1 /*- 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 35 */ 36 37 /* 38 * External virtual filesystem routines 39 */ 40 41 #include <sys/cdefs.h> 42 __FBSDID("$FreeBSD$"); 43 44 #include "opt_ddb.h" 45 #include "opt_mac.h" 46 47 #include <sys/param.h> 48 #include <sys/systm.h> 49 #include <sys/bio.h> 50 #include <sys/buf.h> 51 #include <sys/condvar.h> 52 #include <sys/conf.h> 53 #include <sys/dirent.h> 54 #include <sys/event.h> 55 #include <sys/eventhandler.h> 56 #include <sys/extattr.h> 57 #include <sys/file.h> 58 #include <sys/fcntl.h> 59 #include <sys/jail.h> 60 #include <sys/kdb.h> 61 #include <sys/kernel.h> 62 #include <sys/kthread.h> 63 #include <sys/lockf.h> 64 #include <sys/malloc.h> 65 #include <sys/mount.h> 66 #include <sys/namei.h> 67 #include <sys/priv.h> 68 #include <sys/reboot.h> 69 #include <sys/sleepqueue.h> 70 #include <sys/stat.h> 71 #include <sys/sysctl.h> 72 #include <sys/syslog.h> 73 #include <sys/vmmeter.h> 74 #include <sys/vnode.h> 75 76 #include <machine/stdarg.h> 77 78 #include <security/mac/mac_framework.h> 79 80 #include <vm/vm.h> 81 #include <vm/vm_object.h> 82 #include <vm/vm_extern.h> 83 #include <vm/pmap.h> 84 #include <vm/vm_map.h> 85 #include <vm/vm_page.h> 86 #include <vm/vm_kern.h> 87 #include <vm/uma.h> 88 89 #ifdef DDB 90 #include <ddb/ddb.h> 91 #endif 92 93 #define WI_MPSAFEQ 0 94 #define WI_GIANTQ 1 95 96 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure"); 97 98 static void delmntque(struct vnode *vp); 99 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, 100 int slpflag, int slptimeo); 101 static void syncer_shutdown(void *arg, int howto); 102 static int vtryrecycle(struct vnode *vp); 103 static void vbusy(struct vnode *vp); 104 static void vinactive(struct vnode *, struct thread *); 105 static void v_incr_usecount(struct vnode *); 106 static void v_decr_usecount(struct vnode *); 107 static void v_decr_useonly(struct vnode *); 108 static void v_upgrade_usecount(struct vnode *); 109 static void vfree(struct vnode *); 110 static void vnlru_free(int); 111 static void vgonel(struct vnode *); 112 static void vfs_knllock(void *arg); 113 static void vfs_knlunlock(void *arg); 114 static int vfs_knllocked(void *arg); 115 static void destroy_vpollinfo(struct vpollinfo *vi); 116 117 /* 118 * Enable Giant pushdown based on whether or not the vm is mpsafe in this 119 * build. Without mpsafevm the buffer cache can not run Giant free. 120 */ 121 int mpsafe_vfs = 1; 122 TUNABLE_INT("debug.mpsafevfs", &mpsafe_vfs); 123 SYSCTL_INT(_debug, OID_AUTO, mpsafevfs, CTLFLAG_RD, &mpsafe_vfs, 0, 124 "MPSAFE VFS"); 125 126 /* 127 * Number of vnodes in existence. Increased whenever getnewvnode() 128 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed 129 * vnode. 130 */ 131 static unsigned long numvnodes; 132 133 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 134 135 /* 136 * Conversion tables for conversion from vnode types to inode formats 137 * and back. 138 */ 139 enum vtype iftovt_tab[16] = { 140 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 141 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 142 }; 143 int vttoif_tab[10] = { 144 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 145 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT 146 }; 147 148 /* 149 * List of vnodes that are ready for recycling. 150 */ 151 static TAILQ_HEAD(freelst, vnode) vnode_free_list; 152 153 /* 154 * Free vnode target. Free vnodes may simply be files which have been stat'd 155 * but not read. This is somewhat common, and a small cache of such files 156 * should be kept to avoid recreation costs. 157 */ 158 static u_long wantfreevnodes; 159 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 160 /* Number of vnodes in the free list. */ 161 static u_long freevnodes; 162 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, ""); 163 164 /* 165 * Various variables used for debugging the new implementation of 166 * reassignbuf(). 167 * XXX these are probably of (very) limited utility now. 168 */ 169 static int reassignbufcalls; 170 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, ""); 171 172 /* 173 * Cache for the mount type id assigned to NFS. This is used for 174 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c. 175 */ 176 int nfs_mount_type = -1; 177 178 /* To keep more than one thread at a time from running vfs_getnewfsid */ 179 static struct mtx mntid_mtx; 180 181 /* 182 * Lock for any access to the following: 183 * vnode_free_list 184 * numvnodes 185 * freevnodes 186 */ 187 static struct mtx vnode_free_list_mtx; 188 189 /* Publicly exported FS */ 190 struct nfs_public nfs_pub; 191 192 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ 193 static uma_zone_t vnode_zone; 194 static uma_zone_t vnodepoll_zone; 195 196 /* Set to 1 to print out reclaim of active vnodes */ 197 int prtactive; 198 199 /* 200 * The workitem queue. 201 * 202 * It is useful to delay writes of file data and filesystem metadata 203 * for tens of seconds so that quickly created and deleted files need 204 * not waste disk bandwidth being created and removed. To realize this, 205 * we append vnodes to a "workitem" queue. When running with a soft 206 * updates implementation, most pending metadata dependencies should 207 * not wait for more than a few seconds. Thus, mounted on block devices 208 * are delayed only about a half the time that file data is delayed. 209 * Similarly, directory updates are more critical, so are only delayed 210 * about a third the time that file data is delayed. Thus, there are 211 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 212 * one each second (driven off the filesystem syncer process). The 213 * syncer_delayno variable indicates the next queue that is to be processed. 214 * Items that need to be processed soon are placed in this queue: 215 * 216 * syncer_workitem_pending[syncer_delayno] 217 * 218 * A delay of fifteen seconds is done by placing the request fifteen 219 * entries later in the queue: 220 * 221 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 222 * 223 */ 224 static int syncer_delayno; 225 static long syncer_mask; 226 LIST_HEAD(synclist, bufobj); 227 static struct synclist *syncer_workitem_pending[2]; 228 /* 229 * The sync_mtx protects: 230 * bo->bo_synclist 231 * sync_vnode_count 232 * syncer_delayno 233 * syncer_state 234 * syncer_workitem_pending 235 * syncer_worklist_len 236 * rushjob 237 */ 238 static struct mtx sync_mtx; 239 static struct cv sync_wakeup; 240 241 #define SYNCER_MAXDELAY 32 242 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 243 static int syncdelay = 30; /* max time to delay syncing data */ 244 static int filedelay = 30; /* time to delay syncing files */ 245 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, ""); 246 static int dirdelay = 29; /* time to delay syncing directories */ 247 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, ""); 248 static int metadelay = 28; /* time to delay syncing metadata */ 249 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, ""); 250 static int rushjob; /* number of slots to run ASAP */ 251 static int stat_rush_requests; /* number of times I/O speeded up */ 252 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, ""); 253 254 /* 255 * When shutting down the syncer, run it at four times normal speed. 256 */ 257 #define SYNCER_SHUTDOWN_SPEEDUP 4 258 static int sync_vnode_count; 259 static int syncer_worklist_len; 260 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } 261 syncer_state; 262 263 /* 264 * Number of vnodes we want to exist at any one time. This is mostly used 265 * to size hash tables in vnode-related code. It is normally not used in 266 * getnewvnode(), as wantfreevnodes is normally nonzero.) 267 * 268 * XXX desiredvnodes is historical cruft and should not exist. 269 */ 270 int desiredvnodes; 271 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 272 &desiredvnodes, 0, "Maximum number of vnodes"); 273 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, 274 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)"); 275 static int vnlru_nowhere; 276 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, 277 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); 278 279 /* 280 * Macros to control when a vnode is freed and recycled. All require 281 * the vnode interlock. 282 */ 283 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt) 284 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt) 285 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt) 286 287 288 /* 289 * Initialize the vnode management data structures. 290 */ 291 #ifndef MAXVNODES_MAX 292 #define MAXVNODES_MAX 100000 293 #endif 294 static void 295 vntblinit(void *dummy __unused) 296 { 297 298 /* 299 * Desiredvnodes is a function of the physical memory size and 300 * the kernel's heap size. Specifically, desiredvnodes scales 301 * in proportion to the physical memory size until two fifths 302 * of the kernel's heap size is consumed by vnodes and vm 303 * objects. 304 */ 305 desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size / 306 (5 * (sizeof(struct vm_object) + sizeof(struct vnode)))); 307 if (desiredvnodes > MAXVNODES_MAX) { 308 if (bootverbose) 309 printf("Reducing kern.maxvnodes %d -> %d\n", 310 desiredvnodes, MAXVNODES_MAX); 311 desiredvnodes = MAXVNODES_MAX; 312 } 313 wantfreevnodes = desiredvnodes / 4; 314 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); 315 TAILQ_INIT(&vnode_free_list); 316 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF); 317 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, 318 NULL, NULL, UMA_ALIGN_PTR, 0); 319 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), 320 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 321 /* 322 * Initialize the filesystem syncer. 323 */ 324 syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE, 325 &syncer_mask); 326 syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE, 327 &syncer_mask); 328 syncer_maxdelay = syncer_mask + 1; 329 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); 330 cv_init(&sync_wakeup, "syncer"); 331 } 332 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL); 333 334 335 /* 336 * Mark a mount point as busy. Used to synchronize access and to delay 337 * unmounting. Eventually, mountlist_mtx is not released on failure. 338 */ 339 int 340 vfs_busy(struct mount *mp, int flags) 341 { 342 343 MPASS((flags & ~MBF_MASK) == 0); 344 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags); 345 346 MNT_ILOCK(mp); 347 MNT_REF(mp); 348 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 349 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) { 350 MNT_REL(mp); 351 MNT_IUNLOCK(mp); 352 CTR1(KTR_VFS, "%s: failed busying before sleeping", 353 __func__); 354 return (ENOENT); 355 } 356 if (flags & MBF_MNTLSTLOCK) 357 mtx_unlock(&mountlist_mtx); 358 mp->mnt_kern_flag |= MNTK_MWAIT; 359 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0); 360 MNT_REL(mp); 361 MNT_IUNLOCK(mp); 362 if (flags & MBF_MNTLSTLOCK) 363 mtx_lock(&mountlist_mtx); 364 CTR1(KTR_VFS, "%s: failed busying after sleep", __func__); 365 return (ENOENT); 366 } 367 if (flags & MBF_MNTLSTLOCK) 368 mtx_unlock(&mountlist_mtx); 369 mp->mnt_lockref++; 370 MNT_IUNLOCK(mp); 371 return (0); 372 } 373 374 /* 375 * Free a busy filesystem. 376 */ 377 void 378 vfs_unbusy(struct mount *mp) 379 { 380 381 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 382 MNT_ILOCK(mp); 383 MNT_REL(mp); 384 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref")); 385 mp->mnt_lockref--; 386 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) { 387 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT); 388 CTR1(KTR_VFS, "%s: waking up waiters", __func__); 389 mp->mnt_kern_flag &= ~MNTK_DRAINING; 390 wakeup(&mp->mnt_lockref); 391 } 392 MNT_IUNLOCK(mp); 393 } 394 395 /* 396 * Lookup a mount point by filesystem identifier. 397 */ 398 struct mount * 399 vfs_getvfs(fsid_t *fsid) 400 { 401 struct mount *mp; 402 403 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); 404 mtx_lock(&mountlist_mtx); 405 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 406 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 407 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 408 vfs_ref(mp); 409 mtx_unlock(&mountlist_mtx); 410 return (mp); 411 } 412 } 413 mtx_unlock(&mountlist_mtx); 414 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); 415 return ((struct mount *) 0); 416 } 417 418 /* 419 * Lookup a mount point by filesystem identifier, busying it before 420 * returning. 421 */ 422 struct mount * 423 vfs_busyfs(fsid_t *fsid) 424 { 425 struct mount *mp; 426 int error; 427 428 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); 429 mtx_lock(&mountlist_mtx); 430 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 431 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 432 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 433 error = vfs_busy(mp, MBF_MNTLSTLOCK); 434 if (error) { 435 mtx_unlock(&mountlist_mtx); 436 return (NULL); 437 } 438 return (mp); 439 } 440 } 441 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); 442 mtx_unlock(&mountlist_mtx); 443 return ((struct mount *) 0); 444 } 445 446 /* 447 * Check if a user can access privileged mount options. 448 */ 449 int 450 vfs_suser(struct mount *mp, struct thread *td) 451 { 452 int error; 453 454 /* 455 * If the thread is jailed, but this is not a jail-friendly file 456 * system, deny immediately. 457 */ 458 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred)) 459 return (EPERM); 460 461 /* 462 * If the file system was mounted outside a jail and a jailed thread 463 * tries to access it, deny immediately. 464 */ 465 if (!jailed(mp->mnt_cred) && jailed(td->td_ucred)) 466 return (EPERM); 467 468 /* 469 * If the file system was mounted inside different jail that the jail of 470 * the calling thread, deny immediately. 471 */ 472 if (jailed(mp->mnt_cred) && jailed(td->td_ucred) && 473 mp->mnt_cred->cr_prison != td->td_ucred->cr_prison) { 474 return (EPERM); 475 } 476 477 /* 478 * If file system supports delegated administration, we don't check 479 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified 480 * by the file system itself. 481 * If this is not the user that did original mount, we check for 482 * the PRIV_VFS_MOUNT_OWNER privilege. 483 */ 484 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) && 485 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { 486 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) 487 return (error); 488 } 489 return (0); 490 } 491 492 /* 493 * Get a new unique fsid. Try to make its val[0] unique, since this value 494 * will be used to create fake device numbers for stat(). Also try (but 495 * not so hard) make its val[0] unique mod 2^16, since some emulators only 496 * support 16-bit device numbers. We end up with unique val[0]'s for the 497 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 498 * 499 * Keep in mind that several mounts may be running in parallel. Starting 500 * the search one past where the previous search terminated is both a 501 * micro-optimization and a defense against returning the same fsid to 502 * different mounts. 503 */ 504 void 505 vfs_getnewfsid(struct mount *mp) 506 { 507 static u_int16_t mntid_base; 508 struct mount *nmp; 509 fsid_t tfsid; 510 int mtype; 511 512 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 513 mtx_lock(&mntid_mtx); 514 mtype = mp->mnt_vfc->vfc_typenum; 515 tfsid.val[1] = mtype; 516 mtype = (mtype & 0xFF) << 24; 517 for (;;) { 518 tfsid.val[0] = makedev(255, 519 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 520 mntid_base++; 521 if ((nmp = vfs_getvfs(&tfsid)) == NULL) 522 break; 523 vfs_rel(nmp); 524 } 525 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 526 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 527 mtx_unlock(&mntid_mtx); 528 } 529 530 /* 531 * Knob to control the precision of file timestamps: 532 * 533 * 0 = seconds only; nanoseconds zeroed. 534 * 1 = seconds and nanoseconds, accurate within 1/HZ. 535 * 2 = seconds and nanoseconds, truncated to microseconds. 536 * >=3 = seconds and nanoseconds, maximum precision. 537 */ 538 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 539 540 static int timestamp_precision = TSP_SEC; 541 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 542 ×tamp_precision, 0, ""); 543 544 /* 545 * Get a current timestamp. 546 */ 547 void 548 vfs_timestamp(struct timespec *tsp) 549 { 550 struct timeval tv; 551 552 switch (timestamp_precision) { 553 case TSP_SEC: 554 tsp->tv_sec = time_second; 555 tsp->tv_nsec = 0; 556 break; 557 case TSP_HZ: 558 getnanotime(tsp); 559 break; 560 case TSP_USEC: 561 microtime(&tv); 562 TIMEVAL_TO_TIMESPEC(&tv, tsp); 563 break; 564 case TSP_NSEC: 565 default: 566 nanotime(tsp); 567 break; 568 } 569 } 570 571 /* 572 * Set vnode attributes to VNOVAL 573 */ 574 void 575 vattr_null(struct vattr *vap) 576 { 577 578 vap->va_type = VNON; 579 vap->va_size = VNOVAL; 580 vap->va_bytes = VNOVAL; 581 vap->va_mode = VNOVAL; 582 vap->va_nlink = VNOVAL; 583 vap->va_uid = VNOVAL; 584 vap->va_gid = VNOVAL; 585 vap->va_fsid = VNOVAL; 586 vap->va_fileid = VNOVAL; 587 vap->va_blocksize = VNOVAL; 588 vap->va_rdev = VNOVAL; 589 vap->va_atime.tv_sec = VNOVAL; 590 vap->va_atime.tv_nsec = VNOVAL; 591 vap->va_mtime.tv_sec = VNOVAL; 592 vap->va_mtime.tv_nsec = VNOVAL; 593 vap->va_ctime.tv_sec = VNOVAL; 594 vap->va_ctime.tv_nsec = VNOVAL; 595 vap->va_birthtime.tv_sec = VNOVAL; 596 vap->va_birthtime.tv_nsec = VNOVAL; 597 vap->va_flags = VNOVAL; 598 vap->va_gen = VNOVAL; 599 vap->va_vaflags = 0; 600 } 601 602 /* 603 * This routine is called when we have too many vnodes. It attempts 604 * to free <count> vnodes and will potentially free vnodes that still 605 * have VM backing store (VM backing store is typically the cause 606 * of a vnode blowout so we want to do this). Therefore, this operation 607 * is not considered cheap. 608 * 609 * A number of conditions may prevent a vnode from being reclaimed. 610 * the buffer cache may have references on the vnode, a directory 611 * vnode may still have references due to the namei cache representing 612 * underlying files, or the vnode may be in active use. It is not 613 * desireable to reuse such vnodes. These conditions may cause the 614 * number of vnodes to reach some minimum value regardless of what 615 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 616 */ 617 static int 618 vlrureclaim(struct mount *mp) 619 { 620 struct vnode *vp; 621 int done; 622 int trigger; 623 int usevnodes; 624 int count; 625 626 /* 627 * Calculate the trigger point, don't allow user 628 * screwups to blow us up. This prevents us from 629 * recycling vnodes with lots of resident pages. We 630 * aren't trying to free memory, we are trying to 631 * free vnodes. 632 */ 633 usevnodes = desiredvnodes; 634 if (usevnodes <= 0) 635 usevnodes = 1; 636 trigger = cnt.v_page_count * 2 / usevnodes; 637 done = 0; 638 vn_start_write(NULL, &mp, V_WAIT); 639 MNT_ILOCK(mp); 640 count = mp->mnt_nvnodelistsize / 10 + 1; 641 while (count != 0) { 642 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 643 while (vp != NULL && vp->v_type == VMARKER) 644 vp = TAILQ_NEXT(vp, v_nmntvnodes); 645 if (vp == NULL) 646 break; 647 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 648 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 649 --count; 650 if (!VI_TRYLOCK(vp)) 651 goto next_iter; 652 /* 653 * If it's been deconstructed already, it's still 654 * referenced, or it exceeds the trigger, skip it. 655 */ 656 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) || 657 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL && 658 vp->v_object->resident_page_count > trigger)) { 659 VI_UNLOCK(vp); 660 goto next_iter; 661 } 662 MNT_IUNLOCK(mp); 663 vholdl(vp); 664 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) { 665 vdrop(vp); 666 goto next_iter_mntunlocked; 667 } 668 VI_LOCK(vp); 669 /* 670 * v_usecount may have been bumped after VOP_LOCK() dropped 671 * the vnode interlock and before it was locked again. 672 * 673 * It is not necessary to recheck VI_DOOMED because it can 674 * only be set by another thread that holds both the vnode 675 * lock and vnode interlock. If another thread has the 676 * vnode lock before we get to VOP_LOCK() and obtains the 677 * vnode interlock after VOP_LOCK() drops the vnode 678 * interlock, the other thread will be unable to drop the 679 * vnode lock before our VOP_LOCK() call fails. 680 */ 681 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) || 682 (vp->v_object != NULL && 683 vp->v_object->resident_page_count > trigger)) { 684 VOP_UNLOCK(vp, LK_INTERLOCK); 685 goto next_iter_mntunlocked; 686 } 687 KASSERT((vp->v_iflag & VI_DOOMED) == 0, 688 ("VI_DOOMED unexpectedly detected in vlrureclaim()")); 689 vgonel(vp); 690 VOP_UNLOCK(vp, 0); 691 vdropl(vp); 692 done++; 693 next_iter_mntunlocked: 694 if ((count % 256) != 0) 695 goto relock_mnt; 696 goto yield; 697 next_iter: 698 if ((count % 256) != 0) 699 continue; 700 MNT_IUNLOCK(mp); 701 yield: 702 uio_yield(); 703 relock_mnt: 704 MNT_ILOCK(mp); 705 } 706 MNT_IUNLOCK(mp); 707 vn_finished_write(mp); 708 return done; 709 } 710 711 /* 712 * Attempt to keep the free list at wantfreevnodes length. 713 */ 714 static void 715 vnlru_free(int count) 716 { 717 struct vnode *vp; 718 int vfslocked; 719 720 mtx_assert(&vnode_free_list_mtx, MA_OWNED); 721 for (; count > 0; count--) { 722 vp = TAILQ_FIRST(&vnode_free_list); 723 /* 724 * The list can be modified while the free_list_mtx 725 * has been dropped and vp could be NULL here. 726 */ 727 if (!vp) 728 break; 729 VNASSERT(vp->v_op != NULL, vp, 730 ("vnlru_free: vnode already reclaimed.")); 731 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 732 /* 733 * Don't recycle if we can't get the interlock. 734 */ 735 if (!VI_TRYLOCK(vp)) { 736 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 737 continue; 738 } 739 VNASSERT(VCANRECYCLE(vp), vp, 740 ("vp inconsistent on freelist")); 741 freevnodes--; 742 vp->v_iflag &= ~VI_FREE; 743 vholdl(vp); 744 mtx_unlock(&vnode_free_list_mtx); 745 VI_UNLOCK(vp); 746 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 747 vtryrecycle(vp); 748 VFS_UNLOCK_GIANT(vfslocked); 749 /* 750 * If the recycled succeeded this vdrop will actually free 751 * the vnode. If not it will simply place it back on 752 * the free list. 753 */ 754 vdrop(vp); 755 mtx_lock(&vnode_free_list_mtx); 756 } 757 } 758 /* 759 * Attempt to recycle vnodes in a context that is always safe to block. 760 * Calling vlrurecycle() from the bowels of filesystem code has some 761 * interesting deadlock problems. 762 */ 763 static struct proc *vnlruproc; 764 static int vnlruproc_sig; 765 766 static void 767 vnlru_proc(void) 768 { 769 struct mount *mp, *nmp; 770 int done, vfslocked; 771 struct proc *p = vnlruproc; 772 773 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p, 774 SHUTDOWN_PRI_FIRST); 775 776 for (;;) { 777 kproc_suspend_check(p); 778 mtx_lock(&vnode_free_list_mtx); 779 if (freevnodes > wantfreevnodes) 780 vnlru_free(freevnodes - wantfreevnodes); 781 if (numvnodes <= desiredvnodes * 9 / 10) { 782 vnlruproc_sig = 0; 783 wakeup(&vnlruproc_sig); 784 msleep(vnlruproc, &vnode_free_list_mtx, 785 PVFS|PDROP, "vlruwt", hz); 786 continue; 787 } 788 mtx_unlock(&vnode_free_list_mtx); 789 done = 0; 790 mtx_lock(&mountlist_mtx); 791 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 792 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { 793 nmp = TAILQ_NEXT(mp, mnt_list); 794 continue; 795 } 796 vfslocked = VFS_LOCK_GIANT(mp); 797 done += vlrureclaim(mp); 798 VFS_UNLOCK_GIANT(vfslocked); 799 mtx_lock(&mountlist_mtx); 800 nmp = TAILQ_NEXT(mp, mnt_list); 801 vfs_unbusy(mp); 802 } 803 mtx_unlock(&mountlist_mtx); 804 if (done == 0) { 805 EVENTHANDLER_INVOKE(vfs_lowvnodes, desiredvnodes / 10); 806 #if 0 807 /* These messages are temporary debugging aids */ 808 if (vnlru_nowhere < 5) 809 printf("vnlru process getting nowhere..\n"); 810 else if (vnlru_nowhere == 5) 811 printf("vnlru process messages stopped.\n"); 812 #endif 813 vnlru_nowhere++; 814 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); 815 } else 816 uio_yield(); 817 } 818 } 819 820 static struct kproc_desc vnlru_kp = { 821 "vnlru", 822 vnlru_proc, 823 &vnlruproc 824 }; 825 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, 826 &vnlru_kp); 827 828 /* 829 * Routines having to do with the management of the vnode table. 830 */ 831 832 void 833 vdestroy(struct vnode *vp) 834 { 835 struct bufobj *bo; 836 837 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 838 mtx_lock(&vnode_free_list_mtx); 839 numvnodes--; 840 mtx_unlock(&vnode_free_list_mtx); 841 bo = &vp->v_bufobj; 842 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, 843 ("cleaned vnode still on the free list.")); 844 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); 845 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count")); 846 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); 847 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); 848 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); 849 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); 850 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL")); 851 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); 852 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL")); 853 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); 854 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); 855 VI_UNLOCK(vp); 856 #ifdef MAC 857 mac_vnode_destroy(vp); 858 #endif 859 if (vp->v_pollinfo != NULL) 860 destroy_vpollinfo(vp->v_pollinfo); 861 #ifdef INVARIANTS 862 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */ 863 vp->v_op = NULL; 864 #endif 865 lockdestroy(vp->v_vnlock); 866 mtx_destroy(&vp->v_interlock); 867 mtx_destroy(BO_MTX(bo)); 868 uma_zfree(vnode_zone, vp); 869 } 870 871 /* 872 * Try to recycle a freed vnode. We abort if anyone picks up a reference 873 * before we actually vgone(). This function must be called with the vnode 874 * held to prevent the vnode from being returned to the free list midway 875 * through vgone(). 876 */ 877 static int 878 vtryrecycle(struct vnode *vp) 879 { 880 struct mount *vnmp; 881 882 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 883 VNASSERT(vp->v_holdcnt, vp, 884 ("vtryrecycle: Recycling vp %p without a reference.", vp)); 885 /* 886 * This vnode may found and locked via some other list, if so we 887 * can't recycle it yet. 888 */ 889 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 890 CTR2(KTR_VFS, 891 "%s: impossible to recycle, vp %p lock is already held", 892 __func__, vp); 893 return (EWOULDBLOCK); 894 } 895 /* 896 * Don't recycle if its filesystem is being suspended. 897 */ 898 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { 899 VOP_UNLOCK(vp, 0); 900 CTR2(KTR_VFS, 901 "%s: impossible to recycle, cannot start the write for %p", 902 __func__, vp); 903 return (EBUSY); 904 } 905 /* 906 * If we got this far, we need to acquire the interlock and see if 907 * anyone picked up this vnode from another list. If not, we will 908 * mark it with DOOMED via vgonel() so that anyone who does find it 909 * will skip over it. 910 */ 911 VI_LOCK(vp); 912 if (vp->v_usecount) { 913 VOP_UNLOCK(vp, LK_INTERLOCK); 914 vn_finished_write(vnmp); 915 CTR2(KTR_VFS, 916 "%s: impossible to recycle, %p is already referenced", 917 __func__, vp); 918 return (EBUSY); 919 } 920 if ((vp->v_iflag & VI_DOOMED) == 0) 921 vgonel(vp); 922 VOP_UNLOCK(vp, LK_INTERLOCK); 923 vn_finished_write(vnmp); 924 return (0); 925 } 926 927 /* 928 * Return the next vnode from the free list. 929 */ 930 int 931 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, 932 struct vnode **vpp) 933 { 934 struct vnode *vp = NULL; 935 struct bufobj *bo; 936 937 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag); 938 mtx_lock(&vnode_free_list_mtx); 939 /* 940 * Lend our context to reclaim vnodes if they've exceeded the max. 941 */ 942 if (freevnodes > wantfreevnodes) 943 vnlru_free(1); 944 /* 945 * Wait for available vnodes. 946 */ 947 if (numvnodes > desiredvnodes) { 948 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) { 949 /* 950 * File system is beeing suspended, we cannot risk a 951 * deadlock here, so allocate new vnode anyway. 952 */ 953 if (freevnodes > wantfreevnodes) 954 vnlru_free(freevnodes - wantfreevnodes); 955 goto alloc; 956 } 957 if (vnlruproc_sig == 0) { 958 vnlruproc_sig = 1; /* avoid unnecessary wakeups */ 959 wakeup(vnlruproc); 960 } 961 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS, 962 "vlruwk", hz); 963 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */ 964 if (numvnodes > desiredvnodes) { 965 mtx_unlock(&vnode_free_list_mtx); 966 return (ENFILE); 967 } 968 #endif 969 } 970 alloc: 971 numvnodes++; 972 mtx_unlock(&vnode_free_list_mtx); 973 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO); 974 /* 975 * Setup locks. 976 */ 977 vp->v_vnlock = &vp->v_lock; 978 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); 979 /* 980 * By default, don't allow shared locks unless filesystems 981 * opt-in. 982 */ 983 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE); 984 /* 985 * Initialize bufobj. 986 */ 987 bo = &vp->v_bufobj; 988 bo->__bo_vnode = vp; 989 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF); 990 bo->bo_ops = &buf_ops_bio; 991 bo->bo_private = vp; 992 TAILQ_INIT(&bo->bo_clean.bv_hd); 993 TAILQ_INIT(&bo->bo_dirty.bv_hd); 994 /* 995 * Initialize namecache. 996 */ 997 LIST_INIT(&vp->v_cache_src); 998 TAILQ_INIT(&vp->v_cache_dst); 999 /* 1000 * Finalize various vnode identity bits. 1001 */ 1002 vp->v_type = VNON; 1003 vp->v_tag = tag; 1004 vp->v_op = vops; 1005 v_incr_usecount(vp); 1006 vp->v_data = 0; 1007 #ifdef MAC 1008 mac_vnode_init(vp); 1009 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) 1010 mac_vnode_associate_singlelabel(mp, vp); 1011 else if (mp == NULL && vops != &dead_vnodeops) 1012 printf("NULL mp in getnewvnode()\n"); 1013 #endif 1014 if (mp != NULL) { 1015 bo->bo_bsize = mp->mnt_stat.f_iosize; 1016 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) 1017 vp->v_vflag |= VV_NOKNOTE; 1018 } 1019 1020 *vpp = vp; 1021 return (0); 1022 } 1023 1024 /* 1025 * Delete from old mount point vnode list, if on one. 1026 */ 1027 static void 1028 delmntque(struct vnode *vp) 1029 { 1030 struct mount *mp; 1031 1032 mp = vp->v_mount; 1033 if (mp == NULL) 1034 return; 1035 MNT_ILOCK(mp); 1036 vp->v_mount = NULL; 1037 VNASSERT(mp->mnt_nvnodelistsize > 0, vp, 1038 ("bad mount point vnode list size")); 1039 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1040 mp->mnt_nvnodelistsize--; 1041 MNT_REL(mp); 1042 MNT_IUNLOCK(mp); 1043 } 1044 1045 static void 1046 insmntque_stddtr(struct vnode *vp, void *dtr_arg) 1047 { 1048 1049 vp->v_data = NULL; 1050 vp->v_op = &dead_vnodeops; 1051 /* XXX non mp-safe fs may still call insmntque with vnode 1052 unlocked */ 1053 if (!VOP_ISLOCKED(vp)) 1054 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1055 vgone(vp); 1056 vput(vp); 1057 } 1058 1059 /* 1060 * Insert into list of vnodes for the new mount point, if available. 1061 */ 1062 int 1063 insmntque1(struct vnode *vp, struct mount *mp, 1064 void (*dtr)(struct vnode *, void *), void *dtr_arg) 1065 { 1066 int locked; 1067 1068 KASSERT(vp->v_mount == NULL, 1069 ("insmntque: vnode already on per mount vnode list")); 1070 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); 1071 #ifdef DEBUG_VFS_LOCKS 1072 if (!VFS_NEEDSGIANT(mp)) 1073 ASSERT_VOP_ELOCKED(vp, 1074 "insmntque: mp-safe fs and non-locked vp"); 1075 #endif 1076 MNT_ILOCK(mp); 1077 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 && 1078 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 || 1079 mp->mnt_nvnodelistsize == 0)) { 1080 locked = VOP_ISLOCKED(vp); 1081 if (!locked || (locked == LK_EXCLUSIVE && 1082 (vp->v_vflag & VV_FORCEINSMQ) == 0)) { 1083 MNT_IUNLOCK(mp); 1084 if (dtr != NULL) 1085 dtr(vp, dtr_arg); 1086 return (EBUSY); 1087 } 1088 } 1089 vp->v_mount = mp; 1090 MNT_REF(mp); 1091 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1092 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, 1093 ("neg mount point vnode list size")); 1094 mp->mnt_nvnodelistsize++; 1095 MNT_IUNLOCK(mp); 1096 return (0); 1097 } 1098 1099 int 1100 insmntque(struct vnode *vp, struct mount *mp) 1101 { 1102 1103 return (insmntque1(vp, mp, insmntque_stddtr, NULL)); 1104 } 1105 1106 /* 1107 * Flush out and invalidate all buffers associated with a bufobj 1108 * Called with the underlying object locked. 1109 */ 1110 int 1111 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo) 1112 { 1113 int error; 1114 1115 BO_LOCK(bo); 1116 if (flags & V_SAVE) { 1117 error = bufobj_wwait(bo, slpflag, slptimeo); 1118 if (error) { 1119 BO_UNLOCK(bo); 1120 return (error); 1121 } 1122 if (bo->bo_dirty.bv_cnt > 0) { 1123 BO_UNLOCK(bo); 1124 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0) 1125 return (error); 1126 /* 1127 * XXX We could save a lock/unlock if this was only 1128 * enabled under INVARIANTS 1129 */ 1130 BO_LOCK(bo); 1131 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) 1132 panic("vinvalbuf: dirty bufs"); 1133 } 1134 } 1135 /* 1136 * If you alter this loop please notice that interlock is dropped and 1137 * reacquired in flushbuflist. Special care is needed to ensure that 1138 * no race conditions occur from this. 1139 */ 1140 do { 1141 error = flushbuflist(&bo->bo_clean, 1142 flags, bo, slpflag, slptimeo); 1143 if (error == 0) 1144 error = flushbuflist(&bo->bo_dirty, 1145 flags, bo, slpflag, slptimeo); 1146 if (error != 0 && error != EAGAIN) { 1147 BO_UNLOCK(bo); 1148 return (error); 1149 } 1150 } while (error != 0); 1151 1152 /* 1153 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 1154 * have write I/O in-progress but if there is a VM object then the 1155 * VM object can also have read-I/O in-progress. 1156 */ 1157 do { 1158 bufobj_wwait(bo, 0, 0); 1159 BO_UNLOCK(bo); 1160 if (bo->bo_object != NULL) { 1161 VM_OBJECT_LOCK(bo->bo_object); 1162 vm_object_pip_wait(bo->bo_object, "bovlbx"); 1163 VM_OBJECT_UNLOCK(bo->bo_object); 1164 } 1165 BO_LOCK(bo); 1166 } while (bo->bo_numoutput > 0); 1167 BO_UNLOCK(bo); 1168 1169 /* 1170 * Destroy the copy in the VM cache, too. 1171 */ 1172 if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) { 1173 VM_OBJECT_LOCK(bo->bo_object); 1174 vm_object_page_remove(bo->bo_object, 0, 0, 1175 (flags & V_SAVE) ? TRUE : FALSE); 1176 VM_OBJECT_UNLOCK(bo->bo_object); 1177 } 1178 1179 #ifdef INVARIANTS 1180 BO_LOCK(bo); 1181 if ((flags & (V_ALT | V_NORMAL)) == 0 && 1182 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) 1183 panic("vinvalbuf: flush failed"); 1184 BO_UNLOCK(bo); 1185 #endif 1186 return (0); 1187 } 1188 1189 /* 1190 * Flush out and invalidate all buffers associated with a vnode. 1191 * Called with the underlying object locked. 1192 */ 1193 int 1194 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 1195 { 1196 1197 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); 1198 ASSERT_VOP_LOCKED(vp, "vinvalbuf"); 1199 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo)); 1200 } 1201 1202 /* 1203 * Flush out buffers on the specified list. 1204 * 1205 */ 1206 static int 1207 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, 1208 int slptimeo) 1209 { 1210 struct buf *bp, *nbp; 1211 int retval, error; 1212 daddr_t lblkno; 1213 b_xflags_t xflags; 1214 1215 ASSERT_BO_LOCKED(bo); 1216 1217 retval = 0; 1218 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { 1219 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) || 1220 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) { 1221 continue; 1222 } 1223 lblkno = 0; 1224 xflags = 0; 1225 if (nbp != NULL) { 1226 lblkno = nbp->b_lblkno; 1227 xflags = nbp->b_xflags & 1228 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN); 1229 } 1230 retval = EAGAIN; 1231 error = BUF_TIMELOCK(bp, 1232 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo), 1233 "flushbuf", slpflag, slptimeo); 1234 if (error) { 1235 BO_LOCK(bo); 1236 return (error != ENOLCK ? error : EAGAIN); 1237 } 1238 KASSERT(bp->b_bufobj == bo, 1239 ("bp %p wrong b_bufobj %p should be %p", 1240 bp, bp->b_bufobj, bo)); 1241 if (bp->b_bufobj != bo) { /* XXX: necessary ? */ 1242 BUF_UNLOCK(bp); 1243 BO_LOCK(bo); 1244 return (EAGAIN); 1245 } 1246 /* 1247 * XXX Since there are no node locks for NFS, I 1248 * believe there is a slight chance that a delayed 1249 * write will occur while sleeping just above, so 1250 * check for it. 1251 */ 1252 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 1253 (flags & V_SAVE)) { 1254 bremfree(bp); 1255 bp->b_flags |= B_ASYNC; 1256 bwrite(bp); 1257 BO_LOCK(bo); 1258 return (EAGAIN); /* XXX: why not loop ? */ 1259 } 1260 bremfree(bp); 1261 bp->b_flags |= (B_INVAL | B_RELBUF); 1262 bp->b_flags &= ~B_ASYNC; 1263 brelse(bp); 1264 BO_LOCK(bo); 1265 if (nbp != NULL && 1266 (nbp->b_bufobj != bo || 1267 nbp->b_lblkno != lblkno || 1268 (nbp->b_xflags & 1269 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags)) 1270 break; /* nbp invalid */ 1271 } 1272 return (retval); 1273 } 1274 1275 /* 1276 * Truncate a file's buffer and pages to a specified length. This 1277 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 1278 * sync activity. 1279 */ 1280 int 1281 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td, 1282 off_t length, int blksize) 1283 { 1284 struct buf *bp, *nbp; 1285 int anyfreed; 1286 int trunclbn; 1287 struct bufobj *bo; 1288 1289 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__, 1290 vp, cred, blksize, (uintmax_t)length); 1291 1292 /* 1293 * Round up to the *next* lbn. 1294 */ 1295 trunclbn = (length + blksize - 1) / blksize; 1296 1297 ASSERT_VOP_LOCKED(vp, "vtruncbuf"); 1298 restart: 1299 bo = &vp->v_bufobj; 1300 BO_LOCK(bo); 1301 anyfreed = 1; 1302 for (;anyfreed;) { 1303 anyfreed = 0; 1304 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { 1305 if (bp->b_lblkno < trunclbn) 1306 continue; 1307 if (BUF_LOCK(bp, 1308 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1309 BO_MTX(bo)) == ENOLCK) 1310 goto restart; 1311 1312 bremfree(bp); 1313 bp->b_flags |= (B_INVAL | B_RELBUF); 1314 bp->b_flags &= ~B_ASYNC; 1315 brelse(bp); 1316 anyfreed = 1; 1317 1318 if (nbp != NULL && 1319 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 1320 (nbp->b_vp != vp) || 1321 (nbp->b_flags & B_DELWRI))) { 1322 goto restart; 1323 } 1324 BO_LOCK(bo); 1325 } 1326 1327 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1328 if (bp->b_lblkno < trunclbn) 1329 continue; 1330 if (BUF_LOCK(bp, 1331 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1332 BO_MTX(bo)) == ENOLCK) 1333 goto restart; 1334 bremfree(bp); 1335 bp->b_flags |= (B_INVAL | B_RELBUF); 1336 bp->b_flags &= ~B_ASYNC; 1337 brelse(bp); 1338 anyfreed = 1; 1339 if (nbp != NULL && 1340 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 1341 (nbp->b_vp != vp) || 1342 (nbp->b_flags & B_DELWRI) == 0)) { 1343 goto restart; 1344 } 1345 BO_LOCK(bo); 1346 } 1347 } 1348 1349 if (length > 0) { 1350 restartsync: 1351 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1352 if (bp->b_lblkno > 0) 1353 continue; 1354 /* 1355 * Since we hold the vnode lock this should only 1356 * fail if we're racing with the buf daemon. 1357 */ 1358 if (BUF_LOCK(bp, 1359 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1360 BO_MTX(bo)) == ENOLCK) { 1361 goto restart; 1362 } 1363 VNASSERT((bp->b_flags & B_DELWRI), vp, 1364 ("buf(%p) on dirty queue without DELWRI", bp)); 1365 1366 bremfree(bp); 1367 bawrite(bp); 1368 BO_LOCK(bo); 1369 goto restartsync; 1370 } 1371 } 1372 1373 bufobj_wwait(bo, 0, 0); 1374 BO_UNLOCK(bo); 1375 vnode_pager_setsize(vp, length); 1376 1377 return (0); 1378 } 1379 1380 /* 1381 * buf_splay() - splay tree core for the clean/dirty list of buffers in 1382 * a vnode. 1383 * 1384 * NOTE: We have to deal with the special case of a background bitmap 1385 * buffer, a situation where two buffers will have the same logical 1386 * block offset. We want (1) only the foreground buffer to be accessed 1387 * in a lookup and (2) must differentiate between the foreground and 1388 * background buffer in the splay tree algorithm because the splay 1389 * tree cannot normally handle multiple entities with the same 'index'. 1390 * We accomplish this by adding differentiating flags to the splay tree's 1391 * numerical domain. 1392 */ 1393 static 1394 struct buf * 1395 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root) 1396 { 1397 struct buf dummy; 1398 struct buf *lefttreemax, *righttreemin, *y; 1399 1400 if (root == NULL) 1401 return (NULL); 1402 lefttreemax = righttreemin = &dummy; 1403 for (;;) { 1404 if (lblkno < root->b_lblkno || 1405 (lblkno == root->b_lblkno && 1406 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1407 if ((y = root->b_left) == NULL) 1408 break; 1409 if (lblkno < y->b_lblkno) { 1410 /* Rotate right. */ 1411 root->b_left = y->b_right; 1412 y->b_right = root; 1413 root = y; 1414 if ((y = root->b_left) == NULL) 1415 break; 1416 } 1417 /* Link into the new root's right tree. */ 1418 righttreemin->b_left = root; 1419 righttreemin = root; 1420 } else if (lblkno > root->b_lblkno || 1421 (lblkno == root->b_lblkno && 1422 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) { 1423 if ((y = root->b_right) == NULL) 1424 break; 1425 if (lblkno > y->b_lblkno) { 1426 /* Rotate left. */ 1427 root->b_right = y->b_left; 1428 y->b_left = root; 1429 root = y; 1430 if ((y = root->b_right) == NULL) 1431 break; 1432 } 1433 /* Link into the new root's left tree. */ 1434 lefttreemax->b_right = root; 1435 lefttreemax = root; 1436 } else { 1437 break; 1438 } 1439 root = y; 1440 } 1441 /* Assemble the new root. */ 1442 lefttreemax->b_right = root->b_left; 1443 righttreemin->b_left = root->b_right; 1444 root->b_left = dummy.b_right; 1445 root->b_right = dummy.b_left; 1446 return (root); 1447 } 1448 1449 static void 1450 buf_vlist_remove(struct buf *bp) 1451 { 1452 struct buf *root; 1453 struct bufv *bv; 1454 1455 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); 1456 ASSERT_BO_LOCKED(bp->b_bufobj); 1457 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) != 1458 (BX_VNDIRTY|BX_VNCLEAN), 1459 ("buf_vlist_remove: Buf %p is on two lists", bp)); 1460 if (bp->b_xflags & BX_VNDIRTY) 1461 bv = &bp->b_bufobj->bo_dirty; 1462 else 1463 bv = &bp->b_bufobj->bo_clean; 1464 if (bp != bv->bv_root) { 1465 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root); 1466 KASSERT(root == bp, ("splay lookup failed in remove")); 1467 } 1468 if (bp->b_left == NULL) { 1469 root = bp->b_right; 1470 } else { 1471 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left); 1472 root->b_right = bp->b_right; 1473 } 1474 bv->bv_root = root; 1475 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); 1476 bv->bv_cnt--; 1477 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1478 } 1479 1480 /* 1481 * Add the buffer to the sorted clean or dirty block list using a 1482 * splay tree algorithm. 1483 * 1484 * NOTE: xflags is passed as a constant, optimizing this inline function! 1485 */ 1486 static void 1487 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) 1488 { 1489 struct buf *root; 1490 struct bufv *bv; 1491 1492 ASSERT_BO_LOCKED(bo); 1493 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, 1494 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); 1495 bp->b_xflags |= xflags; 1496 if (xflags & BX_VNDIRTY) 1497 bv = &bo->bo_dirty; 1498 else 1499 bv = &bo->bo_clean; 1500 1501 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root); 1502 if (root == NULL) { 1503 bp->b_left = NULL; 1504 bp->b_right = NULL; 1505 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); 1506 } else if (bp->b_lblkno < root->b_lblkno || 1507 (bp->b_lblkno == root->b_lblkno && 1508 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1509 bp->b_left = root->b_left; 1510 bp->b_right = root; 1511 root->b_left = NULL; 1512 TAILQ_INSERT_BEFORE(root, bp, b_bobufs); 1513 } else { 1514 bp->b_right = root->b_right; 1515 bp->b_left = root; 1516 root->b_right = NULL; 1517 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs); 1518 } 1519 bv->bv_cnt++; 1520 bv->bv_root = bp; 1521 } 1522 1523 /* 1524 * Lookup a buffer using the splay tree. Note that we specifically avoid 1525 * shadow buffers used in background bitmap writes. 1526 * 1527 * This code isn't quite efficient as it could be because we are maintaining 1528 * two sorted lists and do not know which list the block resides in. 1529 * 1530 * During a "make buildworld" the desired buffer is found at one of 1531 * the roots more than 60% of the time. Thus, checking both roots 1532 * before performing either splay eliminates unnecessary splays on the 1533 * first tree splayed. 1534 */ 1535 struct buf * 1536 gbincore(struct bufobj *bo, daddr_t lblkno) 1537 { 1538 struct buf *bp; 1539 1540 ASSERT_BO_LOCKED(bo); 1541 if ((bp = bo->bo_clean.bv_root) != NULL && 1542 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1543 return (bp); 1544 if ((bp = bo->bo_dirty.bv_root) != NULL && 1545 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1546 return (bp); 1547 if ((bp = bo->bo_clean.bv_root) != NULL) { 1548 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp); 1549 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1550 return (bp); 1551 } 1552 if ((bp = bo->bo_dirty.bv_root) != NULL) { 1553 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp); 1554 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1555 return (bp); 1556 } 1557 return (NULL); 1558 } 1559 1560 /* 1561 * Associate a buffer with a vnode. 1562 */ 1563 void 1564 bgetvp(struct vnode *vp, struct buf *bp) 1565 { 1566 struct bufobj *bo; 1567 1568 bo = &vp->v_bufobj; 1569 ASSERT_BO_LOCKED(bo); 1570 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); 1571 1572 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); 1573 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, 1574 ("bgetvp: bp already attached! %p", bp)); 1575 1576 vhold(vp); 1577 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT) 1578 bp->b_flags |= B_NEEDSGIANT; 1579 bp->b_vp = vp; 1580 bp->b_bufobj = bo; 1581 /* 1582 * Insert onto list for new vnode. 1583 */ 1584 buf_vlist_add(bp, bo, BX_VNCLEAN); 1585 } 1586 1587 /* 1588 * Disassociate a buffer from a vnode. 1589 */ 1590 void 1591 brelvp(struct buf *bp) 1592 { 1593 struct bufobj *bo; 1594 struct vnode *vp; 1595 1596 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1597 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 1598 1599 /* 1600 * Delete from old vnode list, if on one. 1601 */ 1602 vp = bp->b_vp; /* XXX */ 1603 bo = bp->b_bufobj; 1604 BO_LOCK(bo); 1605 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1606 buf_vlist_remove(bp); 1607 else 1608 panic("brelvp: Buffer %p not on queue.", bp); 1609 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 1610 bo->bo_flag &= ~BO_ONWORKLST; 1611 mtx_lock(&sync_mtx); 1612 LIST_REMOVE(bo, bo_synclist); 1613 syncer_worklist_len--; 1614 mtx_unlock(&sync_mtx); 1615 } 1616 bp->b_flags &= ~B_NEEDSGIANT; 1617 bp->b_vp = NULL; 1618 bp->b_bufobj = NULL; 1619 BO_UNLOCK(bo); 1620 vdrop(vp); 1621 } 1622 1623 /* 1624 * Add an item to the syncer work queue. 1625 */ 1626 static void 1627 vn_syncer_add_to_worklist(struct bufobj *bo, int delay) 1628 { 1629 int queue, slot; 1630 1631 ASSERT_BO_LOCKED(bo); 1632 1633 mtx_lock(&sync_mtx); 1634 if (bo->bo_flag & BO_ONWORKLST) 1635 LIST_REMOVE(bo, bo_synclist); 1636 else { 1637 bo->bo_flag |= BO_ONWORKLST; 1638 syncer_worklist_len++; 1639 } 1640 1641 if (delay > syncer_maxdelay - 2) 1642 delay = syncer_maxdelay - 2; 1643 slot = (syncer_delayno + delay) & syncer_mask; 1644 1645 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ : 1646 WI_MPSAFEQ; 1647 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo, 1648 bo_synclist); 1649 mtx_unlock(&sync_mtx); 1650 } 1651 1652 static int 1653 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) 1654 { 1655 int error, len; 1656 1657 mtx_lock(&sync_mtx); 1658 len = syncer_worklist_len - sync_vnode_count; 1659 mtx_unlock(&sync_mtx); 1660 error = SYSCTL_OUT(req, &len, sizeof(len)); 1661 return (error); 1662 } 1663 1664 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0, 1665 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); 1666 1667 static struct proc *updateproc; 1668 static void sched_sync(void); 1669 static struct kproc_desc up_kp = { 1670 "syncer", 1671 sched_sync, 1672 &updateproc 1673 }; 1674 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp); 1675 1676 static int 1677 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td) 1678 { 1679 struct vnode *vp; 1680 struct mount *mp; 1681 1682 *bo = LIST_FIRST(slp); 1683 if (*bo == NULL) 1684 return (0); 1685 vp = (*bo)->__bo_vnode; /* XXX */ 1686 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0) 1687 return (1); 1688 /* 1689 * We use vhold in case the vnode does not 1690 * successfully sync. vhold prevents the vnode from 1691 * going away when we unlock the sync_mtx so that 1692 * we can acquire the vnode interlock. 1693 */ 1694 vholdl(vp); 1695 mtx_unlock(&sync_mtx); 1696 VI_UNLOCK(vp); 1697 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { 1698 vdrop(vp); 1699 mtx_lock(&sync_mtx); 1700 return (*bo == LIST_FIRST(slp)); 1701 } 1702 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1703 (void) VOP_FSYNC(vp, MNT_LAZY, td); 1704 VOP_UNLOCK(vp, 0); 1705 vn_finished_write(mp); 1706 BO_LOCK(*bo); 1707 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) { 1708 /* 1709 * Put us back on the worklist. The worklist 1710 * routine will remove us from our current 1711 * position and then add us back in at a later 1712 * position. 1713 */ 1714 vn_syncer_add_to_worklist(*bo, syncdelay); 1715 } 1716 BO_UNLOCK(*bo); 1717 vdrop(vp); 1718 mtx_lock(&sync_mtx); 1719 return (0); 1720 } 1721 1722 /* 1723 * System filesystem synchronizer daemon. 1724 */ 1725 static void 1726 sched_sync(void) 1727 { 1728 struct synclist *gnext, *next; 1729 struct synclist *gslp, *slp; 1730 struct bufobj *bo; 1731 long starttime; 1732 struct thread *td = curthread; 1733 int last_work_seen; 1734 int net_worklist_len; 1735 int syncer_final_iter; 1736 int first_printf; 1737 int error; 1738 1739 last_work_seen = 0; 1740 syncer_final_iter = 0; 1741 first_printf = 1; 1742 syncer_state = SYNCER_RUNNING; 1743 starttime = time_uptime; 1744 td->td_pflags |= TDP_NORUNNINGBUF; 1745 1746 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, 1747 SHUTDOWN_PRI_LAST); 1748 1749 mtx_lock(&sync_mtx); 1750 for (;;) { 1751 if (syncer_state == SYNCER_FINAL_DELAY && 1752 syncer_final_iter == 0) { 1753 mtx_unlock(&sync_mtx); 1754 kproc_suspend_check(td->td_proc); 1755 mtx_lock(&sync_mtx); 1756 } 1757 net_worklist_len = syncer_worklist_len - sync_vnode_count; 1758 if (syncer_state != SYNCER_RUNNING && 1759 starttime != time_uptime) { 1760 if (first_printf) { 1761 printf("\nSyncing disks, vnodes remaining..."); 1762 first_printf = 0; 1763 } 1764 printf("%d ", net_worklist_len); 1765 } 1766 starttime = time_uptime; 1767 1768 /* 1769 * Push files whose dirty time has expired. Be careful 1770 * of interrupt race on slp queue. 1771 * 1772 * Skip over empty worklist slots when shutting down. 1773 */ 1774 do { 1775 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno]; 1776 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno]; 1777 syncer_delayno += 1; 1778 if (syncer_delayno == syncer_maxdelay) 1779 syncer_delayno = 0; 1780 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno]; 1781 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno]; 1782 /* 1783 * If the worklist has wrapped since the 1784 * it was emptied of all but syncer vnodes, 1785 * switch to the FINAL_DELAY state and run 1786 * for one more second. 1787 */ 1788 if (syncer_state == SYNCER_SHUTTING_DOWN && 1789 net_worklist_len == 0 && 1790 last_work_seen == syncer_delayno) { 1791 syncer_state = SYNCER_FINAL_DELAY; 1792 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; 1793 } 1794 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && 1795 LIST_EMPTY(gslp) && syncer_worklist_len > 0); 1796 1797 /* 1798 * Keep track of the last time there was anything 1799 * on the worklist other than syncer vnodes. 1800 * Return to the SHUTTING_DOWN state if any 1801 * new work appears. 1802 */ 1803 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) 1804 last_work_seen = syncer_delayno; 1805 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) 1806 syncer_state = SYNCER_SHUTTING_DOWN; 1807 while (!LIST_EMPTY(slp)) { 1808 error = sync_vnode(slp, &bo, td); 1809 if (error == 1) { 1810 LIST_REMOVE(bo, bo_synclist); 1811 LIST_INSERT_HEAD(next, bo, bo_synclist); 1812 continue; 1813 } 1814 } 1815 if (!LIST_EMPTY(gslp)) { 1816 mtx_unlock(&sync_mtx); 1817 mtx_lock(&Giant); 1818 mtx_lock(&sync_mtx); 1819 while (!LIST_EMPTY(gslp)) { 1820 error = sync_vnode(gslp, &bo, td); 1821 if (error == 1) { 1822 LIST_REMOVE(bo, bo_synclist); 1823 LIST_INSERT_HEAD(gnext, bo, 1824 bo_synclist); 1825 continue; 1826 } 1827 } 1828 mtx_unlock(&Giant); 1829 } 1830 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) 1831 syncer_final_iter--; 1832 /* 1833 * The variable rushjob allows the kernel to speed up the 1834 * processing of the filesystem syncer process. A rushjob 1835 * value of N tells the filesystem syncer to process the next 1836 * N seconds worth of work on its queue ASAP. Currently rushjob 1837 * is used by the soft update code to speed up the filesystem 1838 * syncer process when the incore state is getting so far 1839 * ahead of the disk that the kernel memory pool is being 1840 * threatened with exhaustion. 1841 */ 1842 if (rushjob > 0) { 1843 rushjob -= 1; 1844 continue; 1845 } 1846 /* 1847 * Just sleep for a short period of time between 1848 * iterations when shutting down to allow some I/O 1849 * to happen. 1850 * 1851 * If it has taken us less than a second to process the 1852 * current work, then wait. Otherwise start right over 1853 * again. We can still lose time if any single round 1854 * takes more than two seconds, but it does not really 1855 * matter as we are just trying to generally pace the 1856 * filesystem activity. 1857 */ 1858 if (syncer_state != SYNCER_RUNNING) 1859 cv_timedwait(&sync_wakeup, &sync_mtx, 1860 hz / SYNCER_SHUTDOWN_SPEEDUP); 1861 else if (time_uptime == starttime) 1862 cv_timedwait(&sync_wakeup, &sync_mtx, hz); 1863 } 1864 } 1865 1866 /* 1867 * Request the syncer daemon to speed up its work. 1868 * We never push it to speed up more than half of its 1869 * normal turn time, otherwise it could take over the cpu. 1870 */ 1871 int 1872 speedup_syncer(void) 1873 { 1874 int ret = 0; 1875 1876 mtx_lock(&sync_mtx); 1877 if (rushjob < syncdelay / 2) { 1878 rushjob += 1; 1879 stat_rush_requests += 1; 1880 ret = 1; 1881 } 1882 mtx_unlock(&sync_mtx); 1883 cv_broadcast(&sync_wakeup); 1884 return (ret); 1885 } 1886 1887 /* 1888 * Tell the syncer to speed up its work and run though its work 1889 * list several times, then tell it to shut down. 1890 */ 1891 static void 1892 syncer_shutdown(void *arg, int howto) 1893 { 1894 1895 if (howto & RB_NOSYNC) 1896 return; 1897 mtx_lock(&sync_mtx); 1898 syncer_state = SYNCER_SHUTTING_DOWN; 1899 rushjob = 0; 1900 mtx_unlock(&sync_mtx); 1901 cv_broadcast(&sync_wakeup); 1902 kproc_shutdown(arg, howto); 1903 } 1904 1905 /* 1906 * Reassign a buffer from one vnode to another. 1907 * Used to assign file specific control information 1908 * (indirect blocks) to the vnode to which they belong. 1909 */ 1910 void 1911 reassignbuf(struct buf *bp) 1912 { 1913 struct vnode *vp; 1914 struct bufobj *bo; 1915 int delay; 1916 #ifdef INVARIANTS 1917 struct bufv *bv; 1918 #endif 1919 1920 vp = bp->b_vp; 1921 bo = bp->b_bufobj; 1922 ++reassignbufcalls; 1923 1924 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", 1925 bp, bp->b_vp, bp->b_flags); 1926 /* 1927 * B_PAGING flagged buffers cannot be reassigned because their vp 1928 * is not fully linked in. 1929 */ 1930 if (bp->b_flags & B_PAGING) 1931 panic("cannot reassign paging buffer"); 1932 1933 /* 1934 * Delete from old vnode list, if on one. 1935 */ 1936 BO_LOCK(bo); 1937 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1938 buf_vlist_remove(bp); 1939 else 1940 panic("reassignbuf: Buffer %p not on queue.", bp); 1941 /* 1942 * If dirty, put on list of dirty buffers; otherwise insert onto list 1943 * of clean buffers. 1944 */ 1945 if (bp->b_flags & B_DELWRI) { 1946 if ((bo->bo_flag & BO_ONWORKLST) == 0) { 1947 switch (vp->v_type) { 1948 case VDIR: 1949 delay = dirdelay; 1950 break; 1951 case VCHR: 1952 delay = metadelay; 1953 break; 1954 default: 1955 delay = filedelay; 1956 } 1957 vn_syncer_add_to_worklist(bo, delay); 1958 } 1959 buf_vlist_add(bp, bo, BX_VNDIRTY); 1960 } else { 1961 buf_vlist_add(bp, bo, BX_VNCLEAN); 1962 1963 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 1964 mtx_lock(&sync_mtx); 1965 LIST_REMOVE(bo, bo_synclist); 1966 syncer_worklist_len--; 1967 mtx_unlock(&sync_mtx); 1968 bo->bo_flag &= ~BO_ONWORKLST; 1969 } 1970 } 1971 #ifdef INVARIANTS 1972 bv = &bo->bo_clean; 1973 bp = TAILQ_FIRST(&bv->bv_hd); 1974 KASSERT(bp == NULL || bp->b_bufobj == bo, 1975 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 1976 bp = TAILQ_LAST(&bv->bv_hd, buflists); 1977 KASSERT(bp == NULL || bp->b_bufobj == bo, 1978 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 1979 bv = &bo->bo_dirty; 1980 bp = TAILQ_FIRST(&bv->bv_hd); 1981 KASSERT(bp == NULL || bp->b_bufobj == bo, 1982 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 1983 bp = TAILQ_LAST(&bv->bv_hd, buflists); 1984 KASSERT(bp == NULL || bp->b_bufobj == bo, 1985 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 1986 #endif 1987 BO_UNLOCK(bo); 1988 } 1989 1990 /* 1991 * Increment the use and hold counts on the vnode, taking care to reference 1992 * the driver's usecount if this is a chardev. The vholdl() will remove 1993 * the vnode from the free list if it is presently free. Requires the 1994 * vnode interlock and returns with it held. 1995 */ 1996 static void 1997 v_incr_usecount(struct vnode *vp) 1998 { 1999 2000 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2001 vp->v_usecount++; 2002 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2003 dev_lock(); 2004 vp->v_rdev->si_usecount++; 2005 dev_unlock(); 2006 } 2007 vholdl(vp); 2008 } 2009 2010 /* 2011 * Turn a holdcnt into a use+holdcnt such that only one call to 2012 * v_decr_usecount is needed. 2013 */ 2014 static void 2015 v_upgrade_usecount(struct vnode *vp) 2016 { 2017 2018 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2019 vp->v_usecount++; 2020 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2021 dev_lock(); 2022 vp->v_rdev->si_usecount++; 2023 dev_unlock(); 2024 } 2025 } 2026 2027 /* 2028 * Decrement the vnode use and hold count along with the driver's usecount 2029 * if this is a chardev. The vdropl() below releases the vnode interlock 2030 * as it may free the vnode. 2031 */ 2032 static void 2033 v_decr_usecount(struct vnode *vp) 2034 { 2035 2036 ASSERT_VI_LOCKED(vp, __FUNCTION__); 2037 VNASSERT(vp->v_usecount > 0, vp, 2038 ("v_decr_usecount: negative usecount")); 2039 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2040 vp->v_usecount--; 2041 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2042 dev_lock(); 2043 vp->v_rdev->si_usecount--; 2044 dev_unlock(); 2045 } 2046 vdropl(vp); 2047 } 2048 2049 /* 2050 * Decrement only the use count and driver use count. This is intended to 2051 * be paired with a follow on vdropl() to release the remaining hold count. 2052 * In this way we may vgone() a vnode with a 0 usecount without risk of 2053 * having it end up on a free list because the hold count is kept above 0. 2054 */ 2055 static void 2056 v_decr_useonly(struct vnode *vp) 2057 { 2058 2059 ASSERT_VI_LOCKED(vp, __FUNCTION__); 2060 VNASSERT(vp->v_usecount > 0, vp, 2061 ("v_decr_useonly: negative usecount")); 2062 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2063 vp->v_usecount--; 2064 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2065 dev_lock(); 2066 vp->v_rdev->si_usecount--; 2067 dev_unlock(); 2068 } 2069 } 2070 2071 /* 2072 * Grab a particular vnode from the free list, increment its 2073 * reference count and lock it. VI_DOOMED is set if the vnode 2074 * is being destroyed. Only callers who specify LK_RETRY will 2075 * see doomed vnodes. If inactive processing was delayed in 2076 * vput try to do it here. 2077 */ 2078 int 2079 vget(struct vnode *vp, int flags, struct thread *td) 2080 { 2081 int error; 2082 2083 error = 0; 2084 VFS_ASSERT_GIANT(vp->v_mount); 2085 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 2086 ("vget: invalid lock operation")); 2087 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); 2088 2089 if ((flags & LK_INTERLOCK) == 0) 2090 VI_LOCK(vp); 2091 vholdl(vp); 2092 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) { 2093 vdrop(vp); 2094 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__, 2095 vp); 2096 return (error); 2097 } 2098 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) 2099 panic("vget: vn_lock failed to return ENOENT\n"); 2100 VI_LOCK(vp); 2101 /* Upgrade our holdcnt to a usecount. */ 2102 v_upgrade_usecount(vp); 2103 /* 2104 * We don't guarantee that any particular close will 2105 * trigger inactive processing so just make a best effort 2106 * here at preventing a reference to a removed file. If 2107 * we don't succeed no harm is done. 2108 */ 2109 if (vp->v_iflag & VI_OWEINACT) { 2110 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && 2111 (flags & LK_NOWAIT) == 0) 2112 vinactive(vp, td); 2113 vp->v_iflag &= ~VI_OWEINACT; 2114 } 2115 VI_UNLOCK(vp); 2116 return (0); 2117 } 2118 2119 /* 2120 * Increase the reference count of a vnode. 2121 */ 2122 void 2123 vref(struct vnode *vp) 2124 { 2125 2126 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2127 VI_LOCK(vp); 2128 v_incr_usecount(vp); 2129 VI_UNLOCK(vp); 2130 } 2131 2132 /* 2133 * Return reference count of a vnode. 2134 * 2135 * The results of this call are only guaranteed when some mechanism other 2136 * than the VI lock is used to stop other processes from gaining references 2137 * to the vnode. This may be the case if the caller holds the only reference. 2138 * This is also useful when stale data is acceptable as race conditions may 2139 * be accounted for by some other means. 2140 */ 2141 int 2142 vrefcnt(struct vnode *vp) 2143 { 2144 int usecnt; 2145 2146 VI_LOCK(vp); 2147 usecnt = vp->v_usecount; 2148 VI_UNLOCK(vp); 2149 2150 return (usecnt); 2151 } 2152 2153 2154 /* 2155 * Vnode put/release. 2156 * If count drops to zero, call inactive routine and return to freelist. 2157 */ 2158 void 2159 vrele(struct vnode *vp) 2160 { 2161 struct thread *td = curthread; /* XXX */ 2162 2163 KASSERT(vp != NULL, ("vrele: null vp")); 2164 VFS_ASSERT_GIANT(vp->v_mount); 2165 2166 VI_LOCK(vp); 2167 2168 /* Skip this v_writecount check if we're going to panic below. */ 2169 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp, 2170 ("vrele: missed vn_close")); 2171 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2172 2173 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) && 2174 vp->v_usecount == 1)) { 2175 v_decr_usecount(vp); 2176 return; 2177 } 2178 if (vp->v_usecount != 1) { 2179 #ifdef DIAGNOSTIC 2180 vprint("vrele: negative ref count", vp); 2181 #endif 2182 VI_UNLOCK(vp); 2183 panic("vrele: negative ref cnt"); 2184 } 2185 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp); 2186 /* 2187 * We want to hold the vnode until the inactive finishes to 2188 * prevent vgone() races. We drop the use count here and the 2189 * hold count below when we're done. 2190 */ 2191 v_decr_useonly(vp); 2192 /* 2193 * We must call VOP_INACTIVE with the node locked. Mark 2194 * as VI_DOINGINACT to avoid recursion. 2195 */ 2196 vp->v_iflag |= VI_OWEINACT; 2197 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0) { 2198 VI_LOCK(vp); 2199 if (vp->v_usecount > 0) 2200 vp->v_iflag &= ~VI_OWEINACT; 2201 if (vp->v_iflag & VI_OWEINACT) 2202 vinactive(vp, td); 2203 VOP_UNLOCK(vp, 0); 2204 } else { 2205 VI_LOCK(vp); 2206 if (vp->v_usecount > 0) 2207 vp->v_iflag &= ~VI_OWEINACT; 2208 } 2209 vdropl(vp); 2210 } 2211 2212 /* 2213 * Release an already locked vnode. This give the same effects as 2214 * unlock+vrele(), but takes less time and avoids releasing and 2215 * re-aquiring the lock (as vrele() acquires the lock internally.) 2216 */ 2217 void 2218 vput(struct vnode *vp) 2219 { 2220 struct thread *td = curthread; /* XXX */ 2221 int error; 2222 2223 KASSERT(vp != NULL, ("vput: null vp")); 2224 ASSERT_VOP_LOCKED(vp, "vput"); 2225 VFS_ASSERT_GIANT(vp->v_mount); 2226 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2227 VI_LOCK(vp); 2228 /* Skip this v_writecount check if we're going to panic below. */ 2229 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp, 2230 ("vput: missed vn_close")); 2231 error = 0; 2232 2233 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) && 2234 vp->v_usecount == 1)) { 2235 VOP_UNLOCK(vp, 0); 2236 v_decr_usecount(vp); 2237 return; 2238 } 2239 2240 if (vp->v_usecount != 1) { 2241 #ifdef DIAGNOSTIC 2242 vprint("vput: negative ref count", vp); 2243 #endif 2244 panic("vput: negative ref cnt"); 2245 } 2246 CTR2(KTR_VFS, "%s: return to freelist the vnode %p", __func__, vp); 2247 /* 2248 * We want to hold the vnode until the inactive finishes to 2249 * prevent vgone() races. We drop the use count here and the 2250 * hold count below when we're done. 2251 */ 2252 v_decr_useonly(vp); 2253 vp->v_iflag |= VI_OWEINACT; 2254 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 2255 error = VOP_LOCK(vp, LK_UPGRADE|LK_INTERLOCK|LK_NOWAIT); 2256 VI_LOCK(vp); 2257 if (error) { 2258 if (vp->v_usecount > 0) 2259 vp->v_iflag &= ~VI_OWEINACT; 2260 goto done; 2261 } 2262 } 2263 if (vp->v_usecount > 0) 2264 vp->v_iflag &= ~VI_OWEINACT; 2265 if (vp->v_iflag & VI_OWEINACT) 2266 vinactive(vp, td); 2267 VOP_UNLOCK(vp, 0); 2268 done: 2269 vdropl(vp); 2270 } 2271 2272 /* 2273 * Somebody doesn't want the vnode recycled. 2274 */ 2275 void 2276 vhold(struct vnode *vp) 2277 { 2278 2279 VI_LOCK(vp); 2280 vholdl(vp); 2281 VI_UNLOCK(vp); 2282 } 2283 2284 void 2285 vholdl(struct vnode *vp) 2286 { 2287 2288 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2289 vp->v_holdcnt++; 2290 if (VSHOULDBUSY(vp)) 2291 vbusy(vp); 2292 } 2293 2294 /* 2295 * Note that there is one less who cares about this vnode. vdrop() is the 2296 * opposite of vhold(). 2297 */ 2298 void 2299 vdrop(struct vnode *vp) 2300 { 2301 2302 VI_LOCK(vp); 2303 vdropl(vp); 2304 } 2305 2306 /* 2307 * Drop the hold count of the vnode. If this is the last reference to 2308 * the vnode we will free it if it has been vgone'd otherwise it is 2309 * placed on the free list. 2310 */ 2311 void 2312 vdropl(struct vnode *vp) 2313 { 2314 2315 ASSERT_VI_LOCKED(vp, "vdropl"); 2316 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2317 if (vp->v_holdcnt <= 0) 2318 panic("vdrop: holdcnt %d", vp->v_holdcnt); 2319 vp->v_holdcnt--; 2320 if (vp->v_holdcnt == 0) { 2321 if (vp->v_iflag & VI_DOOMED) { 2322 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, 2323 vp); 2324 vdestroy(vp); 2325 return; 2326 } else 2327 vfree(vp); 2328 } 2329 VI_UNLOCK(vp); 2330 } 2331 2332 /* 2333 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT 2334 * flags. DOINGINACT prevents us from recursing in calls to vinactive. 2335 * OWEINACT tracks whether a vnode missed a call to inactive due to a 2336 * failed lock upgrade. 2337 */ 2338 static void 2339 vinactive(struct vnode *vp, struct thread *td) 2340 { 2341 2342 ASSERT_VOP_ELOCKED(vp, "vinactive"); 2343 ASSERT_VI_LOCKED(vp, "vinactive"); 2344 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, 2345 ("vinactive: recursed on VI_DOINGINACT")); 2346 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2347 vp->v_iflag |= VI_DOINGINACT; 2348 vp->v_iflag &= ~VI_OWEINACT; 2349 VI_UNLOCK(vp); 2350 VOP_INACTIVE(vp, td); 2351 VI_LOCK(vp); 2352 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, 2353 ("vinactive: lost VI_DOINGINACT")); 2354 vp->v_iflag &= ~VI_DOINGINACT; 2355 } 2356 2357 /* 2358 * Remove any vnodes in the vnode table belonging to mount point mp. 2359 * 2360 * If FORCECLOSE is not specified, there should not be any active ones, 2361 * return error if any are found (nb: this is a user error, not a 2362 * system error). If FORCECLOSE is specified, detach any active vnodes 2363 * that are found. 2364 * 2365 * If WRITECLOSE is set, only flush out regular file vnodes open for 2366 * writing. 2367 * 2368 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. 2369 * 2370 * `rootrefs' specifies the base reference count for the root vnode 2371 * of this filesystem. The root vnode is considered busy if its 2372 * v_usecount exceeds this value. On a successful return, vflush(, td) 2373 * will call vrele() on the root vnode exactly rootrefs times. 2374 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 2375 * be zero. 2376 */ 2377 #ifdef DIAGNOSTIC 2378 static int busyprt = 0; /* print out busy vnodes */ 2379 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 2380 #endif 2381 2382 int 2383 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td) 2384 { 2385 struct vnode *vp, *mvp, *rootvp = NULL; 2386 struct vattr vattr; 2387 int busy = 0, error; 2388 2389 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp, 2390 rootrefs, flags); 2391 if (rootrefs > 0) { 2392 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 2393 ("vflush: bad args")); 2394 /* 2395 * Get the filesystem root vnode. We can vput() it 2396 * immediately, since with rootrefs > 0, it won't go away. 2397 */ 2398 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp, td)) != 0) { 2399 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d", 2400 __func__, error); 2401 return (error); 2402 } 2403 vput(rootvp); 2404 2405 } 2406 MNT_ILOCK(mp); 2407 loop: 2408 MNT_VNODE_FOREACH(vp, mp, mvp) { 2409 2410 VI_LOCK(vp); 2411 vholdl(vp); 2412 MNT_IUNLOCK(mp); 2413 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE); 2414 if (error) { 2415 vdrop(vp); 2416 MNT_ILOCK(mp); 2417 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); 2418 goto loop; 2419 } 2420 /* 2421 * Skip over a vnodes marked VV_SYSTEM. 2422 */ 2423 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { 2424 VOP_UNLOCK(vp, 0); 2425 vdrop(vp); 2426 MNT_ILOCK(mp); 2427 continue; 2428 } 2429 /* 2430 * If WRITECLOSE is set, flush out unlinked but still open 2431 * files (even if open only for reading) and regular file 2432 * vnodes open for writing. 2433 */ 2434 if (flags & WRITECLOSE) { 2435 error = VOP_GETATTR(vp, &vattr, td->td_ucred); 2436 VI_LOCK(vp); 2437 2438 if ((vp->v_type == VNON || 2439 (error == 0 && vattr.va_nlink > 0)) && 2440 (vp->v_writecount == 0 || vp->v_type != VREG)) { 2441 VOP_UNLOCK(vp, 0); 2442 vdropl(vp); 2443 MNT_ILOCK(mp); 2444 continue; 2445 } 2446 } else 2447 VI_LOCK(vp); 2448 /* 2449 * With v_usecount == 0, all we need to do is clear out the 2450 * vnode data structures and we are done. 2451 * 2452 * If FORCECLOSE is set, forcibly close the vnode. 2453 */ 2454 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { 2455 VNASSERT(vp->v_usecount == 0 || 2456 (vp->v_type != VCHR && vp->v_type != VBLK), vp, 2457 ("device VNODE %p is FORCECLOSED", vp)); 2458 vgonel(vp); 2459 } else { 2460 busy++; 2461 #ifdef DIAGNOSTIC 2462 if (busyprt) 2463 vprint("vflush: busy vnode", vp); 2464 #endif 2465 } 2466 VOP_UNLOCK(vp, 0); 2467 vdropl(vp); 2468 MNT_ILOCK(mp); 2469 } 2470 MNT_IUNLOCK(mp); 2471 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 2472 /* 2473 * If just the root vnode is busy, and if its refcount 2474 * is equal to `rootrefs', then go ahead and kill it. 2475 */ 2476 VI_LOCK(rootvp); 2477 KASSERT(busy > 0, ("vflush: not busy")); 2478 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, 2479 ("vflush: usecount %d < rootrefs %d", 2480 rootvp->v_usecount, rootrefs)); 2481 if (busy == 1 && rootvp->v_usecount == rootrefs) { 2482 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK); 2483 vgone(rootvp); 2484 VOP_UNLOCK(rootvp, 0); 2485 busy = 0; 2486 } else 2487 VI_UNLOCK(rootvp); 2488 } 2489 if (busy) { 2490 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__, 2491 busy); 2492 return (EBUSY); 2493 } 2494 for (; rootrefs > 0; rootrefs--) 2495 vrele(rootvp); 2496 return (0); 2497 } 2498 2499 /* 2500 * Recycle an unused vnode to the front of the free list. 2501 */ 2502 int 2503 vrecycle(struct vnode *vp, struct thread *td) 2504 { 2505 int recycled; 2506 2507 ASSERT_VOP_ELOCKED(vp, "vrecycle"); 2508 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2509 recycled = 0; 2510 VI_LOCK(vp); 2511 if (vp->v_usecount == 0) { 2512 recycled = 1; 2513 vgonel(vp); 2514 } 2515 VI_UNLOCK(vp); 2516 return (recycled); 2517 } 2518 2519 /* 2520 * Eliminate all activity associated with a vnode 2521 * in preparation for reuse. 2522 */ 2523 void 2524 vgone(struct vnode *vp) 2525 { 2526 VI_LOCK(vp); 2527 vgonel(vp); 2528 VI_UNLOCK(vp); 2529 } 2530 2531 /* 2532 * vgone, with the vp interlock held. 2533 */ 2534 void 2535 vgonel(struct vnode *vp) 2536 { 2537 struct thread *td; 2538 int oweinact; 2539 int active; 2540 struct mount *mp; 2541 2542 ASSERT_VOP_ELOCKED(vp, "vgonel"); 2543 ASSERT_VI_LOCKED(vp, "vgonel"); 2544 VNASSERT(vp->v_holdcnt, vp, 2545 ("vgonel: vp %p has no reference.", vp)); 2546 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2547 td = curthread; 2548 2549 /* 2550 * Don't vgonel if we're already doomed. 2551 */ 2552 if (vp->v_iflag & VI_DOOMED) 2553 return; 2554 vp->v_iflag |= VI_DOOMED; 2555 /* 2556 * Check to see if the vnode is in use. If so, we have to call 2557 * VOP_CLOSE() and VOP_INACTIVE(). 2558 */ 2559 active = vp->v_usecount; 2560 oweinact = (vp->v_iflag & VI_OWEINACT); 2561 VI_UNLOCK(vp); 2562 /* 2563 * Clean out any buffers associated with the vnode. 2564 * If the flush fails, just toss the buffers. 2565 */ 2566 mp = NULL; 2567 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) 2568 (void) vn_start_secondary_write(vp, &mp, V_WAIT); 2569 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) 2570 vinvalbuf(vp, 0, 0, 0); 2571 2572 /* 2573 * If purging an active vnode, it must be closed and 2574 * deactivated before being reclaimed. 2575 */ 2576 if (active) 2577 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); 2578 if (oweinact || active) { 2579 VI_LOCK(vp); 2580 if ((vp->v_iflag & VI_DOINGINACT) == 0) 2581 vinactive(vp, td); 2582 VI_UNLOCK(vp); 2583 } 2584 /* 2585 * Reclaim the vnode. 2586 */ 2587 if (VOP_RECLAIM(vp, td)) 2588 panic("vgone: cannot reclaim"); 2589 if (mp != NULL) 2590 vn_finished_secondary_write(mp); 2591 VNASSERT(vp->v_object == NULL, vp, 2592 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag)); 2593 /* 2594 * Clear the advisory locks and wake up waiting threads. 2595 */ 2596 lf_purgelocks(vp, &(vp->v_lockf)); 2597 /* 2598 * Delete from old mount point vnode list. 2599 */ 2600 delmntque(vp); 2601 cache_purge(vp); 2602 /* 2603 * Done with purge, reset to the standard lock and invalidate 2604 * the vnode. 2605 */ 2606 VI_LOCK(vp); 2607 vp->v_vnlock = &vp->v_lock; 2608 vp->v_op = &dead_vnodeops; 2609 vp->v_tag = "none"; 2610 vp->v_type = VBAD; 2611 } 2612 2613 /* 2614 * Calculate the total number of references to a special device. 2615 */ 2616 int 2617 vcount(struct vnode *vp) 2618 { 2619 int count; 2620 2621 dev_lock(); 2622 count = vp->v_rdev->si_usecount; 2623 dev_unlock(); 2624 return (count); 2625 } 2626 2627 /* 2628 * Same as above, but using the struct cdev *as argument 2629 */ 2630 int 2631 count_dev(struct cdev *dev) 2632 { 2633 int count; 2634 2635 dev_lock(); 2636 count = dev->si_usecount; 2637 dev_unlock(); 2638 return(count); 2639 } 2640 2641 /* 2642 * Print out a description of a vnode. 2643 */ 2644 static char *typename[] = 2645 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", 2646 "VMARKER"}; 2647 2648 void 2649 vn_printf(struct vnode *vp, const char *fmt, ...) 2650 { 2651 va_list ap; 2652 char buf[256], buf2[16]; 2653 u_long flags; 2654 2655 va_start(ap, fmt); 2656 vprintf(fmt, ap); 2657 va_end(ap); 2658 printf("%p: ", (void *)vp); 2659 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]); 2660 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n", 2661 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere); 2662 buf[0] = '\0'; 2663 buf[1] = '\0'; 2664 if (vp->v_vflag & VV_ROOT) 2665 strlcat(buf, "|VV_ROOT", sizeof(buf)); 2666 if (vp->v_vflag & VV_ISTTY) 2667 strlcat(buf, "|VV_ISTTY", sizeof(buf)); 2668 if (vp->v_vflag & VV_NOSYNC) 2669 strlcat(buf, "|VV_NOSYNC", sizeof(buf)); 2670 if (vp->v_vflag & VV_CACHEDLABEL) 2671 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf)); 2672 if (vp->v_vflag & VV_TEXT) 2673 strlcat(buf, "|VV_TEXT", sizeof(buf)); 2674 if (vp->v_vflag & VV_COPYONWRITE) 2675 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf)); 2676 if (vp->v_vflag & VV_SYSTEM) 2677 strlcat(buf, "|VV_SYSTEM", sizeof(buf)); 2678 if (vp->v_vflag & VV_PROCDEP) 2679 strlcat(buf, "|VV_PROCDEP", sizeof(buf)); 2680 if (vp->v_vflag & VV_NOKNOTE) 2681 strlcat(buf, "|VV_NOKNOTE", sizeof(buf)); 2682 if (vp->v_vflag & VV_DELETED) 2683 strlcat(buf, "|VV_DELETED", sizeof(buf)); 2684 if (vp->v_vflag & VV_MD) 2685 strlcat(buf, "|VV_MD", sizeof(buf)); 2686 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | 2687 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP | 2688 VV_NOKNOTE | VV_DELETED | VV_MD); 2689 if (flags != 0) { 2690 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags); 2691 strlcat(buf, buf2, sizeof(buf)); 2692 } 2693 if (vp->v_iflag & VI_MOUNT) 2694 strlcat(buf, "|VI_MOUNT", sizeof(buf)); 2695 if (vp->v_iflag & VI_AGE) 2696 strlcat(buf, "|VI_AGE", sizeof(buf)); 2697 if (vp->v_iflag & VI_DOOMED) 2698 strlcat(buf, "|VI_DOOMED", sizeof(buf)); 2699 if (vp->v_iflag & VI_FREE) 2700 strlcat(buf, "|VI_FREE", sizeof(buf)); 2701 if (vp->v_iflag & VI_OBJDIRTY) 2702 strlcat(buf, "|VI_OBJDIRTY", sizeof(buf)); 2703 if (vp->v_iflag & VI_DOINGINACT) 2704 strlcat(buf, "|VI_DOINGINACT", sizeof(buf)); 2705 if (vp->v_iflag & VI_OWEINACT) 2706 strlcat(buf, "|VI_OWEINACT", sizeof(buf)); 2707 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE | 2708 VI_OBJDIRTY | VI_DOINGINACT | VI_OWEINACT); 2709 if (flags != 0) { 2710 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags); 2711 strlcat(buf, buf2, sizeof(buf)); 2712 } 2713 printf(" flags (%s)\n", buf + 1); 2714 if (mtx_owned(VI_MTX(vp))) 2715 printf(" VI_LOCKed"); 2716 if (vp->v_object != NULL) 2717 printf(" v_object %p ref %d pages %d\n", 2718 vp->v_object, vp->v_object->ref_count, 2719 vp->v_object->resident_page_count); 2720 printf(" "); 2721 lockmgr_printinfo(vp->v_vnlock); 2722 if (vp->v_data != NULL) 2723 VOP_PRINT(vp); 2724 } 2725 2726 #ifdef DDB 2727 /* 2728 * List all of the locked vnodes in the system. 2729 * Called when debugging the kernel. 2730 */ 2731 DB_SHOW_COMMAND(lockedvnods, lockedvnodes) 2732 { 2733 struct mount *mp, *nmp; 2734 struct vnode *vp; 2735 2736 /* 2737 * Note: because this is DDB, we can't obey the locking semantics 2738 * for these structures, which means we could catch an inconsistent 2739 * state and dereference a nasty pointer. Not much to be done 2740 * about that. 2741 */ 2742 db_printf("Locked vnodes\n"); 2743 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2744 nmp = TAILQ_NEXT(mp, mnt_list); 2745 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2746 if (vp->v_type != VMARKER && 2747 VOP_ISLOCKED(vp)) 2748 vprint("", vp); 2749 } 2750 nmp = TAILQ_NEXT(mp, mnt_list); 2751 } 2752 } 2753 2754 /* 2755 * Show details about the given vnode. 2756 */ 2757 DB_SHOW_COMMAND(vnode, db_show_vnode) 2758 { 2759 struct vnode *vp; 2760 2761 if (!have_addr) 2762 return; 2763 vp = (struct vnode *)addr; 2764 vn_printf(vp, "vnode "); 2765 } 2766 2767 /* 2768 * Show details about the given mount point. 2769 */ 2770 DB_SHOW_COMMAND(mount, db_show_mount) 2771 { 2772 struct mount *mp; 2773 struct statfs *sp; 2774 struct vnode *vp; 2775 char buf[512]; 2776 u_int flags; 2777 2778 if (!have_addr) { 2779 /* No address given, print short info about all mount points. */ 2780 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2781 db_printf("%p %s on %s (%s)\n", mp, 2782 mp->mnt_stat.f_mntfromname, 2783 mp->mnt_stat.f_mntonname, 2784 mp->mnt_stat.f_fstypename); 2785 if (db_pager_quit) 2786 break; 2787 } 2788 db_printf("\nMore info: show mount <addr>\n"); 2789 return; 2790 } 2791 2792 mp = (struct mount *)addr; 2793 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, 2794 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); 2795 2796 buf[0] = '\0'; 2797 flags = mp->mnt_flag; 2798 #define MNT_FLAG(flag) do { \ 2799 if (flags & (flag)) { \ 2800 if (buf[0] != '\0') \ 2801 strlcat(buf, ", ", sizeof(buf)); \ 2802 strlcat(buf, (#flag) + 4, sizeof(buf)); \ 2803 flags &= ~(flag); \ 2804 } \ 2805 } while (0) 2806 MNT_FLAG(MNT_RDONLY); 2807 MNT_FLAG(MNT_SYNCHRONOUS); 2808 MNT_FLAG(MNT_NOEXEC); 2809 MNT_FLAG(MNT_NOSUID); 2810 MNT_FLAG(MNT_UNION); 2811 MNT_FLAG(MNT_ASYNC); 2812 MNT_FLAG(MNT_SUIDDIR); 2813 MNT_FLAG(MNT_SOFTDEP); 2814 MNT_FLAG(MNT_NOSYMFOLLOW); 2815 MNT_FLAG(MNT_GJOURNAL); 2816 MNT_FLAG(MNT_MULTILABEL); 2817 MNT_FLAG(MNT_ACLS); 2818 MNT_FLAG(MNT_NOATIME); 2819 MNT_FLAG(MNT_NOCLUSTERR); 2820 MNT_FLAG(MNT_NOCLUSTERW); 2821 MNT_FLAG(MNT_EXRDONLY); 2822 MNT_FLAG(MNT_EXPORTED); 2823 MNT_FLAG(MNT_DEFEXPORTED); 2824 MNT_FLAG(MNT_EXPORTANON); 2825 MNT_FLAG(MNT_EXKERB); 2826 MNT_FLAG(MNT_EXPUBLIC); 2827 MNT_FLAG(MNT_LOCAL); 2828 MNT_FLAG(MNT_QUOTA); 2829 MNT_FLAG(MNT_ROOTFS); 2830 MNT_FLAG(MNT_USER); 2831 MNT_FLAG(MNT_IGNORE); 2832 MNT_FLAG(MNT_UPDATE); 2833 MNT_FLAG(MNT_DELEXPORT); 2834 MNT_FLAG(MNT_RELOAD); 2835 MNT_FLAG(MNT_FORCE); 2836 MNT_FLAG(MNT_SNAPSHOT); 2837 MNT_FLAG(MNT_BYFSID); 2838 #undef MNT_FLAG 2839 if (flags != 0) { 2840 if (buf[0] != '\0') 2841 strlcat(buf, ", ", sizeof(buf)); 2842 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), 2843 "0x%08x", flags); 2844 } 2845 db_printf(" mnt_flag = %s\n", buf); 2846 2847 buf[0] = '\0'; 2848 flags = mp->mnt_kern_flag; 2849 #define MNT_KERN_FLAG(flag) do { \ 2850 if (flags & (flag)) { \ 2851 if (buf[0] != '\0') \ 2852 strlcat(buf, ", ", sizeof(buf)); \ 2853 strlcat(buf, (#flag) + 5, sizeof(buf)); \ 2854 flags &= ~(flag); \ 2855 } \ 2856 } while (0) 2857 MNT_KERN_FLAG(MNTK_UNMOUNTF); 2858 MNT_KERN_FLAG(MNTK_ASYNC); 2859 MNT_KERN_FLAG(MNTK_SOFTDEP); 2860 MNT_KERN_FLAG(MNTK_NOINSMNTQ); 2861 MNT_KERN_FLAG(MNTK_UNMOUNT); 2862 MNT_KERN_FLAG(MNTK_MWAIT); 2863 MNT_KERN_FLAG(MNTK_SUSPEND); 2864 MNT_KERN_FLAG(MNTK_SUSPEND2); 2865 MNT_KERN_FLAG(MNTK_SUSPENDED); 2866 MNT_KERN_FLAG(MNTK_MPSAFE); 2867 MNT_KERN_FLAG(MNTK_NOKNOTE); 2868 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED); 2869 #undef MNT_KERN_FLAG 2870 if (flags != 0) { 2871 if (buf[0] != '\0') 2872 strlcat(buf, ", ", sizeof(buf)); 2873 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), 2874 "0x%08x", flags); 2875 } 2876 db_printf(" mnt_kern_flag = %s\n", buf); 2877 2878 sp = &mp->mnt_stat; 2879 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx " 2880 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju " 2881 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju " 2882 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n", 2883 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags, 2884 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize, 2885 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree, 2886 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files, 2887 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites, 2888 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads, 2889 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax, 2890 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]); 2891 2892 db_printf(" mnt_cred = { uid=%u ruid=%u", 2893 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid); 2894 if (mp->mnt_cred->cr_prison != NULL) 2895 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id); 2896 db_printf(" }\n"); 2897 db_printf(" mnt_ref = %d\n", mp->mnt_ref); 2898 db_printf(" mnt_gen = %d\n", mp->mnt_gen); 2899 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize); 2900 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount); 2901 db_printf(" mnt_noasync = %u\n", mp->mnt_noasync); 2902 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen); 2903 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max); 2904 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed); 2905 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes); 2906 db_printf(" mnt_secondary_accwrites = %d\n", 2907 mp->mnt_secondary_accwrites); 2908 db_printf(" mnt_gjprovider = %s\n", 2909 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL"); 2910 db_printf("\n"); 2911 2912 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2913 if (vp->v_type != VMARKER) { 2914 vn_printf(vp, "vnode "); 2915 if (db_pager_quit) 2916 break; 2917 } 2918 } 2919 } 2920 #endif /* DDB */ 2921 2922 /* 2923 * Fill in a struct xvfsconf based on a struct vfsconf. 2924 */ 2925 static void 2926 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp) 2927 { 2928 2929 strcpy(xvfsp->vfc_name, vfsp->vfc_name); 2930 xvfsp->vfc_typenum = vfsp->vfc_typenum; 2931 xvfsp->vfc_refcount = vfsp->vfc_refcount; 2932 xvfsp->vfc_flags = vfsp->vfc_flags; 2933 /* 2934 * These are unused in userland, we keep them 2935 * to not break binary compatibility. 2936 */ 2937 xvfsp->vfc_vfsops = NULL; 2938 xvfsp->vfc_next = NULL; 2939 } 2940 2941 /* 2942 * Top level filesystem related information gathering. 2943 */ 2944 static int 2945 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) 2946 { 2947 struct vfsconf *vfsp; 2948 struct xvfsconf xvfsp; 2949 int error; 2950 2951 error = 0; 2952 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 2953 bzero(&xvfsp, sizeof(xvfsp)); 2954 vfsconf2x(vfsp, &xvfsp); 2955 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp); 2956 if (error) 2957 break; 2958 } 2959 return (error); 2960 } 2961 2962 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist, 2963 "S,xvfsconf", "List of all configured filesystems"); 2964 2965 #ifndef BURN_BRIDGES 2966 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); 2967 2968 static int 2969 vfs_sysctl(SYSCTL_HANDLER_ARGS) 2970 { 2971 int *name = (int *)arg1 - 1; /* XXX */ 2972 u_int namelen = arg2 + 1; /* XXX */ 2973 struct vfsconf *vfsp; 2974 struct xvfsconf xvfsp; 2975 2976 printf("WARNING: userland calling deprecated sysctl, " 2977 "please rebuild world\n"); 2978 2979 #if 1 || defined(COMPAT_PRELITE2) 2980 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 2981 if (namelen == 1) 2982 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 2983 #endif 2984 2985 switch (name[1]) { 2986 case VFS_MAXTYPENUM: 2987 if (namelen != 2) 2988 return (ENOTDIR); 2989 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 2990 case VFS_CONF: 2991 if (namelen != 3) 2992 return (ENOTDIR); /* overloaded */ 2993 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) 2994 if (vfsp->vfc_typenum == name[2]) 2995 break; 2996 if (vfsp == NULL) 2997 return (EOPNOTSUPP); 2998 bzero(&xvfsp, sizeof(xvfsp)); 2999 vfsconf2x(vfsp, &xvfsp); 3000 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); 3001 } 3002 return (EOPNOTSUPP); 3003 } 3004 3005 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, 3006 vfs_sysctl, "Generic filesystem"); 3007 3008 #if 1 || defined(COMPAT_PRELITE2) 3009 3010 static int 3011 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 3012 { 3013 int error; 3014 struct vfsconf *vfsp; 3015 struct ovfsconf ovfs; 3016 3017 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 3018 bzero(&ovfs, sizeof(ovfs)); 3019 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 3020 strcpy(ovfs.vfc_name, vfsp->vfc_name); 3021 ovfs.vfc_index = vfsp->vfc_typenum; 3022 ovfs.vfc_refcount = vfsp->vfc_refcount; 3023 ovfs.vfc_flags = vfsp->vfc_flags; 3024 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 3025 if (error) 3026 return error; 3027 } 3028 return 0; 3029 } 3030 3031 #endif /* 1 || COMPAT_PRELITE2 */ 3032 #endif /* !BURN_BRIDGES */ 3033 3034 #define KINFO_VNODESLOP 10 3035 #ifdef notyet 3036 /* 3037 * Dump vnode list (via sysctl). 3038 */ 3039 /* ARGSUSED */ 3040 static int 3041 sysctl_vnode(SYSCTL_HANDLER_ARGS) 3042 { 3043 struct xvnode *xvn; 3044 struct mount *mp; 3045 struct vnode *vp; 3046 int error, len, n; 3047 3048 /* 3049 * Stale numvnodes access is not fatal here. 3050 */ 3051 req->lock = 0; 3052 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; 3053 if (!req->oldptr) 3054 /* Make an estimate */ 3055 return (SYSCTL_OUT(req, 0, len)); 3056 3057 error = sysctl_wire_old_buffer(req, 0); 3058 if (error != 0) 3059 return (error); 3060 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); 3061 n = 0; 3062 mtx_lock(&mountlist_mtx); 3063 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 3064 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) 3065 continue; 3066 MNT_ILOCK(mp); 3067 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 3068 if (n == len) 3069 break; 3070 vref(vp); 3071 xvn[n].xv_size = sizeof *xvn; 3072 xvn[n].xv_vnode = vp; 3073 xvn[n].xv_id = 0; /* XXX compat */ 3074 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field 3075 XV_COPY(usecount); 3076 XV_COPY(writecount); 3077 XV_COPY(holdcnt); 3078 XV_COPY(mount); 3079 XV_COPY(numoutput); 3080 XV_COPY(type); 3081 #undef XV_COPY 3082 xvn[n].xv_flag = vp->v_vflag; 3083 3084 switch (vp->v_type) { 3085 case VREG: 3086 case VDIR: 3087 case VLNK: 3088 break; 3089 case VBLK: 3090 case VCHR: 3091 if (vp->v_rdev == NULL) { 3092 vrele(vp); 3093 continue; 3094 } 3095 xvn[n].xv_dev = dev2udev(vp->v_rdev); 3096 break; 3097 case VSOCK: 3098 xvn[n].xv_socket = vp->v_socket; 3099 break; 3100 case VFIFO: 3101 xvn[n].xv_fifo = vp->v_fifoinfo; 3102 break; 3103 case VNON: 3104 case VBAD: 3105 default: 3106 /* shouldn't happen? */ 3107 vrele(vp); 3108 continue; 3109 } 3110 vrele(vp); 3111 ++n; 3112 } 3113 MNT_IUNLOCK(mp); 3114 mtx_lock(&mountlist_mtx); 3115 vfs_unbusy(mp); 3116 if (n == len) 3117 break; 3118 } 3119 mtx_unlock(&mountlist_mtx); 3120 3121 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); 3122 free(xvn, M_TEMP); 3123 return (error); 3124 } 3125 3126 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 3127 0, 0, sysctl_vnode, "S,xvnode", ""); 3128 #endif 3129 3130 /* 3131 * Unmount all filesystems. The list is traversed in reverse order 3132 * of mounting to avoid dependencies. 3133 */ 3134 void 3135 vfs_unmountall(void) 3136 { 3137 struct mount *mp; 3138 struct thread *td; 3139 int error; 3140 3141 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread")); 3142 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__); 3143 td = curthread; 3144 3145 /* 3146 * Since this only runs when rebooting, it is not interlocked. 3147 */ 3148 while(!TAILQ_EMPTY(&mountlist)) { 3149 mp = TAILQ_LAST(&mountlist, mntlist); 3150 error = dounmount(mp, MNT_FORCE, td); 3151 if (error) { 3152 TAILQ_REMOVE(&mountlist, mp, mnt_list); 3153 /* 3154 * XXX: Due to the way in which we mount the root 3155 * file system off of devfs, devfs will generate a 3156 * "busy" warning when we try to unmount it before 3157 * the root. Don't print a warning as a result in 3158 * order to avoid false positive errors that may 3159 * cause needless upset. 3160 */ 3161 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) { 3162 printf("unmount of %s failed (", 3163 mp->mnt_stat.f_mntonname); 3164 if (error == EBUSY) 3165 printf("BUSY)\n"); 3166 else 3167 printf("%d)\n", error); 3168 } 3169 } else { 3170 /* The unmount has removed mp from the mountlist */ 3171 } 3172 } 3173 } 3174 3175 /* 3176 * perform msync on all vnodes under a mount point 3177 * the mount point must be locked. 3178 */ 3179 void 3180 vfs_msync(struct mount *mp, int flags) 3181 { 3182 struct vnode *vp, *mvp; 3183 struct vm_object *obj; 3184 3185 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 3186 MNT_ILOCK(mp); 3187 MNT_VNODE_FOREACH(vp, mp, mvp) { 3188 VI_LOCK(vp); 3189 if ((vp->v_iflag & VI_OBJDIRTY) && 3190 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) { 3191 MNT_IUNLOCK(mp); 3192 if (!vget(vp, 3193 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, 3194 curthread)) { 3195 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */ 3196 vput(vp); 3197 MNT_ILOCK(mp); 3198 continue; 3199 } 3200 3201 obj = vp->v_object; 3202 if (obj != NULL) { 3203 VM_OBJECT_LOCK(obj); 3204 vm_object_page_clean(obj, 0, 0, 3205 flags == MNT_WAIT ? 3206 OBJPC_SYNC : OBJPC_NOSYNC); 3207 VM_OBJECT_UNLOCK(obj); 3208 } 3209 vput(vp); 3210 } 3211 MNT_ILOCK(mp); 3212 } else 3213 VI_UNLOCK(vp); 3214 } 3215 MNT_IUNLOCK(mp); 3216 } 3217 3218 /* 3219 * Mark a vnode as free, putting it up for recycling. 3220 */ 3221 static void 3222 vfree(struct vnode *vp) 3223 { 3224 3225 ASSERT_VI_LOCKED(vp, "vfree"); 3226 mtx_lock(&vnode_free_list_mtx); 3227 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed.")); 3228 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free")); 3229 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't")); 3230 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp, 3231 ("vfree: Freeing doomed vnode")); 3232 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 3233 if (vp->v_iflag & VI_AGE) { 3234 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 3235 } else { 3236 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 3237 } 3238 freevnodes++; 3239 vp->v_iflag &= ~VI_AGE; 3240 vp->v_iflag |= VI_FREE; 3241 mtx_unlock(&vnode_free_list_mtx); 3242 } 3243 3244 /* 3245 * Opposite of vfree() - mark a vnode as in use. 3246 */ 3247 static void 3248 vbusy(struct vnode *vp) 3249 { 3250 ASSERT_VI_LOCKED(vp, "vbusy"); 3251 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free")); 3252 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed.")); 3253 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 3254 3255 mtx_lock(&vnode_free_list_mtx); 3256 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 3257 freevnodes--; 3258 vp->v_iflag &= ~(VI_FREE|VI_AGE); 3259 mtx_unlock(&vnode_free_list_mtx); 3260 } 3261 3262 static void 3263 destroy_vpollinfo(struct vpollinfo *vi) 3264 { 3265 knlist_destroy(&vi->vpi_selinfo.si_note); 3266 mtx_destroy(&vi->vpi_lock); 3267 uma_zfree(vnodepoll_zone, vi); 3268 } 3269 3270 /* 3271 * Initalize per-vnode helper structure to hold poll-related state. 3272 */ 3273 void 3274 v_addpollinfo(struct vnode *vp) 3275 { 3276 struct vpollinfo *vi; 3277 3278 if (vp->v_pollinfo != NULL) 3279 return; 3280 vi = uma_zalloc(vnodepoll_zone, M_WAITOK); 3281 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); 3282 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock, 3283 vfs_knlunlock, vfs_knllocked); 3284 VI_LOCK(vp); 3285 if (vp->v_pollinfo != NULL) { 3286 VI_UNLOCK(vp); 3287 destroy_vpollinfo(vi); 3288 return; 3289 } 3290 vp->v_pollinfo = vi; 3291 VI_UNLOCK(vp); 3292 } 3293 3294 /* 3295 * Record a process's interest in events which might happen to 3296 * a vnode. Because poll uses the historic select-style interface 3297 * internally, this routine serves as both the ``check for any 3298 * pending events'' and the ``record my interest in future events'' 3299 * functions. (These are done together, while the lock is held, 3300 * to avoid race conditions.) 3301 */ 3302 int 3303 vn_pollrecord(struct vnode *vp, struct thread *td, int events) 3304 { 3305 3306 v_addpollinfo(vp); 3307 mtx_lock(&vp->v_pollinfo->vpi_lock); 3308 if (vp->v_pollinfo->vpi_revents & events) { 3309 /* 3310 * This leaves events we are not interested 3311 * in available for the other process which 3312 * which presumably had requested them 3313 * (otherwise they would never have been 3314 * recorded). 3315 */ 3316 events &= vp->v_pollinfo->vpi_revents; 3317 vp->v_pollinfo->vpi_revents &= ~events; 3318 3319 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3320 return (events); 3321 } 3322 vp->v_pollinfo->vpi_events |= events; 3323 selrecord(td, &vp->v_pollinfo->vpi_selinfo); 3324 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3325 return (0); 3326 } 3327 3328 /* 3329 * Routine to create and manage a filesystem syncer vnode. 3330 */ 3331 #define sync_close ((int (*)(struct vop_close_args *))nullop) 3332 static int sync_fsync(struct vop_fsync_args *); 3333 static int sync_inactive(struct vop_inactive_args *); 3334 static int sync_reclaim(struct vop_reclaim_args *); 3335 3336 static struct vop_vector sync_vnodeops = { 3337 .vop_bypass = VOP_EOPNOTSUPP, 3338 .vop_close = sync_close, /* close */ 3339 .vop_fsync = sync_fsync, /* fsync */ 3340 .vop_inactive = sync_inactive, /* inactive */ 3341 .vop_reclaim = sync_reclaim, /* reclaim */ 3342 .vop_lock1 = vop_stdlock, /* lock */ 3343 .vop_unlock = vop_stdunlock, /* unlock */ 3344 .vop_islocked = vop_stdislocked, /* islocked */ 3345 }; 3346 3347 /* 3348 * Create a new filesystem syncer vnode for the specified mount point. 3349 */ 3350 int 3351 vfs_allocate_syncvnode(struct mount *mp) 3352 { 3353 struct vnode *vp; 3354 struct bufobj *bo; 3355 static long start, incr, next; 3356 int error; 3357 3358 /* Allocate a new vnode */ 3359 if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) { 3360 mp->mnt_syncer = NULL; 3361 return (error); 3362 } 3363 vp->v_type = VNON; 3364 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3365 vp->v_vflag |= VV_FORCEINSMQ; 3366 error = insmntque(vp, mp); 3367 if (error != 0) 3368 panic("vfs_allocate_syncvnode: insmntque failed"); 3369 vp->v_vflag &= ~VV_FORCEINSMQ; 3370 VOP_UNLOCK(vp, 0); 3371 /* 3372 * Place the vnode onto the syncer worklist. We attempt to 3373 * scatter them about on the list so that they will go off 3374 * at evenly distributed times even if all the filesystems 3375 * are mounted at once. 3376 */ 3377 next += incr; 3378 if (next == 0 || next > syncer_maxdelay) { 3379 start /= 2; 3380 incr /= 2; 3381 if (start == 0) { 3382 start = syncer_maxdelay / 2; 3383 incr = syncer_maxdelay; 3384 } 3385 next = start; 3386 } 3387 bo = &vp->v_bufobj; 3388 BO_LOCK(bo); 3389 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0); 3390 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ 3391 mtx_lock(&sync_mtx); 3392 sync_vnode_count++; 3393 mtx_unlock(&sync_mtx); 3394 BO_UNLOCK(bo); 3395 mp->mnt_syncer = vp; 3396 return (0); 3397 } 3398 3399 /* 3400 * Do a lazy sync of the filesystem. 3401 */ 3402 static int 3403 sync_fsync(struct vop_fsync_args *ap) 3404 { 3405 struct vnode *syncvp = ap->a_vp; 3406 struct mount *mp = syncvp->v_mount; 3407 int error; 3408 struct bufobj *bo; 3409 3410 /* 3411 * We only need to do something if this is a lazy evaluation. 3412 */ 3413 if (ap->a_waitfor != MNT_LAZY) 3414 return (0); 3415 3416 /* 3417 * Move ourselves to the back of the sync list. 3418 */ 3419 bo = &syncvp->v_bufobj; 3420 BO_LOCK(bo); 3421 vn_syncer_add_to_worklist(bo, syncdelay); 3422 BO_UNLOCK(bo); 3423 3424 /* 3425 * Walk the list of vnodes pushing all that are dirty and 3426 * not already on the sync list. 3427 */ 3428 mtx_lock(&mountlist_mtx); 3429 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) { 3430 mtx_unlock(&mountlist_mtx); 3431 return (0); 3432 } 3433 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { 3434 vfs_unbusy(mp); 3435 return (0); 3436 } 3437 MNT_ILOCK(mp); 3438 mp->mnt_noasync++; 3439 mp->mnt_kern_flag &= ~MNTK_ASYNC; 3440 MNT_IUNLOCK(mp); 3441 vfs_msync(mp, MNT_NOWAIT); 3442 error = VFS_SYNC(mp, MNT_LAZY, ap->a_td); 3443 MNT_ILOCK(mp); 3444 mp->mnt_noasync--; 3445 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 3446 mp->mnt_kern_flag |= MNTK_ASYNC; 3447 MNT_IUNLOCK(mp); 3448 vn_finished_write(mp); 3449 vfs_unbusy(mp); 3450 return (error); 3451 } 3452 3453 /* 3454 * The syncer vnode is no referenced. 3455 */ 3456 static int 3457 sync_inactive(struct vop_inactive_args *ap) 3458 { 3459 3460 vgone(ap->a_vp); 3461 return (0); 3462 } 3463 3464 /* 3465 * The syncer vnode is no longer needed and is being decommissioned. 3466 * 3467 * Modifications to the worklist must be protected by sync_mtx. 3468 */ 3469 static int 3470 sync_reclaim(struct vop_reclaim_args *ap) 3471 { 3472 struct vnode *vp = ap->a_vp; 3473 struct bufobj *bo; 3474 3475 bo = &vp->v_bufobj; 3476 BO_LOCK(bo); 3477 vp->v_mount->mnt_syncer = NULL; 3478 if (bo->bo_flag & BO_ONWORKLST) { 3479 mtx_lock(&sync_mtx); 3480 LIST_REMOVE(bo, bo_synclist); 3481 syncer_worklist_len--; 3482 sync_vnode_count--; 3483 mtx_unlock(&sync_mtx); 3484 bo->bo_flag &= ~BO_ONWORKLST; 3485 } 3486 BO_UNLOCK(bo); 3487 3488 return (0); 3489 } 3490 3491 /* 3492 * Check if vnode represents a disk device 3493 */ 3494 int 3495 vn_isdisk(struct vnode *vp, int *errp) 3496 { 3497 int error; 3498 3499 error = 0; 3500 dev_lock(); 3501 if (vp->v_type != VCHR) 3502 error = ENOTBLK; 3503 else if (vp->v_rdev == NULL) 3504 error = ENXIO; 3505 else if (vp->v_rdev->si_devsw == NULL) 3506 error = ENXIO; 3507 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) 3508 error = ENOTBLK; 3509 dev_unlock(); 3510 if (errp != NULL) 3511 *errp = error; 3512 return (error == 0); 3513 } 3514 3515 /* 3516 * Common filesystem object access control check routine. Accepts a 3517 * vnode's type, "mode", uid and gid, requested access mode, credentials, 3518 * and optional call-by-reference privused argument allowing vaccess() 3519 * to indicate to the caller whether privilege was used to satisfy the 3520 * request (obsoleted). Returns 0 on success, or an errno on failure. 3521 * 3522 * The ifdef'd CAPABILITIES version is here for reference, but is not 3523 * actually used. 3524 */ 3525 int 3526 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, 3527 accmode_t accmode, struct ucred *cred, int *privused) 3528 { 3529 accmode_t dac_granted; 3530 accmode_t priv_granted; 3531 3532 /* 3533 * Look for a normal, non-privileged way to access the file/directory 3534 * as requested. If it exists, go with that. 3535 */ 3536 3537 if (privused != NULL) 3538 *privused = 0; 3539 3540 dac_granted = 0; 3541 3542 /* Check the owner. */ 3543 if (cred->cr_uid == file_uid) { 3544 dac_granted |= VADMIN; 3545 if (file_mode & S_IXUSR) 3546 dac_granted |= VEXEC; 3547 if (file_mode & S_IRUSR) 3548 dac_granted |= VREAD; 3549 if (file_mode & S_IWUSR) 3550 dac_granted |= (VWRITE | VAPPEND); 3551 3552 if ((accmode & dac_granted) == accmode) 3553 return (0); 3554 3555 goto privcheck; 3556 } 3557 3558 /* Otherwise, check the groups (first match) */ 3559 if (groupmember(file_gid, cred)) { 3560 if (file_mode & S_IXGRP) 3561 dac_granted |= VEXEC; 3562 if (file_mode & S_IRGRP) 3563 dac_granted |= VREAD; 3564 if (file_mode & S_IWGRP) 3565 dac_granted |= (VWRITE | VAPPEND); 3566 3567 if ((accmode & dac_granted) == accmode) 3568 return (0); 3569 3570 goto privcheck; 3571 } 3572 3573 /* Otherwise, check everyone else. */ 3574 if (file_mode & S_IXOTH) 3575 dac_granted |= VEXEC; 3576 if (file_mode & S_IROTH) 3577 dac_granted |= VREAD; 3578 if (file_mode & S_IWOTH) 3579 dac_granted |= (VWRITE | VAPPEND); 3580 if ((accmode & dac_granted) == accmode) 3581 return (0); 3582 3583 privcheck: 3584 /* 3585 * Build a privilege mask to determine if the set of privileges 3586 * satisfies the requirements when combined with the granted mask 3587 * from above. For each privilege, if the privilege is required, 3588 * bitwise or the request type onto the priv_granted mask. 3589 */ 3590 priv_granted = 0; 3591 3592 if (type == VDIR) { 3593 /* 3594 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC 3595 * requests, instead of PRIV_VFS_EXEC. 3596 */ 3597 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3598 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0)) 3599 priv_granted |= VEXEC; 3600 } else { 3601 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3602 !priv_check_cred(cred, PRIV_VFS_EXEC, 0)) 3603 priv_granted |= VEXEC; 3604 } 3605 3606 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) && 3607 !priv_check_cred(cred, PRIV_VFS_READ, 0)) 3608 priv_granted |= VREAD; 3609 3610 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) && 3611 !priv_check_cred(cred, PRIV_VFS_WRITE, 0)) 3612 priv_granted |= (VWRITE | VAPPEND); 3613 3614 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) && 3615 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0)) 3616 priv_granted |= VADMIN; 3617 3618 if ((accmode & (priv_granted | dac_granted)) == accmode) { 3619 /* XXX audit: privilege used */ 3620 if (privused != NULL) 3621 *privused = 1; 3622 return (0); 3623 } 3624 3625 return ((accmode & VADMIN) ? EPERM : EACCES); 3626 } 3627 3628 /* 3629 * Credential check based on process requesting service, and per-attribute 3630 * permissions. 3631 */ 3632 int 3633 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, 3634 struct thread *td, accmode_t accmode) 3635 { 3636 3637 /* 3638 * Kernel-invoked always succeeds. 3639 */ 3640 if (cred == NOCRED) 3641 return (0); 3642 3643 /* 3644 * Do not allow privileged processes in jail to directly manipulate 3645 * system attributes. 3646 */ 3647 switch (attrnamespace) { 3648 case EXTATTR_NAMESPACE_SYSTEM: 3649 /* Potentially should be: return (EPERM); */ 3650 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0)); 3651 case EXTATTR_NAMESPACE_USER: 3652 return (VOP_ACCESS(vp, accmode, cred, td)); 3653 default: 3654 return (EPERM); 3655 } 3656 } 3657 3658 #ifdef DEBUG_VFS_LOCKS 3659 /* 3660 * This only exists to supress warnings from unlocked specfs accesses. It is 3661 * no longer ok to have an unlocked VFS. 3662 */ 3663 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \ 3664 (vp)->v_type == VCHR || (vp)->v_type == VBAD) 3665 3666 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ 3667 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, ""); 3668 3669 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ 3670 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, ""); 3671 3672 int vfs_badlock_print = 1; /* Print lock violations. */ 3673 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, ""); 3674 3675 #ifdef KDB 3676 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ 3677 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, ""); 3678 #endif 3679 3680 static void 3681 vfs_badlock(const char *msg, const char *str, struct vnode *vp) 3682 { 3683 3684 #ifdef KDB 3685 if (vfs_badlock_backtrace) 3686 kdb_backtrace(); 3687 #endif 3688 if (vfs_badlock_print) 3689 printf("%s: %p %s\n", str, (void *)vp, msg); 3690 if (vfs_badlock_ddb) 3691 kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); 3692 } 3693 3694 void 3695 assert_vi_locked(struct vnode *vp, const char *str) 3696 { 3697 3698 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) 3699 vfs_badlock("interlock is not locked but should be", str, vp); 3700 } 3701 3702 void 3703 assert_vi_unlocked(struct vnode *vp, const char *str) 3704 { 3705 3706 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) 3707 vfs_badlock("interlock is locked but should not be", str, vp); 3708 } 3709 3710 void 3711 assert_vop_locked(struct vnode *vp, const char *str) 3712 { 3713 3714 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0) 3715 vfs_badlock("is not locked but should be", str, vp); 3716 } 3717 3718 void 3719 assert_vop_unlocked(struct vnode *vp, const char *str) 3720 { 3721 3722 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) 3723 vfs_badlock("is locked but should not be", str, vp); 3724 } 3725 3726 void 3727 assert_vop_elocked(struct vnode *vp, const char *str) 3728 { 3729 3730 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 3731 vfs_badlock("is not exclusive locked but should be", str, vp); 3732 } 3733 3734 #if 0 3735 void 3736 assert_vop_elocked_other(struct vnode *vp, const char *str) 3737 { 3738 3739 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER) 3740 vfs_badlock("is not exclusive locked by another thread", 3741 str, vp); 3742 } 3743 3744 void 3745 assert_vop_slocked(struct vnode *vp, const char *str) 3746 { 3747 3748 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED) 3749 vfs_badlock("is not locked shared but should be", str, vp); 3750 } 3751 #endif /* 0 */ 3752 #endif /* DEBUG_VFS_LOCKS */ 3753 3754 void 3755 vop_rename_pre(void *ap) 3756 { 3757 struct vop_rename_args *a = ap; 3758 3759 #ifdef DEBUG_VFS_LOCKS 3760 if (a->a_tvp) 3761 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); 3762 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); 3763 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); 3764 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); 3765 3766 /* Check the source (from). */ 3767 if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp) 3768 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); 3769 if (a->a_tvp != a->a_fvp) 3770 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); 3771 3772 /* Check the target. */ 3773 if (a->a_tvp) 3774 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); 3775 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); 3776 #endif 3777 if (a->a_tdvp != a->a_fdvp) 3778 vhold(a->a_fdvp); 3779 if (a->a_tvp != a->a_fvp) 3780 vhold(a->a_fvp); 3781 vhold(a->a_tdvp); 3782 if (a->a_tvp) 3783 vhold(a->a_tvp); 3784 } 3785 3786 void 3787 vop_strategy_pre(void *ap) 3788 { 3789 #ifdef DEBUG_VFS_LOCKS 3790 struct vop_strategy_args *a; 3791 struct buf *bp; 3792 3793 a = ap; 3794 bp = a->a_bp; 3795 3796 /* 3797 * Cluster ops lock their component buffers but not the IO container. 3798 */ 3799 if ((bp->b_flags & B_CLUSTER) != 0) 3800 return; 3801 3802 if (!BUF_ISLOCKED(bp)) { 3803 if (vfs_badlock_print) 3804 printf( 3805 "VOP_STRATEGY: bp is not locked but should be\n"); 3806 if (vfs_badlock_ddb) 3807 kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); 3808 } 3809 #endif 3810 } 3811 3812 void 3813 vop_lookup_pre(void *ap) 3814 { 3815 #ifdef DEBUG_VFS_LOCKS 3816 struct vop_lookup_args *a; 3817 struct vnode *dvp; 3818 3819 a = ap; 3820 dvp = a->a_dvp; 3821 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 3822 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 3823 #endif 3824 } 3825 3826 void 3827 vop_lookup_post(void *ap, int rc) 3828 { 3829 #ifdef DEBUG_VFS_LOCKS 3830 struct vop_lookup_args *a; 3831 struct vnode *dvp; 3832 struct vnode *vp; 3833 3834 a = ap; 3835 dvp = a->a_dvp; 3836 vp = *(a->a_vpp); 3837 3838 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 3839 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 3840 3841 if (!rc) 3842 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)"); 3843 #endif 3844 } 3845 3846 void 3847 vop_lock_pre(void *ap) 3848 { 3849 #ifdef DEBUG_VFS_LOCKS 3850 struct vop_lock1_args *a = ap; 3851 3852 if ((a->a_flags & LK_INTERLOCK) == 0) 3853 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 3854 else 3855 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); 3856 #endif 3857 } 3858 3859 void 3860 vop_lock_post(void *ap, int rc) 3861 { 3862 #ifdef DEBUG_VFS_LOCKS 3863 struct vop_lock1_args *a = ap; 3864 3865 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 3866 if (rc == 0) 3867 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); 3868 #endif 3869 } 3870 3871 void 3872 vop_unlock_pre(void *ap) 3873 { 3874 #ifdef DEBUG_VFS_LOCKS 3875 struct vop_unlock_args *a = ap; 3876 3877 if (a->a_flags & LK_INTERLOCK) 3878 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK"); 3879 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); 3880 #endif 3881 } 3882 3883 void 3884 vop_unlock_post(void *ap, int rc) 3885 { 3886 #ifdef DEBUG_VFS_LOCKS 3887 struct vop_unlock_args *a = ap; 3888 3889 if (a->a_flags & LK_INTERLOCK) 3890 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK"); 3891 #endif 3892 } 3893 3894 void 3895 vop_create_post(void *ap, int rc) 3896 { 3897 struct vop_create_args *a = ap; 3898 3899 if (!rc) 3900 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3901 } 3902 3903 void 3904 vop_link_post(void *ap, int rc) 3905 { 3906 struct vop_link_args *a = ap; 3907 3908 if (!rc) { 3909 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK); 3910 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE); 3911 } 3912 } 3913 3914 void 3915 vop_mkdir_post(void *ap, int rc) 3916 { 3917 struct vop_mkdir_args *a = ap; 3918 3919 if (!rc) 3920 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 3921 } 3922 3923 void 3924 vop_mknod_post(void *ap, int rc) 3925 { 3926 struct vop_mknod_args *a = ap; 3927 3928 if (!rc) 3929 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3930 } 3931 3932 void 3933 vop_remove_post(void *ap, int rc) 3934 { 3935 struct vop_remove_args *a = ap; 3936 3937 if (!rc) { 3938 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3939 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 3940 } 3941 } 3942 3943 void 3944 vop_rename_post(void *ap, int rc) 3945 { 3946 struct vop_rename_args *a = ap; 3947 3948 if (!rc) { 3949 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE); 3950 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE); 3951 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); 3952 if (a->a_tvp) 3953 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); 3954 } 3955 if (a->a_tdvp != a->a_fdvp) 3956 vdrop(a->a_fdvp); 3957 if (a->a_tvp != a->a_fvp) 3958 vdrop(a->a_fvp); 3959 vdrop(a->a_tdvp); 3960 if (a->a_tvp) 3961 vdrop(a->a_tvp); 3962 } 3963 3964 void 3965 vop_rmdir_post(void *ap, int rc) 3966 { 3967 struct vop_rmdir_args *a = ap; 3968 3969 if (!rc) { 3970 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 3971 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 3972 } 3973 } 3974 3975 void 3976 vop_setattr_post(void *ap, int rc) 3977 { 3978 struct vop_setattr_args *a = ap; 3979 3980 if (!rc) 3981 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); 3982 } 3983 3984 void 3985 vop_symlink_post(void *ap, int rc) 3986 { 3987 struct vop_symlink_args *a = ap; 3988 3989 if (!rc) 3990 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3991 } 3992 3993 static struct knlist fs_knlist; 3994 3995 static void 3996 vfs_event_init(void *arg) 3997 { 3998 knlist_init(&fs_knlist, NULL, NULL, NULL, NULL); 3999 } 4000 /* XXX - correct order? */ 4001 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); 4002 4003 void 4004 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused) 4005 { 4006 4007 KNOTE_UNLOCKED(&fs_knlist, event); 4008 } 4009 4010 static int filt_fsattach(struct knote *kn); 4011 static void filt_fsdetach(struct knote *kn); 4012 static int filt_fsevent(struct knote *kn, long hint); 4013 4014 struct filterops fs_filtops = 4015 { 0, filt_fsattach, filt_fsdetach, filt_fsevent }; 4016 4017 static int 4018 filt_fsattach(struct knote *kn) 4019 { 4020 4021 kn->kn_flags |= EV_CLEAR; 4022 knlist_add(&fs_knlist, kn, 0); 4023 return (0); 4024 } 4025 4026 static void 4027 filt_fsdetach(struct knote *kn) 4028 { 4029 4030 knlist_remove(&fs_knlist, kn, 0); 4031 } 4032 4033 static int 4034 filt_fsevent(struct knote *kn, long hint) 4035 { 4036 4037 kn->kn_fflags |= hint; 4038 return (kn->kn_fflags != 0); 4039 } 4040 4041 static int 4042 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) 4043 { 4044 struct vfsidctl vc; 4045 int error; 4046 struct mount *mp; 4047 4048 error = SYSCTL_IN(req, &vc, sizeof(vc)); 4049 if (error) 4050 return (error); 4051 if (vc.vc_vers != VFS_CTL_VERS1) 4052 return (EINVAL); 4053 mp = vfs_getvfs(&vc.vc_fsid); 4054 if (mp == NULL) 4055 return (ENOENT); 4056 /* ensure that a specific sysctl goes to the right filesystem. */ 4057 if (strcmp(vc.vc_fstypename, "*") != 0 && 4058 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { 4059 vfs_rel(mp); 4060 return (EINVAL); 4061 } 4062 VCTLTOREQ(&vc, req); 4063 error = VFS_SYSCTL(mp, vc.vc_op, req); 4064 vfs_rel(mp); 4065 return (error); 4066 } 4067 4068 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "", 4069 "Sysctl by fsid"); 4070 4071 /* 4072 * Function to initialize a va_filerev field sensibly. 4073 * XXX: Wouldn't a random number make a lot more sense ?? 4074 */ 4075 u_quad_t 4076 init_va_filerev(void) 4077 { 4078 struct bintime bt; 4079 4080 getbinuptime(&bt); 4081 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); 4082 } 4083 4084 static int filt_vfsread(struct knote *kn, long hint); 4085 static int filt_vfswrite(struct knote *kn, long hint); 4086 static int filt_vfsvnode(struct knote *kn, long hint); 4087 static void filt_vfsdetach(struct knote *kn); 4088 static struct filterops vfsread_filtops = 4089 { 1, NULL, filt_vfsdetach, filt_vfsread }; 4090 static struct filterops vfswrite_filtops = 4091 { 1, NULL, filt_vfsdetach, filt_vfswrite }; 4092 static struct filterops vfsvnode_filtops = 4093 { 1, NULL, filt_vfsdetach, filt_vfsvnode }; 4094 4095 static void 4096 vfs_knllock(void *arg) 4097 { 4098 struct vnode *vp = arg; 4099 4100 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4101 } 4102 4103 static void 4104 vfs_knlunlock(void *arg) 4105 { 4106 struct vnode *vp = arg; 4107 4108 VOP_UNLOCK(vp, 0); 4109 } 4110 4111 static int 4112 vfs_knllocked(void *arg) 4113 { 4114 struct vnode *vp = arg; 4115 4116 return (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 4117 } 4118 4119 int 4120 vfs_kqfilter(struct vop_kqfilter_args *ap) 4121 { 4122 struct vnode *vp = ap->a_vp; 4123 struct knote *kn = ap->a_kn; 4124 struct knlist *knl; 4125 4126 switch (kn->kn_filter) { 4127 case EVFILT_READ: 4128 kn->kn_fop = &vfsread_filtops; 4129 break; 4130 case EVFILT_WRITE: 4131 kn->kn_fop = &vfswrite_filtops; 4132 break; 4133 case EVFILT_VNODE: 4134 kn->kn_fop = &vfsvnode_filtops; 4135 break; 4136 default: 4137 return (EINVAL); 4138 } 4139 4140 kn->kn_hook = (caddr_t)vp; 4141 4142 v_addpollinfo(vp); 4143 if (vp->v_pollinfo == NULL) 4144 return (ENOMEM); 4145 knl = &vp->v_pollinfo->vpi_selinfo.si_note; 4146 knlist_add(knl, kn, 0); 4147 4148 return (0); 4149 } 4150 4151 /* 4152 * Detach knote from vnode 4153 */ 4154 static void 4155 filt_vfsdetach(struct knote *kn) 4156 { 4157 struct vnode *vp = (struct vnode *)kn->kn_hook; 4158 4159 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); 4160 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); 4161 } 4162 4163 /*ARGSUSED*/ 4164 static int 4165 filt_vfsread(struct knote *kn, long hint) 4166 { 4167 struct vnode *vp = (struct vnode *)kn->kn_hook; 4168 struct vattr va; 4169 4170 /* 4171 * filesystem is gone, so set the EOF flag and schedule 4172 * the knote for deletion. 4173 */ 4174 if (hint == NOTE_REVOKE) { 4175 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 4176 return (1); 4177 } 4178 4179 if (VOP_GETATTR(vp, &va, curthread->td_ucred)) 4180 return (0); 4181 4182 kn->kn_data = va.va_size - kn->kn_fp->f_offset; 4183 return (kn->kn_data != 0); 4184 } 4185 4186 /*ARGSUSED*/ 4187 static int 4188 filt_vfswrite(struct knote *kn, long hint) 4189 { 4190 /* 4191 * filesystem is gone, so set the EOF flag and schedule 4192 * the knote for deletion. 4193 */ 4194 if (hint == NOTE_REVOKE) 4195 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 4196 4197 kn->kn_data = 0; 4198 return (1); 4199 } 4200 4201 static int 4202 filt_vfsvnode(struct knote *kn, long hint) 4203 { 4204 if (kn->kn_sfflags & hint) 4205 kn->kn_fflags |= hint; 4206 if (hint == NOTE_REVOKE) { 4207 kn->kn_flags |= EV_EOF; 4208 return (1); 4209 } 4210 return (kn->kn_fflags != 0); 4211 } 4212 4213 int 4214 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) 4215 { 4216 int error; 4217 4218 if (dp->d_reclen > ap->a_uio->uio_resid) 4219 return (ENAMETOOLONG); 4220 error = uiomove(dp, dp->d_reclen, ap->a_uio); 4221 if (error) { 4222 if (ap->a_ncookies != NULL) { 4223 if (ap->a_cookies != NULL) 4224 free(ap->a_cookies, M_TEMP); 4225 ap->a_cookies = NULL; 4226 *ap->a_ncookies = 0; 4227 } 4228 return (error); 4229 } 4230 if (ap->a_ncookies == NULL) 4231 return (0); 4232 4233 KASSERT(ap->a_cookies, 4234 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); 4235 4236 *ap->a_cookies = realloc(*ap->a_cookies, 4237 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); 4238 (*ap->a_cookies)[*ap->a_ncookies] = off; 4239 return (0); 4240 } 4241 4242 /* 4243 * Mark for update the access time of the file if the filesystem 4244 * supports VOP_MARKATIME. This functionality is used by execve and 4245 * mmap, so we want to avoid the I/O implied by directly setting 4246 * va_atime for the sake of efficiency. 4247 */ 4248 void 4249 vfs_mark_atime(struct vnode *vp, struct ucred *cred) 4250 { 4251 4252 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) 4253 (void)VOP_MARKATIME(vp); 4254 } 4255