1 /* $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $ */ 2 3 /*- 4 * SPDX-License-Identifier: BSD-2-Clause-NetBSD 5 * 6 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc. 7 * All rights reserved. 8 * 9 * This code is derived from software contributed to The NetBSD Foundation 10 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code 11 * 2005 program. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 24 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 25 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 26 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 28 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 30 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 31 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 32 * POSSIBILITY OF SUCH DAMAGE. 33 */ 34 35 /* 36 * tmpfs vnode interface. 37 */ 38 #include <sys/cdefs.h> 39 __FBSDID("$FreeBSD$"); 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/dirent.h> 44 #include <sys/fcntl.h> 45 #include <sys/file.h> 46 #include <sys/limits.h> 47 #include <sys/lockf.h> 48 #include <sys/lock.h> 49 #include <sys/mount.h> 50 #include <sys/namei.h> 51 #include <sys/priv.h> 52 #include <sys/proc.h> 53 #include <sys/rwlock.h> 54 #include <sys/sched.h> 55 #include <sys/stat.h> 56 #include <sys/sysctl.h> 57 #include <sys/unistd.h> 58 #include <sys/vnode.h> 59 #include <sys/smr.h> 60 #include <security/audit/audit.h> 61 #include <security/mac/mac_framework.h> 62 63 #include <vm/vm.h> 64 #include <vm/vm_param.h> 65 #include <vm/vm_object.h> 66 67 #include <fs/tmpfs/tmpfs_vnops.h> 68 #include <fs/tmpfs/tmpfs.h> 69 70 SYSCTL_DECL(_vfs_tmpfs); 71 VFS_SMR_DECLARE; 72 73 static volatile int tmpfs_rename_restarts; 74 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, rename_restarts, CTLFLAG_RD, 75 __DEVOLATILE(int *, &tmpfs_rename_restarts), 0, 76 "Times rename had to restart due to lock contention"); 77 78 static int 79 tmpfs_vn_get_ino_alloc(struct mount *mp, void *arg, int lkflags, 80 struct vnode **rvp) 81 { 82 83 return (tmpfs_alloc_vp(mp, arg, lkflags, rvp)); 84 } 85 86 static int 87 tmpfs_lookup1(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) 88 { 89 struct tmpfs_dirent *de; 90 struct tmpfs_node *dnode, *pnode; 91 struct tmpfs_mount *tm; 92 int error; 93 94 /* Caller assumes responsibility for ensuring access (VEXEC). */ 95 dnode = VP_TO_TMPFS_DIR(dvp); 96 *vpp = NULLVP; 97 98 /* We cannot be requesting the parent directory of the root node. */ 99 MPASS(IMPLIES(dnode->tn_type == VDIR && 100 dnode->tn_dir.tn_parent == dnode, 101 !(cnp->cn_flags & ISDOTDOT))); 102 103 TMPFS_ASSERT_LOCKED(dnode); 104 if (dnode->tn_dir.tn_parent == NULL) { 105 error = ENOENT; 106 goto out; 107 } 108 if (cnp->cn_flags & ISDOTDOT) { 109 tm = VFS_TO_TMPFS(dvp->v_mount); 110 pnode = dnode->tn_dir.tn_parent; 111 tmpfs_ref_node(pnode); 112 error = vn_vget_ino_gen(dvp, tmpfs_vn_get_ino_alloc, 113 pnode, cnp->cn_lkflags, vpp); 114 tmpfs_free_node(tm, pnode); 115 if (error != 0) 116 goto out; 117 } else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 118 VREF(dvp); 119 *vpp = dvp; 120 error = 0; 121 } else { 122 de = tmpfs_dir_lookup(dnode, NULL, cnp); 123 if (de != NULL && de->td_node == NULL) 124 cnp->cn_flags |= ISWHITEOUT; 125 if (de == NULL || de->td_node == NULL) { 126 /* 127 * The entry was not found in the directory. 128 * This is OK if we are creating or renaming an 129 * entry and are working on the last component of 130 * the path name. 131 */ 132 if ((cnp->cn_flags & ISLASTCN) && 133 (cnp->cn_nameiop == CREATE || \ 134 cnp->cn_nameiop == RENAME || 135 (cnp->cn_nameiop == DELETE && 136 cnp->cn_flags & DOWHITEOUT && 137 cnp->cn_flags & ISWHITEOUT))) { 138 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, 139 curthread); 140 if (error != 0) 141 goto out; 142 143 error = EJUSTRETURN; 144 } else 145 error = ENOENT; 146 } else { 147 struct tmpfs_node *tnode; 148 149 /* 150 * The entry was found, so get its associated 151 * tmpfs_node. 152 */ 153 tnode = de->td_node; 154 155 /* 156 * If we are not at the last path component and 157 * found a non-directory or non-link entry (which 158 * may itself be pointing to a directory), raise 159 * an error. 160 */ 161 if ((tnode->tn_type != VDIR && 162 tnode->tn_type != VLNK) && 163 !(cnp->cn_flags & ISLASTCN)) { 164 error = ENOTDIR; 165 goto out; 166 } 167 168 /* 169 * If we are deleting or renaming the entry, keep 170 * track of its tmpfs_dirent so that it can be 171 * easily deleted later. 172 */ 173 if ((cnp->cn_flags & ISLASTCN) && 174 (cnp->cn_nameiop == DELETE || 175 cnp->cn_nameiop == RENAME)) { 176 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, 177 curthread); 178 if (error != 0) 179 goto out; 180 181 /* Allocate a new vnode on the matching entry. */ 182 error = tmpfs_alloc_vp(dvp->v_mount, tnode, 183 cnp->cn_lkflags, vpp); 184 if (error != 0) 185 goto out; 186 187 if ((dnode->tn_mode & S_ISTXT) && 188 VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, 189 curthread) && VOP_ACCESS(*vpp, VADMIN, 190 cnp->cn_cred, curthread)) { 191 error = EPERM; 192 vput(*vpp); 193 *vpp = NULL; 194 goto out; 195 } 196 } else { 197 error = tmpfs_alloc_vp(dvp->v_mount, tnode, 198 cnp->cn_lkflags, vpp); 199 if (error != 0) 200 goto out; 201 } 202 } 203 } 204 205 /* 206 * Store the result of this lookup in the cache. Avoid this if the 207 * request was for creation, as it does not improve timings on 208 * emprical tests. 209 */ 210 if ((cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp)) 211 cache_enter(dvp, *vpp, cnp); 212 213 out: 214 /* 215 * If there were no errors, *vpp cannot be null and it must be 216 * locked. 217 */ 218 MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp))); 219 220 return (error); 221 } 222 223 static int 224 tmpfs_cached_lookup(struct vop_cachedlookup_args *v) 225 { 226 227 return (tmpfs_lookup1(v->a_dvp, v->a_vpp, v->a_cnp)); 228 } 229 230 static int 231 tmpfs_lookup(struct vop_lookup_args *v) 232 { 233 struct vnode *dvp = v->a_dvp; 234 struct vnode **vpp = v->a_vpp; 235 struct componentname *cnp = v->a_cnp; 236 int error; 237 238 /* Check accessibility of requested node as a first step. */ 239 error = vn_dir_check_exec(dvp, cnp); 240 if (error != 0) 241 return (error); 242 243 return (tmpfs_lookup1(dvp, vpp, cnp)); 244 } 245 246 static int 247 tmpfs_create(struct vop_create_args *v) 248 { 249 struct vnode *dvp = v->a_dvp; 250 struct vnode **vpp = v->a_vpp; 251 struct componentname *cnp = v->a_cnp; 252 struct vattr *vap = v->a_vap; 253 int error; 254 255 MPASS(vap->va_type == VREG || vap->va_type == VSOCK); 256 257 error = tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL); 258 if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp)) 259 cache_enter(dvp, *vpp, cnp); 260 return (error); 261 } 262 263 static int 264 tmpfs_mknod(struct vop_mknod_args *v) 265 { 266 struct vnode *dvp = v->a_dvp; 267 struct vnode **vpp = v->a_vpp; 268 struct componentname *cnp = v->a_cnp; 269 struct vattr *vap = v->a_vap; 270 271 if (vap->va_type != VBLK && vap->va_type != VCHR && 272 vap->va_type != VFIFO) 273 return (EINVAL); 274 275 return (tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL)); 276 } 277 278 struct fileops tmpfs_fnops; 279 280 static int 281 tmpfs_open(struct vop_open_args *v) 282 { 283 struct vnode *vp; 284 struct tmpfs_node *node; 285 struct file *fp; 286 int error, mode; 287 288 vp = v->a_vp; 289 mode = v->a_mode; 290 node = VP_TO_TMPFS_NODE(vp); 291 292 /* 293 * The file is still active but all its names have been removed 294 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as 295 * it is about to die. 296 */ 297 if (node->tn_links < 1) 298 return (ENOENT); 299 300 /* If the file is marked append-only, deny write requests. */ 301 if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE) 302 error = EPERM; 303 else { 304 error = 0; 305 /* For regular files, the call below is nop. */ 306 KASSERT(vp->v_type != VREG || (node->tn_reg.tn_aobj->flags & 307 OBJ_DEAD) == 0, ("dead object")); 308 vnode_create_vobject(vp, node->tn_size, v->a_td); 309 } 310 311 fp = v->a_fp; 312 MPASS(fp == NULL || fp->f_data == NULL); 313 if (error == 0 && fp != NULL && vp->v_type == VREG) { 314 tmpfs_ref_node(node); 315 finit_vnode(fp, mode, node, &tmpfs_fnops); 316 } 317 318 return (error); 319 } 320 321 static int 322 tmpfs_close(struct vop_close_args *v) 323 { 324 struct vnode *vp = v->a_vp; 325 326 /* Update node times. */ 327 tmpfs_update(vp); 328 329 return (0); 330 } 331 332 int 333 tmpfs_fo_close(struct file *fp, struct thread *td) 334 { 335 struct tmpfs_node *node; 336 337 node = fp->f_data; 338 if (node != NULL) { 339 MPASS(node->tn_type == VREG); 340 tmpfs_free_node(node->tn_reg.tn_tmp, node); 341 } 342 return (vnops.fo_close(fp, td)); 343 } 344 345 /* 346 * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see 347 * the comment above cache_fplookup for details. 348 */ 349 int 350 tmpfs_fplookup_vexec(struct vop_fplookup_vexec_args *v) 351 { 352 struct vnode *vp; 353 struct tmpfs_node *node; 354 struct ucred *cred; 355 mode_t all_x, mode; 356 357 vp = v->a_vp; 358 node = VP_TO_TMPFS_NODE_SMR(vp); 359 if (__predict_false(node == NULL)) 360 return (EAGAIN); 361 362 all_x = S_IXUSR | S_IXGRP | S_IXOTH; 363 mode = atomic_load_short(&node->tn_mode); 364 if (__predict_true((mode & all_x) == all_x)) 365 return (0); 366 367 cred = v->a_cred; 368 return (vaccess_vexec_smr(mode, node->tn_uid, node->tn_gid, cred)); 369 } 370 371 int 372 tmpfs_access(struct vop_access_args *v) 373 { 374 struct vnode *vp = v->a_vp; 375 accmode_t accmode = v->a_accmode; 376 struct ucred *cred = v->a_cred; 377 mode_t all_x = S_IXUSR | S_IXGRP | S_IXOTH; 378 int error; 379 struct tmpfs_node *node; 380 381 MPASS(VOP_ISLOCKED(vp)); 382 383 node = VP_TO_TMPFS_NODE(vp); 384 385 /* 386 * Common case path lookup. 387 */ 388 if (__predict_true(accmode == VEXEC && (node->tn_mode & all_x) == all_x)) 389 return (0); 390 391 switch (vp->v_type) { 392 case VDIR: 393 /* FALLTHROUGH */ 394 case VLNK: 395 /* FALLTHROUGH */ 396 case VREG: 397 if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) { 398 error = EROFS; 399 goto out; 400 } 401 break; 402 403 case VBLK: 404 /* FALLTHROUGH */ 405 case VCHR: 406 /* FALLTHROUGH */ 407 case VSOCK: 408 /* FALLTHROUGH */ 409 case VFIFO: 410 break; 411 412 default: 413 error = EINVAL; 414 goto out; 415 } 416 417 if (accmode & VWRITE && node->tn_flags & IMMUTABLE) { 418 error = EPERM; 419 goto out; 420 } 421 422 error = vaccess(vp->v_type, node->tn_mode, node->tn_uid, node->tn_gid, 423 accmode, cred); 424 425 out: 426 MPASS(VOP_ISLOCKED(vp)); 427 428 return (error); 429 } 430 431 int 432 tmpfs_stat(struct vop_stat_args *v) 433 { 434 struct vnode *vp = v->a_vp; 435 struct stat *sb = v->a_sb; 436 vm_object_t obj; 437 struct tmpfs_node *node; 438 int error; 439 440 node = VP_TO_TMPFS_NODE(vp); 441 442 tmpfs_update_getattr(vp); 443 444 error = vop_stat_helper_pre(v); 445 if (__predict_false(error)) 446 return (error); 447 448 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; 449 sb->st_ino = node->tn_id; 450 sb->st_mode = node->tn_mode | VTTOIF(vp->v_type); 451 sb->st_nlink = node->tn_links; 452 sb->st_uid = node->tn_uid; 453 sb->st_gid = node->tn_gid; 454 sb->st_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ? 455 node->tn_rdev : NODEV; 456 sb->st_size = node->tn_size; 457 sb->st_atim.tv_sec = node->tn_atime.tv_sec; 458 sb->st_atim.tv_nsec = node->tn_atime.tv_nsec; 459 sb->st_mtim.tv_sec = node->tn_mtime.tv_sec; 460 sb->st_mtim.tv_nsec = node->tn_mtime.tv_nsec; 461 sb->st_ctim.tv_sec = node->tn_ctime.tv_sec; 462 sb->st_ctim.tv_nsec = node->tn_ctime.tv_nsec; 463 sb->st_birthtim.tv_sec = node->tn_birthtime.tv_sec; 464 sb->st_birthtim.tv_nsec = node->tn_birthtime.tv_nsec; 465 sb->st_blksize = PAGE_SIZE; 466 sb->st_flags = node->tn_flags; 467 sb->st_gen = node->tn_gen; 468 if (vp->v_type == VREG) { 469 obj = node->tn_reg.tn_aobj; 470 sb->st_blocks = (u_quad_t)obj->resident_page_count * PAGE_SIZE; 471 } else 472 sb->st_blocks = node->tn_size; 473 sb->st_blocks /= S_BLKSIZE; 474 return (vop_stat_helper_post(v, error)); 475 } 476 477 int 478 tmpfs_getattr(struct vop_getattr_args *v) 479 { 480 struct vnode *vp = v->a_vp; 481 struct vattr *vap = v->a_vap; 482 vm_object_t obj; 483 struct tmpfs_node *node; 484 485 node = VP_TO_TMPFS_NODE(vp); 486 487 tmpfs_update_getattr(vp); 488 489 vap->va_type = vp->v_type; 490 vap->va_mode = node->tn_mode; 491 vap->va_nlink = node->tn_links; 492 vap->va_uid = node->tn_uid; 493 vap->va_gid = node->tn_gid; 494 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 495 vap->va_fileid = node->tn_id; 496 vap->va_size = node->tn_size; 497 vap->va_blocksize = PAGE_SIZE; 498 vap->va_atime = node->tn_atime; 499 vap->va_mtime = node->tn_mtime; 500 vap->va_ctime = node->tn_ctime; 501 vap->va_birthtime = node->tn_birthtime; 502 vap->va_gen = node->tn_gen; 503 vap->va_flags = node->tn_flags; 504 vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ? 505 node->tn_rdev : NODEV; 506 if (vp->v_type == VREG) { 507 obj = node->tn_reg.tn_aobj; 508 vap->va_bytes = (u_quad_t)obj->resident_page_count * PAGE_SIZE; 509 } else 510 vap->va_bytes = node->tn_size; 511 vap->va_filerev = 0; 512 513 return (0); 514 } 515 516 int 517 tmpfs_setattr(struct vop_setattr_args *v) 518 { 519 struct vnode *vp = v->a_vp; 520 struct vattr *vap = v->a_vap; 521 struct ucred *cred = v->a_cred; 522 struct thread *td = curthread; 523 524 int error; 525 526 MPASS(VOP_ISLOCKED(vp)); 527 ASSERT_VOP_IN_SEQC(vp); 528 529 error = 0; 530 531 /* Abort if any unsettable attribute is given. */ 532 if (vap->va_type != VNON || 533 vap->va_nlink != VNOVAL || 534 vap->va_fsid != VNOVAL || 535 vap->va_fileid != VNOVAL || 536 vap->va_blocksize != VNOVAL || 537 vap->va_gen != VNOVAL || 538 vap->va_rdev != VNOVAL || 539 vap->va_bytes != VNOVAL) 540 error = EINVAL; 541 542 if (error == 0 && (vap->va_flags != VNOVAL)) 543 error = tmpfs_chflags(vp, vap->va_flags, cred, td); 544 545 if (error == 0 && (vap->va_size != VNOVAL)) 546 error = tmpfs_chsize(vp, vap->va_size, cred, td); 547 548 if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL)) 549 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td); 550 551 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) 552 error = tmpfs_chmod(vp, vap->va_mode, cred, td); 553 554 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL && 555 vap->va_atime.tv_nsec != VNOVAL) || 556 (vap->va_mtime.tv_sec != VNOVAL && 557 vap->va_mtime.tv_nsec != VNOVAL) || 558 (vap->va_birthtime.tv_sec != VNOVAL && 559 vap->va_birthtime.tv_nsec != VNOVAL))) 560 error = tmpfs_chtimes(vp, vap, cred, td); 561 562 /* 563 * Update the node times. We give preference to the error codes 564 * generated by this function rather than the ones that may arise 565 * from tmpfs_update. 566 */ 567 tmpfs_update(vp); 568 569 MPASS(VOP_ISLOCKED(vp)); 570 571 return (error); 572 } 573 574 static int 575 tmpfs_read(struct vop_read_args *v) 576 { 577 struct vnode *vp; 578 struct uio *uio; 579 struct tmpfs_node *node; 580 581 vp = v->a_vp; 582 if (vp->v_type != VREG) 583 return (EISDIR); 584 uio = v->a_uio; 585 if (uio->uio_offset < 0) 586 return (EINVAL); 587 node = VP_TO_TMPFS_NODE(vp); 588 tmpfs_set_accessed(VFS_TO_TMPFS(vp->v_mount), node); 589 return (uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio)); 590 } 591 592 static int 593 tmpfs_read_pgcache(struct vop_read_pgcache_args *v) 594 { 595 struct vnode *vp; 596 struct tmpfs_node *node; 597 vm_object_t object; 598 off_t size; 599 int error; 600 601 vp = v->a_vp; 602 VNPASS((vn_irflag_read(vp) & VIRF_PGREAD) != 0, vp); 603 604 if (v->a_uio->uio_offset < 0) 605 return (EINVAL); 606 607 error = EJUSTRETURN; 608 vfs_smr_enter(); 609 610 node = VP_TO_TMPFS_NODE_SMR(vp); 611 if (node == NULL) 612 goto out_smr; 613 MPASS(node->tn_type == VREG); 614 MPASS(node->tn_refcount >= 1); 615 object = node->tn_reg.tn_aobj; 616 if (object == NULL) 617 goto out_smr; 618 619 MPASS(object->type == tmpfs_pager_type); 620 MPASS((object->flags & (OBJ_ANON | OBJ_DEAD | OBJ_SWAP)) == 621 OBJ_SWAP); 622 if (!VN_IS_DOOMED(vp)) { 623 /* size cannot become shorter due to rangelock. */ 624 size = node->tn_size; 625 tmpfs_set_accessed(node->tn_reg.tn_tmp, node); 626 vfs_smr_exit(); 627 error = uiomove_object(object, size, v->a_uio); 628 return (error); 629 } 630 out_smr: 631 vfs_smr_exit(); 632 return (error); 633 } 634 635 static int 636 tmpfs_write(struct vop_write_args *v) 637 { 638 struct vnode *vp; 639 struct uio *uio; 640 struct tmpfs_node *node; 641 off_t oldsize; 642 int error, ioflag; 643 mode_t newmode; 644 645 vp = v->a_vp; 646 uio = v->a_uio; 647 ioflag = v->a_ioflag; 648 error = 0; 649 node = VP_TO_TMPFS_NODE(vp); 650 oldsize = node->tn_size; 651 652 if (uio->uio_offset < 0 || vp->v_type != VREG) 653 return (EINVAL); 654 if (uio->uio_resid == 0) 655 return (0); 656 if (ioflag & IO_APPEND) 657 uio->uio_offset = node->tn_size; 658 if (uio->uio_offset + uio->uio_resid > 659 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 660 return (EFBIG); 661 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) 662 return (EFBIG); 663 if (uio->uio_offset + uio->uio_resid > node->tn_size) { 664 error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid, 665 FALSE); 666 if (error != 0) 667 goto out; 668 } 669 670 error = uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio); 671 node->tn_status |= TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED; 672 node->tn_accessed = true; 673 if (node->tn_mode & (S_ISUID | S_ISGID)) { 674 if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID)) { 675 newmode = node->tn_mode & ~(S_ISUID | S_ISGID); 676 vn_seqc_write_begin(vp); 677 atomic_store_short(&node->tn_mode, newmode); 678 vn_seqc_write_end(vp); 679 } 680 } 681 if (error != 0) 682 (void)tmpfs_reg_resize(vp, oldsize, TRUE); 683 684 out: 685 MPASS(IMPLIES(error == 0, uio->uio_resid == 0)); 686 MPASS(IMPLIES(error != 0, oldsize == node->tn_size)); 687 688 return (error); 689 } 690 691 static int 692 tmpfs_deallocate(struct vop_deallocate_args *v) 693 { 694 return (tmpfs_reg_punch_hole(v->a_vp, v->a_offset, v->a_len)); 695 } 696 697 static int 698 tmpfs_fsync(struct vop_fsync_args *v) 699 { 700 struct vnode *vp = v->a_vp; 701 702 MPASS(VOP_ISLOCKED(vp)); 703 704 tmpfs_check_mtime(vp); 705 tmpfs_update(vp); 706 707 return (0); 708 } 709 710 static int 711 tmpfs_remove(struct vop_remove_args *v) 712 { 713 struct vnode *dvp = v->a_dvp; 714 struct vnode *vp = v->a_vp; 715 716 int error; 717 struct tmpfs_dirent *de; 718 struct tmpfs_mount *tmp; 719 struct tmpfs_node *dnode; 720 struct tmpfs_node *node; 721 722 MPASS(VOP_ISLOCKED(dvp)); 723 MPASS(VOP_ISLOCKED(vp)); 724 725 if (vp->v_type == VDIR) { 726 error = EISDIR; 727 goto out; 728 } 729 730 dnode = VP_TO_TMPFS_DIR(dvp); 731 node = VP_TO_TMPFS_NODE(vp); 732 tmp = VFS_TO_TMPFS(vp->v_mount); 733 de = tmpfs_dir_lookup(dnode, node, v->a_cnp); 734 MPASS(de != NULL); 735 736 /* Files marked as immutable or append-only cannot be deleted. */ 737 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) || 738 (dnode->tn_flags & APPEND)) { 739 error = EPERM; 740 goto out; 741 } 742 743 /* Remove the entry from the directory; as it is a file, we do not 744 * have to change the number of hard links of the directory. */ 745 tmpfs_dir_detach(dvp, de); 746 if (v->a_cnp->cn_flags & DOWHITEOUT) 747 tmpfs_dir_whiteout_add(dvp, v->a_cnp); 748 749 /* Free the directory entry we just deleted. Note that the node 750 * referred by it will not be removed until the vnode is really 751 * reclaimed. */ 752 tmpfs_free_dirent(tmp, de); 753 754 node->tn_status |= TMPFS_NODE_CHANGED; 755 node->tn_accessed = true; 756 error = 0; 757 758 out: 759 return (error); 760 } 761 762 static int 763 tmpfs_link(struct vop_link_args *v) 764 { 765 struct vnode *dvp = v->a_tdvp; 766 struct vnode *vp = v->a_vp; 767 struct componentname *cnp = v->a_cnp; 768 769 int error; 770 struct tmpfs_dirent *de; 771 struct tmpfs_node *node; 772 773 MPASS(VOP_ISLOCKED(dvp)); 774 MPASS(dvp != vp); /* XXX When can this be false? */ 775 node = VP_TO_TMPFS_NODE(vp); 776 777 /* Ensure that we do not overflow the maximum number of links imposed 778 * by the system. */ 779 MPASS(node->tn_links <= TMPFS_LINK_MAX); 780 if (node->tn_links == TMPFS_LINK_MAX) { 781 error = EMLINK; 782 goto out; 783 } 784 785 /* We cannot create links of files marked immutable or append-only. */ 786 if (node->tn_flags & (IMMUTABLE | APPEND)) { 787 error = EPERM; 788 goto out; 789 } 790 791 /* Allocate a new directory entry to represent the node. */ 792 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node, 793 cnp->cn_nameptr, cnp->cn_namelen, &de); 794 if (error != 0) 795 goto out; 796 797 /* Insert the new directory entry into the appropriate directory. */ 798 if (cnp->cn_flags & ISWHITEOUT) 799 tmpfs_dir_whiteout_remove(dvp, cnp); 800 tmpfs_dir_attach(dvp, de); 801 802 /* vp link count has changed, so update node times. */ 803 node->tn_status |= TMPFS_NODE_CHANGED; 804 tmpfs_update(vp); 805 806 error = 0; 807 808 out: 809 return (error); 810 } 811 812 /* 813 * We acquire all but fdvp locks using non-blocking acquisitions. If we 814 * fail to acquire any lock in the path we will drop all held locks, 815 * acquire the new lock in a blocking fashion, and then release it and 816 * restart the rename. This acquire/release step ensures that we do not 817 * spin on a lock waiting for release. On error release all vnode locks 818 * and decrement references the way tmpfs_rename() would do. 819 */ 820 static int 821 tmpfs_rename_relock(struct vnode *fdvp, struct vnode **fvpp, 822 struct vnode *tdvp, struct vnode **tvpp, 823 struct componentname *fcnp, struct componentname *tcnp) 824 { 825 struct vnode *nvp; 826 struct mount *mp; 827 struct tmpfs_dirent *de; 828 int error, restarts = 0; 829 830 VOP_UNLOCK(tdvp); 831 if (*tvpp != NULL && *tvpp != tdvp) 832 VOP_UNLOCK(*tvpp); 833 mp = fdvp->v_mount; 834 835 relock: 836 restarts += 1; 837 error = vn_lock(fdvp, LK_EXCLUSIVE); 838 if (error) 839 goto releout; 840 if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 841 VOP_UNLOCK(fdvp); 842 error = vn_lock(tdvp, LK_EXCLUSIVE); 843 if (error) 844 goto releout; 845 VOP_UNLOCK(tdvp); 846 goto relock; 847 } 848 /* 849 * Re-resolve fvp to be certain it still exists and fetch the 850 * correct vnode. 851 */ 852 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(fdvp), NULL, fcnp); 853 if (de == NULL) { 854 VOP_UNLOCK(fdvp); 855 VOP_UNLOCK(tdvp); 856 if ((fcnp->cn_flags & ISDOTDOT) != 0 || 857 (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.')) 858 error = EINVAL; 859 else 860 error = ENOENT; 861 goto releout; 862 } 863 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE | LK_NOWAIT, &nvp); 864 if (error != 0) { 865 VOP_UNLOCK(fdvp); 866 VOP_UNLOCK(tdvp); 867 if (error != EBUSY) 868 goto releout; 869 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, &nvp); 870 if (error != 0) 871 goto releout; 872 VOP_UNLOCK(nvp); 873 /* 874 * Concurrent rename race. 875 */ 876 if (nvp == tdvp) { 877 vrele(nvp); 878 error = EINVAL; 879 goto releout; 880 } 881 vrele(*fvpp); 882 *fvpp = nvp; 883 goto relock; 884 } 885 vrele(*fvpp); 886 *fvpp = nvp; 887 VOP_UNLOCK(*fvpp); 888 /* 889 * Re-resolve tvp and acquire the vnode lock if present. 890 */ 891 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(tdvp), NULL, tcnp); 892 /* 893 * If tvp disappeared we just carry on. 894 */ 895 if (de == NULL && *tvpp != NULL) { 896 vrele(*tvpp); 897 *tvpp = NULL; 898 } 899 /* 900 * Get the tvp ino if the lookup succeeded. We may have to restart 901 * if the non-blocking acquire fails. 902 */ 903 if (de != NULL) { 904 nvp = NULL; 905 error = tmpfs_alloc_vp(mp, de->td_node, 906 LK_EXCLUSIVE | LK_NOWAIT, &nvp); 907 if (*tvpp != NULL) 908 vrele(*tvpp); 909 *tvpp = nvp; 910 if (error != 0) { 911 VOP_UNLOCK(fdvp); 912 VOP_UNLOCK(tdvp); 913 if (error != EBUSY) 914 goto releout; 915 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, 916 &nvp); 917 if (error != 0) 918 goto releout; 919 VOP_UNLOCK(nvp); 920 /* 921 * fdvp contains fvp, thus tvp (=fdvp) is not empty. 922 */ 923 if (nvp == fdvp) { 924 error = ENOTEMPTY; 925 goto releout; 926 } 927 goto relock; 928 } 929 } 930 tmpfs_rename_restarts += restarts; 931 932 return (0); 933 934 releout: 935 vrele(fdvp); 936 vrele(*fvpp); 937 vrele(tdvp); 938 if (*tvpp != NULL) 939 vrele(*tvpp); 940 tmpfs_rename_restarts += restarts; 941 942 return (error); 943 } 944 945 static int 946 tmpfs_rename(struct vop_rename_args *v) 947 { 948 struct vnode *fdvp = v->a_fdvp; 949 struct vnode *fvp = v->a_fvp; 950 struct componentname *fcnp = v->a_fcnp; 951 struct vnode *tdvp = v->a_tdvp; 952 struct vnode *tvp = v->a_tvp; 953 struct componentname *tcnp = v->a_tcnp; 954 char *newname; 955 struct tmpfs_dirent *de; 956 struct tmpfs_mount *tmp; 957 struct tmpfs_node *fdnode; 958 struct tmpfs_node *fnode; 959 struct tmpfs_node *tnode; 960 struct tmpfs_node *tdnode; 961 int error; 962 bool want_seqc_end; 963 964 MPASS(VOP_ISLOCKED(tdvp)); 965 MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp))); 966 967 want_seqc_end = false; 968 969 /* 970 * Disallow cross-device renames. 971 * XXX Why isn't this done by the caller? 972 */ 973 if (fvp->v_mount != tdvp->v_mount || 974 (tvp != NULL && fvp->v_mount != tvp->v_mount)) { 975 error = EXDEV; 976 goto out; 977 } 978 979 /* If source and target are the same file, there is nothing to do. */ 980 if (fvp == tvp) { 981 error = 0; 982 goto out; 983 } 984 985 /* 986 * If we need to move the directory between entries, lock the 987 * source so that we can safely operate on it. 988 */ 989 if (fdvp != tdvp && fdvp != tvp) { 990 if (vn_lock(fdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 991 error = tmpfs_rename_relock(fdvp, &fvp, tdvp, &tvp, 992 fcnp, tcnp); 993 if (error != 0) 994 return (error); 995 ASSERT_VOP_ELOCKED(fdvp, 996 "tmpfs_rename: fdvp not locked"); 997 ASSERT_VOP_ELOCKED(tdvp, 998 "tmpfs_rename: tdvp not locked"); 999 if (tvp != NULL) 1000 ASSERT_VOP_ELOCKED(tvp, 1001 "tmpfs_rename: tvp not locked"); 1002 if (fvp == tvp) { 1003 error = 0; 1004 goto out_locked; 1005 } 1006 } 1007 } 1008 1009 if (tvp != NULL) 1010 vn_seqc_write_begin(tvp); 1011 vn_seqc_write_begin(tdvp); 1012 vn_seqc_write_begin(fvp); 1013 vn_seqc_write_begin(fdvp); 1014 want_seqc_end = true; 1015 1016 tmp = VFS_TO_TMPFS(tdvp->v_mount); 1017 tdnode = VP_TO_TMPFS_DIR(tdvp); 1018 tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp); 1019 fdnode = VP_TO_TMPFS_DIR(fdvp); 1020 fnode = VP_TO_TMPFS_NODE(fvp); 1021 de = tmpfs_dir_lookup(fdnode, fnode, fcnp); 1022 1023 /* 1024 * Entry can disappear before we lock fdvp, 1025 * also avoid manipulating '.' and '..' entries. 1026 */ 1027 if (de == NULL) { 1028 if ((fcnp->cn_flags & ISDOTDOT) != 0 || 1029 (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.')) 1030 error = EINVAL; 1031 else 1032 error = ENOENT; 1033 goto out_locked; 1034 } 1035 MPASS(de->td_node == fnode); 1036 1037 /* 1038 * If re-naming a directory to another preexisting directory 1039 * ensure that the target directory is empty so that its 1040 * removal causes no side effects. 1041 * Kern_rename guarantees the destination to be a directory 1042 * if the source is one. 1043 */ 1044 if (tvp != NULL) { 1045 MPASS(tnode != NULL); 1046 1047 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 1048 (tdnode->tn_flags & (APPEND | IMMUTABLE))) { 1049 error = EPERM; 1050 goto out_locked; 1051 } 1052 1053 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) { 1054 if (tnode->tn_size > 0) { 1055 error = ENOTEMPTY; 1056 goto out_locked; 1057 } 1058 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) { 1059 error = ENOTDIR; 1060 goto out_locked; 1061 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) { 1062 error = EISDIR; 1063 goto out_locked; 1064 } else { 1065 MPASS(fnode->tn_type != VDIR && 1066 tnode->tn_type != VDIR); 1067 } 1068 } 1069 1070 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) 1071 || (fdnode->tn_flags & (APPEND | IMMUTABLE))) { 1072 error = EPERM; 1073 goto out_locked; 1074 } 1075 1076 /* 1077 * Ensure that we have enough memory to hold the new name, if it 1078 * has to be changed. 1079 */ 1080 if (fcnp->cn_namelen != tcnp->cn_namelen || 1081 bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) { 1082 newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK); 1083 } else 1084 newname = NULL; 1085 1086 /* 1087 * If the node is being moved to another directory, we have to do 1088 * the move. 1089 */ 1090 if (fdnode != tdnode) { 1091 /* 1092 * In case we are moving a directory, we have to adjust its 1093 * parent to point to the new parent. 1094 */ 1095 if (de->td_node->tn_type == VDIR) { 1096 struct tmpfs_node *n; 1097 1098 /* 1099 * Ensure the target directory is not a child of the 1100 * directory being moved. Otherwise, we'd end up 1101 * with stale nodes. 1102 */ 1103 n = tdnode; 1104 /* 1105 * TMPFS_LOCK guaranties that no nodes are freed while 1106 * traversing the list. Nodes can only be marked as 1107 * removed: tn_parent == NULL. 1108 */ 1109 TMPFS_LOCK(tmp); 1110 TMPFS_NODE_LOCK(n); 1111 while (n != n->tn_dir.tn_parent) { 1112 struct tmpfs_node *parent; 1113 1114 if (n == fnode) { 1115 TMPFS_NODE_UNLOCK(n); 1116 TMPFS_UNLOCK(tmp); 1117 error = EINVAL; 1118 if (newname != NULL) 1119 free(newname, M_TMPFSNAME); 1120 goto out_locked; 1121 } 1122 parent = n->tn_dir.tn_parent; 1123 TMPFS_NODE_UNLOCK(n); 1124 if (parent == NULL) { 1125 n = NULL; 1126 break; 1127 } 1128 TMPFS_NODE_LOCK(parent); 1129 if (parent->tn_dir.tn_parent == NULL) { 1130 TMPFS_NODE_UNLOCK(parent); 1131 n = NULL; 1132 break; 1133 } 1134 n = parent; 1135 } 1136 TMPFS_UNLOCK(tmp); 1137 if (n == NULL) { 1138 error = EINVAL; 1139 if (newname != NULL) 1140 free(newname, M_TMPFSNAME); 1141 goto out_locked; 1142 } 1143 TMPFS_NODE_UNLOCK(n); 1144 1145 /* Adjust the parent pointer. */ 1146 TMPFS_VALIDATE_DIR(fnode); 1147 TMPFS_NODE_LOCK(de->td_node); 1148 de->td_node->tn_dir.tn_parent = tdnode; 1149 TMPFS_NODE_UNLOCK(de->td_node); 1150 1151 /* 1152 * As a result of changing the target of the '..' 1153 * entry, the link count of the source and target 1154 * directories has to be adjusted. 1155 */ 1156 TMPFS_NODE_LOCK(tdnode); 1157 TMPFS_ASSERT_LOCKED(tdnode); 1158 tdnode->tn_links++; 1159 TMPFS_NODE_UNLOCK(tdnode); 1160 1161 TMPFS_NODE_LOCK(fdnode); 1162 TMPFS_ASSERT_LOCKED(fdnode); 1163 fdnode->tn_links--; 1164 TMPFS_NODE_UNLOCK(fdnode); 1165 } 1166 } 1167 1168 /* 1169 * Do the move: just remove the entry from the source directory 1170 * and insert it into the target one. 1171 */ 1172 tmpfs_dir_detach(fdvp, de); 1173 1174 if (fcnp->cn_flags & DOWHITEOUT) 1175 tmpfs_dir_whiteout_add(fdvp, fcnp); 1176 if (tcnp->cn_flags & ISWHITEOUT) 1177 tmpfs_dir_whiteout_remove(tdvp, tcnp); 1178 1179 /* 1180 * If the name has changed, we need to make it effective by changing 1181 * it in the directory entry. 1182 */ 1183 if (newname != NULL) { 1184 MPASS(tcnp->cn_namelen <= MAXNAMLEN); 1185 1186 free(de->ud.td_name, M_TMPFSNAME); 1187 de->ud.td_name = newname; 1188 tmpfs_dirent_init(de, tcnp->cn_nameptr, tcnp->cn_namelen); 1189 1190 fnode->tn_status |= TMPFS_NODE_CHANGED; 1191 tdnode->tn_status |= TMPFS_NODE_MODIFIED; 1192 } 1193 1194 /* 1195 * If we are overwriting an entry, we have to remove the old one 1196 * from the target directory. 1197 */ 1198 if (tvp != NULL) { 1199 struct tmpfs_dirent *tde; 1200 1201 /* Remove the old entry from the target directory. */ 1202 tde = tmpfs_dir_lookup(tdnode, tnode, tcnp); 1203 tmpfs_dir_detach(tdvp, tde); 1204 1205 /* 1206 * Free the directory entry we just deleted. Note that the 1207 * node referred by it will not be removed until the vnode is 1208 * really reclaimed. 1209 */ 1210 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde); 1211 } 1212 1213 tmpfs_dir_attach(tdvp, de); 1214 1215 if (tmpfs_use_nc(fvp)) { 1216 cache_vop_rename(fdvp, fvp, tdvp, tvp, fcnp, tcnp); 1217 } 1218 1219 error = 0; 1220 1221 out_locked: 1222 if (fdvp != tdvp && fdvp != tvp) 1223 VOP_UNLOCK(fdvp); 1224 1225 out: 1226 if (want_seqc_end) { 1227 if (tvp != NULL) 1228 vn_seqc_write_end(tvp); 1229 vn_seqc_write_end(tdvp); 1230 vn_seqc_write_end(fvp); 1231 vn_seqc_write_end(fdvp); 1232 } 1233 1234 /* 1235 * Release target nodes. 1236 * XXX: I don't understand when tdvp can be the same as tvp, but 1237 * other code takes care of this... 1238 */ 1239 if (tdvp == tvp) 1240 vrele(tdvp); 1241 else 1242 vput(tdvp); 1243 if (tvp != NULL) 1244 vput(tvp); 1245 1246 /* Release source nodes. */ 1247 vrele(fdvp); 1248 vrele(fvp); 1249 1250 return (error); 1251 } 1252 1253 static int 1254 tmpfs_mkdir(struct vop_mkdir_args *v) 1255 { 1256 struct vnode *dvp = v->a_dvp; 1257 struct vnode **vpp = v->a_vpp; 1258 struct componentname *cnp = v->a_cnp; 1259 struct vattr *vap = v->a_vap; 1260 1261 MPASS(vap->va_type == VDIR); 1262 1263 return (tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL)); 1264 } 1265 1266 static int 1267 tmpfs_rmdir(struct vop_rmdir_args *v) 1268 { 1269 struct vnode *dvp = v->a_dvp; 1270 struct vnode *vp = v->a_vp; 1271 1272 int error; 1273 struct tmpfs_dirent *de; 1274 struct tmpfs_mount *tmp; 1275 struct tmpfs_node *dnode; 1276 struct tmpfs_node *node; 1277 1278 MPASS(VOP_ISLOCKED(dvp)); 1279 MPASS(VOP_ISLOCKED(vp)); 1280 1281 tmp = VFS_TO_TMPFS(dvp->v_mount); 1282 dnode = VP_TO_TMPFS_DIR(dvp); 1283 node = VP_TO_TMPFS_DIR(vp); 1284 1285 /* Directories with more than two entries ('.' and '..') cannot be 1286 * removed. */ 1287 if (node->tn_size > 0) { 1288 error = ENOTEMPTY; 1289 goto out; 1290 } 1291 1292 if ((dnode->tn_flags & APPEND) 1293 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1294 error = EPERM; 1295 goto out; 1296 } 1297 1298 /* This invariant holds only if we are not trying to remove "..". 1299 * We checked for that above so this is safe now. */ 1300 MPASS(node->tn_dir.tn_parent == dnode); 1301 1302 /* Get the directory entry associated with node (vp). This was 1303 * filled by tmpfs_lookup while looking up the entry. */ 1304 de = tmpfs_dir_lookup(dnode, node, v->a_cnp); 1305 MPASS(TMPFS_DIRENT_MATCHES(de, 1306 v->a_cnp->cn_nameptr, 1307 v->a_cnp->cn_namelen)); 1308 1309 /* Check flags to see if we are allowed to remove the directory. */ 1310 if ((dnode->tn_flags & APPEND) != 0 || 1311 (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) != 0) { 1312 error = EPERM; 1313 goto out; 1314 } 1315 1316 /* Detach the directory entry from the directory (dnode). */ 1317 tmpfs_dir_detach(dvp, de); 1318 if (v->a_cnp->cn_flags & DOWHITEOUT) 1319 tmpfs_dir_whiteout_add(dvp, v->a_cnp); 1320 1321 /* No vnode should be allocated for this entry from this point */ 1322 TMPFS_NODE_LOCK(node); 1323 node->tn_links--; 1324 node->tn_dir.tn_parent = NULL; 1325 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1326 node->tn_accessed = true; 1327 1328 TMPFS_NODE_UNLOCK(node); 1329 1330 TMPFS_NODE_LOCK(dnode); 1331 dnode->tn_links--; 1332 dnode->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1333 dnode->tn_accessed = true; 1334 TMPFS_NODE_UNLOCK(dnode); 1335 1336 if (tmpfs_use_nc(dvp)) { 1337 cache_vop_rmdir(dvp, vp); 1338 } 1339 1340 /* Free the directory entry we just deleted. Note that the node 1341 * referred by it will not be removed until the vnode is really 1342 * reclaimed. */ 1343 tmpfs_free_dirent(tmp, de); 1344 1345 /* Release the deleted vnode (will destroy the node, notify 1346 * interested parties and clean it from the cache). */ 1347 1348 dnode->tn_status |= TMPFS_NODE_CHANGED; 1349 tmpfs_update(dvp); 1350 1351 error = 0; 1352 1353 out: 1354 return (error); 1355 } 1356 1357 static int 1358 tmpfs_symlink(struct vop_symlink_args *v) 1359 { 1360 struct vnode *dvp = v->a_dvp; 1361 struct vnode **vpp = v->a_vpp; 1362 struct componentname *cnp = v->a_cnp; 1363 struct vattr *vap = v->a_vap; 1364 const char *target = v->a_target; 1365 1366 #ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */ 1367 MPASS(vap->va_type == VLNK); 1368 #else 1369 vap->va_type = VLNK; 1370 #endif 1371 1372 return (tmpfs_alloc_file(dvp, vpp, vap, cnp, target)); 1373 } 1374 1375 static int 1376 tmpfs_readdir(struct vop_readdir_args *va) 1377 { 1378 struct vnode *vp; 1379 struct uio *uio; 1380 struct tmpfs_mount *tm; 1381 struct tmpfs_node *node; 1382 uint64_t **cookies; 1383 int *eofflag, *ncookies; 1384 ssize_t startresid; 1385 int error, maxcookies; 1386 1387 vp = va->a_vp; 1388 uio = va->a_uio; 1389 eofflag = va->a_eofflag; 1390 cookies = va->a_cookies; 1391 ncookies = va->a_ncookies; 1392 1393 /* This operation only makes sense on directory nodes. */ 1394 if (vp->v_type != VDIR) 1395 return (ENOTDIR); 1396 1397 maxcookies = 0; 1398 node = VP_TO_TMPFS_DIR(vp); 1399 tm = VFS_TO_TMPFS(vp->v_mount); 1400 1401 startresid = uio->uio_resid; 1402 1403 /* Allocate cookies for NFS and compat modules. */ 1404 if (cookies != NULL && ncookies != NULL) { 1405 maxcookies = howmany(node->tn_size, 1406 sizeof(struct tmpfs_dirent)) + 2; 1407 *cookies = malloc(maxcookies * sizeof(**cookies), M_TEMP, 1408 M_WAITOK); 1409 *ncookies = 0; 1410 } 1411 1412 if (cookies == NULL) 1413 error = tmpfs_dir_getdents(tm, node, uio, 0, NULL, NULL); 1414 else 1415 error = tmpfs_dir_getdents(tm, node, uio, maxcookies, *cookies, 1416 ncookies); 1417 1418 /* Buffer was filled without hitting EOF. */ 1419 if (error == EJUSTRETURN) 1420 error = (uio->uio_resid != startresid) ? 0 : EINVAL; 1421 1422 if (error != 0 && cookies != NULL && ncookies != NULL) { 1423 free(*cookies, M_TEMP); 1424 *cookies = NULL; 1425 *ncookies = 0; 1426 } 1427 1428 if (eofflag != NULL) 1429 *eofflag = 1430 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF); 1431 1432 return (error); 1433 } 1434 1435 static int 1436 tmpfs_readlink(struct vop_readlink_args *v) 1437 { 1438 struct vnode *vp = v->a_vp; 1439 struct uio *uio = v->a_uio; 1440 1441 int error; 1442 struct tmpfs_node *node; 1443 1444 MPASS(uio->uio_offset == 0); 1445 MPASS(vp->v_type == VLNK); 1446 1447 node = VP_TO_TMPFS_NODE(vp); 1448 1449 error = uiomove(node->tn_link_target, MIN(node->tn_size, uio->uio_resid), 1450 uio); 1451 tmpfs_set_accessed(VFS_TO_TMPFS(vp->v_mount), node); 1452 1453 return (error); 1454 } 1455 1456 /* 1457 * VOP_FPLOOKUP_SYMLINK routines are subject to special circumstances, see 1458 * the comment above cache_fplookup for details. 1459 * 1460 * Check tmpfs_alloc_node for tmpfs-specific synchronisation notes. 1461 */ 1462 static int 1463 tmpfs_fplookup_symlink(struct vop_fplookup_symlink_args *v) 1464 { 1465 struct vnode *vp; 1466 struct tmpfs_node *node; 1467 char *symlink; 1468 1469 vp = v->a_vp; 1470 node = VP_TO_TMPFS_NODE_SMR(vp); 1471 if (__predict_false(node == NULL)) 1472 return (EAGAIN); 1473 if (!atomic_load_char(&node->tn_link_smr)) 1474 return (EAGAIN); 1475 symlink = atomic_load_ptr(&node->tn_link_target); 1476 if (symlink == NULL) 1477 return (EAGAIN); 1478 1479 return (cache_symlink_resolve(v->a_fpl, symlink, node->tn_size)); 1480 } 1481 1482 static int 1483 tmpfs_inactive(struct vop_inactive_args *v) 1484 { 1485 struct vnode *vp; 1486 struct tmpfs_node *node; 1487 1488 vp = v->a_vp; 1489 node = VP_TO_TMPFS_NODE(vp); 1490 if (node->tn_links == 0) 1491 vrecycle(vp); 1492 else 1493 tmpfs_check_mtime(vp); 1494 return (0); 1495 } 1496 1497 static int 1498 tmpfs_need_inactive(struct vop_need_inactive_args *ap) 1499 { 1500 struct vnode *vp; 1501 struct tmpfs_node *node; 1502 struct vm_object *obj; 1503 1504 vp = ap->a_vp; 1505 node = VP_TO_TMPFS_NODE(vp); 1506 if (node->tn_links == 0) 1507 goto need; 1508 if (vp->v_type == VREG) { 1509 obj = vp->v_object; 1510 if (obj->generation != obj->cleangeneration) 1511 goto need; 1512 } 1513 return (0); 1514 need: 1515 return (1); 1516 } 1517 1518 int 1519 tmpfs_reclaim(struct vop_reclaim_args *v) 1520 { 1521 struct vnode *vp; 1522 struct tmpfs_mount *tmp; 1523 struct tmpfs_node *node; 1524 bool unlock; 1525 1526 vp = v->a_vp; 1527 node = VP_TO_TMPFS_NODE(vp); 1528 tmp = VFS_TO_TMPFS(vp->v_mount); 1529 1530 if (vp->v_type == VREG) 1531 tmpfs_destroy_vobject(vp, node->tn_reg.tn_aobj); 1532 vp->v_object = NULL; 1533 1534 TMPFS_LOCK(tmp); 1535 TMPFS_NODE_LOCK(node); 1536 tmpfs_free_vp(vp); 1537 1538 /* 1539 * If the node referenced by this vnode was deleted by the user, 1540 * we must free its associated data structures (now that the vnode 1541 * is being reclaimed). 1542 */ 1543 unlock = true; 1544 if (node->tn_links == 0 && 1545 (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) { 1546 node->tn_vpstate = TMPFS_VNODE_DOOMED; 1547 unlock = !tmpfs_free_node_locked(tmp, node, true); 1548 } 1549 1550 if (unlock) { 1551 TMPFS_NODE_UNLOCK(node); 1552 TMPFS_UNLOCK(tmp); 1553 } 1554 1555 MPASS(vp->v_data == NULL); 1556 return (0); 1557 } 1558 1559 int 1560 tmpfs_print(struct vop_print_args *v) 1561 { 1562 struct vnode *vp = v->a_vp; 1563 1564 struct tmpfs_node *node; 1565 1566 node = VP_TO_TMPFS_NODE(vp); 1567 1568 printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%lx, links %jd\n", 1569 node, node->tn_flags, (uintmax_t)node->tn_links); 1570 printf("\tmode 0%o, owner %d, group %d, size %jd, status 0x%x\n", 1571 node->tn_mode, node->tn_uid, node->tn_gid, 1572 (intmax_t)node->tn_size, node->tn_status); 1573 1574 if (vp->v_type == VFIFO) 1575 fifo_printinfo(vp); 1576 1577 printf("\n"); 1578 1579 return (0); 1580 } 1581 1582 int 1583 tmpfs_pathconf(struct vop_pathconf_args *v) 1584 { 1585 struct vnode *vp = v->a_vp; 1586 int name = v->a_name; 1587 long *retval = v->a_retval; 1588 1589 int error; 1590 1591 error = 0; 1592 1593 switch (name) { 1594 case _PC_LINK_MAX: 1595 *retval = TMPFS_LINK_MAX; 1596 break; 1597 1598 case _PC_SYMLINK_MAX: 1599 *retval = MAXPATHLEN; 1600 break; 1601 1602 case _PC_NAME_MAX: 1603 *retval = NAME_MAX; 1604 break; 1605 1606 case _PC_PIPE_BUF: 1607 if (vp->v_type == VDIR || vp->v_type == VFIFO) 1608 *retval = PIPE_BUF; 1609 else 1610 error = EINVAL; 1611 break; 1612 1613 case _PC_CHOWN_RESTRICTED: 1614 *retval = 1; 1615 break; 1616 1617 case _PC_NO_TRUNC: 1618 *retval = 1; 1619 break; 1620 1621 case _PC_SYNC_IO: 1622 *retval = 1; 1623 break; 1624 1625 case _PC_FILESIZEBITS: 1626 *retval = 64; 1627 break; 1628 1629 default: 1630 error = vop_stdpathconf(v); 1631 } 1632 1633 return (error); 1634 } 1635 1636 static int 1637 tmpfs_vptofh(struct vop_vptofh_args *ap) 1638 /* 1639 vop_vptofh { 1640 IN struct vnode *a_vp; 1641 IN struct fid *a_fhp; 1642 }; 1643 */ 1644 { 1645 struct tmpfs_fid_data tfd; 1646 struct tmpfs_node *node; 1647 struct fid *fhp; 1648 1649 node = VP_TO_TMPFS_NODE(ap->a_vp); 1650 fhp = ap->a_fhp; 1651 fhp->fid_len = sizeof(tfd); 1652 1653 /* 1654 * Copy into fid_data from the stack to avoid unaligned pointer use. 1655 * See the comment in sys/mount.h on struct fid for details. 1656 */ 1657 tfd.tfd_id = node->tn_id; 1658 tfd.tfd_gen = node->tn_gen; 1659 memcpy(fhp->fid_data, &tfd, fhp->fid_len); 1660 1661 return (0); 1662 } 1663 1664 static int 1665 tmpfs_whiteout(struct vop_whiteout_args *ap) 1666 { 1667 struct vnode *dvp = ap->a_dvp; 1668 struct componentname *cnp = ap->a_cnp; 1669 struct tmpfs_dirent *de; 1670 1671 switch (ap->a_flags) { 1672 case LOOKUP: 1673 return (0); 1674 case CREATE: 1675 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1676 if (de != NULL) 1677 return (de->td_node == NULL ? 0 : EEXIST); 1678 return (tmpfs_dir_whiteout_add(dvp, cnp)); 1679 case DELETE: 1680 tmpfs_dir_whiteout_remove(dvp, cnp); 1681 return (0); 1682 default: 1683 panic("tmpfs_whiteout: unknown op"); 1684 } 1685 } 1686 1687 static int 1688 tmpfs_vptocnp_dir(struct tmpfs_node *tn, struct tmpfs_node *tnp, 1689 struct tmpfs_dirent **pde) 1690 { 1691 struct tmpfs_dir_cursor dc; 1692 struct tmpfs_dirent *de; 1693 1694 for (de = tmpfs_dir_first(tnp, &dc); de != NULL; 1695 de = tmpfs_dir_next(tnp, &dc)) { 1696 if (de->td_node == tn) { 1697 *pde = de; 1698 return (0); 1699 } 1700 } 1701 return (ENOENT); 1702 } 1703 1704 static int 1705 tmpfs_vptocnp_fill(struct vnode *vp, struct tmpfs_node *tn, 1706 struct tmpfs_node *tnp, char *buf, size_t *buflen, struct vnode **dvp) 1707 { 1708 struct tmpfs_dirent *de; 1709 int error, i; 1710 1711 error = vn_vget_ino_gen(vp, tmpfs_vn_get_ino_alloc, tnp, LK_SHARED, 1712 dvp); 1713 if (error != 0) 1714 return (error); 1715 error = tmpfs_vptocnp_dir(tn, tnp, &de); 1716 if (error == 0) { 1717 i = *buflen; 1718 i -= de->td_namelen; 1719 if (i < 0) { 1720 error = ENOMEM; 1721 } else { 1722 bcopy(de->ud.td_name, buf + i, de->td_namelen); 1723 *buflen = i; 1724 } 1725 } 1726 if (error == 0) { 1727 if (vp != *dvp) 1728 VOP_UNLOCK(*dvp); 1729 } else { 1730 if (vp != *dvp) 1731 vput(*dvp); 1732 else 1733 vrele(vp); 1734 } 1735 return (error); 1736 } 1737 1738 static int 1739 tmpfs_vptocnp(struct vop_vptocnp_args *ap) 1740 { 1741 struct vnode *vp, **dvp; 1742 struct tmpfs_node *tn, *tnp, *tnp1; 1743 struct tmpfs_dirent *de; 1744 struct tmpfs_mount *tm; 1745 char *buf; 1746 size_t *buflen; 1747 int error; 1748 1749 vp = ap->a_vp; 1750 dvp = ap->a_vpp; 1751 buf = ap->a_buf; 1752 buflen = ap->a_buflen; 1753 1754 tm = VFS_TO_TMPFS(vp->v_mount); 1755 tn = VP_TO_TMPFS_NODE(vp); 1756 if (tn->tn_type == VDIR) { 1757 tnp = tn->tn_dir.tn_parent; 1758 if (tnp == NULL) 1759 return (ENOENT); 1760 tmpfs_ref_node(tnp); 1761 error = tmpfs_vptocnp_fill(vp, tn, tn->tn_dir.tn_parent, buf, 1762 buflen, dvp); 1763 tmpfs_free_node(tm, tnp); 1764 return (error); 1765 } 1766 restart: 1767 TMPFS_LOCK(tm); 1768 restart_locked: 1769 LIST_FOREACH_SAFE(tnp, &tm->tm_nodes_used, tn_entries, tnp1) { 1770 if (tnp->tn_type != VDIR) 1771 continue; 1772 TMPFS_NODE_LOCK(tnp); 1773 tmpfs_ref_node(tnp); 1774 1775 /* 1776 * tn_vnode cannot be instantiated while we hold the 1777 * node lock, so the directory cannot be changed while 1778 * we iterate over it. Do this to avoid instantiating 1779 * vnode for directories which cannot point to our 1780 * node. 1781 */ 1782 error = tnp->tn_vnode == NULL ? tmpfs_vptocnp_dir(tn, tnp, 1783 &de) : 0; 1784 1785 if (error == 0) { 1786 TMPFS_NODE_UNLOCK(tnp); 1787 TMPFS_UNLOCK(tm); 1788 error = tmpfs_vptocnp_fill(vp, tn, tnp, buf, buflen, 1789 dvp); 1790 if (error == 0) { 1791 tmpfs_free_node(tm, tnp); 1792 return (0); 1793 } 1794 if (VN_IS_DOOMED(vp)) { 1795 tmpfs_free_node(tm, tnp); 1796 return (ENOENT); 1797 } 1798 TMPFS_LOCK(tm); 1799 TMPFS_NODE_LOCK(tnp); 1800 } 1801 if (tmpfs_free_node_locked(tm, tnp, false)) { 1802 goto restart; 1803 } else { 1804 KASSERT(tnp->tn_refcount > 0, 1805 ("node %p refcount zero", tnp)); 1806 if (tnp->tn_attached) { 1807 tnp1 = LIST_NEXT(tnp, tn_entries); 1808 TMPFS_NODE_UNLOCK(tnp); 1809 } else { 1810 TMPFS_NODE_UNLOCK(tnp); 1811 goto restart_locked; 1812 } 1813 } 1814 } 1815 TMPFS_UNLOCK(tm); 1816 return (ENOENT); 1817 } 1818 1819 /* 1820 * Vnode operations vector used for files stored in a tmpfs file system. 1821 */ 1822 struct vop_vector tmpfs_vnodeop_entries = { 1823 .vop_default = &default_vnodeops, 1824 .vop_lookup = vfs_cache_lookup, 1825 .vop_cachedlookup = tmpfs_cached_lookup, 1826 .vop_create = tmpfs_create, 1827 .vop_mknod = tmpfs_mknod, 1828 .vop_open = tmpfs_open, 1829 .vop_close = tmpfs_close, 1830 .vop_fplookup_vexec = tmpfs_fplookup_vexec, 1831 .vop_fplookup_symlink = tmpfs_fplookup_symlink, 1832 .vop_access = tmpfs_access, 1833 .vop_stat = tmpfs_stat, 1834 .vop_getattr = tmpfs_getattr, 1835 .vop_setattr = tmpfs_setattr, 1836 .vop_read = tmpfs_read, 1837 .vop_read_pgcache = tmpfs_read_pgcache, 1838 .vop_write = tmpfs_write, 1839 .vop_deallocate = tmpfs_deallocate, 1840 .vop_fsync = tmpfs_fsync, 1841 .vop_remove = tmpfs_remove, 1842 .vop_link = tmpfs_link, 1843 .vop_rename = tmpfs_rename, 1844 .vop_mkdir = tmpfs_mkdir, 1845 .vop_rmdir = tmpfs_rmdir, 1846 .vop_symlink = tmpfs_symlink, 1847 .vop_readdir = tmpfs_readdir, 1848 .vop_readlink = tmpfs_readlink, 1849 .vop_inactive = tmpfs_inactive, 1850 .vop_need_inactive = tmpfs_need_inactive, 1851 .vop_reclaim = tmpfs_reclaim, 1852 .vop_print = tmpfs_print, 1853 .vop_pathconf = tmpfs_pathconf, 1854 .vop_vptofh = tmpfs_vptofh, 1855 .vop_whiteout = tmpfs_whiteout, 1856 .vop_bmap = VOP_EOPNOTSUPP, 1857 .vop_vptocnp = tmpfs_vptocnp, 1858 .vop_lock1 = vop_lock, 1859 .vop_unlock = vop_unlock, 1860 .vop_islocked = vop_islocked, 1861 .vop_add_writecount = vop_stdadd_writecount_nomsync, 1862 }; 1863 VFS_VOP_VECTOR_REGISTER(tmpfs_vnodeop_entries); 1864 1865 /* 1866 * Same vector for mounts which do not use namecache. 1867 */ 1868 struct vop_vector tmpfs_vnodeop_nonc_entries = { 1869 .vop_default = &tmpfs_vnodeop_entries, 1870 .vop_lookup = tmpfs_lookup, 1871 }; 1872 VFS_VOP_VECTOR_REGISTER(tmpfs_vnodeop_nonc_entries); 1873