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