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 #include <sys/sysref2.h> 88 89 #include <vm/vm.h> 90 #include <vm/vm_object.h> 91 92 struct mountscan_info { 93 TAILQ_ENTRY(mountscan_info) msi_entry; 94 int msi_how; 95 struct mount *msi_node; 96 }; 97 98 struct vmntvnodescan_info { 99 TAILQ_ENTRY(vmntvnodescan_info) entry; 100 struct vnode *vp; 101 }; 102 103 struct vnlru_info { 104 int pass; 105 }; 106 107 static int vnlru_nowhere = 0; 108 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RD, 109 &vnlru_nowhere, 0, 110 "Number of times the vnlru process ran without success"); 111 112 113 static struct lwkt_token mntid_token; 114 static struct mount dummymount; 115 116 /* note: mountlist exported to pstat */ 117 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 118 static TAILQ_HEAD(,mountscan_info) mountscan_list; 119 static struct lwkt_token mountlist_token; 120 121 static TAILQ_HEAD(,bio_ops) bio_ops_list = TAILQ_HEAD_INITIALIZER(bio_ops_list); 122 123 /* 124 * Called from vfsinit() 125 */ 126 void 127 vfs_mount_init(void) 128 { 129 lwkt_token_init(&mountlist_token, "mntlist"); 130 lwkt_token_init(&mntid_token, "mntid"); 131 TAILQ_INIT(&mountscan_list); 132 mount_init(&dummymount); 133 dummymount.mnt_flag |= MNT_RDONLY; 134 dummymount.mnt_kern_flag |= MNTK_ALL_MPSAFE; 135 } 136 137 /* 138 * Support function called to remove a vnode from the mountlist and 139 * deal with side effects for scans in progress. 140 * 141 * Target mnt_token is held on call. 142 */ 143 static void 144 vremovevnodemnt(struct vnode *vp) 145 { 146 struct vmntvnodescan_info *info; 147 struct mount *mp = vp->v_mount; 148 149 TAILQ_FOREACH(info, &mp->mnt_vnodescan_list, entry) { 150 if (info->vp == vp) 151 info->vp = TAILQ_NEXT(vp, v_nmntvnodes); 152 } 153 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); 154 } 155 156 /* 157 * Allocate a new vnode and associate it with a tag, mount point, and 158 * operations vector. 159 * 160 * A VX locked and refd vnode is returned. The caller should setup the 161 * remaining fields and vx_put() or, if he wishes to leave a vref, 162 * vx_unlock() the vnode. 163 */ 164 int 165 getnewvnode(enum vtagtype tag, struct mount *mp, 166 struct vnode **vpp, int lktimeout, int lkflags) 167 { 168 struct vnode *vp; 169 170 KKASSERT(mp != NULL); 171 172 vp = allocvnode(lktimeout, lkflags); 173 vp->v_tag = tag; 174 vp->v_data = NULL; 175 176 /* 177 * By default the vnode is assigned the mount point's normal 178 * operations vector. 179 */ 180 vp->v_ops = &mp->mnt_vn_use_ops; 181 vp->v_pbuf_count = nswbuf_kva / NSWBUF_SPLIT; 182 183 /* 184 * Placing the vnode on the mount point's queue makes it visible. 185 * VNON prevents it from being messed with, however. 186 */ 187 insmntque(vp, mp); 188 189 /* 190 * A VX locked & refd vnode is returned. 191 */ 192 *vpp = vp; 193 return (0); 194 } 195 196 /* 197 * This function creates vnodes with special operations vectors. The 198 * mount point is optional. 199 * 200 * This routine is being phased out but is still used by vfs_conf to 201 * create vnodes for devices prior to the root mount (with mp == NULL). 202 */ 203 int 204 getspecialvnode(enum vtagtype tag, struct mount *mp, 205 struct vop_ops **ops, 206 struct vnode **vpp, int lktimeout, int lkflags) 207 { 208 struct vnode *vp; 209 210 vp = allocvnode(lktimeout, lkflags); 211 vp->v_tag = tag; 212 vp->v_data = NULL; 213 vp->v_ops = ops; 214 215 if (mp == NULL) 216 mp = &dummymount; 217 218 /* 219 * Placing the vnode on the mount point's queue makes it visible. 220 * VNON prevents it from being messed with, however. 221 */ 222 insmntque(vp, mp); 223 224 /* 225 * A VX locked & refd vnode is returned. 226 */ 227 *vpp = vp; 228 return (0); 229 } 230 231 /* 232 * Interlock against an unmount, return 0 on success, non-zero on failure. 233 * 234 * The passed flag may be 0 or LK_NOWAIT and is only used if an unmount 235 * is in-progress. 236 * 237 * If no unmount is in-progress LK_NOWAIT is ignored. No other flag bits 238 * are used. A shared locked will be obtained and the filesystem will not 239 * be unmountable until the lock is released. 240 */ 241 int 242 vfs_busy(struct mount *mp, int flags) 243 { 244 int lkflags; 245 246 atomic_add_int(&mp->mnt_refs, 1); 247 lwkt_gettoken(&mp->mnt_token); 248 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 249 if (flags & LK_NOWAIT) { 250 lwkt_reltoken(&mp->mnt_token); 251 atomic_add_int(&mp->mnt_refs, -1); 252 return (ENOENT); 253 } 254 /* XXX not MP safe */ 255 mp->mnt_kern_flag |= MNTK_MWAIT; 256 257 /* 258 * Since all busy locks are shared except the exclusive 259 * lock granted when unmounting, the only place that a 260 * wakeup needs to be done is at the release of the 261 * exclusive lock at the end of dounmount. 262 * 263 * WARNING! mp can potentially go away once we release 264 * our ref. 265 */ 266 tsleep((caddr_t)mp, 0, "vfs_busy", 0); 267 lwkt_reltoken(&mp->mnt_token); 268 atomic_add_int(&mp->mnt_refs, -1); 269 return (ENOENT); 270 } 271 lkflags = LK_SHARED; 272 if (lockmgr(&mp->mnt_lock, lkflags)) 273 panic("vfs_busy: unexpected lock failure"); 274 lwkt_reltoken(&mp->mnt_token); 275 return (0); 276 } 277 278 /* 279 * Free a busy filesystem. 280 * 281 * Once refs is decremented the mount point can potentially get ripped 282 * out from under us, but we want to clean up our refs before unlocking 283 * so do a hold/drop around the whole mess. 284 * 285 * This is not in the critical path (I hope). 286 */ 287 void 288 vfs_unbusy(struct mount *mp) 289 { 290 mount_hold(mp); 291 atomic_add_int(&mp->mnt_refs, -1); 292 lockmgr(&mp->mnt_lock, LK_RELEASE); 293 mount_drop(mp); 294 } 295 296 /* 297 * Lookup a filesystem type, and if found allocate and initialize 298 * a mount structure for it. 299 * 300 * Devname is usually updated by mount(8) after booting. 301 */ 302 int 303 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp) 304 { 305 struct vfsconf *vfsp; 306 struct mount *mp; 307 308 if (fstypename == NULL) 309 return (ENODEV); 310 311 vfsp = vfsconf_find_by_name(fstypename); 312 if (vfsp == NULL) 313 return (ENODEV); 314 mp = kmalloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO); 315 mount_init(mp); 316 lockinit(&mp->mnt_lock, "vfslock", VLKTIMEOUT, 0); 317 318 vfs_busy(mp, 0); 319 mp->mnt_vfc = vfsp; 320 mp->mnt_op = vfsp->vfc_vfsops; 321 mp->mnt_pbuf_count = nswbuf_kva / NSWBUF_SPLIT; 322 vfsp->vfc_refcount++; 323 mp->mnt_stat.f_type = vfsp->vfc_typenum; 324 mp->mnt_flag |= MNT_RDONLY; 325 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 326 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 327 copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 328 329 /* 330 * Pre-set MPSAFE flags for VFS_MOUNT() call. 331 */ 332 if (vfsp->vfc_flags & VFCF_MPSAFE) 333 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; 334 335 *mpp = mp; 336 337 return (0); 338 } 339 340 /* 341 * Basic mount structure initialization 342 */ 343 void 344 mount_init(struct mount *mp) 345 { 346 lockinit(&mp->mnt_lock, "vfslock", hz*5, 0); 347 lwkt_token_init(&mp->mnt_token, "permnt"); 348 349 TAILQ_INIT(&mp->mnt_vnodescan_list); 350 TAILQ_INIT(&mp->mnt_nvnodelist); 351 TAILQ_INIT(&mp->mnt_reservedvnlist); 352 TAILQ_INIT(&mp->mnt_jlist); 353 mp->mnt_nvnodelistsize = 0; 354 mp->mnt_flag = 0; 355 mp->mnt_hold = 1; /* hold for umount last drop */ 356 mp->mnt_iosize_max = MAXPHYS; 357 vn_syncer_thr_create(mp); 358 } 359 360 void 361 mount_hold(struct mount *mp) 362 { 363 atomic_add_int(&mp->mnt_hold, 1); 364 } 365 366 void 367 mount_drop(struct mount *mp) 368 { 369 if (atomic_fetchadd_int(&mp->mnt_hold, -1) == 1) { 370 KKASSERT(mp->mnt_refs == 0); 371 kfree(mp, M_MOUNT); 372 } 373 } 374 375 /* 376 * Lookup a mount point by filesystem identifier. 377 */ 378 struct mount * 379 vfs_getvfs(fsid_t *fsid) 380 { 381 struct mount *mp; 382 383 lwkt_gettoken(&mountlist_token); 384 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 385 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 386 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 387 break; 388 } 389 } 390 lwkt_reltoken(&mountlist_token); 391 return (mp); 392 } 393 394 /* 395 * Get a new unique fsid. Try to make its val[0] unique, since this value 396 * will be used to create fake device numbers for stat(). Also try (but 397 * not so hard) make its val[0] unique mod 2^16, since some emulators only 398 * support 16-bit device numbers. We end up with unique val[0]'s for the 399 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 400 * 401 * Keep in mind that several mounts may be running in parallel. Starting 402 * the search one past where the previous search terminated is both a 403 * micro-optimization and a defense against returning the same fsid to 404 * different mounts. 405 */ 406 void 407 vfs_getnewfsid(struct mount *mp) 408 { 409 static u_int16_t mntid_base; 410 fsid_t tfsid; 411 int mtype; 412 413 lwkt_gettoken(&mntid_token); 414 mtype = mp->mnt_vfc->vfc_typenum; 415 tfsid.val[1] = mtype; 416 mtype = (mtype & 0xFF) << 24; 417 for (;;) { 418 tfsid.val[0] = makeudev(255, 419 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 420 mntid_base++; 421 if (vfs_getvfs(&tfsid) == NULL) 422 break; 423 } 424 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 425 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 426 lwkt_reltoken(&mntid_token); 427 } 428 429 /* 430 * Set the FSID for a new mount point to the template. Adjust 431 * the FSID to avoid collisions. 432 */ 433 int 434 vfs_setfsid(struct mount *mp, fsid_t *template) 435 { 436 int didmunge = 0; 437 438 bzero(&mp->mnt_stat.f_fsid, sizeof(mp->mnt_stat.f_fsid)); 439 for (;;) { 440 if (vfs_getvfs(template) == NULL) 441 break; 442 didmunge = 1; 443 ++template->val[1]; 444 } 445 mp->mnt_stat.f_fsid = *template; 446 return(didmunge); 447 } 448 449 /* 450 * This routine is called when we have too many vnodes. It attempts 451 * to free <count> vnodes and will potentially free vnodes that still 452 * have VM backing store (VM backing store is typically the cause 453 * of a vnode blowout so we want to do this). Therefore, this operation 454 * is not considered cheap. 455 * 456 * A number of conditions may prevent a vnode from being reclaimed. 457 * the buffer cache may have references on the vnode, a directory 458 * vnode may still have references due to the namei cache representing 459 * underlying files, or the vnode may be in active use. It is not 460 * desireable to reuse such vnodes. These conditions may cause the 461 * number of vnodes to reach some minimum value regardless of what 462 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 463 */ 464 465 /* 466 * Attempt to recycle vnodes in a context that is always safe to block. 467 * Calling vlrurecycle() from the bowels of file system code has some 468 * interesting deadlock problems. 469 */ 470 static struct thread *vnlruthread; 471 472 static void 473 vnlru_proc(void) 474 { 475 struct thread *td = curthread; 476 477 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td, 478 SHUTDOWN_PRI_FIRST); 479 480 for (;;) { 481 int ncached; 482 483 kproc_suspend_loop(); 484 485 /* 486 * Try to free some vnodes if we have too many. Trigger based 487 * on potentially freeable vnodes but calculate the count 488 * based on total vnodes. 489 * 490 * (long) -> deal with 64 bit machines, intermediate overflow 491 */ 492 ncached = countcachedvnodes(1); 493 if (numvnodes >= maxvnodes * 9 / 10 && 494 ncached + inactivevnodes >= maxvnodes * 5 / 10) { 495 int count = numvnodes - maxvnodes * 9 / 10; 496 497 if (count > (ncached + inactivevnodes) / 100) 498 count = (ncached + inactivevnodes) / 100; 499 if (count < 5) 500 count = 5; 501 freesomevnodes(count); 502 } 503 504 /* 505 * Do non-critical-path (more robust) cache cleaning, 506 * even if vnode counts are nominal, to try to avoid 507 * having to do it in the critical path. 508 */ 509 cache_hysteresis(0); 510 511 /* 512 * Nothing to do if most of our vnodes are already on 513 * the free list. 514 */ 515 ncached = countcachedvnodes(1); 516 if (numvnodes <= maxvnodes * 9 / 10 || 517 ncached + inactivevnodes <= maxvnodes * 5 / 10) { 518 tsleep(vnlruthread, 0, "vlruwt", hz); 519 continue; 520 } 521 } 522 } 523 524 /* 525 * MOUNTLIST FUNCTIONS 526 */ 527 528 /* 529 * mountlist_insert (MP SAFE) 530 * 531 * Add a new mount point to the mount list. 532 */ 533 void 534 mountlist_insert(struct mount *mp, int how) 535 { 536 lwkt_gettoken(&mountlist_token); 537 if (how == MNTINS_FIRST) 538 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 539 else 540 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 541 lwkt_reltoken(&mountlist_token); 542 } 543 544 /* 545 * mountlist_interlock (MP SAFE) 546 * 547 * Execute the specified interlock function with the mountlist token 548 * held. The function will be called in a serialized fashion verses 549 * other functions called through this mechanism. 550 */ 551 int 552 mountlist_interlock(int (*callback)(struct mount *), struct mount *mp) 553 { 554 int error; 555 556 lwkt_gettoken(&mountlist_token); 557 error = callback(mp); 558 lwkt_reltoken(&mountlist_token); 559 return (error); 560 } 561 562 /* 563 * mountlist_boot_getfirst (DURING BOOT ONLY) 564 * 565 * This function returns the first mount on the mountlist, which is 566 * expected to be the root mount. Since no interlocks are obtained 567 * this function is only safe to use during booting. 568 */ 569 570 struct mount * 571 mountlist_boot_getfirst(void) 572 { 573 return(TAILQ_FIRST(&mountlist)); 574 } 575 576 /* 577 * mountlist_remove (MP SAFE) 578 * 579 * Remove a node from the mountlist. If this node is the next scan node 580 * for any active mountlist scans, the active mountlist scan will be 581 * adjusted to skip the node, thus allowing removals during mountlist 582 * scans. 583 */ 584 void 585 mountlist_remove(struct mount *mp) 586 { 587 struct mountscan_info *msi; 588 589 lwkt_gettoken(&mountlist_token); 590 TAILQ_FOREACH(msi, &mountscan_list, msi_entry) { 591 if (msi->msi_node == mp) { 592 if (msi->msi_how & MNTSCAN_FORWARD) 593 msi->msi_node = TAILQ_NEXT(mp, mnt_list); 594 else 595 msi->msi_node = TAILQ_PREV(mp, mntlist, mnt_list); 596 } 597 } 598 TAILQ_REMOVE(&mountlist, mp, mnt_list); 599 lwkt_reltoken(&mountlist_token); 600 } 601 602 /* 603 * mountlist_exists (MP SAFE) 604 * 605 * Checks if a node exists in the mountlist. 606 * This function is mainly used by VFS quota code to check if a 607 * cached nullfs struct mount pointer is still valid at use time 608 * 609 * FIXME: there is no warranty the mp passed to that function 610 * will be the same one used by VFS_ACCOUNT() later 611 */ 612 int 613 mountlist_exists(struct mount *mp) 614 { 615 int node_exists = 0; 616 struct mount* lmp; 617 618 lwkt_gettoken(&mountlist_token); 619 TAILQ_FOREACH(lmp, &mountlist, mnt_list) { 620 if (lmp == mp) { 621 node_exists = 1; 622 break; 623 } 624 } 625 lwkt_reltoken(&mountlist_token); 626 return(node_exists); 627 } 628 629 /* 630 * mountlist_scan (MP SAFE) 631 * 632 * Safely scan the mount points on the mount list. Unless otherwise 633 * specified each mount point will be busied prior to the callback and 634 * unbusied afterwords. The callback may safely remove any mount point 635 * without interfering with the scan. If the current callback 636 * mount is removed the scanner will not attempt to unbusy it. 637 * 638 * If a mount node cannot be busied it is silently skipped. 639 * 640 * The callback return value is aggregated and a total is returned. A return 641 * value of < 0 is not aggregated and will terminate the scan. 642 * 643 * MNTSCAN_FORWARD - the mountlist is scanned in the forward direction 644 * MNTSCAN_REVERSE - the mountlist is scanned in reverse 645 * MNTSCAN_NOBUSY - the scanner will make the callback without busying 646 * the mount node. 647 * 648 * NOTE: mount_hold()/mount_drop() sequence primarily helps us avoid 649 * confusion for the unbusy check, particularly if a kfree/kmalloc 650 * occurs quickly (lots of processes mounting and unmounting at the 651 * same time). 652 */ 653 int 654 mountlist_scan(int (*callback)(struct mount *, void *), void *data, int how) 655 { 656 struct mountscan_info info; 657 struct mount *mp; 658 int count; 659 int res; 660 661 lwkt_gettoken(&mountlist_token); 662 663 info.msi_how = how; 664 info.msi_node = NULL; /* paranoia */ 665 TAILQ_INSERT_TAIL(&mountscan_list, &info, msi_entry); 666 667 res = 0; 668 669 if (how & MNTSCAN_FORWARD) { 670 info.msi_node = TAILQ_FIRST(&mountlist); 671 while ((mp = info.msi_node) != NULL) { 672 mount_hold(mp); 673 if (how & MNTSCAN_NOBUSY) { 674 count = callback(mp, data); 675 } else if (vfs_busy(mp, LK_NOWAIT) == 0) { 676 count = callback(mp, data); 677 if (mp == info.msi_node) 678 vfs_unbusy(mp); 679 } else { 680 count = 0; 681 } 682 mount_drop(mp); 683 if (count < 0) 684 break; 685 res += count; 686 if (mp == info.msi_node) 687 info.msi_node = TAILQ_NEXT(mp, mnt_list); 688 } 689 } else if (how & MNTSCAN_REVERSE) { 690 info.msi_node = TAILQ_LAST(&mountlist, mntlist); 691 while ((mp = info.msi_node) != NULL) { 692 mount_hold(mp); 693 if (how & MNTSCAN_NOBUSY) { 694 count = callback(mp, data); 695 } else if (vfs_busy(mp, LK_NOWAIT) == 0) { 696 count = callback(mp, data); 697 if (mp == info.msi_node) 698 vfs_unbusy(mp); 699 } else { 700 count = 0; 701 } 702 mount_drop(mp); 703 if (count < 0) 704 break; 705 res += count; 706 if (mp == info.msi_node) 707 info.msi_node = TAILQ_PREV(mp, mntlist, mnt_list); 708 } 709 } 710 TAILQ_REMOVE(&mountscan_list, &info, msi_entry); 711 lwkt_reltoken(&mountlist_token); 712 return(res); 713 } 714 715 /* 716 * MOUNT RELATED VNODE FUNCTIONS 717 */ 718 719 static struct kproc_desc vnlru_kp = { 720 "vnlru", 721 vnlru_proc, 722 &vnlruthread 723 }; 724 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp); 725 726 /* 727 * Move a vnode from one mount queue to another. 728 */ 729 void 730 insmntque(struct vnode *vp, struct mount *mp) 731 { 732 struct mount *omp; 733 734 /* 735 * Delete from old mount point vnode list, if on one. 736 */ 737 if ((omp = vp->v_mount) != NULL) { 738 lwkt_gettoken(&omp->mnt_token); 739 KKASSERT(omp == vp->v_mount); 740 KASSERT(omp->mnt_nvnodelistsize > 0, 741 ("bad mount point vnode list size")); 742 vremovevnodemnt(vp); 743 omp->mnt_nvnodelistsize--; 744 lwkt_reltoken(&omp->mnt_token); 745 } 746 747 /* 748 * Insert into list of vnodes for the new mount point, if available. 749 * The 'end' of the LRU list is the vnode prior to mp->mnt_syncer. 750 */ 751 if (mp == NULL) { 752 vp->v_mount = NULL; 753 return; 754 } 755 lwkt_gettoken(&mp->mnt_token); 756 vp->v_mount = mp; 757 if (mp->mnt_syncer) { 758 TAILQ_INSERT_BEFORE(mp->mnt_syncer, vp, v_nmntvnodes); 759 } else { 760 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 761 } 762 mp->mnt_nvnodelistsize++; 763 lwkt_reltoken(&mp->mnt_token); 764 } 765 766 767 /* 768 * Scan the vnodes under a mount point and issue appropriate callbacks. 769 * 770 * The fastfunc() callback is called with just the mountlist token held 771 * (no vnode lock). It may not block and the vnode may be undergoing 772 * modifications while the caller is processing it. The vnode will 773 * not be entirely destroyed, however, due to the fact that the mountlist 774 * token is held. A return value < 0 skips to the next vnode without calling 775 * the slowfunc(), a return value > 0 terminates the loop. 776 * 777 * WARNING! The fastfunc() should not indirect through vp->v_object, the vp 778 * data structure is unstable when called from fastfunc(). 779 * 780 * The slowfunc() callback is called after the vnode has been successfully 781 * locked based on passed flags. The vnode is skipped if it gets rearranged 782 * or destroyed while blocking on the lock. A non-zero return value from 783 * the slow function terminates the loop. The slow function is allowed to 784 * arbitrarily block. The scanning code guarentees consistency of operation 785 * even if the slow function deletes or moves the node, or blocks and some 786 * other thread deletes or moves the node. 787 */ 788 int 789 vmntvnodescan( 790 struct mount *mp, 791 int flags, 792 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data), 793 int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data), 794 void *data 795 ) { 796 struct vmntvnodescan_info info; 797 struct vnode *vp; 798 int r = 0; 799 int maxcount = mp->mnt_nvnodelistsize * 2; 800 int stopcount = 0; 801 int count = 0; 802 803 lwkt_gettoken(&mp->mnt_token); 804 805 /* 806 * If asked to do one pass stop after iterating available vnodes. 807 * Under heavy loads new vnodes can be added while we are scanning, 808 * so this isn't perfect. Create a slop factor of 2x. 809 */ 810 if (flags & VMSC_ONEPASS) 811 stopcount = mp->mnt_nvnodelistsize; 812 813 info.vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 814 TAILQ_INSERT_TAIL(&mp->mnt_vnodescan_list, &info, entry); 815 816 while ((vp = info.vp) != NULL) { 817 if (--maxcount == 0) { 818 kprintf("Warning: excessive fssync iteration\n"); 819 maxcount = mp->mnt_nvnodelistsize * 2; 820 } 821 822 /* 823 * Skip if visible but not ready, or special (e.g. 824 * mp->mnt_syncer) 825 */ 826 if (vp->v_type == VNON) 827 goto next; 828 KKASSERT(vp->v_mount == mp); 829 830 /* 831 * Quick test. A negative return continues the loop without 832 * calling the slow test. 0 continues onto the slow test. 833 * A positive number aborts the loop. 834 */ 835 if (fastfunc) { 836 if ((r = fastfunc(mp, vp, data)) < 0) { 837 r = 0; 838 goto next; 839 } 840 if (r) 841 break; 842 } 843 844 /* 845 * Get a vxlock on the vnode, retry if it has moved or isn't 846 * in the mountlist where we expect it. 847 */ 848 if (slowfunc) { 849 int error; 850 851 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) { 852 case VMSC_GETVP: 853 error = vget(vp, LK_EXCLUSIVE); 854 break; 855 case VMSC_GETVP|VMSC_NOWAIT: 856 error = vget(vp, LK_EXCLUSIVE|LK_NOWAIT); 857 break; 858 case VMSC_GETVX: 859 vx_get(vp); 860 error = 0; 861 break; 862 default: 863 error = 0; 864 break; 865 } 866 if (error) 867 goto next; 868 /* 869 * Do not call the slow function if the vnode is 870 * invalid or if it was ripped out from under us 871 * while we (potentially) blocked. 872 */ 873 if (info.vp == vp && vp->v_type != VNON) 874 r = slowfunc(mp, vp, data); 875 876 /* 877 * Cleanup 878 */ 879 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) { 880 case VMSC_GETVP: 881 case VMSC_GETVP|VMSC_NOWAIT: 882 vput(vp); 883 break; 884 case VMSC_GETVX: 885 vx_put(vp); 886 break; 887 default: 888 break; 889 } 890 if (r != 0) 891 break; 892 } 893 894 next: 895 /* 896 * Yield after some processing. Depending on the number 897 * of vnodes, we might wind up running for a long time. 898 * Because threads are not preemptable, time critical 899 * userland processes might starve. Give them a chance 900 * now and then. 901 */ 902 if (++count == 10000) { 903 /* 904 * We really want to yield a bit, so we simply 905 * sleep a tick 906 */ 907 tsleep(mp, 0, "vnodescn", 1); 908 count = 0; 909 } 910 911 /* 912 * If doing one pass this decrements to zero. If it starts 913 * at zero it is effectively unlimited for the purposes of 914 * this loop. 915 */ 916 if (--stopcount == 0) 917 break; 918 919 /* 920 * Iterate. If the vnode was ripped out from under us 921 * info.vp will already point to the next vnode, otherwise 922 * we have to obtain the next valid vnode ourselves. 923 */ 924 if (info.vp == vp) 925 info.vp = TAILQ_NEXT(vp, v_nmntvnodes); 926 } 927 928 TAILQ_REMOVE(&mp->mnt_vnodescan_list, &info, entry); 929 lwkt_reltoken(&mp->mnt_token); 930 return(r); 931 } 932 933 /* 934 * Remove any vnodes in the vnode table belonging to mount point mp. 935 * 936 * If FORCECLOSE is not specified, there should not be any active ones, 937 * return error if any are found (nb: this is a user error, not a 938 * system error). If FORCECLOSE is specified, detach any active vnodes 939 * that are found. 940 * 941 * If WRITECLOSE is set, only flush out regular file vnodes open for 942 * writing. 943 * 944 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped. 945 * 946 * `rootrefs' specifies the base reference count for the root vnode 947 * of this filesystem. The root vnode is considered busy if its 948 * v_refcnt exceeds this value. On a successful return, vflush() 949 * will call vrele() on the root vnode exactly rootrefs times. 950 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 951 * be zero. 952 */ 953 #ifdef DIAGNOSTIC 954 static int busyprt = 0; /* print out busy vnodes */ 955 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 956 #endif 957 958 static int vflush_scan(struct mount *mp, struct vnode *vp, void *data); 959 960 struct vflush_info { 961 int flags; 962 int busy; 963 thread_t td; 964 }; 965 966 int 967 vflush(struct mount *mp, int rootrefs, int flags) 968 { 969 struct thread *td = curthread; /* XXX */ 970 struct vnode *rootvp = NULL; 971 int error; 972 struct vflush_info vflush_info; 973 974 if (rootrefs > 0) { 975 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 976 ("vflush: bad args")); 977 /* 978 * Get the filesystem root vnode. We can vput() it 979 * immediately, since with rootrefs > 0, it won't go away. 980 */ 981 if ((error = VFS_ROOT(mp, &rootvp)) != 0) { 982 if ((flags & FORCECLOSE) == 0) 983 return (error); 984 rootrefs = 0; 985 /* continue anyway */ 986 } 987 if (rootrefs) 988 vput(rootvp); 989 } 990 991 vflush_info.busy = 0; 992 vflush_info.flags = flags; 993 vflush_info.td = td; 994 vmntvnodescan(mp, VMSC_GETVX, NULL, vflush_scan, &vflush_info); 995 996 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 997 /* 998 * If just the root vnode is busy, and if its refcount 999 * is equal to `rootrefs', then go ahead and kill it. 1000 */ 1001 KASSERT(vflush_info.busy > 0, ("vflush: not busy")); 1002 KASSERT(VREFCNT(rootvp) >= rootrefs, ("vflush: rootrefs")); 1003 if (vflush_info.busy == 1 && VREFCNT(rootvp) == rootrefs) { 1004 vx_lock(rootvp); 1005 vgone_vxlocked(rootvp); 1006 vx_unlock(rootvp); 1007 vflush_info.busy = 0; 1008 } 1009 } 1010 if (vflush_info.busy) 1011 return (EBUSY); 1012 for (; rootrefs > 0; rootrefs--) 1013 vrele(rootvp); 1014 return (0); 1015 } 1016 1017 /* 1018 * The scan callback is made with an VX locked vnode. 1019 */ 1020 static int 1021 vflush_scan(struct mount *mp, struct vnode *vp, void *data) 1022 { 1023 struct vflush_info *info = data; 1024 struct vattr vattr; 1025 int flags = info->flags; 1026 1027 /* 1028 * Generally speaking try to deactivate on 0 refs (catch-all) 1029 */ 1030 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 1031 1032 /* 1033 * Skip over a vnodes marked VSYSTEM. 1034 */ 1035 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1036 return(0); 1037 } 1038 1039 /* 1040 * Do not force-close VCHR or VBLK vnodes 1041 */ 1042 if (vp->v_type == VCHR || vp->v_type == VBLK) 1043 flags &= ~(WRITECLOSE|FORCECLOSE); 1044 1045 /* 1046 * If WRITECLOSE is set, flush out unlinked but still open 1047 * files (even if open only for reading) and regular file 1048 * vnodes open for writing. 1049 */ 1050 if ((flags & WRITECLOSE) && 1051 (vp->v_type == VNON || 1052 (VOP_GETATTR(vp, &vattr) == 0 && 1053 vattr.va_nlink > 0)) && 1054 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1055 return(0); 1056 } 1057 1058 /* 1059 * If we are the only holder (refcnt of 1) or the vnode is in 1060 * termination (refcnt < 0), we can vgone the vnode. 1061 */ 1062 if (VREFCNT(vp) <= 1) { 1063 vgone_vxlocked(vp); 1064 return(0); 1065 } 1066 1067 /* 1068 * If FORCECLOSE is set, forcibly destroy the vnode and then move 1069 * it to a dummymount structure so vop_*() functions don't deref 1070 * a NULL pointer. 1071 */ 1072 if (flags & FORCECLOSE) { 1073 vhold(vp); 1074 vgone_vxlocked(vp); 1075 if (vp->v_mount == NULL) 1076 insmntque(vp, &dummymount); 1077 vdrop(vp); 1078 return(0); 1079 } 1080 if (vp->v_type == VCHR || vp->v_type == VBLK) 1081 kprintf("vflush: Warning, cannot destroy busy device vnode\n"); 1082 #ifdef DIAGNOSTIC 1083 if (busyprt) 1084 vprint("vflush: busy vnode", vp); 1085 #endif 1086 ++info->busy; 1087 return(0); 1088 } 1089 1090 void 1091 add_bio_ops(struct bio_ops *ops) 1092 { 1093 TAILQ_INSERT_TAIL(&bio_ops_list, ops, entry); 1094 } 1095 1096 void 1097 rem_bio_ops(struct bio_ops *ops) 1098 { 1099 TAILQ_REMOVE(&bio_ops_list, ops, entry); 1100 } 1101 1102 /* 1103 * This calls the bio_ops io_sync function either for a mount point 1104 * or generally. 1105 * 1106 * WARNING: softdeps is weirdly coded and just isn't happy unless 1107 * io_sync is called with a NULL mount from the general syncing code. 1108 */ 1109 void 1110 bio_ops_sync(struct mount *mp) 1111 { 1112 struct bio_ops *ops; 1113 1114 if (mp) { 1115 if ((ops = mp->mnt_bioops) != NULL) 1116 ops->io_sync(mp); 1117 } else { 1118 TAILQ_FOREACH(ops, &bio_ops_list, entry) { 1119 ops->io_sync(NULL); 1120 } 1121 } 1122 } 1123 1124 /* 1125 * Lookup a mount point by nch 1126 */ 1127 struct mount * 1128 mount_get_by_nc(struct namecache *ncp) 1129 { 1130 struct mount *mp = NULL; 1131 1132 lwkt_gettoken(&mountlist_token); 1133 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 1134 if (ncp == mp->mnt_ncmountpt.ncp) 1135 break; 1136 } 1137 lwkt_reltoken(&mountlist_token); 1138 return (mp); 1139 } 1140 1141