1 /* 2 * Copyright (c) 2004,2013 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@backplane.com> 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * Copyright (c) 1989, 1993 35 * The Regents of the University of California. All rights reserved. 36 * (c) UNIX System Laboratories, Inc. 37 * All or some portions of this file are derived from material licensed 38 * to the University of California by American Telephone and Telegraph 39 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 40 * the permission of UNIX System Laboratories, Inc. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 */ 66 67 /* 68 * External virtual filesystem routines 69 */ 70 71 #include <sys/param.h> 72 #include <sys/systm.h> 73 #include <sys/kernel.h> 74 #include <sys/malloc.h> 75 #include <sys/mount.h> 76 #include <sys/proc.h> 77 #include <sys/vnode.h> 78 #include <sys/buf.h> 79 #include <sys/eventhandler.h> 80 #include <sys/kthread.h> 81 #include <sys/sysctl.h> 82 83 #include <machine/limits.h> 84 85 #include <sys/buf2.h> 86 #include <sys/thread2.h> 87 88 #include <vm/vm.h> 89 #include <vm/vm_object.h> 90 91 struct mountscan_info { 92 TAILQ_ENTRY(mountscan_info) msi_entry; 93 int msi_how; 94 struct mount *msi_node; 95 }; 96 97 struct vmntvnodescan_info { 98 TAILQ_ENTRY(vmntvnodescan_info) entry; 99 struct vnode *vp; 100 }; 101 102 struct vnlru_info { 103 int pass; 104 }; 105 106 static int vnlru_nowhere = 0; 107 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RD, 108 &vnlru_nowhere, 0, 109 "Number of times the vnlru process ran without success"); 110 111 112 static struct lwkt_token mntid_token; 113 static struct mount dummymount; 114 115 /* note: mountlist exported to pstat */ 116 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 117 static TAILQ_HEAD(,mountscan_info) mountscan_list; 118 static struct lwkt_token mountlist_token; 119 120 static TAILQ_HEAD(,bio_ops) bio_ops_list = TAILQ_HEAD_INITIALIZER(bio_ops_list); 121 122 /* 123 * Called from vfsinit() 124 */ 125 void 126 vfs_mount_init(void) 127 { 128 lwkt_token_init(&mountlist_token, "mntlist"); 129 lwkt_token_init(&mntid_token, "mntid"); 130 TAILQ_INIT(&mountscan_list); 131 mount_init(&dummymount); 132 dummymount.mnt_flag |= MNT_RDONLY; 133 dummymount.mnt_kern_flag |= MNTK_ALL_MPSAFE; 134 } 135 136 /* 137 * Support function called to remove a vnode from the mountlist and 138 * deal with side effects for scans in progress. 139 * 140 * Target mnt_token is held on call. 141 */ 142 static void 143 vremovevnodemnt(struct vnode *vp) 144 { 145 struct vmntvnodescan_info *info; 146 struct mount *mp = vp->v_mount; 147 148 TAILQ_FOREACH(info, &mp->mnt_vnodescan_list, entry) { 149 if (info->vp == vp) 150 info->vp = TAILQ_NEXT(vp, v_nmntvnodes); 151 } 152 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); 153 } 154 155 /* 156 * Allocate a new vnode and associate it with a tag, mount point, and 157 * operations vector. 158 * 159 * A VX locked and refd vnode is returned. The caller should setup the 160 * remaining fields and vx_put() or, if he wishes to leave a vref, 161 * vx_unlock() the vnode. 162 */ 163 int 164 getnewvnode(enum vtagtype tag, struct mount *mp, 165 struct vnode **vpp, int lktimeout, int lkflags) 166 { 167 struct vnode *vp; 168 169 KKASSERT(mp != NULL); 170 171 vp = allocvnode(lktimeout, lkflags); 172 vp->v_tag = tag; 173 vp->v_data = NULL; 174 175 /* 176 * By default the vnode is assigned the mount point's normal 177 * operations vector. 178 */ 179 vp->v_ops = &mp->mnt_vn_use_ops; 180 vp->v_pbuf_count = nswbuf_kva / NSWBUF_SPLIT; 181 182 /* 183 * Placing the vnode on the mount point's queue makes it visible. 184 * VNON prevents it from being messed with, however. 185 */ 186 insmntque(vp, mp); 187 188 /* 189 * A VX locked & refd vnode is returned. 190 */ 191 *vpp = vp; 192 return (0); 193 } 194 195 /* 196 * This function creates vnodes with special operations vectors. The 197 * mount point is optional. 198 * 199 * This routine is being phased out but is still used by vfs_conf to 200 * create vnodes for devices prior to the root mount (with mp == NULL). 201 */ 202 int 203 getspecialvnode(enum vtagtype tag, struct mount *mp, 204 struct vop_ops **ops, 205 struct vnode **vpp, int lktimeout, int lkflags) 206 { 207 struct vnode *vp; 208 209 vp = allocvnode(lktimeout, lkflags); 210 vp->v_tag = tag; 211 vp->v_data = NULL; 212 vp->v_ops = ops; 213 214 if (mp == NULL) 215 mp = &dummymount; 216 217 /* 218 * Placing the vnode on the mount point's queue makes it visible. 219 * VNON prevents it from being messed with, however. 220 */ 221 insmntque(vp, mp); 222 223 /* 224 * A VX locked & refd vnode is returned. 225 */ 226 *vpp = vp; 227 return (0); 228 } 229 230 /* 231 * Interlock against an unmount, return 0 on success, non-zero on failure. 232 * 233 * The passed flag may be 0 or LK_NOWAIT and is only used if an unmount 234 * is in-progress. 235 * 236 * If no unmount is in-progress LK_NOWAIT is ignored. No other flag bits 237 * are used. A shared locked will be obtained and the filesystem will not 238 * be unmountable until the lock is released. 239 */ 240 int 241 vfs_busy(struct mount *mp, int flags) 242 { 243 int lkflags; 244 245 atomic_add_int(&mp->mnt_refs, 1); 246 lwkt_gettoken(&mp->mnt_token); 247 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 248 if (flags & LK_NOWAIT) { 249 lwkt_reltoken(&mp->mnt_token); 250 atomic_add_int(&mp->mnt_refs, -1); 251 return (ENOENT); 252 } 253 /* XXX not MP safe */ 254 mp->mnt_kern_flag |= MNTK_MWAIT; 255 256 /* 257 * Since all busy locks are shared except the exclusive 258 * lock granted when unmounting, the only place that a 259 * wakeup needs to be done is at the release of the 260 * exclusive lock at the end of dounmount. 261 * 262 * WARNING! mp can potentially go away once we release 263 * our ref. 264 */ 265 tsleep((caddr_t)mp, 0, "vfs_busy", 0); 266 lwkt_reltoken(&mp->mnt_token); 267 atomic_add_int(&mp->mnt_refs, -1); 268 return (ENOENT); 269 } 270 lkflags = LK_SHARED; 271 if (lockmgr(&mp->mnt_lock, lkflags)) 272 panic("vfs_busy: unexpected lock failure"); 273 lwkt_reltoken(&mp->mnt_token); 274 return (0); 275 } 276 277 /* 278 * Free a busy filesystem. 279 * 280 * Once refs is decremented the mount point can potentially get ripped 281 * out from under us, but we want to clean up our refs before unlocking 282 * so do a hold/drop around the whole mess. 283 * 284 * This is not in the critical path (I hope). 285 */ 286 void 287 vfs_unbusy(struct mount *mp) 288 { 289 mount_hold(mp); 290 atomic_add_int(&mp->mnt_refs, -1); 291 lockmgr(&mp->mnt_lock, LK_RELEASE); 292 mount_drop(mp); 293 } 294 295 /* 296 * Lookup a filesystem type, and if found allocate and initialize 297 * a mount structure for it. 298 * 299 * Devname is usually updated by mount(8) after booting. 300 */ 301 int 302 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp) 303 { 304 struct vfsconf *vfsp; 305 struct mount *mp; 306 307 if (fstypename == NULL) 308 return (ENODEV); 309 310 vfsp = vfsconf_find_by_name(fstypename); 311 if (vfsp == NULL) 312 return (ENODEV); 313 mp = kmalloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO); 314 mount_init(mp); 315 lockinit(&mp->mnt_lock, "vfslock", VLKTIMEOUT, 0); 316 317 vfs_busy(mp, 0); 318 mp->mnt_vfc = vfsp; 319 mp->mnt_op = vfsp->vfc_vfsops; 320 mp->mnt_pbuf_count = nswbuf_kva / NSWBUF_SPLIT; 321 vfsp->vfc_refcount++; 322 mp->mnt_stat.f_type = vfsp->vfc_typenum; 323 mp->mnt_flag |= MNT_RDONLY; 324 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 325 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 326 copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 327 328 /* 329 * Pre-set MPSAFE flags for VFS_MOUNT() call. 330 */ 331 if (vfsp->vfc_flags & VFCF_MPSAFE) 332 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; 333 334 *mpp = mp; 335 336 return (0); 337 } 338 339 /* 340 * Basic mount structure initialization 341 */ 342 void 343 mount_init(struct mount *mp) 344 { 345 lockinit(&mp->mnt_lock, "vfslock", hz*5, 0); 346 lwkt_token_init(&mp->mnt_token, "permnt"); 347 348 TAILQ_INIT(&mp->mnt_vnodescan_list); 349 TAILQ_INIT(&mp->mnt_nvnodelist); 350 TAILQ_INIT(&mp->mnt_reservedvnlist); 351 TAILQ_INIT(&mp->mnt_jlist); 352 mp->mnt_nvnodelistsize = 0; 353 mp->mnt_flag = 0; 354 mp->mnt_hold = 1; /* hold for umount last drop */ 355 mp->mnt_iosize_max = MAXPHYS; 356 vn_syncer_thr_create(mp); 357 } 358 359 void 360 mount_hold(struct mount *mp) 361 { 362 atomic_add_int(&mp->mnt_hold, 1); 363 } 364 365 void 366 mount_drop(struct mount *mp) 367 { 368 if (atomic_fetchadd_int(&mp->mnt_hold, -1) == 1) { 369 KKASSERT(mp->mnt_refs == 0); 370 kfree(mp, M_MOUNT); 371 } 372 } 373 374 /* 375 * Lookup a mount point by filesystem identifier. 376 * 377 * If not NULL, the returned mp is held and the caller is expected to drop 378 * it via mount_drop(). 379 */ 380 struct mount * 381 vfs_getvfs(fsid_t *fsid) 382 { 383 struct mount *mp; 384 385 lwkt_gettoken_shared(&mountlist_token); 386 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 387 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 388 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 389 mount_hold(mp); 390 break; 391 } 392 } 393 lwkt_reltoken(&mountlist_token); 394 return (mp); 395 } 396 397 /* 398 * Get a new unique fsid. Try to make its val[0] unique, since this value 399 * will be used to create fake device numbers for stat(). Also try (but 400 * not so hard) make its val[0] unique mod 2^16, since some emulators only 401 * support 16-bit device numbers. We end up with unique val[0]'s for the 402 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 403 * 404 * Keep in mind that several mounts may be running in parallel. Starting 405 * the search one past where the previous search terminated is both a 406 * micro-optimization and a defense against returning the same fsid to 407 * different mounts. 408 */ 409 void 410 vfs_getnewfsid(struct mount *mp) 411 { 412 static u_int16_t mntid_base; 413 struct mount *mptmp; 414 fsid_t tfsid; 415 int mtype; 416 417 lwkt_gettoken(&mntid_token); 418 mtype = mp->mnt_vfc->vfc_typenum; 419 tfsid.val[1] = mtype; 420 mtype = (mtype & 0xFF) << 24; 421 for (;;) { 422 tfsid.val[0] = makeudev(255, 423 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 424 mntid_base++; 425 mptmp = vfs_getvfs(&tfsid); 426 if (mptmp == NULL) 427 break; 428 mount_drop(mptmp); 429 } 430 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 431 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 432 lwkt_reltoken(&mntid_token); 433 } 434 435 /* 436 * Set the FSID for a new mount point to the template. Adjust 437 * the FSID to avoid collisions. 438 */ 439 int 440 vfs_setfsid(struct mount *mp, fsid_t *template) 441 { 442 struct mount *mptmp; 443 int didmunge = 0; 444 445 bzero(&mp->mnt_stat.f_fsid, sizeof(mp->mnt_stat.f_fsid)); 446 447 lwkt_gettoken(&mntid_token); 448 for (;;) { 449 mptmp = vfs_getvfs(template); 450 if (mptmp == NULL) 451 break; 452 mount_drop(mptmp); 453 didmunge = 1; 454 ++template->val[1]; 455 } 456 mp->mnt_stat.f_fsid = *template; 457 lwkt_reltoken(&mntid_token); 458 459 return(didmunge); 460 } 461 462 /* 463 * This routine is called when we have too many vnodes. It attempts 464 * to free <count> vnodes and will potentially free vnodes that still 465 * have VM backing store (VM backing store is typically the cause 466 * of a vnode blowout so we want to do this). Therefore, this operation 467 * is not considered cheap. 468 * 469 * A number of conditions may prevent a vnode from being reclaimed. 470 * the buffer cache may have references on the vnode, a directory 471 * vnode may still have references due to the namei cache representing 472 * underlying files, or the vnode may be in active use. It is not 473 * desireable to reuse such vnodes. These conditions may cause the 474 * number of vnodes to reach some minimum value regardless of what 475 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 476 */ 477 478 /* 479 * Attempt to recycle vnodes in a context that is always safe to block. 480 * Calling vlrurecycle() from the bowels of file system code has some 481 * interesting deadlock problems. 482 */ 483 static struct thread *vnlruthread; 484 485 static void 486 vnlru_proc(void) 487 { 488 struct thread *td = curthread; 489 490 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td, 491 SHUTDOWN_PRI_FIRST); 492 493 for (;;) { 494 int ncachedandinactive; 495 496 kproc_suspend_loop(); 497 498 /* 499 * Try to free some vnodes if we have too many. Trigger based 500 * on potentially freeable vnodes but calculate the count 501 * based on total vnodes. 502 * 503 * (long) -> deal with 64 bit machines, intermediate overflow 504 */ 505 synchronizevnodecount(); 506 ncachedandinactive = countcachedandinactivevnodes(); 507 if (numvnodes >= maxvnodes * 9 / 10 && 508 ncachedandinactive >= maxvnodes * 5 / 10) { 509 int count = numvnodes - maxvnodes * 9 / 10; 510 511 if (count > (ncachedandinactive) / 100) 512 count = (ncachedandinactive) / 100; 513 if (count < 5) 514 count = 5; 515 freesomevnodes(count); 516 } 517 518 /* 519 * Do non-critical-path (more robust) cache cleaning, 520 * even if vnode counts are nominal, to try to avoid 521 * having to do it in the critical path. 522 */ 523 cache_hysteresis(0); 524 525 /* 526 * Nothing to do if most of our vnodes are already on 527 * the free list. 528 */ 529 synchronizevnodecount(); 530 ncachedandinactive = countcachedandinactivevnodes(); 531 if (numvnodes <= maxvnodes * 9 / 10 || 532 ncachedandinactive <= maxvnodes * 5 / 10) { 533 tsleep(vnlruthread, 0, "vlruwt", hz); 534 continue; 535 } 536 } 537 } 538 539 /* 540 * MOUNTLIST FUNCTIONS 541 */ 542 543 /* 544 * mountlist_insert (MP SAFE) 545 * 546 * Add a new mount point to the mount list. 547 */ 548 void 549 mountlist_insert(struct mount *mp, int how) 550 { 551 lwkt_gettoken(&mountlist_token); 552 if (how == MNTINS_FIRST) 553 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 554 else 555 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 556 lwkt_reltoken(&mountlist_token); 557 } 558 559 /* 560 * mountlist_interlock (MP SAFE) 561 * 562 * Execute the specified interlock function with the mountlist token 563 * held. The function will be called in a serialized fashion verses 564 * other functions called through this mechanism. 565 * 566 * The function is expected to be very short-lived. 567 */ 568 int 569 mountlist_interlock(int (*callback)(struct mount *), struct mount *mp) 570 { 571 int error; 572 573 lwkt_gettoken(&mountlist_token); 574 error = callback(mp); 575 lwkt_reltoken(&mountlist_token); 576 return (error); 577 } 578 579 /* 580 * mountlist_boot_getfirst (DURING BOOT ONLY) 581 * 582 * This function returns the first mount on the mountlist, which is 583 * expected to be the root mount. Since no interlocks are obtained 584 * this function is only safe to use during booting. 585 */ 586 587 struct mount * 588 mountlist_boot_getfirst(void) 589 { 590 return(TAILQ_FIRST(&mountlist)); 591 } 592 593 /* 594 * mountlist_remove (MP SAFE) 595 * 596 * Remove a node from the mountlist. If this node is the next scan node 597 * for any active mountlist scans, the active mountlist scan will be 598 * adjusted to skip the node, thus allowing removals during mountlist 599 * scans. 600 */ 601 void 602 mountlist_remove(struct mount *mp) 603 { 604 struct mountscan_info *msi; 605 606 lwkt_gettoken(&mountlist_token); 607 TAILQ_FOREACH(msi, &mountscan_list, msi_entry) { 608 if (msi->msi_node == mp) { 609 if (msi->msi_how & MNTSCAN_FORWARD) 610 msi->msi_node = TAILQ_NEXT(mp, mnt_list); 611 else 612 msi->msi_node = TAILQ_PREV(mp, mntlist, 613 mnt_list); 614 } 615 } 616 TAILQ_REMOVE(&mountlist, mp, mnt_list); 617 lwkt_reltoken(&mountlist_token); 618 } 619 620 /* 621 * mountlist_exists (MP SAFE) 622 * 623 * Checks if a node exists in the mountlist. 624 * This function is mainly used by VFS quota code to check if a 625 * cached nullfs struct mount pointer is still valid at use time 626 * 627 * FIXME: there is no warranty the mp passed to that function 628 * will be the same one used by VFS_ACCOUNT() later 629 */ 630 int 631 mountlist_exists(struct mount *mp) 632 { 633 int node_exists = 0; 634 struct mount* lmp; 635 636 lwkt_gettoken_shared(&mountlist_token); 637 TAILQ_FOREACH(lmp, &mountlist, mnt_list) { 638 if (lmp == mp) { 639 node_exists = 1; 640 break; 641 } 642 } 643 lwkt_reltoken(&mountlist_token); 644 645 return(node_exists); 646 } 647 648 /* 649 * mountlist_scan 650 * 651 * Safely scan the mount points on the mount list. Each mountpoint 652 * is held across the callback. The callback is responsible for 653 * acquiring any further tokens or locks. 654 * 655 * Unless otherwise specified each mount point will be busied prior to the 656 * callback and unbusied afterwords. The callback may safely remove any 657 * mount point without interfering with the scan. If the current callback 658 * mount is removed the scanner will not attempt to unbusy it. 659 * 660 * If a mount node cannot be busied it is silently skipped. 661 * 662 * The callback return value is aggregated and a total is returned. A return 663 * value of < 0 is not aggregated and will terminate the scan. 664 * 665 * MNTSCAN_FORWARD - the mountlist is scanned in the forward direction 666 * MNTSCAN_REVERSE - the mountlist is scanned in reverse 667 * MNTSCAN_NOBUSY - the scanner will make the callback without busying 668 * the mount node. 669 * 670 * NOTE: mountlist_token is not held across the callback. 671 */ 672 int 673 mountlist_scan(int (*callback)(struct mount *, void *), void *data, int how) 674 { 675 struct mountscan_info info; 676 struct mount *mp; 677 int count; 678 int res; 679 680 lwkt_gettoken(&mountlist_token); 681 info.msi_how = how; 682 info.msi_node = NULL; /* paranoia */ 683 TAILQ_INSERT_TAIL(&mountscan_list, &info, msi_entry); 684 lwkt_reltoken(&mountlist_token); 685 686 res = 0; 687 lwkt_gettoken_shared(&mountlist_token); 688 689 if (how & MNTSCAN_FORWARD) { 690 info.msi_node = TAILQ_FIRST(&mountlist); 691 while ((mp = info.msi_node) != NULL) { 692 mount_hold(mp); 693 if (how & MNTSCAN_NOBUSY) { 694 lwkt_reltoken(&mountlist_token); 695 count = callback(mp, data); 696 lwkt_gettoken_shared(&mountlist_token); 697 } else if (vfs_busy(mp, LK_NOWAIT) == 0) { 698 lwkt_reltoken(&mountlist_token); 699 count = callback(mp, data); 700 lwkt_gettoken_shared(&mountlist_token); 701 if (mp == info.msi_node) 702 vfs_unbusy(mp); 703 } else { 704 count = 0; 705 } 706 mount_drop(mp); 707 if (count < 0) 708 break; 709 res += count; 710 if (mp == info.msi_node) 711 info.msi_node = TAILQ_NEXT(mp, mnt_list); 712 } 713 } else if (how & MNTSCAN_REVERSE) { 714 info.msi_node = TAILQ_LAST(&mountlist, mntlist); 715 while ((mp = info.msi_node) != NULL) { 716 mount_hold(mp); 717 if (how & MNTSCAN_NOBUSY) { 718 lwkt_reltoken(&mountlist_token); 719 count = callback(mp, data); 720 lwkt_gettoken_shared(&mountlist_token); 721 } else if (vfs_busy(mp, LK_NOWAIT) == 0) { 722 lwkt_reltoken(&mountlist_token); 723 count = callback(mp, data); 724 lwkt_gettoken_shared(&mountlist_token); 725 if (mp == info.msi_node) 726 vfs_unbusy(mp); 727 } else { 728 count = 0; 729 } 730 mount_drop(mp); 731 if (count < 0) 732 break; 733 res += count; 734 if (mp == info.msi_node) 735 info.msi_node = TAILQ_PREV(mp, mntlist, 736 mnt_list); 737 } 738 } 739 lwkt_reltoken(&mountlist_token); 740 741 lwkt_gettoken(&mountlist_token); 742 TAILQ_REMOVE(&mountscan_list, &info, msi_entry); 743 lwkt_reltoken(&mountlist_token); 744 745 return(res); 746 } 747 748 /* 749 * MOUNT RELATED VNODE FUNCTIONS 750 */ 751 752 static struct kproc_desc vnlru_kp = { 753 "vnlru", 754 vnlru_proc, 755 &vnlruthread 756 }; 757 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp); 758 759 /* 760 * Move a vnode from one mount queue to another. 761 */ 762 void 763 insmntque(struct vnode *vp, struct mount *mp) 764 { 765 struct mount *omp; 766 767 /* 768 * Delete from old mount point vnode list, if on one. 769 */ 770 if ((omp = vp->v_mount) != NULL) { 771 lwkt_gettoken(&omp->mnt_token); 772 KKASSERT(omp == vp->v_mount); 773 KASSERT(omp->mnt_nvnodelistsize > 0, 774 ("bad mount point vnode list size")); 775 vremovevnodemnt(vp); 776 omp->mnt_nvnodelistsize--; 777 lwkt_reltoken(&omp->mnt_token); 778 } 779 780 /* 781 * Insert into list of vnodes for the new mount point, if available. 782 * The 'end' of the LRU list is the vnode prior to mp->mnt_syncer. 783 */ 784 if (mp == NULL) { 785 vp->v_mount = NULL; 786 return; 787 } 788 lwkt_gettoken(&mp->mnt_token); 789 vp->v_mount = mp; 790 if (mp->mnt_syncer) { 791 TAILQ_INSERT_BEFORE(mp->mnt_syncer, vp, v_nmntvnodes); 792 } else { 793 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 794 } 795 mp->mnt_nvnodelistsize++; 796 lwkt_reltoken(&mp->mnt_token); 797 } 798 799 800 /* 801 * Scan the vnodes under a mount point and issue appropriate callbacks. 802 * 803 * The fastfunc() callback is called with just the mountlist token held 804 * (no vnode lock). It may not block and the vnode may be undergoing 805 * modifications while the caller is processing it. The vnode will 806 * not be entirely destroyed, however, due to the fact that the mountlist 807 * token is held. A return value < 0 skips to the next vnode without calling 808 * the slowfunc(), a return value > 0 terminates the loop. 809 * 810 * WARNING! The fastfunc() should not indirect through vp->v_object, the vp 811 * data structure is unstable when called from fastfunc(). 812 * 813 * The slowfunc() callback is called after the vnode has been successfully 814 * locked based on passed flags. The vnode is skipped if it gets rearranged 815 * or destroyed while blocking on the lock. A non-zero return value from 816 * the slow function terminates the loop. The slow function is allowed to 817 * arbitrarily block. The scanning code guarentees consistency of operation 818 * even if the slow function deletes or moves the node, or blocks and some 819 * other thread deletes or moves the node. 820 */ 821 int 822 vmntvnodescan( 823 struct mount *mp, 824 int flags, 825 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data), 826 int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data), 827 void *data 828 ) { 829 struct vmntvnodescan_info info; 830 struct vnode *vp; 831 int r = 0; 832 int maxcount = mp->mnt_nvnodelistsize * 2; 833 int stopcount = 0; 834 int count = 0; 835 836 lwkt_gettoken(&mp->mnt_token); 837 838 /* 839 * If asked to do one pass stop after iterating available vnodes. 840 * Under heavy loads new vnodes can be added while we are scanning, 841 * so this isn't perfect. Create a slop factor of 2x. 842 */ 843 if (flags & VMSC_ONEPASS) 844 stopcount = mp->mnt_nvnodelistsize; 845 846 info.vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 847 TAILQ_INSERT_TAIL(&mp->mnt_vnodescan_list, &info, entry); 848 849 while ((vp = info.vp) != NULL) { 850 if (--maxcount == 0) { 851 kprintf("Warning: excessive fssync iteration\n"); 852 maxcount = mp->mnt_nvnodelistsize * 2; 853 } 854 855 /* 856 * Skip if visible but not ready, or special (e.g. 857 * mp->mnt_syncer) 858 */ 859 if (vp->v_type == VNON) 860 goto next; 861 KKASSERT(vp->v_mount == mp); 862 863 /* 864 * Quick test. A negative return continues the loop without 865 * calling the slow test. 0 continues onto the slow test. 866 * A positive number aborts the loop. 867 */ 868 if (fastfunc) { 869 if ((r = fastfunc(mp, vp, data)) < 0) { 870 r = 0; 871 goto next; 872 } 873 if (r) 874 break; 875 } 876 877 /* 878 * Get a vxlock on the vnode, retry if it has moved or isn't 879 * in the mountlist where we expect it. 880 */ 881 if (slowfunc) { 882 int error; 883 884 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) { 885 case VMSC_GETVP: 886 error = vget(vp, LK_EXCLUSIVE); 887 break; 888 case VMSC_GETVP|VMSC_NOWAIT: 889 error = vget(vp, LK_EXCLUSIVE|LK_NOWAIT); 890 break; 891 case VMSC_GETVX: 892 vx_get(vp); 893 error = 0; 894 break; 895 default: 896 error = 0; 897 break; 898 } 899 if (error) 900 goto next; 901 /* 902 * Do not call the slow function if the vnode is 903 * invalid or if it was ripped out from under us 904 * while we (potentially) blocked. 905 */ 906 if (info.vp == vp && vp->v_type != VNON) 907 r = slowfunc(mp, vp, data); 908 909 /* 910 * Cleanup 911 */ 912 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) { 913 case VMSC_GETVP: 914 case VMSC_GETVP|VMSC_NOWAIT: 915 vput(vp); 916 break; 917 case VMSC_GETVX: 918 vx_put(vp); 919 break; 920 default: 921 break; 922 } 923 if (r != 0) 924 break; 925 } 926 927 next: 928 /* 929 * Yield after some processing. Depending on the number 930 * of vnodes, we might wind up running for a long time. 931 * Because threads are not preemptable, time critical 932 * userland processes might starve. Give them a chance 933 * now and then. 934 */ 935 if (++count == 10000) { 936 /* 937 * We really want to yield a bit, so we simply 938 * sleep a tick 939 */ 940 tsleep(mp, 0, "vnodescn", 1); 941 count = 0; 942 } 943 944 /* 945 * If doing one pass this decrements to zero. If it starts 946 * at zero it is effectively unlimited for the purposes of 947 * this loop. 948 */ 949 if (--stopcount == 0) 950 break; 951 952 /* 953 * Iterate. If the vnode was ripped out from under us 954 * info.vp will already point to the next vnode, otherwise 955 * we have to obtain the next valid vnode ourselves. 956 */ 957 if (info.vp == vp) 958 info.vp = TAILQ_NEXT(vp, v_nmntvnodes); 959 } 960 961 TAILQ_REMOVE(&mp->mnt_vnodescan_list, &info, entry); 962 lwkt_reltoken(&mp->mnt_token); 963 return(r); 964 } 965 966 /* 967 * Remove any vnodes in the vnode table belonging to mount point mp. 968 * 969 * If FORCECLOSE is not specified, there should not be any active ones, 970 * return error if any are found (nb: this is a user error, not a 971 * system error). If FORCECLOSE is specified, detach any active vnodes 972 * that are found. 973 * 974 * If WRITECLOSE is set, only flush out regular file vnodes open for 975 * writing. 976 * 977 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped. 978 * 979 * `rootrefs' specifies the base reference count for the root vnode 980 * of this filesystem. The root vnode is considered busy if its 981 * v_refcnt exceeds this value. On a successful return, vflush() 982 * will call vrele() on the root vnode exactly rootrefs times. 983 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 984 * be zero. 985 */ 986 #ifdef DIAGNOSTIC 987 static int busyprt = 0; /* print out busy vnodes */ 988 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 989 #endif 990 991 static int vflush_scan(struct mount *mp, struct vnode *vp, void *data); 992 993 struct vflush_info { 994 int flags; 995 int busy; 996 thread_t td; 997 }; 998 999 int 1000 vflush(struct mount *mp, int rootrefs, int flags) 1001 { 1002 struct thread *td = curthread; /* XXX */ 1003 struct vnode *rootvp = NULL; 1004 int error; 1005 struct vflush_info vflush_info; 1006 1007 if (rootrefs > 0) { 1008 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 1009 ("vflush: bad args")); 1010 /* 1011 * Get the filesystem root vnode. We can vput() it 1012 * immediately, since with rootrefs > 0, it won't go away. 1013 */ 1014 if ((error = VFS_ROOT(mp, &rootvp)) != 0) { 1015 if ((flags & FORCECLOSE) == 0) 1016 return (error); 1017 rootrefs = 0; 1018 /* continue anyway */ 1019 } 1020 if (rootrefs) 1021 vput(rootvp); 1022 } 1023 1024 vflush_info.busy = 0; 1025 vflush_info.flags = flags; 1026 vflush_info.td = td; 1027 vmntvnodescan(mp, VMSC_GETVX, NULL, vflush_scan, &vflush_info); 1028 1029 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 1030 /* 1031 * If just the root vnode is busy, and if its refcount 1032 * is equal to `rootrefs', then go ahead and kill it. 1033 */ 1034 KASSERT(vflush_info.busy > 0, ("vflush: not busy")); 1035 KASSERT(VREFCNT(rootvp) >= rootrefs, ("vflush: rootrefs")); 1036 if (vflush_info.busy == 1 && VREFCNT(rootvp) == rootrefs) { 1037 vx_lock(rootvp); 1038 vgone_vxlocked(rootvp); 1039 vx_unlock(rootvp); 1040 vflush_info.busy = 0; 1041 } 1042 } 1043 if (vflush_info.busy) 1044 return (EBUSY); 1045 for (; rootrefs > 0; rootrefs--) 1046 vrele(rootvp); 1047 return (0); 1048 } 1049 1050 /* 1051 * The scan callback is made with an VX locked vnode. 1052 */ 1053 static int 1054 vflush_scan(struct mount *mp, struct vnode *vp, void *data) 1055 { 1056 struct vflush_info *info = data; 1057 struct vattr vattr; 1058 int flags = info->flags; 1059 1060 /* 1061 * Generally speaking try to deactivate on 0 refs (catch-all) 1062 */ 1063 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 1064 1065 /* 1066 * Skip over a vnodes marked VSYSTEM. 1067 */ 1068 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1069 return(0); 1070 } 1071 1072 /* 1073 * Do not force-close VCHR or VBLK vnodes 1074 */ 1075 if (vp->v_type == VCHR || vp->v_type == VBLK) 1076 flags &= ~(WRITECLOSE|FORCECLOSE); 1077 1078 /* 1079 * If WRITECLOSE is set, flush out unlinked but still open 1080 * files (even if open only for reading) and regular file 1081 * vnodes open for writing. 1082 */ 1083 if ((flags & WRITECLOSE) && 1084 (vp->v_type == VNON || 1085 (VOP_GETATTR(vp, &vattr) == 0 && 1086 vattr.va_nlink > 0)) && 1087 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1088 return(0); 1089 } 1090 1091 /* 1092 * If we are the only holder (refcnt of 1) or the vnode is in 1093 * termination (refcnt < 0), we can vgone the vnode. 1094 */ 1095 if (VREFCNT(vp) <= 1) { 1096 vgone_vxlocked(vp); 1097 return(0); 1098 } 1099 1100 /* 1101 * If FORCECLOSE is set, forcibly destroy the vnode and then move 1102 * it to a dummymount structure so vop_*() functions don't deref 1103 * a NULL pointer. 1104 */ 1105 if (flags & FORCECLOSE) { 1106 vhold(vp); 1107 vgone_vxlocked(vp); 1108 if (vp->v_mount == NULL) 1109 insmntque(vp, &dummymount); 1110 vdrop(vp); 1111 return(0); 1112 } 1113 if (vp->v_type == VCHR || vp->v_type == VBLK) 1114 kprintf("vflush: Warning, cannot destroy busy device vnode\n"); 1115 #ifdef DIAGNOSTIC 1116 if (busyprt) 1117 vprint("vflush: busy vnode", vp); 1118 #endif 1119 ++info->busy; 1120 return(0); 1121 } 1122 1123 void 1124 add_bio_ops(struct bio_ops *ops) 1125 { 1126 TAILQ_INSERT_TAIL(&bio_ops_list, ops, entry); 1127 } 1128 1129 void 1130 rem_bio_ops(struct bio_ops *ops) 1131 { 1132 TAILQ_REMOVE(&bio_ops_list, ops, entry); 1133 } 1134 1135 /* 1136 * This calls the bio_ops io_sync function either for a mount point 1137 * or generally. 1138 * 1139 * WARNING: softdeps is weirdly coded and just isn't happy unless 1140 * io_sync is called with a NULL mount from the general syncing code. 1141 */ 1142 void 1143 bio_ops_sync(struct mount *mp) 1144 { 1145 struct bio_ops *ops; 1146 1147 if (mp) { 1148 if ((ops = mp->mnt_bioops) != NULL) 1149 ops->io_sync(mp); 1150 } else { 1151 TAILQ_FOREACH(ops, &bio_ops_list, entry) { 1152 ops->io_sync(NULL); 1153 } 1154 } 1155 } 1156 1157 /* 1158 * Lookup a mount point by nch 1159 */ 1160 struct mount * 1161 mount_get_by_nc(struct namecache *ncp) 1162 { 1163 struct mount *mp = NULL; 1164 1165 lwkt_gettoken_shared(&mountlist_token); 1166 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 1167 if (ncp == mp->mnt_ncmountpt.ncp) 1168 break; 1169 } 1170 lwkt_reltoken(&mountlist_token); 1171 1172 return (mp); 1173 } 1174 1175