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 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/bio.h> 49 #include <sys/buf.h> 50 #include <sys/condvar.h> 51 #include <sys/conf.h> 52 #include <sys/dirent.h> 53 #include <sys/event.h> 54 #include <sys/eventhandler.h> 55 #include <sys/extattr.h> 56 #include <sys/file.h> 57 #include <sys/fcntl.h> 58 #include <sys/jail.h> 59 #include <sys/kdb.h> 60 #include <sys/kernel.h> 61 #include <sys/kthread.h> 62 #include <sys/lockf.h> 63 #include <sys/malloc.h> 64 #include <sys/mount.h> 65 #include <sys/namei.h> 66 #include <sys/priv.h> 67 #include <sys/reboot.h> 68 #include <sys/sleepqueue.h> 69 #include <sys/stat.h> 70 #include <sys/sysctl.h> 71 #include <sys/syslog.h> 72 #include <sys/vmmeter.h> 73 #include <sys/vnode.h> 74 75 #include <machine/stdarg.h> 76 77 #include <security/mac/mac_framework.h> 78 79 #include <vm/vm.h> 80 #include <vm/vm_object.h> 81 #include <vm/vm_extern.h> 82 #include <vm/pmap.h> 83 #include <vm/vm_map.h> 84 #include <vm/vm_page.h> 85 #include <vm/vm_kern.h> 86 #include <vm/uma.h> 87 88 #ifdef DDB 89 #include <ddb/ddb.h> 90 #endif 91 92 #define WI_MPSAFEQ 0 93 #define WI_GIANTQ 1 94 95 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure"); 96 97 static void delmntque(struct vnode *vp); 98 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, 99 int slpflag, int slptimeo); 100 static void syncer_shutdown(void *arg, int howto); 101 static int vtryrecycle(struct vnode *vp); 102 static void vbusy(struct vnode *vp); 103 static void vinactive(struct vnode *, struct thread *); 104 static void v_incr_usecount(struct vnode *); 105 static void v_decr_usecount(struct vnode *); 106 static void v_decr_useonly(struct vnode *); 107 static void v_upgrade_usecount(struct vnode *); 108 static void vfree(struct vnode *); 109 static void vnlru_free(int); 110 static void vgonel(struct vnode *); 111 static void vfs_knllock(void *arg); 112 static void vfs_knlunlock(void *arg); 113 static void vfs_knl_assert_locked(void *arg); 114 static void vfs_knl_assert_unlocked(void *arg); 115 static void destroy_vpollinfo(struct vpollinfo *vi); 116 117 /* 118 * Number of vnodes in existence. Increased whenever getnewvnode() 119 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed 120 * vnode. 121 */ 122 static unsigned long numvnodes; 123 124 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 125 126 /* 127 * Conversion tables for conversion from vnode types to inode formats 128 * and back. 129 */ 130 enum vtype iftovt_tab[16] = { 131 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 132 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 133 }; 134 int vttoif_tab[10] = { 135 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 136 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT 137 }; 138 139 /* 140 * List of vnodes that are ready for recycling. 141 */ 142 static TAILQ_HEAD(freelst, vnode) vnode_free_list; 143 144 /* 145 * Free vnode target. Free vnodes may simply be files which have been stat'd 146 * but not read. This is somewhat common, and a small cache of such files 147 * should be kept to avoid recreation costs. 148 */ 149 static u_long wantfreevnodes; 150 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 151 /* Number of vnodes in the free list. */ 152 static u_long freevnodes; 153 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, ""); 154 155 static int vlru_allow_cache_src; 156 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW, 157 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode"); 158 159 /* 160 * Various variables used for debugging the new implementation of 161 * reassignbuf(). 162 * XXX these are probably of (very) limited utility now. 163 */ 164 static int reassignbufcalls; 165 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, ""); 166 167 /* 168 * Cache for the mount type id assigned to NFS. This is used for 169 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c. 170 */ 171 int nfs_mount_type = -1; 172 173 /* To keep more than one thread at a time from running vfs_getnewfsid */ 174 static struct mtx mntid_mtx; 175 176 /* 177 * Lock for any access to the following: 178 * vnode_free_list 179 * numvnodes 180 * freevnodes 181 */ 182 static struct mtx vnode_free_list_mtx; 183 184 /* Publicly exported FS */ 185 struct nfs_public nfs_pub; 186 187 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ 188 static uma_zone_t vnode_zone; 189 static uma_zone_t vnodepoll_zone; 190 191 /* Set to 1 to print out reclaim of active vnodes */ 192 int prtactive; 193 194 /* 195 * The workitem queue. 196 * 197 * It is useful to delay writes of file data and filesystem metadata 198 * for tens of seconds so that quickly created and deleted files need 199 * not waste disk bandwidth being created and removed. To realize this, 200 * we append vnodes to a "workitem" queue. When running with a soft 201 * updates implementation, most pending metadata dependencies should 202 * not wait for more than a few seconds. Thus, mounted on block devices 203 * are delayed only about a half the time that file data is delayed. 204 * Similarly, directory updates are more critical, so are only delayed 205 * about a third the time that file data is delayed. Thus, there are 206 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 207 * one each second (driven off the filesystem syncer process). The 208 * syncer_delayno variable indicates the next queue that is to be processed. 209 * Items that need to be processed soon are placed in this queue: 210 * 211 * syncer_workitem_pending[syncer_delayno] 212 * 213 * A delay of fifteen seconds is done by placing the request fifteen 214 * entries later in the queue: 215 * 216 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 217 * 218 */ 219 static int syncer_delayno; 220 static long syncer_mask; 221 LIST_HEAD(synclist, bufobj); 222 static struct synclist *syncer_workitem_pending[2]; 223 /* 224 * The sync_mtx protects: 225 * bo->bo_synclist 226 * sync_vnode_count 227 * syncer_delayno 228 * syncer_state 229 * syncer_workitem_pending 230 * syncer_worklist_len 231 * rushjob 232 */ 233 static struct mtx sync_mtx; 234 static struct cv sync_wakeup; 235 236 #define SYNCER_MAXDELAY 32 237 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 238 static int syncdelay = 30; /* max time to delay syncing data */ 239 static int filedelay = 30; /* time to delay syncing files */ 240 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, ""); 241 static int dirdelay = 29; /* time to delay syncing directories */ 242 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, ""); 243 static int metadelay = 28; /* time to delay syncing metadata */ 244 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, ""); 245 static int rushjob; /* number of slots to run ASAP */ 246 static int stat_rush_requests; /* number of times I/O speeded up */ 247 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, ""); 248 249 /* 250 * When shutting down the syncer, run it at four times normal speed. 251 */ 252 #define SYNCER_SHUTDOWN_SPEEDUP 4 253 static int sync_vnode_count; 254 static int syncer_worklist_len; 255 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } 256 syncer_state; 257 258 /* 259 * Number of vnodes we want to exist at any one time. This is mostly used 260 * to size hash tables in vnode-related code. It is normally not used in 261 * getnewvnode(), as wantfreevnodes is normally nonzero.) 262 * 263 * XXX desiredvnodes is historical cruft and should not exist. 264 */ 265 int desiredvnodes; 266 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 267 &desiredvnodes, 0, "Maximum number of vnodes"); 268 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, 269 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)"); 270 static int vnlru_nowhere; 271 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, 272 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); 273 274 /* 275 * Macros to control when a vnode is freed and recycled. All require 276 * the vnode interlock. 277 */ 278 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt) 279 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt) 280 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt) 281 282 283 /* 284 * Initialize the vnode management data structures. 285 * 286 * Reevaluate the following cap on the number of vnodes after the physical 287 * memory size exceeds 512GB. In the limit, as the physical memory size 288 * grows, the ratio of physical pages to vnodes approaches sixteen to one. 289 */ 290 #ifndef MAXVNODES_MAX 291 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16)) 292 #endif 293 static void 294 vntblinit(void *dummy __unused) 295 { 296 int physvnodes, virtvnodes; 297 298 /* 299 * Desiredvnodes is a function of the physical memory size and the 300 * kernel's heap size. Generally speaking, it scales with the 301 * physical memory size. The ratio of desiredvnodes to physical pages 302 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the 303 * marginal ratio of desiredvnodes to physical pages is one to 304 * sixteen. However, desiredvnodes is limited by the kernel's heap 305 * size. The memory required by desiredvnodes vnodes and vm objects 306 * may not exceed one seventh of the kernel's heap size. 307 */ 308 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4, 309 cnt.v_page_count) / 16; 310 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) + 311 sizeof(struct vnode))); 312 desiredvnodes = min(physvnodes, virtvnodes); 313 if (desiredvnodes > MAXVNODES_MAX) { 314 if (bootverbose) 315 printf("Reducing kern.maxvnodes %d -> %d\n", 316 desiredvnodes, MAXVNODES_MAX); 317 desiredvnodes = MAXVNODES_MAX; 318 } 319 wantfreevnodes = desiredvnodes / 4; 320 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); 321 TAILQ_INIT(&vnode_free_list); 322 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF); 323 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, 324 NULL, NULL, UMA_ALIGN_PTR, 0); 325 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), 326 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 327 /* 328 * Initialize the filesystem syncer. 329 */ 330 syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE, 331 &syncer_mask); 332 syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE, 333 &syncer_mask); 334 syncer_maxdelay = syncer_mask + 1; 335 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); 336 cv_init(&sync_wakeup, "syncer"); 337 } 338 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL); 339 340 341 /* 342 * Mark a mount point as busy. Used to synchronize access and to delay 343 * unmounting. Eventually, mountlist_mtx is not released on failure. 344 */ 345 int 346 vfs_busy(struct mount *mp, int flags) 347 { 348 349 MPASS((flags & ~MBF_MASK) == 0); 350 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags); 351 352 MNT_ILOCK(mp); 353 MNT_REF(mp); 354 /* 355 * If mount point is currenly being unmounted, sleep until the 356 * mount point fate is decided. If thread doing the unmounting fails, 357 * it will clear MNTK_UNMOUNT flag before waking us up, indicating 358 * that this mount point has survived the unmount attempt and vfs_busy 359 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE 360 * flag in addition to MNTK_UNMOUNT, indicating that mount point is 361 * about to be really destroyed. vfs_busy needs to release its 362 * reference on the mount point in this case and return with ENOENT, 363 * telling the caller that mount mount it tried to busy is no longer 364 * valid. 365 */ 366 while (mp->mnt_kern_flag & MNTK_UNMOUNT) { 367 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) { 368 MNT_REL(mp); 369 MNT_IUNLOCK(mp); 370 CTR1(KTR_VFS, "%s: failed busying before sleeping", 371 __func__); 372 return (ENOENT); 373 } 374 if (flags & MBF_MNTLSTLOCK) 375 mtx_unlock(&mountlist_mtx); 376 mp->mnt_kern_flag |= MNTK_MWAIT; 377 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0); 378 if (flags & MBF_MNTLSTLOCK) 379 mtx_lock(&mountlist_mtx); 380 } 381 if (flags & MBF_MNTLSTLOCK) 382 mtx_unlock(&mountlist_mtx); 383 mp->mnt_lockref++; 384 MNT_IUNLOCK(mp); 385 return (0); 386 } 387 388 /* 389 * Free a busy filesystem. 390 */ 391 void 392 vfs_unbusy(struct mount *mp) 393 { 394 395 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 396 MNT_ILOCK(mp); 397 MNT_REL(mp); 398 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref")); 399 mp->mnt_lockref--; 400 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) { 401 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT); 402 CTR1(KTR_VFS, "%s: waking up waiters", __func__); 403 mp->mnt_kern_flag &= ~MNTK_DRAINING; 404 wakeup(&mp->mnt_lockref); 405 } 406 MNT_IUNLOCK(mp); 407 } 408 409 /* 410 * Lookup a mount point by filesystem identifier. 411 */ 412 struct mount * 413 vfs_getvfs(fsid_t *fsid) 414 { 415 struct mount *mp; 416 417 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); 418 mtx_lock(&mountlist_mtx); 419 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 420 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 421 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 422 vfs_ref(mp); 423 mtx_unlock(&mountlist_mtx); 424 return (mp); 425 } 426 } 427 mtx_unlock(&mountlist_mtx); 428 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); 429 return ((struct mount *) 0); 430 } 431 432 /* 433 * Lookup a mount point by filesystem identifier, busying it before 434 * returning. 435 */ 436 struct mount * 437 vfs_busyfs(fsid_t *fsid) 438 { 439 struct mount *mp; 440 int error; 441 442 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); 443 mtx_lock(&mountlist_mtx); 444 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 445 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 446 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 447 error = vfs_busy(mp, MBF_MNTLSTLOCK); 448 if (error) { 449 mtx_unlock(&mountlist_mtx); 450 return (NULL); 451 } 452 return (mp); 453 } 454 } 455 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); 456 mtx_unlock(&mountlist_mtx); 457 return ((struct mount *) 0); 458 } 459 460 /* 461 * Check if a user can access privileged mount options. 462 */ 463 int 464 vfs_suser(struct mount *mp, struct thread *td) 465 { 466 int error; 467 468 /* 469 * If the thread is jailed, but this is not a jail-friendly file 470 * system, deny immediately. 471 */ 472 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred)) 473 return (EPERM); 474 475 /* 476 * If the file system was mounted outside the jail of the calling 477 * thread, deny immediately. 478 */ 479 if (prison_check(td->td_ucred, mp->mnt_cred) != 0) 480 return (EPERM); 481 482 /* 483 * If file system supports delegated administration, we don't check 484 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified 485 * by the file system itself. 486 * If this is not the user that did original mount, we check for 487 * the PRIV_VFS_MOUNT_OWNER privilege. 488 */ 489 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) && 490 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { 491 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) 492 return (error); 493 } 494 return (0); 495 } 496 497 /* 498 * Get a new unique fsid. Try to make its val[0] unique, since this value 499 * will be used to create fake device numbers for stat(). Also try (but 500 * not so hard) make its val[0] unique mod 2^16, since some emulators only 501 * support 16-bit device numbers. We end up with unique val[0]'s for the 502 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 503 * 504 * Keep in mind that several mounts may be running in parallel. Starting 505 * the search one past where the previous search terminated is both a 506 * micro-optimization and a defense against returning the same fsid to 507 * different mounts. 508 */ 509 void 510 vfs_getnewfsid(struct mount *mp) 511 { 512 static uint16_t mntid_base; 513 struct mount *nmp; 514 fsid_t tfsid; 515 int mtype; 516 517 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 518 mtx_lock(&mntid_mtx); 519 mtype = mp->mnt_vfc->vfc_typenum; 520 tfsid.val[1] = mtype; 521 mtype = (mtype & 0xFF) << 24; 522 for (;;) { 523 tfsid.val[0] = makedev(255, 524 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 525 mntid_base++; 526 if ((nmp = vfs_getvfs(&tfsid)) == NULL) 527 break; 528 vfs_rel(nmp); 529 } 530 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 531 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 532 mtx_unlock(&mntid_mtx); 533 } 534 535 /* 536 * Knob to control the precision of file timestamps: 537 * 538 * 0 = seconds only; nanoseconds zeroed. 539 * 1 = seconds and nanoseconds, accurate within 1/HZ. 540 * 2 = seconds and nanoseconds, truncated to microseconds. 541 * >=3 = seconds and nanoseconds, maximum precision. 542 */ 543 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 544 545 static int timestamp_precision = TSP_SEC; 546 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 547 ×tamp_precision, 0, ""); 548 549 /* 550 * Get a current timestamp. 551 */ 552 void 553 vfs_timestamp(struct timespec *tsp) 554 { 555 struct timeval tv; 556 557 switch (timestamp_precision) { 558 case TSP_SEC: 559 tsp->tv_sec = time_second; 560 tsp->tv_nsec = 0; 561 break; 562 case TSP_HZ: 563 getnanotime(tsp); 564 break; 565 case TSP_USEC: 566 microtime(&tv); 567 TIMEVAL_TO_TIMESPEC(&tv, tsp); 568 break; 569 case TSP_NSEC: 570 default: 571 nanotime(tsp); 572 break; 573 } 574 } 575 576 /* 577 * Set vnode attributes to VNOVAL 578 */ 579 void 580 vattr_null(struct vattr *vap) 581 { 582 583 vap->va_type = VNON; 584 vap->va_size = VNOVAL; 585 vap->va_bytes = VNOVAL; 586 vap->va_mode = VNOVAL; 587 vap->va_nlink = VNOVAL; 588 vap->va_uid = VNOVAL; 589 vap->va_gid = VNOVAL; 590 vap->va_fsid = VNOVAL; 591 vap->va_fileid = VNOVAL; 592 vap->va_blocksize = VNOVAL; 593 vap->va_rdev = VNOVAL; 594 vap->va_atime.tv_sec = VNOVAL; 595 vap->va_atime.tv_nsec = VNOVAL; 596 vap->va_mtime.tv_sec = VNOVAL; 597 vap->va_mtime.tv_nsec = VNOVAL; 598 vap->va_ctime.tv_sec = VNOVAL; 599 vap->va_ctime.tv_nsec = VNOVAL; 600 vap->va_birthtime.tv_sec = VNOVAL; 601 vap->va_birthtime.tv_nsec = VNOVAL; 602 vap->va_flags = VNOVAL; 603 vap->va_gen = VNOVAL; 604 vap->va_vaflags = 0; 605 } 606 607 /* 608 * This routine is called when we have too many vnodes. It attempts 609 * to free <count> vnodes and will potentially free vnodes that still 610 * have VM backing store (VM backing store is typically the cause 611 * of a vnode blowout so we want to do this). Therefore, this operation 612 * is not considered cheap. 613 * 614 * A number of conditions may prevent a vnode from being reclaimed. 615 * the buffer cache may have references on the vnode, a directory 616 * vnode may still have references due to the namei cache representing 617 * underlying files, or the vnode may be in active use. It is not 618 * desireable to reuse such vnodes. These conditions may cause the 619 * number of vnodes to reach some minimum value regardless of what 620 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 621 */ 622 static int 623 vlrureclaim(struct mount *mp) 624 { 625 struct vnode *vp; 626 int done; 627 int trigger; 628 int usevnodes; 629 int count; 630 631 /* 632 * Calculate the trigger point, don't allow user 633 * screwups to blow us up. This prevents us from 634 * recycling vnodes with lots of resident pages. We 635 * aren't trying to free memory, we are trying to 636 * free vnodes. 637 */ 638 usevnodes = desiredvnodes; 639 if (usevnodes <= 0) 640 usevnodes = 1; 641 trigger = cnt.v_page_count * 2 / usevnodes; 642 done = 0; 643 vn_start_write(NULL, &mp, V_WAIT); 644 MNT_ILOCK(mp); 645 count = mp->mnt_nvnodelistsize / 10 + 1; 646 while (count != 0) { 647 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 648 while (vp != NULL && vp->v_type == VMARKER) 649 vp = TAILQ_NEXT(vp, v_nmntvnodes); 650 if (vp == NULL) 651 break; 652 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 653 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 654 --count; 655 if (!VI_TRYLOCK(vp)) 656 goto next_iter; 657 /* 658 * If it's been deconstructed already, it's still 659 * referenced, or it exceeds the trigger, skip it. 660 */ 661 if (vp->v_usecount || 662 (!vlru_allow_cache_src && 663 !LIST_EMPTY(&(vp)->v_cache_src)) || 664 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL && 665 vp->v_object->resident_page_count > trigger)) { 666 VI_UNLOCK(vp); 667 goto next_iter; 668 } 669 MNT_IUNLOCK(mp); 670 vholdl(vp); 671 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) { 672 vdrop(vp); 673 goto next_iter_mntunlocked; 674 } 675 VI_LOCK(vp); 676 /* 677 * v_usecount may have been bumped after VOP_LOCK() dropped 678 * the vnode interlock and before it was locked again. 679 * 680 * It is not necessary to recheck VI_DOOMED because it can 681 * only be set by another thread that holds both the vnode 682 * lock and vnode interlock. If another thread has the 683 * vnode lock before we get to VOP_LOCK() and obtains the 684 * vnode interlock after VOP_LOCK() drops the vnode 685 * interlock, the other thread will be unable to drop the 686 * vnode lock before our VOP_LOCK() call fails. 687 */ 688 if (vp->v_usecount || 689 (!vlru_allow_cache_src && 690 !LIST_EMPTY(&(vp)->v_cache_src)) || 691 (vp->v_object != NULL && 692 vp->v_object->resident_page_count > trigger)) { 693 VOP_UNLOCK(vp, LK_INTERLOCK); 694 goto next_iter_mntunlocked; 695 } 696 KASSERT((vp->v_iflag & VI_DOOMED) == 0, 697 ("VI_DOOMED unexpectedly detected in vlrureclaim()")); 698 vgonel(vp); 699 VOP_UNLOCK(vp, 0); 700 vdropl(vp); 701 done++; 702 next_iter_mntunlocked: 703 if ((count % 256) != 0) 704 goto relock_mnt; 705 goto yield; 706 next_iter: 707 if ((count % 256) != 0) 708 continue; 709 MNT_IUNLOCK(mp); 710 yield: 711 uio_yield(); 712 relock_mnt: 713 MNT_ILOCK(mp); 714 } 715 MNT_IUNLOCK(mp); 716 vn_finished_write(mp); 717 return done; 718 } 719 720 /* 721 * Attempt to keep the free list at wantfreevnodes length. 722 */ 723 static void 724 vnlru_free(int count) 725 { 726 struct vnode *vp; 727 int vfslocked; 728 729 mtx_assert(&vnode_free_list_mtx, MA_OWNED); 730 for (; count > 0; count--) { 731 vp = TAILQ_FIRST(&vnode_free_list); 732 /* 733 * The list can be modified while the free_list_mtx 734 * has been dropped and vp could be NULL here. 735 */ 736 if (!vp) 737 break; 738 VNASSERT(vp->v_op != NULL, vp, 739 ("vnlru_free: vnode already reclaimed.")); 740 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 741 /* 742 * Don't recycle if we can't get the interlock. 743 */ 744 if (!VI_TRYLOCK(vp)) { 745 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 746 continue; 747 } 748 VNASSERT(VCANRECYCLE(vp), vp, 749 ("vp inconsistent on freelist")); 750 freevnodes--; 751 vp->v_iflag &= ~VI_FREE; 752 vholdl(vp); 753 mtx_unlock(&vnode_free_list_mtx); 754 VI_UNLOCK(vp); 755 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 756 vtryrecycle(vp); 757 VFS_UNLOCK_GIANT(vfslocked); 758 /* 759 * If the recycled succeeded this vdrop will actually free 760 * the vnode. If not it will simply place it back on 761 * the free list. 762 */ 763 vdrop(vp); 764 mtx_lock(&vnode_free_list_mtx); 765 } 766 } 767 /* 768 * Attempt to recycle vnodes in a context that is always safe to block. 769 * Calling vlrurecycle() from the bowels of filesystem code has some 770 * interesting deadlock problems. 771 */ 772 static struct proc *vnlruproc; 773 static int vnlruproc_sig; 774 775 static void 776 vnlru_proc(void) 777 { 778 struct mount *mp, *nmp; 779 int done, vfslocked; 780 struct proc *p = vnlruproc; 781 782 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p, 783 SHUTDOWN_PRI_FIRST); 784 785 for (;;) { 786 kproc_suspend_check(p); 787 mtx_lock(&vnode_free_list_mtx); 788 if (freevnodes > wantfreevnodes) 789 vnlru_free(freevnodes - wantfreevnodes); 790 if (numvnodes <= desiredvnodes * 9 / 10) { 791 vnlruproc_sig = 0; 792 wakeup(&vnlruproc_sig); 793 msleep(vnlruproc, &vnode_free_list_mtx, 794 PVFS|PDROP, "vlruwt", hz); 795 continue; 796 } 797 mtx_unlock(&vnode_free_list_mtx); 798 done = 0; 799 mtx_lock(&mountlist_mtx); 800 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 801 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { 802 nmp = TAILQ_NEXT(mp, mnt_list); 803 continue; 804 } 805 vfslocked = VFS_LOCK_GIANT(mp); 806 done += vlrureclaim(mp); 807 VFS_UNLOCK_GIANT(vfslocked); 808 mtx_lock(&mountlist_mtx); 809 nmp = TAILQ_NEXT(mp, mnt_list); 810 vfs_unbusy(mp); 811 } 812 mtx_unlock(&mountlist_mtx); 813 if (done == 0) { 814 #if 0 815 /* These messages are temporary debugging aids */ 816 if (vnlru_nowhere < 5) 817 printf("vnlru process getting nowhere..\n"); 818 else if (vnlru_nowhere == 5) 819 printf("vnlru process messages stopped.\n"); 820 #endif 821 vnlru_nowhere++; 822 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); 823 } else 824 uio_yield(); 825 } 826 } 827 828 static struct kproc_desc vnlru_kp = { 829 "vnlru", 830 vnlru_proc, 831 &vnlruproc 832 }; 833 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, 834 &vnlru_kp); 835 836 /* 837 * Routines having to do with the management of the vnode table. 838 */ 839 840 void 841 vdestroy(struct vnode *vp) 842 { 843 struct bufobj *bo; 844 845 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 846 mtx_lock(&vnode_free_list_mtx); 847 numvnodes--; 848 mtx_unlock(&vnode_free_list_mtx); 849 bo = &vp->v_bufobj; 850 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, 851 ("cleaned vnode still on the free list.")); 852 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); 853 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count")); 854 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); 855 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); 856 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); 857 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); 858 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL")); 859 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); 860 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL")); 861 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); 862 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); 863 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for ..")); 864 VI_UNLOCK(vp); 865 #ifdef MAC 866 mac_vnode_destroy(vp); 867 #endif 868 if (vp->v_pollinfo != NULL) 869 destroy_vpollinfo(vp->v_pollinfo); 870 #ifdef INVARIANTS 871 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */ 872 vp->v_op = NULL; 873 #endif 874 lockdestroy(vp->v_vnlock); 875 mtx_destroy(&vp->v_interlock); 876 mtx_destroy(BO_MTX(bo)); 877 uma_zfree(vnode_zone, vp); 878 } 879 880 /* 881 * Try to recycle a freed vnode. We abort if anyone picks up a reference 882 * before we actually vgone(). This function must be called with the vnode 883 * held to prevent the vnode from being returned to the free list midway 884 * through vgone(). 885 */ 886 static int 887 vtryrecycle(struct vnode *vp) 888 { 889 struct mount *vnmp; 890 891 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 892 VNASSERT(vp->v_holdcnt, vp, 893 ("vtryrecycle: Recycling vp %p without a reference.", vp)); 894 /* 895 * This vnode may found and locked via some other list, if so we 896 * can't recycle it yet. 897 */ 898 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 899 CTR2(KTR_VFS, 900 "%s: impossible to recycle, vp %p lock is already held", 901 __func__, vp); 902 return (EWOULDBLOCK); 903 } 904 /* 905 * Don't recycle if its filesystem is being suspended. 906 */ 907 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { 908 VOP_UNLOCK(vp, 0); 909 CTR2(KTR_VFS, 910 "%s: impossible to recycle, cannot start the write for %p", 911 __func__, vp); 912 return (EBUSY); 913 } 914 /* 915 * If we got this far, we need to acquire the interlock and see if 916 * anyone picked up this vnode from another list. If not, we will 917 * mark it with DOOMED via vgonel() so that anyone who does find it 918 * will skip over it. 919 */ 920 VI_LOCK(vp); 921 if (vp->v_usecount) { 922 VOP_UNLOCK(vp, LK_INTERLOCK); 923 vn_finished_write(vnmp); 924 CTR2(KTR_VFS, 925 "%s: impossible to recycle, %p is already referenced", 926 __func__, vp); 927 return (EBUSY); 928 } 929 if ((vp->v_iflag & VI_DOOMED) == 0) 930 vgonel(vp); 931 VOP_UNLOCK(vp, LK_INTERLOCK); 932 vn_finished_write(vnmp); 933 return (0); 934 } 935 936 /* 937 * Return the next vnode from the free list. 938 */ 939 int 940 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, 941 struct vnode **vpp) 942 { 943 struct vnode *vp = NULL; 944 struct bufobj *bo; 945 946 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag); 947 mtx_lock(&vnode_free_list_mtx); 948 /* 949 * Lend our context to reclaim vnodes if they've exceeded the max. 950 */ 951 if (freevnodes > wantfreevnodes) 952 vnlru_free(1); 953 /* 954 * Wait for available vnodes. 955 */ 956 if (numvnodes > desiredvnodes) { 957 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) { 958 /* 959 * File system is beeing suspended, we cannot risk a 960 * deadlock here, so allocate new vnode anyway. 961 */ 962 if (freevnodes > wantfreevnodes) 963 vnlru_free(freevnodes - wantfreevnodes); 964 goto alloc; 965 } 966 if (vnlruproc_sig == 0) { 967 vnlruproc_sig = 1; /* avoid unnecessary wakeups */ 968 wakeup(vnlruproc); 969 } 970 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS, 971 "vlruwk", hz); 972 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */ 973 if (numvnodes > desiredvnodes) { 974 mtx_unlock(&vnode_free_list_mtx); 975 return (ENFILE); 976 } 977 #endif 978 } 979 alloc: 980 numvnodes++; 981 mtx_unlock(&vnode_free_list_mtx); 982 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO); 983 /* 984 * Setup locks. 985 */ 986 vp->v_vnlock = &vp->v_lock; 987 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); 988 /* 989 * By default, don't allow shared locks unless filesystems 990 * opt-in. 991 */ 992 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE); 993 /* 994 * Initialize bufobj. 995 */ 996 bo = &vp->v_bufobj; 997 bo->__bo_vnode = vp; 998 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF); 999 bo->bo_ops = &buf_ops_bio; 1000 bo->bo_private = vp; 1001 TAILQ_INIT(&bo->bo_clean.bv_hd); 1002 TAILQ_INIT(&bo->bo_dirty.bv_hd); 1003 /* 1004 * Initialize namecache. 1005 */ 1006 LIST_INIT(&vp->v_cache_src); 1007 TAILQ_INIT(&vp->v_cache_dst); 1008 /* 1009 * Finalize various vnode identity bits. 1010 */ 1011 vp->v_type = VNON; 1012 vp->v_tag = tag; 1013 vp->v_op = vops; 1014 v_incr_usecount(vp); 1015 vp->v_data = 0; 1016 #ifdef MAC 1017 mac_vnode_init(vp); 1018 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) 1019 mac_vnode_associate_singlelabel(mp, vp); 1020 else if (mp == NULL && vops != &dead_vnodeops) 1021 printf("NULL mp in getnewvnode()\n"); 1022 #endif 1023 if (mp != NULL) { 1024 bo->bo_bsize = mp->mnt_stat.f_iosize; 1025 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) 1026 vp->v_vflag |= VV_NOKNOTE; 1027 } 1028 1029 *vpp = vp; 1030 return (0); 1031 } 1032 1033 /* 1034 * Delete from old mount point vnode list, if on one. 1035 */ 1036 static void 1037 delmntque(struct vnode *vp) 1038 { 1039 struct mount *mp; 1040 1041 mp = vp->v_mount; 1042 if (mp == NULL) 1043 return; 1044 MNT_ILOCK(mp); 1045 vp->v_mount = NULL; 1046 VNASSERT(mp->mnt_nvnodelistsize > 0, vp, 1047 ("bad mount point vnode list size")); 1048 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1049 mp->mnt_nvnodelistsize--; 1050 MNT_REL(mp); 1051 MNT_IUNLOCK(mp); 1052 } 1053 1054 static void 1055 insmntque_stddtr(struct vnode *vp, void *dtr_arg) 1056 { 1057 1058 vp->v_data = NULL; 1059 vp->v_op = &dead_vnodeops; 1060 /* XXX non mp-safe fs may still call insmntque with vnode 1061 unlocked */ 1062 if (!VOP_ISLOCKED(vp)) 1063 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1064 vgone(vp); 1065 vput(vp); 1066 } 1067 1068 /* 1069 * Insert into list of vnodes for the new mount point, if available. 1070 */ 1071 int 1072 insmntque1(struct vnode *vp, struct mount *mp, 1073 void (*dtr)(struct vnode *, void *), void *dtr_arg) 1074 { 1075 int locked; 1076 1077 KASSERT(vp->v_mount == NULL, 1078 ("insmntque: vnode already on per mount vnode list")); 1079 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); 1080 #ifdef DEBUG_VFS_LOCKS 1081 if (!VFS_NEEDSGIANT(mp)) 1082 ASSERT_VOP_ELOCKED(vp, 1083 "insmntque: mp-safe fs and non-locked vp"); 1084 #endif 1085 MNT_ILOCK(mp); 1086 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 && 1087 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 || 1088 mp->mnt_nvnodelistsize == 0)) { 1089 locked = VOP_ISLOCKED(vp); 1090 if (!locked || (locked == LK_EXCLUSIVE && 1091 (vp->v_vflag & VV_FORCEINSMQ) == 0)) { 1092 MNT_IUNLOCK(mp); 1093 if (dtr != NULL) 1094 dtr(vp, dtr_arg); 1095 return (EBUSY); 1096 } 1097 } 1098 vp->v_mount = mp; 1099 MNT_REF(mp); 1100 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1101 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, 1102 ("neg mount point vnode list size")); 1103 mp->mnt_nvnodelistsize++; 1104 MNT_IUNLOCK(mp); 1105 return (0); 1106 } 1107 1108 int 1109 insmntque(struct vnode *vp, struct mount *mp) 1110 { 1111 1112 return (insmntque1(vp, mp, insmntque_stddtr, NULL)); 1113 } 1114 1115 /* 1116 * Flush out and invalidate all buffers associated with a bufobj 1117 * Called with the underlying object locked. 1118 */ 1119 int 1120 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo) 1121 { 1122 int error; 1123 1124 BO_LOCK(bo); 1125 if (flags & V_SAVE) { 1126 error = bufobj_wwait(bo, slpflag, slptimeo); 1127 if (error) { 1128 BO_UNLOCK(bo); 1129 return (error); 1130 } 1131 if (bo->bo_dirty.bv_cnt > 0) { 1132 BO_UNLOCK(bo); 1133 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0) 1134 return (error); 1135 /* 1136 * XXX We could save a lock/unlock if this was only 1137 * enabled under INVARIANTS 1138 */ 1139 BO_LOCK(bo); 1140 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) 1141 panic("vinvalbuf: dirty bufs"); 1142 } 1143 } 1144 /* 1145 * If you alter this loop please notice that interlock is dropped and 1146 * reacquired in flushbuflist. Special care is needed to ensure that 1147 * no race conditions occur from this. 1148 */ 1149 do { 1150 error = flushbuflist(&bo->bo_clean, 1151 flags, bo, slpflag, slptimeo); 1152 if (error == 0) 1153 error = flushbuflist(&bo->bo_dirty, 1154 flags, bo, slpflag, slptimeo); 1155 if (error != 0 && error != EAGAIN) { 1156 BO_UNLOCK(bo); 1157 return (error); 1158 } 1159 } while (error != 0); 1160 1161 /* 1162 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 1163 * have write I/O in-progress but if there is a VM object then the 1164 * VM object can also have read-I/O in-progress. 1165 */ 1166 do { 1167 bufobj_wwait(bo, 0, 0); 1168 BO_UNLOCK(bo); 1169 if (bo->bo_object != NULL) { 1170 VM_OBJECT_LOCK(bo->bo_object); 1171 vm_object_pip_wait(bo->bo_object, "bovlbx"); 1172 VM_OBJECT_UNLOCK(bo->bo_object); 1173 } 1174 BO_LOCK(bo); 1175 } while (bo->bo_numoutput > 0); 1176 BO_UNLOCK(bo); 1177 1178 /* 1179 * Destroy the copy in the VM cache, too. 1180 */ 1181 if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) { 1182 VM_OBJECT_LOCK(bo->bo_object); 1183 vm_object_page_remove(bo->bo_object, 0, 0, 1184 (flags & V_SAVE) ? TRUE : FALSE); 1185 VM_OBJECT_UNLOCK(bo->bo_object); 1186 } 1187 1188 #ifdef INVARIANTS 1189 BO_LOCK(bo); 1190 if ((flags & (V_ALT | V_NORMAL)) == 0 && 1191 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) 1192 panic("vinvalbuf: flush failed"); 1193 BO_UNLOCK(bo); 1194 #endif 1195 return (0); 1196 } 1197 1198 /* 1199 * Flush out and invalidate all buffers associated with a vnode. 1200 * Called with the underlying object locked. 1201 */ 1202 int 1203 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 1204 { 1205 1206 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); 1207 ASSERT_VOP_LOCKED(vp, "vinvalbuf"); 1208 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo)); 1209 } 1210 1211 /* 1212 * Flush out buffers on the specified list. 1213 * 1214 */ 1215 static int 1216 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, 1217 int slptimeo) 1218 { 1219 struct buf *bp, *nbp; 1220 int retval, error; 1221 daddr_t lblkno; 1222 b_xflags_t xflags; 1223 1224 ASSERT_BO_LOCKED(bo); 1225 1226 retval = 0; 1227 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { 1228 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) || 1229 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) { 1230 continue; 1231 } 1232 lblkno = 0; 1233 xflags = 0; 1234 if (nbp != NULL) { 1235 lblkno = nbp->b_lblkno; 1236 xflags = nbp->b_xflags & 1237 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN); 1238 } 1239 retval = EAGAIN; 1240 error = BUF_TIMELOCK(bp, 1241 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo), 1242 "flushbuf", slpflag, slptimeo); 1243 if (error) { 1244 BO_LOCK(bo); 1245 return (error != ENOLCK ? error : EAGAIN); 1246 } 1247 KASSERT(bp->b_bufobj == bo, 1248 ("bp %p wrong b_bufobj %p should be %p", 1249 bp, bp->b_bufobj, bo)); 1250 if (bp->b_bufobj != bo) { /* XXX: necessary ? */ 1251 BUF_UNLOCK(bp); 1252 BO_LOCK(bo); 1253 return (EAGAIN); 1254 } 1255 /* 1256 * XXX Since there are no node locks for NFS, I 1257 * believe there is a slight chance that a delayed 1258 * write will occur while sleeping just above, so 1259 * check for it. 1260 */ 1261 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 1262 (flags & V_SAVE)) { 1263 BO_LOCK(bo); 1264 bremfree(bp); 1265 BO_UNLOCK(bo); 1266 bp->b_flags |= B_ASYNC; 1267 bwrite(bp); 1268 BO_LOCK(bo); 1269 return (EAGAIN); /* XXX: why not loop ? */ 1270 } 1271 BO_LOCK(bo); 1272 bremfree(bp); 1273 BO_UNLOCK(bo); 1274 bp->b_flags |= (B_INVAL | B_RELBUF); 1275 bp->b_flags &= ~B_ASYNC; 1276 brelse(bp); 1277 BO_LOCK(bo); 1278 if (nbp != NULL && 1279 (nbp->b_bufobj != bo || 1280 nbp->b_lblkno != lblkno || 1281 (nbp->b_xflags & 1282 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags)) 1283 break; /* nbp invalid */ 1284 } 1285 return (retval); 1286 } 1287 1288 /* 1289 * Truncate a file's buffer and pages to a specified length. This 1290 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 1291 * sync activity. 1292 */ 1293 int 1294 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td, 1295 off_t length, int blksize) 1296 { 1297 struct buf *bp, *nbp; 1298 int anyfreed; 1299 int trunclbn; 1300 struct bufobj *bo; 1301 1302 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__, 1303 vp, cred, blksize, (uintmax_t)length); 1304 1305 /* 1306 * Round up to the *next* lbn. 1307 */ 1308 trunclbn = (length + blksize - 1) / blksize; 1309 1310 ASSERT_VOP_LOCKED(vp, "vtruncbuf"); 1311 restart: 1312 bo = &vp->v_bufobj; 1313 BO_LOCK(bo); 1314 anyfreed = 1; 1315 for (;anyfreed;) { 1316 anyfreed = 0; 1317 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { 1318 if (bp->b_lblkno < trunclbn) 1319 continue; 1320 if (BUF_LOCK(bp, 1321 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1322 BO_MTX(bo)) == ENOLCK) 1323 goto restart; 1324 1325 BO_LOCK(bo); 1326 bremfree(bp); 1327 BO_UNLOCK(bo); 1328 bp->b_flags |= (B_INVAL | B_RELBUF); 1329 bp->b_flags &= ~B_ASYNC; 1330 brelse(bp); 1331 anyfreed = 1; 1332 1333 if (nbp != NULL && 1334 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 1335 (nbp->b_vp != vp) || 1336 (nbp->b_flags & B_DELWRI))) { 1337 goto restart; 1338 } 1339 BO_LOCK(bo); 1340 } 1341 1342 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1343 if (bp->b_lblkno < trunclbn) 1344 continue; 1345 if (BUF_LOCK(bp, 1346 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1347 BO_MTX(bo)) == ENOLCK) 1348 goto restart; 1349 BO_LOCK(bo); 1350 bremfree(bp); 1351 BO_UNLOCK(bo); 1352 bp->b_flags |= (B_INVAL | B_RELBUF); 1353 bp->b_flags &= ~B_ASYNC; 1354 brelse(bp); 1355 anyfreed = 1; 1356 if (nbp != NULL && 1357 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 1358 (nbp->b_vp != vp) || 1359 (nbp->b_flags & B_DELWRI) == 0)) { 1360 goto restart; 1361 } 1362 BO_LOCK(bo); 1363 } 1364 } 1365 1366 if (length > 0) { 1367 restartsync: 1368 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1369 if (bp->b_lblkno > 0) 1370 continue; 1371 /* 1372 * Since we hold the vnode lock this should only 1373 * fail if we're racing with the buf daemon. 1374 */ 1375 if (BUF_LOCK(bp, 1376 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1377 BO_MTX(bo)) == ENOLCK) { 1378 goto restart; 1379 } 1380 VNASSERT((bp->b_flags & B_DELWRI), vp, 1381 ("buf(%p) on dirty queue without DELWRI", bp)); 1382 1383 BO_LOCK(bo); 1384 bremfree(bp); 1385 BO_UNLOCK(bo); 1386 bawrite(bp); 1387 BO_LOCK(bo); 1388 goto restartsync; 1389 } 1390 } 1391 1392 bufobj_wwait(bo, 0, 0); 1393 BO_UNLOCK(bo); 1394 vnode_pager_setsize(vp, length); 1395 1396 return (0); 1397 } 1398 1399 /* 1400 * buf_splay() - splay tree core for the clean/dirty list of buffers in 1401 * a vnode. 1402 * 1403 * NOTE: We have to deal with the special case of a background bitmap 1404 * buffer, a situation where two buffers will have the same logical 1405 * block offset. We want (1) only the foreground buffer to be accessed 1406 * in a lookup and (2) must differentiate between the foreground and 1407 * background buffer in the splay tree algorithm because the splay 1408 * tree cannot normally handle multiple entities with the same 'index'. 1409 * We accomplish this by adding differentiating flags to the splay tree's 1410 * numerical domain. 1411 */ 1412 static 1413 struct buf * 1414 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root) 1415 { 1416 struct buf dummy; 1417 struct buf *lefttreemax, *righttreemin, *y; 1418 1419 if (root == NULL) 1420 return (NULL); 1421 lefttreemax = righttreemin = &dummy; 1422 for (;;) { 1423 if (lblkno < root->b_lblkno || 1424 (lblkno == root->b_lblkno && 1425 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1426 if ((y = root->b_left) == NULL) 1427 break; 1428 if (lblkno < y->b_lblkno) { 1429 /* Rotate right. */ 1430 root->b_left = y->b_right; 1431 y->b_right = root; 1432 root = y; 1433 if ((y = root->b_left) == NULL) 1434 break; 1435 } 1436 /* Link into the new root's right tree. */ 1437 righttreemin->b_left = root; 1438 righttreemin = root; 1439 } else if (lblkno > root->b_lblkno || 1440 (lblkno == root->b_lblkno && 1441 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) { 1442 if ((y = root->b_right) == NULL) 1443 break; 1444 if (lblkno > y->b_lblkno) { 1445 /* Rotate left. */ 1446 root->b_right = y->b_left; 1447 y->b_left = root; 1448 root = y; 1449 if ((y = root->b_right) == NULL) 1450 break; 1451 } 1452 /* Link into the new root's left tree. */ 1453 lefttreemax->b_right = root; 1454 lefttreemax = root; 1455 } else { 1456 break; 1457 } 1458 root = y; 1459 } 1460 /* Assemble the new root. */ 1461 lefttreemax->b_right = root->b_left; 1462 righttreemin->b_left = root->b_right; 1463 root->b_left = dummy.b_right; 1464 root->b_right = dummy.b_left; 1465 return (root); 1466 } 1467 1468 static void 1469 buf_vlist_remove(struct buf *bp) 1470 { 1471 struct buf *root; 1472 struct bufv *bv; 1473 1474 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); 1475 ASSERT_BO_LOCKED(bp->b_bufobj); 1476 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) != 1477 (BX_VNDIRTY|BX_VNCLEAN), 1478 ("buf_vlist_remove: Buf %p is on two lists", bp)); 1479 if (bp->b_xflags & BX_VNDIRTY) 1480 bv = &bp->b_bufobj->bo_dirty; 1481 else 1482 bv = &bp->b_bufobj->bo_clean; 1483 if (bp != bv->bv_root) { 1484 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root); 1485 KASSERT(root == bp, ("splay lookup failed in remove")); 1486 } 1487 if (bp->b_left == NULL) { 1488 root = bp->b_right; 1489 } else { 1490 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left); 1491 root->b_right = bp->b_right; 1492 } 1493 bv->bv_root = root; 1494 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); 1495 bv->bv_cnt--; 1496 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1497 } 1498 1499 /* 1500 * Add the buffer to the sorted clean or dirty block list using a 1501 * splay tree algorithm. 1502 * 1503 * NOTE: xflags is passed as a constant, optimizing this inline function! 1504 */ 1505 static void 1506 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) 1507 { 1508 struct buf *root; 1509 struct bufv *bv; 1510 1511 ASSERT_BO_LOCKED(bo); 1512 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, 1513 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); 1514 bp->b_xflags |= xflags; 1515 if (xflags & BX_VNDIRTY) 1516 bv = &bo->bo_dirty; 1517 else 1518 bv = &bo->bo_clean; 1519 1520 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root); 1521 if (root == NULL) { 1522 bp->b_left = NULL; 1523 bp->b_right = NULL; 1524 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); 1525 } else if (bp->b_lblkno < root->b_lblkno || 1526 (bp->b_lblkno == root->b_lblkno && 1527 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1528 bp->b_left = root->b_left; 1529 bp->b_right = root; 1530 root->b_left = NULL; 1531 TAILQ_INSERT_BEFORE(root, bp, b_bobufs); 1532 } else { 1533 bp->b_right = root->b_right; 1534 bp->b_left = root; 1535 root->b_right = NULL; 1536 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs); 1537 } 1538 bv->bv_cnt++; 1539 bv->bv_root = bp; 1540 } 1541 1542 /* 1543 * Lookup a buffer using the splay tree. Note that we specifically avoid 1544 * shadow buffers used in background bitmap writes. 1545 * 1546 * This code isn't quite efficient as it could be because we are maintaining 1547 * two sorted lists and do not know which list the block resides in. 1548 * 1549 * During a "make buildworld" the desired buffer is found at one of 1550 * the roots more than 60% of the time. Thus, checking both roots 1551 * before performing either splay eliminates unnecessary splays on the 1552 * first tree splayed. 1553 */ 1554 struct buf * 1555 gbincore(struct bufobj *bo, daddr_t lblkno) 1556 { 1557 struct buf *bp; 1558 1559 ASSERT_BO_LOCKED(bo); 1560 if ((bp = bo->bo_clean.bv_root) != NULL && 1561 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1562 return (bp); 1563 if ((bp = bo->bo_dirty.bv_root) != NULL && 1564 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1565 return (bp); 1566 if ((bp = bo->bo_clean.bv_root) != NULL) { 1567 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp); 1568 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1569 return (bp); 1570 } 1571 if ((bp = bo->bo_dirty.bv_root) != NULL) { 1572 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp); 1573 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1574 return (bp); 1575 } 1576 return (NULL); 1577 } 1578 1579 /* 1580 * Associate a buffer with a vnode. 1581 */ 1582 void 1583 bgetvp(struct vnode *vp, struct buf *bp) 1584 { 1585 struct bufobj *bo; 1586 1587 bo = &vp->v_bufobj; 1588 ASSERT_BO_LOCKED(bo); 1589 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); 1590 1591 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); 1592 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, 1593 ("bgetvp: bp already attached! %p", bp)); 1594 1595 vhold(vp); 1596 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT) 1597 bp->b_flags |= B_NEEDSGIANT; 1598 bp->b_vp = vp; 1599 bp->b_bufobj = bo; 1600 /* 1601 * Insert onto list for new vnode. 1602 */ 1603 buf_vlist_add(bp, bo, BX_VNCLEAN); 1604 } 1605 1606 /* 1607 * Disassociate a buffer from a vnode. 1608 */ 1609 void 1610 brelvp(struct buf *bp) 1611 { 1612 struct bufobj *bo; 1613 struct vnode *vp; 1614 1615 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1616 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 1617 1618 /* 1619 * Delete from old vnode list, if on one. 1620 */ 1621 vp = bp->b_vp; /* XXX */ 1622 bo = bp->b_bufobj; 1623 BO_LOCK(bo); 1624 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1625 buf_vlist_remove(bp); 1626 else 1627 panic("brelvp: Buffer %p not on queue.", bp); 1628 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 1629 bo->bo_flag &= ~BO_ONWORKLST; 1630 mtx_lock(&sync_mtx); 1631 LIST_REMOVE(bo, bo_synclist); 1632 syncer_worklist_len--; 1633 mtx_unlock(&sync_mtx); 1634 } 1635 bp->b_flags &= ~B_NEEDSGIANT; 1636 bp->b_vp = NULL; 1637 bp->b_bufobj = NULL; 1638 BO_UNLOCK(bo); 1639 vdrop(vp); 1640 } 1641 1642 /* 1643 * Add an item to the syncer work queue. 1644 */ 1645 static void 1646 vn_syncer_add_to_worklist(struct bufobj *bo, int delay) 1647 { 1648 int queue, slot; 1649 1650 ASSERT_BO_LOCKED(bo); 1651 1652 mtx_lock(&sync_mtx); 1653 if (bo->bo_flag & BO_ONWORKLST) 1654 LIST_REMOVE(bo, bo_synclist); 1655 else { 1656 bo->bo_flag |= BO_ONWORKLST; 1657 syncer_worklist_len++; 1658 } 1659 1660 if (delay > syncer_maxdelay - 2) 1661 delay = syncer_maxdelay - 2; 1662 slot = (syncer_delayno + delay) & syncer_mask; 1663 1664 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ : 1665 WI_MPSAFEQ; 1666 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo, 1667 bo_synclist); 1668 mtx_unlock(&sync_mtx); 1669 } 1670 1671 static int 1672 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) 1673 { 1674 int error, len; 1675 1676 mtx_lock(&sync_mtx); 1677 len = syncer_worklist_len - sync_vnode_count; 1678 mtx_unlock(&sync_mtx); 1679 error = SYSCTL_OUT(req, &len, sizeof(len)); 1680 return (error); 1681 } 1682 1683 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0, 1684 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); 1685 1686 static struct proc *updateproc; 1687 static void sched_sync(void); 1688 static struct kproc_desc up_kp = { 1689 "syncer", 1690 sched_sync, 1691 &updateproc 1692 }; 1693 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp); 1694 1695 static int 1696 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td) 1697 { 1698 struct vnode *vp; 1699 struct mount *mp; 1700 1701 *bo = LIST_FIRST(slp); 1702 if (*bo == NULL) 1703 return (0); 1704 vp = (*bo)->__bo_vnode; /* XXX */ 1705 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0) 1706 return (1); 1707 /* 1708 * We use vhold in case the vnode does not 1709 * successfully sync. vhold prevents the vnode from 1710 * going away when we unlock the sync_mtx so that 1711 * we can acquire the vnode interlock. 1712 */ 1713 vholdl(vp); 1714 mtx_unlock(&sync_mtx); 1715 VI_UNLOCK(vp); 1716 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { 1717 vdrop(vp); 1718 mtx_lock(&sync_mtx); 1719 return (*bo == LIST_FIRST(slp)); 1720 } 1721 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1722 (void) VOP_FSYNC(vp, MNT_LAZY, td); 1723 VOP_UNLOCK(vp, 0); 1724 vn_finished_write(mp); 1725 BO_LOCK(*bo); 1726 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) { 1727 /* 1728 * Put us back on the worklist. The worklist 1729 * routine will remove us from our current 1730 * position and then add us back in at a later 1731 * position. 1732 */ 1733 vn_syncer_add_to_worklist(*bo, syncdelay); 1734 } 1735 BO_UNLOCK(*bo); 1736 vdrop(vp); 1737 mtx_lock(&sync_mtx); 1738 return (0); 1739 } 1740 1741 /* 1742 * System filesystem synchronizer daemon. 1743 */ 1744 static void 1745 sched_sync(void) 1746 { 1747 struct synclist *gnext, *next; 1748 struct synclist *gslp, *slp; 1749 struct bufobj *bo; 1750 long starttime; 1751 struct thread *td = curthread; 1752 int last_work_seen; 1753 int net_worklist_len; 1754 int syncer_final_iter; 1755 int first_printf; 1756 int error; 1757 1758 last_work_seen = 0; 1759 syncer_final_iter = 0; 1760 first_printf = 1; 1761 syncer_state = SYNCER_RUNNING; 1762 starttime = time_uptime; 1763 td->td_pflags |= TDP_NORUNNINGBUF; 1764 1765 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, 1766 SHUTDOWN_PRI_LAST); 1767 1768 mtx_lock(&sync_mtx); 1769 for (;;) { 1770 if (syncer_state == SYNCER_FINAL_DELAY && 1771 syncer_final_iter == 0) { 1772 mtx_unlock(&sync_mtx); 1773 kproc_suspend_check(td->td_proc); 1774 mtx_lock(&sync_mtx); 1775 } 1776 net_worklist_len = syncer_worklist_len - sync_vnode_count; 1777 if (syncer_state != SYNCER_RUNNING && 1778 starttime != time_uptime) { 1779 if (first_printf) { 1780 printf("\nSyncing disks, vnodes remaining..."); 1781 first_printf = 0; 1782 } 1783 printf("%d ", net_worklist_len); 1784 } 1785 starttime = time_uptime; 1786 1787 /* 1788 * Push files whose dirty time has expired. Be careful 1789 * of interrupt race on slp queue. 1790 * 1791 * Skip over empty worklist slots when shutting down. 1792 */ 1793 do { 1794 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno]; 1795 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno]; 1796 syncer_delayno += 1; 1797 if (syncer_delayno == syncer_maxdelay) 1798 syncer_delayno = 0; 1799 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno]; 1800 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno]; 1801 /* 1802 * If the worklist has wrapped since the 1803 * it was emptied of all but syncer vnodes, 1804 * switch to the FINAL_DELAY state and run 1805 * for one more second. 1806 */ 1807 if (syncer_state == SYNCER_SHUTTING_DOWN && 1808 net_worklist_len == 0 && 1809 last_work_seen == syncer_delayno) { 1810 syncer_state = SYNCER_FINAL_DELAY; 1811 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; 1812 } 1813 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && 1814 LIST_EMPTY(gslp) && syncer_worklist_len > 0); 1815 1816 /* 1817 * Keep track of the last time there was anything 1818 * on the worklist other than syncer vnodes. 1819 * Return to the SHUTTING_DOWN state if any 1820 * new work appears. 1821 */ 1822 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) 1823 last_work_seen = syncer_delayno; 1824 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) 1825 syncer_state = SYNCER_SHUTTING_DOWN; 1826 while (!LIST_EMPTY(slp)) { 1827 error = sync_vnode(slp, &bo, td); 1828 if (error == 1) { 1829 LIST_REMOVE(bo, bo_synclist); 1830 LIST_INSERT_HEAD(next, bo, bo_synclist); 1831 continue; 1832 } 1833 } 1834 if (!LIST_EMPTY(gslp)) { 1835 mtx_unlock(&sync_mtx); 1836 mtx_lock(&Giant); 1837 mtx_lock(&sync_mtx); 1838 while (!LIST_EMPTY(gslp)) { 1839 error = sync_vnode(gslp, &bo, td); 1840 if (error == 1) { 1841 LIST_REMOVE(bo, bo_synclist); 1842 LIST_INSERT_HEAD(gnext, bo, 1843 bo_synclist); 1844 continue; 1845 } 1846 } 1847 mtx_unlock(&Giant); 1848 } 1849 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) 1850 syncer_final_iter--; 1851 /* 1852 * The variable rushjob allows the kernel to speed up the 1853 * processing of the filesystem syncer process. A rushjob 1854 * value of N tells the filesystem syncer to process the next 1855 * N seconds worth of work on its queue ASAP. Currently rushjob 1856 * is used by the soft update code to speed up the filesystem 1857 * syncer process when the incore state is getting so far 1858 * ahead of the disk that the kernel memory pool is being 1859 * threatened with exhaustion. 1860 */ 1861 if (rushjob > 0) { 1862 rushjob -= 1; 1863 continue; 1864 } 1865 /* 1866 * Just sleep for a short period of time between 1867 * iterations when shutting down to allow some I/O 1868 * to happen. 1869 * 1870 * If it has taken us less than a second to process the 1871 * current work, then wait. Otherwise start right over 1872 * again. We can still lose time if any single round 1873 * takes more than two seconds, but it does not really 1874 * matter as we are just trying to generally pace the 1875 * filesystem activity. 1876 */ 1877 if (syncer_state != SYNCER_RUNNING) 1878 cv_timedwait(&sync_wakeup, &sync_mtx, 1879 hz / SYNCER_SHUTDOWN_SPEEDUP); 1880 else if (time_uptime == starttime) 1881 cv_timedwait(&sync_wakeup, &sync_mtx, hz); 1882 } 1883 } 1884 1885 /* 1886 * Request the syncer daemon to speed up its work. 1887 * We never push it to speed up more than half of its 1888 * normal turn time, otherwise it could take over the cpu. 1889 */ 1890 int 1891 speedup_syncer(void) 1892 { 1893 int ret = 0; 1894 1895 mtx_lock(&sync_mtx); 1896 if (rushjob < syncdelay / 2) { 1897 rushjob += 1; 1898 stat_rush_requests += 1; 1899 ret = 1; 1900 } 1901 mtx_unlock(&sync_mtx); 1902 cv_broadcast(&sync_wakeup); 1903 return (ret); 1904 } 1905 1906 /* 1907 * Tell the syncer to speed up its work and run though its work 1908 * list several times, then tell it to shut down. 1909 */ 1910 static void 1911 syncer_shutdown(void *arg, int howto) 1912 { 1913 1914 if (howto & RB_NOSYNC) 1915 return; 1916 mtx_lock(&sync_mtx); 1917 syncer_state = SYNCER_SHUTTING_DOWN; 1918 rushjob = 0; 1919 mtx_unlock(&sync_mtx); 1920 cv_broadcast(&sync_wakeup); 1921 kproc_shutdown(arg, howto); 1922 } 1923 1924 /* 1925 * Reassign a buffer from one vnode to another. 1926 * Used to assign file specific control information 1927 * (indirect blocks) to the vnode to which they belong. 1928 */ 1929 void 1930 reassignbuf(struct buf *bp) 1931 { 1932 struct vnode *vp; 1933 struct bufobj *bo; 1934 int delay; 1935 #ifdef INVARIANTS 1936 struct bufv *bv; 1937 #endif 1938 1939 vp = bp->b_vp; 1940 bo = bp->b_bufobj; 1941 ++reassignbufcalls; 1942 1943 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", 1944 bp, bp->b_vp, bp->b_flags); 1945 /* 1946 * B_PAGING flagged buffers cannot be reassigned because their vp 1947 * is not fully linked in. 1948 */ 1949 if (bp->b_flags & B_PAGING) 1950 panic("cannot reassign paging buffer"); 1951 1952 /* 1953 * Delete from old vnode list, if on one. 1954 */ 1955 BO_LOCK(bo); 1956 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1957 buf_vlist_remove(bp); 1958 else 1959 panic("reassignbuf: Buffer %p not on queue.", bp); 1960 /* 1961 * If dirty, put on list of dirty buffers; otherwise insert onto list 1962 * of clean buffers. 1963 */ 1964 if (bp->b_flags & B_DELWRI) { 1965 if ((bo->bo_flag & BO_ONWORKLST) == 0) { 1966 switch (vp->v_type) { 1967 case VDIR: 1968 delay = dirdelay; 1969 break; 1970 case VCHR: 1971 delay = metadelay; 1972 break; 1973 default: 1974 delay = filedelay; 1975 } 1976 vn_syncer_add_to_worklist(bo, delay); 1977 } 1978 buf_vlist_add(bp, bo, BX_VNDIRTY); 1979 } else { 1980 buf_vlist_add(bp, bo, BX_VNCLEAN); 1981 1982 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 1983 mtx_lock(&sync_mtx); 1984 LIST_REMOVE(bo, bo_synclist); 1985 syncer_worklist_len--; 1986 mtx_unlock(&sync_mtx); 1987 bo->bo_flag &= ~BO_ONWORKLST; 1988 } 1989 } 1990 #ifdef INVARIANTS 1991 bv = &bo->bo_clean; 1992 bp = TAILQ_FIRST(&bv->bv_hd); 1993 KASSERT(bp == NULL || bp->b_bufobj == bo, 1994 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 1995 bp = TAILQ_LAST(&bv->bv_hd, buflists); 1996 KASSERT(bp == NULL || bp->b_bufobj == bo, 1997 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 1998 bv = &bo->bo_dirty; 1999 bp = TAILQ_FIRST(&bv->bv_hd); 2000 KASSERT(bp == NULL || bp->b_bufobj == bo, 2001 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2002 bp = TAILQ_LAST(&bv->bv_hd, buflists); 2003 KASSERT(bp == NULL || bp->b_bufobj == bo, 2004 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2005 #endif 2006 BO_UNLOCK(bo); 2007 } 2008 2009 /* 2010 * Increment the use and hold counts on the vnode, taking care to reference 2011 * the driver's usecount if this is a chardev. The vholdl() will remove 2012 * the vnode from the free list if it is presently free. Requires the 2013 * vnode interlock and returns with it held. 2014 */ 2015 static void 2016 v_incr_usecount(struct vnode *vp) 2017 { 2018 2019 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2020 vp->v_usecount++; 2021 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2022 dev_lock(); 2023 vp->v_rdev->si_usecount++; 2024 dev_unlock(); 2025 } 2026 vholdl(vp); 2027 } 2028 2029 /* 2030 * Turn a holdcnt into a use+holdcnt such that only one call to 2031 * v_decr_usecount is needed. 2032 */ 2033 static void 2034 v_upgrade_usecount(struct vnode *vp) 2035 { 2036 2037 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2038 vp->v_usecount++; 2039 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2040 dev_lock(); 2041 vp->v_rdev->si_usecount++; 2042 dev_unlock(); 2043 } 2044 } 2045 2046 /* 2047 * Decrement the vnode use and hold count along with the driver's usecount 2048 * if this is a chardev. The vdropl() below releases the vnode interlock 2049 * as it may free the vnode. 2050 */ 2051 static void 2052 v_decr_usecount(struct vnode *vp) 2053 { 2054 2055 ASSERT_VI_LOCKED(vp, __FUNCTION__); 2056 VNASSERT(vp->v_usecount > 0, vp, 2057 ("v_decr_usecount: negative usecount")); 2058 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2059 vp->v_usecount--; 2060 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2061 dev_lock(); 2062 vp->v_rdev->si_usecount--; 2063 dev_unlock(); 2064 } 2065 vdropl(vp); 2066 } 2067 2068 /* 2069 * Decrement only the use count and driver use count. This is intended to 2070 * be paired with a follow on vdropl() to release the remaining hold count. 2071 * In this way we may vgone() a vnode with a 0 usecount without risk of 2072 * having it end up on a free list because the hold count is kept above 0. 2073 */ 2074 static void 2075 v_decr_useonly(struct vnode *vp) 2076 { 2077 2078 ASSERT_VI_LOCKED(vp, __FUNCTION__); 2079 VNASSERT(vp->v_usecount > 0, vp, 2080 ("v_decr_useonly: negative usecount")); 2081 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2082 vp->v_usecount--; 2083 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2084 dev_lock(); 2085 vp->v_rdev->si_usecount--; 2086 dev_unlock(); 2087 } 2088 } 2089 2090 /* 2091 * Grab a particular vnode from the free list, increment its 2092 * reference count and lock it. VI_DOOMED is set if the vnode 2093 * is being destroyed. Only callers who specify LK_RETRY will 2094 * see doomed vnodes. If inactive processing was delayed in 2095 * vput try to do it here. 2096 */ 2097 int 2098 vget(struct vnode *vp, int flags, struct thread *td) 2099 { 2100 int error; 2101 2102 error = 0; 2103 VFS_ASSERT_GIANT(vp->v_mount); 2104 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 2105 ("vget: invalid lock operation")); 2106 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); 2107 2108 if ((flags & LK_INTERLOCK) == 0) 2109 VI_LOCK(vp); 2110 vholdl(vp); 2111 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) { 2112 vdrop(vp); 2113 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__, 2114 vp); 2115 return (error); 2116 } 2117 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) 2118 panic("vget: vn_lock failed to return ENOENT\n"); 2119 VI_LOCK(vp); 2120 /* Upgrade our holdcnt to a usecount. */ 2121 v_upgrade_usecount(vp); 2122 /* 2123 * We don't guarantee that any particular close will 2124 * trigger inactive processing so just make a best effort 2125 * here at preventing a reference to a removed file. If 2126 * we don't succeed no harm is done. 2127 */ 2128 if (vp->v_iflag & VI_OWEINACT) { 2129 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && 2130 (flags & LK_NOWAIT) == 0) 2131 vinactive(vp, td); 2132 vp->v_iflag &= ~VI_OWEINACT; 2133 } 2134 VI_UNLOCK(vp); 2135 return (0); 2136 } 2137 2138 /* 2139 * Increase the reference count of a vnode. 2140 */ 2141 void 2142 vref(struct vnode *vp) 2143 { 2144 2145 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2146 VI_LOCK(vp); 2147 v_incr_usecount(vp); 2148 VI_UNLOCK(vp); 2149 } 2150 2151 /* 2152 * Return reference count of a vnode. 2153 * 2154 * The results of this call are only guaranteed when some mechanism other 2155 * than the VI lock is used to stop other processes from gaining references 2156 * to the vnode. This may be the case if the caller holds the only reference. 2157 * This is also useful when stale data is acceptable as race conditions may 2158 * be accounted for by some other means. 2159 */ 2160 int 2161 vrefcnt(struct vnode *vp) 2162 { 2163 int usecnt; 2164 2165 VI_LOCK(vp); 2166 usecnt = vp->v_usecount; 2167 VI_UNLOCK(vp); 2168 2169 return (usecnt); 2170 } 2171 2172 #define VPUTX_VRELE 1 2173 #define VPUTX_VPUT 2 2174 #define VPUTX_VUNREF 3 2175 2176 static void 2177 vputx(struct vnode *vp, int func) 2178 { 2179 int error; 2180 2181 KASSERT(vp != NULL, ("vputx: null vp")); 2182 if (func == VPUTX_VUNREF) 2183 ASSERT_VOP_ELOCKED(vp, "vunref"); 2184 else if (func == VPUTX_VPUT) 2185 ASSERT_VOP_LOCKED(vp, "vput"); 2186 else 2187 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func")); 2188 VFS_ASSERT_GIANT(vp->v_mount); 2189 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2190 VI_LOCK(vp); 2191 2192 /* Skip this v_writecount check if we're going to panic below. */ 2193 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp, 2194 ("vputx: missed vn_close")); 2195 error = 0; 2196 2197 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) && 2198 vp->v_usecount == 1)) { 2199 if (func == VPUTX_VPUT) 2200 VOP_UNLOCK(vp, 0); 2201 v_decr_usecount(vp); 2202 return; 2203 } 2204 2205 if (vp->v_usecount != 1) { 2206 vprint("vputx: negative ref count", vp); 2207 panic("vputx: negative ref cnt"); 2208 } 2209 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp); 2210 /* 2211 * We want to hold the vnode until the inactive finishes to 2212 * prevent vgone() races. We drop the use count here and the 2213 * hold count below when we're done. 2214 */ 2215 v_decr_useonly(vp); 2216 /* 2217 * We must call VOP_INACTIVE with the node locked. Mark 2218 * as VI_DOINGINACT to avoid recursion. 2219 */ 2220 vp->v_iflag |= VI_OWEINACT; 2221 if (func == VPUTX_VRELE) { 2222 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2223 VI_LOCK(vp); 2224 } else if (func == VPUTX_VPUT && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 2225 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | LK_NOWAIT); 2226 VI_LOCK(vp); 2227 } 2228 if (vp->v_usecount > 0) 2229 vp->v_iflag &= ~VI_OWEINACT; 2230 if (error == 0) { 2231 if (vp->v_iflag & VI_OWEINACT) 2232 vinactive(vp, curthread); 2233 if (func != VPUTX_VUNREF) 2234 VOP_UNLOCK(vp, 0); 2235 } 2236 vdropl(vp); 2237 } 2238 2239 /* 2240 * Vnode put/release. 2241 * If count drops to zero, call inactive routine and return to freelist. 2242 */ 2243 void 2244 vrele(struct vnode *vp) 2245 { 2246 2247 vputx(vp, VPUTX_VRELE); 2248 } 2249 2250 /* 2251 * Release an already locked vnode. This give the same effects as 2252 * unlock+vrele(), but takes less time and avoids releasing and 2253 * re-aquiring the lock (as vrele() acquires the lock internally.) 2254 */ 2255 void 2256 vput(struct vnode *vp) 2257 { 2258 2259 vputx(vp, VPUTX_VPUT); 2260 } 2261 2262 /* 2263 * Release an exclusively locked vnode. Do not unlock the vnode lock. 2264 */ 2265 void 2266 vunref(struct vnode *vp) 2267 { 2268 2269 vputx(vp, VPUTX_VUNREF); 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)) != 0) { 2399 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d", 2400 __func__, error); 2401 return (error); 2402 } 2403 vput(rootvp); 2404 } 2405 MNT_ILOCK(mp); 2406 loop: 2407 MNT_VNODE_FOREACH(vp, mp, mvp) { 2408 VI_LOCK(vp); 2409 vholdl(vp); 2410 MNT_IUNLOCK(mp); 2411 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE); 2412 if (error) { 2413 vdrop(vp); 2414 MNT_ILOCK(mp); 2415 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); 2416 goto loop; 2417 } 2418 /* 2419 * Skip over a vnodes marked VV_SYSTEM. 2420 */ 2421 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { 2422 VOP_UNLOCK(vp, 0); 2423 vdrop(vp); 2424 MNT_ILOCK(mp); 2425 continue; 2426 } 2427 /* 2428 * If WRITECLOSE is set, flush out unlinked but still open 2429 * files (even if open only for reading) and regular file 2430 * vnodes open for writing. 2431 */ 2432 if (flags & WRITECLOSE) { 2433 error = VOP_GETATTR(vp, &vattr, td->td_ucred); 2434 VI_LOCK(vp); 2435 2436 if ((vp->v_type == VNON || 2437 (error == 0 && vattr.va_nlink > 0)) && 2438 (vp->v_writecount == 0 || vp->v_type != VREG)) { 2439 VOP_UNLOCK(vp, 0); 2440 vdropl(vp); 2441 MNT_ILOCK(mp); 2442 continue; 2443 } 2444 } else 2445 VI_LOCK(vp); 2446 /* 2447 * With v_usecount == 0, all we need to do is clear out the 2448 * vnode data structures and we are done. 2449 * 2450 * If FORCECLOSE is set, forcibly close the vnode. 2451 */ 2452 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { 2453 VNASSERT(vp->v_usecount == 0 || 2454 (vp->v_type != VCHR && vp->v_type != VBLK), vp, 2455 ("device VNODE %p is FORCECLOSED", vp)); 2456 vgonel(vp); 2457 } else { 2458 busy++; 2459 #ifdef DIAGNOSTIC 2460 if (busyprt) 2461 vprint("vflush: busy vnode", vp); 2462 #endif 2463 } 2464 VOP_UNLOCK(vp, 0); 2465 vdropl(vp); 2466 MNT_ILOCK(mp); 2467 } 2468 MNT_IUNLOCK(mp); 2469 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 2470 /* 2471 * If just the root vnode is busy, and if its refcount 2472 * is equal to `rootrefs', then go ahead and kill it. 2473 */ 2474 VI_LOCK(rootvp); 2475 KASSERT(busy > 0, ("vflush: not busy")); 2476 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, 2477 ("vflush: usecount %d < rootrefs %d", 2478 rootvp->v_usecount, rootrefs)); 2479 if (busy == 1 && rootvp->v_usecount == rootrefs) { 2480 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK); 2481 vgone(rootvp); 2482 VOP_UNLOCK(rootvp, 0); 2483 busy = 0; 2484 } else 2485 VI_UNLOCK(rootvp); 2486 } 2487 if (busy) { 2488 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__, 2489 busy); 2490 return (EBUSY); 2491 } 2492 for (; rootrefs > 0; rootrefs--) 2493 vrele(rootvp); 2494 return (0); 2495 } 2496 2497 /* 2498 * Recycle an unused vnode to the front of the free list. 2499 */ 2500 int 2501 vrecycle(struct vnode *vp, struct thread *td) 2502 { 2503 int recycled; 2504 2505 ASSERT_VOP_ELOCKED(vp, "vrecycle"); 2506 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2507 recycled = 0; 2508 VI_LOCK(vp); 2509 if (vp->v_usecount == 0) { 2510 recycled = 1; 2511 vgonel(vp); 2512 } 2513 VI_UNLOCK(vp); 2514 return (recycled); 2515 } 2516 2517 /* 2518 * Eliminate all activity associated with a vnode 2519 * in preparation for reuse. 2520 */ 2521 void 2522 vgone(struct vnode *vp) 2523 { 2524 VI_LOCK(vp); 2525 vgonel(vp); 2526 VI_UNLOCK(vp); 2527 } 2528 2529 /* 2530 * vgone, with the vp interlock held. 2531 */ 2532 void 2533 vgonel(struct vnode *vp) 2534 { 2535 struct thread *td; 2536 int oweinact; 2537 int active; 2538 struct mount *mp; 2539 2540 ASSERT_VOP_ELOCKED(vp, "vgonel"); 2541 ASSERT_VI_LOCKED(vp, "vgonel"); 2542 VNASSERT(vp->v_holdcnt, vp, 2543 ("vgonel: vp %p has no reference.", vp)); 2544 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2545 td = curthread; 2546 2547 /* 2548 * Don't vgonel if we're already doomed. 2549 */ 2550 if (vp->v_iflag & VI_DOOMED) 2551 return; 2552 vp->v_iflag |= VI_DOOMED; 2553 /* 2554 * Check to see if the vnode is in use. If so, we have to call 2555 * VOP_CLOSE() and VOP_INACTIVE(). 2556 */ 2557 active = vp->v_usecount; 2558 oweinact = (vp->v_iflag & VI_OWEINACT); 2559 VI_UNLOCK(vp); 2560 /* 2561 * Clean out any buffers associated with the vnode. 2562 * If the flush fails, just toss the buffers. 2563 */ 2564 mp = NULL; 2565 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) 2566 (void) vn_start_secondary_write(vp, &mp, V_WAIT); 2567 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) 2568 vinvalbuf(vp, 0, 0, 0); 2569 2570 /* 2571 * If purging an active vnode, it must be closed and 2572 * deactivated before being reclaimed. 2573 */ 2574 if (active) 2575 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); 2576 if (oweinact || active) { 2577 VI_LOCK(vp); 2578 if ((vp->v_iflag & VI_DOINGINACT) == 0) 2579 vinactive(vp, td); 2580 VI_UNLOCK(vp); 2581 } 2582 /* 2583 * Reclaim the vnode. 2584 */ 2585 if (VOP_RECLAIM(vp, td)) 2586 panic("vgone: cannot reclaim"); 2587 if (mp != NULL) 2588 vn_finished_secondary_write(mp); 2589 VNASSERT(vp->v_object == NULL, vp, 2590 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag)); 2591 /* 2592 * Clear the advisory locks and wake up waiting threads. 2593 */ 2594 (void)VOP_ADVLOCKPURGE(vp); 2595 /* 2596 * Delete from old mount point vnode list. 2597 */ 2598 delmntque(vp); 2599 cache_purge(vp); 2600 /* 2601 * Done with purge, reset to the standard lock and invalidate 2602 * the vnode. 2603 */ 2604 VI_LOCK(vp); 2605 vp->v_vnlock = &vp->v_lock; 2606 vp->v_op = &dead_vnodeops; 2607 vp->v_tag = "none"; 2608 vp->v_type = VBAD; 2609 } 2610 2611 /* 2612 * Calculate the total number of references to a special device. 2613 */ 2614 int 2615 vcount(struct vnode *vp) 2616 { 2617 int count; 2618 2619 dev_lock(); 2620 count = vp->v_rdev->si_usecount; 2621 dev_unlock(); 2622 return (count); 2623 } 2624 2625 /* 2626 * Same as above, but using the struct cdev *as argument 2627 */ 2628 int 2629 count_dev(struct cdev *dev) 2630 { 2631 int count; 2632 2633 dev_lock(); 2634 count = dev->si_usecount; 2635 dev_unlock(); 2636 return(count); 2637 } 2638 2639 /* 2640 * Print out a description of a vnode. 2641 */ 2642 static char *typename[] = 2643 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", 2644 "VMARKER"}; 2645 2646 void 2647 vn_printf(struct vnode *vp, const char *fmt, ...) 2648 { 2649 va_list ap; 2650 char buf[256], buf2[16]; 2651 u_long flags; 2652 2653 va_start(ap, fmt); 2654 vprintf(fmt, ap); 2655 va_end(ap); 2656 printf("%p: ", (void *)vp); 2657 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]); 2658 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n", 2659 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere); 2660 buf[0] = '\0'; 2661 buf[1] = '\0'; 2662 if (vp->v_vflag & VV_ROOT) 2663 strlcat(buf, "|VV_ROOT", sizeof(buf)); 2664 if (vp->v_vflag & VV_ISTTY) 2665 strlcat(buf, "|VV_ISTTY", sizeof(buf)); 2666 if (vp->v_vflag & VV_NOSYNC) 2667 strlcat(buf, "|VV_NOSYNC", sizeof(buf)); 2668 if (vp->v_vflag & VV_CACHEDLABEL) 2669 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf)); 2670 if (vp->v_vflag & VV_TEXT) 2671 strlcat(buf, "|VV_TEXT", sizeof(buf)); 2672 if (vp->v_vflag & VV_COPYONWRITE) 2673 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf)); 2674 if (vp->v_vflag & VV_SYSTEM) 2675 strlcat(buf, "|VV_SYSTEM", sizeof(buf)); 2676 if (vp->v_vflag & VV_PROCDEP) 2677 strlcat(buf, "|VV_PROCDEP", sizeof(buf)); 2678 if (vp->v_vflag & VV_NOKNOTE) 2679 strlcat(buf, "|VV_NOKNOTE", sizeof(buf)); 2680 if (vp->v_vflag & VV_DELETED) 2681 strlcat(buf, "|VV_DELETED", sizeof(buf)); 2682 if (vp->v_vflag & VV_MD) 2683 strlcat(buf, "|VV_MD", sizeof(buf)); 2684 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | 2685 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP | 2686 VV_NOKNOTE | VV_DELETED | VV_MD); 2687 if (flags != 0) { 2688 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags); 2689 strlcat(buf, buf2, sizeof(buf)); 2690 } 2691 if (vp->v_iflag & VI_MOUNT) 2692 strlcat(buf, "|VI_MOUNT", sizeof(buf)); 2693 if (vp->v_iflag & VI_AGE) 2694 strlcat(buf, "|VI_AGE", sizeof(buf)); 2695 if (vp->v_iflag & VI_DOOMED) 2696 strlcat(buf, "|VI_DOOMED", sizeof(buf)); 2697 if (vp->v_iflag & VI_FREE) 2698 strlcat(buf, "|VI_FREE", sizeof(buf)); 2699 if (vp->v_iflag & VI_DOINGINACT) 2700 strlcat(buf, "|VI_DOINGINACT", sizeof(buf)); 2701 if (vp->v_iflag & VI_OWEINACT) 2702 strlcat(buf, "|VI_OWEINACT", sizeof(buf)); 2703 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE | 2704 VI_DOINGINACT | VI_OWEINACT); 2705 if (flags != 0) { 2706 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags); 2707 strlcat(buf, buf2, sizeof(buf)); 2708 } 2709 printf(" flags (%s)\n", buf + 1); 2710 if (mtx_owned(VI_MTX(vp))) 2711 printf(" VI_LOCKed"); 2712 if (vp->v_object != NULL) 2713 printf(" v_object %p ref %d pages %d\n", 2714 vp->v_object, vp->v_object->ref_count, 2715 vp->v_object->resident_page_count); 2716 printf(" "); 2717 lockmgr_printinfo(vp->v_vnlock); 2718 if (vp->v_data != NULL) 2719 VOP_PRINT(vp); 2720 } 2721 2722 #ifdef DDB 2723 /* 2724 * List all of the locked vnodes in the system. 2725 * Called when debugging the kernel. 2726 */ 2727 DB_SHOW_COMMAND(lockedvnods, lockedvnodes) 2728 { 2729 struct mount *mp, *nmp; 2730 struct vnode *vp; 2731 2732 /* 2733 * Note: because this is DDB, we can't obey the locking semantics 2734 * for these structures, which means we could catch an inconsistent 2735 * state and dereference a nasty pointer. Not much to be done 2736 * about that. 2737 */ 2738 db_printf("Locked vnodes\n"); 2739 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2740 nmp = TAILQ_NEXT(mp, mnt_list); 2741 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2742 if (vp->v_type != VMARKER && 2743 VOP_ISLOCKED(vp)) 2744 vprint("", vp); 2745 } 2746 nmp = TAILQ_NEXT(mp, mnt_list); 2747 } 2748 } 2749 2750 /* 2751 * Show details about the given vnode. 2752 */ 2753 DB_SHOW_COMMAND(vnode, db_show_vnode) 2754 { 2755 struct vnode *vp; 2756 2757 if (!have_addr) 2758 return; 2759 vp = (struct vnode *)addr; 2760 vn_printf(vp, "vnode "); 2761 } 2762 2763 /* 2764 * Show details about the given mount point. 2765 */ 2766 DB_SHOW_COMMAND(mount, db_show_mount) 2767 { 2768 struct mount *mp; 2769 struct vfsopt *opt; 2770 struct statfs *sp; 2771 struct vnode *vp; 2772 char buf[512]; 2773 u_int flags; 2774 2775 if (!have_addr) { 2776 /* No address given, print short info about all mount points. */ 2777 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2778 db_printf("%p %s on %s (%s)\n", mp, 2779 mp->mnt_stat.f_mntfromname, 2780 mp->mnt_stat.f_mntonname, 2781 mp->mnt_stat.f_fstypename); 2782 if (db_pager_quit) 2783 break; 2784 } 2785 db_printf("\nMore info: show mount <addr>\n"); 2786 return; 2787 } 2788 2789 mp = (struct mount *)addr; 2790 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, 2791 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); 2792 2793 buf[0] = '\0'; 2794 flags = mp->mnt_flag; 2795 #define MNT_FLAG(flag) do { \ 2796 if (flags & (flag)) { \ 2797 if (buf[0] != '\0') \ 2798 strlcat(buf, ", ", sizeof(buf)); \ 2799 strlcat(buf, (#flag) + 4, sizeof(buf)); \ 2800 flags &= ~(flag); \ 2801 } \ 2802 } while (0) 2803 MNT_FLAG(MNT_RDONLY); 2804 MNT_FLAG(MNT_SYNCHRONOUS); 2805 MNT_FLAG(MNT_NOEXEC); 2806 MNT_FLAG(MNT_NOSUID); 2807 MNT_FLAG(MNT_UNION); 2808 MNT_FLAG(MNT_ASYNC); 2809 MNT_FLAG(MNT_SUIDDIR); 2810 MNT_FLAG(MNT_SOFTDEP); 2811 MNT_FLAG(MNT_NOSYMFOLLOW); 2812 MNT_FLAG(MNT_GJOURNAL); 2813 MNT_FLAG(MNT_MULTILABEL); 2814 MNT_FLAG(MNT_ACLS); 2815 MNT_FLAG(MNT_NOATIME); 2816 MNT_FLAG(MNT_NOCLUSTERR); 2817 MNT_FLAG(MNT_NOCLUSTERW); 2818 MNT_FLAG(MNT_NFS4ACLS); 2819 MNT_FLAG(MNT_EXRDONLY); 2820 MNT_FLAG(MNT_EXPORTED); 2821 MNT_FLAG(MNT_DEFEXPORTED); 2822 MNT_FLAG(MNT_EXPORTANON); 2823 MNT_FLAG(MNT_EXKERB); 2824 MNT_FLAG(MNT_EXPUBLIC); 2825 MNT_FLAG(MNT_LOCAL); 2826 MNT_FLAG(MNT_QUOTA); 2827 MNT_FLAG(MNT_ROOTFS); 2828 MNT_FLAG(MNT_USER); 2829 MNT_FLAG(MNT_IGNORE); 2830 MNT_FLAG(MNT_UPDATE); 2831 MNT_FLAG(MNT_DELEXPORT); 2832 MNT_FLAG(MNT_RELOAD); 2833 MNT_FLAG(MNT_FORCE); 2834 MNT_FLAG(MNT_SNAPSHOT); 2835 MNT_FLAG(MNT_BYFSID); 2836 MNT_FLAG(MNT_SOFTDEP); 2837 #undef MNT_FLAG 2838 if (flags != 0) { 2839 if (buf[0] != '\0') 2840 strlcat(buf, ", ", sizeof(buf)); 2841 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), 2842 "0x%08x", flags); 2843 } 2844 db_printf(" mnt_flag = %s\n", buf); 2845 2846 buf[0] = '\0'; 2847 flags = mp->mnt_kern_flag; 2848 #define MNT_KERN_FLAG(flag) do { \ 2849 if (flags & (flag)) { \ 2850 if (buf[0] != '\0') \ 2851 strlcat(buf, ", ", sizeof(buf)); \ 2852 strlcat(buf, (#flag) + 5, sizeof(buf)); \ 2853 flags &= ~(flag); \ 2854 } \ 2855 } while (0) 2856 MNT_KERN_FLAG(MNTK_UNMOUNTF); 2857 MNT_KERN_FLAG(MNTK_ASYNC); 2858 MNT_KERN_FLAG(MNTK_SOFTDEP); 2859 MNT_KERN_FLAG(MNTK_NOINSMNTQ); 2860 MNT_KERN_FLAG(MNTK_DRAINING); 2861 MNT_KERN_FLAG(MNTK_REFEXPIRE); 2862 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED); 2863 MNT_KERN_FLAG(MNTK_SHARED_WRITES); 2864 MNT_KERN_FLAG(MNTK_UNMOUNT); 2865 MNT_KERN_FLAG(MNTK_MWAIT); 2866 MNT_KERN_FLAG(MNTK_SUSPEND); 2867 MNT_KERN_FLAG(MNTK_SUSPEND2); 2868 MNT_KERN_FLAG(MNTK_SUSPENDED); 2869 MNT_KERN_FLAG(MNTK_MPSAFE); 2870 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED); 2871 MNT_KERN_FLAG(MNTK_NOKNOTE); 2872 #undef MNT_KERN_FLAG 2873 if (flags != 0) { 2874 if (buf[0] != '\0') 2875 strlcat(buf, ", ", sizeof(buf)); 2876 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), 2877 "0x%08x", flags); 2878 } 2879 db_printf(" mnt_kern_flag = %s\n", buf); 2880 2881 db_printf(" mnt_opt = "); 2882 opt = TAILQ_FIRST(mp->mnt_opt); 2883 if (opt != NULL) { 2884 db_printf("%s", opt->name); 2885 opt = TAILQ_NEXT(opt, link); 2886 while (opt != NULL) { 2887 db_printf(", %s", opt->name); 2888 opt = TAILQ_NEXT(opt, link); 2889 } 2890 } 2891 db_printf("\n"); 2892 2893 sp = &mp->mnt_stat; 2894 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx " 2895 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju " 2896 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju " 2897 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n", 2898 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags, 2899 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize, 2900 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree, 2901 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files, 2902 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites, 2903 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads, 2904 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax, 2905 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]); 2906 2907 db_printf(" mnt_cred = { uid=%u ruid=%u", 2908 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid); 2909 if (jailed(mp->mnt_cred)) 2910 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id); 2911 db_printf(" }\n"); 2912 db_printf(" mnt_ref = %d\n", mp->mnt_ref); 2913 db_printf(" mnt_gen = %d\n", mp->mnt_gen); 2914 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize); 2915 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount); 2916 db_printf(" mnt_noasync = %u\n", mp->mnt_noasync); 2917 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen); 2918 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max); 2919 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed); 2920 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes); 2921 db_printf(" mnt_secondary_accwrites = %d\n", 2922 mp->mnt_secondary_accwrites); 2923 db_printf(" mnt_gjprovider = %s\n", 2924 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL"); 2925 db_printf("\n"); 2926 2927 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2928 if (vp->v_type != VMARKER) { 2929 vn_printf(vp, "vnode "); 2930 if (db_pager_quit) 2931 break; 2932 } 2933 } 2934 } 2935 #endif /* DDB */ 2936 2937 /* 2938 * Fill in a struct xvfsconf based on a struct vfsconf. 2939 */ 2940 static void 2941 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp) 2942 { 2943 2944 strcpy(xvfsp->vfc_name, vfsp->vfc_name); 2945 xvfsp->vfc_typenum = vfsp->vfc_typenum; 2946 xvfsp->vfc_refcount = vfsp->vfc_refcount; 2947 xvfsp->vfc_flags = vfsp->vfc_flags; 2948 /* 2949 * These are unused in userland, we keep them 2950 * to not break binary compatibility. 2951 */ 2952 xvfsp->vfc_vfsops = NULL; 2953 xvfsp->vfc_next = NULL; 2954 } 2955 2956 /* 2957 * Top level filesystem related information gathering. 2958 */ 2959 static int 2960 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) 2961 { 2962 struct vfsconf *vfsp; 2963 struct xvfsconf xvfsp; 2964 int error; 2965 2966 error = 0; 2967 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 2968 bzero(&xvfsp, sizeof(xvfsp)); 2969 vfsconf2x(vfsp, &xvfsp); 2970 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp); 2971 if (error) 2972 break; 2973 } 2974 return (error); 2975 } 2976 2977 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist, 2978 "S,xvfsconf", "List of all configured filesystems"); 2979 2980 #ifndef BURN_BRIDGES 2981 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); 2982 2983 static int 2984 vfs_sysctl(SYSCTL_HANDLER_ARGS) 2985 { 2986 int *name = (int *)arg1 - 1; /* XXX */ 2987 u_int namelen = arg2 + 1; /* XXX */ 2988 struct vfsconf *vfsp; 2989 struct xvfsconf xvfsp; 2990 2991 printf("WARNING: userland calling deprecated sysctl, " 2992 "please rebuild world\n"); 2993 2994 #if 1 || defined(COMPAT_PRELITE2) 2995 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 2996 if (namelen == 1) 2997 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 2998 #endif 2999 3000 switch (name[1]) { 3001 case VFS_MAXTYPENUM: 3002 if (namelen != 2) 3003 return (ENOTDIR); 3004 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 3005 case VFS_CONF: 3006 if (namelen != 3) 3007 return (ENOTDIR); /* overloaded */ 3008 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) 3009 if (vfsp->vfc_typenum == name[2]) 3010 break; 3011 if (vfsp == NULL) 3012 return (EOPNOTSUPP); 3013 bzero(&xvfsp, sizeof(xvfsp)); 3014 vfsconf2x(vfsp, &xvfsp); 3015 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); 3016 } 3017 return (EOPNOTSUPP); 3018 } 3019 3020 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, 3021 vfs_sysctl, "Generic filesystem"); 3022 3023 #if 1 || defined(COMPAT_PRELITE2) 3024 3025 static int 3026 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 3027 { 3028 int error; 3029 struct vfsconf *vfsp; 3030 struct ovfsconf ovfs; 3031 3032 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 3033 bzero(&ovfs, sizeof(ovfs)); 3034 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 3035 strcpy(ovfs.vfc_name, vfsp->vfc_name); 3036 ovfs.vfc_index = vfsp->vfc_typenum; 3037 ovfs.vfc_refcount = vfsp->vfc_refcount; 3038 ovfs.vfc_flags = vfsp->vfc_flags; 3039 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 3040 if (error) 3041 return error; 3042 } 3043 return 0; 3044 } 3045 3046 #endif /* 1 || COMPAT_PRELITE2 */ 3047 #endif /* !BURN_BRIDGES */ 3048 3049 #define KINFO_VNODESLOP 10 3050 #ifdef notyet 3051 /* 3052 * Dump vnode list (via sysctl). 3053 */ 3054 /* ARGSUSED */ 3055 static int 3056 sysctl_vnode(SYSCTL_HANDLER_ARGS) 3057 { 3058 struct xvnode *xvn; 3059 struct mount *mp; 3060 struct vnode *vp; 3061 int error, len, n; 3062 3063 /* 3064 * Stale numvnodes access is not fatal here. 3065 */ 3066 req->lock = 0; 3067 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; 3068 if (!req->oldptr) 3069 /* Make an estimate */ 3070 return (SYSCTL_OUT(req, 0, len)); 3071 3072 error = sysctl_wire_old_buffer(req, 0); 3073 if (error != 0) 3074 return (error); 3075 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); 3076 n = 0; 3077 mtx_lock(&mountlist_mtx); 3078 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 3079 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) 3080 continue; 3081 MNT_ILOCK(mp); 3082 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 3083 if (n == len) 3084 break; 3085 vref(vp); 3086 xvn[n].xv_size = sizeof *xvn; 3087 xvn[n].xv_vnode = vp; 3088 xvn[n].xv_id = 0; /* XXX compat */ 3089 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field 3090 XV_COPY(usecount); 3091 XV_COPY(writecount); 3092 XV_COPY(holdcnt); 3093 XV_COPY(mount); 3094 XV_COPY(numoutput); 3095 XV_COPY(type); 3096 #undef XV_COPY 3097 xvn[n].xv_flag = vp->v_vflag; 3098 3099 switch (vp->v_type) { 3100 case VREG: 3101 case VDIR: 3102 case VLNK: 3103 break; 3104 case VBLK: 3105 case VCHR: 3106 if (vp->v_rdev == NULL) { 3107 vrele(vp); 3108 continue; 3109 } 3110 xvn[n].xv_dev = dev2udev(vp->v_rdev); 3111 break; 3112 case VSOCK: 3113 xvn[n].xv_socket = vp->v_socket; 3114 break; 3115 case VFIFO: 3116 xvn[n].xv_fifo = vp->v_fifoinfo; 3117 break; 3118 case VNON: 3119 case VBAD: 3120 default: 3121 /* shouldn't happen? */ 3122 vrele(vp); 3123 continue; 3124 } 3125 vrele(vp); 3126 ++n; 3127 } 3128 MNT_IUNLOCK(mp); 3129 mtx_lock(&mountlist_mtx); 3130 vfs_unbusy(mp); 3131 if (n == len) 3132 break; 3133 } 3134 mtx_unlock(&mountlist_mtx); 3135 3136 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); 3137 free(xvn, M_TEMP); 3138 return (error); 3139 } 3140 3141 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 3142 0, 0, sysctl_vnode, "S,xvnode", ""); 3143 #endif 3144 3145 /* 3146 * Unmount all filesystems. The list is traversed in reverse order 3147 * of mounting to avoid dependencies. 3148 */ 3149 void 3150 vfs_unmountall(void) 3151 { 3152 struct mount *mp; 3153 struct thread *td; 3154 int error; 3155 3156 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread")); 3157 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__); 3158 td = curthread; 3159 3160 /* 3161 * Since this only runs when rebooting, it is not interlocked. 3162 */ 3163 while(!TAILQ_EMPTY(&mountlist)) { 3164 mp = TAILQ_LAST(&mountlist, mntlist); 3165 error = dounmount(mp, MNT_FORCE, td); 3166 if (error) { 3167 TAILQ_REMOVE(&mountlist, mp, mnt_list); 3168 /* 3169 * XXX: Due to the way in which we mount the root 3170 * file system off of devfs, devfs will generate a 3171 * "busy" warning when we try to unmount it before 3172 * the root. Don't print a warning as a result in 3173 * order to avoid false positive errors that may 3174 * cause needless upset. 3175 */ 3176 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) { 3177 printf("unmount of %s failed (", 3178 mp->mnt_stat.f_mntonname); 3179 if (error == EBUSY) 3180 printf("BUSY)\n"); 3181 else 3182 printf("%d)\n", error); 3183 } 3184 } else { 3185 /* The unmount has removed mp from the mountlist */ 3186 } 3187 } 3188 } 3189 3190 /* 3191 * perform msync on all vnodes under a mount point 3192 * the mount point must be locked. 3193 */ 3194 void 3195 vfs_msync(struct mount *mp, int flags) 3196 { 3197 struct vnode *vp, *mvp; 3198 struct vm_object *obj; 3199 3200 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 3201 MNT_ILOCK(mp); 3202 MNT_VNODE_FOREACH(vp, mp, mvp) { 3203 VI_LOCK(vp); 3204 obj = vp->v_object; 3205 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 && 3206 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) { 3207 MNT_IUNLOCK(mp); 3208 if (!vget(vp, 3209 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, 3210 curthread)) { 3211 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */ 3212 vput(vp); 3213 MNT_ILOCK(mp); 3214 continue; 3215 } 3216 3217 obj = vp->v_object; 3218 if (obj != NULL) { 3219 VM_OBJECT_LOCK(obj); 3220 vm_object_page_clean(obj, 0, 0, 3221 flags == MNT_WAIT ? 3222 OBJPC_SYNC : OBJPC_NOSYNC); 3223 VM_OBJECT_UNLOCK(obj); 3224 } 3225 vput(vp); 3226 } 3227 MNT_ILOCK(mp); 3228 } else 3229 VI_UNLOCK(vp); 3230 } 3231 MNT_IUNLOCK(mp); 3232 } 3233 3234 /* 3235 * Mark a vnode as free, putting it up for recycling. 3236 */ 3237 static void 3238 vfree(struct vnode *vp) 3239 { 3240 3241 ASSERT_VI_LOCKED(vp, "vfree"); 3242 mtx_lock(&vnode_free_list_mtx); 3243 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed.")); 3244 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free")); 3245 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't")); 3246 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp, 3247 ("vfree: Freeing doomed vnode")); 3248 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 3249 if (vp->v_iflag & VI_AGE) { 3250 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 3251 } else { 3252 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 3253 } 3254 freevnodes++; 3255 vp->v_iflag &= ~VI_AGE; 3256 vp->v_iflag |= VI_FREE; 3257 mtx_unlock(&vnode_free_list_mtx); 3258 } 3259 3260 /* 3261 * Opposite of vfree() - mark a vnode as in use. 3262 */ 3263 static void 3264 vbusy(struct vnode *vp) 3265 { 3266 ASSERT_VI_LOCKED(vp, "vbusy"); 3267 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free")); 3268 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed.")); 3269 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 3270 3271 mtx_lock(&vnode_free_list_mtx); 3272 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 3273 freevnodes--; 3274 vp->v_iflag &= ~(VI_FREE|VI_AGE); 3275 mtx_unlock(&vnode_free_list_mtx); 3276 } 3277 3278 static void 3279 destroy_vpollinfo(struct vpollinfo *vi) 3280 { 3281 knlist_destroy(&vi->vpi_selinfo.si_note); 3282 mtx_destroy(&vi->vpi_lock); 3283 uma_zfree(vnodepoll_zone, vi); 3284 } 3285 3286 /* 3287 * Initalize per-vnode helper structure to hold poll-related state. 3288 */ 3289 void 3290 v_addpollinfo(struct vnode *vp) 3291 { 3292 struct vpollinfo *vi; 3293 3294 if (vp->v_pollinfo != NULL) 3295 return; 3296 vi = uma_zalloc(vnodepoll_zone, M_WAITOK); 3297 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); 3298 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock, 3299 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked); 3300 VI_LOCK(vp); 3301 if (vp->v_pollinfo != NULL) { 3302 VI_UNLOCK(vp); 3303 destroy_vpollinfo(vi); 3304 return; 3305 } 3306 vp->v_pollinfo = vi; 3307 VI_UNLOCK(vp); 3308 } 3309 3310 /* 3311 * Record a process's interest in events which might happen to 3312 * a vnode. Because poll uses the historic select-style interface 3313 * internally, this routine serves as both the ``check for any 3314 * pending events'' and the ``record my interest in future events'' 3315 * functions. (These are done together, while the lock is held, 3316 * to avoid race conditions.) 3317 */ 3318 int 3319 vn_pollrecord(struct vnode *vp, struct thread *td, int events) 3320 { 3321 3322 v_addpollinfo(vp); 3323 mtx_lock(&vp->v_pollinfo->vpi_lock); 3324 if (vp->v_pollinfo->vpi_revents & events) { 3325 /* 3326 * This leaves events we are not interested 3327 * in available for the other process which 3328 * which presumably had requested them 3329 * (otherwise they would never have been 3330 * recorded). 3331 */ 3332 events &= vp->v_pollinfo->vpi_revents; 3333 vp->v_pollinfo->vpi_revents &= ~events; 3334 3335 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3336 return (events); 3337 } 3338 vp->v_pollinfo->vpi_events |= events; 3339 selrecord(td, &vp->v_pollinfo->vpi_selinfo); 3340 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3341 return (0); 3342 } 3343 3344 /* 3345 * Routine to create and manage a filesystem syncer vnode. 3346 */ 3347 #define sync_close ((int (*)(struct vop_close_args *))nullop) 3348 static int sync_fsync(struct vop_fsync_args *); 3349 static int sync_inactive(struct vop_inactive_args *); 3350 static int sync_reclaim(struct vop_reclaim_args *); 3351 3352 static struct vop_vector sync_vnodeops = { 3353 .vop_bypass = VOP_EOPNOTSUPP, 3354 .vop_close = sync_close, /* close */ 3355 .vop_fsync = sync_fsync, /* fsync */ 3356 .vop_inactive = sync_inactive, /* inactive */ 3357 .vop_reclaim = sync_reclaim, /* reclaim */ 3358 .vop_lock1 = vop_stdlock, /* lock */ 3359 .vop_unlock = vop_stdunlock, /* unlock */ 3360 .vop_islocked = vop_stdislocked, /* islocked */ 3361 }; 3362 3363 /* 3364 * Create a new filesystem syncer vnode for the specified mount point. 3365 */ 3366 void 3367 vfs_allocate_syncvnode(struct mount *mp) 3368 { 3369 struct vnode *vp; 3370 struct bufobj *bo; 3371 static long start, incr, next; 3372 int error; 3373 3374 /* Allocate a new vnode */ 3375 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp); 3376 if (error != 0) 3377 panic("vfs_allocate_syncvnode: getnewvnode() failed"); 3378 vp->v_type = VNON; 3379 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3380 vp->v_vflag |= VV_FORCEINSMQ; 3381 error = insmntque(vp, mp); 3382 if (error != 0) 3383 panic("vfs_allocate_syncvnode: insmntque() failed"); 3384 vp->v_vflag &= ~VV_FORCEINSMQ; 3385 VOP_UNLOCK(vp, 0); 3386 /* 3387 * Place the vnode onto the syncer worklist. We attempt to 3388 * scatter them about on the list so that they will go off 3389 * at evenly distributed times even if all the filesystems 3390 * are mounted at once. 3391 */ 3392 next += incr; 3393 if (next == 0 || next > syncer_maxdelay) { 3394 start /= 2; 3395 incr /= 2; 3396 if (start == 0) { 3397 start = syncer_maxdelay / 2; 3398 incr = syncer_maxdelay; 3399 } 3400 next = start; 3401 } 3402 bo = &vp->v_bufobj; 3403 BO_LOCK(bo); 3404 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0); 3405 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ 3406 mtx_lock(&sync_mtx); 3407 sync_vnode_count++; 3408 if (mp->mnt_syncer == NULL) { 3409 mp->mnt_syncer = vp; 3410 vp = NULL; 3411 } 3412 mtx_unlock(&sync_mtx); 3413 BO_UNLOCK(bo); 3414 if (vp != NULL) { 3415 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3416 vgone(vp); 3417 vput(vp); 3418 } 3419 } 3420 3421 void 3422 vfs_deallocate_syncvnode(struct mount *mp) 3423 { 3424 struct vnode *vp; 3425 3426 mtx_lock(&sync_mtx); 3427 vp = mp->mnt_syncer; 3428 if (vp != NULL) 3429 mp->mnt_syncer = NULL; 3430 mtx_unlock(&sync_mtx); 3431 if (vp != NULL) 3432 vrele(vp); 3433 } 3434 3435 /* 3436 * Do a lazy sync of the filesystem. 3437 */ 3438 static int 3439 sync_fsync(struct vop_fsync_args *ap) 3440 { 3441 struct vnode *syncvp = ap->a_vp; 3442 struct mount *mp = syncvp->v_mount; 3443 int error; 3444 struct bufobj *bo; 3445 3446 /* 3447 * We only need to do something if this is a lazy evaluation. 3448 */ 3449 if (ap->a_waitfor != MNT_LAZY) 3450 return (0); 3451 3452 /* 3453 * Move ourselves to the back of the sync list. 3454 */ 3455 bo = &syncvp->v_bufobj; 3456 BO_LOCK(bo); 3457 vn_syncer_add_to_worklist(bo, syncdelay); 3458 BO_UNLOCK(bo); 3459 3460 /* 3461 * Walk the list of vnodes pushing all that are dirty and 3462 * not already on the sync list. 3463 */ 3464 mtx_lock(&mountlist_mtx); 3465 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) { 3466 mtx_unlock(&mountlist_mtx); 3467 return (0); 3468 } 3469 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { 3470 vfs_unbusy(mp); 3471 return (0); 3472 } 3473 MNT_ILOCK(mp); 3474 mp->mnt_noasync++; 3475 mp->mnt_kern_flag &= ~MNTK_ASYNC; 3476 MNT_IUNLOCK(mp); 3477 vfs_msync(mp, MNT_NOWAIT); 3478 error = VFS_SYNC(mp, MNT_LAZY); 3479 MNT_ILOCK(mp); 3480 mp->mnt_noasync--; 3481 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 3482 mp->mnt_kern_flag |= MNTK_ASYNC; 3483 MNT_IUNLOCK(mp); 3484 vn_finished_write(mp); 3485 vfs_unbusy(mp); 3486 return (error); 3487 } 3488 3489 /* 3490 * The syncer vnode is no referenced. 3491 */ 3492 static int 3493 sync_inactive(struct vop_inactive_args *ap) 3494 { 3495 3496 vgone(ap->a_vp); 3497 return (0); 3498 } 3499 3500 /* 3501 * The syncer vnode is no longer needed and is being decommissioned. 3502 * 3503 * Modifications to the worklist must be protected by sync_mtx. 3504 */ 3505 static int 3506 sync_reclaim(struct vop_reclaim_args *ap) 3507 { 3508 struct vnode *vp = ap->a_vp; 3509 struct bufobj *bo; 3510 3511 bo = &vp->v_bufobj; 3512 BO_LOCK(bo); 3513 mtx_lock(&sync_mtx); 3514 if (vp->v_mount->mnt_syncer == vp) 3515 vp->v_mount->mnt_syncer = NULL; 3516 if (bo->bo_flag & BO_ONWORKLST) { 3517 LIST_REMOVE(bo, bo_synclist); 3518 syncer_worklist_len--; 3519 sync_vnode_count--; 3520 bo->bo_flag &= ~BO_ONWORKLST; 3521 } 3522 mtx_unlock(&sync_mtx); 3523 BO_UNLOCK(bo); 3524 3525 return (0); 3526 } 3527 3528 /* 3529 * Check if vnode represents a disk device 3530 */ 3531 int 3532 vn_isdisk(struct vnode *vp, int *errp) 3533 { 3534 int error; 3535 3536 error = 0; 3537 dev_lock(); 3538 if (vp->v_type != VCHR) 3539 error = ENOTBLK; 3540 else if (vp->v_rdev == NULL) 3541 error = ENXIO; 3542 else if (vp->v_rdev->si_devsw == NULL) 3543 error = ENXIO; 3544 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) 3545 error = ENOTBLK; 3546 dev_unlock(); 3547 if (errp != NULL) 3548 *errp = error; 3549 return (error == 0); 3550 } 3551 3552 /* 3553 * Common filesystem object access control check routine. Accepts a 3554 * vnode's type, "mode", uid and gid, requested access mode, credentials, 3555 * and optional call-by-reference privused argument allowing vaccess() 3556 * to indicate to the caller whether privilege was used to satisfy the 3557 * request (obsoleted). Returns 0 on success, or an errno on failure. 3558 * 3559 * The ifdef'd CAPABILITIES version is here for reference, but is not 3560 * actually used. 3561 */ 3562 int 3563 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, 3564 accmode_t accmode, struct ucred *cred, int *privused) 3565 { 3566 accmode_t dac_granted; 3567 accmode_t priv_granted; 3568 3569 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0, 3570 ("invalid bit in accmode")); 3571 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE), 3572 ("VAPPEND without VWRITE")); 3573 3574 /* 3575 * Look for a normal, non-privileged way to access the file/directory 3576 * as requested. If it exists, go with that. 3577 */ 3578 3579 if (privused != NULL) 3580 *privused = 0; 3581 3582 dac_granted = 0; 3583 3584 /* Check the owner. */ 3585 if (cred->cr_uid == file_uid) { 3586 dac_granted |= VADMIN; 3587 if (file_mode & S_IXUSR) 3588 dac_granted |= VEXEC; 3589 if (file_mode & S_IRUSR) 3590 dac_granted |= VREAD; 3591 if (file_mode & S_IWUSR) 3592 dac_granted |= (VWRITE | VAPPEND); 3593 3594 if ((accmode & dac_granted) == accmode) 3595 return (0); 3596 3597 goto privcheck; 3598 } 3599 3600 /* Otherwise, check the groups (first match) */ 3601 if (groupmember(file_gid, cred)) { 3602 if (file_mode & S_IXGRP) 3603 dac_granted |= VEXEC; 3604 if (file_mode & S_IRGRP) 3605 dac_granted |= VREAD; 3606 if (file_mode & S_IWGRP) 3607 dac_granted |= (VWRITE | VAPPEND); 3608 3609 if ((accmode & dac_granted) == accmode) 3610 return (0); 3611 3612 goto privcheck; 3613 } 3614 3615 /* Otherwise, check everyone else. */ 3616 if (file_mode & S_IXOTH) 3617 dac_granted |= VEXEC; 3618 if (file_mode & S_IROTH) 3619 dac_granted |= VREAD; 3620 if (file_mode & S_IWOTH) 3621 dac_granted |= (VWRITE | VAPPEND); 3622 if ((accmode & dac_granted) == accmode) 3623 return (0); 3624 3625 privcheck: 3626 /* 3627 * Build a privilege mask to determine if the set of privileges 3628 * satisfies the requirements when combined with the granted mask 3629 * from above. For each privilege, if the privilege is required, 3630 * bitwise or the request type onto the priv_granted mask. 3631 */ 3632 priv_granted = 0; 3633 3634 if (type == VDIR) { 3635 /* 3636 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC 3637 * requests, instead of PRIV_VFS_EXEC. 3638 */ 3639 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3640 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0)) 3641 priv_granted |= VEXEC; 3642 } else { 3643 /* 3644 * Ensure that at least one execute bit is on. Otherwise, 3645 * a privileged user will always succeed, and we don't want 3646 * this to happen unless the file really is executable. 3647 */ 3648 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3649 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 && 3650 !priv_check_cred(cred, PRIV_VFS_EXEC, 0)) 3651 priv_granted |= VEXEC; 3652 } 3653 3654 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) && 3655 !priv_check_cred(cred, PRIV_VFS_READ, 0)) 3656 priv_granted |= VREAD; 3657 3658 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) && 3659 !priv_check_cred(cred, PRIV_VFS_WRITE, 0)) 3660 priv_granted |= (VWRITE | VAPPEND); 3661 3662 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) && 3663 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0)) 3664 priv_granted |= VADMIN; 3665 3666 if ((accmode & (priv_granted | dac_granted)) == accmode) { 3667 /* XXX audit: privilege used */ 3668 if (privused != NULL) 3669 *privused = 1; 3670 return (0); 3671 } 3672 3673 return ((accmode & VADMIN) ? EPERM : EACCES); 3674 } 3675 3676 /* 3677 * Credential check based on process requesting service, and per-attribute 3678 * permissions. 3679 */ 3680 int 3681 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, 3682 struct thread *td, accmode_t accmode) 3683 { 3684 3685 /* 3686 * Kernel-invoked always succeeds. 3687 */ 3688 if (cred == NOCRED) 3689 return (0); 3690 3691 /* 3692 * Do not allow privileged processes in jail to directly manipulate 3693 * system attributes. 3694 */ 3695 switch (attrnamespace) { 3696 case EXTATTR_NAMESPACE_SYSTEM: 3697 /* Potentially should be: return (EPERM); */ 3698 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0)); 3699 case EXTATTR_NAMESPACE_USER: 3700 return (VOP_ACCESS(vp, accmode, cred, td)); 3701 default: 3702 return (EPERM); 3703 } 3704 } 3705 3706 #ifdef DEBUG_VFS_LOCKS 3707 /* 3708 * This only exists to supress warnings from unlocked specfs accesses. It is 3709 * no longer ok to have an unlocked VFS. 3710 */ 3711 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \ 3712 (vp)->v_type == VCHR || (vp)->v_type == VBAD) 3713 3714 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ 3715 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, ""); 3716 3717 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ 3718 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, ""); 3719 3720 int vfs_badlock_print = 1; /* Print lock violations. */ 3721 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, ""); 3722 3723 #ifdef KDB 3724 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ 3725 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, ""); 3726 #endif 3727 3728 static void 3729 vfs_badlock(const char *msg, const char *str, struct vnode *vp) 3730 { 3731 3732 #ifdef KDB 3733 if (vfs_badlock_backtrace) 3734 kdb_backtrace(); 3735 #endif 3736 if (vfs_badlock_print) 3737 printf("%s: %p %s\n", str, (void *)vp, msg); 3738 if (vfs_badlock_ddb) 3739 kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); 3740 } 3741 3742 void 3743 assert_vi_locked(struct vnode *vp, const char *str) 3744 { 3745 3746 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) 3747 vfs_badlock("interlock is not locked but should be", str, vp); 3748 } 3749 3750 void 3751 assert_vi_unlocked(struct vnode *vp, const char *str) 3752 { 3753 3754 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) 3755 vfs_badlock("interlock is locked but should not be", str, vp); 3756 } 3757 3758 void 3759 assert_vop_locked(struct vnode *vp, const char *str) 3760 { 3761 3762 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0) 3763 vfs_badlock("is not locked but should be", str, vp); 3764 } 3765 3766 void 3767 assert_vop_unlocked(struct vnode *vp, const char *str) 3768 { 3769 3770 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) 3771 vfs_badlock("is locked but should not be", str, vp); 3772 } 3773 3774 void 3775 assert_vop_elocked(struct vnode *vp, const char *str) 3776 { 3777 3778 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 3779 vfs_badlock("is not exclusive locked but should be", str, vp); 3780 } 3781 3782 #if 0 3783 void 3784 assert_vop_elocked_other(struct vnode *vp, const char *str) 3785 { 3786 3787 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER) 3788 vfs_badlock("is not exclusive locked by another thread", 3789 str, vp); 3790 } 3791 3792 void 3793 assert_vop_slocked(struct vnode *vp, const char *str) 3794 { 3795 3796 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED) 3797 vfs_badlock("is not locked shared but should be", str, vp); 3798 } 3799 #endif /* 0 */ 3800 #endif /* DEBUG_VFS_LOCKS */ 3801 3802 void 3803 vop_rename_fail(struct vop_rename_args *ap) 3804 { 3805 3806 if (ap->a_tvp != NULL) 3807 vput(ap->a_tvp); 3808 if (ap->a_tdvp == ap->a_tvp) 3809 vrele(ap->a_tdvp); 3810 else 3811 vput(ap->a_tdvp); 3812 vrele(ap->a_fdvp); 3813 vrele(ap->a_fvp); 3814 } 3815 3816 void 3817 vop_rename_pre(void *ap) 3818 { 3819 struct vop_rename_args *a = ap; 3820 3821 #ifdef DEBUG_VFS_LOCKS 3822 if (a->a_tvp) 3823 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); 3824 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); 3825 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); 3826 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); 3827 3828 /* Check the source (from). */ 3829 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock && 3830 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock)) 3831 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); 3832 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock) 3833 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); 3834 3835 /* Check the target. */ 3836 if (a->a_tvp) 3837 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); 3838 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); 3839 #endif 3840 if (a->a_tdvp != a->a_fdvp) 3841 vhold(a->a_fdvp); 3842 if (a->a_tvp != a->a_fvp) 3843 vhold(a->a_fvp); 3844 vhold(a->a_tdvp); 3845 if (a->a_tvp) 3846 vhold(a->a_tvp); 3847 } 3848 3849 void 3850 vop_strategy_pre(void *ap) 3851 { 3852 #ifdef DEBUG_VFS_LOCKS 3853 struct vop_strategy_args *a; 3854 struct buf *bp; 3855 3856 a = ap; 3857 bp = a->a_bp; 3858 3859 /* 3860 * Cluster ops lock their component buffers but not the IO container. 3861 */ 3862 if ((bp->b_flags & B_CLUSTER) != 0) 3863 return; 3864 3865 if (!BUF_ISLOCKED(bp)) { 3866 if (vfs_badlock_print) 3867 printf( 3868 "VOP_STRATEGY: bp is not locked but should be\n"); 3869 if (vfs_badlock_ddb) 3870 kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); 3871 } 3872 #endif 3873 } 3874 3875 void 3876 vop_lookup_pre(void *ap) 3877 { 3878 #ifdef DEBUG_VFS_LOCKS 3879 struct vop_lookup_args *a; 3880 struct vnode *dvp; 3881 3882 a = ap; 3883 dvp = a->a_dvp; 3884 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 3885 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 3886 #endif 3887 } 3888 3889 void 3890 vop_lookup_post(void *ap, int rc) 3891 { 3892 #ifdef DEBUG_VFS_LOCKS 3893 struct vop_lookup_args *a; 3894 struct vnode *dvp; 3895 struct vnode *vp; 3896 3897 a = ap; 3898 dvp = a->a_dvp; 3899 vp = *(a->a_vpp); 3900 3901 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 3902 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 3903 3904 if (!rc) 3905 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)"); 3906 #endif 3907 } 3908 3909 void 3910 vop_lock_pre(void *ap) 3911 { 3912 #ifdef DEBUG_VFS_LOCKS 3913 struct vop_lock1_args *a = ap; 3914 3915 if ((a->a_flags & LK_INTERLOCK) == 0) 3916 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 3917 else 3918 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); 3919 #endif 3920 } 3921 3922 void 3923 vop_lock_post(void *ap, int rc) 3924 { 3925 #ifdef DEBUG_VFS_LOCKS 3926 struct vop_lock1_args *a = ap; 3927 3928 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 3929 if (rc == 0) 3930 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); 3931 #endif 3932 } 3933 3934 void 3935 vop_unlock_pre(void *ap) 3936 { 3937 #ifdef DEBUG_VFS_LOCKS 3938 struct vop_unlock_args *a = ap; 3939 3940 if (a->a_flags & LK_INTERLOCK) 3941 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK"); 3942 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); 3943 #endif 3944 } 3945 3946 void 3947 vop_unlock_post(void *ap, int rc) 3948 { 3949 #ifdef DEBUG_VFS_LOCKS 3950 struct vop_unlock_args *a = ap; 3951 3952 if (a->a_flags & LK_INTERLOCK) 3953 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK"); 3954 #endif 3955 } 3956 3957 void 3958 vop_create_post(void *ap, int rc) 3959 { 3960 struct vop_create_args *a = ap; 3961 3962 if (!rc) 3963 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3964 } 3965 3966 void 3967 vop_link_post(void *ap, int rc) 3968 { 3969 struct vop_link_args *a = ap; 3970 3971 if (!rc) { 3972 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK); 3973 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE); 3974 } 3975 } 3976 3977 void 3978 vop_mkdir_post(void *ap, int rc) 3979 { 3980 struct vop_mkdir_args *a = ap; 3981 3982 if (!rc) 3983 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 3984 } 3985 3986 void 3987 vop_mknod_post(void *ap, int rc) 3988 { 3989 struct vop_mknod_args *a = ap; 3990 3991 if (!rc) 3992 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3993 } 3994 3995 void 3996 vop_remove_post(void *ap, int rc) 3997 { 3998 struct vop_remove_args *a = ap; 3999 4000 if (!rc) { 4001 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4002 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 4003 } 4004 } 4005 4006 void 4007 vop_rename_post(void *ap, int rc) 4008 { 4009 struct vop_rename_args *a = ap; 4010 4011 if (!rc) { 4012 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE); 4013 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE); 4014 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); 4015 if (a->a_tvp) 4016 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); 4017 } 4018 if (a->a_tdvp != a->a_fdvp) 4019 vdrop(a->a_fdvp); 4020 if (a->a_tvp != a->a_fvp) 4021 vdrop(a->a_fvp); 4022 vdrop(a->a_tdvp); 4023 if (a->a_tvp) 4024 vdrop(a->a_tvp); 4025 } 4026 4027 void 4028 vop_rmdir_post(void *ap, int rc) 4029 { 4030 struct vop_rmdir_args *a = ap; 4031 4032 if (!rc) { 4033 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 4034 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 4035 } 4036 } 4037 4038 void 4039 vop_setattr_post(void *ap, int rc) 4040 { 4041 struct vop_setattr_args *a = ap; 4042 4043 if (!rc) 4044 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); 4045 } 4046 4047 void 4048 vop_symlink_post(void *ap, int rc) 4049 { 4050 struct vop_symlink_args *a = ap; 4051 4052 if (!rc) 4053 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4054 } 4055 4056 static struct knlist fs_knlist; 4057 4058 static void 4059 vfs_event_init(void *arg) 4060 { 4061 knlist_init_mtx(&fs_knlist, NULL); 4062 } 4063 /* XXX - correct order? */ 4064 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); 4065 4066 void 4067 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused) 4068 { 4069 4070 KNOTE_UNLOCKED(&fs_knlist, event); 4071 } 4072 4073 static int filt_fsattach(struct knote *kn); 4074 static void filt_fsdetach(struct knote *kn); 4075 static int filt_fsevent(struct knote *kn, long hint); 4076 4077 struct filterops fs_filtops = { 4078 .f_isfd = 0, 4079 .f_attach = filt_fsattach, 4080 .f_detach = filt_fsdetach, 4081 .f_event = filt_fsevent 4082 }; 4083 4084 static int 4085 filt_fsattach(struct knote *kn) 4086 { 4087 4088 kn->kn_flags |= EV_CLEAR; 4089 knlist_add(&fs_knlist, kn, 0); 4090 return (0); 4091 } 4092 4093 static void 4094 filt_fsdetach(struct knote *kn) 4095 { 4096 4097 knlist_remove(&fs_knlist, kn, 0); 4098 } 4099 4100 static int 4101 filt_fsevent(struct knote *kn, long hint) 4102 { 4103 4104 kn->kn_fflags |= hint; 4105 return (kn->kn_fflags != 0); 4106 } 4107 4108 static int 4109 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) 4110 { 4111 struct vfsidctl vc; 4112 int error; 4113 struct mount *mp; 4114 4115 error = SYSCTL_IN(req, &vc, sizeof(vc)); 4116 if (error) 4117 return (error); 4118 if (vc.vc_vers != VFS_CTL_VERS1) 4119 return (EINVAL); 4120 mp = vfs_getvfs(&vc.vc_fsid); 4121 if (mp == NULL) 4122 return (ENOENT); 4123 /* ensure that a specific sysctl goes to the right filesystem. */ 4124 if (strcmp(vc.vc_fstypename, "*") != 0 && 4125 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { 4126 vfs_rel(mp); 4127 return (EINVAL); 4128 } 4129 VCTLTOREQ(&vc, req); 4130 error = VFS_SYSCTL(mp, vc.vc_op, req); 4131 vfs_rel(mp); 4132 return (error); 4133 } 4134 4135 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "", 4136 "Sysctl by fsid"); 4137 4138 /* 4139 * Function to initialize a va_filerev field sensibly. 4140 * XXX: Wouldn't a random number make a lot more sense ?? 4141 */ 4142 u_quad_t 4143 init_va_filerev(void) 4144 { 4145 struct bintime bt; 4146 4147 getbinuptime(&bt); 4148 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); 4149 } 4150 4151 static int filt_vfsread(struct knote *kn, long hint); 4152 static int filt_vfswrite(struct knote *kn, long hint); 4153 static int filt_vfsvnode(struct knote *kn, long hint); 4154 static void filt_vfsdetach(struct knote *kn); 4155 static struct filterops vfsread_filtops = { 4156 .f_isfd = 1, 4157 .f_detach = filt_vfsdetach, 4158 .f_event = filt_vfsread 4159 }; 4160 static struct filterops vfswrite_filtops = { 4161 .f_isfd = 1, 4162 .f_detach = filt_vfsdetach, 4163 .f_event = filt_vfswrite 4164 }; 4165 static struct filterops vfsvnode_filtops = { 4166 .f_isfd = 1, 4167 .f_detach = filt_vfsdetach, 4168 .f_event = filt_vfsvnode 4169 }; 4170 4171 static void 4172 vfs_knllock(void *arg) 4173 { 4174 struct vnode *vp = arg; 4175 4176 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4177 } 4178 4179 static void 4180 vfs_knlunlock(void *arg) 4181 { 4182 struct vnode *vp = arg; 4183 4184 VOP_UNLOCK(vp, 0); 4185 } 4186 4187 static void 4188 vfs_knl_assert_locked(void *arg) 4189 { 4190 #ifdef DEBUG_VFS_LOCKS 4191 struct vnode *vp = arg; 4192 4193 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked"); 4194 #endif 4195 } 4196 4197 static void 4198 vfs_knl_assert_unlocked(void *arg) 4199 { 4200 #ifdef DEBUG_VFS_LOCKS 4201 struct vnode *vp = arg; 4202 4203 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked"); 4204 #endif 4205 } 4206 4207 int 4208 vfs_kqfilter(struct vop_kqfilter_args *ap) 4209 { 4210 struct vnode *vp = ap->a_vp; 4211 struct knote *kn = ap->a_kn; 4212 struct knlist *knl; 4213 4214 switch (kn->kn_filter) { 4215 case EVFILT_READ: 4216 kn->kn_fop = &vfsread_filtops; 4217 break; 4218 case EVFILT_WRITE: 4219 kn->kn_fop = &vfswrite_filtops; 4220 break; 4221 case EVFILT_VNODE: 4222 kn->kn_fop = &vfsvnode_filtops; 4223 break; 4224 default: 4225 return (EINVAL); 4226 } 4227 4228 kn->kn_hook = (caddr_t)vp; 4229 4230 v_addpollinfo(vp); 4231 if (vp->v_pollinfo == NULL) 4232 return (ENOMEM); 4233 knl = &vp->v_pollinfo->vpi_selinfo.si_note; 4234 knlist_add(knl, kn, 0); 4235 4236 return (0); 4237 } 4238 4239 /* 4240 * Detach knote from vnode 4241 */ 4242 static void 4243 filt_vfsdetach(struct knote *kn) 4244 { 4245 struct vnode *vp = (struct vnode *)kn->kn_hook; 4246 4247 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); 4248 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); 4249 } 4250 4251 /*ARGSUSED*/ 4252 static int 4253 filt_vfsread(struct knote *kn, long hint) 4254 { 4255 struct vnode *vp = (struct vnode *)kn->kn_hook; 4256 struct vattr va; 4257 int res; 4258 4259 /* 4260 * filesystem is gone, so set the EOF flag and schedule 4261 * the knote for deletion. 4262 */ 4263 if (hint == NOTE_REVOKE) { 4264 VI_LOCK(vp); 4265 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 4266 VI_UNLOCK(vp); 4267 return (1); 4268 } 4269 4270 if (VOP_GETATTR(vp, &va, curthread->td_ucred)) 4271 return (0); 4272 4273 VI_LOCK(vp); 4274 kn->kn_data = va.va_size - kn->kn_fp->f_offset; 4275 res = (kn->kn_data != 0); 4276 VI_UNLOCK(vp); 4277 return (res); 4278 } 4279 4280 /*ARGSUSED*/ 4281 static int 4282 filt_vfswrite(struct knote *kn, long hint) 4283 { 4284 struct vnode *vp = (struct vnode *)kn->kn_hook; 4285 4286 VI_LOCK(vp); 4287 4288 /* 4289 * filesystem is gone, so set the EOF flag and schedule 4290 * the knote for deletion. 4291 */ 4292 if (hint == NOTE_REVOKE) 4293 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 4294 4295 kn->kn_data = 0; 4296 VI_UNLOCK(vp); 4297 return (1); 4298 } 4299 4300 static int 4301 filt_vfsvnode(struct knote *kn, long hint) 4302 { 4303 struct vnode *vp = (struct vnode *)kn->kn_hook; 4304 int res; 4305 4306 VI_LOCK(vp); 4307 if (kn->kn_sfflags & hint) 4308 kn->kn_fflags |= hint; 4309 if (hint == NOTE_REVOKE) { 4310 kn->kn_flags |= EV_EOF; 4311 VI_UNLOCK(vp); 4312 return (1); 4313 } 4314 res = (kn->kn_fflags != 0); 4315 VI_UNLOCK(vp); 4316 return (res); 4317 } 4318 4319 int 4320 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) 4321 { 4322 int error; 4323 4324 if (dp->d_reclen > ap->a_uio->uio_resid) 4325 return (ENAMETOOLONG); 4326 error = uiomove(dp, dp->d_reclen, ap->a_uio); 4327 if (error) { 4328 if (ap->a_ncookies != NULL) { 4329 if (ap->a_cookies != NULL) 4330 free(ap->a_cookies, M_TEMP); 4331 ap->a_cookies = NULL; 4332 *ap->a_ncookies = 0; 4333 } 4334 return (error); 4335 } 4336 if (ap->a_ncookies == NULL) 4337 return (0); 4338 4339 KASSERT(ap->a_cookies, 4340 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); 4341 4342 *ap->a_cookies = realloc(*ap->a_cookies, 4343 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); 4344 (*ap->a_cookies)[*ap->a_ncookies] = off; 4345 return (0); 4346 } 4347 4348 /* 4349 * Mark for update the access time of the file if the filesystem 4350 * supports VOP_MARKATIME. This functionality is used by execve and 4351 * mmap, so we want to avoid the I/O implied by directly setting 4352 * va_atime for the sake of efficiency. 4353 */ 4354 void 4355 vfs_mark_atime(struct vnode *vp, struct ucred *cred) 4356 { 4357 struct mount *mp; 4358 4359 mp = vp->v_mount; 4360 VFS_ASSERT_GIANT(mp); 4361 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime"); 4362 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) 4363 (void)VOP_MARKATIME(vp); 4364 } 4365 4366 /* 4367 * The purpose of this routine is to remove granularity from accmode_t, 4368 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE, 4369 * VADMIN and VAPPEND. 4370 * 4371 * If it returns 0, the caller is supposed to continue with the usual 4372 * access checks using 'accmode' as modified by this routine. If it 4373 * returns nonzero value, the caller is supposed to return that value 4374 * as errno. 4375 * 4376 * Note that after this routine runs, accmode may be zero. 4377 */ 4378 int 4379 vfs_unixify_accmode(accmode_t *accmode) 4380 { 4381 /* 4382 * There is no way to specify explicit "deny" rule using 4383 * file mode or POSIX.1e ACLs. 4384 */ 4385 if (*accmode & VEXPLICIT_DENY) { 4386 *accmode = 0; 4387 return (0); 4388 } 4389 4390 /* 4391 * None of these can be translated into usual access bits. 4392 * Also, the common case for NFSv4 ACLs is to not contain 4393 * either of these bits. Caller should check for VWRITE 4394 * on the containing directory instead. 4395 */ 4396 if (*accmode & (VDELETE_CHILD | VDELETE)) 4397 return (EPERM); 4398 4399 if (*accmode & VADMIN_PERMS) { 4400 *accmode &= ~VADMIN_PERMS; 4401 *accmode |= VADMIN; 4402 } 4403 4404 /* 4405 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL 4406 * or VSYNCHRONIZE using file mode or POSIX.1e ACL. 4407 */ 4408 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE); 4409 4410 return (0); 4411 } 4412