1 /*- 2 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc. 3 * All rights reserved. 4 * 5 * This code is derived from software contributed to The NetBSD Foundation 6 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code 7 * 2005 program. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 19 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 20 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 21 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 22 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 28 * POSSIBILITY OF SUCH DAMAGE. 29 * 30 * $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $ 31 */ 32 33 /* 34 * tmpfs vnode interface. 35 */ 36 37 #include <sys/kernel.h> 38 #include <sys/kern_syscall.h> 39 #include <sys/param.h> 40 #include <sys/uio.h> 41 #include <sys/fcntl.h> 42 #include <sys/lockf.h> 43 #include <sys/priv.h> 44 #include <sys/proc.h> 45 #include <sys/resourcevar.h> 46 #include <sys/sched.h> 47 #include <sys/stat.h> 48 #include <sys/systm.h> 49 #include <sys/sysctl.h> 50 #include <sys/unistd.h> 51 #include <sys/vfsops.h> 52 #include <sys/vnode.h> 53 #include <sys/mountctl.h> 54 55 #include <vm/vm.h> 56 #include <vm/vm_extern.h> 57 #include <vm/vm_object.h> 58 #include <vm/vm_page.h> 59 #include <vm/vm_pageout.h> 60 #include <vm/vm_pager.h> 61 #include <vm/swap_pager.h> 62 63 #include <sys/buf2.h> 64 #include <vm/vm_page2.h> 65 66 #include <vfs/fifofs/fifo.h> 67 #include <vfs/tmpfs/tmpfs_vnops.h> 68 #include "tmpfs.h" 69 70 static void tmpfs_strategy_done(struct bio *bio); 71 static void tmpfs_move_pages(vm_object_t src, vm_object_t dst, int movflags); 72 73 /* 74 * bufcache_mode: 75 * 0 Normal page queue operation on flush. Try to keep in memory. 76 * 1 Try to cache on flush to swap (default). 77 * 2 Always page to swap (not recommended). 78 */ 79 __read_mostly static int tmpfs_cluster_rd_enable = 1; 80 __read_mostly static int tmpfs_cluster_wr_enable = 1; 81 __read_mostly int tmpfs_bufcache_mode = 1; 82 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "TMPFS filesystem"); 83 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, cluster_rd_enable, CTLFLAG_RW, 84 &tmpfs_cluster_rd_enable, 0, ""); 85 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, cluster_wr_enable, CTLFLAG_RW, 86 &tmpfs_cluster_wr_enable, 0, ""); 87 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, bufcache_mode, CTLFLAG_RW, 88 &tmpfs_bufcache_mode, 0, ""); 89 90 #define TMPFS_MOVF_FROMBACKING 0x0001 91 #define TMPFS_MOVF_DEACTIVATE 0x0002 92 93 94 static __inline 95 void 96 tmpfs_knote(struct vnode *vp, int flags) 97 { 98 if (flags) 99 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags); 100 } 101 102 103 /* --------------------------------------------------------------------- */ 104 105 static int 106 tmpfs_nresolve(struct vop_nresolve_args *ap) 107 { 108 struct vnode *dvp = ap->a_dvp; 109 struct vnode *vp = NULL; 110 struct namecache *ncp = ap->a_nch->ncp; 111 struct tmpfs_node *tnode; 112 struct tmpfs_dirent *de; 113 struct tmpfs_node *dnode; 114 int error; 115 116 dnode = VP_TO_TMPFS_DIR(dvp); 117 118 TMPFS_NODE_LOCK_SH(dnode); 119 loop: 120 de = tmpfs_dir_lookup(dnode, NULL, ncp); 121 if (de == NULL) { 122 error = ENOENT; 123 } else { 124 /* 125 * Allocate a vnode for the node we found. Use 126 * tmpfs_alloc_vp()'s deadlock handling mode. 127 */ 128 tnode = de->td_node; 129 error = tmpfs_alloc_vp(dvp->v_mount, dnode, tnode, 130 LK_EXCLUSIVE | LK_RETRY, &vp); 131 if (error == EAGAIN) 132 goto loop; 133 if (error) 134 goto out; 135 KKASSERT(vp); 136 } 137 138 out: 139 TMPFS_NODE_UNLOCK(dnode); 140 141 if ((dnode->tn_status & TMPFS_NODE_ACCESSED) == 0) { 142 TMPFS_NODE_LOCK(dnode); 143 dnode->tn_status |= TMPFS_NODE_ACCESSED; 144 TMPFS_NODE_UNLOCK(dnode); 145 } 146 147 /* 148 * Store the result of this lookup in the cache. Avoid this if the 149 * request was for creation, as it does not improve timings on 150 * emprical tests. 151 */ 152 if (vp) { 153 vn_unlock(vp); 154 cache_setvp(ap->a_nch, vp); 155 vrele(vp); 156 } else if (error == ENOENT) { 157 cache_setvp(ap->a_nch, NULL); 158 } 159 return (error); 160 } 161 162 static int 163 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *ap) 164 { 165 struct vnode *dvp = ap->a_dvp; 166 struct vnode **vpp = ap->a_vpp; 167 struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp); 168 struct ucred *cred = ap->a_cred; 169 int error; 170 171 *vpp = NULL; 172 173 /* Check accessibility of requested node as a first step. */ 174 error = VOP_ACCESS(dvp, VEXEC, cred); 175 if (error != 0) 176 return error; 177 178 if (dnode->tn_dir.tn_parent != NULL) { 179 /* Allocate a new vnode on the matching entry. */ 180 error = tmpfs_alloc_vp(dvp->v_mount, 181 NULL, dnode->tn_dir.tn_parent, 182 LK_EXCLUSIVE | LK_RETRY, vpp); 183 184 if (*vpp) 185 vn_unlock(*vpp); 186 } 187 return (*vpp == NULL) ? ENOENT : 0; 188 } 189 190 /* --------------------------------------------------------------------- */ 191 192 static int 193 tmpfs_ncreate(struct vop_ncreate_args *ap) 194 { 195 struct vnode *dvp = ap->a_dvp; 196 struct vnode **vpp = ap->a_vpp; 197 struct namecache *ncp = ap->a_nch->ncp; 198 struct vattr *vap = ap->a_vap; 199 struct ucred *cred = ap->a_cred; 200 int error; 201 202 KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK); 203 204 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL); 205 if (error == 0) { 206 cache_setunresolved(ap->a_nch); 207 cache_setvp(ap->a_nch, *vpp); 208 tmpfs_knote(dvp, NOTE_WRITE); 209 } 210 return (error); 211 } 212 /* --------------------------------------------------------------------- */ 213 214 static int 215 tmpfs_nmknod(struct vop_nmknod_args *ap) 216 { 217 struct vnode *dvp = ap->a_dvp; 218 struct vnode **vpp = ap->a_vpp; 219 struct namecache *ncp = ap->a_nch->ncp; 220 struct vattr *vap = ap->a_vap; 221 struct ucred *cred = ap->a_cred; 222 int error; 223 224 if (vap->va_type != VBLK && vap->va_type != VCHR && 225 vap->va_type != VFIFO) { 226 return (EINVAL); 227 } 228 229 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL); 230 if (error == 0) { 231 cache_setunresolved(ap->a_nch); 232 cache_setvp(ap->a_nch, *vpp); 233 tmpfs_knote(dvp, NOTE_WRITE); 234 } 235 return error; 236 } 237 238 /* --------------------------------------------------------------------- */ 239 240 static int 241 tmpfs_open(struct vop_open_args *ap) 242 { 243 struct vnode *vp = ap->a_vp; 244 int mode = ap->a_mode; 245 struct tmpfs_node *node; 246 int error; 247 248 node = VP_TO_TMPFS_NODE(vp); 249 250 #if 0 251 /* The file is still active but all its names have been removed 252 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as 253 * it is about to die. */ 254 if (node->tn_links < 1) 255 return (ENOENT); 256 #endif 257 258 /* If the file is marked append-only, deny write requests. */ 259 if ((node->tn_flags & APPEND) && 260 (mode & (FWRITE | O_APPEND)) == FWRITE) { 261 error = EPERM; 262 } else { 263 if (node->tn_reg.tn_pages_in_aobj) { 264 TMPFS_NODE_LOCK(node); 265 if (node->tn_reg.tn_pages_in_aobj) { 266 tmpfs_move_pages(node->tn_reg.tn_aobj, 267 vp->v_object, 268 TMPFS_MOVF_FROMBACKING); 269 node->tn_reg.tn_pages_in_aobj = 0; 270 } 271 TMPFS_NODE_UNLOCK(node); 272 } 273 error = vop_stdopen(ap); 274 } 275 276 return (error); 277 } 278 279 /* --------------------------------------------------------------------- */ 280 281 static int 282 tmpfs_close(struct vop_close_args *ap) 283 { 284 struct vnode *vp = ap->a_vp; 285 struct tmpfs_node *node; 286 int error; 287 288 node = VP_TO_TMPFS_NODE(vp); 289 290 if (node->tn_links > 0) { 291 /* 292 * Update node times. No need to do it if the node has 293 * been deleted, because it will vanish after we return. 294 */ 295 tmpfs_update(vp); 296 } 297 298 error = vop_stdclose(ap); 299 300 return (error); 301 } 302 303 /* --------------------------------------------------------------------- */ 304 305 int 306 tmpfs_access(struct vop_access_args *ap) 307 { 308 struct vnode *vp = ap->a_vp; 309 int error; 310 struct tmpfs_node *node; 311 312 node = VP_TO_TMPFS_NODE(vp); 313 314 switch (vp->v_type) { 315 case VDIR: 316 /* FALLTHROUGH */ 317 case VLNK: 318 /* FALLTHROUGH */ 319 case VREG: 320 if ((ap->a_mode & VWRITE) && 321 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 322 error = EROFS; 323 goto out; 324 } 325 break; 326 327 case VBLK: 328 /* FALLTHROUGH */ 329 case VCHR: 330 /* FALLTHROUGH */ 331 case VSOCK: 332 /* FALLTHROUGH */ 333 case VFIFO: 334 break; 335 336 default: 337 error = EINVAL; 338 goto out; 339 } 340 341 if ((ap->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) { 342 error = EPERM; 343 goto out; 344 } 345 346 error = vop_helper_access(ap, node->tn_uid, node->tn_gid, 347 node->tn_mode, 0); 348 out: 349 return error; 350 } 351 352 /* --------------------------------------------------------------------- */ 353 354 int 355 tmpfs_getattr(struct vop_getattr_args *ap) 356 { 357 struct vnode *vp = ap->a_vp; 358 struct vattr *vap = ap->a_vap; 359 struct tmpfs_node *node; 360 361 node = VP_TO_TMPFS_NODE(vp); 362 363 tmpfs_update(vp); 364 365 TMPFS_NODE_LOCK_SH(node); 366 vap->va_type = vp->v_type; 367 vap->va_mode = node->tn_mode; 368 vap->va_nlink = node->tn_links; 369 vap->va_uid = node->tn_uid; 370 vap->va_gid = node->tn_gid; 371 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 372 vap->va_fileid = node->tn_id; 373 vap->va_size = node->tn_size; 374 vap->va_blocksize = PAGE_SIZE; 375 vap->va_atime.tv_sec = node->tn_atime; 376 vap->va_atime.tv_nsec = node->tn_atimensec; 377 vap->va_mtime.tv_sec = node->tn_mtime; 378 vap->va_mtime.tv_nsec = node->tn_mtimensec; 379 vap->va_ctime.tv_sec = node->tn_ctime; 380 vap->va_ctime.tv_nsec = node->tn_ctimensec; 381 vap->va_gen = node->tn_gen; 382 vap->va_flags = node->tn_flags; 383 if (vp->v_type == VBLK || vp->v_type == VCHR) { 384 vap->va_rmajor = umajor(node->tn_rdev); 385 vap->va_rminor = uminor(node->tn_rdev); 386 } 387 vap->va_bytes = round_page(node->tn_size); 388 vap->va_filerev = 0; 389 TMPFS_NODE_UNLOCK(node); 390 391 return 0; 392 } 393 394 /* --------------------------------------------------------------------- */ 395 396 int 397 tmpfs_setattr(struct vop_setattr_args *ap) 398 { 399 struct vnode *vp = ap->a_vp; 400 struct vattr *vap = ap->a_vap; 401 struct ucred *cred = ap->a_cred; 402 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp); 403 int error = 0; 404 int kflags = 0; 405 406 TMPFS_NODE_LOCK(node); 407 if (error == 0 && (vap->va_flags != VNOVAL)) { 408 error = tmpfs_chflags(vp, vap->va_flags, cred); 409 kflags |= NOTE_ATTRIB; 410 } 411 412 if (error == 0 && (vap->va_size != VNOVAL)) { 413 /* restore any saved pages before proceeding */ 414 if (node->tn_reg.tn_pages_in_aobj) { 415 tmpfs_move_pages(node->tn_reg.tn_aobj, vp->v_object, 416 TMPFS_MOVF_FROMBACKING | 417 TMPFS_MOVF_DEACTIVATE); 418 node->tn_reg.tn_pages_in_aobj = 0; 419 } 420 if (vap->va_size > node->tn_size) 421 kflags |= NOTE_WRITE | NOTE_EXTEND; 422 else 423 kflags |= NOTE_WRITE; 424 error = tmpfs_chsize(vp, vap->va_size, cred); 425 } 426 427 if (error == 0 && (vap->va_uid != (uid_t)VNOVAL || 428 vap->va_gid != (gid_t)VNOVAL)) { 429 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred); 430 kflags |= NOTE_ATTRIB; 431 } 432 433 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) { 434 error = tmpfs_chmod(vp, vap->va_mode, cred); 435 kflags |= NOTE_ATTRIB; 436 } 437 438 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL && 439 vap->va_atime.tv_nsec != VNOVAL) || 440 (vap->va_mtime.tv_sec != VNOVAL && 441 vap->va_mtime.tv_nsec != VNOVAL) )) { 442 error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime, 443 vap->va_vaflags, cred); 444 kflags |= NOTE_ATTRIB; 445 } 446 447 /* 448 * Update the node times. We give preference to the error codes 449 * generated by this function rather than the ones that may arise 450 * from tmpfs_update. 451 */ 452 tmpfs_update(vp); 453 TMPFS_NODE_UNLOCK(node); 454 tmpfs_knote(vp, kflags); 455 456 return (error); 457 } 458 459 /* --------------------------------------------------------------------- */ 460 461 /* 462 * fsync is usually a NOP, but we must take action when unmounting or 463 * when recycling. 464 */ 465 static int 466 tmpfs_fsync(struct vop_fsync_args *ap) 467 { 468 struct tmpfs_node *node; 469 struct vnode *vp = ap->a_vp; 470 471 node = VP_TO_TMPFS_NODE(vp); 472 473 /* 474 * tmpfs vnodes typically remain dirty, avoid long syncer scans 475 * by forcing removal from the syncer list. 476 */ 477 vn_syncer_remove(vp, 1); 478 479 tmpfs_update(vp); 480 if (vp->v_type == VREG) { 481 if (vp->v_flag & VRECLAIMED) { 482 if (node->tn_links == 0) 483 tmpfs_truncate(vp, 0); 484 else 485 vfsync(ap->a_vp, ap->a_waitfor, 1, NULL, NULL); 486 } 487 } 488 489 return 0; 490 } 491 492 /* --------------------------------------------------------------------- */ 493 494 static int 495 tmpfs_read(struct vop_read_args *ap) 496 { 497 struct buf *bp; 498 struct vnode *vp = ap->a_vp; 499 struct uio *uio = ap->a_uio; 500 struct tmpfs_node *node; 501 off_t base_offset; 502 size_t offset; 503 size_t len; 504 size_t resid; 505 int error; 506 int seqcount; 507 508 /* 509 * Check the basics 510 */ 511 if (uio->uio_offset < 0) 512 return (EINVAL); 513 if (vp->v_type != VREG) 514 return (EINVAL); 515 516 /* 517 * Extract node, try to shortcut the operation through 518 * the VM page cache, allowing us to avoid buffer cache 519 * overheads. 520 */ 521 node = VP_TO_TMPFS_NODE(vp); 522 resid = uio->uio_resid; 523 seqcount = ap->a_ioflag >> IO_SEQSHIFT; 524 error = vop_helper_read_shortcut(ap); 525 if (error) 526 return error; 527 if (uio->uio_resid == 0) { 528 if (resid) 529 goto finished; 530 return error; 531 } 532 533 /* 534 * restore any saved pages before proceeding 535 */ 536 if (node->tn_reg.tn_pages_in_aobj) { 537 TMPFS_NODE_LOCK(node); 538 if (node->tn_reg.tn_pages_in_aobj) { 539 tmpfs_move_pages(node->tn_reg.tn_aobj, vp->v_object, 540 TMPFS_MOVF_FROMBACKING); 541 node->tn_reg.tn_pages_in_aobj = 0; 542 } 543 TMPFS_NODE_UNLOCK(node); 544 } 545 546 /* 547 * Fall-through to our normal read code. 548 */ 549 while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) { 550 /* 551 * Use buffer cache I/O (via tmpfs_strategy) 552 */ 553 offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64; 554 base_offset = (off_t)uio->uio_offset - offset; 555 bp = getcacheblk(vp, base_offset, 556 node->tn_blksize, GETBLK_KVABIO); 557 if (bp == NULL) { 558 if (tmpfs_cluster_rd_enable) { 559 error = cluster_readx(vp, node->tn_size, 560 base_offset, 561 node->tn_blksize, 562 B_NOTMETA | B_KVABIO, 563 uio->uio_resid, 564 seqcount * MAXBSIZE, 565 &bp); 566 } else { 567 error = bread_kvabio(vp, base_offset, 568 node->tn_blksize, &bp); 569 } 570 if (error) { 571 brelse(bp); 572 kprintf("tmpfs_read bread error %d\n", error); 573 break; 574 } 575 576 /* 577 * tmpfs pretty much fiddles directly with the VM 578 * system, don't let it exhaust it or we won't play 579 * nice with other processes. 580 * 581 * Only do this if the VOP is coming from a normal 582 * read/write. The VM system handles the case for 583 * UIO_NOCOPY. 584 */ 585 if (uio->uio_segflg != UIO_NOCOPY) 586 vm_wait_nominal(); 587 } 588 bp->b_flags |= B_CLUSTEROK; 589 bkvasync(bp); 590 591 /* 592 * Figure out how many bytes we can actually copy this loop. 593 */ 594 len = node->tn_blksize - offset; 595 if (len > uio->uio_resid) 596 len = uio->uio_resid; 597 if (len > node->tn_size - uio->uio_offset) 598 len = (size_t)(node->tn_size - uio->uio_offset); 599 600 error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio); 601 bqrelse(bp); 602 if (error) { 603 kprintf("tmpfs_read uiomove error %d\n", error); 604 break; 605 } 606 } 607 608 finished: 609 if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) { 610 TMPFS_NODE_LOCK(node); 611 node->tn_status |= TMPFS_NODE_ACCESSED; 612 TMPFS_NODE_UNLOCK(node); 613 } 614 return (error); 615 } 616 617 static int 618 tmpfs_write(struct vop_write_args *ap) 619 { 620 struct buf *bp; 621 struct vnode *vp = ap->a_vp; 622 struct uio *uio = ap->a_uio; 623 struct thread *td = uio->uio_td; 624 struct tmpfs_node *node; 625 boolean_t extended; 626 off_t oldsize; 627 int error; 628 off_t base_offset; 629 size_t offset; 630 size_t len; 631 struct rlimit limit; 632 int trivial = 0; 633 int kflags = 0; 634 int seqcount; 635 636 error = 0; 637 if (uio->uio_resid == 0) { 638 return error; 639 } 640 641 node = VP_TO_TMPFS_NODE(vp); 642 643 if (vp->v_type != VREG) 644 return (EINVAL); 645 seqcount = ap->a_ioflag >> IO_SEQSHIFT; 646 647 TMPFS_NODE_LOCK(node); 648 649 /* 650 * restore any saved pages before proceeding 651 */ 652 if (node->tn_reg.tn_pages_in_aobj) { 653 tmpfs_move_pages(node->tn_reg.tn_aobj, vp->v_object, 654 TMPFS_MOVF_FROMBACKING); 655 node->tn_reg.tn_pages_in_aobj = 0; 656 } 657 658 oldsize = node->tn_size; 659 if (ap->a_ioflag & IO_APPEND) 660 uio->uio_offset = node->tn_size; 661 662 /* 663 * Check for illegal write offsets. 664 */ 665 if (uio->uio_offset + uio->uio_resid > 666 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) { 667 error = EFBIG; 668 goto done; 669 } 670 671 /* 672 * NOTE: Ignore if UIO does not come from a user thread (e.g. VN). 673 */ 674 if (vp->v_type == VREG && td != NULL && td->td_lwp != NULL) { 675 error = kern_getrlimit(RLIMIT_FSIZE, &limit); 676 if (error) 677 goto done; 678 if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) { 679 ksignal(td->td_proc, SIGXFSZ); 680 error = EFBIG; 681 goto done; 682 } 683 } 684 685 /* 686 * Extend the file's size if necessary 687 */ 688 extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size); 689 690 while (uio->uio_resid > 0) { 691 /* 692 * Don't completely blow out running buffer I/O 693 * when being hit from the pageout daemon. 694 */ 695 if (uio->uio_segflg == UIO_NOCOPY && 696 (ap->a_ioflag & IO_RECURSE) == 0) { 697 bwillwrite(node->tn_blksize); 698 } 699 700 /* 701 * Use buffer cache I/O (via tmpfs_strategy) 702 * 703 * Calculate the maximum bytes we can write to the buffer at 704 * this offset (after resizing). 705 */ 706 offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64; 707 base_offset = (off_t)uio->uio_offset - offset; 708 len = uio->uio_resid; 709 if (len > TMPFS_BLKSIZE - offset) 710 len = TMPFS_BLKSIZE - offset; 711 712 if ((uio->uio_offset + len) > node->tn_size) { 713 trivial = (uio->uio_offset <= node->tn_size); 714 error = tmpfs_reg_resize(vp, uio->uio_offset + len, 715 trivial); 716 if (error) 717 break; 718 } 719 720 /* 721 * Read to fill in any gaps. Theoretically we could 722 * optimize this if the write covers the entire buffer 723 * and is not a UIO_NOCOPY write, however this can lead 724 * to a security violation exposing random kernel memory 725 * (whatever junk was in the backing VM pages before). 726 * 727 * So just use bread() to do the right thing. 728 */ 729 error = bread_kvabio(vp, base_offset, node->tn_blksize, &bp); 730 bkvasync(bp); 731 error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio); 732 if (error) { 733 kprintf("tmpfs_write uiomove error %d\n", error); 734 brelse(bp); 735 break; 736 } 737 738 if (uio->uio_offset > node->tn_size) { 739 node->tn_size = uio->uio_offset; 740 kflags |= NOTE_EXTEND; 741 } 742 kflags |= NOTE_WRITE; 743 744 /* 745 * UIO_NOCOPY is a sensitive state due to potentially being 746 * issued from the pageout daemon while in a low-memory 747 * situation. However, in order to cluster the I/O nicely 748 * (e.g. 64KB+ writes instead of 16KB writes), we still try 749 * to follow the same semantics that any other filesystem 750 * might use. 751 * 752 * For the normal case we buwrite(), dirtying the underlying 753 * VM pages instead of dirtying the buffer and releasing the 754 * buffer as a clean buffer. This allows tmpfs to use 755 * essentially all available memory to cache file data. 756 * If we used bdwrite() the buffer cache would wind up 757 * flushing the data to swap too quickly. 758 * 759 * But because tmpfs can seriously load the VM system we 760 * fall-back to using bdwrite() when free memory starts 761 * to get low. This shifts the load away from the VM system 762 * and makes tmpfs act more like a normal filesystem with 763 * regards to disk activity. 764 * 765 * tmpfs pretty much fiddles directly with the VM 766 * system, don't let it exhaust it or we won't play 767 * nice with other processes. Only do this if the 768 * VOP is coming from a normal read/write. The VM system 769 * handles the case for UIO_NOCOPY. 770 */ 771 bp->b_flags |= B_CLUSTEROK; 772 if (uio->uio_segflg == UIO_NOCOPY) { 773 /* 774 * Flush from the pageout daemon, deal with 775 * potentially very heavy tmpfs write activity 776 * causing long stalls in the pageout daemon 777 * before pages get to free/cache. 778 * 779 * (a) Under severe pressure setting B_DIRECT will 780 * cause a buffer release to try to free the 781 * underlying pages. 782 * 783 * (b) Under modest memory pressure the B_RELBUF 784 * alone is sufficient to get the pages moved 785 * to the cache. We could also force this by 786 * setting B_NOTMETA but that might have other 787 * unintended side-effects (e.g. setting 788 * PG_NOTMETA on the VM page). 789 * 790 * (c) For the pageout->putpages->generic_putpages-> 791 * UIO_NOCOPY-write (here), issuing an immediate 792 * write prevents any real clustering from 793 * happening because the buffers probably aren't 794 * (yet) marked dirty, or lost due to prior use 795 * of buwrite(). Try to use the normal 796 * cluster_write() mechanism for performance. 797 * 798 * Hopefully this will unblock the VM system more 799 * quickly under extreme tmpfs write load. 800 */ 801 if (vm_page_count_min(vm_page_free_hysteresis)) 802 bp->b_flags |= B_DIRECT; 803 bp->b_flags |= B_AGE | B_RELBUF | B_TTC; 804 bp->b_act_count = 0; /* buffer->deactivate pgs */ 805 if (tmpfs_cluster_wr_enable && 806 (ap->a_ioflag & (IO_SYNC | IO_DIRECT)) == 0) { 807 cluster_write(bp, node->tn_size, 808 node->tn_blksize, seqcount); 809 } else { 810 cluster_awrite(bp); 811 } 812 } else if (vm_pages_needed || vm_paging_needed(0) || 813 tmpfs_bufcache_mode >= 2) { 814 /* 815 * If the pageout daemon is running we cycle the 816 * write through the buffer cache normally to 817 * pipeline the flush, thus avoiding adding any 818 * more memory pressure to the pageout daemon. 819 */ 820 bp->b_act_count = 0; /* buffer->deactivate pgs */ 821 if (tmpfs_cluster_wr_enable) { 822 cluster_write(bp, node->tn_size, 823 node->tn_blksize, seqcount); 824 } else { 825 bdwrite(bp); 826 } 827 } else { 828 /* 829 * Otherwise run the buffer directly through to the 830 * backing VM store, leaving the buffer clean so 831 * buffer limits do not force early flushes to swap. 832 */ 833 buwrite(bp); 834 /*vm_wait_nominal();*/ 835 } 836 837 if (bp->b_error) { 838 kprintf("tmpfs_write bwrite error %d\n", bp->b_error); 839 break; 840 } 841 } 842 843 if (error) { 844 if (extended) { 845 (void)tmpfs_reg_resize(vp, oldsize, trivial); 846 kflags &= ~NOTE_EXTEND; 847 } 848 goto done; 849 } 850 851 /* 852 * Currently we don't set the mtime on files modified via mmap() 853 * because we can't tell the difference between those modifications 854 * and an attempt by the pageout daemon to flush tmpfs pages to 855 * swap. 856 * 857 * This is because in order to defer flushes as long as possible 858 * buwrite() works by marking the underlying VM pages dirty in 859 * order to be able to dispose of the buffer cache buffer without 860 * flushing it. 861 */ 862 if (uio->uio_segflg == UIO_NOCOPY) { 863 if (vp->v_flag & VLASTWRITETS) { 864 node->tn_mtime = vp->v_lastwrite_ts.tv_sec; 865 node->tn_mtimensec = vp->v_lastwrite_ts.tv_nsec; 866 } 867 } else { 868 node->tn_status |= TMPFS_NODE_MODIFIED; 869 vclrflags(vp, VLASTWRITETS); 870 } 871 872 if (extended) 873 node->tn_status |= TMPFS_NODE_CHANGED; 874 875 if (node->tn_mode & (S_ISUID | S_ISGID)) { 876 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) 877 node->tn_mode &= ~(S_ISUID | S_ISGID); 878 } 879 done: 880 TMPFS_NODE_UNLOCK(node); 881 if (kflags) 882 tmpfs_knote(vp, kflags); 883 884 return(error); 885 } 886 887 static int 888 tmpfs_advlock(struct vop_advlock_args *ap) 889 { 890 struct tmpfs_node *node; 891 struct vnode *vp = ap->a_vp; 892 int error; 893 894 node = VP_TO_TMPFS_NODE(vp); 895 error = (lf_advlock(ap, &node->tn_advlock, node->tn_size)); 896 897 return (error); 898 } 899 900 /* 901 * The strategy function is typically only called when memory pressure 902 * forces the system to attempt to pageout pages. It can also be called 903 * by [n]vtruncbuf() when a truncation cuts a page in half. Normal write 904 * operations 905 * 906 * We set VKVABIO for VREG files so bp->b_data may not be synchronized to 907 * our cpu. swap_pager_strategy() is all we really use, and it directly 908 * supports this. 909 */ 910 static int 911 tmpfs_strategy(struct vop_strategy_args *ap) 912 { 913 struct bio *bio = ap->a_bio; 914 struct bio *nbio; 915 struct buf *bp = bio->bio_buf; 916 struct vnode *vp = ap->a_vp; 917 struct tmpfs_node *node; 918 vm_object_t uobj; 919 vm_page_t m; 920 int i; 921 922 if (vp->v_type != VREG) { 923 bp->b_resid = bp->b_bcount; 924 bp->b_flags |= B_ERROR | B_INVAL; 925 bp->b_error = EINVAL; 926 biodone(bio); 927 return(0); 928 } 929 930 node = VP_TO_TMPFS_NODE(vp); 931 932 uobj = node->tn_reg.tn_aobj; 933 934 /* 935 * Don't bother flushing to swap if there is no swap, just 936 * ensure that the pages are marked as needing a commit (still). 937 */ 938 if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) { 939 for (i = 0; i < bp->b_xio.xio_npages; ++i) { 940 m = bp->b_xio.xio_pages[i]; 941 vm_page_need_commit(m); 942 } 943 bp->b_resid = 0; 944 bp->b_error = 0; 945 biodone(bio); 946 } else { 947 /* 948 * Tell the buffer cache to try to recycle the pages 949 * to PQ_CACHE on release. 950 */ 951 if (tmpfs_bufcache_mode >= 2 || 952 (tmpfs_bufcache_mode == 1 && vm_paging_needed(0))) { 953 bp->b_flags |= B_TTC; 954 } 955 nbio = push_bio(bio); 956 nbio->bio_done = tmpfs_strategy_done; 957 nbio->bio_offset = bio->bio_offset; 958 swap_pager_strategy(uobj, nbio); 959 } 960 return 0; 961 } 962 963 /* 964 * If we were unable to commit the pages to swap make sure they are marked 965 * as needing a commit (again). If we were, clear the flag to allow the 966 * pages to be freed. 967 * 968 * Do not error-out the buffer. In particular, vinvalbuf() needs to 969 * always work. 970 */ 971 static void 972 tmpfs_strategy_done(struct bio *bio) 973 { 974 struct buf *bp; 975 vm_page_t m; 976 int i; 977 978 bp = bio->bio_buf; 979 980 if (bp->b_flags & B_ERROR) { 981 bp->b_flags &= ~B_ERROR; 982 bp->b_error = 0; 983 bp->b_resid = 0; 984 for (i = 0; i < bp->b_xio.xio_npages; ++i) { 985 m = bp->b_xio.xio_pages[i]; 986 vm_page_need_commit(m); 987 } 988 } else { 989 for (i = 0; i < bp->b_xio.xio_npages; ++i) { 990 m = bp->b_xio.xio_pages[i]; 991 vm_page_clear_commit(m); 992 } 993 } 994 bio = pop_bio(bio); 995 biodone(bio); 996 } 997 998 /* 999 * To make write clustering work well make the backing store look 1000 * contiguous to the cluster_*() code. The swap_strategy() function 1001 * will take it from there. 1002 * 1003 * Use MAXBSIZE-sized chunks as a micro-optimization to make random 1004 * flushes leave full-sized gaps. 1005 */ 1006 static int 1007 tmpfs_bmap(struct vop_bmap_args *ap) 1008 { 1009 if (ap->a_doffsetp != NULL) 1010 *ap->a_doffsetp = ap->a_loffset; 1011 if (ap->a_runp != NULL) 1012 *ap->a_runp = MAXBSIZE - (ap->a_loffset & (MAXBSIZE - 1)); 1013 if (ap->a_runb != NULL) 1014 *ap->a_runb = ap->a_loffset & (MAXBSIZE - 1); 1015 1016 return 0; 1017 } 1018 1019 /* --------------------------------------------------------------------- */ 1020 1021 static int 1022 tmpfs_nremove(struct vop_nremove_args *ap) 1023 { 1024 struct vnode *dvp = ap->a_dvp; 1025 struct namecache *ncp = ap->a_nch->ncp; 1026 struct vnode *vp; 1027 int error; 1028 struct tmpfs_dirent *de; 1029 struct tmpfs_mount *tmp; 1030 struct tmpfs_node *dnode; 1031 struct tmpfs_node *node; 1032 1033 /* 1034 * We have to acquire the vp from ap->a_nch because we will likely 1035 * unresolve the namecache entry, and a vrele/vput is needed to 1036 * trigger the tmpfs_inactive/tmpfs_reclaim sequence. 1037 * 1038 * We have to use vget to clear any inactive state on the vnode, 1039 * otherwise the vnode may remain inactive and thus tmpfs_inactive 1040 * will not get called when we release it. 1041 */ 1042 error = cache_vget(ap->a_nch, ap->a_cred, LK_SHARED, &vp); 1043 KKASSERT(vp->v_mount == dvp->v_mount); 1044 KKASSERT(error == 0); 1045 vn_unlock(vp); 1046 1047 if (vp->v_type == VDIR) { 1048 error = EISDIR; 1049 goto out2; 1050 } 1051 1052 dnode = VP_TO_TMPFS_DIR(dvp); 1053 node = VP_TO_TMPFS_NODE(vp); 1054 tmp = VFS_TO_TMPFS(vp->v_mount); 1055 1056 TMPFS_NODE_LOCK(dnode); 1057 de = tmpfs_dir_lookup(dnode, node, ncp); 1058 if (de == NULL) { 1059 error = ENOENT; 1060 TMPFS_NODE_UNLOCK(dnode); 1061 goto out; 1062 } 1063 1064 /* Files marked as immutable or append-only cannot be deleted. */ 1065 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) || 1066 (dnode->tn_flags & APPEND)) { 1067 error = EPERM; 1068 TMPFS_NODE_UNLOCK(dnode); 1069 goto out; 1070 } 1071 1072 /* Remove the entry from the directory; as it is a file, we do not 1073 * have to change the number of hard links of the directory. */ 1074 tmpfs_dir_detach(dnode, de); 1075 TMPFS_NODE_UNLOCK(dnode); 1076 1077 /* Free the directory entry we just deleted. Note that the node 1078 * referred by it will not be removed until the vnode is really 1079 * reclaimed. */ 1080 tmpfs_free_dirent(tmp, de); 1081 1082 if (node->tn_links > 0) { 1083 TMPFS_NODE_LOCK(node); 1084 node->tn_status |= TMPFS_NODE_CHANGED; 1085 TMPFS_NODE_UNLOCK(node); 1086 } 1087 1088 cache_unlink(ap->a_nch); 1089 tmpfs_knote(vp, NOTE_DELETE); 1090 error = 0; 1091 1092 out: 1093 if (error == 0) 1094 tmpfs_knote(dvp, NOTE_WRITE); 1095 out2: 1096 vrele(vp); 1097 1098 return error; 1099 } 1100 1101 /* --------------------------------------------------------------------- */ 1102 1103 static int 1104 tmpfs_nlink(struct vop_nlink_args *ap) 1105 { 1106 struct vnode *dvp = ap->a_dvp; 1107 struct vnode *vp = ap->a_vp; 1108 struct namecache *ncp = ap->a_nch->ncp; 1109 struct tmpfs_dirent *de; 1110 struct tmpfs_node *node; 1111 struct tmpfs_node *dnode; 1112 int error; 1113 1114 KKASSERT(dvp != vp); /* XXX When can this be false? */ 1115 1116 node = VP_TO_TMPFS_NODE(vp); 1117 dnode = VP_TO_TMPFS_NODE(dvp); 1118 TMPFS_NODE_LOCK(dnode); 1119 1120 /* XXX: Why aren't the following two tests done by the caller? */ 1121 1122 /* Hard links of directories are forbidden. */ 1123 if (vp->v_type == VDIR) { 1124 error = EPERM; 1125 goto out; 1126 } 1127 1128 /* Cannot create cross-device links. */ 1129 if (dvp->v_mount != vp->v_mount) { 1130 error = EXDEV; 1131 goto out; 1132 } 1133 1134 /* Ensure that we do not overflow the maximum number of links imposed 1135 * by the system. */ 1136 KKASSERT(node->tn_links <= LINK_MAX); 1137 if (node->tn_links >= LINK_MAX) { 1138 error = EMLINK; 1139 goto out; 1140 } 1141 1142 /* We cannot create links of files marked immutable or append-only. */ 1143 if (node->tn_flags & (IMMUTABLE | APPEND)) { 1144 error = EPERM; 1145 goto out; 1146 } 1147 1148 /* Allocate a new directory entry to represent the node. */ 1149 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node, 1150 ncp->nc_name, ncp->nc_nlen, &de); 1151 if (error != 0) 1152 goto out; 1153 1154 /* Insert the new directory entry into the appropriate directory. */ 1155 tmpfs_dir_attach(dnode, de); 1156 1157 /* vp link count has changed, so update node times. */ 1158 1159 TMPFS_NODE_LOCK(node); 1160 node->tn_status |= TMPFS_NODE_CHANGED; 1161 TMPFS_NODE_UNLOCK(node); 1162 tmpfs_update(vp); 1163 1164 tmpfs_knote(vp, NOTE_LINK); 1165 cache_setunresolved(ap->a_nch); 1166 cache_setvp(ap->a_nch, vp); 1167 error = 0; 1168 1169 out: 1170 TMPFS_NODE_UNLOCK(dnode); 1171 if (error == 0) 1172 tmpfs_knote(dvp, NOTE_WRITE); 1173 return error; 1174 } 1175 1176 /* --------------------------------------------------------------------- */ 1177 1178 static int 1179 tmpfs_nrename(struct vop_nrename_args *ap) 1180 { 1181 struct vnode *fdvp = ap->a_fdvp; 1182 struct namecache *fncp = ap->a_fnch->ncp; 1183 struct vnode *fvp = fncp->nc_vp; 1184 struct vnode *tdvp = ap->a_tdvp; 1185 struct namecache *tncp = ap->a_tnch->ncp; 1186 struct vnode *tvp; 1187 struct tmpfs_dirent *de, *tde; 1188 struct tmpfs_mount *tmp; 1189 struct tmpfs_node *fdnode; 1190 struct tmpfs_node *fnode; 1191 struct tmpfs_node *tnode; 1192 struct tmpfs_node *tdnode; 1193 char *newname; 1194 char *oldname; 1195 int error; 1196 1197 KKASSERT(fdvp->v_mount == fvp->v_mount); 1198 1199 /* 1200 * Because tvp can get overwritten we have to vget it instead of 1201 * just vref or use it, otherwise it's VINACTIVE flag may not get 1202 * cleared and the node won't get destroyed. 1203 */ 1204 error = cache_vget(ap->a_tnch, ap->a_cred, LK_SHARED, &tvp); 1205 if (error == 0) { 1206 tnode = VP_TO_TMPFS_NODE(tvp); 1207 vn_unlock(tvp); 1208 } else { 1209 tnode = NULL; 1210 } 1211 1212 /* Disallow cross-device renames. 1213 * XXX Why isn't this done by the caller? */ 1214 if (fvp->v_mount != tdvp->v_mount || 1215 (tvp != NULL && fvp->v_mount != tvp->v_mount)) { 1216 error = EXDEV; 1217 goto out; 1218 } 1219 1220 tmp = VFS_TO_TMPFS(tdvp->v_mount); 1221 tdnode = VP_TO_TMPFS_DIR(tdvp); 1222 1223 /* If source and target are the same file, there is nothing to do. */ 1224 if (fvp == tvp) { 1225 error = 0; 1226 goto out; 1227 } 1228 1229 fdnode = VP_TO_TMPFS_DIR(fdvp); 1230 fnode = VP_TO_TMPFS_NODE(fvp); 1231 TMPFS_NODE_LOCK(fdnode); 1232 de = tmpfs_dir_lookup(fdnode, fnode, fncp); 1233 TMPFS_NODE_UNLOCK(fdnode); /* XXX depend on namecache lock */ 1234 1235 /* Avoid manipulating '.' and '..' entries. */ 1236 if (de == NULL) { 1237 error = ENOENT; 1238 goto out_locked; 1239 } 1240 KKASSERT(de->td_node == fnode); 1241 1242 /* 1243 * If replacing an entry in the target directory and that entry 1244 * is a directory, it must be empty. 1245 * 1246 * Kern_rename gurantees the destination to be a directory 1247 * if the source is one (it does?). 1248 */ 1249 if (tvp != NULL) { 1250 KKASSERT(tnode != NULL); 1251 1252 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 1253 (tdnode->tn_flags & (APPEND | IMMUTABLE))) { 1254 error = EPERM; 1255 goto out_locked; 1256 } 1257 1258 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) { 1259 if (tnode->tn_size > 0) { 1260 error = ENOTEMPTY; 1261 goto out_locked; 1262 } 1263 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) { 1264 error = ENOTDIR; 1265 goto out_locked; 1266 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) { 1267 error = EISDIR; 1268 goto out_locked; 1269 } else { 1270 KKASSERT(fnode->tn_type != VDIR && 1271 tnode->tn_type != VDIR); 1272 } 1273 } 1274 1275 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 1276 (fdnode->tn_flags & (APPEND | IMMUTABLE))) { 1277 error = EPERM; 1278 goto out_locked; 1279 } 1280 1281 /* 1282 * Ensure that we have enough memory to hold the new name, if it 1283 * has to be changed. 1284 */ 1285 if (fncp->nc_nlen != tncp->nc_nlen || 1286 bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) { 1287 newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone, 1288 M_WAITOK | M_NULLOK); 1289 if (newname == NULL) { 1290 error = ENOSPC; 1291 goto out_locked; 1292 } 1293 bcopy(tncp->nc_name, newname, tncp->nc_nlen); 1294 newname[tncp->nc_nlen] = '\0'; 1295 } else { 1296 newname = NULL; 1297 } 1298 1299 /* 1300 * Unlink entry from source directory. Note that the kernel has 1301 * already checked for illegal recursion cases (renaming a directory 1302 * into a subdirectory of itself). 1303 */ 1304 if (fdnode != tdnode) { 1305 tmpfs_dir_detach(fdnode, de); 1306 } else { 1307 /* XXX depend on namecache lock */ 1308 TMPFS_NODE_LOCK(fdnode); 1309 KKASSERT(de == tmpfs_dir_lookup(fdnode, fnode, fncp)); 1310 RB_REMOVE(tmpfs_dirtree, &fdnode->tn_dir.tn_dirtree, de); 1311 RB_REMOVE(tmpfs_dirtree_cookie, 1312 &fdnode->tn_dir.tn_cookietree, de); 1313 TMPFS_NODE_UNLOCK(fdnode); 1314 } 1315 1316 /* 1317 * Handle any name change. Swap with newname, we will 1318 * deallocate it at the end. 1319 */ 1320 if (newname != NULL) { 1321 #if 0 1322 TMPFS_NODE_LOCK(fnode); 1323 fnode->tn_status |= TMPFS_NODE_CHANGED; 1324 TMPFS_NODE_UNLOCK(fnode); 1325 #endif 1326 oldname = de->td_name; 1327 de->td_name = newname; 1328 de->td_namelen = (uint16_t)tncp->nc_nlen; 1329 newname = oldname; 1330 } 1331 1332 /* 1333 * If we are overwriting an entry, we have to remove the old one 1334 * from the target directory. 1335 */ 1336 if (tvp != NULL) { 1337 /* Remove the old entry from the target directory. */ 1338 TMPFS_NODE_LOCK(tdnode); 1339 tde = tmpfs_dir_lookup(tdnode, tnode, tncp); 1340 tmpfs_dir_detach(tdnode, tde); 1341 TMPFS_NODE_UNLOCK(tdnode); 1342 tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE); 1343 1344 /* 1345 * Free the directory entry we just deleted. Note that the 1346 * node referred by it will not be removed until the vnode is 1347 * really reclaimed. 1348 */ 1349 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde); 1350 /*cache_inval_vp(tvp, CINV_DESTROY);*/ 1351 } 1352 1353 /* 1354 * Link entry to target directory. If the entry 1355 * represents a directory move the parent linkage 1356 * as well. 1357 */ 1358 if (fdnode != tdnode) { 1359 if (de->td_node->tn_type == VDIR) { 1360 TMPFS_VALIDATE_DIR(fnode); 1361 } 1362 tmpfs_dir_attach(tdnode, de); 1363 } else { 1364 TMPFS_NODE_LOCK(tdnode); 1365 tdnode->tn_status |= TMPFS_NODE_MODIFIED; 1366 RB_INSERT(tmpfs_dirtree, &tdnode->tn_dir.tn_dirtree, de); 1367 RB_INSERT(tmpfs_dirtree_cookie, 1368 &tdnode->tn_dir.tn_cookietree, de); 1369 TMPFS_NODE_UNLOCK(tdnode); 1370 } 1371 1372 /* 1373 * Finish up 1374 */ 1375 if (newname) { 1376 kfree(newname, tmp->tm_name_zone); 1377 newname = NULL; 1378 } 1379 cache_rename(ap->a_fnch, ap->a_tnch); 1380 tmpfs_knote(ap->a_fdvp, NOTE_WRITE); 1381 tmpfs_knote(ap->a_tdvp, NOTE_WRITE); 1382 if (fnode->tn_vnode) 1383 tmpfs_knote(fnode->tn_vnode, NOTE_RENAME); 1384 error = 0; 1385 1386 out_locked: 1387 ; 1388 out: 1389 if (tvp) 1390 vrele(tvp); 1391 return error; 1392 } 1393 1394 /* --------------------------------------------------------------------- */ 1395 1396 static int 1397 tmpfs_nmkdir(struct vop_nmkdir_args *ap) 1398 { 1399 struct vnode *dvp = ap->a_dvp; 1400 struct vnode **vpp = ap->a_vpp; 1401 struct namecache *ncp = ap->a_nch->ncp; 1402 struct vattr *vap = ap->a_vap; 1403 struct ucred *cred = ap->a_cred; 1404 int error; 1405 1406 KKASSERT(vap->va_type == VDIR); 1407 1408 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL); 1409 if (error == 0) { 1410 cache_setunresolved(ap->a_nch); 1411 cache_setvp(ap->a_nch, *vpp); 1412 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK); 1413 } 1414 return error; 1415 } 1416 1417 /* --------------------------------------------------------------------- */ 1418 1419 static int 1420 tmpfs_nrmdir(struct vop_nrmdir_args *ap) 1421 { 1422 struct vnode *dvp = ap->a_dvp; 1423 struct namecache *ncp = ap->a_nch->ncp; 1424 struct vnode *vp; 1425 struct tmpfs_dirent *de; 1426 struct tmpfs_mount *tmp; 1427 struct tmpfs_node *dnode; 1428 struct tmpfs_node *node; 1429 int error; 1430 1431 /* 1432 * We have to acquire the vp from ap->a_nch because we will likely 1433 * unresolve the namecache entry, and a vrele/vput is needed to 1434 * trigger the tmpfs_inactive/tmpfs_reclaim sequence. 1435 * 1436 * We have to use vget to clear any inactive state on the vnode, 1437 * otherwise the vnode may remain inactive and thus tmpfs_inactive 1438 * will not get called when we release it. 1439 */ 1440 error = cache_vget(ap->a_nch, ap->a_cred, LK_SHARED, &vp); 1441 KKASSERT(error == 0); 1442 vn_unlock(vp); 1443 1444 /* 1445 * Prevalidate so we don't hit an assertion later 1446 */ 1447 if (vp->v_type != VDIR) { 1448 error = ENOTDIR; 1449 goto out; 1450 } 1451 1452 tmp = VFS_TO_TMPFS(dvp->v_mount); 1453 dnode = VP_TO_TMPFS_DIR(dvp); 1454 node = VP_TO_TMPFS_DIR(vp); 1455 1456 /* 1457 * Directories with more than two entries ('.' and '..') cannot 1458 * be removed. 1459 */ 1460 if (node->tn_size > 0) { 1461 error = ENOTEMPTY; 1462 goto out; 1463 } 1464 1465 if ((dnode->tn_flags & APPEND) 1466 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1467 error = EPERM; 1468 goto out; 1469 } 1470 1471 /* 1472 * This invariant holds only if we are not trying to 1473 * remove "..". We checked for that above so this is safe now. 1474 */ 1475 KKASSERT(node->tn_dir.tn_parent == dnode); 1476 1477 /* 1478 * Get the directory entry associated with node (vp). This 1479 * was filled by tmpfs_lookup while looking up the entry. 1480 */ 1481 TMPFS_NODE_LOCK(dnode); 1482 de = tmpfs_dir_lookup(dnode, node, ncp); 1483 KKASSERT(TMPFS_DIRENT_MATCHES(de, ncp->nc_name, ncp->nc_nlen)); 1484 1485 /* Check flags to see if we are allowed to remove the directory. */ 1486 if ((dnode->tn_flags & APPEND) || 1487 node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) { 1488 error = EPERM; 1489 TMPFS_NODE_UNLOCK(dnode); 1490 goto out; 1491 } 1492 1493 /* Detach the directory entry from the directory (dnode). */ 1494 tmpfs_dir_detach(dnode, de); 1495 TMPFS_NODE_UNLOCK(dnode); 1496 1497 /* No vnode should be allocated for this entry from this point */ 1498 TMPFS_NODE_LOCK(dnode); 1499 TMPFS_ASSERT_ELOCKED(dnode); 1500 TMPFS_NODE_LOCK(node); 1501 TMPFS_ASSERT_ELOCKED(node); 1502 1503 /* 1504 * Must set parent linkage to NULL (tested by ncreate to disallow 1505 * the creation of new files/dirs in a deleted directory) 1506 */ 1507 node->tn_status |= TMPFS_NODE_CHANGED; 1508 1509 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 1510 TMPFS_NODE_MODIFIED; 1511 1512 TMPFS_NODE_UNLOCK(node); 1513 TMPFS_NODE_UNLOCK(dnode); 1514 1515 /* Free the directory entry we just deleted. Note that the node 1516 * referred by it will not be removed until the vnode is really 1517 * reclaimed. */ 1518 tmpfs_free_dirent(tmp, de); 1519 1520 /* Release the deleted vnode (will destroy the node, notify 1521 * interested parties and clean it from the cache). */ 1522 1523 TMPFS_NODE_LOCK(dnode); 1524 dnode->tn_status |= TMPFS_NODE_CHANGED; 1525 TMPFS_NODE_UNLOCK(dnode); 1526 tmpfs_update(dvp); 1527 1528 cache_unlink(ap->a_nch); 1529 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK); 1530 error = 0; 1531 1532 out: 1533 vrele(vp); 1534 1535 return error; 1536 } 1537 1538 /* --------------------------------------------------------------------- */ 1539 1540 static int 1541 tmpfs_nsymlink(struct vop_nsymlink_args *ap) 1542 { 1543 struct vnode *dvp = ap->a_dvp; 1544 struct vnode **vpp = ap->a_vpp; 1545 struct namecache *ncp = ap->a_nch->ncp; 1546 struct vattr *vap = ap->a_vap; 1547 struct ucred *cred = ap->a_cred; 1548 char *target = ap->a_target; 1549 int error; 1550 1551 vap->va_type = VLNK; 1552 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target); 1553 if (error == 0) { 1554 tmpfs_knote(*vpp, NOTE_WRITE); 1555 cache_setunresolved(ap->a_nch); 1556 cache_setvp(ap->a_nch, *vpp); 1557 } 1558 return error; 1559 } 1560 1561 /* --------------------------------------------------------------------- */ 1562 1563 static int 1564 tmpfs_readdir(struct vop_readdir_args *ap) 1565 { 1566 struct vnode *vp = ap->a_vp; 1567 struct uio *uio = ap->a_uio; 1568 int *eofflag = ap->a_eofflag; 1569 off_t **cookies = ap->a_cookies; 1570 int *ncookies = ap->a_ncookies; 1571 struct tmpfs_mount *tmp; 1572 int error; 1573 off_t startoff; 1574 off_t cnt = 0; 1575 struct tmpfs_node *node; 1576 1577 /* This operation only makes sense on directory nodes. */ 1578 if (vp->v_type != VDIR) { 1579 return ENOTDIR; 1580 } 1581 1582 tmp = VFS_TO_TMPFS(vp->v_mount); 1583 node = VP_TO_TMPFS_DIR(vp); 1584 startoff = uio->uio_offset; 1585 1586 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) { 1587 error = tmpfs_dir_getdotdent(node, uio); 1588 if (error != 0) { 1589 TMPFS_NODE_LOCK_SH(node); 1590 goto outok; 1591 } 1592 cnt++; 1593 } 1594 1595 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) { 1596 /* may lock parent, cannot hold node lock */ 1597 error = tmpfs_dir_getdotdotdent(tmp, node, uio); 1598 if (error != 0) { 1599 TMPFS_NODE_LOCK_SH(node); 1600 goto outok; 1601 } 1602 cnt++; 1603 } 1604 1605 TMPFS_NODE_LOCK_SH(node); 1606 error = tmpfs_dir_getdents(node, uio, &cnt); 1607 1608 outok: 1609 KKASSERT(error >= -1); 1610 1611 if (error == -1) 1612 error = 0; 1613 1614 if (eofflag != NULL) 1615 *eofflag = 1616 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF); 1617 1618 /* Update NFS-related variables. */ 1619 if (error == 0 && cookies != NULL && ncookies != NULL) { 1620 off_t i; 1621 off_t off = startoff; 1622 struct tmpfs_dirent *de = NULL; 1623 1624 *ncookies = cnt; 1625 *cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK); 1626 1627 for (i = 0; i < cnt; i++) { 1628 KKASSERT(off != TMPFS_DIRCOOKIE_EOF); 1629 if (off == TMPFS_DIRCOOKIE_DOT) { 1630 off = TMPFS_DIRCOOKIE_DOTDOT; 1631 } else { 1632 if (off == TMPFS_DIRCOOKIE_DOTDOT) { 1633 de = RB_MIN(tmpfs_dirtree_cookie, 1634 &node->tn_dir.tn_cookietree); 1635 } else if (de != NULL) { 1636 de = RB_NEXT(tmpfs_dirtree_cookie, 1637 &node->tn_dir.tn_cookietree, de); 1638 } else { 1639 de = tmpfs_dir_lookupbycookie(node, 1640 off); 1641 KKASSERT(de != NULL); 1642 de = RB_NEXT(tmpfs_dirtree_cookie, 1643 &node->tn_dir.tn_cookietree, de); 1644 } 1645 if (de == NULL) 1646 off = TMPFS_DIRCOOKIE_EOF; 1647 else 1648 off = tmpfs_dircookie(de); 1649 } 1650 (*cookies)[i] = off; 1651 } 1652 KKASSERT(uio->uio_offset == off); 1653 } 1654 TMPFS_NODE_UNLOCK(node); 1655 1656 if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) { 1657 TMPFS_NODE_LOCK(node); 1658 node->tn_status |= TMPFS_NODE_ACCESSED; 1659 TMPFS_NODE_UNLOCK(node); 1660 } 1661 return error; 1662 } 1663 1664 /* --------------------------------------------------------------------- */ 1665 1666 static int 1667 tmpfs_readlink(struct vop_readlink_args *ap) 1668 { 1669 struct vnode *vp = ap->a_vp; 1670 struct uio *uio = ap->a_uio; 1671 int error; 1672 struct tmpfs_node *node; 1673 1674 KKASSERT(uio->uio_offset == 0); 1675 KKASSERT(vp->v_type == VLNK); 1676 1677 node = VP_TO_TMPFS_NODE(vp); 1678 TMPFS_NODE_LOCK_SH(node); 1679 error = uiomove(node->tn_link, 1680 MIN(node->tn_size, uio->uio_resid), uio); 1681 TMPFS_NODE_UNLOCK(node); 1682 if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) { 1683 TMPFS_NODE_LOCK(node); 1684 node->tn_status |= TMPFS_NODE_ACCESSED; 1685 TMPFS_NODE_UNLOCK(node); 1686 } 1687 return error; 1688 } 1689 1690 /* --------------------------------------------------------------------- */ 1691 1692 static int 1693 tmpfs_inactive(struct vop_inactive_args *ap) 1694 { 1695 struct vnode *vp = ap->a_vp; 1696 struct tmpfs_node *node; 1697 struct mount *mp; 1698 1699 mp = vp->v_mount; 1700 lwkt_gettoken(&mp->mnt_token); 1701 node = VP_TO_TMPFS_NODE(vp); 1702 1703 /* 1704 * Degenerate case 1705 */ 1706 if (node == NULL) { 1707 vrecycle(vp); 1708 lwkt_reltoken(&mp->mnt_token); 1709 return(0); 1710 } 1711 1712 /* 1713 * Get rid of unreferenced deleted vnodes sooner rather than 1714 * later so the data memory can be recovered immediately. 1715 * 1716 * We must truncate the vnode to prevent the normal reclamation 1717 * path from flushing the data for the removed file to disk. 1718 */ 1719 TMPFS_NODE_LOCK(node); 1720 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 && 1721 node->tn_links == 0) 1722 { 1723 node->tn_vpstate = TMPFS_VNODE_DOOMED; 1724 TMPFS_NODE_UNLOCK(node); 1725 if (node->tn_type == VREG) 1726 tmpfs_truncate(vp, 0); 1727 vrecycle(vp); 1728 } else { 1729 /* 1730 * We must retain any VM pages belonging to the vnode's 1731 * object as the vnode will destroy the object during a 1732 * later reclaim. We call vinvalbuf(V_SAVE) to clean 1733 * out the buffer cache. 1734 * 1735 * On DragonFlyBSD, vnodes are not immediately deactivated 1736 * on the 1->0 refs, so this is a relatively optimal 1737 * operation. We have to do this in tmpfs_inactive() 1738 * because the pages will have already been thrown away 1739 * at the time tmpfs_reclaim() is called. 1740 */ 1741 if (node->tn_type == VREG && 1742 node->tn_reg.tn_pages_in_aobj == 0) { 1743 vinvalbuf(vp, V_SAVE, 0, 0); 1744 KKASSERT(RB_EMPTY(&vp->v_rbdirty_tree)); 1745 KKASSERT(RB_EMPTY(&vp->v_rbclean_tree)); 1746 tmpfs_move_pages(vp->v_object, node->tn_reg.tn_aobj, 1747 TMPFS_MOVF_DEACTIVATE); 1748 node->tn_reg.tn_pages_in_aobj = 1; 1749 } 1750 1751 TMPFS_NODE_UNLOCK(node); 1752 } 1753 lwkt_reltoken(&mp->mnt_token); 1754 1755 return 0; 1756 } 1757 1758 /* --------------------------------------------------------------------- */ 1759 1760 int 1761 tmpfs_reclaim(struct vop_reclaim_args *ap) 1762 { 1763 struct vnode *vp = ap->a_vp; 1764 struct tmpfs_mount *tmp; 1765 struct tmpfs_node *node; 1766 struct mount *mp; 1767 1768 mp = vp->v_mount; 1769 lwkt_gettoken(&mp->mnt_token); 1770 1771 node = VP_TO_TMPFS_NODE(vp); 1772 tmp = VFS_TO_TMPFS(vp->v_mount); 1773 KKASSERT(mp == tmp->tm_mount); 1774 1775 TMPFS_NODE_LOCK(node); 1776 KKASSERT(node->tn_vnode == vp); 1777 node->tn_vnode = NULL; 1778 vp->v_data = NULL; 1779 1780 /* 1781 * If the node referenced by this vnode was deleted by the 1782 * user, we must free its associated data structures now that 1783 * the vnode is being reclaimed. 1784 * 1785 * Directories have an extra link ref. 1786 */ 1787 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 && 1788 node->tn_links == 0) { 1789 node->tn_vpstate = TMPFS_VNODE_DOOMED; 1790 tmpfs_free_node(tmp, node); 1791 /* eats the lock */ 1792 } else { 1793 TMPFS_NODE_UNLOCK(node); 1794 } 1795 lwkt_reltoken(&mp->mnt_token); 1796 1797 KKASSERT(vp->v_data == NULL); 1798 return 0; 1799 } 1800 1801 /* --------------------------------------------------------------------- */ 1802 1803 static int 1804 tmpfs_mountctl(struct vop_mountctl_args *ap) 1805 { 1806 struct tmpfs_mount *tmp; 1807 struct mount *mp; 1808 int rc; 1809 1810 mp = ap->a_head.a_ops->head.vv_mount; 1811 lwkt_gettoken(&mp->mnt_token); 1812 1813 switch (ap->a_op) { 1814 case (MOUNTCTL_SET_EXPORT): 1815 tmp = (struct tmpfs_mount *) mp->mnt_data; 1816 1817 if (ap->a_ctllen != sizeof(struct export_args)) 1818 rc = (EINVAL); 1819 else 1820 rc = vfs_export(mp, &tmp->tm_export, 1821 (const struct export_args *) ap->a_ctl); 1822 break; 1823 default: 1824 rc = vop_stdmountctl(ap); 1825 break; 1826 } 1827 1828 lwkt_reltoken(&mp->mnt_token); 1829 return (rc); 1830 } 1831 1832 /* --------------------------------------------------------------------- */ 1833 1834 static int 1835 tmpfs_print(struct vop_print_args *ap) 1836 { 1837 struct vnode *vp = ap->a_vp; 1838 1839 struct tmpfs_node *node; 1840 1841 node = VP_TO_TMPFS_NODE(vp); 1842 1843 kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n", 1844 node, node->tn_flags, node->tn_links); 1845 kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n", 1846 node->tn_mode, node->tn_uid, node->tn_gid, 1847 (uintmax_t)node->tn_size, node->tn_status); 1848 1849 if (vp->v_type == VFIFO) 1850 fifo_printinfo(vp); 1851 1852 kprintf("\n"); 1853 1854 return 0; 1855 } 1856 1857 /* --------------------------------------------------------------------- */ 1858 1859 static int 1860 tmpfs_pathconf(struct vop_pathconf_args *ap) 1861 { 1862 struct vnode *vp = ap->a_vp; 1863 int name = ap->a_name; 1864 register_t *retval = ap->a_retval; 1865 struct tmpfs_mount *tmp; 1866 int error; 1867 1868 error = 0; 1869 1870 switch (name) { 1871 case _PC_CHOWN_RESTRICTED: 1872 *retval = 1; 1873 break; 1874 1875 case _PC_FILESIZEBITS: 1876 tmp = VFS_TO_TMPFS(vp->v_mount); 1877 *retval = max(32, flsll(tmp->tm_pages_max * PAGE_SIZE) + 1); 1878 break; 1879 1880 case _PC_LINK_MAX: 1881 *retval = LINK_MAX; 1882 break; 1883 1884 case _PC_NAME_MAX: 1885 *retval = NAME_MAX; 1886 break; 1887 1888 case _PC_NO_TRUNC: 1889 *retval = 1; 1890 break; 1891 1892 case _PC_PATH_MAX: 1893 *retval = PATH_MAX; 1894 break; 1895 1896 case _PC_PIPE_BUF: 1897 *retval = PIPE_BUF; 1898 break; 1899 1900 case _PC_SYNC_IO: 1901 *retval = 1; 1902 break; 1903 1904 case _PC_2_SYMLINKS: 1905 *retval = 1; 1906 break; 1907 1908 default: 1909 error = EINVAL; 1910 } 1911 1912 return error; 1913 } 1914 1915 /************************************************************************ 1916 * KQFILTER OPS * 1917 ************************************************************************/ 1918 1919 static void filt_tmpfsdetach(struct knote *kn); 1920 static int filt_tmpfsread(struct knote *kn, long hint); 1921 static int filt_tmpfswrite(struct knote *kn, long hint); 1922 static int filt_tmpfsvnode(struct knote *kn, long hint); 1923 1924 static struct filterops tmpfsread_filtops = 1925 { FILTEROP_ISFD | FILTEROP_MPSAFE, 1926 NULL, filt_tmpfsdetach, filt_tmpfsread }; 1927 static struct filterops tmpfswrite_filtops = 1928 { FILTEROP_ISFD | FILTEROP_MPSAFE, 1929 NULL, filt_tmpfsdetach, filt_tmpfswrite }; 1930 static struct filterops tmpfsvnode_filtops = 1931 { FILTEROP_ISFD | FILTEROP_MPSAFE, 1932 NULL, filt_tmpfsdetach, filt_tmpfsvnode }; 1933 1934 static int 1935 tmpfs_kqfilter (struct vop_kqfilter_args *ap) 1936 { 1937 struct vnode *vp = ap->a_vp; 1938 struct knote *kn = ap->a_kn; 1939 1940 switch (kn->kn_filter) { 1941 case EVFILT_READ: 1942 kn->kn_fop = &tmpfsread_filtops; 1943 break; 1944 case EVFILT_WRITE: 1945 kn->kn_fop = &tmpfswrite_filtops; 1946 break; 1947 case EVFILT_VNODE: 1948 kn->kn_fop = &tmpfsvnode_filtops; 1949 break; 1950 default: 1951 return (EOPNOTSUPP); 1952 } 1953 1954 kn->kn_hook = (caddr_t)vp; 1955 1956 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 1957 1958 return(0); 1959 } 1960 1961 static void 1962 filt_tmpfsdetach(struct knote *kn) 1963 { 1964 struct vnode *vp = (void *)kn->kn_hook; 1965 1966 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 1967 } 1968 1969 static int 1970 filt_tmpfsread(struct knote *kn, long hint) 1971 { 1972 struct vnode *vp = (void *)kn->kn_hook; 1973 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp); 1974 off_t off; 1975 1976 if (hint == NOTE_REVOKE) { 1977 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 1978 return(1); 1979 } 1980 1981 /* 1982 * Interlock against MP races when performing this function. 1983 */ 1984 TMPFS_NODE_LOCK_SH(node); 1985 off = node->tn_size - kn->kn_fp->f_offset; 1986 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 1987 if (kn->kn_sfflags & NOTE_OLDAPI) { 1988 TMPFS_NODE_UNLOCK(node); 1989 return(1); 1990 } 1991 if (kn->kn_data == 0) { 1992 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 1993 } 1994 TMPFS_NODE_UNLOCK(node); 1995 return (kn->kn_data != 0); 1996 } 1997 1998 static int 1999 filt_tmpfswrite(struct knote *kn, long hint) 2000 { 2001 if (hint == NOTE_REVOKE) 2002 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2003 kn->kn_data = 0; 2004 return (1); 2005 } 2006 2007 static int 2008 filt_tmpfsvnode(struct knote *kn, long hint) 2009 { 2010 if (kn->kn_sfflags & hint) 2011 kn->kn_fflags |= hint; 2012 if (hint == NOTE_REVOKE) { 2013 kn->kn_flags |= (EV_EOF | EV_NODATA); 2014 return (1); 2015 } 2016 return (kn->kn_fflags != 0); 2017 } 2018 2019 /* 2020 * Helper to move VM pages between objects 2021 * 2022 * NOTE: The vm_page_rename() dirties the page, so we can clear the 2023 * PG_NEED_COMMIT flag. If the pages are being moved into tn_aobj, 2024 * the pageout daemon will be able to page them out. 2025 */ 2026 static int 2027 tmpfs_move_pages_callback(vm_page_t p, void *data) 2028 { 2029 struct rb_vm_page_scan_info *info = data; 2030 vm_pindex_t pindex; 2031 2032 pindex = p->pindex; 2033 if (vm_page_busy_try(p, TRUE)) { 2034 vm_page_sleep_busy(p, TRUE, "tpgmov"); 2035 info->error = -1; 2036 return -1; 2037 } 2038 if (p->object != info->object || p->pindex != pindex) { 2039 vm_page_wakeup(p); 2040 info->error = -1; 2041 return -1; 2042 } 2043 2044 if ((info->pagerflags & TMPFS_MOVF_FROMBACKING) && 2045 (p->flags & PG_SWAPPED) && 2046 (p->flags & PG_NEED_COMMIT) == 0 && 2047 p->dirty == 0) { 2048 /* 2049 * If the page in the backing aobj was paged out to swap 2050 * it will be clean and it is better to free it rather 2051 * than re-dirty it. We will assume that the page was 2052 * paged out to swap for a reason! 2053 * 2054 * This helps avoid unnecessary swap thrashing on the page. 2055 */ 2056 vm_page_free(p); 2057 } else if ((info->pagerflags & TMPFS_MOVF_FROMBACKING) == 0 && 2058 (p->flags & PG_NEED_COMMIT) == 0 && 2059 p->dirty == 0) { 2060 /* 2061 * If the page associated with the vnode was cleaned via 2062 * a tmpfs_strategy() call, it exists as a swap block in 2063 * aobj and it is again better to free it rather than 2064 * re-dirty it. We will assume that the page was 2065 * paged out to swap for a reason! 2066 * 2067 * This helps avoid unnecessary swap thrashing on the page. 2068 */ 2069 vm_page_free(p); 2070 } else { 2071 /* 2072 * Rename the page, which will also ensure that it is flagged 2073 * as dirty and check whether a swap block association exists 2074 * in the target object or not, setting appropriate flags if 2075 * it does. 2076 */ 2077 vm_page_rename(p, info->dest_object, pindex); 2078 vm_page_clear_commit(p); 2079 if (info->pagerflags & TMPFS_MOVF_DEACTIVATE) 2080 vm_page_deactivate(p); 2081 vm_page_wakeup(p); 2082 /* page automaticaly made dirty */ 2083 } 2084 2085 return 0; 2086 } 2087 2088 static 2089 void 2090 tmpfs_move_pages(vm_object_t src, vm_object_t dst, int movflags) 2091 { 2092 struct rb_vm_page_scan_info info; 2093 2094 vm_object_hold(src); 2095 vm_object_hold(dst); 2096 info.object = src; 2097 info.dest_object = dst; 2098 info.pagerflags = movflags; 2099 do { 2100 if (src->paging_in_progress) 2101 vm_object_pip_wait(src, "objtfs"); 2102 info.error = 1; 2103 vm_page_rb_tree_RB_SCAN(&src->rb_memq, NULL, 2104 tmpfs_move_pages_callback, &info); 2105 } while (info.error < 0 || !RB_EMPTY(&src->rb_memq) || 2106 src->paging_in_progress); 2107 vm_object_drop(dst); 2108 vm_object_drop(src); 2109 } 2110 2111 /* --------------------------------------------------------------------- */ 2112 2113 /* 2114 * vnode operations vector used for files stored in a tmpfs file system. 2115 */ 2116 struct vop_ops tmpfs_vnode_vops = { 2117 .vop_default = vop_defaultop, 2118 .vop_getpages = vop_stdgetpages, 2119 .vop_putpages = vop_stdputpages, 2120 .vop_ncreate = tmpfs_ncreate, 2121 .vop_nresolve = tmpfs_nresolve, 2122 .vop_nlookupdotdot = tmpfs_nlookupdotdot, 2123 .vop_nmknod = tmpfs_nmknod, 2124 .vop_open = tmpfs_open, 2125 .vop_close = tmpfs_close, 2126 .vop_access = tmpfs_access, 2127 .vop_getattr = tmpfs_getattr, 2128 .vop_setattr = tmpfs_setattr, 2129 .vop_read = tmpfs_read, 2130 .vop_write = tmpfs_write, 2131 .vop_fsync = tmpfs_fsync, 2132 .vop_mountctl = tmpfs_mountctl, 2133 .vop_nremove = tmpfs_nremove, 2134 .vop_nlink = tmpfs_nlink, 2135 .vop_nrename = tmpfs_nrename, 2136 .vop_nmkdir = tmpfs_nmkdir, 2137 .vop_nrmdir = tmpfs_nrmdir, 2138 .vop_nsymlink = tmpfs_nsymlink, 2139 .vop_readdir = tmpfs_readdir, 2140 .vop_readlink = tmpfs_readlink, 2141 .vop_inactive = tmpfs_inactive, 2142 .vop_reclaim = tmpfs_reclaim, 2143 .vop_print = tmpfs_print, 2144 .vop_pathconf = tmpfs_pathconf, 2145 .vop_bmap = tmpfs_bmap, 2146 .vop_strategy = tmpfs_strategy, 2147 .vop_advlock = tmpfs_advlock, 2148 .vop_kqfilter = tmpfs_kqfilter 2149 }; 2150