&lfs_lock); 1728 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1729 mutex_exit(&lfs_lock); 1730 } 1731 1732 /* 1733 * we're relying on the fact that genfs_getpages() always read in 1734 * entire filesystem blocks. 1735 */ 1736 return genfs_getpages(v); 1737 } 1738 1739 /* 1740 * Wait for a page to become unbusy, possibly printing diagnostic messages 1741 * as well. 1742 * 1743 * Called with vp->v_interlock held; return with it held. 1744 */ 1745 static void 1746 wait_for_page(struct vnode *vp, struct vm_page *pg, const char *label) 1747 { 1748 if ((pg->flags & PG_BUSY) == 0) 1749 return; /* Nothing to wait for! */ 1750 1751 #if defined(DEBUG) && defined(UVM_PAGE_TRKOWN) 1752 static struct vm_page *lastpg; 1753 1754 if (label != NULL && pg != lastpg) { 1755 if (pg->owner_tag) { 1756 printf("lfs_putpages[%d.%d]: %s: page %p owner %d.%d [%s]\n", 1757 curproc->p_pid, curlwp->l_lid, label, 1758 pg, pg->owner, pg->lowner, pg->owner_tag); 1759 } else { 1760 printf("lfs_putpages[%d.%d]: %s: page %p unowned?!\n", 1761 curproc->p_pid, curlwp->l_lid, label, pg); 1762 } 1763 } 1764 lastpg = pg; 1765 #endif 1766 1767 pg->flags |= PG_WANTED; 1768 UVM_UNLOCK_AND_WAIT(pg, vp->v_interlock, 0, "lfsput", 0); 1769 mutex_enter(vp->v_interlock); 1770 } 1771 1772 /* 1773 * This routine is called by lfs_putpages() when it can't complete the 1774 * write because a page is busy. This means that either (1) someone, 1775 * possibly the pagedaemon, is looking at this page, and will give it up 1776 * presently; or (2) we ourselves are holding the page busy in the 1777 * process of being written (either gathered or actually on its way to 1778 * disk). We don't need to give up the segment lock, but we might need 1779 * to call lfs_writeseg() to expedite the page's journey to disk. 1780 * 1781 * Called with vp->v_interlock held; return with it held. 1782 */ 1783 /* #define BUSYWAIT */ 1784 static void 1785 write_and_wait(struct lfs *fs, struct vnode *vp, struct vm_page *pg, 1786 int seglocked, const char *label) 1787 { 1788 #ifndef BUSYWAIT 1789 struct inode *ip = VTOI(vp); 1790 struct segment *sp = fs->lfs_sp; 1791 int count = 0; 1792 1793 if (pg == NULL) 1794 return; 1795 1796 while (pg->flags & PG_BUSY && 1797 pg->uobject == &vp->v_uobj) { 1798 mutex_exit(vp->v_interlock); 1799 if (sp->cbpp - sp->bpp > 1) { 1800 /* Write gathered pages */ 1801 lfs_updatemeta(sp); 1802 lfs_release_finfo(fs); 1803 (void) lfs_writeseg(fs, sp); 1804 1805 /* 1806 * Reinitialize FIP 1807 */ 1808 KASSERT(sp->vp == vp); 1809 lfs_acquire_finfo(fs, ip->i_number, 1810 ip->i_gen); 1811 } 1812 ++count; 1813 mutex_enter(vp->v_interlock); 1814 wait_for_page(vp, pg, label); 1815 } 1816 if (label != NULL && count > 1) 1817 printf("lfs_putpages[%d]: %s: %sn = %d\n", curproc->p_pid, 1818 label, (count > 0 ? "looping, " : ""), count); 1819 #else 1820 preempt(1); 1821 #endif 1822 } 1823 1824 /* 1825 * Make sure that for all pages in every block in the given range, 1826 * either all are dirty or all are clean. If any of the pages 1827 * we've seen so far are dirty, put the vnode on the paging chain, 1828 * and mark it IN_PAGING. 1829 * 1830 * If checkfirst != 0, don't check all the pages but return at the 1831 * first dirty page. 1832 */ 1833 static int 1834 check_dirty(struct lfs *fs, struct vnode *vp, 1835 off_t startoffset, off_t endoffset, off_t blkeof, 1836 int flags, int checkfirst, struct vm_page **pgp) 1837 { 1838 int by_list; 1839 struct vm_page *curpg = NULL; /* XXX: gcc */ 1840 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1841 off_t soff = 0; /* XXX: gcc */ 1842 voff_t off; 1843 int i; 1844 int nonexistent; 1845 int any_dirty; /* number of dirty pages */ 1846 int dirty; /* number of dirty pages in a block */ 1847 int tdirty; 1848 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1849 int pagedaemon = (curlwp == uvm.pagedaemon_lwp); 1850 1851 ASSERT_MAYBE_SEGLOCK(fs); 1852 top: 1853 by_list = (vp->v_uobj.uo_npages <= 1854 ((endoffset - startoffset) >> PAGE_SHIFT) * 1855 UVM_PAGE_TREE_PENALTY); 1856 any_dirty = 0; 1857 1858 if (by_list) { 1859 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1860 } else { 1861 soff = startoffset; 1862 } 1863 while (by_list || soff < MIN(blkeof, endoffset)) { 1864 if (by_list) { 1865 /* 1866 * Find the first page in a block. Skip 1867 * blocks outside our area of interest or beyond 1868 * the end of file. 1869 */ 1870 KASSERT(curpg == NULL 1871 || (curpg->flags & PG_MARKER) == 0); 1872 if (pages_per_block > 1) { 1873 while (curpg && 1874 ((curpg->offset & fs->lfs_bmask) || 1875 curpg->offset >= vp->v_size || 1876 curpg->offset >= endoffset)) { 1877 curpg = TAILQ_NEXT(curpg, listq.queue); 1878 KASSERT(curpg == NULL || 1879 (curpg->flags & PG_MARKER) == 0); 1880 } 1881 } 1882 if (curpg == NULL) 1883 break; 1884 soff = curpg->offset; 1885 } 1886 1887 /* 1888 * Mark all pages in extended range busy; find out if any 1889 * of them are dirty. 1890 */ 1891 nonexistent = dirty = 0; 1892 for (i = 0; i == 0 || i < pages_per_block; i++) { 1893 if (by_list && pages_per_block <= 1) { 1894 pgs[i] = pg = curpg; 1895 } else { 1896 off = soff + (i << PAGE_SHIFT); 1897 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1898 if (pg == NULL) { 1899 ++nonexistent; 1900 continue; 1901 } 1902 } 1903 KASSERT(pg != NULL); 1904 1905 /* 1906 * If we're holding the segment lock, we can deadlock 1907 * against a process that has our page and is waiting 1908 * for the cleaner, while the cleaner waits for the 1909 * segment lock. Just bail in that case. 1910 */ 1911 if ((pg->flags & PG_BUSY) && 1912 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1913 if (i > 0) 1914 uvm_page_unbusy(pgs, i); 1915 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1916 if (pgp) 1917 *pgp = pg; 1918 return -1; 1919 } 1920 1921 while (pg->flags & PG_BUSY) { 1922 wait_for_page(vp, pg, NULL); 1923 if (i > 0) 1924 uvm_page_unbusy(pgs, i); 1925 goto top; 1926 } 1927 pg->flags |= PG_BUSY; 1928 UVM_PAGE_OWN(pg, "lfs_putpages"); 1929 1930 pmap_page_protect(pg, VM_PROT_NONE); 1931 tdirty = (pmap_clear_modify(pg) || 1932 (pg->flags & PG_CLEAN) == 0); 1933 dirty += tdirty; 1934 } 1935 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1936 if (by_list) { 1937 curpg = TAILQ_NEXT(curpg, listq.queue); 1938 } else { 1939 soff += fs->lfs_bsize; 1940 } 1941 continue; 1942 } 1943 1944 any_dirty += dirty; 1945 KASSERT(nonexistent == 0); 1946 1947 /* 1948 * If any are dirty make all dirty; unbusy them, 1949 * but if we were asked to clean, wire them so that 1950 * the pagedaemon doesn't bother us about them while 1951 * they're on their way to disk. 1952 */ 1953 for (i = 0; i == 0 || i < pages_per_block; i++) { 1954 pg = pgs[i]; 1955 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1956 if (dirty) { 1957 pg->flags &= ~PG_CLEAN; 1958 if (flags & PGO_FREE) { 1959 /* 1960 * Wire the page so that 1961 * pdaemon doesn't see it again. 1962 */ 1963 mutex_enter(&uvm_pageqlock); 1964 uvm_pagewire(pg); 1965 mutex_exit(&uvm_pageqlock); 1966 1967 /* Suspended write flag */ 1968 pg->flags |= PG_DELWRI; 1969 } 1970 } 1971 if (pg->flags & PG_WANTED) 1972 wakeup(pg); 1973 pg->flags &= ~(PG_WANTED|PG_BUSY); 1974 UVM_PAGE_OWN(pg, NULL); 1975 } 1976 1977 if (checkfirst && any_dirty) 1978 break; 1979 1980 if (by_list) { 1981 curpg = TAILQ_NEXT(curpg, listq.queue); 1982 } else { 1983 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1984 } 1985 } 1986 1987 return any_dirty; 1988 } 1989 1990 /* 1991 * lfs_putpages functions like genfs_putpages except that 1992 * 1993 * (1) It needs to bounds-check the incoming requests to ensure that 1994 * they are block-aligned; if they are not, expand the range and 1995 * do the right thing in case, e.g., the requested range is clean 1996 * but the expanded range is dirty. 1997 * 1998 * (2) It needs to explicitly send blocks to be written when it is done. 1999 * If VOP_PUTPAGES is called without the seglock held, we simply take 2000 * the seglock and let lfs_segunlock wait for us. 2001 * XXX There might be a bad situation if we have to flush a vnode while 2002 * XXX lfs_markv is in operation. As of this writing we panic in this 2003 * XXX case. 2004 * 2005 * Assumptions: 2006 * 2007 * (1) The caller does not hold any pages in this vnode busy. If it does, 2008 * there is a danger that when we expand the page range and busy the 2009 * pages we will deadlock. 2010 * 2011 * (2) We are called with vp->v_interlock held; we must return with it 2012 * released. 2013 * 2014 * (3) We don't absolutely have to free pages right away, provided that 2015 * the request does not have PGO_SYNCIO. When the pagedaemon gives 2016 * us a request with PGO_FREE, we take the pages out of the paging 2017 * queue and wake up the writer, which will handle freeing them for us. 2018 * 2019 * We ensure that for any filesystem block, all pages for that 2020 * block are either resident or not, even if those pages are higher 2021 * than EOF; that means that we will be getting requests to free 2022 * "unused" pages above EOF all the time, and should ignore them. 2023 * 2024 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 2025 * into has been set up for us by lfs_writefile. If not, we will 2026 * have to handle allocating and/or freeing an finfo entry. 2027 * 2028 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 2029 */ 2030 2031 /* How many times to loop before we should start to worry */ 2032 #define TOOMANY 4 2033 2034 int 2035 lfs_putpages(void *v) 2036 { 2037 int error; 2038 struct vop_putpages_args /* { 2039 struct vnode *a_vp; 2040 voff_t a_offlo; 2041 voff_t a_offhi; 2042 int a_flags; 2043 } */ *ap = v; 2044 struct vnode *vp; 2045 struct inode *ip; 2046 struct lfs *fs; 2047 struct segment *sp; 2048 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 2049 off_t off, max_endoffset; 2050 bool seglocked, sync, pagedaemon; 2051 struct vm_page *pg, *busypg; 2052 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 2053 #ifdef DEBUG 2054 int debug_n_again, debug_n_dirtyclean; 2055 #endif 2056 2057 vp = ap->a_vp; 2058 ip = VTOI(vp); 2059 fs = ip->i_lfs; 2060 sync = (ap->a_flags & PGO_SYNCIO) != 0; 2061 pagedaemon = (curlwp == uvm.pagedaemon_lwp); 2062 2063 /* Putpages does nothing for metadata. */ 2064 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 2065 mutex_exit(vp->v_interlock); 2066 return 0; 2067 } 2068 2069 /* 2070 * If there are no pages, don't do anything. 2071 */ 2072 if (vp->v_uobj.uo_npages == 0) { 2073 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 2074 (vp->v_iflag & VI_ONWORKLST) && 2075 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 2076 vp->v_iflag &= ~VI_WRMAPDIRTY; 2077 vn_syncer_remove_from_worklist(vp); 2078 } 2079 mutex_exit(vp->v_interlock); 2080 2081 /* Remove us from paging queue, if we were on it */ 2082 mutex_enter(&lfs_lock); 2083 if (ip->i_flags & IN_PAGING) { 2084 ip->i_flags &= ~IN_PAGING; 2085 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2086 } 2087 mutex_exit(&lfs_lock); 2088 return 0; 2089 } 2090 2091 blkeof = blkroundup(fs, ip->i_size); 2092 2093 /* 2094 * Ignore requests to free pages past EOF but in the same block 2095 * as EOF, unless the request is synchronous. (If the request is 2096 * sync, it comes from lfs_truncate.) 2097 * XXXUBC Make these pages look "active" so the pagedaemon won't 2098 * XXXUBC bother us with them again. 2099 */ 2100 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 2101 origoffset = ap->a_offlo; 2102 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 2103 pg = uvm_pagelookup(&vp->v_uobj, off); 2104 KASSERT(pg != NULL); 2105 while (pg->flags & PG_BUSY) { 2106 pg->flags |= PG_WANTED; 2107 UVM_UNLOCK_AND_WAIT(pg, vp->v_interlock, 0, 2108 "lfsput2", 0); 2109 mutex_enter(vp->v_interlock); 2110 } 2111 mutex_enter(&uvm_pageqlock); 2112 uvm_pageactivate(pg); 2113 mutex_exit(&uvm_pageqlock); 2114 } 2115 ap->a_offlo = blkeof; 2116 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 2117 mutex_exit(vp->v_interlock); 2118 return 0; 2119 } 2120 } 2121 2122 /* 2123 * Extend page range to start and end at block boundaries. 2124 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 2125 */ 2126 origoffset = ap->a_offlo; 2127 origendoffset = ap->a_offhi; 2128 startoffset = origoffset & ~(fs->lfs_bmask); 2129 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 2130 << fs->lfs_bshift; 2131 2132 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2133 endoffset = max_endoffset; 2134 origendoffset = endoffset; 2135 } else { 2136 origendoffset = round_page(ap->a_offhi); 2137 endoffset = round_page(blkroundup(fs, origendoffset)); 2138 } 2139 2140 KASSERT(startoffset > 0 || endoffset >= startoffset); 2141 if (startoffset == endoffset) { 2142 /* Nothing to do, why were we called? */ 2143 mutex_exit(vp->v_interlock); 2144 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 2145 PRId64 "\n", startoffset)); 2146 return 0; 2147 } 2148 2149 ap->a_offlo = startoffset; 2150 ap->a_offhi = endoffset; 2151 2152 /* 2153 * If not cleaning, just send the pages through genfs_putpages 2154 * to be returned to the pool. 2155 */ 2156 if (!(ap->a_flags & PGO_CLEANIT)) 2157 return genfs_putpages(v); 2158 2159 /* Set PGO_BUSYFAIL to avoid deadlocks */ 2160 ap->a_flags |= PGO_BUSYFAIL; 2161 2162 /* 2163 * Likewise, if we are asked to clean but the pages are not 2164 * dirty, we can just free them using genfs_putpages. 2165 */ 2166 #ifdef DEBUG 2167 debug_n_dirtyclean = 0; 2168 #endif 2169 do { 2170 int r; 2171 2172 /* Count the number of dirty pages */ 2173 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2174 ap->a_flags, 1, NULL); 2175 if (r < 0) { 2176 /* Pages are busy with another process */ 2177 mutex_exit(vp->v_interlock); 2178 return EDEADLK; 2179 } 2180 if (r > 0) /* Some pages are dirty */ 2181 break; 2182 2183 /* 2184 * Sometimes pages are dirtied between the time that 2185 * we check and the time we try to clean them. 2186 * Instruct lfs_gop_write to return EDEADLK in this case 2187 * so we can write them properly. 2188 */ 2189 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2190 r = genfs_do_putpages(vp, startoffset, endoffset, 2191 ap->a_flags & ~PGO_SYNCIO, &busypg); 2192 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2193 if (r != EDEADLK) 2194 return r; 2195 2196 /* One of the pages was busy. Start over. */ 2197 mutex_enter(vp->v_interlock); 2198 wait_for_page(vp, busypg, "dirtyclean"); 2199 #ifdef DEBUG 2200 ++debug_n_dirtyclean; 2201 #endif 2202 } while(1); 2203 2204 #ifdef DEBUG 2205 if (debug_n_dirtyclean > TOOMANY) 2206 printf("lfs_putpages: dirtyclean: looping, n = %d\n", 2207 debug_n_dirtyclean); 2208 #endif 2209 2210 /* 2211 * Dirty and asked to clean. 2212 * 2213 * Pagedaemon can't actually write LFS pages; wake up 2214 * the writer to take care of that. The writer will 2215 * notice the pager inode queue and act on that. 2216 * 2217 * XXX We must drop the vp->interlock before taking the lfs_lock or we 2218 * get a nasty deadlock with lfs_flush_pchain(). 2219 */ 2220 if (pagedaemon) { 2221 mutex_exit(vp->v_interlock); 2222 mutex_enter(&lfs_lock); 2223 if (!(ip->i_flags & IN_PAGING)) { 2224 ip->i_flags |= IN_PAGING; 2225 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2226 } 2227 wakeup(&lfs_writer_daemon); 2228 mutex_exit(&lfs_lock); 2229 preempt(); 2230 return EWOULDBLOCK; 2231 } 2232 2233 /* 2234 * If this is a file created in a recent dirop, we can't flush its 2235 * inode until the dirop is complete. Drain dirops, then flush the 2236 * filesystem (taking care of any other pending dirops while we're 2237 * at it). 2238 */ 2239 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2240 (vp->v_uflag & VU_DIROP)) { 2241 int locked; 2242 2243 DLOG((DLOG_PAGE, "lfs_putpages: flushing VU_DIROP\n")); 2244 /* XXX VOP_ISLOCKED() may not be used for lock decisions. */ 2245 locked = (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 2246 mutex_exit(vp->v_interlock); 2247 lfs_writer_enter(fs, "ppdirop"); 2248 if (locked) 2249 VOP_UNLOCK(vp); /* XXX why? */ 2250 2251 mutex_enter(&lfs_lock); 2252 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2253 mutex_exit(&lfs_lock); 2254 2255 if (locked) 2256 VOP_LOCK(vp, LK_EXCLUSIVE); 2257 mutex_enter(vp->v_interlock); 2258 lfs_writer_leave(fs); 2259 2260 /* XXX the flush should have taken care of this one too! */ 2261 } 2262 2263 /* 2264 * This is it. We are going to write some pages. From here on 2265 * down it's all just mechanics. 2266 * 2267 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2268 */ 2269 ap->a_flags &= ~PGO_SYNCIO; 2270 2271 /* 2272 * If we've already got the seglock, flush the node and return. 2273 * The FIP has already been set up for us by lfs_writefile, 2274 * and FIP cleanup and lfs_updatemeta will also be done there, 2275 * unless genfs_putpages returns EDEADLK; then we must flush 2276 * what we have, and correct FIP and segment header accounting. 2277 */ 2278 get_seglock: 2279 /* 2280 * If we are not called with the segment locked, lock it. 2281 * Account for a new FIP in the segment header, and set sp->vp. 2282 * (This should duplicate the setup at the top of lfs_writefile().) 2283 */ 2284 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2285 if (!seglocked) { 2286 mutex_exit(vp->v_interlock); 2287 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2288 if (error != 0) 2289 return error; 2290 mutex_enter(vp->v_interlock); 2291 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2292 } 2293 sp = fs->lfs_sp; 2294 KASSERT(sp->vp == NULL); 2295 sp->vp = vp; 2296 2297 /* 2298 * Ensure that the partial segment is marked SS_DIROP if this 2299 * vnode is a DIROP. 2300 */ 2301 if (!seglocked && vp->v_uflag & VU_DIROP) 2302 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2303 2304 /* 2305 * Loop over genfs_putpages until all pages are gathered. 2306 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2307 * Whenever we lose the interlock we have to rerun check_dirty, as 2308 * well, since more pages might have been dirtied in our absence. 2309 */ 2310 #ifdef DEBUG 2311 debug_n_again = 0; 2312 #endif 2313 do { 2314 busypg = NULL; 2315 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2316 ap->a_flags, 0, &busypg) < 0) { 2317 mutex_exit(vp->v_interlock); 2318 2319 mutex_enter(vp->v_interlock); 2320 write_and_wait(fs, vp, busypg, seglocked, NULL); 2321 if (!seglocked) { 2322 mutex_exit(vp->v_interlock); 2323 lfs_release_finfo(fs); 2324 lfs_segunlock(fs); 2325 mutex_enter(vp->v_interlock); 2326 } 2327 sp->vp = NULL; 2328 goto get_seglock; 2329 } 2330 2331 busypg = NULL; 2332 error = genfs_do_putpages(vp, startoffset, endoffset, 2333 ap->a_flags, &busypg); 2334 2335 if (error == EDEADLK || error == EAGAIN) { 2336 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2337 " %d ino %d off %x (seg %d)\n", error, 2338 ip->i_number, fs->lfs_offset, 2339 dtosn(fs, fs->lfs_offset))); 2340 2341 mutex_enter(vp->v_interlock); 2342 write_and_wait(fs, vp, busypg, seglocked, "again"); 2343 } 2344 #ifdef DEBUG 2345 ++debug_n_again; 2346 #endif 2347 } while (error == EDEADLK); 2348 #ifdef DEBUG 2349 if (debug_n_again > TOOMANY) 2350 printf("lfs_putpages: again: looping, n = %d\n", debug_n_again); 2351 #endif 2352 2353 KASSERT(sp != NULL && sp->vp == vp); 2354 if (!seglocked) { 2355 sp->vp = NULL; 2356 2357 /* Write indirect blocks as well */ 2358 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2359 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2360 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2361 2362 KASSERT(sp->vp == NULL); 2363 sp->vp = vp; 2364 } 2365 2366 /* 2367 * Blocks are now gathered into a segment waiting to be written. 2368 * All that's left to do is update metadata, and write them. 2369 */ 2370 lfs_updatemeta(sp); 2371 KASSERT(sp->vp == vp); 2372 sp->vp = NULL; 2373 2374 /* 2375 * If we were called from lfs_writefile, we don't need to clean up 2376 * the FIP or unlock the segment lock. We're done. 2377 */ 2378 if (seglocked) 2379 return error; 2380 2381 /* Clean up FIP and send it to disk. */ 2382 lfs_release_finfo(fs); 2383 lfs_writeseg(fs, fs->lfs_sp); 2384 2385 /* 2386 * Remove us from paging queue if we wrote all our pages. 2387 */ 2388 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2389 mutex_enter(&lfs_lock); 2390 if (ip->i_flags & IN_PAGING) { 2391 ip->i_flags &= ~IN_PAGING; 2392 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2393 } 2394 mutex_exit(&lfs_lock); 2395 } 2396 2397 /* 2398 * XXX - with the malloc/copy writeseg, the pages are freed by now 2399 * even if we don't wait (e.g. if we hold a nested lock). This 2400 * will not be true if we stop using malloc/copy. 2401 */ 2402 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2403 lfs_segunlock(fs); 2404 2405 /* 2406 * Wait for v_numoutput to drop to zero. The seglock should 2407 * take care of this, but there is a slight possibility that 2408 * aiodoned might not have got around to our buffers yet. 2409 */ 2410 if (sync) { 2411 mutex_enter(vp->v_interlock); 2412 while (vp->v_numoutput > 0) { 2413 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2414 " num %d\n", ip->i_number, vp->v_numoutput)); 2415 cv_wait(&vp->v_cv, vp->v_interlock); 2416 } 2417 mutex_exit(vp->v_interlock); 2418 } 2419 return error; 2420 } 2421 2422 /* 2423 * Return the last logical file offset that should be written for this file 2424 * if we're doing a write that ends at "size". If writing, we need to know 2425 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2426 * to know about entire blocks. 2427 */ 2428 void 2429 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2430 { 2431 struct inode *ip = VTOI(vp); 2432 struct lfs *fs = ip->i_lfs; 2433 daddr_t olbn, nlbn; 2434 2435 olbn = lblkno(fs, ip->i_size); 2436 nlbn = lblkno(fs, size); 2437 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2438 *eobp = fragroundup(fs, size); 2439 } else { 2440 *eobp = blkroundup(fs, size); 2441 } 2442 } 2443 2444 #ifdef DEBUG 2445 void lfs_dump_vop(void *); 2446 2447 void 2448 lfs_dump_vop(void *v) 2449 { 2450 struct vop_putpages_args /* { 2451 struct vnode *a_vp; 2452 voff_t a_offlo; 2453 voff_t a_offhi; 2454 int a_flags; 2455 } */ *ap = v; 2456 2457 #ifdef DDB 2458 vfs_vnode_print(ap->a_vp, 0, printf); 2459 #endif 2460 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2461 } 2462 #endif 2463 2464 int 2465 lfs_mmap(void *v) 2466 { 2467 struct vop_mmap_args /* { 2468 const struct vnodeop_desc *a_desc; 2469 struct vnode *a_vp; 2470 vm_prot_t a_prot; 2471 kauth_cred_t a_cred; 2472 } */ *ap = v; 2473 2474 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2475 return EOPNOTSUPP; 2476 return ufs_mmap(v); 2477 } 2478
Indexes created Tue Sep 30 20:09:53 GMT 2025