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