1 /* $NetBSD: vfs_subr.c,v 1.378 2009/05/03 16:52:54 pooka Exp $ */ 2 3 /*- 4 * Copyright (c) 1997, 1998, 2004, 2005, 2007, 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center, by Charles M. Hannum, and by Andrew Doran. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 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 * @(#)vfs_subr.c 8.13 (Berkeley) 4/18/94 67 */ 68 69 /* 70 * Note on v_usecount and locking: 71 * 72 * At nearly all points it is known that v_usecount could be zero, the 73 * vnode interlock will be held. 74 * 75 * To change v_usecount away from zero, the interlock must be held. To 76 * change from a non-zero value to zero, again the interlock must be 77 * held. 78 * 79 * Changing the usecount from a non-zero value to a non-zero value can 80 * safely be done using atomic operations, without the interlock held. 81 */ 82 83 #include <sys/cdefs.h> 84 __KERNEL_RCSID(0, "$NetBSD: vfs_subr.c,v 1.378 2009/05/03 16:52:54 pooka Exp $"); 85 86 #include "opt_ddb.h" 87 #include "opt_compat_netbsd.h" 88 #include "opt_compat_43.h" 89 90 #include <sys/param.h> 91 #include <sys/systm.h> 92 #include <sys/conf.h> 93 #include <sys/proc.h> 94 #include <sys/kernel.h> 95 #include <sys/mount.h> 96 #include <sys/fcntl.h> 97 #include <sys/vnode.h> 98 #include <sys/stat.h> 99 #include <sys/namei.h> 100 #include <sys/ucred.h> 101 #include <sys/buf.h> 102 #include <sys/errno.h> 103 #include <sys/kmem.h> 104 #include <sys/syscallargs.h> 105 #include <sys/device.h> 106 #include <sys/filedesc.h> 107 #include <sys/kauth.h> 108 #include <sys/atomic.h> 109 #include <sys/kthread.h> 110 #include <sys/wapbl.h> 111 112 #include <miscfs/specfs/specdev.h> 113 #include <miscfs/syncfs/syncfs.h> 114 115 #include <uvm/uvm.h> 116 #include <uvm/uvm_readahead.h> 117 #include <uvm/uvm_ddb.h> 118 119 #include <sys/sysctl.h> 120 121 const enum vtype iftovt_tab[16] = { 122 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 123 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 124 }; 125 const int vttoif_tab[9] = { 126 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 127 S_IFSOCK, S_IFIFO, S_IFMT, 128 }; 129 130 /* 131 * Insq/Remq for the vnode usage lists. 132 */ 133 #define bufinsvn(bp, dp) LIST_INSERT_HEAD(dp, bp, b_vnbufs) 134 #define bufremvn(bp) { \ 135 LIST_REMOVE(bp, b_vnbufs); \ 136 (bp)->b_vnbufs.le_next = NOLIST; \ 137 } 138 139 int doforce = 1; /* 1 => permit forcible unmounting */ 140 int prtactive = 0; /* 1 => print out reclaim of active vnodes */ 141 142 static vnodelst_t vnode_free_list = TAILQ_HEAD_INITIALIZER(vnode_free_list); 143 static vnodelst_t vnode_hold_list = TAILQ_HEAD_INITIALIZER(vnode_hold_list); 144 static vnodelst_t vrele_list = TAILQ_HEAD_INITIALIZER(vrele_list); 145 146 struct mntlist mountlist = /* mounted filesystem list */ 147 CIRCLEQ_HEAD_INITIALIZER(mountlist); 148 149 u_int numvnodes; 150 static specificdata_domain_t mount_specificdata_domain; 151 152 static int vrele_pending; 153 static int vrele_gen; 154 static kmutex_t vrele_lock; 155 static kcondvar_t vrele_cv; 156 static lwp_t *vrele_lwp; 157 158 kmutex_t mountlist_lock; 159 kmutex_t mntid_lock; 160 kmutex_t mntvnode_lock; 161 kmutex_t vnode_free_list_lock; 162 kmutex_t vfs_list_lock; 163 164 static pool_cache_t vnode_cache; 165 166 /* 167 * These define the root filesystem and device. 168 */ 169 struct vnode *rootvnode; 170 struct device *root_device; /* root device */ 171 172 /* 173 * Local declarations. 174 */ 175 176 static void vrele_thread(void *); 177 static void insmntque(vnode_t *, struct mount *); 178 static int getdevvp(dev_t, vnode_t **, enum vtype); 179 static vnode_t *getcleanvnode(void); 180 void vpanic(vnode_t *, const char *); 181 182 #ifdef DEBUG 183 void printlockedvnodes(void); 184 #endif 185 186 #ifdef DIAGNOSTIC 187 void 188 vpanic(vnode_t *vp, const char *msg) 189 { 190 191 vprint(NULL, vp); 192 panic("%s\n", msg); 193 } 194 #else 195 #define vpanic(vp, msg) /* nothing */ 196 #endif 197 198 void 199 vn_init1(void) 200 { 201 202 vnode_cache = pool_cache_init(sizeof(struct vnode), 0, 0, 0, "vnodepl", 203 NULL, IPL_NONE, NULL, NULL, NULL); 204 KASSERT(vnode_cache != NULL); 205 206 /* Create deferred release thread. */ 207 mutex_init(&vrele_lock, MUTEX_DEFAULT, IPL_NONE); 208 cv_init(&vrele_cv, "vrele"); 209 if (kthread_create(PRI_VM, KTHREAD_MPSAFE, NULL, vrele_thread, 210 NULL, &vrele_lwp, "vrele")) 211 panic("fork vrele"); 212 } 213 214 /* 215 * Initialize the vnode management data structures. 216 */ 217 void 218 vntblinit(void) 219 { 220 221 mutex_init(&mountlist_lock, MUTEX_DEFAULT, IPL_NONE); 222 mutex_init(&mntid_lock, MUTEX_DEFAULT, IPL_NONE); 223 mutex_init(&mntvnode_lock, MUTEX_DEFAULT, IPL_NONE); 224 mutex_init(&vnode_free_list_lock, MUTEX_DEFAULT, IPL_NONE); 225 mutex_init(&vfs_list_lock, MUTEX_DEFAULT, IPL_NONE); 226 227 mount_specificdata_domain = specificdata_domain_create(); 228 229 /* Initialize the filesystem syncer. */ 230 vn_initialize_syncerd(); 231 vn_init1(); 232 } 233 234 int 235 vfs_drainvnodes(long target, struct lwp *l) 236 { 237 238 while (numvnodes > target) { 239 vnode_t *vp; 240 241 mutex_enter(&vnode_free_list_lock); 242 vp = getcleanvnode(); 243 if (vp == NULL) 244 return EBUSY; /* give up */ 245 ungetnewvnode(vp); 246 } 247 248 return 0; 249 } 250 251 /* 252 * Lookup a mount point by filesystem identifier. 253 * 254 * XXX Needs to add a reference to the mount point. 255 */ 256 struct mount * 257 vfs_getvfs(fsid_t *fsid) 258 { 259 struct mount *mp; 260 261 mutex_enter(&mountlist_lock); 262 CIRCLEQ_FOREACH(mp, &mountlist, mnt_list) { 263 if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] && 264 mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) { 265 mutex_exit(&mountlist_lock); 266 return (mp); 267 } 268 } 269 mutex_exit(&mountlist_lock); 270 return ((struct mount *)0); 271 } 272 273 /* 274 * Drop a reference to a mount structure, freeing if the last reference. 275 */ 276 void 277 vfs_destroy(struct mount *mp) 278 { 279 280 if (__predict_true((int)atomic_dec_uint_nv(&mp->mnt_refcnt) > 0)) { 281 return; 282 } 283 284 /* 285 * Nothing else has visibility of the mount: we can now 286 * free the data structures. 287 */ 288 KASSERT(mp->mnt_refcnt == 0); 289 specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref); 290 rw_destroy(&mp->mnt_unmounting); 291 mutex_destroy(&mp->mnt_updating); 292 mutex_destroy(&mp->mnt_renamelock); 293 if (mp->mnt_op != NULL) { 294 vfs_delref(mp->mnt_op); 295 } 296 kmem_free(mp, sizeof(*mp)); 297 } 298 299 /* 300 * grab a vnode from freelist and clean it. 301 */ 302 vnode_t * 303 getcleanvnode(void) 304 { 305 vnode_t *vp; 306 vnodelst_t *listhd; 307 308 KASSERT(mutex_owned(&vnode_free_list_lock)); 309 310 retry: 311 listhd = &vnode_free_list; 312 try_nextlist: 313 TAILQ_FOREACH(vp, listhd, v_freelist) { 314 /* 315 * It's safe to test v_usecount and v_iflag 316 * without holding the interlock here, since 317 * these vnodes should never appear on the 318 * lists. 319 */ 320 if (vp->v_usecount != 0) { 321 vpanic(vp, "free vnode isn't"); 322 } 323 if ((vp->v_iflag & VI_CLEAN) != 0) { 324 vpanic(vp, "clean vnode on freelist"); 325 } 326 if (vp->v_freelisthd != listhd) { 327 printf("vnode sez %p, listhd %p\n", vp->v_freelisthd, listhd); 328 vpanic(vp, "list head mismatch"); 329 } 330 if (!mutex_tryenter(&vp->v_interlock)) 331 continue; 332 /* 333 * Our lwp might hold the underlying vnode 334 * locked, so don't try to reclaim a VI_LAYER 335 * node if it's locked. 336 */ 337 if ((vp->v_iflag & VI_XLOCK) == 0 && 338 ((vp->v_iflag & VI_LAYER) == 0 || VOP_ISLOCKED(vp) == 0)) { 339 break; 340 } 341 mutex_exit(&vp->v_interlock); 342 } 343 344 if (vp == NULL) { 345 if (listhd == &vnode_free_list) { 346 listhd = &vnode_hold_list; 347 goto try_nextlist; 348 } 349 mutex_exit(&vnode_free_list_lock); 350 return NULL; 351 } 352 353 /* Remove it from the freelist. */ 354 TAILQ_REMOVE(listhd, vp, v_freelist); 355 vp->v_freelisthd = NULL; 356 mutex_exit(&vnode_free_list_lock); 357 358 /* 359 * The vnode is still associated with a file system, so we must 360 * clean it out before reusing it. We need to add a reference 361 * before doing this. If the vnode gains another reference while 362 * being cleaned out then we lose - retry. 363 */ 364 atomic_inc_uint(&vp->v_usecount); 365 vclean(vp, DOCLOSE); 366 if (vp->v_usecount == 1) { 367 /* We're about to dirty it. */ 368 vp->v_iflag &= ~VI_CLEAN; 369 mutex_exit(&vp->v_interlock); 370 if (vp->v_type == VBLK || vp->v_type == VCHR) { 371 spec_node_destroy(vp); 372 } 373 vp->v_type = VNON; 374 } else { 375 /* 376 * Don't return to freelist - the holder of the last 377 * reference will destroy it. 378 */ 379 vrelel(vp, 0); /* releases vp->v_interlock */ 380 mutex_enter(&vnode_free_list_lock); 381 goto retry; 382 } 383 384 if (vp->v_data != NULL || vp->v_uobj.uo_npages != 0 || 385 !TAILQ_EMPTY(&vp->v_uobj.memq)) { 386 vpanic(vp, "cleaned vnode isn't"); 387 } 388 if (vp->v_numoutput != 0) { 389 vpanic(vp, "clean vnode has pending I/O's"); 390 } 391 if ((vp->v_iflag & VI_ONWORKLST) != 0) { 392 vpanic(vp, "clean vnode on syncer list"); 393 } 394 395 return vp; 396 } 397 398 /* 399 * Mark a mount point as busy, and gain a new reference to it. Used to 400 * prevent the file system from being unmounted during critical sections. 401 * 402 * => The caller must hold a pre-existing reference to the mount. 403 * => Will fail if the file system is being unmounted, or is unmounted. 404 */ 405 int 406 vfs_busy(struct mount *mp, struct mount **nextp) 407 { 408 409 KASSERT(mp->mnt_refcnt > 0); 410 411 if (__predict_false(!rw_tryenter(&mp->mnt_unmounting, RW_READER))) { 412 if (nextp != NULL) { 413 KASSERT(mutex_owned(&mountlist_lock)); 414 *nextp = CIRCLEQ_NEXT(mp, mnt_list); 415 } 416 return EBUSY; 417 } 418 if (__predict_false((mp->mnt_iflag & IMNT_GONE) != 0)) { 419 rw_exit(&mp->mnt_unmounting); 420 if (nextp != NULL) { 421 KASSERT(mutex_owned(&mountlist_lock)); 422 *nextp = CIRCLEQ_NEXT(mp, mnt_list); 423 } 424 return ENOENT; 425 } 426 if (nextp != NULL) { 427 mutex_exit(&mountlist_lock); 428 } 429 atomic_inc_uint(&mp->mnt_refcnt); 430 return 0; 431 } 432 433 /* 434 * Unbusy a busy filesystem. 435 * 436 * => If keepref is true, preserve reference added by vfs_busy(). 437 * => If nextp != NULL, acquire mountlist_lock. 438 */ 439 void 440 vfs_unbusy(struct mount *mp, bool keepref, struct mount **nextp) 441 { 442 443 KASSERT(mp->mnt_refcnt > 0); 444 445 if (nextp != NULL) { 446 mutex_enter(&mountlist_lock); 447 } 448 rw_exit(&mp->mnt_unmounting); 449 if (!keepref) { 450 vfs_destroy(mp); 451 } 452 if (nextp != NULL) { 453 KASSERT(mutex_owned(&mountlist_lock)); 454 *nextp = CIRCLEQ_NEXT(mp, mnt_list); 455 } 456 } 457 458 struct mount * 459 vfs_mountalloc(struct vfsops *vfsops, struct vnode *vp) 460 { 461 int error; 462 struct mount *mp; 463 464 mp = kmem_zalloc(sizeof(*mp), KM_SLEEP); 465 if (mp == NULL) 466 return NULL; 467 468 mp->mnt_op = vfsops; 469 mp->mnt_refcnt = 1; 470 TAILQ_INIT(&mp->mnt_vnodelist); 471 rw_init(&mp->mnt_unmounting); 472 mutex_init(&mp->mnt_renamelock, MUTEX_DEFAULT, IPL_NONE); 473 mutex_init(&mp->mnt_updating, MUTEX_DEFAULT, IPL_NONE); 474 error = vfs_busy(mp, NULL); 475 KASSERT(error == 0); 476 mp->mnt_vnodecovered = vp; 477 mount_initspecific(mp); 478 479 return mp; 480 } 481 482 /* 483 * Lookup a filesystem type, and if found allocate and initialize 484 * a mount structure for it. 485 * 486 * Devname is usually updated by mount(8) after booting. 487 */ 488 int 489 vfs_rootmountalloc(const char *fstypename, const char *devname, 490 struct mount **mpp) 491 { 492 struct vfsops *vfsp = NULL; 493 struct mount *mp; 494 495 mutex_enter(&vfs_list_lock); 496 LIST_FOREACH(vfsp, &vfs_list, vfs_list) 497 if (!strncmp(vfsp->vfs_name, fstypename, 498 sizeof(mp->mnt_stat.f_fstypename))) 499 break; 500 if (vfsp == NULL) { 501 mutex_exit(&vfs_list_lock); 502 return (ENODEV); 503 } 504 vfsp->vfs_refcount++; 505 mutex_exit(&vfs_list_lock); 506 507 if ((mp = vfs_mountalloc(vfsp, NULL)) == NULL) 508 return ENOMEM; 509 mp->mnt_flag = MNT_RDONLY; 510 (void)strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name, 511 sizeof(mp->mnt_stat.f_fstypename)); 512 mp->mnt_stat.f_mntonname[0] = '/'; 513 mp->mnt_stat.f_mntonname[1] = '\0'; 514 mp->mnt_stat.f_mntfromname[sizeof(mp->mnt_stat.f_mntfromname) - 1] = 515 '\0'; 516 (void)copystr(devname, mp->mnt_stat.f_mntfromname, 517 sizeof(mp->mnt_stat.f_mntfromname) - 1, 0); 518 *mpp = mp; 519 return (0); 520 } 521 522 /* 523 * Routines having to do with the management of the vnode table. 524 */ 525 extern int (**dead_vnodeop_p)(void *); 526 527 /* 528 * Return the next vnode from the free list. 529 */ 530 int 531 getnewvnode(enum vtagtype tag, struct mount *mp, int (**vops)(void *), 532 vnode_t **vpp) 533 { 534 struct uvm_object *uobj; 535 static int toggle; 536 vnode_t *vp; 537 int error = 0, tryalloc; 538 539 try_again: 540 if (mp != NULL) { 541 /* 542 * Mark filesystem busy while we're creating a 543 * vnode. If unmount is in progress, this will 544 * fail. 545 */ 546 error = vfs_busy(mp, NULL); 547 if (error) 548 return error; 549 } 550 551 /* 552 * We must choose whether to allocate a new vnode or recycle an 553 * existing one. The criterion for allocating a new one is that 554 * the total number of vnodes is less than the number desired or 555 * there are no vnodes on either free list. Generally we only 556 * want to recycle vnodes that have no buffers associated with 557 * them, so we look first on the vnode_free_list. If it is empty, 558 * we next consider vnodes with referencing buffers on the 559 * vnode_hold_list. The toggle ensures that half the time we 560 * will use a buffer from the vnode_hold_list, and half the time 561 * we will allocate a new one unless the list has grown to twice 562 * the desired size. We are reticent to recycle vnodes from the 563 * vnode_hold_list because we will lose the identity of all its 564 * referencing buffers. 565 */ 566 567 vp = NULL; 568 569 mutex_enter(&vnode_free_list_lock); 570 571 toggle ^= 1; 572 if (numvnodes > 2 * desiredvnodes) 573 toggle = 0; 574 575 tryalloc = numvnodes < desiredvnodes || 576 (TAILQ_FIRST(&vnode_free_list) == NULL && 577 (TAILQ_FIRST(&vnode_hold_list) == NULL || toggle)); 578 579 if (tryalloc) { 580 numvnodes++; 581 mutex_exit(&vnode_free_list_lock); 582 if ((vp = vnalloc(NULL)) == NULL) { 583 mutex_enter(&vnode_free_list_lock); 584 numvnodes--; 585 } else 586 vp->v_usecount = 1; 587 } 588 589 if (vp == NULL) { 590 vp = getcleanvnode(); 591 if (vp == NULL) { 592 if (mp != NULL) { 593 vfs_unbusy(mp, false, NULL); 594 } 595 if (tryalloc) { 596 printf("WARNING: unable to allocate new " 597 "vnode, retrying...\n"); 598 kpause("newvn", false, hz, NULL); 599 goto try_again; 600 } 601 tablefull("vnode", "increase kern.maxvnodes or NVNODE"); 602 *vpp = 0; 603 return (ENFILE); 604 } 605 vp->v_iflag = 0; 606 vp->v_vflag = 0; 607 vp->v_uflag = 0; 608 vp->v_socket = NULL; 609 } 610 611 KASSERT(vp->v_usecount == 1); 612 KASSERT(vp->v_freelisthd == NULL); 613 KASSERT(LIST_EMPTY(&vp->v_nclist)); 614 KASSERT(LIST_EMPTY(&vp->v_dnclist)); 615 616 vp->v_type = VNON; 617 vp->v_vnlock = &vp->v_lock; 618 vp->v_tag = tag; 619 vp->v_op = vops; 620 insmntque(vp, mp); 621 *vpp = vp; 622 vp->v_data = 0; 623 624 /* 625 * initialize uvm_object within vnode. 626 */ 627 628 uobj = &vp->v_uobj; 629 KASSERT(uobj->pgops == &uvm_vnodeops); 630 KASSERT(uobj->uo_npages == 0); 631 KASSERT(TAILQ_FIRST(&uobj->memq) == NULL); 632 vp->v_size = vp->v_writesize = VSIZENOTSET; 633 634 if (mp != NULL) { 635 if ((mp->mnt_iflag & IMNT_MPSAFE) != 0) 636 vp->v_vflag |= VV_MPSAFE; 637 vfs_unbusy(mp, true, NULL); 638 } 639 640 return (0); 641 } 642 643 /* 644 * This is really just the reverse of getnewvnode(). Needed for 645 * VFS_VGET functions who may need to push back a vnode in case 646 * of a locking race. 647 */ 648 void 649 ungetnewvnode(vnode_t *vp) 650 { 651 652 KASSERT(vp->v_usecount == 1); 653 KASSERT(vp->v_data == NULL); 654 KASSERT(vp->v_freelisthd == NULL); 655 656 mutex_enter(&vp->v_interlock); 657 vp->v_iflag |= VI_CLEAN; 658 vrelel(vp, 0); 659 } 660 661 /* 662 * Allocate a new, uninitialized vnode. If 'mp' is non-NULL, this is a 663 * marker vnode and we are prepared to wait for the allocation. 664 */ 665 vnode_t * 666 vnalloc(struct mount *mp) 667 { 668 vnode_t *vp; 669 670 vp = pool_cache_get(vnode_cache, (mp != NULL ? PR_WAITOK : PR_NOWAIT)); 671 if (vp == NULL) { 672 return NULL; 673 } 674 675 memset(vp, 0, sizeof(*vp)); 676 UVM_OBJ_INIT(&vp->v_uobj, &uvm_vnodeops, 0); 677 cv_init(&vp->v_cv, "vnode"); 678 /* 679 * done by memset() above. 680 * LIST_INIT(&vp->v_nclist); 681 * LIST_INIT(&vp->v_dnclist); 682 */ 683 684 if (mp != NULL) { 685 vp->v_mount = mp; 686 vp->v_type = VBAD; 687 vp->v_iflag = VI_MARKER; 688 } else { 689 rw_init(&vp->v_lock.vl_lock); 690 } 691 692 return vp; 693 } 694 695 /* 696 * Free an unused, unreferenced vnode. 697 */ 698 void 699 vnfree(vnode_t *vp) 700 { 701 702 KASSERT(vp->v_usecount == 0); 703 704 if ((vp->v_iflag & VI_MARKER) == 0) { 705 rw_destroy(&vp->v_lock.vl_lock); 706 mutex_enter(&vnode_free_list_lock); 707 numvnodes--; 708 mutex_exit(&vnode_free_list_lock); 709 } 710 711 UVM_OBJ_DESTROY(&vp->v_uobj); 712 cv_destroy(&vp->v_cv); 713 pool_cache_put(vnode_cache, vp); 714 } 715 716 /* 717 * Remove a vnode from its freelist. 718 */ 719 static inline void 720 vremfree(vnode_t *vp) 721 { 722 723 KASSERT(mutex_owned(&vp->v_interlock)); 724 KASSERT(vp->v_usecount == 0); 725 726 /* 727 * Note that the reference count must not change until 728 * the vnode is removed. 729 */ 730 mutex_enter(&vnode_free_list_lock); 731 if (vp->v_holdcnt > 0) { 732 KASSERT(vp->v_freelisthd == &vnode_hold_list); 733 } else { 734 KASSERT(vp->v_freelisthd == &vnode_free_list); 735 } 736 TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist); 737 vp->v_freelisthd = NULL; 738 mutex_exit(&vnode_free_list_lock); 739 } 740 741 /* 742 * Move a vnode from one mount queue to another. 743 */ 744 static void 745 insmntque(vnode_t *vp, struct mount *mp) 746 { 747 struct mount *omp; 748 749 #ifdef DIAGNOSTIC 750 if ((mp != NULL) && 751 (mp->mnt_iflag & IMNT_UNMOUNT) && 752 vp->v_tag != VT_VFS) { 753 panic("insmntque into dying filesystem"); 754 } 755 #endif 756 757 mutex_enter(&mntvnode_lock); 758 /* 759 * Delete from old mount point vnode list, if on one. 760 */ 761 if ((omp = vp->v_mount) != NULL) 762 TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vp, v_mntvnodes); 763 /* 764 * Insert into list of vnodes for the new mount point, if 765 * available. The caller must take a reference on the mount 766 * structure and donate to the vnode. 767 */ 768 if ((vp->v_mount = mp) != NULL) 769 TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vp, v_mntvnodes); 770 mutex_exit(&mntvnode_lock); 771 772 if (omp != NULL) { 773 /* Release reference to old mount. */ 774 vfs_destroy(omp); 775 } 776 } 777 778 /* 779 * Wait for a vnode (typically with VI_XLOCK set) to be cleaned or 780 * recycled. 781 */ 782 void 783 vwait(vnode_t *vp, int flags) 784 { 785 786 KASSERT(mutex_owned(&vp->v_interlock)); 787 KASSERT(vp->v_usecount != 0); 788 789 while ((vp->v_iflag & flags) != 0) 790 cv_wait(&vp->v_cv, &vp->v_interlock); 791 } 792 793 /* 794 * Insert a marker vnode into a mount's vnode list, after the 795 * specified vnode. mntvnode_lock must be held. 796 */ 797 void 798 vmark(vnode_t *mvp, vnode_t *vp) 799 { 800 struct mount *mp; 801 802 mp = mvp->v_mount; 803 804 KASSERT(mutex_owned(&mntvnode_lock)); 805 KASSERT((mvp->v_iflag & VI_MARKER) != 0); 806 KASSERT(vp->v_mount == mp); 807 808 TAILQ_INSERT_AFTER(&mp->mnt_vnodelist, vp, mvp, v_mntvnodes); 809 } 810 811 /* 812 * Remove a marker vnode from a mount's vnode list, and return 813 * a pointer to the next vnode in the list. mntvnode_lock must 814 * be held. 815 */ 816 vnode_t * 817 vunmark(vnode_t *mvp) 818 { 819 vnode_t *vp; 820 struct mount *mp; 821 822 mp = mvp->v_mount; 823 824 KASSERT(mutex_owned(&mntvnode_lock)); 825 KASSERT((mvp->v_iflag & VI_MARKER) != 0); 826 827 vp = TAILQ_NEXT(mvp, v_mntvnodes); 828 TAILQ_REMOVE(&mp->mnt_vnodelist, mvp, v_mntvnodes); 829 830 KASSERT(vp == NULL || vp->v_mount == mp); 831 832 return vp; 833 } 834 835 /* 836 * Update outstanding I/O count and do wakeup if requested. 837 */ 838 void 839 vwakeup(struct buf *bp) 840 { 841 struct vnode *vp; 842 843 if ((vp = bp->b_vp) == NULL) 844 return; 845 846 KASSERT(bp->b_objlock == &vp->v_interlock); 847 KASSERT(mutex_owned(bp->b_objlock)); 848 849 if (--vp->v_numoutput < 0) 850 panic("vwakeup: neg numoutput, vp %p", vp); 851 if (vp->v_numoutput == 0) 852 cv_broadcast(&vp->v_cv); 853 } 854 855 /* 856 * Flush out and invalidate all buffers associated with a vnode. 857 * Called with the underlying vnode locked, which should prevent new dirty 858 * buffers from being queued. 859 */ 860 int 861 vinvalbuf(struct vnode *vp, int flags, kauth_cred_t cred, struct lwp *l, 862 bool catch, int slptimeo) 863 { 864 struct buf *bp, *nbp; 865 int error; 866 int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO | 867 (flags & V_SAVE ? PGO_CLEANIT | PGO_RECLAIM : 0); 868 869 /* XXXUBC this doesn't look at flags or slp* */ 870 mutex_enter(&vp->v_interlock); 871 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 872 if (error) { 873 return error; 874 } 875 876 if (flags & V_SAVE) { 877 error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0); 878 if (error) 879 return (error); 880 KASSERT(LIST_EMPTY(&vp->v_dirtyblkhd)); 881 } 882 883 mutex_enter(&bufcache_lock); 884 restart: 885 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 886 nbp = LIST_NEXT(bp, b_vnbufs); 887 error = bbusy(bp, catch, slptimeo, NULL); 888 if (error != 0) { 889 if (error == EPASSTHROUGH) 890 goto restart; 891 mutex_exit(&bufcache_lock); 892 return (error); 893 } 894 brelsel(bp, BC_INVAL | BC_VFLUSH); 895 } 896 897 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 898 nbp = LIST_NEXT(bp, b_vnbufs); 899 error = bbusy(bp, catch, slptimeo, NULL); 900 if (error != 0) { 901 if (error == EPASSTHROUGH) 902 goto restart; 903 mutex_exit(&bufcache_lock); 904 return (error); 905 } 906 /* 907 * XXX Since there are no node locks for NFS, I believe 908 * there is a slight chance that a delayed write will 909 * occur while sleeping just above, so check for it. 910 */ 911 if ((bp->b_oflags & BO_DELWRI) && (flags & V_SAVE)) { 912 #ifdef DEBUG 913 printf("buffer still DELWRI\n"); 914 #endif 915 bp->b_cflags |= BC_BUSY | BC_VFLUSH; 916 mutex_exit(&bufcache_lock); 917 VOP_BWRITE(bp); 918 mutex_enter(&bufcache_lock); 919 goto restart; 920 } 921 brelsel(bp, BC_INVAL | BC_VFLUSH); 922 } 923 924 #ifdef DIAGNOSTIC 925 if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd)) 926 panic("vinvalbuf: flush failed, vp %p", vp); 927 #endif 928 929 mutex_exit(&bufcache_lock); 930 931 return (0); 932 } 933 934 /* 935 * Destroy any in core blocks past the truncation length. 936 * Called with the underlying vnode locked, which should prevent new dirty 937 * buffers from being queued. 938 */ 939 int 940 vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo) 941 { 942 struct buf *bp, *nbp; 943 int error; 944 voff_t off; 945 946 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 947 mutex_enter(&vp->v_interlock); 948 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 949 if (error) { 950 return error; 951 } 952 953 mutex_enter(&bufcache_lock); 954 restart: 955 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 956 nbp = LIST_NEXT(bp, b_vnbufs); 957 if (bp->b_lblkno < lbn) 958 continue; 959 error = bbusy(bp, catch, slptimeo, NULL); 960 if (error != 0) { 961 if (error == EPASSTHROUGH) 962 goto restart; 963 mutex_exit(&bufcache_lock); 964 return (error); 965 } 966 brelsel(bp, BC_INVAL | BC_VFLUSH); 967 } 968 969 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 970 nbp = LIST_NEXT(bp, b_vnbufs); 971 if (bp->b_lblkno < lbn) 972 continue; 973 error = bbusy(bp, catch, slptimeo, NULL); 974 if (error != 0) { 975 if (error == EPASSTHROUGH) 976 goto restart; 977 mutex_exit(&bufcache_lock); 978 return (error); 979 } 980 brelsel(bp, BC_INVAL | BC_VFLUSH); 981 } 982 mutex_exit(&bufcache_lock); 983 984 return (0); 985 } 986 987 /* 988 * Flush all dirty buffers from a vnode. 989 * Called with the underlying vnode locked, which should prevent new dirty 990 * buffers from being queued. 991 */ 992 void 993 vflushbuf(struct vnode *vp, int sync) 994 { 995 struct buf *bp, *nbp; 996 int flags = PGO_CLEANIT | PGO_ALLPAGES | (sync ? PGO_SYNCIO : 0); 997 bool dirty; 998 999 mutex_enter(&vp->v_interlock); 1000 (void) VOP_PUTPAGES(vp, 0, 0, flags); 1001 1002 loop: 1003 mutex_enter(&bufcache_lock); 1004 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 1005 nbp = LIST_NEXT(bp, b_vnbufs); 1006 if ((bp->b_cflags & BC_BUSY)) 1007 continue; 1008 if ((bp->b_oflags & BO_DELWRI) == 0) 1009 panic("vflushbuf: not dirty, bp %p", bp); 1010 bp->b_cflags |= BC_BUSY | BC_VFLUSH; 1011 mutex_exit(&bufcache_lock); 1012 /* 1013 * Wait for I/O associated with indirect blocks to complete, 1014 * since there is no way to quickly wait for them below. 1015 */ 1016 if (bp->b_vp == vp || sync == 0) 1017 (void) bawrite(bp); 1018 else 1019 (void) bwrite(bp); 1020 goto loop; 1021 } 1022 mutex_exit(&bufcache_lock); 1023 1024 if (sync == 0) 1025 return; 1026 1027 mutex_enter(&vp->v_interlock); 1028 while (vp->v_numoutput != 0) 1029 cv_wait(&vp->v_cv, &vp->v_interlock); 1030 dirty = !LIST_EMPTY(&vp->v_dirtyblkhd); 1031 mutex_exit(&vp->v_interlock); 1032 1033 if (dirty) { 1034 vprint("vflushbuf: dirty", vp); 1035 goto loop; 1036 } 1037 } 1038 1039 /* 1040 * Create a vnode for a block device. 1041 * Used for root filesystem and swap areas. 1042 * Also used for memory file system special devices. 1043 */ 1044 int 1045 bdevvp(dev_t dev, vnode_t **vpp) 1046 { 1047 1048 return (getdevvp(dev, vpp, VBLK)); 1049 } 1050 1051 /* 1052 * Create a vnode for a character device. 1053 * Used for kernfs and some console handling. 1054 */ 1055 int 1056 cdevvp(dev_t dev, vnode_t **vpp) 1057 { 1058 1059 return (getdevvp(dev, vpp, VCHR)); 1060 } 1061 1062 /* 1063 * Associate a buffer with a vnode. There must already be a hold on 1064 * the vnode. 1065 */ 1066 void 1067 bgetvp(struct vnode *vp, struct buf *bp) 1068 { 1069 1070 KASSERT(bp->b_vp == NULL); 1071 KASSERT(bp->b_objlock == &buffer_lock); 1072 KASSERT(mutex_owned(&vp->v_interlock)); 1073 KASSERT(mutex_owned(&bufcache_lock)); 1074 KASSERT((bp->b_cflags & BC_BUSY) != 0); 1075 KASSERT(!cv_has_waiters(&bp->b_done)); 1076 1077 vholdl(vp); 1078 bp->b_vp = vp; 1079 if (vp->v_type == VBLK || vp->v_type == VCHR) 1080 bp->b_dev = vp->v_rdev; 1081 else 1082 bp->b_dev = NODEV; 1083 1084 /* 1085 * Insert onto list for new vnode. 1086 */ 1087 bufinsvn(bp, &vp->v_cleanblkhd); 1088 bp->b_objlock = &vp->v_interlock; 1089 } 1090 1091 /* 1092 * Disassociate a buffer from a vnode. 1093 */ 1094 void 1095 brelvp(struct buf *bp) 1096 { 1097 struct vnode *vp = bp->b_vp; 1098 1099 KASSERT(vp != NULL); 1100 KASSERT(bp->b_objlock == &vp->v_interlock); 1101 KASSERT(mutex_owned(&vp->v_interlock)); 1102 KASSERT(mutex_owned(&bufcache_lock)); 1103 KASSERT((bp->b_cflags & BC_BUSY) != 0); 1104 KASSERT(!cv_has_waiters(&bp->b_done)); 1105 1106 /* 1107 * Delete from old vnode list, if on one. 1108 */ 1109 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 1110 bufremvn(bp); 1111 1112 if (TAILQ_EMPTY(&vp->v_uobj.memq) && (vp->v_iflag & VI_ONWORKLST) && 1113 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 1114 vp->v_iflag &= ~VI_WRMAPDIRTY; 1115 vn_syncer_remove_from_worklist(vp); 1116 } 1117 1118 bp->b_objlock = &buffer_lock; 1119 bp->b_vp = NULL; 1120 holdrelel(vp); 1121 } 1122 1123 /* 1124 * Reassign a buffer from one vnode list to another. 1125 * The list reassignment must be within the same vnode. 1126 * Used to assign file specific control information 1127 * (indirect blocks) to the list to which they belong. 1128 */ 1129 void 1130 reassignbuf(struct buf *bp, struct vnode *vp) 1131 { 1132 struct buflists *listheadp; 1133 int delayx; 1134 1135 KASSERT(mutex_owned(&bufcache_lock)); 1136 KASSERT(bp->b_objlock == &vp->v_interlock); 1137 KASSERT(mutex_owned(&vp->v_interlock)); 1138 KASSERT((bp->b_cflags & BC_BUSY) != 0); 1139 1140 /* 1141 * Delete from old vnode list, if on one. 1142 */ 1143 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 1144 bufremvn(bp); 1145 1146 /* 1147 * If dirty, put on list of dirty buffers; 1148 * otherwise insert onto list of clean buffers. 1149 */ 1150 if ((bp->b_oflags & BO_DELWRI) == 0) { 1151 listheadp = &vp->v_cleanblkhd; 1152 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 1153 (vp->v_iflag & VI_ONWORKLST) && 1154 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 1155 vp->v_iflag &= ~VI_WRMAPDIRTY; 1156 vn_syncer_remove_from_worklist(vp); 1157 } 1158 } else { 1159 listheadp = &vp->v_dirtyblkhd; 1160 if ((vp->v_iflag & VI_ONWORKLST) == 0) { 1161 switch (vp->v_type) { 1162 case VDIR: 1163 delayx = dirdelay; 1164 break; 1165 case VBLK: 1166 if (vp->v_specmountpoint != NULL) { 1167 delayx = metadelay; 1168 break; 1169 } 1170 /* fall through */ 1171 default: 1172 delayx = filedelay; 1173 break; 1174 } 1175 if (!vp->v_mount || 1176 (vp->v_mount->mnt_flag & MNT_ASYNC) == 0) 1177 vn_syncer_add_to_worklist(vp, delayx); 1178 } 1179 } 1180 bufinsvn(bp, listheadp); 1181 } 1182 1183 /* 1184 * Create a vnode for a device. 1185 * Used by bdevvp (block device) for root file system etc., 1186 * and by cdevvp (character device) for console and kernfs. 1187 */ 1188 static int 1189 getdevvp(dev_t dev, vnode_t **vpp, enum vtype type) 1190 { 1191 vnode_t *vp; 1192 vnode_t *nvp; 1193 int error; 1194 1195 if (dev == NODEV) { 1196 *vpp = NULL; 1197 return (0); 1198 } 1199 error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp); 1200 if (error) { 1201 *vpp = NULL; 1202 return (error); 1203 } 1204 vp = nvp; 1205 vp->v_type = type; 1206 vp->v_vflag |= VV_MPSAFE; 1207 uvm_vnp_setsize(vp, 0); 1208 spec_node_init(vp, dev); 1209 *vpp = vp; 1210 return (0); 1211 } 1212 1213 /* 1214 * Try to gain a reference to a vnode, without acquiring its interlock. 1215 * The caller must hold a lock that will prevent the vnode from being 1216 * recycled or freed. 1217 */ 1218 bool 1219 vtryget(vnode_t *vp) 1220 { 1221 u_int use, next; 1222 1223 /* 1224 * If the vnode is being freed, don't make life any harder 1225 * for vclean() by adding another reference without waiting. 1226 * This is not strictly necessary, but we'll do it anyway. 1227 */ 1228 if (__predict_false((vp->v_iflag & (VI_XLOCK | VI_FREEING)) != 0)) { 1229 return false; 1230 } 1231 for (use = vp->v_usecount;; use = next) { 1232 if (use == 0) { 1233 /* Need interlock held if first reference. */ 1234 return false; 1235 } 1236 next = atomic_cas_uint(&vp->v_usecount, use, use + 1); 1237 if (__predict_true(next == use)) { 1238 return true; 1239 } 1240 } 1241 } 1242 1243 /* 1244 * Grab a particular vnode from the free list, increment its 1245 * reference count and lock it. If the vnode lock bit is set the 1246 * vnode is being eliminated in vgone. In that case, we can not 1247 * grab the vnode, so the process is awakened when the transition is 1248 * completed, and an error returned to indicate that the vnode is no 1249 * longer usable (possibly having been changed to a new file system type). 1250 */ 1251 int 1252 vget(vnode_t *vp, int flags) 1253 { 1254 int error; 1255 1256 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1257 1258 if ((flags & LK_INTERLOCK) == 0) 1259 mutex_enter(&vp->v_interlock); 1260 1261 /* 1262 * Before adding a reference, we must remove the vnode 1263 * from its freelist. 1264 */ 1265 if (vp->v_usecount == 0) { 1266 vremfree(vp); 1267 vp->v_usecount = 1; 1268 } else { 1269 atomic_inc_uint(&vp->v_usecount); 1270 } 1271 1272 /* 1273 * If the vnode is in the process of being cleaned out for 1274 * another use, we wait for the cleaning to finish and then 1275 * return failure. Cleaning is determined by checking if 1276 * the VI_XLOCK or VI_FREEING flags are set. 1277 */ 1278 if ((vp->v_iflag & (VI_XLOCK | VI_FREEING)) != 0) { 1279 if ((flags & LK_NOWAIT) != 0) { 1280 vrelel(vp, 0); 1281 return EBUSY; 1282 } 1283 vwait(vp, VI_XLOCK | VI_FREEING); 1284 vrelel(vp, 0); 1285 return ENOENT; 1286 } 1287 if (flags & LK_TYPE_MASK) { 1288 error = vn_lock(vp, flags | LK_INTERLOCK); 1289 if (error != 0) { 1290 vrele(vp); 1291 } 1292 return error; 1293 } 1294 mutex_exit(&vp->v_interlock); 1295 return 0; 1296 } 1297 1298 /* 1299 * vput(), just unlock and vrele() 1300 */ 1301 void 1302 vput(vnode_t *vp) 1303 { 1304 1305 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1306 1307 VOP_UNLOCK(vp, 0); 1308 vrele(vp); 1309 } 1310 1311 /* 1312 * Try to drop reference on a vnode. Abort if we are releasing the 1313 * last reference. Note: this _must_ succeed if not the last reference. 1314 */ 1315 static inline bool 1316 vtryrele(vnode_t *vp) 1317 { 1318 u_int use, next; 1319 1320 for (use = vp->v_usecount;; use = next) { 1321 if (use == 1) { 1322 return false; 1323 } 1324 next = atomic_cas_uint(&vp->v_usecount, use, use - 1); 1325 if (__predict_true(next == use)) { 1326 return true; 1327 } 1328 } 1329 } 1330 1331 /* 1332 * Vnode release. If reference count drops to zero, call inactive 1333 * routine and either return to freelist or free to the pool. 1334 */ 1335 void 1336 vrelel(vnode_t *vp, int flags) 1337 { 1338 bool recycle, defer; 1339 int error; 1340 1341 KASSERT(mutex_owned(&vp->v_interlock)); 1342 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1343 KASSERT(vp->v_freelisthd == NULL); 1344 1345 if (__predict_false(vp->v_op == dead_vnodeop_p && 1346 (vp->v_iflag & (VI_CLEAN|VI_XLOCK)) == 0)) { 1347 vpanic(vp, "dead but not clean"); 1348 } 1349 1350 /* 1351 * If not the last reference, just drop the reference count 1352 * and unlock. 1353 */ 1354 if (vtryrele(vp)) { 1355 vp->v_iflag |= VI_INACTREDO; 1356 mutex_exit(&vp->v_interlock); 1357 return; 1358 } 1359 if (vp->v_usecount <= 0 || vp->v_writecount != 0) { 1360 vpanic(vp, "vrelel: bad ref count"); 1361 } 1362 1363 KASSERT((vp->v_iflag & VI_XLOCK) == 0); 1364 1365 /* 1366 * If not clean, deactivate the vnode, but preserve 1367 * our reference across the call to VOP_INACTIVE(). 1368 */ 1369 retry: 1370 if ((vp->v_iflag & VI_CLEAN) == 0) { 1371 recycle = false; 1372 vp->v_iflag |= VI_INACTNOW; 1373 1374 /* 1375 * XXX This ugly block can be largely eliminated if 1376 * locking is pushed down into the file systems. 1377 */ 1378 if (curlwp == uvm.pagedaemon_lwp) { 1379 /* The pagedaemon can't wait around; defer. */ 1380 defer = true; 1381 } else if (curlwp == vrele_lwp) { 1382 /* We have to try harder. */ 1383 vp->v_iflag &= ~VI_INACTREDO; 1384 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK | 1385 LK_RETRY); 1386 if (error != 0) { 1387 /* XXX */ 1388 vpanic(vp, "vrele: unable to lock %p"); 1389 } 1390 defer = false; 1391 } else if ((vp->v_iflag & VI_LAYER) != 0) { 1392 /* 1393 * Acquiring the stack's lock in vclean() even 1394 * for an honest vput/vrele is dangerous because 1395 * our caller may hold other vnode locks; defer. 1396 */ 1397 defer = true; 1398 } else { 1399 /* If we can't acquire the lock, then defer. */ 1400 vp->v_iflag &= ~VI_INACTREDO; 1401 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK | 1402 LK_NOWAIT); 1403 if (error != 0) { 1404 defer = true; 1405 mutex_enter(&vp->v_interlock); 1406 } else { 1407 defer = false; 1408 } 1409 } 1410 1411 if (defer) { 1412 /* 1413 * Defer reclaim to the kthread; it's not safe to 1414 * clean it here. We donate it our last reference. 1415 */ 1416 KASSERT(mutex_owned(&vp->v_interlock)); 1417 KASSERT((vp->v_iflag & VI_INACTPEND) == 0); 1418 vp->v_iflag &= ~VI_INACTNOW; 1419 vp->v_iflag |= VI_INACTPEND; 1420 mutex_enter(&vrele_lock); 1421 TAILQ_INSERT_TAIL(&vrele_list, vp, v_freelist); 1422 if (++vrele_pending > (desiredvnodes >> 8)) 1423 cv_signal(&vrele_cv); 1424 mutex_exit(&vrele_lock); 1425 mutex_exit(&vp->v_interlock); 1426 return; 1427 } 1428 1429 #ifdef DIAGNOSTIC 1430 if ((vp->v_type == VBLK || vp->v_type == VCHR) && 1431 vp->v_specnode != NULL && vp->v_specnode->sn_opencnt != 0) { 1432 vprint("vrelel: missing VOP_CLOSE()", vp); 1433 } 1434 #endif 1435 1436 /* 1437 * The vnode can gain another reference while being 1438 * deactivated. If VOP_INACTIVE() indicates that 1439 * the described file has been deleted, then recycle 1440 * the vnode irrespective of additional references. 1441 * Another thread may be waiting to re-use the on-disk 1442 * inode. 1443 * 1444 * Note that VOP_INACTIVE() will drop the vnode lock. 1445 */ 1446 VOP_INACTIVE(vp, &recycle); 1447 mutex_enter(&vp->v_interlock); 1448 vp->v_iflag &= ~VI_INACTNOW; 1449 if (!recycle) { 1450 if (vtryrele(vp)) { 1451 mutex_exit(&vp->v_interlock); 1452 return; 1453 } 1454 1455 /* 1456 * If we grew another reference while 1457 * VOP_INACTIVE() was underway, retry. 1458 */ 1459 if ((vp->v_iflag & VI_INACTREDO) != 0) { 1460 goto retry; 1461 } 1462 } 1463 1464 /* Take care of space accounting. */ 1465 if (vp->v_iflag & VI_EXECMAP) { 1466 atomic_add_int(&uvmexp.execpages, 1467 -vp->v_uobj.uo_npages); 1468 atomic_add_int(&uvmexp.filepages, 1469 vp->v_uobj.uo_npages); 1470 } 1471 vp->v_iflag &= ~(VI_TEXT|VI_EXECMAP|VI_WRMAP); 1472 vp->v_vflag &= ~VV_MAPPED; 1473 1474 /* 1475 * Recycle the vnode if the file is now unused (unlinked), 1476 * otherwise just free it. 1477 */ 1478 if (recycle) { 1479 vclean(vp, DOCLOSE); 1480 } 1481 KASSERT(vp->v_usecount > 0); 1482 } 1483 1484 if (atomic_dec_uint_nv(&vp->v_usecount) != 0) { 1485 /* Gained another reference while being reclaimed. */ 1486 mutex_exit(&vp->v_interlock); 1487 return; 1488 } 1489 1490 if ((vp->v_iflag & VI_CLEAN) != 0) { 1491 /* 1492 * It's clean so destroy it. It isn't referenced 1493 * anywhere since it has been reclaimed. 1494 */ 1495 KASSERT(vp->v_holdcnt == 0); 1496 KASSERT(vp->v_writecount == 0); 1497 mutex_exit(&vp->v_interlock); 1498 insmntque(vp, NULL); 1499 if (vp->v_type == VBLK || vp->v_type == VCHR) { 1500 spec_node_destroy(vp); 1501 } 1502 vnfree(vp); 1503 } else { 1504 /* 1505 * Otherwise, put it back onto the freelist. It 1506 * can't be destroyed while still associated with 1507 * a file system. 1508 */ 1509 mutex_enter(&vnode_free_list_lock); 1510 if (vp->v_holdcnt > 0) { 1511 vp->v_freelisthd = &vnode_hold_list; 1512 } else { 1513 vp->v_freelisthd = &vnode_free_list; 1514 } 1515 TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist); 1516 mutex_exit(&vnode_free_list_lock); 1517 mutex_exit(&vp->v_interlock); 1518 } 1519 } 1520 1521 void 1522 vrele(vnode_t *vp) 1523 { 1524 1525 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1526 1527 if ((vp->v_iflag & VI_INACTNOW) == 0 && vtryrele(vp)) { 1528 return; 1529 } 1530 mutex_enter(&vp->v_interlock); 1531 vrelel(vp, 0); 1532 } 1533 1534 static void 1535 vrele_thread(void *cookie) 1536 { 1537 vnode_t *vp; 1538 1539 for (;;) { 1540 mutex_enter(&vrele_lock); 1541 while (TAILQ_EMPTY(&vrele_list)) { 1542 vrele_gen++; 1543 cv_broadcast(&vrele_cv); 1544 cv_timedwait(&vrele_cv, &vrele_lock, hz); 1545 } 1546 vp = TAILQ_FIRST(&vrele_list); 1547 TAILQ_REMOVE(&vrele_list, vp, v_freelist); 1548 vrele_pending--; 1549 mutex_exit(&vrele_lock); 1550 1551 /* 1552 * If not the last reference, then ignore the vnode 1553 * and look for more work. 1554 */ 1555 mutex_enter(&vp->v_interlock); 1556 KASSERT((vp->v_iflag & VI_INACTPEND) != 0); 1557 vp->v_iflag &= ~VI_INACTPEND; 1558 vrelel(vp, 0); 1559 } 1560 } 1561 1562 /* 1563 * Page or buffer structure gets a reference. 1564 * Called with v_interlock held. 1565 */ 1566 void 1567 vholdl(vnode_t *vp) 1568 { 1569 1570 KASSERT(mutex_owned(&vp->v_interlock)); 1571 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1572 1573 if (vp->v_holdcnt++ == 0 && vp->v_usecount == 0) { 1574 mutex_enter(&vnode_free_list_lock); 1575 KASSERT(vp->v_freelisthd == &vnode_free_list); 1576 TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist); 1577 vp->v_freelisthd = &vnode_hold_list; 1578 TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist); 1579 mutex_exit(&vnode_free_list_lock); 1580 } 1581 } 1582 1583 /* 1584 * Page or buffer structure frees a reference. 1585 * Called with v_interlock held. 1586 */ 1587 void 1588 holdrelel(vnode_t *vp) 1589 { 1590 1591 KASSERT(mutex_owned(&vp->v_interlock)); 1592 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1593 1594 if (vp->v_holdcnt <= 0) { 1595 vpanic(vp, "holdrelel: holdcnt vp %p"); 1596 } 1597 1598 vp->v_holdcnt--; 1599 if (vp->v_holdcnt == 0 && vp->v_usecount == 0) { 1600 mutex_enter(&vnode_free_list_lock); 1601 KASSERT(vp->v_freelisthd == &vnode_hold_list); 1602 TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist); 1603 vp->v_freelisthd = &vnode_free_list; 1604 TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist); 1605 mutex_exit(&vnode_free_list_lock); 1606 } 1607 } 1608 1609 /* 1610 * Vnode reference, where a reference is already held by some other 1611 * object (for example, a file structure). 1612 */ 1613 void 1614 vref(vnode_t *vp) 1615 { 1616 1617 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1618 KASSERT(vp->v_usecount != 0); 1619 1620 atomic_inc_uint(&vp->v_usecount); 1621 } 1622 1623 /* 1624 * Remove any vnodes in the vnode table belonging to mount point mp. 1625 * 1626 * If FORCECLOSE is not specified, there should not be any active ones, 1627 * return error if any are found (nb: this is a user error, not a 1628 * system error). If FORCECLOSE is specified, detach any active vnodes 1629 * that are found. 1630 * 1631 * If WRITECLOSE is set, only flush out regular file vnodes open for 1632 * writing. 1633 * 1634 * SKIPSYSTEM causes any vnodes marked V_SYSTEM to be skipped. 1635 */ 1636 #ifdef DEBUG 1637 int busyprt = 0; /* print out busy vnodes */ 1638 struct ctldebug debug1 = { "busyprt", &busyprt }; 1639 #endif 1640 1641 static vnode_t * 1642 vflushnext(vnode_t *mvp, int *when) 1643 { 1644 1645 if (hardclock_ticks > *when) { 1646 mutex_exit(&mntvnode_lock); 1647 yield(); 1648 mutex_enter(&mntvnode_lock); 1649 *when = hardclock_ticks + hz / 10; 1650 } 1651 1652 return vunmark(mvp); 1653 } 1654 1655 int 1656 vflush(struct mount *mp, vnode_t *skipvp, int flags) 1657 { 1658 vnode_t *vp, *mvp; 1659 int busy = 0, when = 0, gen; 1660 1661 /* 1662 * First, flush out any vnode references from vrele_list. 1663 */ 1664 mutex_enter(&vrele_lock); 1665 gen = vrele_gen; 1666 while (vrele_pending && gen == vrele_gen) { 1667 cv_broadcast(&vrele_cv); 1668 cv_wait(&vrele_cv, &vrele_lock); 1669 } 1670 mutex_exit(&vrele_lock); 1671 1672 /* Allocate a marker vnode. */ 1673 if ((mvp = vnalloc(mp)) == NULL) 1674 return (ENOMEM); 1675 1676 /* 1677 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone() 1678 * and vclean() are called 1679 */ 1680 mutex_enter(&mntvnode_lock); 1681 for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp != NULL; 1682 vp = vflushnext(mvp, &when)) { 1683 vmark(mvp, vp); 1684 if (vp->v_mount != mp || vismarker(vp)) 1685 continue; 1686 /* 1687 * Skip over a selected vnode. 1688 */ 1689 if (vp == skipvp) 1690 continue; 1691 mutex_enter(&vp->v_interlock); 1692 /* 1693 * Ignore clean but still referenced vnodes. 1694 */ 1695 if ((vp->v_iflag & VI_CLEAN) != 0) { 1696 mutex_exit(&vp->v_interlock); 1697 continue; 1698 } 1699 /* 1700 * Skip over a vnodes marked VSYSTEM. 1701 */ 1702 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { 1703 mutex_exit(&vp->v_interlock); 1704 continue; 1705 } 1706 /* 1707 * If WRITECLOSE is set, only flush out regular file 1708 * vnodes open for writing. 1709 */ 1710 if ((flags & WRITECLOSE) && 1711 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1712 mutex_exit(&vp->v_interlock); 1713 continue; 1714 } 1715 /* 1716 * With v_usecount == 0, all we need to do is clear 1717 * out the vnode data structures and we are done. 1718 */ 1719 if (vp->v_usecount == 0) { 1720 mutex_exit(&mntvnode_lock); 1721 vremfree(vp); 1722 vp->v_usecount = 1; 1723 vclean(vp, DOCLOSE); 1724 vrelel(vp, 0); 1725 mutex_enter(&mntvnode_lock); 1726 continue; 1727 } 1728 /* 1729 * If FORCECLOSE is set, forcibly close the vnode. 1730 * For block or character devices, revert to an 1731 * anonymous device. For all other files, just 1732 * kill them. 1733 */ 1734 if (flags & FORCECLOSE) { 1735 mutex_exit(&mntvnode_lock); 1736 atomic_inc_uint(&vp->v_usecount); 1737 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1738 vclean(vp, DOCLOSE); 1739 vrelel(vp, 0); 1740 } else { 1741 vclean(vp, 0); 1742 vp->v_op = spec_vnodeop_p; /* XXXSMP */ 1743 mutex_exit(&vp->v_interlock); 1744 /* 1745 * The vnode isn't clean, but still resides 1746 * on the mount list. Remove it. XXX This 1747 * is a bit dodgy. 1748 */ 1749 insmntque(vp, NULL); 1750 vrele(vp); 1751 } 1752 mutex_enter(&mntvnode_lock); 1753 continue; 1754 } 1755 #ifdef DEBUG 1756 if (busyprt) 1757 vprint("vflush: busy vnode", vp); 1758 #endif 1759 mutex_exit(&vp->v_interlock); 1760 busy++; 1761 } 1762 mutex_exit(&mntvnode_lock); 1763 vnfree(mvp); 1764 if (busy) 1765 return (EBUSY); 1766 return (0); 1767 } 1768 1769 /* 1770 * Disassociate the underlying file system from a vnode. 1771 * 1772 * Must be called with the interlock held, and will return with it held. 1773 */ 1774 void 1775 vclean(vnode_t *vp, int flags) 1776 { 1777 lwp_t *l = curlwp; 1778 bool recycle, active; 1779 int error; 1780 1781 KASSERT(mutex_owned(&vp->v_interlock)); 1782 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1783 KASSERT(vp->v_usecount != 0); 1784 1785 /* If cleaning is already in progress wait until done and return. */ 1786 if (vp->v_iflag & VI_XLOCK) { 1787 vwait(vp, VI_XLOCK); 1788 return; 1789 } 1790 1791 /* If already clean, nothing to do. */ 1792 if ((vp->v_iflag & VI_CLEAN) != 0) { 1793 return; 1794 } 1795 1796 /* 1797 * Prevent the vnode from being recycled or brought into use 1798 * while we clean it out. 1799 */ 1800 vp->v_iflag |= VI_XLOCK; 1801 if (vp->v_iflag & VI_EXECMAP) { 1802 atomic_add_int(&uvmexp.execpages, -vp->v_uobj.uo_npages); 1803 atomic_add_int(&uvmexp.filepages, vp->v_uobj.uo_npages); 1804 } 1805 vp->v_iflag &= ~(VI_TEXT|VI_EXECMAP); 1806 active = (vp->v_usecount > 1); 1807 1808 /* XXXAD should not lock vnode under layer */ 1809 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 1810 1811 /* 1812 * Clean out any cached data associated with the vnode. 1813 * If purging an active vnode, it must be closed and 1814 * deactivated before being reclaimed. Note that the 1815 * VOP_INACTIVE will unlock the vnode. 1816 */ 1817 if (flags & DOCLOSE) { 1818 error = vinvalbuf(vp, V_SAVE, NOCRED, l, 0, 0); 1819 if (error != 0) { 1820 /* XXX, fix vn_start_write's grab of mp and use that. */ 1821 1822 if (wapbl_vphaswapbl(vp)) 1823 WAPBL_DISCARD(wapbl_vptomp(vp)); 1824 error = vinvalbuf(vp, 0, NOCRED, l, 0, 0); 1825 } 1826 KASSERT(error == 0); 1827 KASSERT((vp->v_iflag & VI_ONWORKLST) == 0); 1828 if (active && (vp->v_type == VBLK || vp->v_type == VCHR)) { 1829 spec_node_revoke(vp); 1830 } 1831 } 1832 if (active) { 1833 VOP_INACTIVE(vp, &recycle); 1834 } else { 1835 /* 1836 * Any other processes trying to obtain this lock must first 1837 * wait for VI_XLOCK to clear, then call the new lock operation. 1838 */ 1839 VOP_UNLOCK(vp, 0); 1840 } 1841 1842 /* Disassociate the underlying file system from the vnode. */ 1843 if (VOP_RECLAIM(vp)) { 1844 vpanic(vp, "vclean: cannot reclaim"); 1845 } 1846 1847 KASSERT(vp->v_uobj.uo_npages == 0); 1848 if (vp->v_type == VREG && vp->v_ractx != NULL) { 1849 uvm_ra_freectx(vp->v_ractx); 1850 vp->v_ractx = NULL; 1851 } 1852 cache_purge(vp); 1853 1854 /* Done with purge, notify sleepers of the grim news. */ 1855 mutex_enter(&vp->v_interlock); 1856 vp->v_op = dead_vnodeop_p; 1857 vp->v_tag = VT_NON; 1858 vp->v_vnlock = &vp->v_lock; 1859 KNOTE(&vp->v_klist, NOTE_REVOKE); 1860 vp->v_iflag &= ~(VI_XLOCK | VI_FREEING); 1861 vp->v_vflag &= ~VV_LOCKSWORK; 1862 if ((flags & DOCLOSE) != 0) { 1863 vp->v_iflag |= VI_CLEAN; 1864 } 1865 cv_broadcast(&vp->v_cv); 1866 1867 KASSERT((vp->v_iflag & VI_ONWORKLST) == 0); 1868 } 1869 1870 /* 1871 * Recycle an unused vnode to the front of the free list. 1872 * Release the passed interlock if the vnode will be recycled. 1873 */ 1874 int 1875 vrecycle(vnode_t *vp, kmutex_t *inter_lkp, struct lwp *l) 1876 { 1877 1878 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1879 1880 mutex_enter(&vp->v_interlock); 1881 if (vp->v_usecount != 0) { 1882 mutex_exit(&vp->v_interlock); 1883 return (0); 1884 } 1885 if (inter_lkp) 1886 mutex_exit(inter_lkp); 1887 vremfree(vp); 1888 vp->v_usecount = 1; 1889 vclean(vp, DOCLOSE); 1890 vrelel(vp, 0); 1891 return (1); 1892 } 1893 1894 /* 1895 * Eliminate all activity associated with a vnode in preparation for 1896 * reuse. Drops a reference from the vnode. 1897 */ 1898 void 1899 vgone(vnode_t *vp) 1900 { 1901 1902 mutex_enter(&vp->v_interlock); 1903 vclean(vp, DOCLOSE); 1904 vrelel(vp, 0); 1905 } 1906 1907 /* 1908 * Lookup a vnode by device number. 1909 */ 1910 int 1911 vfinddev(dev_t dev, enum vtype type, vnode_t **vpp) 1912 { 1913 vnode_t *vp; 1914 int rc = 0; 1915 1916 mutex_enter(&device_lock); 1917 for (vp = specfs_hash[SPECHASH(dev)]; vp; vp = vp->v_specnext) { 1918 if (dev != vp->v_rdev || type != vp->v_type) 1919 continue; 1920 *vpp = vp; 1921 rc = 1; 1922 break; 1923 } 1924 mutex_exit(&device_lock); 1925 return (rc); 1926 } 1927 1928 /* 1929 * Revoke all the vnodes corresponding to the specified minor number 1930 * range (endpoints inclusive) of the specified major. 1931 */ 1932 void 1933 vdevgone(int maj, int minl, int minh, enum vtype type) 1934 { 1935 vnode_t *vp, **vpp; 1936 dev_t dev; 1937 int mn; 1938 1939 vp = NULL; /* XXX gcc */ 1940 1941 mutex_enter(&device_lock); 1942 for (mn = minl; mn <= minh; mn++) { 1943 dev = makedev(maj, mn); 1944 vpp = &specfs_hash[SPECHASH(dev)]; 1945 for (vp = *vpp; vp != NULL;) { 1946 mutex_enter(&vp->v_interlock); 1947 if ((vp->v_iflag & VI_CLEAN) != 0 || 1948 dev != vp->v_rdev || type != vp->v_type) { 1949 mutex_exit(&vp->v_interlock); 1950 vp = vp->v_specnext; 1951 continue; 1952 } 1953 mutex_exit(&device_lock); 1954 if (vget(vp, LK_INTERLOCK) == 0) { 1955 VOP_REVOKE(vp, REVOKEALL); 1956 vrele(vp); 1957 } 1958 mutex_enter(&device_lock); 1959 vp = *vpp; 1960 } 1961 } 1962 mutex_exit(&device_lock); 1963 } 1964 1965 /* 1966 * Calculate the total number of references to a special device. 1967 */ 1968 int 1969 vcount(vnode_t *vp) 1970 { 1971 int count; 1972 1973 mutex_enter(&device_lock); 1974 mutex_enter(&vp->v_interlock); 1975 if (vp->v_specnode == NULL) { 1976 count = vp->v_usecount - ((vp->v_iflag & VI_INACTPEND) != 0); 1977 mutex_exit(&vp->v_interlock); 1978 mutex_exit(&device_lock); 1979 return (count); 1980 } 1981 mutex_exit(&vp->v_interlock); 1982 count = vp->v_specnode->sn_dev->sd_opencnt; 1983 mutex_exit(&device_lock); 1984 return (count); 1985 } 1986 1987 /* 1988 * Eliminate all activity associated with the requested vnode 1989 * and with all vnodes aliased to the requested vnode. 1990 */ 1991 void 1992 vrevoke(vnode_t *vp) 1993 { 1994 vnode_t *vq, **vpp; 1995 enum vtype type; 1996 dev_t dev; 1997 1998 KASSERT(vp->v_usecount > 0); 1999 2000 mutex_enter(&vp->v_interlock); 2001 if ((vp->v_iflag & VI_CLEAN) != 0) { 2002 mutex_exit(&vp->v_interlock); 2003 return; 2004 } else if (vp->v_type != VBLK && vp->v_type != VCHR) { 2005 atomic_inc_uint(&vp->v_usecount); 2006 vclean(vp, DOCLOSE); 2007 vrelel(vp, 0); 2008 return; 2009 } else { 2010 dev = vp->v_rdev; 2011 type = vp->v_type; 2012 mutex_exit(&vp->v_interlock); 2013 } 2014 2015 vpp = &specfs_hash[SPECHASH(dev)]; 2016 mutex_enter(&device_lock); 2017 for (vq = *vpp; vq != NULL;) { 2018 /* If clean or being cleaned, then ignore it. */ 2019 mutex_enter(&vq->v_interlock); 2020 if ((vq->v_iflag & (VI_CLEAN | VI_XLOCK)) != 0 || 2021 vq->v_rdev != dev || vq->v_type != type) { 2022 mutex_exit(&vq->v_interlock); 2023 vq = vq->v_specnext; 2024 continue; 2025 } 2026 mutex_exit(&device_lock); 2027 if (vq->v_usecount == 0) { 2028 vremfree(vq); 2029 vq->v_usecount = 1; 2030 } else { 2031 atomic_inc_uint(&vq->v_usecount); 2032 } 2033 vclean(vq, DOCLOSE); 2034 vrelel(vq, 0); 2035 mutex_enter(&device_lock); 2036 vq = *vpp; 2037 } 2038 mutex_exit(&device_lock); 2039 } 2040 2041 /* 2042 * sysctl helper routine to return list of supported fstypes 2043 */ 2044 int 2045 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS) 2046 { 2047 char bf[sizeof(((struct statvfs *)NULL)->f_fstypename)]; 2048 char *where = oldp; 2049 struct vfsops *v; 2050 size_t needed, left, slen; 2051 int error, first; 2052 2053 if (newp != NULL) 2054 return (EPERM); 2055 if (namelen != 0) 2056 return (EINVAL); 2057 2058 first = 1; 2059 error = 0; 2060 needed = 0; 2061 left = *oldlenp; 2062 2063 sysctl_unlock(); 2064 mutex_enter(&vfs_list_lock); 2065 LIST_FOREACH(v, &vfs_list, vfs_list) { 2066 if (where == NULL) 2067 needed += strlen(v->vfs_name) + 1; 2068 else { 2069 memset(bf, 0, sizeof(bf)); 2070 if (first) { 2071 strncpy(bf, v->vfs_name, sizeof(bf)); 2072 first = 0; 2073 } else { 2074 bf[0] = ' '; 2075 strncpy(bf + 1, v->vfs_name, sizeof(bf) - 1); 2076 } 2077 bf[sizeof(bf)-1] = '\0'; 2078 slen = strlen(bf); 2079 if (left < slen + 1) 2080 break; 2081 v->vfs_refcount++; 2082 mutex_exit(&vfs_list_lock); 2083 /* +1 to copy out the trailing NUL byte */ 2084 error = copyout(bf, where, slen + 1); 2085 mutex_enter(&vfs_list_lock); 2086 v->vfs_refcount--; 2087 if (error) 2088 break; 2089 where += slen; 2090 needed += slen; 2091 left -= slen; 2092 } 2093 } 2094 mutex_exit(&vfs_list_lock); 2095 sysctl_relock(); 2096 *oldlenp = needed; 2097 return (error); 2098 } 2099 2100 2101 int kinfo_vdebug = 1; 2102 int kinfo_vgetfailed; 2103 #define KINFO_VNODESLOP 10 2104 /* 2105 * Dump vnode list (via sysctl). 2106 * Copyout address of vnode followed by vnode. 2107 */ 2108 /* ARGSUSED */ 2109 int 2110 sysctl_kern_vnode(SYSCTLFN_ARGS) 2111 { 2112 char *where = oldp; 2113 size_t *sizep = oldlenp; 2114 struct mount *mp, *nmp; 2115 vnode_t *vp, *mvp, vbuf; 2116 char *bp = where, *savebp; 2117 char *ewhere; 2118 int error; 2119 2120 if (namelen != 0) 2121 return (EOPNOTSUPP); 2122 if (newp != NULL) 2123 return (EPERM); 2124 2125 #define VPTRSZ sizeof(vnode_t *) 2126 #define VNODESZ sizeof(vnode_t) 2127 if (where == NULL) { 2128 *sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ); 2129 return (0); 2130 } 2131 ewhere = where + *sizep; 2132 2133 sysctl_unlock(); 2134 mutex_enter(&mountlist_lock); 2135 for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist; 2136 mp = nmp) { 2137 if (vfs_busy(mp, &nmp)) { 2138 continue; 2139 } 2140 savebp = bp; 2141 /* Allocate a marker vnode. */ 2142 mvp = vnalloc(mp); 2143 /* Should never fail for mp != NULL */ 2144 KASSERT(mvp != NULL); 2145 mutex_enter(&mntvnode_lock); 2146 for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = vunmark(mvp)) { 2147 vmark(mvp, vp); 2148 /* 2149 * Check that the vp is still associated with 2150 * this filesystem. RACE: could have been 2151 * recycled onto the same filesystem. 2152 */ 2153 if (vp->v_mount != mp || vismarker(vp)) 2154 continue; 2155 if (bp + VPTRSZ + VNODESZ > ewhere) { 2156 (void)vunmark(mvp); 2157 mutex_exit(&mntvnode_lock); 2158 vnfree(mvp); 2159 sysctl_relock(); 2160 *sizep = bp - where; 2161 return (ENOMEM); 2162 } 2163 memcpy(&vbuf, vp, VNODESZ); 2164 mutex_exit(&mntvnode_lock); 2165 if ((error = copyout(&vp, bp, VPTRSZ)) || 2166 (error = copyout(&vbuf, bp + VPTRSZ, VNODESZ))) { 2167 mutex_enter(&mntvnode_lock); 2168 (void)vunmark(mvp); 2169 mutex_exit(&mntvnode_lock); 2170 vnfree(mvp); 2171 sysctl_relock(); 2172 return (error); 2173 } 2174 bp += VPTRSZ + VNODESZ; 2175 mutex_enter(&mntvnode_lock); 2176 } 2177 mutex_exit(&mntvnode_lock); 2178 vnfree(mvp); 2179 vfs_unbusy(mp, false, &nmp); 2180 } 2181 mutex_exit(&mountlist_lock); 2182 sysctl_relock(); 2183 2184 *sizep = bp - where; 2185 return (0); 2186 } 2187 2188 /* 2189 * Remove clean vnodes from a mountpoint's vnode list. 2190 */ 2191 void 2192 vfs_scrubvnlist(struct mount *mp) 2193 { 2194 vnode_t *vp, *nvp; 2195 2196 retry: 2197 mutex_enter(&mntvnode_lock); 2198 for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) { 2199 nvp = TAILQ_NEXT(vp, v_mntvnodes); 2200 mutex_enter(&vp->v_interlock); 2201 if ((vp->v_iflag & VI_CLEAN) != 0) { 2202 TAILQ_REMOVE(&mp->mnt_vnodelist, vp, v_mntvnodes); 2203 vp->v_mount = NULL; 2204 mutex_exit(&mntvnode_lock); 2205 mutex_exit(&vp->v_interlock); 2206 vfs_destroy(mp); 2207 goto retry; 2208 } 2209 mutex_exit(&vp->v_interlock); 2210 } 2211 mutex_exit(&mntvnode_lock); 2212 } 2213 2214 /* 2215 * Check to see if a filesystem is mounted on a block device. 2216 */ 2217 int 2218 vfs_mountedon(vnode_t *vp) 2219 { 2220 vnode_t *vq; 2221 int error = 0; 2222 2223 if (vp->v_type != VBLK) 2224 return ENOTBLK; 2225 if (vp->v_specmountpoint != NULL) 2226 return (EBUSY); 2227 mutex_enter(&device_lock); 2228 for (vq = specfs_hash[SPECHASH(vp->v_rdev)]; vq != NULL; 2229 vq = vq->v_specnext) { 2230 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) 2231 continue; 2232 if (vq->v_specmountpoint != NULL) { 2233 error = EBUSY; 2234 break; 2235 } 2236 } 2237 mutex_exit(&device_lock); 2238 return (error); 2239 } 2240 2241 /* 2242 * Unmount all file systems. 2243 * We traverse the list in reverse order under the assumption that doing so 2244 * will avoid needing to worry about dependencies. 2245 */ 2246 bool 2247 vfs_unmountall(struct lwp *l) 2248 { 2249 printf("unmounting file systems..."); 2250 return vfs_unmountall1(l, true, true); 2251 } 2252 2253 bool 2254 vfs_unmountall1(struct lwp *l, bool force, bool verbose) 2255 { 2256 struct mount *mp, *nmp; 2257 bool any_error, progress; 2258 int error; 2259 2260 for (any_error = false, mp = CIRCLEQ_LAST(&mountlist); 2261 !CIRCLEQ_EMPTY(&mountlist); 2262 mp = nmp) { 2263 nmp = CIRCLEQ_PREV(mp, mnt_list); 2264 #ifdef DEBUG 2265 printf("\nunmounting %s (%s)...", 2266 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname); 2267 #endif 2268 atomic_inc_uint(&mp->mnt_refcnt); 2269 if ((error = dounmount(mp, force ? MNT_FORCE : 0, l)) == 0) 2270 progress = true; 2271 else { 2272 if (verbose) { 2273 printf("unmount of %s failed with error %d\n", 2274 mp->mnt_stat.f_mntonname, error); 2275 } 2276 any_error = true; 2277 } 2278 } 2279 if (verbose) 2280 printf(" done\n"); 2281 if (any_error && verbose) 2282 printf("WARNING: some file systems would not unmount\n"); 2283 return progress; 2284 } 2285 2286 /* 2287 * Sync and unmount file systems before shutting down. 2288 */ 2289 void 2290 vfs_shutdown(void) 2291 { 2292 struct lwp *l; 2293 2294 /* XXX we're certainly not running in lwp0's context! */ 2295 l = (curlwp == NULL) ? &lwp0 : curlwp; 2296 2297 printf("syncing disks... "); 2298 2299 /* remove user processes from run queue */ 2300 suspendsched(); 2301 (void) spl0(); 2302 2303 /* avoid coming back this way again if we panic. */ 2304 doing_shutdown = 1; 2305 2306 sys_sync(l, NULL, NULL); 2307 2308 /* Wait for sync to finish. */ 2309 if (buf_syncwait() != 0) { 2310 #if defined(DDB) && defined(DEBUG_HALT_BUSY) 2311 Debugger(); 2312 #endif 2313 printf("giving up\n"); 2314 return; 2315 } else 2316 printf("done\n"); 2317 2318 /* 2319 * If we've panic'd, don't make the situation potentially 2320 * worse by unmounting the file systems. 2321 */ 2322 if (panicstr != NULL) 2323 return; 2324 2325 /* Release inodes held by texts before update. */ 2326 #ifdef notdef 2327 vnshutdown(); 2328 #endif 2329 /* Unmount file systems. */ 2330 vfs_unmountall(l); 2331 } 2332 2333 /* 2334 * Mount the root file system. If the operator didn't specify a 2335 * file system to use, try all possible file systems until one 2336 * succeeds. 2337 */ 2338 int 2339 vfs_mountroot(void) 2340 { 2341 struct vfsops *v; 2342 int error = ENODEV; 2343 2344 if (root_device == NULL) 2345 panic("vfs_mountroot: root device unknown"); 2346 2347 switch (device_class(root_device)) { 2348 case DV_IFNET: 2349 if (rootdev != NODEV) 2350 panic("vfs_mountroot: rootdev set for DV_IFNET " 2351 "(0x%llx -> %llu,%llu)", 2352 (unsigned long long)rootdev, 2353 (unsigned long long)major(rootdev), 2354 (unsigned long long)minor(rootdev)); 2355 break; 2356 2357 case DV_DISK: 2358 if (rootdev == NODEV) 2359 panic("vfs_mountroot: rootdev not set for DV_DISK"); 2360 if (bdevvp(rootdev, &rootvp)) 2361 panic("vfs_mountroot: can't get vnode for rootdev"); 2362 error = VOP_OPEN(rootvp, FREAD, FSCRED); 2363 if (error) { 2364 printf("vfs_mountroot: can't open root device\n"); 2365 return (error); 2366 } 2367 break; 2368 2369 default: 2370 printf("%s: inappropriate for root file system\n", 2371 device_xname(root_device)); 2372 return (ENODEV); 2373 } 2374 2375 /* 2376 * If user specified a root fs type, use it. Make sure the 2377 * specified type exists and has a mount_root() 2378 */ 2379 if (strcmp(rootfstype, ROOT_FSTYPE_ANY) != 0) { 2380 v = vfs_getopsbyname(rootfstype); 2381 error = EFTYPE; 2382 if (v != NULL) { 2383 if (v->vfs_mountroot != NULL) { 2384 error = (v->vfs_mountroot)(); 2385 } 2386 v->vfs_refcount--; 2387 } 2388 goto done; 2389 } 2390 2391 /* 2392 * Try each file system currently configured into the kernel. 2393 */ 2394 mutex_enter(&vfs_list_lock); 2395 LIST_FOREACH(v, &vfs_list, vfs_list) { 2396 if (v->vfs_mountroot == NULL) 2397 continue; 2398 #ifdef DEBUG 2399 aprint_normal("mountroot: trying %s...\n", v->vfs_name); 2400 #endif 2401 v->vfs_refcount++; 2402 mutex_exit(&vfs_list_lock); 2403 error = (*v->vfs_mountroot)(); 2404 mutex_enter(&vfs_list_lock); 2405 v->vfs_refcount--; 2406 if (!error) { 2407 aprint_normal("root file system type: %s\n", 2408 v->vfs_name); 2409 break; 2410 } 2411 } 2412 mutex_exit(&vfs_list_lock); 2413 2414 if (v == NULL) { 2415 printf("no file system for %s", device_xname(root_device)); 2416 if (device_class(root_device) == DV_DISK) 2417 printf(" (dev 0x%llx)", (unsigned long long)rootdev); 2418 printf("\n"); 2419 error = EFTYPE; 2420 } 2421 2422 done: 2423 if (error && device_class(root_device) == DV_DISK) { 2424 VOP_CLOSE(rootvp, FREAD, FSCRED); 2425 vrele(rootvp); 2426 } 2427 return (error); 2428 } 2429 2430 /* 2431 * Get a new unique fsid 2432 */ 2433 void 2434 vfs_getnewfsid(struct mount *mp) 2435 { 2436 static u_short xxxfs_mntid; 2437 fsid_t tfsid; 2438 int mtype; 2439 2440 mutex_enter(&mntid_lock); 2441 mtype = makefstype(mp->mnt_op->vfs_name); 2442 mp->mnt_stat.f_fsidx.__fsid_val[0] = makedev(mtype, 0); 2443 mp->mnt_stat.f_fsidx.__fsid_val[1] = mtype; 2444 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 2445 if (xxxfs_mntid == 0) 2446 ++xxxfs_mntid; 2447 tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid); 2448 tfsid.__fsid_val[1] = mtype; 2449 if (!CIRCLEQ_EMPTY(&mountlist)) { 2450 while (vfs_getvfs(&tfsid)) { 2451 tfsid.__fsid_val[0]++; 2452 xxxfs_mntid++; 2453 } 2454 } 2455 mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0]; 2456 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 2457 mutex_exit(&mntid_lock); 2458 } 2459 2460 /* 2461 * Make a 'unique' number from a mount type name. 2462 */ 2463 long 2464 makefstype(const char *type) 2465 { 2466 long rv; 2467 2468 for (rv = 0; *type; type++) { 2469 rv <<= 2; 2470 rv ^= *type; 2471 } 2472 return rv; 2473 } 2474 2475 /* 2476 * Set vnode attributes to VNOVAL 2477 */ 2478 void 2479 vattr_null(struct vattr *vap) 2480 { 2481 2482 vap->va_type = VNON; 2483 2484 /* 2485 * Assign individually so that it is safe even if size and 2486 * sign of each member are varied. 2487 */ 2488 vap->va_mode = VNOVAL; 2489 vap->va_nlink = VNOVAL; 2490 vap->va_uid = VNOVAL; 2491 vap->va_gid = VNOVAL; 2492 vap->va_fsid = VNOVAL; 2493 vap->va_fileid = VNOVAL; 2494 vap->va_size = VNOVAL; 2495 vap->va_blocksize = VNOVAL; 2496 vap->va_atime.tv_sec = 2497 vap->va_mtime.tv_sec = 2498 vap->va_ctime.tv_sec = 2499 vap->va_birthtime.tv_sec = VNOVAL; 2500 vap->va_atime.tv_nsec = 2501 vap->va_mtime.tv_nsec = 2502 vap->va_ctime.tv_nsec = 2503 vap->va_birthtime.tv_nsec = VNOVAL; 2504 vap->va_gen = VNOVAL; 2505 vap->va_flags = VNOVAL; 2506 vap->va_rdev = VNOVAL; 2507 vap->va_bytes = VNOVAL; 2508 vap->va_vaflags = 0; 2509 } 2510 2511 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0])) 2512 #define ARRAY_PRINT(idx, arr) \ 2513 ((unsigned int)(idx) < ARRAY_SIZE(arr) ? (arr)[(idx)] : "UNKNOWN") 2514 2515 const char * const vnode_tags[] = { VNODE_TAGS }; 2516 const char * const vnode_types[] = { VNODE_TYPES }; 2517 const char vnode_flagbits[] = VNODE_FLAGBITS; 2518 2519 /* 2520 * Print out a description of a vnode. 2521 */ 2522 void 2523 vprint(const char *label, struct vnode *vp) 2524 { 2525 struct vnlock *vl; 2526 char bf[96]; 2527 int flag; 2528 2529 vl = (vp->v_vnlock != NULL ? vp->v_vnlock : &vp->v_lock); 2530 flag = vp->v_iflag | vp->v_vflag | vp->v_uflag; 2531 snprintb(bf, sizeof(bf), vnode_flagbits, flag); 2532 2533 if (label != NULL) 2534 printf("%s: ", label); 2535 printf("vnode @ %p, flags (%s)\n\ttag %s(%d), type %s(%d), " 2536 "usecount %d, writecount %d, holdcount %d\n" 2537 "\tfreelisthd %p, mount %p, data %p lock %p recursecnt %d\n", 2538 vp, bf, ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag, 2539 ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type, 2540 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, 2541 vp->v_freelisthd, vp->v_mount, vp->v_data, vl, vl->vl_recursecnt); 2542 if (vp->v_data != NULL) { 2543 printf("\t"); 2544 VOP_PRINT(vp); 2545 } 2546 } 2547 2548 #ifdef DEBUG 2549 /* 2550 * List all of the locked vnodes in the system. 2551 * Called when debugging the kernel. 2552 */ 2553 void 2554 printlockedvnodes(void) 2555 { 2556 struct mount *mp, *nmp; 2557 struct vnode *vp; 2558 2559 printf("Locked vnodes\n"); 2560 mutex_enter(&mountlist_lock); 2561 for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist; 2562 mp = nmp) { 2563 if (vfs_busy(mp, &nmp)) { 2564 continue; 2565 } 2566 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2567 if (VOP_ISLOCKED(vp)) 2568 vprint(NULL, vp); 2569 } 2570 mutex_enter(&mountlist_lock); 2571 vfs_unbusy(mp, false, &nmp); 2572 } 2573 mutex_exit(&mountlist_lock); 2574 } 2575 #endif 2576 2577 /* 2578 * Do the usual access checking. 2579 * file_mode, uid and gid are from the vnode in question, 2580 * while acc_mode and cred are from the VOP_ACCESS parameter list 2581 */ 2582 int 2583 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid, 2584 mode_t acc_mode, kauth_cred_t cred) 2585 { 2586 mode_t mask; 2587 int error, ismember; 2588 2589 /* 2590 * Super-user always gets read/write access, but execute access depends 2591 * on at least one execute bit being set. 2592 */ 2593 if (kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, NULL) == 0) { 2594 if ((acc_mode & VEXEC) && type != VDIR && 2595 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 2596 return (EACCES); 2597 return (0); 2598 } 2599 2600 mask = 0; 2601 2602 /* Otherwise, check the owner. */ 2603 if (kauth_cred_geteuid(cred) == uid) { 2604 if (acc_mode & VEXEC) 2605 mask |= S_IXUSR; 2606 if (acc_mode & VREAD) 2607 mask |= S_IRUSR; 2608 if (acc_mode & VWRITE) 2609 mask |= S_IWUSR; 2610 return ((file_mode & mask) == mask ? 0 : EACCES); 2611 } 2612 2613 /* Otherwise, check the groups. */ 2614 error = kauth_cred_ismember_gid(cred, gid, &ismember); 2615 if (error) 2616 return (error); 2617 if (kauth_cred_getegid(cred) == gid || ismember) { 2618 if (acc_mode & VEXEC) 2619 mask |= S_IXGRP; 2620 if (acc_mode & VREAD) 2621 mask |= S_IRGRP; 2622 if (acc_mode & VWRITE) 2623 mask |= S_IWGRP; 2624 return ((file_mode & mask) == mask ? 0 : EACCES); 2625 } 2626 2627 /* Otherwise, check everyone else. */ 2628 if (acc_mode & VEXEC) 2629 mask |= S_IXOTH; 2630 if (acc_mode & VREAD) 2631 mask |= S_IROTH; 2632 if (acc_mode & VWRITE) 2633 mask |= S_IWOTH; 2634 return ((file_mode & mask) == mask ? 0 : EACCES); 2635 } 2636 2637 /* 2638 * Given a file system name, look up the vfsops for that 2639 * file system, or return NULL if file system isn't present 2640 * in the kernel. 2641 */ 2642 struct vfsops * 2643 vfs_getopsbyname(const char *name) 2644 { 2645 struct vfsops *v; 2646 2647 mutex_enter(&vfs_list_lock); 2648 LIST_FOREACH(v, &vfs_list, vfs_list) { 2649 if (strcmp(v->vfs_name, name) == 0) 2650 break; 2651 } 2652 if (v != NULL) 2653 v->vfs_refcount++; 2654 mutex_exit(&vfs_list_lock); 2655 2656 return (v); 2657 } 2658 2659 void 2660 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp) 2661 { 2662 const struct statvfs *mbp; 2663 2664 if (sbp == (mbp = &mp->mnt_stat)) 2665 return; 2666 2667 (void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx)); 2668 sbp->f_fsid = mbp->f_fsid; 2669 sbp->f_owner = mbp->f_owner; 2670 sbp->f_flag = mbp->f_flag; 2671 sbp->f_syncwrites = mbp->f_syncwrites; 2672 sbp->f_asyncwrites = mbp->f_asyncwrites; 2673 sbp->f_syncreads = mbp->f_syncreads; 2674 sbp->f_asyncreads = mbp->f_asyncreads; 2675 (void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare)); 2676 (void)memcpy(sbp->f_fstypename, mbp->f_fstypename, 2677 sizeof(sbp->f_fstypename)); 2678 (void)memcpy(sbp->f_mntonname, mbp->f_mntonname, 2679 sizeof(sbp->f_mntonname)); 2680 (void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, 2681 sizeof(sbp->f_mntfromname)); 2682 sbp->f_namemax = mbp->f_namemax; 2683 } 2684 2685 int 2686 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom, 2687 const char *vfsname, struct mount *mp, struct lwp *l) 2688 { 2689 int error; 2690 size_t size; 2691 struct statvfs *sfs = &mp->mnt_stat; 2692 int (*fun)(const void *, void *, size_t, size_t *); 2693 2694 (void)strlcpy(mp->mnt_stat.f_fstypename, vfsname, 2695 sizeof(mp->mnt_stat.f_fstypename)); 2696 2697 if (onp) { 2698 struct cwdinfo *cwdi = l->l_proc->p_cwdi; 2699 fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr; 2700 if (cwdi->cwdi_rdir != NULL) { 2701 size_t len; 2702 char *bp; 2703 char *path = PNBUF_GET(); 2704 2705 bp = path + MAXPATHLEN; 2706 *--bp = '\0'; 2707 rw_enter(&cwdi->cwdi_lock, RW_READER); 2708 error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp, 2709 path, MAXPATHLEN / 2, 0, l); 2710 rw_exit(&cwdi->cwdi_lock); 2711 if (error) { 2712 PNBUF_PUT(path); 2713 return error; 2714 } 2715 2716 len = strlen(bp); 2717 if (len > sizeof(sfs->f_mntonname) - 1) 2718 len = sizeof(sfs->f_mntonname) - 1; 2719 (void)strncpy(sfs->f_mntonname, bp, len); 2720 PNBUF_PUT(path); 2721 2722 if (len < sizeof(sfs->f_mntonname) - 1) { 2723 error = (*fun)(onp, &sfs->f_mntonname[len], 2724 sizeof(sfs->f_mntonname) - len - 1, &size); 2725 if (error) 2726 return error; 2727 size += len; 2728 } else { 2729 size = len; 2730 } 2731 } else { 2732 error = (*fun)(onp, &sfs->f_mntonname, 2733 sizeof(sfs->f_mntonname) - 1, &size); 2734 if (error) 2735 return error; 2736 } 2737 (void)memset(sfs->f_mntonname + size, 0, 2738 sizeof(sfs->f_mntonname) - size); 2739 } 2740 2741 if (fromp) { 2742 fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr; 2743 error = (*fun)(fromp, sfs->f_mntfromname, 2744 sizeof(sfs->f_mntfromname) - 1, &size); 2745 if (error) 2746 return error; 2747 (void)memset(sfs->f_mntfromname + size, 0, 2748 sizeof(sfs->f_mntfromname) - size); 2749 } 2750 return 0; 2751 } 2752 2753 void 2754 vfs_timestamp(struct timespec *ts) 2755 { 2756 2757 nanotime(ts); 2758 } 2759 2760 time_t rootfstime; /* recorded root fs time, if known */ 2761 void 2762 setrootfstime(time_t t) 2763 { 2764 rootfstime = t; 2765 } 2766 2767 /* 2768 * Sham lock manager for vnodes. This is a temporary measure. 2769 */ 2770 int 2771 vlockmgr(struct vnlock *vl, int flags) 2772 { 2773 2774 KASSERT((flags & ~(LK_CANRECURSE | LK_NOWAIT | LK_TYPE_MASK)) == 0); 2775 2776 switch (flags & LK_TYPE_MASK) { 2777 case LK_SHARED: 2778 if (rw_tryenter(&vl->vl_lock, RW_READER)) { 2779 return 0; 2780 } 2781 if ((flags & LK_NOWAIT) != 0) { 2782 return EBUSY; 2783 } 2784 rw_enter(&vl->vl_lock, RW_READER); 2785 return 0; 2786 2787 case LK_EXCLUSIVE: 2788 if (rw_tryenter(&vl->vl_lock, RW_WRITER)) { 2789 return 0; 2790 } 2791 if ((vl->vl_canrecurse || (flags & LK_CANRECURSE) != 0) && 2792 rw_write_held(&vl->vl_lock)) { 2793 vl->vl_recursecnt++; 2794 return 0; 2795 } 2796 if ((flags & LK_NOWAIT) != 0) { 2797 return EBUSY; 2798 } 2799 rw_enter(&vl->vl_lock, RW_WRITER); 2800 return 0; 2801 2802 case LK_RELEASE: 2803 if (vl->vl_recursecnt != 0) { 2804 KASSERT(rw_write_held(&vl->vl_lock)); 2805 vl->vl_recursecnt--; 2806 return 0; 2807 } 2808 rw_exit(&vl->vl_lock); 2809 return 0; 2810 2811 default: 2812 panic("vlockmgr: flags %x", flags); 2813 } 2814 } 2815 2816 int 2817 vlockstatus(struct vnlock *vl) 2818 { 2819 2820 if (rw_write_held(&vl->vl_lock)) { 2821 return LK_EXCLUSIVE; 2822 } 2823 if (rw_read_held(&vl->vl_lock)) { 2824 return LK_SHARED; 2825 } 2826 return 0; 2827 } 2828 2829 /* 2830 * mount_specific_key_create -- 2831 * Create a key for subsystem mount-specific data. 2832 */ 2833 int 2834 mount_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor) 2835 { 2836 2837 return (specificdata_key_create(mount_specificdata_domain, keyp, dtor)); 2838 } 2839 2840 /* 2841 * mount_specific_key_delete -- 2842 * Delete a key for subsystem mount-specific data. 2843 */ 2844 void 2845 mount_specific_key_delete(specificdata_key_t key) 2846 { 2847 2848 specificdata_key_delete(mount_specificdata_domain, key); 2849 } 2850 2851 /* 2852 * mount_initspecific -- 2853 * Initialize a mount's specificdata container. 2854 */ 2855 void 2856 mount_initspecific(struct mount *mp) 2857 { 2858 int error; 2859 2860 error = specificdata_init(mount_specificdata_domain, 2861 &mp->mnt_specdataref); 2862 KASSERT(error == 0); 2863 } 2864 2865 /* 2866 * mount_finispecific -- 2867 * Finalize a mount's specificdata container. 2868 */ 2869 void 2870 mount_finispecific(struct mount *mp) 2871 { 2872 2873 specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref); 2874 } 2875 2876 /* 2877 * mount_getspecific -- 2878 * Return mount-specific data corresponding to the specified key. 2879 */ 2880 void * 2881 mount_getspecific(struct mount *mp, specificdata_key_t key) 2882 { 2883 2884 return (specificdata_getspecific(mount_specificdata_domain, 2885 &mp->mnt_specdataref, key)); 2886 } 2887 2888 /* 2889 * mount_setspecific -- 2890 * Set mount-specific data corresponding to the specified key. 2891 */ 2892 void 2893 mount_setspecific(struct mount *mp, specificdata_key_t key, void *data) 2894 { 2895 2896 specificdata_setspecific(mount_specificdata_domain, 2897 &mp->mnt_specdataref, key, data); 2898 } 2899 2900 int 2901 VFS_MOUNT(struct mount *mp, const char *a, void *b, size_t *c) 2902 { 2903 int error; 2904 2905 KERNEL_LOCK(1, NULL); 2906 error = (*(mp->mnt_op->vfs_mount))(mp, a, b, c); 2907 KERNEL_UNLOCK_ONE(NULL); 2908 2909 return error; 2910 } 2911 2912 int 2913 VFS_START(struct mount *mp, int a) 2914 { 2915 int error; 2916 2917 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 2918 KERNEL_LOCK(1, NULL); 2919 } 2920 error = (*(mp->mnt_op->vfs_start))(mp, a); 2921 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 2922 KERNEL_UNLOCK_ONE(NULL); 2923 } 2924 2925 return error; 2926 } 2927 2928 int 2929 VFS_UNMOUNT(struct mount *mp, int a) 2930 { 2931 int error; 2932 2933 KERNEL_LOCK(1, NULL); 2934 error = (*(mp->mnt_op->vfs_unmount))(mp, a); 2935 KERNEL_UNLOCK_ONE(NULL); 2936 2937 return error; 2938 } 2939 2940 int 2941 VFS_ROOT(struct mount *mp, struct vnode **a) 2942 { 2943 int error; 2944 2945 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 2946 KERNEL_LOCK(1, NULL); 2947 } 2948 error = (*(mp->mnt_op->vfs_root))(mp, a); 2949 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 2950 KERNEL_UNLOCK_ONE(NULL); 2951 } 2952 2953 return error; 2954 } 2955 2956 int 2957 VFS_QUOTACTL(struct mount *mp, int a, uid_t b, void *c) 2958 { 2959 int error; 2960 2961 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 2962 KERNEL_LOCK(1, NULL); 2963 } 2964 error = (*(mp->mnt_op->vfs_quotactl))(mp, a, b, c); 2965 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 2966 KERNEL_UNLOCK_ONE(NULL); 2967 } 2968 2969 return error; 2970 } 2971 2972 int 2973 VFS_STATVFS(struct mount *mp, struct statvfs *a) 2974 { 2975 int error; 2976 2977 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 2978 KERNEL_LOCK(1, NULL); 2979 } 2980 error = (*(mp->mnt_op->vfs_statvfs))(mp, a); 2981 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 2982 KERNEL_UNLOCK_ONE(NULL); 2983 } 2984 2985 return error; 2986 } 2987 2988 int 2989 VFS_SYNC(struct mount *mp, int a, struct kauth_cred *b) 2990 { 2991 int error; 2992 2993 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 2994 KERNEL_LOCK(1, NULL); 2995 } 2996 error = (*(mp->mnt_op->vfs_sync))(mp, a, b); 2997 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 2998 KERNEL_UNLOCK_ONE(NULL); 2999 } 3000 3001 return error; 3002 } 3003 3004 int 3005 VFS_FHTOVP(struct mount *mp, struct fid *a, struct vnode **b) 3006 { 3007 int error; 3008 3009 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3010 KERNEL_LOCK(1, NULL); 3011 } 3012 error = (*(mp->mnt_op->vfs_fhtovp))(mp, a, b); 3013 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3014 KERNEL_UNLOCK_ONE(NULL); 3015 } 3016 3017 return error; 3018 } 3019 3020 int 3021 VFS_VPTOFH(struct vnode *vp, struct fid *a, size_t *b) 3022 { 3023 int error; 3024 3025 if ((vp->v_vflag & VV_MPSAFE) == 0) { 3026 KERNEL_LOCK(1, NULL); 3027 } 3028 error = (*(vp->v_mount->mnt_op->vfs_vptofh))(vp, a, b); 3029 if ((vp->v_vflag & VV_MPSAFE) == 0) { 3030 KERNEL_UNLOCK_ONE(NULL); 3031 } 3032 3033 return error; 3034 } 3035 3036 int 3037 VFS_SNAPSHOT(struct mount *mp, struct vnode *a, struct timespec *b) 3038 { 3039 int error; 3040 3041 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3042 KERNEL_LOCK(1, NULL); 3043 } 3044 error = (*(mp->mnt_op->vfs_snapshot))(mp, a, b); 3045 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3046 KERNEL_UNLOCK_ONE(NULL); 3047 } 3048 3049 return error; 3050 } 3051 3052 int 3053 VFS_EXTATTRCTL(struct mount *mp, int a, struct vnode *b, int c, const char *d) 3054 { 3055 int error; 3056 3057 KERNEL_LOCK(1, NULL); /* XXXSMP check ffs */ 3058 error = (*(mp->mnt_op->vfs_extattrctl))(mp, a, b, c, d); 3059 KERNEL_UNLOCK_ONE(NULL); /* XXX */ 3060 3061 return error; 3062 } 3063 3064 int 3065 VFS_SUSPENDCTL(struct mount *mp, int a) 3066 { 3067 int error; 3068 3069 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3070 KERNEL_LOCK(1, NULL); 3071 } 3072 error = (*(mp->mnt_op->vfs_suspendctl))(mp, a); 3073 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3074 KERNEL_UNLOCK_ONE(NULL); 3075 } 3076 3077 return error; 3078 } 3079 3080 #if defined(DDB) || defined(DEBUGPRINT) 3081 static const char buf_flagbits[] = BUF_FLAGBITS; 3082 3083 void 3084 vfs_buf_print(struct buf *bp, int full, void (*pr)(const char *, ...)) 3085 { 3086 char bf[1024]; 3087 3088 (*pr)(" vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" rawblkno 0x%" 3089 PRIx64 " dev 0x%x\n", 3090 bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_rawblkno, bp->b_dev); 3091 3092 snprintb(bf, sizeof(bf), 3093 buf_flagbits, bp->b_flags | bp->b_oflags | bp->b_cflags); 3094 (*pr)(" error %d flags 0x%s\n", bp->b_error, bf); 3095 3096 (*pr)(" bufsize 0x%lx bcount 0x%lx resid 0x%lx\n", 3097 bp->b_bufsize, bp->b_bcount, bp->b_resid); 3098 (*pr)(" data %p saveaddr %p\n", 3099 bp->b_data, bp->b_saveaddr); 3100 (*pr)(" iodone %p objlock %p\n", bp->b_iodone, bp->b_objlock); 3101 } 3102 3103 3104 void 3105 vfs_vnode_print(struct vnode *vp, int full, void (*pr)(const char *, ...)) 3106 { 3107 char bf[256]; 3108 3109 uvm_object_printit(&vp->v_uobj, full, pr); 3110 snprintb(bf, sizeof(bf), 3111 vnode_flagbits, vp->v_iflag | vp->v_vflag | vp->v_uflag); 3112 (*pr)("\nVNODE flags %s\n", bf); 3113 (*pr)("mp %p numoutput %d size 0x%llx writesize 0x%llx\n", 3114 vp->v_mount, vp->v_numoutput, vp->v_size, vp->v_writesize); 3115 3116 (*pr)("data %p writecount %ld holdcnt %ld\n", 3117 vp->v_data, vp->v_writecount, vp->v_holdcnt); 3118 3119 (*pr)("tag %s(%d) type %s(%d) mount %p typedata %p\n", 3120 ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag, 3121 ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type, 3122 vp->v_mount, vp->v_mountedhere); 3123 3124 (*pr)("v_lock %p v_vnlock %p\n", &vp->v_lock, vp->v_vnlock); 3125 3126 if (full) { 3127 struct buf *bp; 3128 3129 (*pr)("clean bufs:\n"); 3130 LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) { 3131 (*pr)(" bp %p\n", bp); 3132 vfs_buf_print(bp, full, pr); 3133 } 3134 3135 (*pr)("dirty bufs:\n"); 3136 LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) { 3137 (*pr)(" bp %p\n", bp); 3138 vfs_buf_print(bp, full, pr); 3139 } 3140 } 3141 } 3142 3143 void 3144 vfs_mount_print(struct mount *mp, int full, void (*pr)(const char *, ...)) 3145 { 3146 char sbuf[256]; 3147 3148 (*pr)("vnodecovered = %p syncer = %p data = %p\n", 3149 mp->mnt_vnodecovered,mp->mnt_syncer,mp->mnt_data); 3150 3151 (*pr)("fs_bshift %d dev_bshift = %d\n", 3152 mp->mnt_fs_bshift,mp->mnt_dev_bshift); 3153 3154 snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_flag); 3155 (*pr)("flag = %s\n", sbuf); 3156 3157 snprintb(sbuf, sizeof(sbuf), __IMNT_FLAG_BITS, mp->mnt_iflag); 3158 (*pr)("iflag = %s\n", sbuf); 3159 3160 (*pr)("refcnt = %d unmounting @ %p updating @ %p\n", mp->mnt_refcnt, 3161 &mp->mnt_unmounting, &mp->mnt_updating); 3162 3163 (*pr)("statvfs cache:\n"); 3164 (*pr)("\tbsize = %lu\n",mp->mnt_stat.f_bsize); 3165 (*pr)("\tfrsize = %lu\n",mp->mnt_stat.f_frsize); 3166 (*pr)("\tiosize = %lu\n",mp->mnt_stat.f_iosize); 3167 3168 (*pr)("\tblocks = %"PRIu64"\n",mp->mnt_stat.f_blocks); 3169 (*pr)("\tbfree = %"PRIu64"\n",mp->mnt_stat.f_bfree); 3170 (*pr)("\tbavail = %"PRIu64"\n",mp->mnt_stat.f_bavail); 3171 (*pr)("\tbresvd = %"PRIu64"\n",mp->mnt_stat.f_bresvd); 3172 3173 (*pr)("\tfiles = %"PRIu64"\n",mp->mnt_stat.f_files); 3174 (*pr)("\tffree = %"PRIu64"\n",mp->mnt_stat.f_ffree); 3175 (*pr)("\tfavail = %"PRIu64"\n",mp->mnt_stat.f_favail); 3176 (*pr)("\tfresvd = %"PRIu64"\n",mp->mnt_stat.f_fresvd); 3177 3178 (*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n", 3179 mp->mnt_stat.f_fsidx.__fsid_val[0], 3180 mp->mnt_stat.f_fsidx.__fsid_val[1]); 3181 3182 (*pr)("\towner = %"PRIu32"\n",mp->mnt_stat.f_owner); 3183 (*pr)("\tnamemax = %lu\n",mp->mnt_stat.f_namemax); 3184 3185 snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_stat.f_flag); 3186 3187 (*pr)("\tflag = %s\n",sbuf); 3188 (*pr)("\tsyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_syncwrites); 3189 (*pr)("\tasyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_asyncwrites); 3190 (*pr)("\tsyncreads = %" PRIu64 "\n",mp->mnt_stat.f_syncreads); 3191 (*pr)("\tasyncreads = %" PRIu64 "\n",mp->mnt_stat.f_asyncreads); 3192 (*pr)("\tfstypename = %s\n",mp->mnt_stat.f_fstypename); 3193 (*pr)("\tmntonname = %s\n",mp->mnt_stat.f_mntonname); 3194 (*pr)("\tmntfromname = %s\n",mp->mnt_stat.f_mntfromname); 3195 3196 { 3197 int cnt = 0; 3198 struct vnode *vp; 3199 (*pr)("locked vnodes ="); 3200 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 3201 if (VOP_ISLOCKED(vp)) { 3202 if ((++cnt % 6) == 0) { 3203 (*pr)(" %p,\n\t", vp); 3204 } else { 3205 (*pr)(" %p,", vp); 3206 } 3207 } 3208 } 3209 (*pr)("\n"); 3210 } 3211 3212 if (full) { 3213 int cnt = 0; 3214 struct vnode *vp; 3215 (*pr)("all vnodes ="); 3216 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 3217 if (!TAILQ_NEXT(vp, v_mntvnodes)) { 3218 (*pr)(" %p", vp); 3219 } else if ((++cnt % 6) == 0) { 3220 (*pr)(" %p,\n\t", vp); 3221 } else { 3222 (*pr)(" %p,", vp); 3223 } 3224 } 3225 (*pr)("\n", vp); 3226 } 3227 } 3228 #endif /* DDB || DEBUGPRINT */ 3229 3230