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/sched.h> 69 #include <sys/sleepqueue.h> 70 #include <sys/stat.h> 71 #include <sys/sysctl.h> 72 #include <sys/syslog.h> 73 #include <sys/vmmeter.h> 74 #include <sys/vnode.h> 75 76 #include <machine/stdarg.h> 77 78 #include <security/mac/mac_framework.h> 79 80 #include <vm/vm.h> 81 #include <vm/vm_object.h> 82 #include <vm/vm_extern.h> 83 #include <vm/pmap.h> 84 #include <vm/vm_map.h> 85 #include <vm/vm_page.h> 86 #include <vm/vm_kern.h> 87 #include <vm/uma.h> 88 89 #ifdef DDB 90 #include <ddb/ddb.h> 91 #endif 92 93 #define WI_MPSAFEQ 0 94 #define WI_GIANTQ 1 95 96 static void delmntque(struct vnode *vp); 97 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, 98 int slpflag, int slptimeo); 99 static void syncer_shutdown(void *arg, int howto); 100 static int vtryrecycle(struct vnode *vp); 101 static void vbusy(struct vnode *vp); 102 static void vinactive(struct vnode *, struct thread *); 103 static void v_incr_usecount(struct vnode *); 104 static void v_decr_usecount(struct vnode *); 105 static void v_decr_useonly(struct vnode *); 106 static void v_upgrade_usecount(struct vnode *); 107 static void vfree(struct vnode *); 108 static void vnlru_free(int); 109 static void vgonel(struct vnode *); 110 static void vfs_knllock(void *arg); 111 static void vfs_knlunlock(void *arg); 112 static void vfs_knl_assert_locked(void *arg); 113 static void vfs_knl_assert_unlocked(void *arg); 114 static void destroy_vpollinfo(struct vpollinfo *vi); 115 116 /* 117 * Number of vnodes in existence. Increased whenever getnewvnode() 118 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed 119 * vnode. 120 */ 121 static unsigned long numvnodes; 122 123 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, 124 "Number of vnodes in existence"); 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_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 151 /* Number of vnodes in the free list. */ 152 static u_long freevnodes; 153 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, 154 "Number of vnodes in the free list"); 155 156 static int vlru_allow_cache_src; 157 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW, 158 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode"); 159 160 /* 161 * Various variables used for debugging the new implementation of 162 * reassignbuf(). 163 * XXX these are probably of (very) limited utility now. 164 */ 165 static int reassignbufcalls; 166 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, 167 "Number of calls to reassignbuf"); 168 169 /* 170 * Cache for the mount type id assigned to NFS. This is used for 171 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c. 172 */ 173 int nfs_mount_type = -1; 174 175 /* To keep more than one thread at a time from running vfs_getnewfsid */ 176 static struct mtx mntid_mtx; 177 178 /* 179 * Lock for any access to the following: 180 * vnode_free_list 181 * numvnodes 182 * freevnodes 183 */ 184 static struct mtx vnode_free_list_mtx; 185 186 /* Publicly exported FS */ 187 struct nfs_public nfs_pub; 188 189 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ 190 static uma_zone_t vnode_zone; 191 static uma_zone_t vnodepoll_zone; 192 193 /* 194 * The workitem queue. 195 * 196 * It is useful to delay writes of file data and filesystem metadata 197 * for tens of seconds so that quickly created and deleted files need 198 * not waste disk bandwidth being created and removed. To realize this, 199 * we append vnodes to a "workitem" queue. When running with a soft 200 * updates implementation, most pending metadata dependencies should 201 * not wait for more than a few seconds. Thus, mounted on block devices 202 * are delayed only about a half the time that file data is delayed. 203 * Similarly, directory updates are more critical, so are only delayed 204 * about a third the time that file data is delayed. Thus, there are 205 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 206 * one each second (driven off the filesystem syncer process). The 207 * syncer_delayno variable indicates the next queue that is to be processed. 208 * Items that need to be processed soon are placed in this queue: 209 * 210 * syncer_workitem_pending[syncer_delayno] 211 * 212 * A delay of fifteen seconds is done by placing the request fifteen 213 * entries later in the queue: 214 * 215 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 216 * 217 */ 218 static int syncer_delayno; 219 static long syncer_mask; 220 LIST_HEAD(synclist, bufobj); 221 static struct synclist *syncer_workitem_pending[2]; 222 /* 223 * The sync_mtx protects: 224 * bo->bo_synclist 225 * sync_vnode_count 226 * syncer_delayno 227 * syncer_state 228 * syncer_workitem_pending 229 * syncer_worklist_len 230 * rushjob 231 */ 232 static struct mtx sync_mtx; 233 static struct cv sync_wakeup; 234 235 #define SYNCER_MAXDELAY 32 236 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 237 static int syncdelay = 30; /* max time to delay syncing data */ 238 static int filedelay = 30; /* time to delay syncing files */ 239 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, 240 "Time to delay syncing files (in seconds)"); 241 static int dirdelay = 29; /* time to delay syncing directories */ 242 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, 243 "Time to delay syncing directories (in seconds)"); 244 static int metadelay = 28; /* time to delay syncing metadata */ 245 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, 246 "Time to delay syncing metadata (in seconds)"); 247 static int rushjob; /* number of slots to run ASAP */ 248 static int stat_rush_requests; /* number of times I/O speeded up */ 249 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, 250 "Number of times I/O speeded up (rush requests)"); 251 252 /* 253 * When shutting down the syncer, run it at four times normal speed. 254 */ 255 #define SYNCER_SHUTDOWN_SPEEDUP 4 256 static int sync_vnode_count; 257 static int syncer_worklist_len; 258 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } 259 syncer_state; 260 261 /* 262 * Number of vnodes we want to exist at any one time. This is mostly used 263 * to size hash tables in vnode-related code. It is normally not used in 264 * getnewvnode(), as wantfreevnodes is normally nonzero.) 265 * 266 * XXX desiredvnodes is historical cruft and should not exist. 267 */ 268 int desiredvnodes; 269 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 270 &desiredvnodes, 0, "Maximum number of vnodes"); 271 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, 272 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)"); 273 static int vnlru_nowhere; 274 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, 275 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); 276 277 /* 278 * Macros to control when a vnode is freed and recycled. All require 279 * the vnode interlock. 280 */ 281 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt) 282 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt) 283 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt) 284 285 286 /* 287 * Initialize the vnode management data structures. 288 * 289 * Reevaluate the following cap on the number of vnodes after the physical 290 * memory size exceeds 512GB. In the limit, as the physical memory size 291 * grows, the ratio of physical pages to vnodes approaches sixteen to one. 292 */ 293 #ifndef MAXVNODES_MAX 294 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16)) 295 #endif 296 static void 297 vntblinit(void *dummy __unused) 298 { 299 int physvnodes, virtvnodes; 300 301 /* 302 * Desiredvnodes is a function of the physical memory size and the 303 * kernel's heap size. Generally speaking, it scales with the 304 * physical memory size. The ratio of desiredvnodes to physical pages 305 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the 306 * marginal ratio of desiredvnodes to physical pages is one to 307 * sixteen. However, desiredvnodes is limited by the kernel's heap 308 * size. The memory required by desiredvnodes vnodes and vm objects 309 * may not exceed one seventh of the kernel's heap size. 310 */ 311 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4, 312 cnt.v_page_count) / 16; 313 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) + 314 sizeof(struct vnode))); 315 desiredvnodes = min(physvnodes, virtvnodes); 316 if (desiredvnodes > MAXVNODES_MAX) { 317 if (bootverbose) 318 printf("Reducing kern.maxvnodes %d -> %d\n", 319 desiredvnodes, MAXVNODES_MAX); 320 desiredvnodes = MAXVNODES_MAX; 321 } 322 wantfreevnodes = desiredvnodes / 4; 323 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); 324 TAILQ_INIT(&vnode_free_list); 325 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF); 326 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, 327 NULL, NULL, UMA_ALIGN_PTR, 0); 328 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), 329 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 330 /* 331 * Initialize the filesystem syncer. 332 */ 333 syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE, 334 &syncer_mask); 335 syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE, 336 &syncer_mask); 337 syncer_maxdelay = syncer_mask + 1; 338 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); 339 cv_init(&sync_wakeup, "syncer"); 340 } 341 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL); 342 343 344 /* 345 * Mark a mount point as busy. Used to synchronize access and to delay 346 * unmounting. Eventually, mountlist_mtx is not released on failure. 347 */ 348 int 349 vfs_busy(struct mount *mp, int flags) 350 { 351 352 MPASS((flags & ~MBF_MASK) == 0); 353 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags); 354 355 MNT_ILOCK(mp); 356 MNT_REF(mp); 357 /* 358 * If mount point is currenly being unmounted, sleep until the 359 * mount point fate is decided. If thread doing the unmounting fails, 360 * it will clear MNTK_UNMOUNT flag before waking us up, indicating 361 * that this mount point has survived the unmount attempt and vfs_busy 362 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE 363 * flag in addition to MNTK_UNMOUNT, indicating that mount point is 364 * about to be really destroyed. vfs_busy needs to release its 365 * reference on the mount point in this case and return with ENOENT, 366 * telling the caller that mount mount it tried to busy is no longer 367 * valid. 368 */ 369 while (mp->mnt_kern_flag & MNTK_UNMOUNT) { 370 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) { 371 MNT_REL(mp); 372 MNT_IUNLOCK(mp); 373 CTR1(KTR_VFS, "%s: failed busying before sleeping", 374 __func__); 375 return (ENOENT); 376 } 377 if (flags & MBF_MNTLSTLOCK) 378 mtx_unlock(&mountlist_mtx); 379 mp->mnt_kern_flag |= MNTK_MWAIT; 380 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0); 381 if (flags & MBF_MNTLSTLOCK) 382 mtx_lock(&mountlist_mtx); 383 } 384 if (flags & MBF_MNTLSTLOCK) 385 mtx_unlock(&mountlist_mtx); 386 mp->mnt_lockref++; 387 MNT_IUNLOCK(mp); 388 return (0); 389 } 390 391 /* 392 * Free a busy filesystem. 393 */ 394 void 395 vfs_unbusy(struct mount *mp) 396 { 397 398 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 399 MNT_ILOCK(mp); 400 MNT_REL(mp); 401 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref")); 402 mp->mnt_lockref--; 403 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) { 404 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT); 405 CTR1(KTR_VFS, "%s: waking up waiters", __func__); 406 mp->mnt_kern_flag &= ~MNTK_DRAINING; 407 wakeup(&mp->mnt_lockref); 408 } 409 MNT_IUNLOCK(mp); 410 } 411 412 /* 413 * Lookup a mount point by filesystem identifier. 414 */ 415 struct mount * 416 vfs_getvfs(fsid_t *fsid) 417 { 418 struct mount *mp; 419 420 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); 421 mtx_lock(&mountlist_mtx); 422 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 423 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 424 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 425 vfs_ref(mp); 426 mtx_unlock(&mountlist_mtx); 427 return (mp); 428 } 429 } 430 mtx_unlock(&mountlist_mtx); 431 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); 432 return ((struct mount *) 0); 433 } 434 435 /* 436 * Lookup a mount point by filesystem identifier, busying it before 437 * returning. 438 */ 439 struct mount * 440 vfs_busyfs(fsid_t *fsid) 441 { 442 struct mount *mp; 443 int error; 444 445 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); 446 mtx_lock(&mountlist_mtx); 447 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 448 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 449 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 450 error = vfs_busy(mp, MBF_MNTLSTLOCK); 451 if (error) { 452 mtx_unlock(&mountlist_mtx); 453 return (NULL); 454 } 455 return (mp); 456 } 457 } 458 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); 459 mtx_unlock(&mountlist_mtx); 460 return ((struct mount *) 0); 461 } 462 463 /* 464 * Check if a user can access privileged mount options. 465 */ 466 int 467 vfs_suser(struct mount *mp, struct thread *td) 468 { 469 int error; 470 471 /* 472 * If the thread is jailed, but this is not a jail-friendly file 473 * system, deny immediately. 474 */ 475 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred)) 476 return (EPERM); 477 478 /* 479 * If the file system was mounted outside the jail of the calling 480 * thread, deny immediately. 481 */ 482 if (prison_check(td->td_ucred, mp->mnt_cred) != 0) 483 return (EPERM); 484 485 /* 486 * If file system supports delegated administration, we don't check 487 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified 488 * by the file system itself. 489 * If this is not the user that did original mount, we check for 490 * the PRIV_VFS_MOUNT_OWNER privilege. 491 */ 492 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) && 493 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { 494 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) 495 return (error); 496 } 497 return (0); 498 } 499 500 /* 501 * Get a new unique fsid. Try to make its val[0] unique, since this value 502 * will be used to create fake device numbers for stat(). Also try (but 503 * not so hard) make its val[0] unique mod 2^16, since some emulators only 504 * support 16-bit device numbers. We end up with unique val[0]'s for the 505 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 506 * 507 * Keep in mind that several mounts may be running in parallel. Starting 508 * the search one past where the previous search terminated is both a 509 * micro-optimization and a defense against returning the same fsid to 510 * different mounts. 511 */ 512 void 513 vfs_getnewfsid(struct mount *mp) 514 { 515 static uint16_t mntid_base; 516 struct mount *nmp; 517 fsid_t tfsid; 518 int mtype; 519 520 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 521 mtx_lock(&mntid_mtx); 522 mtype = mp->mnt_vfc->vfc_typenum; 523 tfsid.val[1] = mtype; 524 mtype = (mtype & 0xFF) << 24; 525 for (;;) { 526 tfsid.val[0] = makedev(255, 527 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 528 mntid_base++; 529 if ((nmp = vfs_getvfs(&tfsid)) == NULL) 530 break; 531 vfs_rel(nmp); 532 } 533 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 534 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 535 mtx_unlock(&mntid_mtx); 536 } 537 538 /* 539 * Knob to control the precision of file timestamps: 540 * 541 * 0 = seconds only; nanoseconds zeroed. 542 * 1 = seconds and nanoseconds, accurate within 1/HZ. 543 * 2 = seconds and nanoseconds, truncated to microseconds. 544 * >=3 = seconds and nanoseconds, maximum precision. 545 */ 546 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 547 548 static int timestamp_precision = TSP_SEC; 549 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 550 ×tamp_precision, 0, "File timestamp precision (0: seconds, " 551 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, " 552 "3+: sec + ns (max. precision))"); 553 554 /* 555 * Get a current timestamp. 556 */ 557 void 558 vfs_timestamp(struct timespec *tsp) 559 { 560 struct timeval tv; 561 562 switch (timestamp_precision) { 563 case TSP_SEC: 564 tsp->tv_sec = time_second; 565 tsp->tv_nsec = 0; 566 break; 567 case TSP_HZ: 568 getnanotime(tsp); 569 break; 570 case TSP_USEC: 571 microtime(&tv); 572 TIMEVAL_TO_TIMESPEC(&tv, tsp); 573 break; 574 case TSP_NSEC: 575 default: 576 nanotime(tsp); 577 break; 578 } 579 } 580 581 /* 582 * Set vnode attributes to VNOVAL 583 */ 584 void 585 vattr_null(struct vattr *vap) 586 { 587 588 vap->va_type = VNON; 589 vap->va_size = VNOVAL; 590 vap->va_bytes = VNOVAL; 591 vap->va_mode = VNOVAL; 592 vap->va_nlink = VNOVAL; 593 vap->va_uid = VNOVAL; 594 vap->va_gid = VNOVAL; 595 vap->va_fsid = VNOVAL; 596 vap->va_fileid = VNOVAL; 597 vap->va_blocksize = VNOVAL; 598 vap->va_rdev = VNOVAL; 599 vap->va_atime.tv_sec = VNOVAL; 600 vap->va_atime.tv_nsec = VNOVAL; 601 vap->va_mtime.tv_sec = VNOVAL; 602 vap->va_mtime.tv_nsec = VNOVAL; 603 vap->va_ctime.tv_sec = VNOVAL; 604 vap->va_ctime.tv_nsec = VNOVAL; 605 vap->va_birthtime.tv_sec = VNOVAL; 606 vap->va_birthtime.tv_nsec = VNOVAL; 607 vap->va_flags = VNOVAL; 608 vap->va_gen = VNOVAL; 609 vap->va_vaflags = 0; 610 } 611 612 /* 613 * This routine is called when we have too many vnodes. It attempts 614 * to free <count> vnodes and will potentially free vnodes that still 615 * have VM backing store (VM backing store is typically the cause 616 * of a vnode blowout so we want to do this). Therefore, this operation 617 * is not considered cheap. 618 * 619 * A number of conditions may prevent a vnode from being reclaimed. 620 * the buffer cache may have references on the vnode, a directory 621 * vnode may still have references due to the namei cache representing 622 * underlying files, or the vnode may be in active use. It is not 623 * desireable to reuse such vnodes. These conditions may cause the 624 * number of vnodes to reach some minimum value regardless of what 625 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 626 */ 627 static int 628 vlrureclaim(struct mount *mp) 629 { 630 struct vnode *vp; 631 int done; 632 int trigger; 633 int usevnodes; 634 int count; 635 636 /* 637 * Calculate the trigger point, don't allow user 638 * screwups to blow us up. This prevents us from 639 * recycling vnodes with lots of resident pages. We 640 * aren't trying to free memory, we are trying to 641 * free vnodes. 642 */ 643 usevnodes = desiredvnodes; 644 if (usevnodes <= 0) 645 usevnodes = 1; 646 trigger = cnt.v_page_count * 2 / usevnodes; 647 done = 0; 648 vn_start_write(NULL, &mp, V_WAIT); 649 MNT_ILOCK(mp); 650 count = mp->mnt_nvnodelistsize / 10 + 1; 651 while (count != 0) { 652 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 653 while (vp != NULL && vp->v_type == VMARKER) 654 vp = TAILQ_NEXT(vp, v_nmntvnodes); 655 if (vp == NULL) 656 break; 657 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 658 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 659 --count; 660 if (!VI_TRYLOCK(vp)) 661 goto next_iter; 662 /* 663 * If it's been deconstructed already, it's still 664 * referenced, or it exceeds the trigger, skip it. 665 */ 666 if (vp->v_usecount || 667 (!vlru_allow_cache_src && 668 !LIST_EMPTY(&(vp)->v_cache_src)) || 669 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL && 670 vp->v_object->resident_page_count > trigger)) { 671 VI_UNLOCK(vp); 672 goto next_iter; 673 } 674 MNT_IUNLOCK(mp); 675 vholdl(vp); 676 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) { 677 vdrop(vp); 678 goto next_iter_mntunlocked; 679 } 680 VI_LOCK(vp); 681 /* 682 * v_usecount may have been bumped after VOP_LOCK() dropped 683 * the vnode interlock and before it was locked again. 684 * 685 * It is not necessary to recheck VI_DOOMED because it can 686 * only be set by another thread that holds both the vnode 687 * lock and vnode interlock. If another thread has the 688 * vnode lock before we get to VOP_LOCK() and obtains the 689 * vnode interlock after VOP_LOCK() drops the vnode 690 * interlock, the other thread will be unable to drop the 691 * vnode lock before our VOP_LOCK() call fails. 692 */ 693 if (vp->v_usecount || 694 (!vlru_allow_cache_src && 695 !LIST_EMPTY(&(vp)->v_cache_src)) || 696 (vp->v_object != NULL && 697 vp->v_object->resident_page_count > trigger)) { 698 VOP_UNLOCK(vp, LK_INTERLOCK); 699 goto next_iter_mntunlocked; 700 } 701 KASSERT((vp->v_iflag & VI_DOOMED) == 0, 702 ("VI_DOOMED unexpectedly detected in vlrureclaim()")); 703 vgonel(vp); 704 VOP_UNLOCK(vp, 0); 705 vdropl(vp); 706 done++; 707 next_iter_mntunlocked: 708 if (!should_yield()) 709 goto relock_mnt; 710 goto yield; 711 next_iter: 712 if (!should_yield()) 713 continue; 714 MNT_IUNLOCK(mp); 715 yield: 716 kern_yield(-1); 717 relock_mnt: 718 MNT_ILOCK(mp); 719 } 720 MNT_IUNLOCK(mp); 721 vn_finished_write(mp); 722 return done; 723 } 724 725 /* 726 * Attempt to keep the free list at wantfreevnodes length. 727 */ 728 static void 729 vnlru_free(int count) 730 { 731 struct vnode *vp; 732 int vfslocked; 733 734 mtx_assert(&vnode_free_list_mtx, MA_OWNED); 735 for (; count > 0; count--) { 736 vp = TAILQ_FIRST(&vnode_free_list); 737 /* 738 * The list can be modified while the free_list_mtx 739 * has been dropped and vp could be NULL here. 740 */ 741 if (!vp) 742 break; 743 VNASSERT(vp->v_op != NULL, vp, 744 ("vnlru_free: vnode already reclaimed.")); 745 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 746 /* 747 * Don't recycle if we can't get the interlock. 748 */ 749 if (!VI_TRYLOCK(vp)) { 750 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 751 continue; 752 } 753 VNASSERT(VCANRECYCLE(vp), vp, 754 ("vp inconsistent on freelist")); 755 freevnodes--; 756 vp->v_iflag &= ~VI_FREE; 757 vholdl(vp); 758 mtx_unlock(&vnode_free_list_mtx); 759 VI_UNLOCK(vp); 760 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 761 vtryrecycle(vp); 762 VFS_UNLOCK_GIANT(vfslocked); 763 /* 764 * If the recycled succeeded this vdrop will actually free 765 * the vnode. If not it will simply place it back on 766 * the free list. 767 */ 768 vdrop(vp); 769 mtx_lock(&vnode_free_list_mtx); 770 } 771 } 772 /* 773 * Attempt to recycle vnodes in a context that is always safe to block. 774 * Calling vlrurecycle() from the bowels of filesystem code has some 775 * interesting deadlock problems. 776 */ 777 static struct proc *vnlruproc; 778 static int vnlruproc_sig; 779 780 static void 781 vnlru_proc(void) 782 { 783 struct mount *mp, *nmp; 784 int done, vfslocked; 785 struct proc *p = vnlruproc; 786 787 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p, 788 SHUTDOWN_PRI_FIRST); 789 790 for (;;) { 791 kproc_suspend_check(p); 792 mtx_lock(&vnode_free_list_mtx); 793 if (freevnodes > wantfreevnodes) 794 vnlru_free(freevnodes - wantfreevnodes); 795 if (numvnodes <= desiredvnodes * 9 / 10) { 796 vnlruproc_sig = 0; 797 wakeup(&vnlruproc_sig); 798 msleep(vnlruproc, &vnode_free_list_mtx, 799 PVFS|PDROP, "vlruwt", hz); 800 continue; 801 } 802 mtx_unlock(&vnode_free_list_mtx); 803 done = 0; 804 mtx_lock(&mountlist_mtx); 805 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 806 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { 807 nmp = TAILQ_NEXT(mp, mnt_list); 808 continue; 809 } 810 vfslocked = VFS_LOCK_GIANT(mp); 811 done += vlrureclaim(mp); 812 VFS_UNLOCK_GIANT(vfslocked); 813 mtx_lock(&mountlist_mtx); 814 nmp = TAILQ_NEXT(mp, mnt_list); 815 vfs_unbusy(mp); 816 } 817 mtx_unlock(&mountlist_mtx); 818 if (done == 0) { 819 #if 0 820 /* These messages are temporary debugging aids */ 821 if (vnlru_nowhere < 5) 822 printf("vnlru process getting nowhere..\n"); 823 else if (vnlru_nowhere == 5) 824 printf("vnlru process messages stopped.\n"); 825 #endif 826 vnlru_nowhere++; 827 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); 828 } else 829 kern_yield(-1); 830 } 831 } 832 833 static struct kproc_desc vnlru_kp = { 834 "vnlru", 835 vnlru_proc, 836 &vnlruproc 837 }; 838 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, 839 &vnlru_kp); 840 841 /* 842 * Routines having to do with the management of the vnode table. 843 */ 844 845 void 846 vdestroy(struct vnode *vp) 847 { 848 struct bufobj *bo; 849 850 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 851 mtx_lock(&vnode_free_list_mtx); 852 numvnodes--; 853 mtx_unlock(&vnode_free_list_mtx); 854 bo = &vp->v_bufobj; 855 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, 856 ("cleaned vnode still on the free list.")); 857 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); 858 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count")); 859 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); 860 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); 861 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); 862 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); 863 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL")); 864 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); 865 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL")); 866 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); 867 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); 868 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for ..")); 869 VI_UNLOCK(vp); 870 #ifdef MAC 871 mac_vnode_destroy(vp); 872 #endif 873 if (vp->v_pollinfo != NULL) 874 destroy_vpollinfo(vp->v_pollinfo); 875 #ifdef INVARIANTS 876 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */ 877 vp->v_op = NULL; 878 #endif 879 lockdestroy(vp->v_vnlock); 880 mtx_destroy(&vp->v_interlock); 881 mtx_destroy(BO_MTX(bo)); 882 uma_zfree(vnode_zone, vp); 883 } 884 885 /* 886 * Try to recycle a freed vnode. We abort if anyone picks up a reference 887 * before we actually vgone(). This function must be called with the vnode 888 * held to prevent the vnode from being returned to the free list midway 889 * through vgone(). 890 */ 891 static int 892 vtryrecycle(struct vnode *vp) 893 { 894 struct mount *vnmp; 895 896 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 897 VNASSERT(vp->v_holdcnt, vp, 898 ("vtryrecycle: Recycling vp %p without a reference.", vp)); 899 /* 900 * This vnode may found and locked via some other list, if so we 901 * can't recycle it yet. 902 */ 903 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 904 CTR2(KTR_VFS, 905 "%s: impossible to recycle, vp %p lock is already held", 906 __func__, vp); 907 return (EWOULDBLOCK); 908 } 909 /* 910 * Don't recycle if its filesystem is being suspended. 911 */ 912 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { 913 VOP_UNLOCK(vp, 0); 914 CTR2(KTR_VFS, 915 "%s: impossible to recycle, cannot start the write for %p", 916 __func__, vp); 917 return (EBUSY); 918 } 919 /* 920 * If we got this far, we need to acquire the interlock and see if 921 * anyone picked up this vnode from another list. If not, we will 922 * mark it with DOOMED via vgonel() so that anyone who does find it 923 * will skip over it. 924 */ 925 VI_LOCK(vp); 926 if (vp->v_usecount) { 927 VOP_UNLOCK(vp, LK_INTERLOCK); 928 vn_finished_write(vnmp); 929 CTR2(KTR_VFS, 930 "%s: impossible to recycle, %p is already referenced", 931 __func__, vp); 932 return (EBUSY); 933 } 934 if ((vp->v_iflag & VI_DOOMED) == 0) 935 vgonel(vp); 936 VOP_UNLOCK(vp, LK_INTERLOCK); 937 vn_finished_write(vnmp); 938 return (0); 939 } 940 941 /* 942 * Return the next vnode from the free list. 943 */ 944 int 945 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, 946 struct vnode **vpp) 947 { 948 struct vnode *vp = NULL; 949 struct bufobj *bo; 950 951 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag); 952 mtx_lock(&vnode_free_list_mtx); 953 /* 954 * Lend our context to reclaim vnodes if they've exceeded the max. 955 */ 956 if (freevnodes > wantfreevnodes) 957 vnlru_free(1); 958 /* 959 * Wait for available vnodes. 960 */ 961 if (numvnodes > desiredvnodes) { 962 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) { 963 /* 964 * File system is beeing suspended, we cannot risk a 965 * deadlock here, so allocate new vnode anyway. 966 */ 967 if (freevnodes > wantfreevnodes) 968 vnlru_free(freevnodes - wantfreevnodes); 969 goto alloc; 970 } 971 if (vnlruproc_sig == 0) { 972 vnlruproc_sig = 1; /* avoid unnecessary wakeups */ 973 wakeup(vnlruproc); 974 } 975 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS, 976 "vlruwk", hz); 977 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */ 978 if (numvnodes > desiredvnodes) { 979 mtx_unlock(&vnode_free_list_mtx); 980 return (ENFILE); 981 } 982 #endif 983 } 984 alloc: 985 numvnodes++; 986 mtx_unlock(&vnode_free_list_mtx); 987 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO); 988 /* 989 * Setup locks. 990 */ 991 vp->v_vnlock = &vp->v_lock; 992 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); 993 /* 994 * By default, don't allow shared locks unless filesystems 995 * opt-in. 996 */ 997 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE); 998 /* 999 * Initialize bufobj. 1000 */ 1001 bo = &vp->v_bufobj; 1002 bo->__bo_vnode = vp; 1003 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF); 1004 bo->bo_ops = &buf_ops_bio; 1005 bo->bo_private = vp; 1006 TAILQ_INIT(&bo->bo_clean.bv_hd); 1007 TAILQ_INIT(&bo->bo_dirty.bv_hd); 1008 /* 1009 * Initialize namecache. 1010 */ 1011 LIST_INIT(&vp->v_cache_src); 1012 TAILQ_INIT(&vp->v_cache_dst); 1013 /* 1014 * Finalize various vnode identity bits. 1015 */ 1016 vp->v_type = VNON; 1017 vp->v_tag = tag; 1018 vp->v_op = vops; 1019 v_incr_usecount(vp); 1020 vp->v_data = 0; 1021 #ifdef MAC 1022 mac_vnode_init(vp); 1023 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) 1024 mac_vnode_associate_singlelabel(mp, vp); 1025 else if (mp == NULL && vops != &dead_vnodeops) 1026 printf("NULL mp in getnewvnode()\n"); 1027 #endif 1028 if (mp != NULL) { 1029 bo->bo_bsize = mp->mnt_stat.f_iosize; 1030 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) 1031 vp->v_vflag |= VV_NOKNOTE; 1032 } 1033 1034 *vpp = vp; 1035 return (0); 1036 } 1037 1038 /* 1039 * Delete from old mount point vnode list, if on one. 1040 */ 1041 static void 1042 delmntque(struct vnode *vp) 1043 { 1044 struct mount *mp; 1045 1046 mp = vp->v_mount; 1047 if (mp == NULL) 1048 return; 1049 MNT_ILOCK(mp); 1050 vp->v_mount = NULL; 1051 VNASSERT(mp->mnt_nvnodelistsize > 0, vp, 1052 ("bad mount point vnode list size")); 1053 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1054 mp->mnt_nvnodelistsize--; 1055 MNT_REL(mp); 1056 MNT_IUNLOCK(mp); 1057 } 1058 1059 static void 1060 insmntque_stddtr(struct vnode *vp, void *dtr_arg) 1061 { 1062 1063 vp->v_data = NULL; 1064 vp->v_op = &dead_vnodeops; 1065 /* XXX non mp-safe fs may still call insmntque with vnode 1066 unlocked */ 1067 if (!VOP_ISLOCKED(vp)) 1068 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1069 vgone(vp); 1070 vput(vp); 1071 } 1072 1073 /* 1074 * Insert into list of vnodes for the new mount point, if available. 1075 */ 1076 int 1077 insmntque1(struct vnode *vp, struct mount *mp, 1078 void (*dtr)(struct vnode *, void *), void *dtr_arg) 1079 { 1080 int locked; 1081 1082 KASSERT(vp->v_mount == NULL, 1083 ("insmntque: vnode already on per mount vnode list")); 1084 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); 1085 #ifdef DEBUG_VFS_LOCKS 1086 if (!VFS_NEEDSGIANT(mp)) 1087 ASSERT_VOP_ELOCKED(vp, 1088 "insmntque: mp-safe fs and non-locked vp"); 1089 #endif 1090 MNT_ILOCK(mp); 1091 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 && 1092 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 || 1093 mp->mnt_nvnodelistsize == 0)) { 1094 locked = VOP_ISLOCKED(vp); 1095 if (!locked || (locked == LK_EXCLUSIVE && 1096 (vp->v_vflag & VV_FORCEINSMQ) == 0)) { 1097 MNT_IUNLOCK(mp); 1098 if (dtr != NULL) 1099 dtr(vp, dtr_arg); 1100 return (EBUSY); 1101 } 1102 } 1103 vp->v_mount = mp; 1104 MNT_REF(mp); 1105 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1106 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, 1107 ("neg mount point vnode list size")); 1108 mp->mnt_nvnodelistsize++; 1109 MNT_IUNLOCK(mp); 1110 return (0); 1111 } 1112 1113 int 1114 insmntque(struct vnode *vp, struct mount *mp) 1115 { 1116 1117 return (insmntque1(vp, mp, insmntque_stddtr, NULL)); 1118 } 1119 1120 /* 1121 * Flush out and invalidate all buffers associated with a bufobj 1122 * Called with the underlying object locked. 1123 */ 1124 int 1125 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo) 1126 { 1127 int error; 1128 1129 BO_LOCK(bo); 1130 if (flags & V_SAVE) { 1131 error = bufobj_wwait(bo, slpflag, slptimeo); 1132 if (error) { 1133 BO_UNLOCK(bo); 1134 return (error); 1135 } 1136 if (bo->bo_dirty.bv_cnt > 0) { 1137 BO_UNLOCK(bo); 1138 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0) 1139 return (error); 1140 /* 1141 * XXX We could save a lock/unlock if this was only 1142 * enabled under INVARIANTS 1143 */ 1144 BO_LOCK(bo); 1145 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) 1146 panic("vinvalbuf: dirty bufs"); 1147 } 1148 } 1149 /* 1150 * If you alter this loop please notice that interlock is dropped and 1151 * reacquired in flushbuflist. Special care is needed to ensure that 1152 * no race conditions occur from this. 1153 */ 1154 do { 1155 error = flushbuflist(&bo->bo_clean, 1156 flags, bo, slpflag, slptimeo); 1157 if (error == 0) 1158 error = flushbuflist(&bo->bo_dirty, 1159 flags, bo, slpflag, slptimeo); 1160 if (error != 0 && error != EAGAIN) { 1161 BO_UNLOCK(bo); 1162 return (error); 1163 } 1164 } while (error != 0); 1165 1166 /* 1167 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 1168 * have write I/O in-progress but if there is a VM object then the 1169 * VM object can also have read-I/O in-progress. 1170 */ 1171 do { 1172 bufobj_wwait(bo, 0, 0); 1173 BO_UNLOCK(bo); 1174 if (bo->bo_object != NULL) { 1175 VM_OBJECT_LOCK(bo->bo_object); 1176 vm_object_pip_wait(bo->bo_object, "bovlbx"); 1177 VM_OBJECT_UNLOCK(bo->bo_object); 1178 } 1179 BO_LOCK(bo); 1180 } while (bo->bo_numoutput > 0); 1181 BO_UNLOCK(bo); 1182 1183 /* 1184 * Destroy the copy in the VM cache, too. 1185 */ 1186 if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) { 1187 VM_OBJECT_LOCK(bo->bo_object); 1188 vm_object_page_remove(bo->bo_object, 0, 0, 1189 (flags & V_SAVE) ? TRUE : FALSE); 1190 VM_OBJECT_UNLOCK(bo->bo_object); 1191 } 1192 1193 #ifdef INVARIANTS 1194 BO_LOCK(bo); 1195 if ((flags & (V_ALT | V_NORMAL)) == 0 && 1196 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) 1197 panic("vinvalbuf: flush failed"); 1198 BO_UNLOCK(bo); 1199 #endif 1200 return (0); 1201 } 1202 1203 /* 1204 * Flush out and invalidate all buffers associated with a vnode. 1205 * Called with the underlying object locked. 1206 */ 1207 int 1208 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 1209 { 1210 1211 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); 1212 ASSERT_VOP_LOCKED(vp, "vinvalbuf"); 1213 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo)); 1214 } 1215 1216 /* 1217 * Flush out buffers on the specified list. 1218 * 1219 */ 1220 static int 1221 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, 1222 int slptimeo) 1223 { 1224 struct buf *bp, *nbp; 1225 int retval, error; 1226 daddr_t lblkno; 1227 b_xflags_t xflags; 1228 1229 ASSERT_BO_LOCKED(bo); 1230 1231 retval = 0; 1232 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { 1233 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) || 1234 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) { 1235 continue; 1236 } 1237 lblkno = 0; 1238 xflags = 0; 1239 if (nbp != NULL) { 1240 lblkno = nbp->b_lblkno; 1241 xflags = nbp->b_xflags & 1242 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN); 1243 } 1244 retval = EAGAIN; 1245 error = BUF_TIMELOCK(bp, 1246 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo), 1247 "flushbuf", slpflag, slptimeo); 1248 if (error) { 1249 BO_LOCK(bo); 1250 return (error != ENOLCK ? error : EAGAIN); 1251 } 1252 KASSERT(bp->b_bufobj == bo, 1253 ("bp %p wrong b_bufobj %p should be %p", 1254 bp, bp->b_bufobj, bo)); 1255 if (bp->b_bufobj != bo) { /* XXX: necessary ? */ 1256 BUF_UNLOCK(bp); 1257 BO_LOCK(bo); 1258 return (EAGAIN); 1259 } 1260 /* 1261 * XXX Since there are no node locks for NFS, I 1262 * believe there is a slight chance that a delayed 1263 * write will occur while sleeping just above, so 1264 * check for it. 1265 */ 1266 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 1267 (flags & V_SAVE)) { 1268 BO_LOCK(bo); 1269 bremfree(bp); 1270 BO_UNLOCK(bo); 1271 bp->b_flags |= B_ASYNC; 1272 bwrite(bp); 1273 BO_LOCK(bo); 1274 return (EAGAIN); /* XXX: why not loop ? */ 1275 } 1276 BO_LOCK(bo); 1277 bremfree(bp); 1278 BO_UNLOCK(bo); 1279 bp->b_flags |= (B_INVAL | B_RELBUF); 1280 bp->b_flags &= ~B_ASYNC; 1281 brelse(bp); 1282 BO_LOCK(bo); 1283 if (nbp != NULL && 1284 (nbp->b_bufobj != bo || 1285 nbp->b_lblkno != lblkno || 1286 (nbp->b_xflags & 1287 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags)) 1288 break; /* nbp invalid */ 1289 } 1290 return (retval); 1291 } 1292 1293 /* 1294 * Truncate a file's buffer and pages to a specified length. This 1295 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 1296 * sync activity. 1297 */ 1298 int 1299 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td, 1300 off_t length, int blksize) 1301 { 1302 struct buf *bp, *nbp; 1303 int anyfreed; 1304 int trunclbn; 1305 struct bufobj *bo; 1306 1307 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__, 1308 vp, cred, blksize, (uintmax_t)length); 1309 1310 /* 1311 * Round up to the *next* lbn. 1312 */ 1313 trunclbn = (length + blksize - 1) / blksize; 1314 1315 ASSERT_VOP_LOCKED(vp, "vtruncbuf"); 1316 restart: 1317 bo = &vp->v_bufobj; 1318 BO_LOCK(bo); 1319 anyfreed = 1; 1320 for (;anyfreed;) { 1321 anyfreed = 0; 1322 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { 1323 if (bp->b_lblkno < trunclbn) 1324 continue; 1325 if (BUF_LOCK(bp, 1326 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1327 BO_MTX(bo)) == ENOLCK) 1328 goto restart; 1329 1330 BO_LOCK(bo); 1331 bremfree(bp); 1332 BO_UNLOCK(bo); 1333 bp->b_flags |= (B_INVAL | B_RELBUF); 1334 bp->b_flags &= ~B_ASYNC; 1335 brelse(bp); 1336 anyfreed = 1; 1337 1338 BO_LOCK(bo); 1339 if (nbp != NULL && 1340 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 1341 (nbp->b_vp != vp) || 1342 (nbp->b_flags & B_DELWRI))) { 1343 BO_UNLOCK(bo); 1344 goto restart; 1345 } 1346 } 1347 1348 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1349 if (bp->b_lblkno < trunclbn) 1350 continue; 1351 if (BUF_LOCK(bp, 1352 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1353 BO_MTX(bo)) == ENOLCK) 1354 goto restart; 1355 BO_LOCK(bo); 1356 bremfree(bp); 1357 BO_UNLOCK(bo); 1358 bp->b_flags |= (B_INVAL | B_RELBUF); 1359 bp->b_flags &= ~B_ASYNC; 1360 brelse(bp); 1361 anyfreed = 1; 1362 1363 BO_LOCK(bo); 1364 if (nbp != NULL && 1365 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 1366 (nbp->b_vp != vp) || 1367 (nbp->b_flags & B_DELWRI) == 0)) { 1368 BO_UNLOCK(bo); 1369 goto restart; 1370 } 1371 } 1372 } 1373 1374 if (length > 0) { 1375 restartsync: 1376 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1377 if (bp->b_lblkno > 0) 1378 continue; 1379 /* 1380 * Since we hold the vnode lock this should only 1381 * fail if we're racing with the buf daemon. 1382 */ 1383 if (BUF_LOCK(bp, 1384 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1385 BO_MTX(bo)) == ENOLCK) { 1386 goto restart; 1387 } 1388 VNASSERT((bp->b_flags & B_DELWRI), vp, 1389 ("buf(%p) on dirty queue without DELWRI", bp)); 1390 1391 BO_LOCK(bo); 1392 bremfree(bp); 1393 BO_UNLOCK(bo); 1394 bawrite(bp); 1395 BO_LOCK(bo); 1396 goto restartsync; 1397 } 1398 } 1399 1400 bufobj_wwait(bo, 0, 0); 1401 BO_UNLOCK(bo); 1402 vnode_pager_setsize(vp, length); 1403 1404 return (0); 1405 } 1406 1407 /* 1408 * buf_splay() - splay tree core for the clean/dirty list of buffers in 1409 * a vnode. 1410 * 1411 * NOTE: We have to deal with the special case of a background bitmap 1412 * buffer, a situation where two buffers will have the same logical 1413 * block offset. We want (1) only the foreground buffer to be accessed 1414 * in a lookup and (2) must differentiate between the foreground and 1415 * background buffer in the splay tree algorithm because the splay 1416 * tree cannot normally handle multiple entities with the same 'index'. 1417 * We accomplish this by adding differentiating flags to the splay tree's 1418 * numerical domain. 1419 */ 1420 static 1421 struct buf * 1422 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root) 1423 { 1424 struct buf dummy; 1425 struct buf *lefttreemax, *righttreemin, *y; 1426 1427 if (root == NULL) 1428 return (NULL); 1429 lefttreemax = righttreemin = &dummy; 1430 for (;;) { 1431 if (lblkno < root->b_lblkno || 1432 (lblkno == root->b_lblkno && 1433 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1434 if ((y = root->b_left) == NULL) 1435 break; 1436 if (lblkno < y->b_lblkno) { 1437 /* Rotate right. */ 1438 root->b_left = y->b_right; 1439 y->b_right = root; 1440 root = y; 1441 if ((y = root->b_left) == NULL) 1442 break; 1443 } 1444 /* Link into the new root's right tree. */ 1445 righttreemin->b_left = root; 1446 righttreemin = root; 1447 } else if (lblkno > root->b_lblkno || 1448 (lblkno == root->b_lblkno && 1449 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) { 1450 if ((y = root->b_right) == NULL) 1451 break; 1452 if (lblkno > y->b_lblkno) { 1453 /* Rotate left. */ 1454 root->b_right = y->b_left; 1455 y->b_left = root; 1456 root = y; 1457 if ((y = root->b_right) == NULL) 1458 break; 1459 } 1460 /* Link into the new root's left tree. */ 1461 lefttreemax->b_right = root; 1462 lefttreemax = root; 1463 } else { 1464 break; 1465 } 1466 root = y; 1467 } 1468 /* Assemble the new root. */ 1469 lefttreemax->b_right = root->b_left; 1470 righttreemin->b_left = root->b_right; 1471 root->b_left = dummy.b_right; 1472 root->b_right = dummy.b_left; 1473 return (root); 1474 } 1475 1476 static void 1477 buf_vlist_remove(struct buf *bp) 1478 { 1479 struct buf *root; 1480 struct bufv *bv; 1481 1482 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); 1483 ASSERT_BO_LOCKED(bp->b_bufobj); 1484 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) != 1485 (BX_VNDIRTY|BX_VNCLEAN), 1486 ("buf_vlist_remove: Buf %p is on two lists", bp)); 1487 if (bp->b_xflags & BX_VNDIRTY) 1488 bv = &bp->b_bufobj->bo_dirty; 1489 else 1490 bv = &bp->b_bufobj->bo_clean; 1491 if (bp != bv->bv_root) { 1492 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root); 1493 KASSERT(root == bp, ("splay lookup failed in remove")); 1494 } 1495 if (bp->b_left == NULL) { 1496 root = bp->b_right; 1497 } else { 1498 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left); 1499 root->b_right = bp->b_right; 1500 } 1501 bv->bv_root = root; 1502 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); 1503 bv->bv_cnt--; 1504 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1505 } 1506 1507 /* 1508 * Add the buffer to the sorted clean or dirty block list using a 1509 * splay tree algorithm. 1510 * 1511 * NOTE: xflags is passed as a constant, optimizing this inline function! 1512 */ 1513 static void 1514 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) 1515 { 1516 struct buf *root; 1517 struct bufv *bv; 1518 1519 ASSERT_BO_LOCKED(bo); 1520 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, 1521 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); 1522 bp->b_xflags |= xflags; 1523 if (xflags & BX_VNDIRTY) 1524 bv = &bo->bo_dirty; 1525 else 1526 bv = &bo->bo_clean; 1527 1528 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root); 1529 if (root == NULL) { 1530 bp->b_left = NULL; 1531 bp->b_right = NULL; 1532 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); 1533 } else if (bp->b_lblkno < root->b_lblkno || 1534 (bp->b_lblkno == root->b_lblkno && 1535 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1536 bp->b_left = root->b_left; 1537 bp->b_right = root; 1538 root->b_left = NULL; 1539 TAILQ_INSERT_BEFORE(root, bp, b_bobufs); 1540 } else { 1541 bp->b_right = root->b_right; 1542 bp->b_left = root; 1543 root->b_right = NULL; 1544 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs); 1545 } 1546 bv->bv_cnt++; 1547 bv->bv_root = bp; 1548 } 1549 1550 /* 1551 * Lookup a buffer using the splay tree. Note that we specifically avoid 1552 * shadow buffers used in background bitmap writes. 1553 * 1554 * This code isn't quite efficient as it could be because we are maintaining 1555 * two sorted lists and do not know which list the block resides in. 1556 * 1557 * During a "make buildworld" the desired buffer is found at one of 1558 * the roots more than 60% of the time. Thus, checking both roots 1559 * before performing either splay eliminates unnecessary splays on the 1560 * first tree splayed. 1561 */ 1562 struct buf * 1563 gbincore(struct bufobj *bo, daddr_t lblkno) 1564 { 1565 struct buf *bp; 1566 1567 ASSERT_BO_LOCKED(bo); 1568 if ((bp = bo->bo_clean.bv_root) != NULL && 1569 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1570 return (bp); 1571 if ((bp = bo->bo_dirty.bv_root) != NULL && 1572 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1573 return (bp); 1574 if ((bp = bo->bo_clean.bv_root) != NULL) { 1575 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp); 1576 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1577 return (bp); 1578 } 1579 if ((bp = bo->bo_dirty.bv_root) != NULL) { 1580 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp); 1581 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1582 return (bp); 1583 } 1584 return (NULL); 1585 } 1586 1587 /* 1588 * Associate a buffer with a vnode. 1589 */ 1590 void 1591 bgetvp(struct vnode *vp, struct buf *bp) 1592 { 1593 struct bufobj *bo; 1594 1595 bo = &vp->v_bufobj; 1596 ASSERT_BO_LOCKED(bo); 1597 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); 1598 1599 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); 1600 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, 1601 ("bgetvp: bp already attached! %p", bp)); 1602 1603 vhold(vp); 1604 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT) 1605 bp->b_flags |= B_NEEDSGIANT; 1606 bp->b_vp = vp; 1607 bp->b_bufobj = bo; 1608 /* 1609 * Insert onto list for new vnode. 1610 */ 1611 buf_vlist_add(bp, bo, BX_VNCLEAN); 1612 } 1613 1614 /* 1615 * Disassociate a buffer from a vnode. 1616 */ 1617 void 1618 brelvp(struct buf *bp) 1619 { 1620 struct bufobj *bo; 1621 struct vnode *vp; 1622 1623 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1624 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 1625 1626 /* 1627 * Delete from old vnode list, if on one. 1628 */ 1629 vp = bp->b_vp; /* XXX */ 1630 bo = bp->b_bufobj; 1631 BO_LOCK(bo); 1632 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1633 buf_vlist_remove(bp); 1634 else 1635 panic("brelvp: Buffer %p not on queue.", bp); 1636 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 1637 bo->bo_flag &= ~BO_ONWORKLST; 1638 mtx_lock(&sync_mtx); 1639 LIST_REMOVE(bo, bo_synclist); 1640 syncer_worklist_len--; 1641 mtx_unlock(&sync_mtx); 1642 } 1643 bp->b_flags &= ~B_NEEDSGIANT; 1644 bp->b_vp = NULL; 1645 bp->b_bufobj = NULL; 1646 BO_UNLOCK(bo); 1647 vdrop(vp); 1648 } 1649 1650 /* 1651 * Add an item to the syncer work queue. 1652 */ 1653 static void 1654 vn_syncer_add_to_worklist(struct bufobj *bo, int delay) 1655 { 1656 int queue, slot; 1657 1658 ASSERT_BO_LOCKED(bo); 1659 1660 mtx_lock(&sync_mtx); 1661 if (bo->bo_flag & BO_ONWORKLST) 1662 LIST_REMOVE(bo, bo_synclist); 1663 else { 1664 bo->bo_flag |= BO_ONWORKLST; 1665 syncer_worklist_len++; 1666 } 1667 1668 if (delay > syncer_maxdelay - 2) 1669 delay = syncer_maxdelay - 2; 1670 slot = (syncer_delayno + delay) & syncer_mask; 1671 1672 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ : 1673 WI_MPSAFEQ; 1674 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo, 1675 bo_synclist); 1676 mtx_unlock(&sync_mtx); 1677 } 1678 1679 static int 1680 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) 1681 { 1682 int error, len; 1683 1684 mtx_lock(&sync_mtx); 1685 len = syncer_worklist_len - sync_vnode_count; 1686 mtx_unlock(&sync_mtx); 1687 error = SYSCTL_OUT(req, &len, sizeof(len)); 1688 return (error); 1689 } 1690 1691 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0, 1692 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); 1693 1694 static struct proc *updateproc; 1695 static void sched_sync(void); 1696 static struct kproc_desc up_kp = { 1697 "syncer", 1698 sched_sync, 1699 &updateproc 1700 }; 1701 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp); 1702 1703 static int 1704 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td) 1705 { 1706 struct vnode *vp; 1707 struct mount *mp; 1708 1709 *bo = LIST_FIRST(slp); 1710 if (*bo == NULL) 1711 return (0); 1712 vp = (*bo)->__bo_vnode; /* XXX */ 1713 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0) 1714 return (1); 1715 /* 1716 * We use vhold in case the vnode does not 1717 * successfully sync. vhold prevents the vnode from 1718 * going away when we unlock the sync_mtx so that 1719 * we can acquire the vnode interlock. 1720 */ 1721 vholdl(vp); 1722 mtx_unlock(&sync_mtx); 1723 VI_UNLOCK(vp); 1724 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { 1725 vdrop(vp); 1726 mtx_lock(&sync_mtx); 1727 return (*bo == LIST_FIRST(slp)); 1728 } 1729 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1730 (void) VOP_FSYNC(vp, MNT_LAZY, td); 1731 VOP_UNLOCK(vp, 0); 1732 vn_finished_write(mp); 1733 BO_LOCK(*bo); 1734 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) { 1735 /* 1736 * Put us back on the worklist. The worklist 1737 * routine will remove us from our current 1738 * position and then add us back in at a later 1739 * position. 1740 */ 1741 vn_syncer_add_to_worklist(*bo, syncdelay); 1742 } 1743 BO_UNLOCK(*bo); 1744 vdrop(vp); 1745 mtx_lock(&sync_mtx); 1746 return (0); 1747 } 1748 1749 /* 1750 * System filesystem synchronizer daemon. 1751 */ 1752 static void 1753 sched_sync(void) 1754 { 1755 struct synclist *gnext, *next; 1756 struct synclist *gslp, *slp; 1757 struct bufobj *bo; 1758 long starttime; 1759 struct thread *td = curthread; 1760 int last_work_seen; 1761 int net_worklist_len; 1762 int syncer_final_iter; 1763 int first_printf; 1764 int error; 1765 1766 last_work_seen = 0; 1767 syncer_final_iter = 0; 1768 first_printf = 1; 1769 syncer_state = SYNCER_RUNNING; 1770 starttime = time_uptime; 1771 td->td_pflags |= TDP_NORUNNINGBUF; 1772 1773 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, 1774 SHUTDOWN_PRI_LAST); 1775 1776 mtx_lock(&sync_mtx); 1777 for (;;) { 1778 if (syncer_state == SYNCER_FINAL_DELAY && 1779 syncer_final_iter == 0) { 1780 mtx_unlock(&sync_mtx); 1781 kproc_suspend_check(td->td_proc); 1782 mtx_lock(&sync_mtx); 1783 } 1784 net_worklist_len = syncer_worklist_len - sync_vnode_count; 1785 if (syncer_state != SYNCER_RUNNING && 1786 starttime != time_uptime) { 1787 if (first_printf) { 1788 printf("\nSyncing disks, vnodes remaining..."); 1789 first_printf = 0; 1790 } 1791 printf("%d ", net_worklist_len); 1792 } 1793 starttime = time_uptime; 1794 1795 /* 1796 * Push files whose dirty time has expired. Be careful 1797 * of interrupt race on slp queue. 1798 * 1799 * Skip over empty worklist slots when shutting down. 1800 */ 1801 do { 1802 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno]; 1803 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno]; 1804 syncer_delayno += 1; 1805 if (syncer_delayno == syncer_maxdelay) 1806 syncer_delayno = 0; 1807 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno]; 1808 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno]; 1809 /* 1810 * If the worklist has wrapped since the 1811 * it was emptied of all but syncer vnodes, 1812 * switch to the FINAL_DELAY state and run 1813 * for one more second. 1814 */ 1815 if (syncer_state == SYNCER_SHUTTING_DOWN && 1816 net_worklist_len == 0 && 1817 last_work_seen == syncer_delayno) { 1818 syncer_state = SYNCER_FINAL_DELAY; 1819 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; 1820 } 1821 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && 1822 LIST_EMPTY(gslp) && syncer_worklist_len > 0); 1823 1824 /* 1825 * Keep track of the last time there was anything 1826 * on the worklist other than syncer vnodes. 1827 * Return to the SHUTTING_DOWN state if any 1828 * new work appears. 1829 */ 1830 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) 1831 last_work_seen = syncer_delayno; 1832 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) 1833 syncer_state = SYNCER_SHUTTING_DOWN; 1834 while (!LIST_EMPTY(slp)) { 1835 error = sync_vnode(slp, &bo, td); 1836 if (error == 1) { 1837 LIST_REMOVE(bo, bo_synclist); 1838 LIST_INSERT_HEAD(next, bo, bo_synclist); 1839 continue; 1840 } 1841 } 1842 if (!LIST_EMPTY(gslp)) { 1843 mtx_unlock(&sync_mtx); 1844 mtx_lock(&Giant); 1845 mtx_lock(&sync_mtx); 1846 while (!LIST_EMPTY(gslp)) { 1847 error = sync_vnode(gslp, &bo, td); 1848 if (error == 1) { 1849 LIST_REMOVE(bo, bo_synclist); 1850 LIST_INSERT_HEAD(gnext, bo, 1851 bo_synclist); 1852 continue; 1853 } 1854 } 1855 mtx_unlock(&Giant); 1856 } 1857 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) 1858 syncer_final_iter--; 1859 /* 1860 * The variable rushjob allows the kernel to speed up the 1861 * processing of the filesystem syncer process. A rushjob 1862 * value of N tells the filesystem syncer to process the next 1863 * N seconds worth of work on its queue ASAP. Currently rushjob 1864 * is used by the soft update code to speed up the filesystem 1865 * syncer process when the incore state is getting so far 1866 * ahead of the disk that the kernel memory pool is being 1867 * threatened with exhaustion. 1868 */ 1869 if (rushjob > 0) { 1870 rushjob -= 1; 1871 continue; 1872 } 1873 /* 1874 * Just sleep for a short period of time between 1875 * iterations when shutting down to allow some I/O 1876 * to happen. 1877 * 1878 * If it has taken us less than a second to process the 1879 * current work, then wait. Otherwise start right over 1880 * again. We can still lose time if any single round 1881 * takes more than two seconds, but it does not really 1882 * matter as we are just trying to generally pace the 1883 * filesystem activity. 1884 */ 1885 if (syncer_state != SYNCER_RUNNING || 1886 time_uptime == starttime) { 1887 thread_lock(td); 1888 sched_prio(td, PPAUSE); 1889 thread_unlock(td); 1890 } 1891 if (syncer_state != SYNCER_RUNNING) 1892 cv_timedwait(&sync_wakeup, &sync_mtx, 1893 hz / SYNCER_SHUTDOWN_SPEEDUP); 1894 else if (time_uptime == starttime) 1895 cv_timedwait(&sync_wakeup, &sync_mtx, hz); 1896 } 1897 } 1898 1899 /* 1900 * Request the syncer daemon to speed up its work. 1901 * We never push it to speed up more than half of its 1902 * normal turn time, otherwise it could take over the cpu. 1903 */ 1904 int 1905 speedup_syncer(void) 1906 { 1907 int ret = 0; 1908 1909 mtx_lock(&sync_mtx); 1910 if (rushjob < syncdelay / 2) { 1911 rushjob += 1; 1912 stat_rush_requests += 1; 1913 ret = 1; 1914 } 1915 mtx_unlock(&sync_mtx); 1916 cv_broadcast(&sync_wakeup); 1917 return (ret); 1918 } 1919 1920 /* 1921 * Tell the syncer to speed up its work and run though its work 1922 * list several times, then tell it to shut down. 1923 */ 1924 static void 1925 syncer_shutdown(void *arg, int howto) 1926 { 1927 1928 if (howto & RB_NOSYNC) 1929 return; 1930 mtx_lock(&sync_mtx); 1931 syncer_state = SYNCER_SHUTTING_DOWN; 1932 rushjob = 0; 1933 mtx_unlock(&sync_mtx); 1934 cv_broadcast(&sync_wakeup); 1935 kproc_shutdown(arg, howto); 1936 } 1937 1938 /* 1939 * Reassign a buffer from one vnode to another. 1940 * Used to assign file specific control information 1941 * (indirect blocks) to the vnode to which they belong. 1942 */ 1943 void 1944 reassignbuf(struct buf *bp) 1945 { 1946 struct vnode *vp; 1947 struct bufobj *bo; 1948 int delay; 1949 #ifdef INVARIANTS 1950 struct bufv *bv; 1951 #endif 1952 1953 vp = bp->b_vp; 1954 bo = bp->b_bufobj; 1955 ++reassignbufcalls; 1956 1957 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", 1958 bp, bp->b_vp, bp->b_flags); 1959 /* 1960 * B_PAGING flagged buffers cannot be reassigned because their vp 1961 * is not fully linked in. 1962 */ 1963 if (bp->b_flags & B_PAGING) 1964 panic("cannot reassign paging buffer"); 1965 1966 /* 1967 * Delete from old vnode list, if on one. 1968 */ 1969 BO_LOCK(bo); 1970 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1971 buf_vlist_remove(bp); 1972 else 1973 panic("reassignbuf: Buffer %p not on queue.", bp); 1974 /* 1975 * If dirty, put on list of dirty buffers; otherwise insert onto list 1976 * of clean buffers. 1977 */ 1978 if (bp->b_flags & B_DELWRI) { 1979 if ((bo->bo_flag & BO_ONWORKLST) == 0) { 1980 switch (vp->v_type) { 1981 case VDIR: 1982 delay = dirdelay; 1983 break; 1984 case VCHR: 1985 delay = metadelay; 1986 break; 1987 default: 1988 delay = filedelay; 1989 } 1990 vn_syncer_add_to_worklist(bo, delay); 1991 } 1992 buf_vlist_add(bp, bo, BX_VNDIRTY); 1993 } else { 1994 buf_vlist_add(bp, bo, BX_VNCLEAN); 1995 1996 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 1997 mtx_lock(&sync_mtx); 1998 LIST_REMOVE(bo, bo_synclist); 1999 syncer_worklist_len--; 2000 mtx_unlock(&sync_mtx); 2001 bo->bo_flag &= ~BO_ONWORKLST; 2002 } 2003 } 2004 #ifdef INVARIANTS 2005 bv = &bo->bo_clean; 2006 bp = TAILQ_FIRST(&bv->bv_hd); 2007 KASSERT(bp == NULL || bp->b_bufobj == bo, 2008 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2009 bp = TAILQ_LAST(&bv->bv_hd, buflists); 2010 KASSERT(bp == NULL || bp->b_bufobj == bo, 2011 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2012 bv = &bo->bo_dirty; 2013 bp = TAILQ_FIRST(&bv->bv_hd); 2014 KASSERT(bp == NULL || bp->b_bufobj == bo, 2015 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2016 bp = TAILQ_LAST(&bv->bv_hd, buflists); 2017 KASSERT(bp == NULL || bp->b_bufobj == bo, 2018 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2019 #endif 2020 BO_UNLOCK(bo); 2021 } 2022 2023 /* 2024 * Increment the use and hold counts on the vnode, taking care to reference 2025 * the driver's usecount if this is a chardev. The vholdl() will remove 2026 * the vnode from the free list if it is presently free. Requires the 2027 * vnode interlock and returns with it held. 2028 */ 2029 static void 2030 v_incr_usecount(struct vnode *vp) 2031 { 2032 2033 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2034 vp->v_usecount++; 2035 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2036 dev_lock(); 2037 vp->v_rdev->si_usecount++; 2038 dev_unlock(); 2039 } 2040 vholdl(vp); 2041 } 2042 2043 /* 2044 * Turn a holdcnt into a use+holdcnt such that only one call to 2045 * v_decr_usecount is needed. 2046 */ 2047 static void 2048 v_upgrade_usecount(struct vnode *vp) 2049 { 2050 2051 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2052 vp->v_usecount++; 2053 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2054 dev_lock(); 2055 vp->v_rdev->si_usecount++; 2056 dev_unlock(); 2057 } 2058 } 2059 2060 /* 2061 * Decrement the vnode use and hold count along with the driver's usecount 2062 * if this is a chardev. The vdropl() below releases the vnode interlock 2063 * as it may free the vnode. 2064 */ 2065 static void 2066 v_decr_usecount(struct vnode *vp) 2067 { 2068 2069 ASSERT_VI_LOCKED(vp, __FUNCTION__); 2070 VNASSERT(vp->v_usecount > 0, vp, 2071 ("v_decr_usecount: negative usecount")); 2072 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2073 vp->v_usecount--; 2074 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2075 dev_lock(); 2076 vp->v_rdev->si_usecount--; 2077 dev_unlock(); 2078 } 2079 vdropl(vp); 2080 } 2081 2082 /* 2083 * Decrement only the use count and driver use count. This is intended to 2084 * be paired with a follow on vdropl() to release the remaining hold count. 2085 * In this way we may vgone() a vnode with a 0 usecount without risk of 2086 * having it end up on a free list because the hold count is kept above 0. 2087 */ 2088 static void 2089 v_decr_useonly(struct vnode *vp) 2090 { 2091 2092 ASSERT_VI_LOCKED(vp, __FUNCTION__); 2093 VNASSERT(vp->v_usecount > 0, vp, 2094 ("v_decr_useonly: negative usecount")); 2095 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2096 vp->v_usecount--; 2097 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2098 dev_lock(); 2099 vp->v_rdev->si_usecount--; 2100 dev_unlock(); 2101 } 2102 } 2103 2104 /* 2105 * Grab a particular vnode from the free list, increment its 2106 * reference count and lock it. VI_DOOMED is set if the vnode 2107 * is being destroyed. Only callers who specify LK_RETRY will 2108 * see doomed vnodes. If inactive processing was delayed in 2109 * vput try to do it here. 2110 */ 2111 int 2112 vget(struct vnode *vp, int flags, struct thread *td) 2113 { 2114 int error; 2115 2116 error = 0; 2117 VFS_ASSERT_GIANT(vp->v_mount); 2118 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 2119 ("vget: invalid lock operation")); 2120 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); 2121 2122 if ((flags & LK_INTERLOCK) == 0) 2123 VI_LOCK(vp); 2124 vholdl(vp); 2125 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) { 2126 vdrop(vp); 2127 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__, 2128 vp); 2129 return (error); 2130 } 2131 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) 2132 panic("vget: vn_lock failed to return ENOENT\n"); 2133 VI_LOCK(vp); 2134 /* Upgrade our holdcnt to a usecount. */ 2135 v_upgrade_usecount(vp); 2136 /* 2137 * We don't guarantee that any particular close will 2138 * trigger inactive processing so just make a best effort 2139 * here at preventing a reference to a removed file. If 2140 * we don't succeed no harm is done. 2141 */ 2142 if (vp->v_iflag & VI_OWEINACT) { 2143 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && 2144 (flags & LK_NOWAIT) == 0) 2145 vinactive(vp, td); 2146 vp->v_iflag &= ~VI_OWEINACT; 2147 } 2148 VI_UNLOCK(vp); 2149 return (0); 2150 } 2151 2152 /* 2153 * Increase the reference count of a vnode. 2154 */ 2155 void 2156 vref(struct vnode *vp) 2157 { 2158 2159 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2160 VI_LOCK(vp); 2161 v_incr_usecount(vp); 2162 VI_UNLOCK(vp); 2163 } 2164 2165 /* 2166 * Return reference count of a vnode. 2167 * 2168 * The results of this call are only guaranteed when some mechanism other 2169 * than the VI lock is used to stop other processes from gaining references 2170 * to the vnode. This may be the case if the caller holds the only reference. 2171 * This is also useful when stale data is acceptable as race conditions may 2172 * be accounted for by some other means. 2173 */ 2174 int 2175 vrefcnt(struct vnode *vp) 2176 { 2177 int usecnt; 2178 2179 VI_LOCK(vp); 2180 usecnt = vp->v_usecount; 2181 VI_UNLOCK(vp); 2182 2183 return (usecnt); 2184 } 2185 2186 #define VPUTX_VRELE 1 2187 #define VPUTX_VPUT 2 2188 #define VPUTX_VUNREF 3 2189 2190 static void 2191 vputx(struct vnode *vp, int func) 2192 { 2193 int error; 2194 2195 KASSERT(vp != NULL, ("vputx: null vp")); 2196 if (func == VPUTX_VUNREF) 2197 ASSERT_VOP_LOCKED(vp, "vunref"); 2198 else if (func == VPUTX_VPUT) 2199 ASSERT_VOP_LOCKED(vp, "vput"); 2200 else 2201 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func")); 2202 VFS_ASSERT_GIANT(vp->v_mount); 2203 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2204 VI_LOCK(vp); 2205 2206 /* Skip this v_writecount check if we're going to panic below. */ 2207 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp, 2208 ("vputx: missed vn_close")); 2209 error = 0; 2210 2211 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) && 2212 vp->v_usecount == 1)) { 2213 if (func == VPUTX_VPUT) 2214 VOP_UNLOCK(vp, 0); 2215 v_decr_usecount(vp); 2216 return; 2217 } 2218 2219 if (vp->v_usecount != 1) { 2220 vprint("vputx: negative ref count", vp); 2221 panic("vputx: negative ref cnt"); 2222 } 2223 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp); 2224 /* 2225 * We want to hold the vnode until the inactive finishes to 2226 * prevent vgone() races. We drop the use count here and the 2227 * hold count below when we're done. 2228 */ 2229 v_decr_useonly(vp); 2230 /* 2231 * We must call VOP_INACTIVE with the node locked. Mark 2232 * as VI_DOINGINACT to avoid recursion. 2233 */ 2234 vp->v_iflag |= VI_OWEINACT; 2235 switch (func) { 2236 case VPUTX_VRELE: 2237 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2238 VI_LOCK(vp); 2239 break; 2240 case VPUTX_VPUT: 2241 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 2242 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | 2243 LK_NOWAIT); 2244 VI_LOCK(vp); 2245 } 2246 break; 2247 case VPUTX_VUNREF: 2248 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 2249 error = EBUSY; 2250 break; 2251 } 2252 if (vp->v_usecount > 0) 2253 vp->v_iflag &= ~VI_OWEINACT; 2254 if (error == 0) { 2255 if (vp->v_iflag & VI_OWEINACT) 2256 vinactive(vp, curthread); 2257 if (func != VPUTX_VUNREF) 2258 VOP_UNLOCK(vp, 0); 2259 } 2260 vdropl(vp); 2261 } 2262 2263 /* 2264 * Vnode put/release. 2265 * If count drops to zero, call inactive routine and return to freelist. 2266 */ 2267 void 2268 vrele(struct vnode *vp) 2269 { 2270 2271 vputx(vp, VPUTX_VRELE); 2272 } 2273 2274 /* 2275 * Release an already locked vnode. This give the same effects as 2276 * unlock+vrele(), but takes less time and avoids releasing and 2277 * re-aquiring the lock (as vrele() acquires the lock internally.) 2278 */ 2279 void 2280 vput(struct vnode *vp) 2281 { 2282 2283 vputx(vp, VPUTX_VPUT); 2284 } 2285 2286 /* 2287 * Release an exclusively locked vnode. Do not unlock the vnode lock. 2288 */ 2289 void 2290 vunref(struct vnode *vp) 2291 { 2292 2293 vputx(vp, VPUTX_VUNREF); 2294 } 2295 2296 /* 2297 * Somebody doesn't want the vnode recycled. 2298 */ 2299 void 2300 vhold(struct vnode *vp) 2301 { 2302 2303 VI_LOCK(vp); 2304 vholdl(vp); 2305 VI_UNLOCK(vp); 2306 } 2307 2308 void 2309 vholdl(struct vnode *vp) 2310 { 2311 2312 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2313 vp->v_holdcnt++; 2314 if (VSHOULDBUSY(vp)) 2315 vbusy(vp); 2316 } 2317 2318 /* 2319 * Note that there is one less who cares about this vnode. vdrop() is the 2320 * opposite of vhold(). 2321 */ 2322 void 2323 vdrop(struct vnode *vp) 2324 { 2325 2326 VI_LOCK(vp); 2327 vdropl(vp); 2328 } 2329 2330 /* 2331 * Drop the hold count of the vnode. If this is the last reference to 2332 * the vnode we will free it if it has been vgone'd otherwise it is 2333 * placed on the free list. 2334 */ 2335 void 2336 vdropl(struct vnode *vp) 2337 { 2338 2339 ASSERT_VI_LOCKED(vp, "vdropl"); 2340 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2341 if (vp->v_holdcnt <= 0) 2342 panic("vdrop: holdcnt %d", vp->v_holdcnt); 2343 vp->v_holdcnt--; 2344 if (vp->v_holdcnt == 0) { 2345 if (vp->v_iflag & VI_DOOMED) { 2346 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, 2347 vp); 2348 vdestroy(vp); 2349 return; 2350 } else 2351 vfree(vp); 2352 } 2353 VI_UNLOCK(vp); 2354 } 2355 2356 /* 2357 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT 2358 * flags. DOINGINACT prevents us from recursing in calls to vinactive. 2359 * OWEINACT tracks whether a vnode missed a call to inactive due to a 2360 * failed lock upgrade. 2361 */ 2362 static void 2363 vinactive(struct vnode *vp, struct thread *td) 2364 { 2365 2366 ASSERT_VOP_ELOCKED(vp, "vinactive"); 2367 ASSERT_VI_LOCKED(vp, "vinactive"); 2368 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, 2369 ("vinactive: recursed on VI_DOINGINACT")); 2370 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2371 vp->v_iflag |= VI_DOINGINACT; 2372 vp->v_iflag &= ~VI_OWEINACT; 2373 VI_UNLOCK(vp); 2374 VOP_INACTIVE(vp, td); 2375 VI_LOCK(vp); 2376 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, 2377 ("vinactive: lost VI_DOINGINACT")); 2378 vp->v_iflag &= ~VI_DOINGINACT; 2379 } 2380 2381 /* 2382 * Remove any vnodes in the vnode table belonging to mount point mp. 2383 * 2384 * If FORCECLOSE is not specified, there should not be any active ones, 2385 * return error if any are found (nb: this is a user error, not a 2386 * system error). If FORCECLOSE is specified, detach any active vnodes 2387 * that are found. 2388 * 2389 * If WRITECLOSE is set, only flush out regular file vnodes open for 2390 * writing. 2391 * 2392 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. 2393 * 2394 * `rootrefs' specifies the base reference count for the root vnode 2395 * of this filesystem. The root vnode is considered busy if its 2396 * v_usecount exceeds this value. On a successful return, vflush(, td) 2397 * will call vrele() on the root vnode exactly rootrefs times. 2398 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 2399 * be zero. 2400 */ 2401 #ifdef DIAGNOSTIC 2402 static int busyprt = 0; /* print out busy vnodes */ 2403 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes"); 2404 #endif 2405 2406 int 2407 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td) 2408 { 2409 struct vnode *vp, *mvp, *rootvp = NULL; 2410 struct vattr vattr; 2411 int busy = 0, error; 2412 2413 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp, 2414 rootrefs, flags); 2415 if (rootrefs > 0) { 2416 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 2417 ("vflush: bad args")); 2418 /* 2419 * Get the filesystem root vnode. We can vput() it 2420 * immediately, since with rootrefs > 0, it won't go away. 2421 */ 2422 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) { 2423 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d", 2424 __func__, error); 2425 return (error); 2426 } 2427 vput(rootvp); 2428 } 2429 MNT_ILOCK(mp); 2430 loop: 2431 MNT_VNODE_FOREACH(vp, mp, mvp) { 2432 VI_LOCK(vp); 2433 vholdl(vp); 2434 MNT_IUNLOCK(mp); 2435 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE); 2436 if (error) { 2437 vdrop(vp); 2438 MNT_ILOCK(mp); 2439 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); 2440 goto loop; 2441 } 2442 /* 2443 * Skip over a vnodes marked VV_SYSTEM. 2444 */ 2445 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { 2446 VOP_UNLOCK(vp, 0); 2447 vdrop(vp); 2448 MNT_ILOCK(mp); 2449 continue; 2450 } 2451 /* 2452 * If WRITECLOSE is set, flush out unlinked but still open 2453 * files (even if open only for reading) and regular file 2454 * vnodes open for writing. 2455 */ 2456 if (flags & WRITECLOSE) { 2457 error = VOP_GETATTR(vp, &vattr, td->td_ucred); 2458 VI_LOCK(vp); 2459 2460 if ((vp->v_type == VNON || 2461 (error == 0 && vattr.va_nlink > 0)) && 2462 (vp->v_writecount == 0 || vp->v_type != VREG)) { 2463 VOP_UNLOCK(vp, 0); 2464 vdropl(vp); 2465 MNT_ILOCK(mp); 2466 continue; 2467 } 2468 } else 2469 VI_LOCK(vp); 2470 /* 2471 * With v_usecount == 0, all we need to do is clear out the 2472 * vnode data structures and we are done. 2473 * 2474 * If FORCECLOSE is set, forcibly close the vnode. 2475 */ 2476 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { 2477 VNASSERT(vp->v_usecount == 0 || 2478 (vp->v_type != VCHR && vp->v_type != VBLK), vp, 2479 ("device VNODE %p is FORCECLOSED", vp)); 2480 vgonel(vp); 2481 } else { 2482 busy++; 2483 #ifdef DIAGNOSTIC 2484 if (busyprt) 2485 vprint("vflush: busy vnode", vp); 2486 #endif 2487 } 2488 VOP_UNLOCK(vp, 0); 2489 vdropl(vp); 2490 MNT_ILOCK(mp); 2491 } 2492 MNT_IUNLOCK(mp); 2493 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 2494 /* 2495 * If just the root vnode is busy, and if its refcount 2496 * is equal to `rootrefs', then go ahead and kill it. 2497 */ 2498 VI_LOCK(rootvp); 2499 KASSERT(busy > 0, ("vflush: not busy")); 2500 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, 2501 ("vflush: usecount %d < rootrefs %d", 2502 rootvp->v_usecount, rootrefs)); 2503 if (busy == 1 && rootvp->v_usecount == rootrefs) { 2504 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK); 2505 vgone(rootvp); 2506 VOP_UNLOCK(rootvp, 0); 2507 busy = 0; 2508 } else 2509 VI_UNLOCK(rootvp); 2510 } 2511 if (busy) { 2512 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__, 2513 busy); 2514 return (EBUSY); 2515 } 2516 for (; rootrefs > 0; rootrefs--) 2517 vrele(rootvp); 2518 return (0); 2519 } 2520 2521 /* 2522 * Recycle an unused vnode to the front of the free list. 2523 */ 2524 int 2525 vrecycle(struct vnode *vp, struct thread *td) 2526 { 2527 int recycled; 2528 2529 ASSERT_VOP_ELOCKED(vp, "vrecycle"); 2530 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2531 recycled = 0; 2532 VI_LOCK(vp); 2533 if (vp->v_usecount == 0) { 2534 recycled = 1; 2535 vgonel(vp); 2536 } 2537 VI_UNLOCK(vp); 2538 return (recycled); 2539 } 2540 2541 /* 2542 * Eliminate all activity associated with a vnode 2543 * in preparation for reuse. 2544 */ 2545 void 2546 vgone(struct vnode *vp) 2547 { 2548 VI_LOCK(vp); 2549 vgonel(vp); 2550 VI_UNLOCK(vp); 2551 } 2552 2553 /* 2554 * vgone, with the vp interlock held. 2555 */ 2556 void 2557 vgonel(struct vnode *vp) 2558 { 2559 struct thread *td; 2560 int oweinact; 2561 int active; 2562 struct mount *mp; 2563 2564 ASSERT_VOP_ELOCKED(vp, "vgonel"); 2565 ASSERT_VI_LOCKED(vp, "vgonel"); 2566 VNASSERT(vp->v_holdcnt, vp, 2567 ("vgonel: vp %p has no reference.", vp)); 2568 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2569 td = curthread; 2570 2571 /* 2572 * Don't vgonel if we're already doomed. 2573 */ 2574 if (vp->v_iflag & VI_DOOMED) 2575 return; 2576 vp->v_iflag |= VI_DOOMED; 2577 /* 2578 * Check to see if the vnode is in use. If so, we have to call 2579 * VOP_CLOSE() and VOP_INACTIVE(). 2580 */ 2581 active = vp->v_usecount; 2582 oweinact = (vp->v_iflag & VI_OWEINACT); 2583 VI_UNLOCK(vp); 2584 /* 2585 * Clean out any buffers associated with the vnode. 2586 * If the flush fails, just toss the buffers. 2587 */ 2588 mp = NULL; 2589 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) 2590 (void) vn_start_secondary_write(vp, &mp, V_WAIT); 2591 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) 2592 vinvalbuf(vp, 0, 0, 0); 2593 2594 /* 2595 * If purging an active vnode, it must be closed and 2596 * deactivated before being reclaimed. 2597 */ 2598 if (active) 2599 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); 2600 if (oweinact || active) { 2601 VI_LOCK(vp); 2602 if ((vp->v_iflag & VI_DOINGINACT) == 0) 2603 vinactive(vp, td); 2604 VI_UNLOCK(vp); 2605 } 2606 /* 2607 * Reclaim the vnode. 2608 */ 2609 if (VOP_RECLAIM(vp, td)) 2610 panic("vgone: cannot reclaim"); 2611 if (mp != NULL) 2612 vn_finished_secondary_write(mp); 2613 VNASSERT(vp->v_object == NULL, vp, 2614 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag)); 2615 /* 2616 * Clear the advisory locks and wake up waiting threads. 2617 */ 2618 (void)VOP_ADVLOCKPURGE(vp); 2619 /* 2620 * Delete from old mount point vnode list. 2621 */ 2622 delmntque(vp); 2623 cache_purge(vp); 2624 /* 2625 * Done with purge, reset to the standard lock and invalidate 2626 * the vnode. 2627 */ 2628 VI_LOCK(vp); 2629 vp->v_vnlock = &vp->v_lock; 2630 vp->v_op = &dead_vnodeops; 2631 vp->v_tag = "none"; 2632 vp->v_type = VBAD; 2633 } 2634 2635 /* 2636 * Calculate the total number of references to a special device. 2637 */ 2638 int 2639 vcount(struct vnode *vp) 2640 { 2641 int count; 2642 2643 dev_lock(); 2644 count = vp->v_rdev->si_usecount; 2645 dev_unlock(); 2646 return (count); 2647 } 2648 2649 /* 2650 * Same as above, but using the struct cdev *as argument 2651 */ 2652 int 2653 count_dev(struct cdev *dev) 2654 { 2655 int count; 2656 2657 dev_lock(); 2658 count = dev->si_usecount; 2659 dev_unlock(); 2660 return(count); 2661 } 2662 2663 /* 2664 * Print out a description of a vnode. 2665 */ 2666 static char *typename[] = 2667 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", 2668 "VMARKER"}; 2669 2670 void 2671 vn_printf(struct vnode *vp, const char *fmt, ...) 2672 { 2673 va_list ap; 2674 char buf[256], buf2[16]; 2675 u_long flags; 2676 2677 va_start(ap, fmt); 2678 vprintf(fmt, ap); 2679 va_end(ap); 2680 printf("%p: ", (void *)vp); 2681 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]); 2682 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n", 2683 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere); 2684 buf[0] = '\0'; 2685 buf[1] = '\0'; 2686 if (vp->v_vflag & VV_ROOT) 2687 strlcat(buf, "|VV_ROOT", sizeof(buf)); 2688 if (vp->v_vflag & VV_ISTTY) 2689 strlcat(buf, "|VV_ISTTY", sizeof(buf)); 2690 if (vp->v_vflag & VV_NOSYNC) 2691 strlcat(buf, "|VV_NOSYNC", sizeof(buf)); 2692 if (vp->v_vflag & VV_CACHEDLABEL) 2693 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf)); 2694 if (vp->v_vflag & VV_TEXT) 2695 strlcat(buf, "|VV_TEXT", sizeof(buf)); 2696 if (vp->v_vflag & VV_COPYONWRITE) 2697 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf)); 2698 if (vp->v_vflag & VV_SYSTEM) 2699 strlcat(buf, "|VV_SYSTEM", sizeof(buf)); 2700 if (vp->v_vflag & VV_PROCDEP) 2701 strlcat(buf, "|VV_PROCDEP", sizeof(buf)); 2702 if (vp->v_vflag & VV_NOKNOTE) 2703 strlcat(buf, "|VV_NOKNOTE", sizeof(buf)); 2704 if (vp->v_vflag & VV_DELETED) 2705 strlcat(buf, "|VV_DELETED", sizeof(buf)); 2706 if (vp->v_vflag & VV_MD) 2707 strlcat(buf, "|VV_MD", sizeof(buf)); 2708 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | 2709 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP | 2710 VV_NOKNOTE | VV_DELETED | VV_MD); 2711 if (flags != 0) { 2712 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags); 2713 strlcat(buf, buf2, sizeof(buf)); 2714 } 2715 if (vp->v_iflag & VI_MOUNT) 2716 strlcat(buf, "|VI_MOUNT", sizeof(buf)); 2717 if (vp->v_iflag & VI_AGE) 2718 strlcat(buf, "|VI_AGE", sizeof(buf)); 2719 if (vp->v_iflag & VI_DOOMED) 2720 strlcat(buf, "|VI_DOOMED", sizeof(buf)); 2721 if (vp->v_iflag & VI_FREE) 2722 strlcat(buf, "|VI_FREE", sizeof(buf)); 2723 if (vp->v_iflag & VI_DOINGINACT) 2724 strlcat(buf, "|VI_DOINGINACT", sizeof(buf)); 2725 if (vp->v_iflag & VI_OWEINACT) 2726 strlcat(buf, "|VI_OWEINACT", sizeof(buf)); 2727 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE | 2728 VI_DOINGINACT | VI_OWEINACT); 2729 if (flags != 0) { 2730 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags); 2731 strlcat(buf, buf2, sizeof(buf)); 2732 } 2733 printf(" flags (%s)\n", buf + 1); 2734 if (mtx_owned(VI_MTX(vp))) 2735 printf(" VI_LOCKed"); 2736 if (vp->v_object != NULL) 2737 printf(" v_object %p ref %d pages %d\n", 2738 vp->v_object, vp->v_object->ref_count, 2739 vp->v_object->resident_page_count); 2740 printf(" "); 2741 lockmgr_printinfo(vp->v_vnlock); 2742 if (vp->v_data != NULL) 2743 VOP_PRINT(vp); 2744 } 2745 2746 #ifdef DDB 2747 /* 2748 * List all of the locked vnodes in the system. 2749 * Called when debugging the kernel. 2750 */ 2751 DB_SHOW_COMMAND(lockedvnods, lockedvnodes) 2752 { 2753 struct mount *mp, *nmp; 2754 struct vnode *vp; 2755 2756 /* 2757 * Note: because this is DDB, we can't obey the locking semantics 2758 * for these structures, which means we could catch an inconsistent 2759 * state and dereference a nasty pointer. Not much to be done 2760 * about that. 2761 */ 2762 db_printf("Locked vnodes\n"); 2763 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2764 nmp = TAILQ_NEXT(mp, mnt_list); 2765 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2766 if (vp->v_type != VMARKER && 2767 VOP_ISLOCKED(vp)) 2768 vprint("", vp); 2769 } 2770 nmp = TAILQ_NEXT(mp, mnt_list); 2771 } 2772 } 2773 2774 /* 2775 * Show details about the given vnode. 2776 */ 2777 DB_SHOW_COMMAND(vnode, db_show_vnode) 2778 { 2779 struct vnode *vp; 2780 2781 if (!have_addr) 2782 return; 2783 vp = (struct vnode *)addr; 2784 vn_printf(vp, "vnode "); 2785 } 2786 2787 /* 2788 * Show details about the given mount point. 2789 */ 2790 DB_SHOW_COMMAND(mount, db_show_mount) 2791 { 2792 struct mount *mp; 2793 struct vfsopt *opt; 2794 struct statfs *sp; 2795 struct vnode *vp; 2796 char buf[512]; 2797 u_int flags; 2798 2799 if (!have_addr) { 2800 /* No address given, print short info about all mount points. */ 2801 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2802 db_printf("%p %s on %s (%s)\n", mp, 2803 mp->mnt_stat.f_mntfromname, 2804 mp->mnt_stat.f_mntonname, 2805 mp->mnt_stat.f_fstypename); 2806 if (db_pager_quit) 2807 break; 2808 } 2809 db_printf("\nMore info: show mount <addr>\n"); 2810 return; 2811 } 2812 2813 mp = (struct mount *)addr; 2814 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, 2815 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); 2816 2817 buf[0] = '\0'; 2818 flags = mp->mnt_flag; 2819 #define MNT_FLAG(flag) do { \ 2820 if (flags & (flag)) { \ 2821 if (buf[0] != '\0') \ 2822 strlcat(buf, ", ", sizeof(buf)); \ 2823 strlcat(buf, (#flag) + 4, sizeof(buf)); \ 2824 flags &= ~(flag); \ 2825 } \ 2826 } while (0) 2827 MNT_FLAG(MNT_RDONLY); 2828 MNT_FLAG(MNT_SYNCHRONOUS); 2829 MNT_FLAG(MNT_NOEXEC); 2830 MNT_FLAG(MNT_NOSUID); 2831 MNT_FLAG(MNT_UNION); 2832 MNT_FLAG(MNT_ASYNC); 2833 MNT_FLAG(MNT_SUIDDIR); 2834 MNT_FLAG(MNT_SOFTDEP); 2835 MNT_FLAG(MNT_NOSYMFOLLOW); 2836 MNT_FLAG(MNT_GJOURNAL); 2837 MNT_FLAG(MNT_MULTILABEL); 2838 MNT_FLAG(MNT_ACLS); 2839 MNT_FLAG(MNT_NOATIME); 2840 MNT_FLAG(MNT_NOCLUSTERR); 2841 MNT_FLAG(MNT_NOCLUSTERW); 2842 MNT_FLAG(MNT_NFS4ACLS); 2843 MNT_FLAG(MNT_EXRDONLY); 2844 MNT_FLAG(MNT_EXPORTED); 2845 MNT_FLAG(MNT_DEFEXPORTED); 2846 MNT_FLAG(MNT_EXPORTANON); 2847 MNT_FLAG(MNT_EXKERB); 2848 MNT_FLAG(MNT_EXPUBLIC); 2849 MNT_FLAG(MNT_LOCAL); 2850 MNT_FLAG(MNT_QUOTA); 2851 MNT_FLAG(MNT_ROOTFS); 2852 MNT_FLAG(MNT_USER); 2853 MNT_FLAG(MNT_IGNORE); 2854 MNT_FLAG(MNT_UPDATE); 2855 MNT_FLAG(MNT_DELEXPORT); 2856 MNT_FLAG(MNT_RELOAD); 2857 MNT_FLAG(MNT_FORCE); 2858 MNT_FLAG(MNT_SNAPSHOT); 2859 MNT_FLAG(MNT_BYFSID); 2860 MNT_FLAG(MNT_SOFTDEP); 2861 #undef MNT_FLAG 2862 if (flags != 0) { 2863 if (buf[0] != '\0') 2864 strlcat(buf, ", ", sizeof(buf)); 2865 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), 2866 "0x%08x", flags); 2867 } 2868 db_printf(" mnt_flag = %s\n", buf); 2869 2870 buf[0] = '\0'; 2871 flags = mp->mnt_kern_flag; 2872 #define MNT_KERN_FLAG(flag) do { \ 2873 if (flags & (flag)) { \ 2874 if (buf[0] != '\0') \ 2875 strlcat(buf, ", ", sizeof(buf)); \ 2876 strlcat(buf, (#flag) + 5, sizeof(buf)); \ 2877 flags &= ~(flag); \ 2878 } \ 2879 } while (0) 2880 MNT_KERN_FLAG(MNTK_UNMOUNTF); 2881 MNT_KERN_FLAG(MNTK_ASYNC); 2882 MNT_KERN_FLAG(MNTK_SOFTDEP); 2883 MNT_KERN_FLAG(MNTK_NOINSMNTQ); 2884 MNT_KERN_FLAG(MNTK_DRAINING); 2885 MNT_KERN_FLAG(MNTK_REFEXPIRE); 2886 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED); 2887 MNT_KERN_FLAG(MNTK_SHARED_WRITES); 2888 MNT_KERN_FLAG(MNTK_SUJ); 2889 MNT_KERN_FLAG(MNTK_UNMOUNT); 2890 MNT_KERN_FLAG(MNTK_MWAIT); 2891 MNT_KERN_FLAG(MNTK_SUSPEND); 2892 MNT_KERN_FLAG(MNTK_SUSPEND2); 2893 MNT_KERN_FLAG(MNTK_SUSPENDED); 2894 MNT_KERN_FLAG(MNTK_MPSAFE); 2895 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED); 2896 MNT_KERN_FLAG(MNTK_NOKNOTE); 2897 #undef MNT_KERN_FLAG 2898 if (flags != 0) { 2899 if (buf[0] != '\0') 2900 strlcat(buf, ", ", sizeof(buf)); 2901 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), 2902 "0x%08x", flags); 2903 } 2904 db_printf(" mnt_kern_flag = %s\n", buf); 2905 2906 db_printf(" mnt_opt = "); 2907 opt = TAILQ_FIRST(mp->mnt_opt); 2908 if (opt != NULL) { 2909 db_printf("%s", opt->name); 2910 opt = TAILQ_NEXT(opt, link); 2911 while (opt != NULL) { 2912 db_printf(", %s", opt->name); 2913 opt = TAILQ_NEXT(opt, link); 2914 } 2915 } 2916 db_printf("\n"); 2917 2918 sp = &mp->mnt_stat; 2919 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx " 2920 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju " 2921 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju " 2922 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n", 2923 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags, 2924 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize, 2925 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree, 2926 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files, 2927 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites, 2928 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads, 2929 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax, 2930 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]); 2931 2932 db_printf(" mnt_cred = { uid=%u ruid=%u", 2933 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid); 2934 if (jailed(mp->mnt_cred)) 2935 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id); 2936 db_printf(" }\n"); 2937 db_printf(" mnt_ref = %d\n", mp->mnt_ref); 2938 db_printf(" mnt_gen = %d\n", mp->mnt_gen); 2939 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize); 2940 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount); 2941 db_printf(" mnt_noasync = %u\n", mp->mnt_noasync); 2942 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen); 2943 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max); 2944 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed); 2945 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes); 2946 db_printf(" mnt_secondary_accwrites = %d\n", 2947 mp->mnt_secondary_accwrites); 2948 db_printf(" mnt_gjprovider = %s\n", 2949 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL"); 2950 db_printf("\n"); 2951 2952 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2953 if (vp->v_type != VMARKER) { 2954 vn_printf(vp, "vnode "); 2955 if (db_pager_quit) 2956 break; 2957 } 2958 } 2959 } 2960 #endif /* DDB */ 2961 2962 /* 2963 * Fill in a struct xvfsconf based on a struct vfsconf. 2964 */ 2965 static void 2966 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp) 2967 { 2968 2969 strcpy(xvfsp->vfc_name, vfsp->vfc_name); 2970 xvfsp->vfc_typenum = vfsp->vfc_typenum; 2971 xvfsp->vfc_refcount = vfsp->vfc_refcount; 2972 xvfsp->vfc_flags = vfsp->vfc_flags; 2973 /* 2974 * These are unused in userland, we keep them 2975 * to not break binary compatibility. 2976 */ 2977 xvfsp->vfc_vfsops = NULL; 2978 xvfsp->vfc_next = NULL; 2979 } 2980 2981 /* 2982 * Top level filesystem related information gathering. 2983 */ 2984 static int 2985 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) 2986 { 2987 struct vfsconf *vfsp; 2988 struct xvfsconf xvfsp; 2989 int error; 2990 2991 error = 0; 2992 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 2993 bzero(&xvfsp, sizeof(xvfsp)); 2994 vfsconf2x(vfsp, &xvfsp); 2995 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp); 2996 if (error) 2997 break; 2998 } 2999 return (error); 3000 } 3001 3002 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD, 3003 NULL, 0, sysctl_vfs_conflist, 3004 "S,xvfsconf", "List of all configured filesystems"); 3005 3006 #ifndef BURN_BRIDGES 3007 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); 3008 3009 static int 3010 vfs_sysctl(SYSCTL_HANDLER_ARGS) 3011 { 3012 int *name = (int *)arg1 - 1; /* XXX */ 3013 u_int namelen = arg2 + 1; /* XXX */ 3014 struct vfsconf *vfsp; 3015 struct xvfsconf xvfsp; 3016 3017 printf("WARNING: userland calling deprecated sysctl, " 3018 "please rebuild world\n"); 3019 3020 #if 1 || defined(COMPAT_PRELITE2) 3021 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 3022 if (namelen == 1) 3023 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 3024 #endif 3025 3026 switch (name[1]) { 3027 case VFS_MAXTYPENUM: 3028 if (namelen != 2) 3029 return (ENOTDIR); 3030 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 3031 case VFS_CONF: 3032 if (namelen != 3) 3033 return (ENOTDIR); /* overloaded */ 3034 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) 3035 if (vfsp->vfc_typenum == name[2]) 3036 break; 3037 if (vfsp == NULL) 3038 return (EOPNOTSUPP); 3039 bzero(&xvfsp, sizeof(xvfsp)); 3040 vfsconf2x(vfsp, &xvfsp); 3041 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); 3042 } 3043 return (EOPNOTSUPP); 3044 } 3045 3046 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, 3047 vfs_sysctl, "Generic filesystem"); 3048 3049 #if 1 || defined(COMPAT_PRELITE2) 3050 3051 static int 3052 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 3053 { 3054 int error; 3055 struct vfsconf *vfsp; 3056 struct ovfsconf ovfs; 3057 3058 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 3059 bzero(&ovfs, sizeof(ovfs)); 3060 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 3061 strcpy(ovfs.vfc_name, vfsp->vfc_name); 3062 ovfs.vfc_index = vfsp->vfc_typenum; 3063 ovfs.vfc_refcount = vfsp->vfc_refcount; 3064 ovfs.vfc_flags = vfsp->vfc_flags; 3065 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 3066 if (error) 3067 return error; 3068 } 3069 return 0; 3070 } 3071 3072 #endif /* 1 || COMPAT_PRELITE2 */ 3073 #endif /* !BURN_BRIDGES */ 3074 3075 #define KINFO_VNODESLOP 10 3076 #ifdef notyet 3077 /* 3078 * Dump vnode list (via sysctl). 3079 */ 3080 /* ARGSUSED */ 3081 static int 3082 sysctl_vnode(SYSCTL_HANDLER_ARGS) 3083 { 3084 struct xvnode *xvn; 3085 struct mount *mp; 3086 struct vnode *vp; 3087 int error, len, n; 3088 3089 /* 3090 * Stale numvnodes access is not fatal here. 3091 */ 3092 req->lock = 0; 3093 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; 3094 if (!req->oldptr) 3095 /* Make an estimate */ 3096 return (SYSCTL_OUT(req, 0, len)); 3097 3098 error = sysctl_wire_old_buffer(req, 0); 3099 if (error != 0) 3100 return (error); 3101 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); 3102 n = 0; 3103 mtx_lock(&mountlist_mtx); 3104 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 3105 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) 3106 continue; 3107 MNT_ILOCK(mp); 3108 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 3109 if (n == len) 3110 break; 3111 vref(vp); 3112 xvn[n].xv_size = sizeof *xvn; 3113 xvn[n].xv_vnode = vp; 3114 xvn[n].xv_id = 0; /* XXX compat */ 3115 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field 3116 XV_COPY(usecount); 3117 XV_COPY(writecount); 3118 XV_COPY(holdcnt); 3119 XV_COPY(mount); 3120 XV_COPY(numoutput); 3121 XV_COPY(type); 3122 #undef XV_COPY 3123 xvn[n].xv_flag = vp->v_vflag; 3124 3125 switch (vp->v_type) { 3126 case VREG: 3127 case VDIR: 3128 case VLNK: 3129 break; 3130 case VBLK: 3131 case VCHR: 3132 if (vp->v_rdev == NULL) { 3133 vrele(vp); 3134 continue; 3135 } 3136 xvn[n].xv_dev = dev2udev(vp->v_rdev); 3137 break; 3138 case VSOCK: 3139 xvn[n].xv_socket = vp->v_socket; 3140 break; 3141 case VFIFO: 3142 xvn[n].xv_fifo = vp->v_fifoinfo; 3143 break; 3144 case VNON: 3145 case VBAD: 3146 default: 3147 /* shouldn't happen? */ 3148 vrele(vp); 3149 continue; 3150 } 3151 vrele(vp); 3152 ++n; 3153 } 3154 MNT_IUNLOCK(mp); 3155 mtx_lock(&mountlist_mtx); 3156 vfs_unbusy(mp); 3157 if (n == len) 3158 break; 3159 } 3160 mtx_unlock(&mountlist_mtx); 3161 3162 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); 3163 free(xvn, M_TEMP); 3164 return (error); 3165 } 3166 3167 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 3168 0, 0, sysctl_vnode, "S,xvnode", ""); 3169 #endif 3170 3171 /* 3172 * Unmount all filesystems. The list is traversed in reverse order 3173 * of mounting to avoid dependencies. 3174 */ 3175 void 3176 vfs_unmountall(void) 3177 { 3178 struct mount *mp; 3179 struct thread *td; 3180 int error; 3181 3182 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread")); 3183 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__); 3184 td = curthread; 3185 3186 /* 3187 * Since this only runs when rebooting, it is not interlocked. 3188 */ 3189 while(!TAILQ_EMPTY(&mountlist)) { 3190 mp = TAILQ_LAST(&mountlist, mntlist); 3191 error = dounmount(mp, MNT_FORCE, td); 3192 if (error) { 3193 TAILQ_REMOVE(&mountlist, mp, mnt_list); 3194 /* 3195 * XXX: Due to the way in which we mount the root 3196 * file system off of devfs, devfs will generate a 3197 * "busy" warning when we try to unmount it before 3198 * the root. Don't print a warning as a result in 3199 * order to avoid false positive errors that may 3200 * cause needless upset. 3201 */ 3202 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) { 3203 printf("unmount of %s failed (", 3204 mp->mnt_stat.f_mntonname); 3205 if (error == EBUSY) 3206 printf("BUSY)\n"); 3207 else 3208 printf("%d)\n", error); 3209 } 3210 } else { 3211 /* The unmount has removed mp from the mountlist */ 3212 } 3213 } 3214 } 3215 3216 /* 3217 * perform msync on all vnodes under a mount point 3218 * the mount point must be locked. 3219 */ 3220 void 3221 vfs_msync(struct mount *mp, int flags) 3222 { 3223 struct vnode *vp, *mvp; 3224 struct vm_object *obj; 3225 3226 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 3227 MNT_ILOCK(mp); 3228 MNT_VNODE_FOREACH(vp, mp, mvp) { 3229 VI_LOCK(vp); 3230 obj = vp->v_object; 3231 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 && 3232 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) { 3233 MNT_IUNLOCK(mp); 3234 if (!vget(vp, 3235 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, 3236 curthread)) { 3237 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */ 3238 vput(vp); 3239 MNT_ILOCK(mp); 3240 continue; 3241 } 3242 3243 obj = vp->v_object; 3244 if (obj != NULL) { 3245 VM_OBJECT_LOCK(obj); 3246 vm_object_page_clean(obj, 0, 0, 3247 flags == MNT_WAIT ? 3248 OBJPC_SYNC : OBJPC_NOSYNC); 3249 VM_OBJECT_UNLOCK(obj); 3250 } 3251 vput(vp); 3252 } 3253 MNT_ILOCK(mp); 3254 } else 3255 VI_UNLOCK(vp); 3256 } 3257 MNT_IUNLOCK(mp); 3258 } 3259 3260 /* 3261 * Mark a vnode as free, putting it up for recycling. 3262 */ 3263 static void 3264 vfree(struct vnode *vp) 3265 { 3266 3267 ASSERT_VI_LOCKED(vp, "vfree"); 3268 mtx_lock(&vnode_free_list_mtx); 3269 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed.")); 3270 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free")); 3271 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't")); 3272 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp, 3273 ("vfree: Freeing doomed vnode")); 3274 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 3275 if (vp->v_iflag & VI_AGE) { 3276 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 3277 } else { 3278 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 3279 } 3280 freevnodes++; 3281 vp->v_iflag &= ~VI_AGE; 3282 vp->v_iflag |= VI_FREE; 3283 mtx_unlock(&vnode_free_list_mtx); 3284 } 3285 3286 /* 3287 * Opposite of vfree() - mark a vnode as in use. 3288 */ 3289 static void 3290 vbusy(struct vnode *vp) 3291 { 3292 ASSERT_VI_LOCKED(vp, "vbusy"); 3293 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free")); 3294 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed.")); 3295 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 3296 3297 mtx_lock(&vnode_free_list_mtx); 3298 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 3299 freevnodes--; 3300 vp->v_iflag &= ~(VI_FREE|VI_AGE); 3301 mtx_unlock(&vnode_free_list_mtx); 3302 } 3303 3304 static void 3305 destroy_vpollinfo(struct vpollinfo *vi) 3306 { 3307 knlist_destroy(&vi->vpi_selinfo.si_note); 3308 mtx_destroy(&vi->vpi_lock); 3309 uma_zfree(vnodepoll_zone, vi); 3310 } 3311 3312 /* 3313 * Initalize per-vnode helper structure to hold poll-related state. 3314 */ 3315 void 3316 v_addpollinfo(struct vnode *vp) 3317 { 3318 struct vpollinfo *vi; 3319 3320 if (vp->v_pollinfo != NULL) 3321 return; 3322 vi = uma_zalloc(vnodepoll_zone, M_WAITOK); 3323 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); 3324 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock, 3325 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked); 3326 VI_LOCK(vp); 3327 if (vp->v_pollinfo != NULL) { 3328 VI_UNLOCK(vp); 3329 destroy_vpollinfo(vi); 3330 return; 3331 } 3332 vp->v_pollinfo = vi; 3333 VI_UNLOCK(vp); 3334 } 3335 3336 /* 3337 * Record a process's interest in events which might happen to 3338 * a vnode. Because poll uses the historic select-style interface 3339 * internally, this routine serves as both the ``check for any 3340 * pending events'' and the ``record my interest in future events'' 3341 * functions. (These are done together, while the lock is held, 3342 * to avoid race conditions.) 3343 */ 3344 int 3345 vn_pollrecord(struct vnode *vp, struct thread *td, int events) 3346 { 3347 3348 v_addpollinfo(vp); 3349 mtx_lock(&vp->v_pollinfo->vpi_lock); 3350 if (vp->v_pollinfo->vpi_revents & events) { 3351 /* 3352 * This leaves events we are not interested 3353 * in available for the other process which 3354 * which presumably had requested them 3355 * (otherwise they would never have been 3356 * recorded). 3357 */ 3358 events &= vp->v_pollinfo->vpi_revents; 3359 vp->v_pollinfo->vpi_revents &= ~events; 3360 3361 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3362 return (events); 3363 } 3364 vp->v_pollinfo->vpi_events |= events; 3365 selrecord(td, &vp->v_pollinfo->vpi_selinfo); 3366 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3367 return (0); 3368 } 3369 3370 /* 3371 * Routine to create and manage a filesystem syncer vnode. 3372 */ 3373 #define sync_close ((int (*)(struct vop_close_args *))nullop) 3374 static int sync_fsync(struct vop_fsync_args *); 3375 static int sync_inactive(struct vop_inactive_args *); 3376 static int sync_reclaim(struct vop_reclaim_args *); 3377 3378 static struct vop_vector sync_vnodeops = { 3379 .vop_bypass = VOP_EOPNOTSUPP, 3380 .vop_close = sync_close, /* close */ 3381 .vop_fsync = sync_fsync, /* fsync */ 3382 .vop_inactive = sync_inactive, /* inactive */ 3383 .vop_reclaim = sync_reclaim, /* reclaim */ 3384 .vop_lock1 = vop_stdlock, /* lock */ 3385 .vop_unlock = vop_stdunlock, /* unlock */ 3386 .vop_islocked = vop_stdislocked, /* islocked */ 3387 }; 3388 3389 /* 3390 * Create a new filesystem syncer vnode for the specified mount point. 3391 */ 3392 void 3393 vfs_allocate_syncvnode(struct mount *mp) 3394 { 3395 struct vnode *vp; 3396 struct bufobj *bo; 3397 static long start, incr, next; 3398 int error; 3399 3400 /* Allocate a new vnode */ 3401 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp); 3402 if (error != 0) 3403 panic("vfs_allocate_syncvnode: getnewvnode() failed"); 3404 vp->v_type = VNON; 3405 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3406 vp->v_vflag |= VV_FORCEINSMQ; 3407 error = insmntque(vp, mp); 3408 if (error != 0) 3409 panic("vfs_allocate_syncvnode: insmntque() failed"); 3410 vp->v_vflag &= ~VV_FORCEINSMQ; 3411 VOP_UNLOCK(vp, 0); 3412 /* 3413 * Place the vnode onto the syncer worklist. We attempt to 3414 * scatter them about on the list so that they will go off 3415 * at evenly distributed times even if all the filesystems 3416 * are mounted at once. 3417 */ 3418 next += incr; 3419 if (next == 0 || next > syncer_maxdelay) { 3420 start /= 2; 3421 incr /= 2; 3422 if (start == 0) { 3423 start = syncer_maxdelay / 2; 3424 incr = syncer_maxdelay; 3425 } 3426 next = start; 3427 } 3428 bo = &vp->v_bufobj; 3429 BO_LOCK(bo); 3430 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0); 3431 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ 3432 mtx_lock(&sync_mtx); 3433 sync_vnode_count++; 3434 if (mp->mnt_syncer == NULL) { 3435 mp->mnt_syncer = vp; 3436 vp = NULL; 3437 } 3438 mtx_unlock(&sync_mtx); 3439 BO_UNLOCK(bo); 3440 if (vp != NULL) { 3441 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3442 vgone(vp); 3443 vput(vp); 3444 } 3445 } 3446 3447 void 3448 vfs_deallocate_syncvnode(struct mount *mp) 3449 { 3450 struct vnode *vp; 3451 3452 mtx_lock(&sync_mtx); 3453 vp = mp->mnt_syncer; 3454 if (vp != NULL) 3455 mp->mnt_syncer = NULL; 3456 mtx_unlock(&sync_mtx); 3457 if (vp != NULL) 3458 vrele(vp); 3459 } 3460 3461 /* 3462 * Do a lazy sync of the filesystem. 3463 */ 3464 static int 3465 sync_fsync(struct vop_fsync_args *ap) 3466 { 3467 struct vnode *syncvp = ap->a_vp; 3468 struct mount *mp = syncvp->v_mount; 3469 int error; 3470 struct bufobj *bo; 3471 3472 /* 3473 * We only need to do something if this is a lazy evaluation. 3474 */ 3475 if (ap->a_waitfor != MNT_LAZY) 3476 return (0); 3477 3478 /* 3479 * Move ourselves to the back of the sync list. 3480 */ 3481 bo = &syncvp->v_bufobj; 3482 BO_LOCK(bo); 3483 vn_syncer_add_to_worklist(bo, syncdelay); 3484 BO_UNLOCK(bo); 3485 3486 /* 3487 * Walk the list of vnodes pushing all that are dirty and 3488 * not already on the sync list. 3489 */ 3490 mtx_lock(&mountlist_mtx); 3491 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) { 3492 mtx_unlock(&mountlist_mtx); 3493 return (0); 3494 } 3495 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { 3496 vfs_unbusy(mp); 3497 return (0); 3498 } 3499 MNT_ILOCK(mp); 3500 mp->mnt_noasync++; 3501 mp->mnt_kern_flag &= ~MNTK_ASYNC; 3502 MNT_IUNLOCK(mp); 3503 vfs_msync(mp, MNT_NOWAIT); 3504 error = VFS_SYNC(mp, MNT_LAZY); 3505 MNT_ILOCK(mp); 3506 mp->mnt_noasync--; 3507 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 3508 mp->mnt_kern_flag |= MNTK_ASYNC; 3509 MNT_IUNLOCK(mp); 3510 vn_finished_write(mp); 3511 vfs_unbusy(mp); 3512 return (error); 3513 } 3514 3515 /* 3516 * The syncer vnode is no referenced. 3517 */ 3518 static int 3519 sync_inactive(struct vop_inactive_args *ap) 3520 { 3521 3522 vgone(ap->a_vp); 3523 return (0); 3524 } 3525 3526 /* 3527 * The syncer vnode is no longer needed and is being decommissioned. 3528 * 3529 * Modifications to the worklist must be protected by sync_mtx. 3530 */ 3531 static int 3532 sync_reclaim(struct vop_reclaim_args *ap) 3533 { 3534 struct vnode *vp = ap->a_vp; 3535 struct bufobj *bo; 3536 3537 bo = &vp->v_bufobj; 3538 BO_LOCK(bo); 3539 mtx_lock(&sync_mtx); 3540 if (vp->v_mount->mnt_syncer == vp) 3541 vp->v_mount->mnt_syncer = NULL; 3542 if (bo->bo_flag & BO_ONWORKLST) { 3543 LIST_REMOVE(bo, bo_synclist); 3544 syncer_worklist_len--; 3545 sync_vnode_count--; 3546 bo->bo_flag &= ~BO_ONWORKLST; 3547 } 3548 mtx_unlock(&sync_mtx); 3549 BO_UNLOCK(bo); 3550 3551 return (0); 3552 } 3553 3554 /* 3555 * Check if vnode represents a disk device 3556 */ 3557 int 3558 vn_isdisk(struct vnode *vp, int *errp) 3559 { 3560 int error; 3561 3562 error = 0; 3563 dev_lock(); 3564 if (vp->v_type != VCHR) 3565 error = ENOTBLK; 3566 else if (vp->v_rdev == NULL) 3567 error = ENXIO; 3568 else if (vp->v_rdev->si_devsw == NULL) 3569 error = ENXIO; 3570 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) 3571 error = ENOTBLK; 3572 dev_unlock(); 3573 if (errp != NULL) 3574 *errp = error; 3575 return (error == 0); 3576 } 3577 3578 /* 3579 * Common filesystem object access control check routine. Accepts a 3580 * vnode's type, "mode", uid and gid, requested access mode, credentials, 3581 * and optional call-by-reference privused argument allowing vaccess() 3582 * to indicate to the caller whether privilege was used to satisfy the 3583 * request (obsoleted). Returns 0 on success, or an errno on failure. 3584 * 3585 * The ifdef'd CAPABILITIES version is here for reference, but is not 3586 * actually used. 3587 */ 3588 int 3589 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, 3590 accmode_t accmode, struct ucred *cred, int *privused) 3591 { 3592 accmode_t dac_granted; 3593 accmode_t priv_granted; 3594 3595 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0, 3596 ("invalid bit in accmode")); 3597 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE), 3598 ("VAPPEND without VWRITE")); 3599 3600 /* 3601 * Look for a normal, non-privileged way to access the file/directory 3602 * as requested. If it exists, go with that. 3603 */ 3604 3605 if (privused != NULL) 3606 *privused = 0; 3607 3608 dac_granted = 0; 3609 3610 /* Check the owner. */ 3611 if (cred->cr_uid == file_uid) { 3612 dac_granted |= VADMIN; 3613 if (file_mode & S_IXUSR) 3614 dac_granted |= VEXEC; 3615 if (file_mode & S_IRUSR) 3616 dac_granted |= VREAD; 3617 if (file_mode & S_IWUSR) 3618 dac_granted |= (VWRITE | VAPPEND); 3619 3620 if ((accmode & dac_granted) == accmode) 3621 return (0); 3622 3623 goto privcheck; 3624 } 3625 3626 /* Otherwise, check the groups (first match) */ 3627 if (groupmember(file_gid, cred)) { 3628 if (file_mode & S_IXGRP) 3629 dac_granted |= VEXEC; 3630 if (file_mode & S_IRGRP) 3631 dac_granted |= VREAD; 3632 if (file_mode & S_IWGRP) 3633 dac_granted |= (VWRITE | VAPPEND); 3634 3635 if ((accmode & dac_granted) == accmode) 3636 return (0); 3637 3638 goto privcheck; 3639 } 3640 3641 /* Otherwise, check everyone else. */ 3642 if (file_mode & S_IXOTH) 3643 dac_granted |= VEXEC; 3644 if (file_mode & S_IROTH) 3645 dac_granted |= VREAD; 3646 if (file_mode & S_IWOTH) 3647 dac_granted |= (VWRITE | VAPPEND); 3648 if ((accmode & dac_granted) == accmode) 3649 return (0); 3650 3651 privcheck: 3652 /* 3653 * Build a privilege mask to determine if the set of privileges 3654 * satisfies the requirements when combined with the granted mask 3655 * from above. For each privilege, if the privilege is required, 3656 * bitwise or the request type onto the priv_granted mask. 3657 */ 3658 priv_granted = 0; 3659 3660 if (type == VDIR) { 3661 /* 3662 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC 3663 * requests, instead of PRIV_VFS_EXEC. 3664 */ 3665 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3666 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0)) 3667 priv_granted |= VEXEC; 3668 } else { 3669 /* 3670 * Ensure that at least one execute bit is on. Otherwise, 3671 * a privileged user will always succeed, and we don't want 3672 * this to happen unless the file really is executable. 3673 */ 3674 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3675 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 && 3676 !priv_check_cred(cred, PRIV_VFS_EXEC, 0)) 3677 priv_granted |= VEXEC; 3678 } 3679 3680 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) && 3681 !priv_check_cred(cred, PRIV_VFS_READ, 0)) 3682 priv_granted |= VREAD; 3683 3684 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) && 3685 !priv_check_cred(cred, PRIV_VFS_WRITE, 0)) 3686 priv_granted |= (VWRITE | VAPPEND); 3687 3688 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) && 3689 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0)) 3690 priv_granted |= VADMIN; 3691 3692 if ((accmode & (priv_granted | dac_granted)) == accmode) { 3693 /* XXX audit: privilege used */ 3694 if (privused != NULL) 3695 *privused = 1; 3696 return (0); 3697 } 3698 3699 return ((accmode & VADMIN) ? EPERM : EACCES); 3700 } 3701 3702 /* 3703 * Credential check based on process requesting service, and per-attribute 3704 * permissions. 3705 */ 3706 int 3707 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, 3708 struct thread *td, accmode_t accmode) 3709 { 3710 3711 /* 3712 * Kernel-invoked always succeeds. 3713 */ 3714 if (cred == NOCRED) 3715 return (0); 3716 3717 /* 3718 * Do not allow privileged processes in jail to directly manipulate 3719 * system attributes. 3720 */ 3721 switch (attrnamespace) { 3722 case EXTATTR_NAMESPACE_SYSTEM: 3723 /* Potentially should be: return (EPERM); */ 3724 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0)); 3725 case EXTATTR_NAMESPACE_USER: 3726 return (VOP_ACCESS(vp, accmode, cred, td)); 3727 default: 3728 return (EPERM); 3729 } 3730 } 3731 3732 #ifdef DEBUG_VFS_LOCKS 3733 /* 3734 * This only exists to supress warnings from unlocked specfs accesses. It is 3735 * no longer ok to have an unlocked VFS. 3736 */ 3737 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \ 3738 (vp)->v_type == VCHR || (vp)->v_type == VBAD) 3739 3740 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ 3741 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, 3742 "Drop into debugger on lock violation"); 3743 3744 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ 3745 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 3746 0, "Check for interlock across VOPs"); 3747 3748 int vfs_badlock_print = 1; /* Print lock violations. */ 3749 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 3750 0, "Print lock violations"); 3751 3752 #ifdef KDB 3753 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ 3754 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, 3755 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations"); 3756 #endif 3757 3758 static void 3759 vfs_badlock(const char *msg, const char *str, struct vnode *vp) 3760 { 3761 3762 #ifdef KDB 3763 if (vfs_badlock_backtrace) 3764 kdb_backtrace(); 3765 #endif 3766 if (vfs_badlock_print) 3767 printf("%s: %p %s\n", str, (void *)vp, msg); 3768 if (vfs_badlock_ddb) 3769 kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); 3770 } 3771 3772 void 3773 assert_vi_locked(struct vnode *vp, const char *str) 3774 { 3775 3776 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) 3777 vfs_badlock("interlock is not locked but should be", str, vp); 3778 } 3779 3780 void 3781 assert_vi_unlocked(struct vnode *vp, const char *str) 3782 { 3783 3784 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) 3785 vfs_badlock("interlock is locked but should not be", str, vp); 3786 } 3787 3788 void 3789 assert_vop_locked(struct vnode *vp, const char *str) 3790 { 3791 3792 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0) 3793 vfs_badlock("is not locked but should be", str, vp); 3794 } 3795 3796 void 3797 assert_vop_unlocked(struct vnode *vp, const char *str) 3798 { 3799 3800 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) 3801 vfs_badlock("is locked but should not be", str, vp); 3802 } 3803 3804 void 3805 assert_vop_elocked(struct vnode *vp, const char *str) 3806 { 3807 3808 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 3809 vfs_badlock("is not exclusive locked but should be", str, vp); 3810 } 3811 3812 #if 0 3813 void 3814 assert_vop_elocked_other(struct vnode *vp, const char *str) 3815 { 3816 3817 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER) 3818 vfs_badlock("is not exclusive locked by another thread", 3819 str, vp); 3820 } 3821 3822 void 3823 assert_vop_slocked(struct vnode *vp, const char *str) 3824 { 3825 3826 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED) 3827 vfs_badlock("is not locked shared but should be", str, vp); 3828 } 3829 #endif /* 0 */ 3830 #endif /* DEBUG_VFS_LOCKS */ 3831 3832 void 3833 vop_rename_fail(struct vop_rename_args *ap) 3834 { 3835 3836 if (ap->a_tvp != NULL) 3837 vput(ap->a_tvp); 3838 if (ap->a_tdvp == ap->a_tvp) 3839 vrele(ap->a_tdvp); 3840 else 3841 vput(ap->a_tdvp); 3842 vrele(ap->a_fdvp); 3843 vrele(ap->a_fvp); 3844 } 3845 3846 void 3847 vop_rename_pre(void *ap) 3848 { 3849 struct vop_rename_args *a = ap; 3850 3851 #ifdef DEBUG_VFS_LOCKS 3852 if (a->a_tvp) 3853 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); 3854 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); 3855 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); 3856 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); 3857 3858 /* Check the source (from). */ 3859 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock && 3860 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock)) 3861 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); 3862 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock) 3863 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); 3864 3865 /* Check the target. */ 3866 if (a->a_tvp) 3867 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); 3868 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); 3869 #endif 3870 if (a->a_tdvp != a->a_fdvp) 3871 vhold(a->a_fdvp); 3872 if (a->a_tvp != a->a_fvp) 3873 vhold(a->a_fvp); 3874 vhold(a->a_tdvp); 3875 if (a->a_tvp) 3876 vhold(a->a_tvp); 3877 } 3878 3879 void 3880 vop_strategy_pre(void *ap) 3881 { 3882 #ifdef DEBUG_VFS_LOCKS 3883 struct vop_strategy_args *a; 3884 struct buf *bp; 3885 3886 a = ap; 3887 bp = a->a_bp; 3888 3889 /* 3890 * Cluster ops lock their component buffers but not the IO container. 3891 */ 3892 if ((bp->b_flags & B_CLUSTER) != 0) 3893 return; 3894 3895 if (panicstr == NULL && !BUF_ISLOCKED(bp)) { 3896 if (vfs_badlock_print) 3897 printf( 3898 "VOP_STRATEGY: bp is not locked but should be\n"); 3899 if (vfs_badlock_ddb) 3900 kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); 3901 } 3902 #endif 3903 } 3904 3905 void 3906 vop_lookup_pre(void *ap) 3907 { 3908 #ifdef DEBUG_VFS_LOCKS 3909 struct vop_lookup_args *a; 3910 struct vnode *dvp; 3911 3912 a = ap; 3913 dvp = a->a_dvp; 3914 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 3915 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 3916 #endif 3917 } 3918 3919 void 3920 vop_lookup_post(void *ap, int rc) 3921 { 3922 #ifdef DEBUG_VFS_LOCKS 3923 struct vop_lookup_args *a; 3924 struct vnode *dvp; 3925 struct vnode *vp; 3926 3927 a = ap; 3928 dvp = a->a_dvp; 3929 vp = *(a->a_vpp); 3930 3931 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 3932 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 3933 3934 if (!rc) 3935 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)"); 3936 #endif 3937 } 3938 3939 void 3940 vop_lock_pre(void *ap) 3941 { 3942 #ifdef DEBUG_VFS_LOCKS 3943 struct vop_lock1_args *a = ap; 3944 3945 if ((a->a_flags & LK_INTERLOCK) == 0) 3946 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 3947 else 3948 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); 3949 #endif 3950 } 3951 3952 void 3953 vop_lock_post(void *ap, int rc) 3954 { 3955 #ifdef DEBUG_VFS_LOCKS 3956 struct vop_lock1_args *a = ap; 3957 3958 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 3959 if (rc == 0) 3960 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); 3961 #endif 3962 } 3963 3964 void 3965 vop_unlock_pre(void *ap) 3966 { 3967 #ifdef DEBUG_VFS_LOCKS 3968 struct vop_unlock_args *a = ap; 3969 3970 if (a->a_flags & LK_INTERLOCK) 3971 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK"); 3972 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); 3973 #endif 3974 } 3975 3976 void 3977 vop_unlock_post(void *ap, int rc) 3978 { 3979 #ifdef DEBUG_VFS_LOCKS 3980 struct vop_unlock_args *a = ap; 3981 3982 if (a->a_flags & LK_INTERLOCK) 3983 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK"); 3984 #endif 3985 } 3986 3987 void 3988 vop_create_post(void *ap, int rc) 3989 { 3990 struct vop_create_args *a = ap; 3991 3992 if (!rc) 3993 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3994 } 3995 3996 void 3997 vop_link_post(void *ap, int rc) 3998 { 3999 struct vop_link_args *a = ap; 4000 4001 if (!rc) { 4002 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK); 4003 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE); 4004 } 4005 } 4006 4007 void 4008 vop_mkdir_post(void *ap, int rc) 4009 { 4010 struct vop_mkdir_args *a = ap; 4011 4012 if (!rc) 4013 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 4014 } 4015 4016 void 4017 vop_mknod_post(void *ap, int rc) 4018 { 4019 struct vop_mknod_args *a = ap; 4020 4021 if (!rc) 4022 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4023 } 4024 4025 void 4026 vop_remove_post(void *ap, int rc) 4027 { 4028 struct vop_remove_args *a = ap; 4029 4030 if (!rc) { 4031 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4032 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 4033 } 4034 } 4035 4036 void 4037 vop_rename_post(void *ap, int rc) 4038 { 4039 struct vop_rename_args *a = ap; 4040 4041 if (!rc) { 4042 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE); 4043 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE); 4044 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); 4045 if (a->a_tvp) 4046 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); 4047 } 4048 if (a->a_tdvp != a->a_fdvp) 4049 vdrop(a->a_fdvp); 4050 if (a->a_tvp != a->a_fvp) 4051 vdrop(a->a_fvp); 4052 vdrop(a->a_tdvp); 4053 if (a->a_tvp) 4054 vdrop(a->a_tvp); 4055 } 4056 4057 void 4058 vop_rmdir_post(void *ap, int rc) 4059 { 4060 struct vop_rmdir_args *a = ap; 4061 4062 if (!rc) { 4063 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 4064 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 4065 } 4066 } 4067 4068 void 4069 vop_setattr_post(void *ap, int rc) 4070 { 4071 struct vop_setattr_args *a = ap; 4072 4073 if (!rc) 4074 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); 4075 } 4076 4077 void 4078 vop_symlink_post(void *ap, int rc) 4079 { 4080 struct vop_symlink_args *a = ap; 4081 4082 if (!rc) 4083 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4084 } 4085 4086 static struct knlist fs_knlist; 4087 4088 static void 4089 vfs_event_init(void *arg) 4090 { 4091 knlist_init_mtx(&fs_knlist, NULL); 4092 } 4093 /* XXX - correct order? */ 4094 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); 4095 4096 void 4097 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused) 4098 { 4099 4100 KNOTE_UNLOCKED(&fs_knlist, event); 4101 } 4102 4103 static int filt_fsattach(struct knote *kn); 4104 static void filt_fsdetach(struct knote *kn); 4105 static int filt_fsevent(struct knote *kn, long hint); 4106 4107 struct filterops fs_filtops = { 4108 .f_isfd = 0, 4109 .f_attach = filt_fsattach, 4110 .f_detach = filt_fsdetach, 4111 .f_event = filt_fsevent 4112 }; 4113 4114 static int 4115 filt_fsattach(struct knote *kn) 4116 { 4117 4118 kn->kn_flags |= EV_CLEAR; 4119 knlist_add(&fs_knlist, kn, 0); 4120 return (0); 4121 } 4122 4123 static void 4124 filt_fsdetach(struct knote *kn) 4125 { 4126 4127 knlist_remove(&fs_knlist, kn, 0); 4128 } 4129 4130 static int 4131 filt_fsevent(struct knote *kn, long hint) 4132 { 4133 4134 kn->kn_fflags |= hint; 4135 return (kn->kn_fflags != 0); 4136 } 4137 4138 static int 4139 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) 4140 { 4141 struct vfsidctl vc; 4142 int error; 4143 struct mount *mp; 4144 4145 error = SYSCTL_IN(req, &vc, sizeof(vc)); 4146 if (error) 4147 return (error); 4148 if (vc.vc_vers != VFS_CTL_VERS1) 4149 return (EINVAL); 4150 mp = vfs_getvfs(&vc.vc_fsid); 4151 if (mp == NULL) 4152 return (ENOENT); 4153 /* ensure that a specific sysctl goes to the right filesystem. */ 4154 if (strcmp(vc.vc_fstypename, "*") != 0 && 4155 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { 4156 vfs_rel(mp); 4157 return (EINVAL); 4158 } 4159 VCTLTOREQ(&vc, req); 4160 error = VFS_SYSCTL(mp, vc.vc_op, req); 4161 vfs_rel(mp); 4162 return (error); 4163 } 4164 4165 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR, 4166 NULL, 0, sysctl_vfs_ctl, "", 4167 "Sysctl by fsid"); 4168 4169 /* 4170 * Function to initialize a va_filerev field sensibly. 4171 * XXX: Wouldn't a random number make a lot more sense ?? 4172 */ 4173 u_quad_t 4174 init_va_filerev(void) 4175 { 4176 struct bintime bt; 4177 4178 getbinuptime(&bt); 4179 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); 4180 } 4181 4182 static int filt_vfsread(struct knote *kn, long hint); 4183 static int filt_vfswrite(struct knote *kn, long hint); 4184 static int filt_vfsvnode(struct knote *kn, long hint); 4185 static void filt_vfsdetach(struct knote *kn); 4186 static struct filterops vfsread_filtops = { 4187 .f_isfd = 1, 4188 .f_detach = filt_vfsdetach, 4189 .f_event = filt_vfsread 4190 }; 4191 static struct filterops vfswrite_filtops = { 4192 .f_isfd = 1, 4193 .f_detach = filt_vfsdetach, 4194 .f_event = filt_vfswrite 4195 }; 4196 static struct filterops vfsvnode_filtops = { 4197 .f_isfd = 1, 4198 .f_detach = filt_vfsdetach, 4199 .f_event = filt_vfsvnode 4200 }; 4201 4202 static void 4203 vfs_knllock(void *arg) 4204 { 4205 struct vnode *vp = arg; 4206 4207 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4208 } 4209 4210 static void 4211 vfs_knlunlock(void *arg) 4212 { 4213 struct vnode *vp = arg; 4214 4215 VOP_UNLOCK(vp, 0); 4216 } 4217 4218 static void 4219 vfs_knl_assert_locked(void *arg) 4220 { 4221 #ifdef DEBUG_VFS_LOCKS 4222 struct vnode *vp = arg; 4223 4224 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked"); 4225 #endif 4226 } 4227 4228 static void 4229 vfs_knl_assert_unlocked(void *arg) 4230 { 4231 #ifdef DEBUG_VFS_LOCKS 4232 struct vnode *vp = arg; 4233 4234 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked"); 4235 #endif 4236 } 4237 4238 int 4239 vfs_kqfilter(struct vop_kqfilter_args *ap) 4240 { 4241 struct vnode *vp = ap->a_vp; 4242 struct knote *kn = ap->a_kn; 4243 struct knlist *knl; 4244 4245 switch (kn->kn_filter) { 4246 case EVFILT_READ: 4247 kn->kn_fop = &vfsread_filtops; 4248 break; 4249 case EVFILT_WRITE: 4250 kn->kn_fop = &vfswrite_filtops; 4251 break; 4252 case EVFILT_VNODE: 4253 kn->kn_fop = &vfsvnode_filtops; 4254 break; 4255 default: 4256 return (EINVAL); 4257 } 4258 4259 kn->kn_hook = (caddr_t)vp; 4260 4261 v_addpollinfo(vp); 4262 if (vp->v_pollinfo == NULL) 4263 return (ENOMEM); 4264 knl = &vp->v_pollinfo->vpi_selinfo.si_note; 4265 knlist_add(knl, kn, 0); 4266 4267 return (0); 4268 } 4269 4270 /* 4271 * Detach knote from vnode 4272 */ 4273 static void 4274 filt_vfsdetach(struct knote *kn) 4275 { 4276 struct vnode *vp = (struct vnode *)kn->kn_hook; 4277 4278 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); 4279 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); 4280 } 4281 4282 /*ARGSUSED*/ 4283 static int 4284 filt_vfsread(struct knote *kn, long hint) 4285 { 4286 struct vnode *vp = (struct vnode *)kn->kn_hook; 4287 struct vattr va; 4288 int res; 4289 4290 /* 4291 * filesystem is gone, so set the EOF flag and schedule 4292 * the knote for deletion. 4293 */ 4294 if (hint == NOTE_REVOKE) { 4295 VI_LOCK(vp); 4296 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 4297 VI_UNLOCK(vp); 4298 return (1); 4299 } 4300 4301 if (VOP_GETATTR(vp, &va, curthread->td_ucred)) 4302 return (0); 4303 4304 VI_LOCK(vp); 4305 kn->kn_data = va.va_size - kn->kn_fp->f_offset; 4306 res = (kn->kn_data != 0); 4307 VI_UNLOCK(vp); 4308 return (res); 4309 } 4310 4311 /*ARGSUSED*/ 4312 static int 4313 filt_vfswrite(struct knote *kn, long hint) 4314 { 4315 struct vnode *vp = (struct vnode *)kn->kn_hook; 4316 4317 VI_LOCK(vp); 4318 4319 /* 4320 * filesystem is gone, so set the EOF flag and schedule 4321 * the knote for deletion. 4322 */ 4323 if (hint == NOTE_REVOKE) 4324 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 4325 4326 kn->kn_data = 0; 4327 VI_UNLOCK(vp); 4328 return (1); 4329 } 4330 4331 static int 4332 filt_vfsvnode(struct knote *kn, long hint) 4333 { 4334 struct vnode *vp = (struct vnode *)kn->kn_hook; 4335 int res; 4336 4337 VI_LOCK(vp); 4338 if (kn->kn_sfflags & hint) 4339 kn->kn_fflags |= hint; 4340 if (hint == NOTE_REVOKE) { 4341 kn->kn_flags |= EV_EOF; 4342 VI_UNLOCK(vp); 4343 return (1); 4344 } 4345 res = (kn->kn_fflags != 0); 4346 VI_UNLOCK(vp); 4347 return (res); 4348 } 4349 4350 int 4351 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) 4352 { 4353 int error; 4354 4355 if (dp->d_reclen > ap->a_uio->uio_resid) 4356 return (ENAMETOOLONG); 4357 error = uiomove(dp, dp->d_reclen, ap->a_uio); 4358 if (error) { 4359 if (ap->a_ncookies != NULL) { 4360 if (ap->a_cookies != NULL) 4361 free(ap->a_cookies, M_TEMP); 4362 ap->a_cookies = NULL; 4363 *ap->a_ncookies = 0; 4364 } 4365 return (error); 4366 } 4367 if (ap->a_ncookies == NULL) 4368 return (0); 4369 4370 KASSERT(ap->a_cookies, 4371 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); 4372 4373 *ap->a_cookies = realloc(*ap->a_cookies, 4374 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); 4375 (*ap->a_cookies)[*ap->a_ncookies] = off; 4376 return (0); 4377 } 4378 4379 /* 4380 * Mark for update the access time of the file if the filesystem 4381 * supports VOP_MARKATIME. This functionality is used by execve and 4382 * mmap, so we want to avoid the I/O implied by directly setting 4383 * va_atime for the sake of efficiency. 4384 */ 4385 void 4386 vfs_mark_atime(struct vnode *vp, struct ucred *cred) 4387 { 4388 struct mount *mp; 4389 4390 mp = vp->v_mount; 4391 VFS_ASSERT_GIANT(mp); 4392 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime"); 4393 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) 4394 (void)VOP_MARKATIME(vp); 4395 } 4396 4397 /* 4398 * The purpose of this routine is to remove granularity from accmode_t, 4399 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE, 4400 * VADMIN and VAPPEND. 4401 * 4402 * If it returns 0, the caller is supposed to continue with the usual 4403 * access checks using 'accmode' as modified by this routine. If it 4404 * returns nonzero value, the caller is supposed to return that value 4405 * as errno. 4406 * 4407 * Note that after this routine runs, accmode may be zero. 4408 */ 4409 int 4410 vfs_unixify_accmode(accmode_t *accmode) 4411 { 4412 /* 4413 * There is no way to specify explicit "deny" rule using 4414 * file mode or POSIX.1e ACLs. 4415 */ 4416 if (*accmode & VEXPLICIT_DENY) { 4417 *accmode = 0; 4418 return (0); 4419 } 4420 4421 /* 4422 * None of these can be translated into usual access bits. 4423 * Also, the common case for NFSv4 ACLs is to not contain 4424 * either of these bits. Caller should check for VWRITE 4425 * on the containing directory instead. 4426 */ 4427 if (*accmode & (VDELETE_CHILD | VDELETE)) 4428 return (EPERM); 4429 4430 if (*accmode & VADMIN_PERMS) { 4431 *accmode &= ~VADMIN_PERMS; 4432 *accmode |= VADMIN; 4433 } 4434 4435 /* 4436 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL 4437 * or VSYNCHRONIZE using file mode or POSIX.1e ACL. 4438 */ 4439 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE); 4440 4441 return (0); 4442 } 4443