lfs_vnops.c revision 1.235
1 /* $NetBSD: lfs_vnops.c,v 1.235 2011/06/12 03:36:01 rmind 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 1.235 2011/06/12 03:36:01 rmind 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; 609 610 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 611 vput(ap->a_dvp); 612 return error; 613 } 614 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 615 ap->a_dvp, vpp, ap->a_cnp); 616 617 /* Either way we're done with the dirop at this point */ 618 SET_ENDOP_CREATE_AP(ap, "mknod"); 619 620 if (error) 621 return (error); 622 623 ip = VTOI(*vpp); 624 mp = (*vpp)->v_mount; 625 ino = ip->i_number; 626 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 627 if (vap->va_rdev != VNOVAL) { 628 /* 629 * Want to be able to use this to make badblock 630 * inodes, so don't truncate the dev number. 631 */ 632 #if 0 633 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 634 UFS_MPNEEDSWAP((*vpp)->v_mount)); 635 #else 636 ip->i_ffs1_rdev = vap->va_rdev; 637 #endif 638 } 639 640 /* 641 * Call fsync to write the vnode so that we don't have to deal with 642 * flushing it when it's marked VU_DIROP|VI_XLOCK. 643 * 644 * XXX KS - If we can't flush we also can't call vgone(), so must 645 * return. But, that leaves this vnode in limbo, also not good. 646 * Can this ever happen (barring hardware failure)? 647 */ 648 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0)) != 0) { 649 panic("lfs_mknod: couldn't fsync (ino %llu)", 650 (unsigned long long)ino); 651 /* return (error); */ 652 } 653 /* 654 * Remove vnode so that it will be reloaded by VFS_VGET and 655 * checked to see if it is an alias of an existing entry in 656 * the inode cache. 657 */ 658 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 659 660 VOP_UNLOCK(*vpp); 661 (*vpp)->v_type = VNON; 662 vgone(*vpp); 663 error = VFS_VGET(mp, ino, vpp); 664 665 if (error != 0) { 666 *vpp = NULL; 667 return (error); 668 } 669 return (0); 670 } 671 672 int 673 lfs_create(void *v) 674 { 675 struct vop_create_args /* { 676 struct vnode *a_dvp; 677 struct vnode **a_vpp; 678 struct componentname *a_cnp; 679 struct vattr *a_vap; 680 } */ *ap = v; 681 int error; 682 683 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 684 vput(ap->a_dvp); 685 return error; 686 } 687 error = ufs_create(ap); 688 SET_ENDOP_CREATE_AP(ap, "create"); 689 return (error); 690 } 691 692 int 693 lfs_mkdir(void *v) 694 { 695 struct vop_mkdir_args /* { 696 struct vnode *a_dvp; 697 struct vnode **a_vpp; 698 struct componentname *a_cnp; 699 struct vattr *a_vap; 700 } */ *ap = v; 701 int error; 702 703 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 704 vput(ap->a_dvp); 705 return error; 706 } 707 error = ufs_mkdir(ap); 708 SET_ENDOP_CREATE_AP(ap, "mkdir"); 709 return (error); 710 } 711 712 int 713 lfs_remove(void *v) 714 { 715 struct vop_remove_args /* { 716 struct vnode *a_dvp; 717 struct vnode *a_vp; 718 struct componentname *a_cnp; 719 } */ *ap = v; 720 struct vnode *dvp, *vp; 721 struct inode *ip; 722 int error; 723 724 dvp = ap->a_dvp; 725 vp = ap->a_vp; 726 ip = VTOI(vp); 727 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 728 if (dvp == vp) 729 vrele(vp); 730 else 731 vput(vp); 732 vput(dvp); 733 return error; 734 } 735 error = ufs_remove(ap); 736 if (ip->i_nlink == 0) 737 lfs_orphan(ip->i_lfs, ip->i_number); 738 SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove"); 739 return (error); 740 } 741 742 int 743 lfs_rmdir(void *v) 744 { 745 struct vop_rmdir_args /* { 746 struct vnodeop_desc *a_desc; 747 struct vnode *a_dvp; 748 struct vnode *a_vp; 749 struct componentname *a_cnp; 750 } */ *ap = v; 751 struct vnode *vp; 752 struct inode *ip; 753 int error; 754 755 vp = ap->a_vp; 756 ip = VTOI(vp); 757 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 758 if (ap->a_dvp == vp) 759 vrele(ap->a_dvp); 760 else 761 vput(ap->a_dvp); 762 vput(vp); 763 return error; 764 } 765 error = ufs_rmdir(ap); 766 if (ip->i_nlink == 0) 767 lfs_orphan(ip->i_lfs, ip->i_number); 768 SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 769 return (error); 770 } 771 772 int 773 lfs_link(void *v) 774 { 775 struct vop_link_args /* { 776 struct vnode *a_dvp; 777 struct vnode *a_vp; 778 struct componentname *a_cnp; 779 } */ *ap = v; 780 int error; 781 struct vnode **vpp = NULL; 782 783 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 784 vput(ap->a_dvp); 785 return error; 786 } 787 error = ufs_link(ap); 788 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 789 return (error); 790 } 791 792 int 793 lfs_rename(void *v) 794 { 795 struct vop_rename_args /* { 796 struct vnode *a_fdvp; 797 struct vnode *a_fvp; 798 struct componentname *a_fcnp; 799 struct vnode *a_tdvp; 800 struct vnode *a_tvp; 801 struct componentname *a_tcnp; 802 } */ *ap = v; 803 struct vnode *tvp, *fvp, *tdvp, *fdvp; 804 struct componentname *tcnp, *fcnp; 805 int error; 806 struct lfs *fs; 807 808 fs = VTOI(ap->a_fdvp)->i_lfs; 809 tvp = ap->a_tvp; 810 tdvp = ap->a_tdvp; 811 tcnp = ap->a_tcnp; 812 fvp = ap->a_fvp; 813 fdvp = ap->a_fdvp; 814 fcnp = ap->a_fcnp; 815 816 /* 817 * Check for cross-device rename. 818 * If it is, we don't want to set dirops, just error out. 819 * (In particular note that MARK_VNODE(tdvp) will DTWT on 820 * a cross-device rename.) 821 * 822 * Copied from ufs_rename. 823 */ 824 if ((fvp->v_mount != tdvp->v_mount) || 825 (tvp && (fvp->v_mount != tvp->v_mount))) { 826 error = EXDEV; 827 goto errout; 828 } 829 830 /* 831 * Check to make sure we're not renaming a vnode onto itself 832 * (deleting a hard link by renaming one name onto another); 833 * if we are we can't recursively call VOP_REMOVE since that 834 * would leave us with an unaccounted-for number of live dirops. 835 * 836 * Inline the relevant section of ufs_rename here, *before* 837 * calling SET_DIROP_REMOVE. 838 */ 839 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 840 (VTOI(tdvp)->i_flags & APPEND))) { 841 error = EPERM; 842 goto errout; 843 } 844 if (fvp == tvp) { 845 if (fvp->v_type == VDIR) { 846 error = EINVAL; 847 goto errout; 848 } 849 850 /* Release destination completely. */ 851 VOP_ABORTOP(tdvp, tcnp); 852 vput(tdvp); 853 vput(tvp); 854 855 /* Delete source. */ 856 vrele(fvp); 857 fcnp->cn_flags &= ~(MODMASK); 858 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 859 fcnp->cn_nameiop = DELETE; 860 vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY); 861 if ((error = relookup(fdvp, &fvp, fcnp, 0))) { 862 vput(fdvp); 863 return (error); 864 } 865 return (VOP_REMOVE(fdvp, fvp, fcnp)); 866 } 867 868 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 869 goto errout; 870 MARK_VNODE(fdvp); 871 MARK_VNODE(fvp); 872 873 error = ufs_rename(ap); 874 UNMARK_VNODE(fdvp); 875 UNMARK_VNODE(fvp); 876 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 877 return (error); 878 879 errout: 880 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 881 if (tdvp == tvp) 882 vrele(tdvp); 883 else 884 vput(tdvp); 885 if (tvp) 886 vput(tvp); 887 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 888 vrele(fdvp); 889 vrele(fvp); 890 return (error); 891 } 892 893 /* XXX hack to avoid calling ITIMES in getattr */ 894 int 895 lfs_getattr(void *v) 896 { 897 struct vop_getattr_args /* { 898 struct vnode *a_vp; 899 struct vattr *a_vap; 900 kauth_cred_t a_cred; 901 } */ *ap = v; 902 struct vnode *vp = ap->a_vp; 903 struct inode *ip = VTOI(vp); 904 struct vattr *vap = ap->a_vap; 905 struct lfs *fs = ip->i_lfs; 906 /* 907 * Copy from inode table 908 */ 909 vap->va_fsid = ip->i_dev; 910 vap->va_fileid = ip->i_number; 911 vap->va_mode = ip->i_mode & ~IFMT; 912 vap->va_nlink = ip->i_nlink; 913 vap->va_uid = ip->i_uid; 914 vap->va_gid = ip->i_gid; 915 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 916 vap->va_size = vp->v_size; 917 vap->va_atime.tv_sec = ip->i_ffs1_atime; 918 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 919 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 920 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 921 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 922 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 923 vap->va_flags = ip->i_flags; 924 vap->va_gen = ip->i_gen; 925 /* this doesn't belong here */ 926 if (vp->v_type == VBLK) 927 vap->va_blocksize = BLKDEV_IOSIZE; 928 else if (vp->v_type == VCHR) 929 vap->va_blocksize = MAXBSIZE; 930 else 931 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 932 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 933 vap->va_type = vp->v_type; 934 vap->va_filerev = ip->i_modrev; 935 return (0); 936 } 937 938 /* 939 * Check to make sure the inode blocks won't choke the buffer 940 * cache, then call ufs_setattr as usual. 941 */ 942 int 943 lfs_setattr(void *v) 944 { 945 struct vop_setattr_args /* { 946 struct vnode *a_vp; 947 struct vattr *a_vap; 948 kauth_cred_t a_cred; 949 } */ *ap = v; 950 struct vnode *vp = ap->a_vp; 951 952 lfs_check(vp, LFS_UNUSED_LBN, 0); 953 return ufs_setattr(v); 954 } 955 956 /* 957 * Release the block we hold on lfs_newseg wrapping. Called on file close, 958 * or explicitly from LFCNWRAPGO. Called with the interlock held. 959 */ 960 static int 961 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor) 962 { 963 if (fs->lfs_stoplwp != curlwp) 964 return EBUSY; 965 966 fs->lfs_stoplwp = NULL; 967 cv_signal(&fs->lfs_stopcv); 968 969 KASSERT(fs->lfs_nowrap > 0); 970 if (fs->lfs_nowrap <= 0) { 971 return 0; 972 } 973 974 if (--fs->lfs_nowrap == 0) { 975 log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt); 976 wakeup(&fs->lfs_wrappass); 977 lfs_wakeup_cleaner(fs); 978 } 979 if (waitfor) { 980 mtsleep(&fs->lfs_nextseg, PCATCH | PUSER, "segment", 981 0, &lfs_lock); 982 } 983 984 return 0; 985 } 986 987 /* 988 * Close called 989 */ 990 /* ARGSUSED */ 991 int 992 lfs_close(void *v) 993 { 994 struct vop_close_args /* { 995 struct vnode *a_vp; 996 int a_fflag; 997 kauth_cred_t a_cred; 998 } */ *ap = v; 999 struct vnode *vp = ap->a_vp; 1000 struct inode *ip = VTOI(vp); 1001 struct lfs *fs = ip->i_lfs; 1002 1003 if ((ip->i_number == ROOTINO || ip->i_number == LFS_IFILE_INUM) && 1004 fs->lfs_stoplwp == curlwp) { 1005 mutex_enter(&lfs_lock); 1006 log(LOG_NOTICE, "lfs_close: releasing log wrap control\n"); 1007 lfs_wrapgo(fs, ip, 0); 1008 mutex_exit(&lfs_lock); 1009 } 1010 1011 if (vp == ip->i_lfs->lfs_ivnode && 1012 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 1013 return 0; 1014 1015 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 1016 LFS_ITIMES(ip, NULL, NULL, NULL); 1017 } 1018 return (0); 1019 } 1020 1021 /* 1022 * Close wrapper for special devices. 1023 * 1024 * Update the times on the inode then do device close. 1025 */ 1026 int 1027 lfsspec_close(void *v) 1028 { 1029 struct vop_close_args /* { 1030 struct vnode *a_vp; 1031 int a_fflag; 1032 kauth_cred_t a_cred; 1033 } */ *ap = v; 1034 struct vnode *vp; 1035 struct inode *ip; 1036 1037 vp = ap->a_vp; 1038 ip = VTOI(vp); 1039 if (vp->v_usecount > 1) { 1040 LFS_ITIMES(ip, NULL, NULL, NULL); 1041 } 1042 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1043 } 1044 1045 /* 1046 * Close wrapper for fifo's. 1047 * 1048 * Update the times on the inode then do device close. 1049 */ 1050 int 1051 lfsfifo_close(void *v) 1052 { 1053 struct vop_close_args /* { 1054 struct vnode *a_vp; 1055 int a_fflag; 1056 kauth_cred_ a_cred; 1057 } */ *ap = v; 1058 struct vnode *vp; 1059 struct inode *ip; 1060 1061 vp = ap->a_vp; 1062 ip = VTOI(vp); 1063 if (ap->a_vp->v_usecount > 1) { 1064 LFS_ITIMES(ip, NULL, NULL, NULL); 1065 } 1066 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1067 } 1068 1069 /* 1070 * Reclaim an inode so that it can be used for other purposes. 1071 */ 1072 1073 int 1074 lfs_reclaim(void *v) 1075 { 1076 struct vop_reclaim_args /* { 1077 struct vnode *a_vp; 1078 } */ *ap = v; 1079 struct vnode *vp = ap->a_vp; 1080 struct inode *ip = VTOI(vp); 1081 struct lfs *fs = ip->i_lfs; 1082 int error; 1083 1084 /* 1085 * The inode must be freed and updated before being removed 1086 * from its hash chain. Other threads trying to gain a hold 1087 * on the inode will be stalled because it is locked (VI_XLOCK). 1088 */ 1089 if (ip->i_nlink <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) 1090 lfs_vfree(vp, ip->i_number, ip->i_omode); 1091 1092 mutex_enter(&lfs_lock); 1093 LFS_CLR_UINO(ip, IN_ALLMOD); 1094 mutex_exit(&lfs_lock); 1095 if ((error = ufs_reclaim(vp))) 1096 return (error); 1097 1098 /* 1099 * Take us off the paging and/or dirop queues if we were on them. 1100 * We shouldn't be on them. 1101 */ 1102 mutex_enter(&lfs_lock); 1103 if (ip->i_flags & IN_PAGING) { 1104 log(LOG_WARNING, "%s: reclaimed vnode is IN_PAGING\n", 1105 fs->lfs_fsmnt); 1106 ip->i_flags &= ~IN_PAGING; 1107 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1108 } 1109 if (vp->v_uflag & VU_DIROP) { 1110 panic("reclaimed vnode is VU_DIROP"); 1111 vp->v_uflag &= ~VU_DIROP; 1112 TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain); 1113 } 1114 mutex_exit(&lfs_lock); 1115 1116 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1117 lfs_deregister_all(vp); 1118 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1119 ip->inode_ext.lfs = NULL; 1120 genfs_node_destroy(vp); 1121 pool_put(&lfs_inode_pool, vp->v_data); 1122 vp->v_data = NULL; 1123 return (0); 1124 } 1125 1126 /* 1127 * Read a block from a storage device. 1128 * In order to avoid reading blocks that are in the process of being 1129 * written by the cleaner---and hence are not mutexed by the normal 1130 * buffer cache / page cache mechanisms---check for collisions before 1131 * reading. 1132 * 1133 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1134 * the active cleaner test. 1135 * 1136 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1137 */ 1138 int 1139 lfs_strategy(void *v) 1140 { 1141 struct vop_strategy_args /* { 1142 struct vnode *a_vp; 1143 struct buf *a_bp; 1144 } */ *ap = v; 1145 struct buf *bp; 1146 struct lfs *fs; 1147 struct vnode *vp; 1148 struct inode *ip; 1149 daddr_t tbn; 1150 int i, sn, error, slept; 1151 1152 bp = ap->a_bp; 1153 vp = ap->a_vp; 1154 ip = VTOI(vp); 1155 fs = ip->i_lfs; 1156 1157 /* lfs uses its strategy routine only for read */ 1158 KASSERT(bp->b_flags & B_READ); 1159 1160 if (vp->v_type == VBLK || vp->v_type == VCHR) 1161 panic("lfs_strategy: spec"); 1162 KASSERT(bp->b_bcount != 0); 1163 if (bp->b_blkno == bp->b_lblkno) { 1164 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1165 NULL); 1166 if (error) { 1167 bp->b_error = error; 1168 bp->b_resid = bp->b_bcount; 1169 biodone(bp); 1170 return (error); 1171 } 1172 if ((long)bp->b_blkno == -1) /* no valid data */ 1173 clrbuf(bp); 1174 } 1175 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1176 bp->b_resid = bp->b_bcount; 1177 biodone(bp); 1178 return (0); 1179 } 1180 1181 slept = 1; 1182 mutex_enter(&lfs_lock); 1183 while (slept && fs->lfs_seglock) { 1184 mutex_exit(&lfs_lock); 1185 /* 1186 * Look through list of intervals. 1187 * There will only be intervals to look through 1188 * if the cleaner holds the seglock. 1189 * Since the cleaner is synchronous, we can trust 1190 * the list of intervals to be current. 1191 */ 1192 tbn = dbtofsb(fs, bp->b_blkno); 1193 sn = dtosn(fs, tbn); 1194 slept = 0; 1195 for (i = 0; i < fs->lfs_cleanind; i++) { 1196 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1197 tbn >= fs->lfs_cleanint[i]) { 1198 DLOG((DLOG_CLEAN, 1199 "lfs_strategy: ino %d lbn %" PRId64 1200 " ind %d sn %d fsb %" PRIx32 1201 " given sn %d fsb %" PRIx64 "\n", 1202 ip->i_number, bp->b_lblkno, i, 1203 dtosn(fs, fs->lfs_cleanint[i]), 1204 fs->lfs_cleanint[i], sn, tbn)); 1205 DLOG((DLOG_CLEAN, 1206 "lfs_strategy: sleeping on ino %d lbn %" 1207 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1208 mutex_enter(&lfs_lock); 1209 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1210 /* Cleaner can't wait for itself */ 1211 mtsleep(&fs->lfs_iocount, 1212 (PRIBIO + 1) | PNORELOCK, 1213 "clean2", 0, 1214 &lfs_lock); 1215 slept = 1; 1216 break; 1217 } else if (fs->lfs_seglock) { 1218 mtsleep(&fs->lfs_seglock, 1219 (PRIBIO + 1) | PNORELOCK, 1220 "clean1", 0, 1221 &lfs_lock); 1222 slept = 1; 1223 break; 1224 } 1225 mutex_exit(&lfs_lock); 1226 } 1227 } 1228 mutex_enter(&lfs_lock); 1229 } 1230 mutex_exit(&lfs_lock); 1231 1232 vp = ip->i_devvp; 1233 VOP_STRATEGY(vp, bp); 1234 return (0); 1235 } 1236 1237 void 1238 lfs_flush_dirops(struct lfs *fs) 1239 { 1240 struct inode *ip, *nip; 1241 struct vnode *vp; 1242 extern int lfs_dostats; 1243 struct segment *sp; 1244 1245 ASSERT_MAYBE_SEGLOCK(fs); 1246 KASSERT(fs->lfs_nadirop == 0); 1247 1248 if (fs->lfs_ronly) 1249 return; 1250 1251 mutex_enter(&lfs_lock); 1252 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1253 mutex_exit(&lfs_lock); 1254 return; 1255 } else 1256 mutex_exit(&lfs_lock); 1257 1258 if (lfs_dostats) 1259 ++lfs_stats.flush_invoked; 1260 1261 /* 1262 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1263 * Technically this is a checkpoint (the on-disk state is valid) 1264 * even though we are leaving out all the file data. 1265 */ 1266 lfs_imtime(fs); 1267 lfs_seglock(fs, SEGM_CKP); 1268 sp = fs->lfs_sp; 1269 1270 /* 1271 * lfs_writevnodes, optimized to get dirops out of the way. 1272 * Only write dirops, and don't flush files' pages, only 1273 * blocks from the directories. 1274 * 1275 * We don't need to vref these files because they are 1276 * dirops and so hold an extra reference until the 1277 * segunlock clears them of that status. 1278 * 1279 * We don't need to check for IN_ADIROP because we know that 1280 * no dirops are active. 1281 * 1282 */ 1283 mutex_enter(&lfs_lock); 1284 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1285 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1286 mutex_exit(&lfs_lock); 1287 vp = ITOV(ip); 1288 1289 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1290 1291 /* 1292 * All writes to directories come from dirops; all 1293 * writes to files' direct blocks go through the page 1294 * cache, which we're not touching. Reads to files 1295 * and/or directories will not be affected by writing 1296 * directory blocks inodes and file inodes. So we don't 1297 * really need to lock. If we don't lock, though, 1298 * make sure that we don't clear IN_MODIFIED 1299 * unnecessarily. 1300 */ 1301 if (vp->v_iflag & VI_XLOCK) { 1302 mutex_enter(&lfs_lock); 1303 continue; 1304 } 1305 /* XXX see below 1306 * waslocked = VOP_ISLOCKED(vp); 1307 */ 1308 if (vp->v_type != VREG && 1309 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1310 lfs_writefile(fs, sp, vp); 1311 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1312 !(ip->i_flag & IN_ALLMOD)) { 1313 mutex_enter(&lfs_lock); 1314 LFS_SET_UINO(ip, IN_MODIFIED); 1315 mutex_exit(&lfs_lock); 1316 } 1317 } 1318 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1319 (void) lfs_writeinode(fs, sp, ip); 1320 mutex_enter(&lfs_lock); 1321 /* 1322 * XXX 1323 * LK_EXCLOTHER is dead -- what is intended here? 1324 * if (waslocked == LK_EXCLOTHER) 1325 * LFS_SET_UINO(ip, IN_MODIFIED); 1326 */ 1327 } 1328 mutex_exit(&lfs_lock); 1329 /* We've written all the dirops there are */ 1330 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1331 lfs_finalize_fs_seguse(fs); 1332 (void) lfs_writeseg(fs, sp); 1333 lfs_segunlock(fs); 1334 } 1335 1336 /* 1337 * Flush all vnodes for which the pagedaemon has requested pageouts. 1338 * Skip over any files that are marked VU_DIROP (since lfs_flush_dirop() 1339 * has just run, this would be an error). If we have to skip a vnode 1340 * for any reason, just skip it; if we have to wait for the cleaner, 1341 * abort. The writer daemon will call us again later. 1342 */ 1343 void 1344 lfs_flush_pchain(struct lfs *fs) 1345 { 1346 struct inode *ip, *nip; 1347 struct vnode *vp; 1348 extern int lfs_dostats; 1349 struct segment *sp; 1350 int error; 1351 1352 ASSERT_NO_SEGLOCK(fs); 1353 1354 if (fs->lfs_ronly) 1355 return; 1356 1357 mutex_enter(&lfs_lock); 1358 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1359 mutex_exit(&lfs_lock); 1360 return; 1361 } else 1362 mutex_exit(&lfs_lock); 1363 1364 /* Get dirops out of the way */ 1365 lfs_flush_dirops(fs); 1366 1367 if (lfs_dostats) 1368 ++lfs_stats.flush_invoked; 1369 1370 /* 1371 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1372 */ 1373 lfs_imtime(fs); 1374 lfs_seglock(fs, 0); 1375 sp = fs->lfs_sp; 1376 1377 /* 1378 * lfs_writevnodes, optimized to clear pageout requests. 1379 * Only write non-dirop files that are in the pageout queue. 1380 * We're very conservative about what we write; we want to be 1381 * fast and async. 1382 */ 1383 mutex_enter(&lfs_lock); 1384 top: 1385 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1386 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1387 vp = ITOV(ip); 1388 1389 if (!(ip->i_flags & IN_PAGING)) 1390 goto top; 1391 1392 mutex_enter(vp->v_interlock); 1393 if ((vp->v_iflag & VI_XLOCK) || (vp->v_uflag & VU_DIROP) != 0) { 1394 mutex_exit(vp->v_interlock); 1395 continue; 1396 } 1397 if (vp->v_type != VREG) { 1398 mutex_exit(vp->v_interlock); 1399 continue; 1400 } 1401 if (lfs_vref(vp)) 1402 continue; 1403 mutex_exit(&lfs_lock); 1404 1405 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_RETRY) != 0) { 1406 lfs_vunref(vp); 1407 mutex_enter(&lfs_lock); 1408 continue; 1409 } 1410 1411 error = lfs_writefile(fs, sp, vp); 1412 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1413 !(ip->i_flag & IN_ALLMOD)) { 1414 mutex_enter(&lfs_lock); 1415 LFS_SET_UINO(ip, IN_MODIFIED); 1416 mutex_exit(&lfs_lock); 1417 } 1418 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1419 (void) lfs_writeinode(fs, sp, ip); 1420 1421 VOP_UNLOCK(vp); 1422 lfs_vunref(vp); 1423 1424 if (error == EAGAIN) { 1425 lfs_writeseg(fs, sp); 1426 mutex_enter(&lfs_lock); 1427 break; 1428 } 1429 mutex_enter(&lfs_lock); 1430 } 1431 mutex_exit(&lfs_lock); 1432 (void) lfs_writeseg(fs, sp); 1433 lfs_segunlock(fs); 1434 } 1435 1436 /* 1437 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1438 */ 1439 int 1440 lfs_fcntl(void *v) 1441 { 1442 struct vop_fcntl_args /* { 1443 struct vnode *a_vp; 1444 u_int a_command; 1445 void * a_data; 1446 int a_fflag; 1447 kauth_cred_t a_cred; 1448 } */ *ap = v; 1449 struct timeval tv; 1450 struct timeval *tvp; 1451 BLOCK_INFO *blkiov; 1452 CLEANERINFO *cip; 1453 SEGUSE *sup; 1454 int blkcnt, error, oclean; 1455 size_t fh_size; 1456 struct lfs_fcntl_markv blkvp; 1457 struct lwp *l; 1458 fsid_t *fsidp; 1459 struct lfs *fs; 1460 struct buf *bp; 1461 fhandle_t *fhp; 1462 daddr_t off; 1463 1464 /* Only respect LFS fcntls on fs root or Ifile */ 1465 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1466 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1467 return ufs_fcntl(v); 1468 } 1469 1470 /* Avoid locking a draining lock */ 1471 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1472 return ESHUTDOWN; 1473 } 1474 1475 /* LFS control and monitoring fcntls are available only to root */ 1476 l = curlwp; 1477 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1478 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1479 NULL)) != 0) 1480 return (error); 1481 1482 fs = VTOI(ap->a_vp)->i_lfs; 1483 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1484 1485 error = 0; 1486 switch ((int)ap->a_command) { 1487 case LFCNSEGWAITALL_COMPAT_50: 1488 case LFCNSEGWAITALL_COMPAT: 1489 fsidp = NULL; 1490 /* FALLSTHROUGH */ 1491 case LFCNSEGWAIT_COMPAT_50: 1492 case LFCNSEGWAIT_COMPAT: 1493 { 1494 struct timeval50 *tvp50 1495 = (struct timeval50 *)ap->a_data; 1496 timeval50_to_timeval(tvp50, &tv); 1497 tvp = &tv; 1498 } 1499 goto segwait_common; 1500 case LFCNSEGWAITALL: 1501 fsidp = NULL; 1502 /* FALLSTHROUGH */ 1503 case LFCNSEGWAIT: 1504 tvp = (struct timeval *)ap->a_data; 1505 segwait_common: 1506 mutex_enter(&lfs_lock); 1507 ++fs->lfs_sleepers; 1508 mutex_exit(&lfs_lock); 1509 1510 error = lfs_segwait(fsidp, tvp); 1511 1512 mutex_enter(&lfs_lock); 1513 if (--fs->lfs_sleepers == 0) 1514 wakeup(&fs->lfs_sleepers); 1515 mutex_exit(&lfs_lock); 1516 return error; 1517 1518 case LFCNBMAPV: 1519 case LFCNMARKV: 1520 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1521 1522 blkcnt = blkvp.blkcnt; 1523 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1524 return (EINVAL); 1525 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1526 if ((error = copyin(blkvp.blkiov, blkiov, 1527 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1528 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1529 return error; 1530 } 1531 1532 mutex_enter(&lfs_lock); 1533 ++fs->lfs_sleepers; 1534 mutex_exit(&lfs_lock); 1535 if (ap->a_command == LFCNBMAPV) 1536 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1537 else /* LFCNMARKV */ 1538 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1539 if (error == 0) 1540 error = copyout(blkiov, blkvp.blkiov, 1541 blkcnt * sizeof(BLOCK_INFO)); 1542 mutex_enter(&lfs_lock); 1543 if (--fs->lfs_sleepers == 0) 1544 wakeup(&fs->lfs_sleepers); 1545 mutex_exit(&lfs_lock); 1546 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1547 return error; 1548 1549 case LFCNRECLAIM: 1550 /* 1551 * Flush dirops and write Ifile, allowing empty segments 1552 * to be immediately reclaimed. 1553 */ 1554 lfs_writer_enter(fs, "pndirop"); 1555 off = fs->lfs_offset; 1556 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1557 lfs_flush_dirops(fs); 1558 LFS_CLEANERINFO(cip, fs, bp); 1559 oclean = cip->clean; 1560 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1561 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1562 fs->lfs_sp->seg_flags |= SEGM_PROT; 1563 lfs_segunlock(fs); 1564 lfs_writer_leave(fs); 1565 1566 #ifdef DEBUG 1567 LFS_CLEANERINFO(cip, fs, bp); 1568 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1569 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1570 fs->lfs_offset - off, cip->clean - oclean, 1571 fs->lfs_activesb)); 1572 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1573 #endif 1574 1575 return 0; 1576 1577 case LFCNIFILEFH_COMPAT: 1578 /* Return the filehandle of the Ifile */ 1579 if ((error = kauth_authorize_system(l->l_cred, 1580 KAUTH_SYSTEM_FILEHANDLE, 0, NULL, NULL, NULL)) != 0) 1581 return (error); 1582 fhp = (struct fhandle *)ap->a_data; 1583 fhp->fh_fsid = *fsidp; 1584 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1585 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1586 1587 case LFCNIFILEFH_COMPAT2: 1588 case LFCNIFILEFH: 1589 /* Return the filehandle of the Ifile */ 1590 fhp = (struct fhandle *)ap->a_data; 1591 fhp->fh_fsid = *fsidp; 1592 fh_size = sizeof(struct lfs_fhandle) - 1593 offsetof(fhandle_t, fh_fid); 1594 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1595 1596 case LFCNREWIND: 1597 /* Move lfs_offset to the lowest-numbered segment */ 1598 return lfs_rewind(fs, *(int *)ap->a_data); 1599 1600 case LFCNINVAL: 1601 /* Mark a segment SEGUSE_INVAL */ 1602 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1603 if (sup->su_nbytes > 0) { 1604 brelse(bp, 0); 1605 lfs_unset_inval_all(fs); 1606 return EBUSY; 1607 } 1608 sup->su_flags |= SEGUSE_INVAL; 1609 VOP_BWRITE(bp); 1610 return 0; 1611 1612 case LFCNRESIZE: 1613 /* Resize the filesystem */ 1614 return lfs_resize_fs(fs, *(int *)ap->a_data); 1615 1616 case LFCNWRAPSTOP: 1617 case LFCNWRAPSTOP_COMPAT: 1618 /* 1619 * Hold lfs_newseg at segment 0; if requested, sleep until 1620 * the filesystem wraps around. To support external agents 1621 * (dump, fsck-based regression test) that need to look at 1622 * a snapshot of the filesystem, without necessarily 1623 * requiring that all fs activity stops. 1624 */ 1625 if (fs->lfs_stoplwp == curlwp) 1626 return EALREADY; 1627 1628 mutex_enter(&lfs_lock); 1629 while (fs->lfs_stoplwp != NULL) 1630 cv_wait(&fs->lfs_stopcv, &lfs_lock); 1631 fs->lfs_stoplwp = curlwp; 1632 if (fs->lfs_nowrap == 0) 1633 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1634 ++fs->lfs_nowrap; 1635 if (*(int *)ap->a_data == 1 1636 || ap->a_command == LFCNWRAPSTOP_COMPAT) { 1637 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1638 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1639 "segwrap", 0, &lfs_lock); 1640 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1641 if (error) { 1642 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1643 } 1644 } 1645 mutex_exit(&lfs_lock); 1646 return 0; 1647 1648 case LFCNWRAPGO: 1649 case LFCNWRAPGO_COMPAT: 1650 /* 1651 * Having done its work, the agent wakes up the writer. 1652 * If the argument is 1, it sleeps until a new segment 1653 * is selected. 1654 */ 1655 mutex_enter(&lfs_lock); 1656 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1657 ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1658 *((int *)ap->a_data)); 1659 mutex_exit(&lfs_lock); 1660 return error; 1661 1662 case LFCNWRAPPASS: 1663 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1664 return EALREADY; 1665 mutex_enter(&lfs_lock); 1666 if (fs->lfs_stoplwp != curlwp) { 1667 mutex_exit(&lfs_lock); 1668 return EALREADY; 1669 } 1670 if (fs->lfs_nowrap == 0) { 1671 mutex_exit(&lfs_lock); 1672 return EBUSY; 1673 } 1674 fs->lfs_wrappass = 1; 1675 wakeup(&fs->lfs_wrappass); 1676 /* Wait for the log to wrap, if asked */ 1677 if (*(int *)ap->a_data) { 1678 mutex_enter(ap->a_vp->v_interlock); 1679 lfs_vref(ap->a_vp); 1680 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1681 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1682 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1683 "segwrap", 0, &lfs_lock); 1684 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1685 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1686 lfs_vunref(ap->a_vp); 1687 } 1688 mutex_exit(&lfs_lock); 1689 return error; 1690 1691 case LFCNWRAPSTATUS: 1692 mutex_enter(&lfs_lock); 1693 *(int *)ap->a_data = fs->lfs_wrapstatus; 1694 mutex_exit(&lfs_lock); 1695 return 0; 1696 1697 default: 1698 return ufs_fcntl(v); 1699 } 1700 return 0; 1701 } 1702 1703 int 1704 lfs_getpages(void *v) 1705 { 1706 struct vop_getpages_args /* { 1707 struct vnode *a_vp; 1708 voff_t a_offset; 1709 struct vm_page **a_m; 1710 int *a_count; 1711 int a_centeridx; 1712 vm_prot_t a_access_type; 1713 int a_advice; 1714 int a_flags; 1715 } */ *ap = v; 1716 1717 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1718 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1719 return EPERM; 1720 } 1721 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1722 mutex_enter(&lfs_lock); 1723 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1724 mutex_exit(&lfs_lock); 1725 } 1726 1727 /* 1728 * we're relying on the fact that genfs_getpages() always read in 1729 * entire filesystem blocks. 1730 */ 1731 return genfs_getpages(v); 1732 } 1733 1734 /* 1735 * Wait for a page to become unbusy, possibly printing diagnostic messages 1736 * as well. 1737 * 1738 * Called with vp->v_interlock held; return with it held. 1739 */ 1740 static void 1741 wait_for_page(struct vnode *vp, struct vm_page *pg, const char *label) 1742 { 1743 if ((pg->flags & PG_BUSY) == 0) 1744 return; /* Nothing to wait for! */ 1745 1746 #if defined(DEBUG) && defined(UVM_PAGE_TRKOWN) 1747 static struct vm_page *lastpg; 1748 1749 if (label != NULL && pg != lastpg) { 1750 if (pg->owner_tag) { 1751 printf("lfs_putpages[%d.%d]: %s: page %p owner %d.%d [%s]\n", 1752 curproc->p_pid, curlwp->l_lid, label, 1753 pg, pg->owner, pg->lowner, pg->owner_tag); 1754 } else { 1755 printf("lfs_putpages[%d.%d]: %s: page %p unowned?!\n", 1756 curproc->p_pid, curlwp->l_lid, label, pg); 1757 } 1758 } 1759 lastpg = pg; 1760 #endif 1761 1762 pg->flags |= PG_WANTED; 1763 UVM_UNLOCK_AND_WAIT(pg, vp->v_interlock, 0, "lfsput", 0); 1764 mutex_enter(vp->v_interlock); 1765 } 1766 1767 /* 1768 * This routine is called by lfs_putpages() when it can't complete the 1769 * write because a page is busy. This means that either (1) someone, 1770 * possibly the pagedaemon, is looking at this page, and will give it up 1771 * presently; or (2) we ourselves are holding the page busy in the 1772 * process of being written (either gathered or actually on its way to 1773 * disk). We don't need to give up the segment lock, but we might need 1774 * to call lfs_writeseg() to expedite the page's journey to disk. 1775 * 1776 * Called with vp->v_interlock held; return with it held. 1777 */ 1778 /* #define BUSYWAIT */ 1779 static void 1780 write_and_wait(struct lfs *fs, struct vnode *vp, struct vm_page *pg, 1781 int seglocked, const char *label) 1782 { 1783 #ifndef BUSYWAIT 1784 struct inode *ip = VTOI(vp); 1785 struct segment *sp = fs->lfs_sp; 1786 int count = 0; 1787 1788 if (pg == NULL) 1789 return; 1790 1791 while (pg->flags & PG_BUSY && 1792 pg->uobject == &vp->v_uobj) { 1793 mutex_exit(vp->v_interlock); 1794 if (sp->cbpp - sp->bpp > 1) { 1795 /* Write gathered pages */ 1796 lfs_updatemeta(sp); 1797 lfs_release_finfo(fs); 1798 (void) lfs_writeseg(fs, sp); 1799 1800 /* 1801 * Reinitialize FIP 1802 */ 1803 KASSERT(sp->vp == vp); 1804 lfs_acquire_finfo(fs, ip->i_number, 1805 ip->i_gen); 1806 } 1807 ++count; 1808 mutex_enter(vp->v_interlock); 1809 wait_for_page(vp, pg, label); 1810 } 1811 if (label != NULL && count > 1) 1812 printf("lfs_putpages[%d]: %s: %sn = %d\n", curproc->p_pid, 1813 label, (count > 0 ? "looping, " : ""), count); 1814 #else 1815 preempt(1); 1816 #endif 1817 } 1818 1819 /* 1820 * Make sure that for all pages in every block in the given range, 1821 * either all are dirty or all are clean. If any of the pages 1822 * we've seen so far are dirty, put the vnode on the paging chain, 1823 * and mark it IN_PAGING. 1824 * 1825 * If checkfirst != 0, don't check all the pages but return at the 1826 * first dirty page. 1827 */ 1828 static int 1829 check_dirty(struct lfs *fs, struct vnode *vp, 1830 off_t startoffset, off_t endoffset, off_t blkeof, 1831 int flags, int checkfirst, struct vm_page **pgp) 1832 { 1833 int by_list; 1834 struct vm_page *curpg = NULL; /* XXX: gcc */ 1835 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1836 off_t soff = 0; /* XXX: gcc */ 1837 voff_t off; 1838 int i; 1839 int nonexistent; 1840 int any_dirty; /* number of dirty pages */ 1841 int dirty; /* number of dirty pages in a block */ 1842 int tdirty; 1843 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1844 int pagedaemon = (curlwp == uvm.pagedaemon_lwp); 1845 1846 ASSERT_MAYBE_SEGLOCK(fs); 1847 top: 1848 by_list = (vp->v_uobj.uo_npages <= 1849 ((endoffset - startoffset) >> PAGE_SHIFT) * 1850 UVM_PAGE_TREE_PENALTY); 1851 any_dirty = 0; 1852 1853 if (by_list) { 1854 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1855 } else { 1856 soff = startoffset; 1857 } 1858 while (by_list || soff < MIN(blkeof, endoffset)) { 1859 if (by_list) { 1860 /* 1861 * Find the first page in a block. Skip 1862 * blocks outside our area of interest or beyond 1863 * the end of file. 1864 */ 1865 KASSERT(curpg == NULL 1866 || (curpg->flags & PG_MARKER) == 0); 1867 if (pages_per_block > 1) { 1868 while (curpg && 1869 ((curpg->offset & fs->lfs_bmask) || 1870 curpg->offset >= vp->v_size || 1871 curpg->offset >= endoffset)) { 1872 curpg = TAILQ_NEXT(curpg, listq.queue); 1873 KASSERT(curpg == NULL || 1874 (curpg->flags & PG_MARKER) == 0); 1875 } 1876 } 1877 if (curpg == NULL) 1878 break; 1879 soff = curpg->offset; 1880 } 1881 1882 /* 1883 * Mark all pages in extended range busy; find out if any 1884 * of them are dirty. 1885 */ 1886 nonexistent = dirty = 0; 1887 for (i = 0; i == 0 || i < pages_per_block; i++) { 1888 if (by_list && pages_per_block <= 1) { 1889 pgs[i] = pg = curpg; 1890 } else { 1891 off = soff + (i << PAGE_SHIFT); 1892 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1893 if (pg == NULL) { 1894 ++nonexistent; 1895 continue; 1896 } 1897 } 1898 KASSERT(pg != NULL); 1899 1900 /* 1901 * If we're holding the segment lock, we can deadlock 1902 * against a process that has our page and is waiting 1903 * for the cleaner, while the cleaner waits for the 1904 * segment lock. Just bail in that case. 1905 */ 1906 if ((pg->flags & PG_BUSY) && 1907 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1908 if (i > 0) 1909 uvm_page_unbusy(pgs, i); 1910 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1911 if (pgp) 1912 *pgp = pg; 1913 return -1; 1914 } 1915 1916 while (pg->flags & PG_BUSY) { 1917 wait_for_page(vp, pg, NULL); 1918 if (i > 0) 1919 uvm_page_unbusy(pgs, i); 1920 goto top; 1921 } 1922 pg->flags |= PG_BUSY; 1923 UVM_PAGE_OWN(pg, "lfs_putpages"); 1924 1925 pmap_page_protect(pg, VM_PROT_NONE); 1926 tdirty = (pmap_clear_modify(pg) || 1927 (pg->flags & PG_CLEAN) == 0); 1928 dirty += tdirty; 1929 } 1930 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1931 if (by_list) { 1932 curpg = TAILQ_NEXT(curpg, listq.queue); 1933 } else { 1934 soff += fs->lfs_bsize; 1935 } 1936 continue; 1937 } 1938 1939 any_dirty += dirty; 1940 KASSERT(nonexistent == 0); 1941 1942 /* 1943 * If any are dirty make all dirty; unbusy them, 1944 * but if we were asked to clean, wire them so that 1945 * the pagedaemon doesn't bother us about them while 1946 * they're on their way to disk. 1947 */ 1948 for (i = 0; i == 0 || i < pages_per_block; i++) { 1949 pg = pgs[i]; 1950 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1951 if (dirty) { 1952 pg->flags &= ~PG_CLEAN; 1953 if (flags & PGO_FREE) { 1954 /* 1955 * Wire the page so that 1956 * pdaemon doesn't see it again. 1957 */ 1958 mutex_enter(&uvm_pageqlock); 1959 uvm_pagewire(pg); 1960 mutex_exit(&uvm_pageqlock); 1961 1962 /* Suspended write flag */ 1963 pg->flags |= PG_DELWRI; 1964 } 1965 } 1966 if (pg->flags & PG_WANTED) 1967 wakeup(pg); 1968 pg->flags &= ~(PG_WANTED|PG_BUSY); 1969 UVM_PAGE_OWN(pg, NULL); 1970 } 1971 1972 if (checkfirst && any_dirty) 1973 break; 1974 1975 if (by_list) { 1976 curpg = TAILQ_NEXT(curpg, listq.queue); 1977 } else { 1978 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1979 } 1980 } 1981 1982 return any_dirty; 1983 } 1984 1985 /* 1986 * lfs_putpages functions like genfs_putpages except that 1987 * 1988 * (1) It needs to bounds-check the incoming requests to ensure that 1989 * they are block-aligned; if they are not, expand the range and 1990 * do the right thing in case, e.g., the requested range is clean 1991 * but the expanded range is dirty. 1992 * 1993 * (2) It needs to explicitly send blocks to be written when it is done. 1994 * If VOP_PUTPAGES is called without the seglock held, we simply take 1995 * the seglock and let lfs_segunlock wait for us. 1996 * XXX There might be a bad situation if we have to flush a vnode while 1997 * XXX lfs_markv is in operation. As of this writing we panic in this 1998 * XXX case. 1999 * 2000 * Assumptions: 2001 * 2002 * (1) The caller does not hold any pages in this vnode busy. If it does, 2003 * there is a danger that when we expand the page range and busy the 2004 * pages we will deadlock. 2005 * 2006 * (2) We are called with vp->v_interlock held; we must return with it 2007 * released. 2008 * 2009 * (3) We don't absolutely have to free pages right away, provided that 2010 * the request does not have PGO_SYNCIO. When the pagedaemon gives 2011 * us a request with PGO_FREE, we take the pages out of the paging 2012 * queue and wake up the writer, which will handle freeing them for us. 2013 * 2014 * We ensure that for any filesystem block, all pages for that 2015 * block are either resident or not, even if those pages are higher 2016 * than EOF; that means that we will be getting requests to free 2017 * "unused" pages above EOF all the time, and should ignore them. 2018 * 2019 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 2020 * into has been set up for us by lfs_writefile. If not, we will 2021 * have to handle allocating and/or freeing an finfo entry. 2022 * 2023 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 2024 */ 2025 2026 /* How many times to loop before we should start to worry */ 2027 #define TOOMANY 4 2028 2029 int 2030 lfs_putpages(void *v) 2031 { 2032 int error; 2033 struct vop_putpages_args /* { 2034 struct vnode *a_vp; 2035 voff_t a_offlo; 2036 voff_t a_offhi; 2037 int a_flags; 2038 } */ *ap = v; 2039 struct vnode *vp; 2040 struct inode *ip; 2041 struct lfs *fs; 2042 struct segment *sp; 2043 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 2044 off_t off, max_endoffset; 2045 bool seglocked, sync, pagedaemon; 2046 struct vm_page *pg, *busypg; 2047 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 2048 #ifdef DEBUG 2049 int debug_n_again, debug_n_dirtyclean; 2050 #endif 2051 2052 vp = ap->a_vp; 2053 ip = VTOI(vp); 2054 fs = ip->i_lfs; 2055 sync = (ap->a_flags & PGO_SYNCIO) != 0; 2056 pagedaemon = (curlwp == uvm.pagedaemon_lwp); 2057 2058 /* Putpages does nothing for metadata. */ 2059 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 2060 mutex_exit(vp->v_interlock); 2061 return 0; 2062 } 2063 2064 /* 2065 * If there are no pages, don't do anything. 2066 */ 2067 if (vp->v_uobj.uo_npages == 0) { 2068 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 2069 (vp->v_iflag & VI_ONWORKLST) && 2070 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 2071 vp->v_iflag &= ~VI_WRMAPDIRTY; 2072 vn_syncer_remove_from_worklist(vp); 2073 } 2074 mutex_exit(vp->v_interlock); 2075 2076 /* Remove us from paging queue, if we were on it */ 2077 mutex_enter(&lfs_lock); 2078 if (ip->i_flags & IN_PAGING) { 2079 ip->i_flags &= ~IN_PAGING; 2080 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2081 } 2082 mutex_exit(&lfs_lock); 2083 return 0; 2084 } 2085 2086 blkeof = blkroundup(fs, ip->i_size); 2087 2088 /* 2089 * Ignore requests to free pages past EOF but in the same block 2090 * as EOF, unless the request is synchronous. (If the request is 2091 * sync, it comes from lfs_truncate.) 2092 * XXXUBC Make these pages look "active" so the pagedaemon won't 2093 * XXXUBC bother us with them again. 2094 */ 2095 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 2096 origoffset = ap->a_offlo; 2097 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 2098 pg = uvm_pagelookup(&vp->v_uobj, off); 2099 KASSERT(pg != NULL); 2100 while (pg->flags & PG_BUSY) { 2101 pg->flags |= PG_WANTED; 2102 UVM_UNLOCK_AND_WAIT(pg, vp->v_interlock, 0, 2103 "lfsput2", 0); 2104 mutex_enter(vp->v_interlock); 2105 } 2106 mutex_enter(&uvm_pageqlock); 2107 uvm_pageactivate(pg); 2108 mutex_exit(&uvm_pageqlock); 2109 } 2110 ap->a_offlo = blkeof; 2111 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 2112 mutex_exit(vp->v_interlock); 2113 return 0; 2114 } 2115 } 2116 2117 /* 2118 * Extend page range to start and end at block boundaries. 2119 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 2120 */ 2121 origoffset = ap->a_offlo; 2122 origendoffset = ap->a_offhi; 2123 startoffset = origoffset & ~(fs->lfs_bmask); 2124 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 2125 << fs->lfs_bshift; 2126 2127 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2128 endoffset = max_endoffset; 2129 origendoffset = endoffset; 2130 } else { 2131 origendoffset = round_page(ap->a_offhi); 2132 endoffset = round_page(blkroundup(fs, origendoffset)); 2133 } 2134 2135 KASSERT(startoffset > 0 || endoffset >= startoffset); 2136 if (startoffset == endoffset) { 2137 /* Nothing to do, why were we called? */ 2138 mutex_exit(vp->v_interlock); 2139 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 2140 PRId64 "\n", startoffset)); 2141 return 0; 2142 } 2143 2144 ap->a_offlo = startoffset; 2145 ap->a_offhi = endoffset; 2146 2147 /* 2148 * If not cleaning, just send the pages through genfs_putpages 2149 * to be returned to the pool. 2150 */ 2151 if (!(ap->a_flags & PGO_CLEANIT)) 2152 return genfs_putpages(v); 2153 2154 /* Set PGO_BUSYFAIL to avoid deadlocks */ 2155 ap->a_flags |= PGO_BUSYFAIL; 2156 2157 /* 2158 * Likewise, if we are asked to clean but the pages are not 2159 * dirty, we can just free them using genfs_putpages. 2160 */ 2161 #ifdef DEBUG 2162 debug_n_dirtyclean = 0; 2163 #endif 2164 do { 2165 int r; 2166 2167 /* Count the number of dirty pages */ 2168 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2169 ap->a_flags, 1, NULL); 2170 if (r < 0) { 2171 /* Pages are busy with another process */ 2172 mutex_exit(vp->v_interlock); 2173 return EDEADLK; 2174 } 2175 if (r > 0) /* Some pages are dirty */ 2176 break; 2177 2178 /* 2179 * Sometimes pages are dirtied between the time that 2180 * we check and the time we try to clean them. 2181 * Instruct lfs_gop_write to return EDEADLK in this case 2182 * so we can write them properly. 2183 */ 2184 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2185 r = genfs_do_putpages(vp, startoffset, endoffset, 2186 ap->a_flags & ~PGO_SYNCIO, &busypg); 2187 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2188 if (r != EDEADLK) 2189 return r; 2190 2191 /* One of the pages was busy. Start over. */ 2192 mutex_enter(vp->v_interlock); 2193 wait_for_page(vp, busypg, "dirtyclean"); 2194 #ifdef DEBUG 2195 ++debug_n_dirtyclean; 2196 #endif 2197 } while(1); 2198 2199 #ifdef DEBUG 2200 if (debug_n_dirtyclean > TOOMANY) 2201 printf("lfs_putpages: dirtyclean: looping, n = %d\n", 2202 debug_n_dirtyclean); 2203 #endif 2204 2205 /* 2206 * Dirty and asked to clean. 2207 * 2208 * Pagedaemon can't actually write LFS pages; wake up 2209 * the writer to take care of that. The writer will 2210 * notice the pager inode queue and act on that. 2211 * 2212 * XXX We must drop the vp->interlock before taking the lfs_lock or we 2213 * get a nasty deadlock with lfs_flush_pchain(). 2214 */ 2215 if (pagedaemon) { 2216 mutex_exit(vp->v_interlock); 2217 mutex_enter(&lfs_lock); 2218 if (!(ip->i_flags & IN_PAGING)) { 2219 ip->i_flags |= IN_PAGING; 2220 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2221 } 2222 wakeup(&lfs_writer_daemon); 2223 mutex_exit(&lfs_lock); 2224 preempt(); 2225 return EWOULDBLOCK; 2226 } 2227 2228 /* 2229 * If this is a file created in a recent dirop, we can't flush its 2230 * inode until the dirop is complete. Drain dirops, then flush the 2231 * filesystem (taking care of any other pending dirops while we're 2232 * at it). 2233 */ 2234 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2235 (vp->v_uflag & VU_DIROP)) { 2236 int locked; 2237 2238 DLOG((DLOG_PAGE, "lfs_putpages: flushing VU_DIROP\n")); 2239 /* XXX VOP_ISLOCKED() may not be used for lock decisions. */ 2240 locked = (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 2241 mutex_exit(vp->v_interlock); 2242 lfs_writer_enter(fs, "ppdirop"); 2243 if (locked) 2244 VOP_UNLOCK(vp); /* XXX why? */ 2245 2246 mutex_enter(&lfs_lock); 2247 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2248 mutex_exit(&lfs_lock); 2249 2250 if (locked) 2251 VOP_LOCK(vp, LK_EXCLUSIVE); 2252 mutex_enter(vp->v_interlock); 2253 lfs_writer_leave(fs); 2254 2255 /* XXX the flush should have taken care of this one too! */ 2256 } 2257 2258 /* 2259 * This is it. We are going to write some pages. From here on 2260 * down it's all just mechanics. 2261 * 2262 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2263 */ 2264 ap->a_flags &= ~PGO_SYNCIO; 2265 2266 /* 2267 * If we've already got the seglock, flush the node and return. 2268 * The FIP has already been set up for us by lfs_writefile, 2269 * and FIP cleanup and lfs_updatemeta will also be done there, 2270 * unless genfs_putpages returns EDEADLK; then we must flush 2271 * what we have, and correct FIP and segment header accounting. 2272 */ 2273 get_seglock: 2274 /* 2275 * If we are not called with the segment locked, lock it. 2276 * Account for a new FIP in the segment header, and set sp->vp. 2277 * (This should duplicate the setup at the top of lfs_writefile().) 2278 */ 2279 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2280 if (!seglocked) { 2281 mutex_exit(vp->v_interlock); 2282 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2283 if (error != 0) 2284 return error; 2285 mutex_enter(vp->v_interlock); 2286 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2287 } 2288 sp = fs->lfs_sp; 2289 KASSERT(sp->vp == NULL); 2290 sp->vp = vp; 2291 2292 /* 2293 * Ensure that the partial segment is marked SS_DIROP if this 2294 * vnode is a DIROP. 2295 */ 2296 if (!seglocked && vp->v_uflag & VU_DIROP) 2297 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2298 2299 /* 2300 * Loop over genfs_putpages until all pages are gathered. 2301 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2302 * Whenever we lose the interlock we have to rerun check_dirty, as 2303 * well, since more pages might have been dirtied in our absence. 2304 */ 2305 #ifdef DEBUG 2306 debug_n_again = 0; 2307 #endif 2308 do { 2309 busypg = NULL; 2310 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2311 ap->a_flags, 0, &busypg) < 0) { 2312 mutex_exit(vp->v_interlock); 2313 2314 mutex_enter(vp->v_interlock); 2315 write_and_wait(fs, vp, busypg, seglocked, NULL); 2316 if (!seglocked) { 2317 mutex_exit(vp->v_interlock); 2318 lfs_release_finfo(fs); 2319 lfs_segunlock(fs); 2320 mutex_enter(vp->v_interlock); 2321 } 2322 sp->vp = NULL; 2323 goto get_seglock; 2324 } 2325 2326 busypg = NULL; 2327 error = genfs_do_putpages(vp, startoffset, endoffset, 2328 ap->a_flags, &busypg); 2329 2330 if (error == EDEADLK || error == EAGAIN) { 2331 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2332 " %d ino %d off %x (seg %d)\n", error, 2333 ip->i_number, fs->lfs_offset, 2334 dtosn(fs, fs->lfs_offset))); 2335 2336 mutex_enter(vp->v_interlock); 2337 write_and_wait(fs, vp, busypg, seglocked, "again"); 2338 } 2339 #ifdef DEBUG 2340 ++debug_n_again; 2341 #endif 2342 } while (error == EDEADLK); 2343 #ifdef DEBUG 2344 if (debug_n_again > TOOMANY) 2345 printf("lfs_putpages: again: looping, n = %d\n", debug_n_again); 2346 #endif 2347 2348 KASSERT(sp != NULL && sp->vp == vp); 2349 if (!seglocked) { 2350 sp->vp = NULL; 2351 2352 /* Write indirect blocks as well */ 2353 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2354 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2355 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2356 2357 KASSERT(sp->vp == NULL); 2358 sp->vp = vp; 2359 } 2360 2361 /* 2362 * Blocks are now gathered into a segment waiting to be written. 2363 * All that's left to do is update metadata, and write them. 2364 */ 2365 lfs_updatemeta(sp); 2366 KASSERT(sp->vp == vp); 2367 sp->vp = NULL; 2368 2369 /* 2370 * If we were called from lfs_writefile, we don't need to clean up 2371 * the FIP or unlock the segment lock. We're done. 2372 */ 2373 if (seglocked) 2374 return error; 2375 2376 /* Clean up FIP and send it to disk. */ 2377 lfs_release_finfo(fs); 2378 lfs_writeseg(fs, fs->lfs_sp); 2379 2380 /* 2381 * Remove us from paging queue if we wrote all our pages. 2382 */ 2383 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2384 mutex_enter(&lfs_lock); 2385 if (ip->i_flags & IN_PAGING) { 2386 ip->i_flags &= ~IN_PAGING; 2387 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2388 } 2389 mutex_exit(&lfs_lock); 2390 } 2391 2392 /* 2393 * XXX - with the malloc/copy writeseg, the pages are freed by now 2394 * even if we don't wait (e.g. if we hold a nested lock). This 2395 * will not be true if we stop using malloc/copy. 2396 */ 2397 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2398 lfs_segunlock(fs); 2399 2400 /* 2401 * Wait for v_numoutput to drop to zero. The seglock should 2402 * take care of this, but there is a slight possibility that 2403 * aiodoned might not have got around to our buffers yet. 2404 */ 2405 if (sync) { 2406 mutex_enter(vp->v_interlock); 2407 while (vp->v_numoutput > 0) { 2408 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2409 " num %d\n", ip->i_number, vp->v_numoutput)); 2410 cv_wait(&vp->v_cv, vp->v_interlock); 2411 } 2412 mutex_exit(vp->v_interlock); 2413 } 2414 return error; 2415 } 2416 2417 /* 2418 * Return the last logical file offset that should be written for this file 2419 * if we're doing a write that ends at "size". If writing, we need to know 2420 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2421 * to know about entire blocks. 2422 */ 2423 void 2424 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2425 { 2426 struct inode *ip = VTOI(vp); 2427 struct lfs *fs = ip->i_lfs; 2428 daddr_t olbn, nlbn; 2429 2430 olbn = lblkno(fs, ip->i_size); 2431 nlbn = lblkno(fs, size); 2432 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2433 *eobp = fragroundup(fs, size); 2434 } else { 2435 *eobp = blkroundup(fs, size); 2436 } 2437 } 2438 2439 #ifdef DEBUG 2440 void lfs_dump_vop(void *); 2441 2442 void 2443 lfs_dump_vop(void *v) 2444 { 2445 struct vop_putpages_args /* { 2446 struct vnode *a_vp; 2447 voff_t a_offlo; 2448 voff_t a_offhi; 2449 int a_flags; 2450 } */ *ap = v; 2451 2452 #ifdef DDB 2453 vfs_vnode_print(ap->a_vp, 0, printf); 2454 #endif 2455 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2456 } 2457 #endif 2458 2459 int 2460 lfs_mmap(void *v) 2461 { 2462 struct vop_mmap_args /* { 2463 const struct vnodeop_desc *a_desc; 2464 struct vnode *a_vp; 2465 vm_prot_t a_prot; 2466 kauth_cred_t a_cred; 2467 } */ *ap = v; 2468 2469 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2470 return EOPNOTSUPP; 2471 return ufs_mmap(v); 2472 } 2473
Indexes created Mon Sep 29 13:09:55 GMT 2025