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