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 ncached; 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 ncached = countcachedvnodes(1); 506 if (numvnodes >= maxvnodes * 9 / 10 && 507 ncached + inactivevnodes >= maxvnodes * 5 / 10) { 508 int count = numvnodes - maxvnodes * 9 / 10; 509 510 if (count > (ncached + inactivevnodes) / 100) 511 count = (ncached + inactivevnodes) / 100; 512 if (count < 5) 513 count = 5; 514 freesomevnodes(count); 515 } 516 517 /* 518 * Do non-critical-path (more robust) cache cleaning, 519 * even if vnode counts are nominal, to try to avoid 520 * having to do it in the critical path. 521 */ 522 cache_hysteresis(0); 523 524 /* 525 * Nothing to do if most of our vnodes are already on 526 * the free list. 527 */ 528 ncached = countcachedvnodes(1); 529 if (numvnodes <= maxvnodes * 9 / 10 || 530 ncached + inactivevnodes <= maxvnodes * 5 / 10) { 531 tsleep(vnlruthread, 0, "vlruwt", hz); 532 continue; 533 } 534 } 535 } 536 537 /* 538 * MOUNTLIST FUNCTIONS 539 */ 540 541 /* 542 * mountlist_insert (MP SAFE) 543 * 544 * Add a new mount point to the mount list. 545 */ 546 void 547 mountlist_insert(struct mount *mp, int how) 548 { 549 lwkt_gettoken(&mountlist_token); 550 if (how == MNTINS_FIRST) 551 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 552 else 553 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 554 lwkt_reltoken(&mountlist_token); 555 } 556 557 /* 558 * mountlist_interlock (MP SAFE) 559 * 560 * Execute the specified interlock function with the mountlist token 561 * held. The function will be called in a serialized fashion verses 562 * other functions called through this mechanism. 563 * 564 * The function is expected to be very short-lived. 565 */ 566 int 567 mountlist_interlock(int (*callback)(struct mount *), struct mount *mp) 568 { 569 int error; 570 571 lwkt_gettoken(&mountlist_token); 572 error = callback(mp); 573 lwkt_reltoken(&mountlist_token); 574 return (error); 575 } 576 577 /* 578 * mountlist_boot_getfirst (DURING BOOT ONLY) 579 * 580 * This function returns the first mount on the mountlist, which is 581 * expected to be the root mount. Since no interlocks are obtained 582 * this function is only safe to use during booting. 583 */ 584 585 struct mount * 586 mountlist_boot_getfirst(void) 587 { 588 return(TAILQ_FIRST(&mountlist)); 589 } 590 591 /* 592 * mountlist_remove (MP SAFE) 593 * 594 * Remove a node from the mountlist. If this node is the next scan node 595 * for any active mountlist scans, the active mountlist scan will be 596 * adjusted to skip the node, thus allowing removals during mountlist 597 * scans. 598 */ 599 void 600 mountlist_remove(struct mount *mp) 601 { 602 struct mountscan_info *msi; 603 604 lwkt_gettoken(&mountlist_token); 605 TAILQ_FOREACH(msi, &mountscan_list, msi_entry) { 606 if (msi->msi_node == mp) { 607 if (msi->msi_how & MNTSCAN_FORWARD) 608 msi->msi_node = TAILQ_NEXT(mp, mnt_list); 609 else 610 msi->msi_node = TAILQ_PREV(mp, mntlist, 611 mnt_list); 612 } 613 } 614 TAILQ_REMOVE(&mountlist, mp, mnt_list); 615 lwkt_reltoken(&mountlist_token); 616 } 617 618 /* 619 * mountlist_exists (MP SAFE) 620 * 621 * Checks if a node exists in the mountlist. 622 * This function is mainly used by VFS quota code to check if a 623 * cached nullfs struct mount pointer is still valid at use time 624 * 625 * FIXME: there is no warranty the mp passed to that function 626 * will be the same one used by VFS_ACCOUNT() later 627 */ 628 int 629 mountlist_exists(struct mount *mp) 630 { 631 int node_exists = 0; 632 struct mount* lmp; 633 634 lwkt_gettoken_shared(&mountlist_token); 635 TAILQ_FOREACH(lmp, &mountlist, mnt_list) { 636 if (lmp == mp) { 637 node_exists = 1; 638 break; 639 } 640 } 641 lwkt_reltoken(&mountlist_token); 642 643 return(node_exists); 644 } 645 646 /* 647 * mountlist_scan 648 * 649 * Safely scan the mount points on the mount list. Each mountpoint 650 * is held across the callback. The callback is responsible for 651 * acquiring any further tokens or locks. 652 * 653 * Unless otherwise specified each mount point will be busied prior to the 654 * callback and unbusied afterwords. The callback may safely remove any 655 * mount point without interfering with the scan. If the current callback 656 * mount is removed the scanner will not attempt to unbusy it. 657 * 658 * If a mount node cannot be busied it is silently skipped. 659 * 660 * The callback return value is aggregated and a total is returned. A return 661 * value of < 0 is not aggregated and will terminate the scan. 662 * 663 * MNTSCAN_FORWARD - the mountlist is scanned in the forward direction 664 * MNTSCAN_REVERSE - the mountlist is scanned in reverse 665 * MNTSCAN_NOBUSY - the scanner will make the callback without busying 666 * the mount node. 667 * 668 * NOTE: mountlist_token is not held across the callback. 669 */ 670 int 671 mountlist_scan(int (*callback)(struct mount *, void *), void *data, int how) 672 { 673 struct mountscan_info info; 674 struct mount *mp; 675 int count; 676 int res; 677 678 lwkt_gettoken(&mountlist_token); 679 info.msi_how = how; 680 info.msi_node = NULL; /* paranoia */ 681 TAILQ_INSERT_TAIL(&mountscan_list, &info, msi_entry); 682 lwkt_reltoken(&mountlist_token); 683 684 res = 0; 685 lwkt_gettoken_shared(&mountlist_token); 686 687 if (how & MNTSCAN_FORWARD) { 688 info.msi_node = TAILQ_FIRST(&mountlist); 689 while ((mp = info.msi_node) != NULL) { 690 mount_hold(mp); 691 if (how & MNTSCAN_NOBUSY) { 692 lwkt_reltoken(&mountlist_token); 693 count = callback(mp, data); 694 lwkt_gettoken_shared(&mountlist_token); 695 } else if (vfs_busy(mp, LK_NOWAIT) == 0) { 696 lwkt_reltoken(&mountlist_token); 697 count = callback(mp, data); 698 lwkt_gettoken_shared(&mountlist_token); 699 if (mp == info.msi_node) 700 vfs_unbusy(mp); 701 } else { 702 count = 0; 703 } 704 mount_drop(mp); 705 if (count < 0) 706 break; 707 res += count; 708 if (mp == info.msi_node) 709 info.msi_node = TAILQ_NEXT(mp, mnt_list); 710 } 711 } else if (how & MNTSCAN_REVERSE) { 712 info.msi_node = TAILQ_LAST(&mountlist, mntlist); 713 while ((mp = info.msi_node) != NULL) { 714 mount_hold(mp); 715 if (how & MNTSCAN_NOBUSY) { 716 lwkt_reltoken(&mountlist_token); 717 count = callback(mp, data); 718 lwkt_gettoken_shared(&mountlist_token); 719 } else if (vfs_busy(mp, LK_NOWAIT) == 0) { 720 lwkt_reltoken(&mountlist_token); 721 count = callback(mp, data); 722 lwkt_gettoken_shared(&mountlist_token); 723 if (mp == info.msi_node) 724 vfs_unbusy(mp); 725 } else { 726 count = 0; 727 } 728 mount_drop(mp); 729 if (count < 0) 730 break; 731 res += count; 732 if (mp == info.msi_node) 733 info.msi_node = TAILQ_PREV(mp, mntlist, 734 mnt_list); 735 } 736 } 737 lwkt_reltoken(&mountlist_token); 738 739 lwkt_gettoken(&mountlist_token); 740 TAILQ_REMOVE(&mountscan_list, &info, msi_entry); 741 lwkt_reltoken(&mountlist_token); 742 743 return(res); 744 } 745 746 /* 747 * MOUNT RELATED VNODE FUNCTIONS 748 */ 749 750 static struct kproc_desc vnlru_kp = { 751 "vnlru", 752 vnlru_proc, 753 &vnlruthread 754 }; 755 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp); 756 757 /* 758 * Move a vnode from one mount queue to another. 759 */ 760 void 761 insmntque(struct vnode *vp, struct mount *mp) 762 { 763 struct mount *omp; 764 765 /* 766 * Delete from old mount point vnode list, if on one. 767 */ 768 if ((omp = vp->v_mount) != NULL) { 769 lwkt_gettoken(&omp->mnt_token); 770 KKASSERT(omp == vp->v_mount); 771 KASSERT(omp->mnt_nvnodelistsize > 0, 772 ("bad mount point vnode list size")); 773 vremovevnodemnt(vp); 774 omp->mnt_nvnodelistsize--; 775 lwkt_reltoken(&omp->mnt_token); 776 } 777 778 /* 779 * Insert into list of vnodes for the new mount point, if available. 780 * The 'end' of the LRU list is the vnode prior to mp->mnt_syncer. 781 */ 782 if (mp == NULL) { 783 vp->v_mount = NULL; 784 return; 785 } 786 lwkt_gettoken(&mp->mnt_token); 787 vp->v_mount = mp; 788 if (mp->mnt_syncer) { 789 TAILQ_INSERT_BEFORE(mp->mnt_syncer, vp, v_nmntvnodes); 790 } else { 791 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 792 } 793 mp->mnt_nvnodelistsize++; 794 lwkt_reltoken(&mp->mnt_token); 795 } 796 797 798 /* 799 * Scan the vnodes under a mount point and issue appropriate callbacks. 800 * 801 * The fastfunc() callback is called with just the mountlist token held 802 * (no vnode lock). It may not block and the vnode may be undergoing 803 * modifications while the caller is processing it. The vnode will 804 * not be entirely destroyed, however, due to the fact that the mountlist 805 * token is held. A return value < 0 skips to the next vnode without calling 806 * the slowfunc(), a return value > 0 terminates the loop. 807 * 808 * WARNING! The fastfunc() should not indirect through vp->v_object, the vp 809 * data structure is unstable when called from fastfunc(). 810 * 811 * The slowfunc() callback is called after the vnode has been successfully 812 * locked based on passed flags. The vnode is skipped if it gets rearranged 813 * or destroyed while blocking on the lock. A non-zero return value from 814 * the slow function terminates the loop. The slow function is allowed to 815 * arbitrarily block. The scanning code guarentees consistency of operation 816 * even if the slow function deletes or moves the node, or blocks and some 817 * other thread deletes or moves the node. 818 */ 819 int 820 vmntvnodescan( 821 struct mount *mp, 822 int flags, 823 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data), 824 int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data), 825 void *data 826 ) { 827 struct vmntvnodescan_info info; 828 struct vnode *vp; 829 int r = 0; 830 int maxcount = mp->mnt_nvnodelistsize * 2; 831 int stopcount = 0; 832 int count = 0; 833 834 lwkt_gettoken(&mp->mnt_token); 835 836 /* 837 * If asked to do one pass stop after iterating available vnodes. 838 * Under heavy loads new vnodes can be added while we are scanning, 839 * so this isn't perfect. Create a slop factor of 2x. 840 */ 841 if (flags & VMSC_ONEPASS) 842 stopcount = mp->mnt_nvnodelistsize; 843 844 info.vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 845 TAILQ_INSERT_TAIL(&mp->mnt_vnodescan_list, &info, entry); 846 847 while ((vp = info.vp) != NULL) { 848 if (--maxcount == 0) { 849 kprintf("Warning: excessive fssync iteration\n"); 850 maxcount = mp->mnt_nvnodelistsize * 2; 851 } 852 853 /* 854 * Skip if visible but not ready, or special (e.g. 855 * mp->mnt_syncer) 856 */ 857 if (vp->v_type == VNON) 858 goto next; 859 KKASSERT(vp->v_mount == mp); 860 861 /* 862 * Quick test. A negative return continues the loop without 863 * calling the slow test. 0 continues onto the slow test. 864 * A positive number aborts the loop. 865 */ 866 if (fastfunc) { 867 if ((r = fastfunc(mp, vp, data)) < 0) { 868 r = 0; 869 goto next; 870 } 871 if (r) 872 break; 873 } 874 875 /* 876 * Get a vxlock on the vnode, retry if it has moved or isn't 877 * in the mountlist where we expect it. 878 */ 879 if (slowfunc) { 880 int error; 881 882 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) { 883 case VMSC_GETVP: 884 error = vget(vp, LK_EXCLUSIVE); 885 break; 886 case VMSC_GETVP|VMSC_NOWAIT: 887 error = vget(vp, LK_EXCLUSIVE|LK_NOWAIT); 888 break; 889 case VMSC_GETVX: 890 vx_get(vp); 891 error = 0; 892 break; 893 default: 894 error = 0; 895 break; 896 } 897 if (error) 898 goto next; 899 /* 900 * Do not call the slow function if the vnode is 901 * invalid or if it was ripped out from under us 902 * while we (potentially) blocked. 903 */ 904 if (info.vp == vp && vp->v_type != VNON) 905 r = slowfunc(mp, vp, data); 906 907 /* 908 * Cleanup 909 */ 910 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) { 911 case VMSC_GETVP: 912 case VMSC_GETVP|VMSC_NOWAIT: 913 vput(vp); 914 break; 915 case VMSC_GETVX: 916 vx_put(vp); 917 break; 918 default: 919 break; 920 } 921 if (r != 0) 922 break; 923 } 924 925 next: 926 /* 927 * Yield after some processing. Depending on the number 928 * of vnodes, we might wind up running for a long time. 929 * Because threads are not preemptable, time critical 930 * userland processes might starve. Give them a chance 931 * now and then. 932 */ 933 if (++count == 10000) { 934 /* 935 * We really want to yield a bit, so we simply 936 * sleep a tick 937 */ 938 tsleep(mp, 0, "vnodescn", 1); 939 count = 0; 940 } 941 942 /* 943 * If doing one pass this decrements to zero. If it starts 944 * at zero it is effectively unlimited for the purposes of 945 * this loop. 946 */ 947 if (--stopcount == 0) 948 break; 949 950 /* 951 * Iterate. If the vnode was ripped out from under us 952 * info.vp will already point to the next vnode, otherwise 953 * we have to obtain the next valid vnode ourselves. 954 */ 955 if (info.vp == vp) 956 info.vp = TAILQ_NEXT(vp, v_nmntvnodes); 957 } 958 959 TAILQ_REMOVE(&mp->mnt_vnodescan_list, &info, entry); 960 lwkt_reltoken(&mp->mnt_token); 961 return(r); 962 } 963 964 /* 965 * Remove any vnodes in the vnode table belonging to mount point mp. 966 * 967 * If FORCECLOSE is not specified, there should not be any active ones, 968 * return error if any are found (nb: this is a user error, not a 969 * system error). If FORCECLOSE is specified, detach any active vnodes 970 * that are found. 971 * 972 * If WRITECLOSE is set, only flush out regular file vnodes open for 973 * writing. 974 * 975 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped. 976 * 977 * `rootrefs' specifies the base reference count for the root vnode 978 * of this filesystem. The root vnode is considered busy if its 979 * v_refcnt exceeds this value. On a successful return, vflush() 980 * will call vrele() on the root vnode exactly rootrefs times. 981 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 982 * be zero. 983 */ 984 #ifdef DIAGNOSTIC 985 static int busyprt = 0; /* print out busy vnodes */ 986 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 987 #endif 988 989 static int vflush_scan(struct mount *mp, struct vnode *vp, void *data); 990 991 struct vflush_info { 992 int flags; 993 int busy; 994 thread_t td; 995 }; 996 997 int 998 vflush(struct mount *mp, int rootrefs, int flags) 999 { 1000 struct thread *td = curthread; /* XXX */ 1001 struct vnode *rootvp = NULL; 1002 int error; 1003 struct vflush_info vflush_info; 1004 1005 if (rootrefs > 0) { 1006 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 1007 ("vflush: bad args")); 1008 /* 1009 * Get the filesystem root vnode. We can vput() it 1010 * immediately, since with rootrefs > 0, it won't go away. 1011 */ 1012 if ((error = VFS_ROOT(mp, &rootvp)) != 0) { 1013 if ((flags & FORCECLOSE) == 0) 1014 return (error); 1015 rootrefs = 0; 1016 /* continue anyway */ 1017 } 1018 if (rootrefs) 1019 vput(rootvp); 1020 } 1021 1022 vflush_info.busy = 0; 1023 vflush_info.flags = flags; 1024 vflush_info.td = td; 1025 vmntvnodescan(mp, VMSC_GETVX, NULL, vflush_scan, &vflush_info); 1026 1027 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 1028 /* 1029 * If just the root vnode is busy, and if its refcount 1030 * is equal to `rootrefs', then go ahead and kill it. 1031 */ 1032 KASSERT(vflush_info.busy > 0, ("vflush: not busy")); 1033 KASSERT(VREFCNT(rootvp) >= rootrefs, ("vflush: rootrefs")); 1034 if (vflush_info.busy == 1 && VREFCNT(rootvp) == rootrefs) { 1035 vx_lock(rootvp); 1036 vgone_vxlocked(rootvp); 1037 vx_unlock(rootvp); 1038 vflush_info.busy = 0; 1039 } 1040 } 1041 if (vflush_info.busy) 1042 return (EBUSY); 1043 for (; rootrefs > 0; rootrefs--) 1044 vrele(rootvp); 1045 return (0); 1046 } 1047 1048 /* 1049 * The scan callback is made with an VX locked vnode. 1050 */ 1051 static int 1052 vflush_scan(struct mount *mp, struct vnode *vp, void *data) 1053 { 1054 struct vflush_info *info = data; 1055 struct vattr vattr; 1056 int flags = info->flags; 1057 1058 /* 1059 * Generally speaking try to deactivate on 0 refs (catch-all) 1060 */ 1061 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 1062 1063 /* 1064 * Skip over a vnodes marked VSYSTEM. 1065 */ 1066 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1067 return(0); 1068 } 1069 1070 /* 1071 * Do not force-close VCHR or VBLK vnodes 1072 */ 1073 if (vp->v_type == VCHR || vp->v_type == VBLK) 1074 flags &= ~(WRITECLOSE|FORCECLOSE); 1075 1076 /* 1077 * If WRITECLOSE is set, flush out unlinked but still open 1078 * files (even if open only for reading) and regular file 1079 * vnodes open for writing. 1080 */ 1081 if ((flags & WRITECLOSE) && 1082 (vp->v_type == VNON || 1083 (VOP_GETATTR(vp, &vattr) == 0 && 1084 vattr.va_nlink > 0)) && 1085 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1086 return(0); 1087 } 1088 1089 /* 1090 * If we are the only holder (refcnt of 1) or the vnode is in 1091 * termination (refcnt < 0), we can vgone the vnode. 1092 */ 1093 if (VREFCNT(vp) <= 1) { 1094 vgone_vxlocked(vp); 1095 return(0); 1096 } 1097 1098 /* 1099 * If FORCECLOSE is set, forcibly destroy the vnode and then move 1100 * it to a dummymount structure so vop_*() functions don't deref 1101 * a NULL pointer. 1102 */ 1103 if (flags & FORCECLOSE) { 1104 vhold(vp); 1105 vgone_vxlocked(vp); 1106 if (vp->v_mount == NULL) 1107 insmntque(vp, &dummymount); 1108 vdrop(vp); 1109 return(0); 1110 } 1111 if (vp->v_type == VCHR || vp->v_type == VBLK) 1112 kprintf("vflush: Warning, cannot destroy busy device vnode\n"); 1113 #ifdef DIAGNOSTIC 1114 if (busyprt) 1115 vprint("vflush: busy vnode", vp); 1116 #endif 1117 ++info->busy; 1118 return(0); 1119 } 1120 1121 void 1122 add_bio_ops(struct bio_ops *ops) 1123 { 1124 TAILQ_INSERT_TAIL(&bio_ops_list, ops, entry); 1125 } 1126 1127 void 1128 rem_bio_ops(struct bio_ops *ops) 1129 { 1130 TAILQ_REMOVE(&bio_ops_list, ops, entry); 1131 } 1132 1133 /* 1134 * This calls the bio_ops io_sync function either for a mount point 1135 * or generally. 1136 * 1137 * WARNING: softdeps is weirdly coded and just isn't happy unless 1138 * io_sync is called with a NULL mount from the general syncing code. 1139 */ 1140 void 1141 bio_ops_sync(struct mount *mp) 1142 { 1143 struct bio_ops *ops; 1144 1145 if (mp) { 1146 if ((ops = mp->mnt_bioops) != NULL) 1147 ops->io_sync(mp); 1148 } else { 1149 TAILQ_FOREACH(ops, &bio_ops_list, entry) { 1150 ops->io_sync(NULL); 1151 } 1152 } 1153 } 1154 1155 /* 1156 * Lookup a mount point by nch 1157 */ 1158 struct mount * 1159 mount_get_by_nc(struct namecache *ncp) 1160 { 1161 struct mount *mp = NULL; 1162 1163 lwkt_gettoken_shared(&mountlist_token); 1164 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 1165 if (ncp == mp->mnt_ncmountpt.ncp) 1166 break; 1167 } 1168 lwkt_reltoken(&mountlist_token); 1169 1170 return (mp); 1171 } 1172 1173