1 /* 2 * Copyright (c) 2011-2015 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@dragonflybsd.org> 6 * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org> 7 * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression) 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 * 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in 17 * the documentation and/or other materials provided with the 18 * distribution. 19 * 3. Neither the name of The DragonFly Project nor the names of its 20 * contributors may be used to endorse or promote products derived 21 * from this software without specific, prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 26 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 27 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 28 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 29 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 30 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 31 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 32 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 33 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 */ 36 /* 37 * Kernel Filesystem interface 38 * 39 * NOTE! local ipdata pointers must be reloaded on any modifying operation 40 * to the inode as its underlying chain may have changed. 41 */ 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/kernel.h> 46 #include <sys/fcntl.h> 47 #include <sys/buf.h> 48 #include <sys/proc.h> 49 #include <sys/namei.h> 50 #include <sys/mount.h> 51 #include <sys/vnode.h> 52 #include <sys/mountctl.h> 53 #include <sys/dirent.h> 54 #include <sys/uio.h> 55 #include <sys/objcache.h> 56 #include <sys/event.h> 57 #include <sys/file.h> 58 #include <vfs/fifofs/fifo.h> 59 60 #include "hammer2.h" 61 62 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, 63 int seqcount); 64 static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio, 65 int ioflag, int seqcount); 66 static void hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize); 67 static void hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize); 68 69 struct objcache *cache_xops; 70 71 static __inline 72 void 73 hammer2_knote(struct vnode *vp, int flags) 74 { 75 if (flags) 76 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags); 77 } 78 79 /* 80 * Last reference to a vnode is going away but it is still cached. 81 */ 82 static 83 int 84 hammer2_vop_inactive(struct vop_inactive_args *ap) 85 { 86 hammer2_inode_t *ip; 87 struct vnode *vp; 88 89 LOCKSTART; 90 vp = ap->a_vp; 91 ip = VTOI(vp); 92 93 /* 94 * Degenerate case 95 */ 96 if (ip == NULL) { 97 vrecycle(vp); 98 LOCKSTOP; 99 return (0); 100 } 101 102 /* 103 * Check for deleted inodes and recycle immediately on the last 104 * release. Be sure to destroy any left-over buffer cache buffers 105 * so we do not waste time trying to flush them. 106 * 107 * Note that deleting the file block chains under the inode chain 108 * would just be a waste of energy, so don't do it. 109 * 110 * WARNING: nvtruncbuf() can only be safely called without the inode 111 * lock held due to the way our write thread works. 112 */ 113 if (ip->flags & HAMMER2_INODE_ISUNLINKED) { 114 hammer2_key_t lbase; 115 int nblksize; 116 117 /* 118 * Detect updates to the embedded data which may be 119 * synchronized by the strategy code. Simply mark the 120 * inode modified so it gets picked up by our normal flush. 121 */ 122 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL); 123 nvtruncbuf(vp, 0, nblksize, 0, 0); 124 vrecycle(vp); 125 } 126 LOCKSTOP; 127 return (0); 128 } 129 130 /* 131 * Reclaim a vnode so that it can be reused; after the inode is 132 * disassociated, the filesystem must manage it alone. 133 */ 134 static 135 int 136 hammer2_vop_reclaim(struct vop_reclaim_args *ap) 137 { 138 hammer2_inode_t *ip; 139 hammer2_pfs_t *pmp; 140 struct vnode *vp; 141 142 LOCKSTART; 143 vp = ap->a_vp; 144 ip = VTOI(vp); 145 if (ip == NULL) { 146 LOCKSTOP; 147 return(0); 148 } 149 pmp = ip->pmp; 150 151 /* 152 * The final close of a deleted file or directory marks it for 153 * destruction. The DELETED flag allows the flusher to shortcut 154 * any modified blocks still unflushed (that is, just ignore them). 155 * 156 * HAMMER2 usually does not try to optimize the freemap by returning 157 * deleted blocks to it as it does not usually know how many snapshots 158 * might be referencing portions of the file/dir. 159 */ 160 vp->v_data = NULL; 161 ip->vp = NULL; 162 163 /* 164 * NOTE! We do not attempt to flush chains here, flushing is 165 * really fragile and could also deadlock. 166 */ 167 vclrisdirty(vp); 168 169 /* 170 * An unlinked inode may have been relinked to the ihidden directory. 171 * This occurs if the inode was unlinked while open. Reclamation of 172 * these inodes requires processing we cannot safely do here so add 173 * the inode to the sideq in that situation. 174 * 175 * A modified inode may require chain synchronization which will no 176 * longer be driven by a sync or fsync without the vnode, also use 177 * the sideq for that. 178 * 179 * A reclaim can occur at any time so we cannot safely start a 180 * transaction to handle reclamation of unlinked files. Instead, 181 * the ip is left with a reference and placed on a linked list and 182 * handled later on. 183 */ 184 185 if ((ip->flags & (HAMMER2_INODE_ISUNLINKED | 186 HAMMER2_INODE_MODIFIED | 187 HAMMER2_INODE_RESIZED)) && 188 (ip->flags & HAMMER2_INODE_ISDELETED) == 0) { 189 hammer2_inode_sideq_t *ipul; 190 191 ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO); 192 ipul->ip = ip; 193 194 hammer2_spin_ex(&pmp->list_spin); 195 if ((ip->flags & HAMMER2_INODE_ONSIDEQ) == 0) { 196 /* ref -> sideq */ 197 atomic_set_int(&ip->flags, HAMMER2_INODE_ONSIDEQ); 198 TAILQ_INSERT_TAIL(&pmp->sideq, ipul, entry); 199 hammer2_spin_unex(&pmp->list_spin); 200 } else { 201 hammer2_spin_unex(&pmp->list_spin); 202 kfree(ipul, pmp->minode); 203 hammer2_inode_drop(ip); /* vp ref */ 204 } 205 /* retain ref from vp for ipul */ 206 } else { 207 hammer2_inode_drop(ip); /* vp ref */ 208 } 209 210 /* 211 * XXX handle background sync when ip dirty, kernel will no longer 212 * notify us regarding this inode because there is no longer a 213 * vnode attached to it. 214 */ 215 216 LOCKSTOP; 217 return (0); 218 } 219 220 static 221 int 222 hammer2_vop_fsync(struct vop_fsync_args *ap) 223 { 224 hammer2_inode_t *ip; 225 struct vnode *vp; 226 227 LOCKSTART; 228 vp = ap->a_vp; 229 ip = VTOI(vp); 230 231 #if 0 232 /* XXX can't do this yet */ 233 hammer2_trans_init(ip->pmp, HAMMER2_TRANS_ISFLUSH); 234 vfsync(vp, ap->a_waitfor, 1, NULL, NULL); 235 #endif 236 hammer2_trans_init(ip->pmp, 0); 237 vfsync(vp, ap->a_waitfor, 1, NULL, NULL); 238 239 /* 240 * Calling chain_flush here creates a lot of duplicative 241 * COW operations due to non-optimal vnode ordering. 242 * 243 * Only do it for an actual fsync() syscall. The other forms 244 * which call this function will eventually call chain_flush 245 * on the volume root as a catch-all, which is far more optimal. 246 */ 247 hammer2_inode_lock(ip, 0); 248 if (ip->flags & HAMMER2_INODE_MODIFIED) 249 hammer2_inode_chain_sync(ip); 250 hammer2_inode_unlock(ip); 251 hammer2_trans_done(ip->pmp); 252 253 LOCKSTOP; 254 return (0); 255 } 256 257 static 258 int 259 hammer2_vop_access(struct vop_access_args *ap) 260 { 261 hammer2_inode_t *ip = VTOI(ap->a_vp); 262 uid_t uid; 263 gid_t gid; 264 int error; 265 266 LOCKSTART; 267 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED); 268 uid = hammer2_to_unix_xid(&ip->meta.uid); 269 gid = hammer2_to_unix_xid(&ip->meta.gid); 270 error = vop_helper_access(ap, uid, gid, ip->meta.mode, ip->meta.uflags); 271 hammer2_inode_unlock(ip); 272 273 LOCKSTOP; 274 return (error); 275 } 276 277 static 278 int 279 hammer2_vop_getattr(struct vop_getattr_args *ap) 280 { 281 hammer2_pfs_t *pmp; 282 hammer2_inode_t *ip; 283 struct vnode *vp; 284 struct vattr *vap; 285 hammer2_chain_t *chain; 286 int i; 287 288 LOCKSTART; 289 vp = ap->a_vp; 290 vap = ap->a_vap; 291 292 ip = VTOI(vp); 293 pmp = ip->pmp; 294 295 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED); 296 297 vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0]; 298 vap->va_fileid = ip->meta.inum; 299 vap->va_mode = ip->meta.mode; 300 vap->va_nlink = ip->meta.nlinks; 301 vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid); 302 vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid); 303 vap->va_rmajor = 0; 304 vap->va_rminor = 0; 305 vap->va_size = ip->meta.size; /* protected by shared lock */ 306 vap->va_blocksize = HAMMER2_PBUFSIZE; 307 vap->va_flags = ip->meta.uflags; 308 hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime); 309 hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime); 310 hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime); 311 vap->va_gen = 1; 312 vap->va_bytes = 0; 313 if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY) { 314 /* 315 * Can't really calculate directory use sans the files under 316 * it, just assume one block for now. 317 */ 318 vap->va_bytes += HAMMER2_INODE_BYTES; 319 } else { 320 for (i = 0; i < ip->cluster.nchains; ++i) { 321 if ((chain = ip->cluster.array[i].chain) != NULL) { 322 if (vap->va_bytes < chain->bref.data_count) 323 vap->va_bytes = chain->bref.data_count; 324 } 325 } 326 } 327 vap->va_type = hammer2_get_vtype(ip->meta.type); 328 vap->va_filerev = 0; 329 vap->va_uid_uuid = ip->meta.uid; 330 vap->va_gid_uuid = ip->meta.gid; 331 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID | 332 VA_FSID_UUID_VALID; 333 334 hammer2_inode_unlock(ip); 335 336 LOCKSTOP; 337 return (0); 338 } 339 340 static 341 int 342 hammer2_vop_setattr(struct vop_setattr_args *ap) 343 { 344 hammer2_inode_t *ip; 345 struct vnode *vp; 346 struct vattr *vap; 347 int error; 348 int kflags = 0; 349 uint64_t ctime; 350 351 LOCKSTART; 352 vp = ap->a_vp; 353 vap = ap->a_vap; 354 hammer2_update_time(&ctime); 355 356 ip = VTOI(vp); 357 358 if (ip->pmp->ronly) { 359 LOCKSTOP; 360 return(EROFS); 361 } 362 363 hammer2_pfs_memory_wait(ip->pmp); 364 hammer2_trans_init(ip->pmp, 0); 365 hammer2_inode_lock(ip, 0); 366 error = 0; 367 368 if (vap->va_flags != VNOVAL) { 369 uint32_t flags; 370 371 flags = ip->meta.uflags; 372 error = vop_helper_setattr_flags(&flags, vap->va_flags, 373 hammer2_to_unix_xid(&ip->meta.uid), 374 ap->a_cred); 375 if (error == 0) { 376 if (ip->meta.uflags != flags) { 377 hammer2_inode_modify(ip); 378 ip->meta.uflags = flags; 379 ip->meta.ctime = ctime; 380 kflags |= NOTE_ATTRIB; 381 } 382 if (ip->meta.uflags & (IMMUTABLE | APPEND)) { 383 error = 0; 384 goto done; 385 } 386 } 387 goto done; 388 } 389 if (ip->meta.uflags & (IMMUTABLE | APPEND)) { 390 error = EPERM; 391 goto done; 392 } 393 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { 394 mode_t cur_mode = ip->meta.mode; 395 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid); 396 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid); 397 uuid_t uuid_uid; 398 uuid_t uuid_gid; 399 400 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid, 401 ap->a_cred, 402 &cur_uid, &cur_gid, &cur_mode); 403 if (error == 0) { 404 hammer2_guid_to_uuid(&uuid_uid, cur_uid); 405 hammer2_guid_to_uuid(&uuid_gid, cur_gid); 406 if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) || 407 bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) || 408 ip->meta.mode != cur_mode 409 ) { 410 hammer2_inode_modify(ip); 411 ip->meta.uid = uuid_uid; 412 ip->meta.gid = uuid_gid; 413 ip->meta.mode = cur_mode; 414 ip->meta.ctime = ctime; 415 } 416 kflags |= NOTE_ATTRIB; 417 } 418 } 419 420 /* 421 * Resize the file 422 */ 423 if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) { 424 switch(vp->v_type) { 425 case VREG: 426 if (vap->va_size == ip->meta.size) 427 break; 428 if (vap->va_size < ip->meta.size) { 429 hammer2_mtx_ex(&ip->truncate_lock); 430 hammer2_truncate_file(ip, vap->va_size); 431 hammer2_mtx_unlock(&ip->truncate_lock); 432 } else { 433 hammer2_extend_file(ip, vap->va_size); 434 } 435 hammer2_inode_modify(ip); 436 ip->meta.mtime = ctime; 437 break; 438 default: 439 error = EINVAL; 440 goto done; 441 } 442 } 443 #if 0 444 /* atime not supported */ 445 if (vap->va_atime.tv_sec != VNOVAL) { 446 hammer2_inode_modify(ip); 447 ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime); 448 kflags |= NOTE_ATTRIB; 449 } 450 #endif 451 if (vap->va_mode != (mode_t)VNOVAL) { 452 mode_t cur_mode = ip->meta.mode; 453 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid); 454 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid); 455 456 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred, 457 cur_uid, cur_gid, &cur_mode); 458 if (error == 0 && ip->meta.mode != cur_mode) { 459 hammer2_inode_modify(ip); 460 ip->meta.mode = cur_mode; 461 ip->meta.ctime = ctime; 462 kflags |= NOTE_ATTRIB; 463 } 464 } 465 466 if (vap->va_mtime.tv_sec != VNOVAL) { 467 hammer2_inode_modify(ip); 468 ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime); 469 kflags |= NOTE_ATTRIB; 470 } 471 472 done: 473 /* 474 * If a truncation occurred we must call inode_fsync() now in order 475 * to trim the related data chains, otherwise a later expansion can 476 * cause havoc. 477 * 478 * If an extend occured that changed the DIRECTDATA state, we must 479 * call inode_fsync now in order to prepare the inode's indirect 480 * block table. 481 */ 482 if (ip->flags & HAMMER2_INODE_RESIZED) 483 hammer2_inode_chain_sync(ip); 484 485 /* 486 * Cleanup. 487 */ 488 hammer2_inode_unlock(ip); 489 hammer2_trans_done(ip->pmp); 490 hammer2_knote(ip->vp, kflags); 491 492 LOCKSTOP; 493 return (error); 494 } 495 496 static 497 int 498 hammer2_vop_readdir(struct vop_readdir_args *ap) 499 { 500 hammer2_xop_readdir_t *xop; 501 hammer2_blockref_t bref; 502 hammer2_inode_t *ip; 503 hammer2_tid_t inum; 504 hammer2_key_t lkey; 505 struct uio *uio; 506 off_t *cookies; 507 off_t saveoff; 508 int cookie_index; 509 int ncookies; 510 int error; 511 int eofflag; 512 int dtype; 513 int r; 514 515 LOCKSTART; 516 ip = VTOI(ap->a_vp); 517 uio = ap->a_uio; 518 saveoff = uio->uio_offset; 519 eofflag = 0; 520 error = 0; 521 522 /* 523 * Setup cookies directory entry cookies if requested 524 */ 525 if (ap->a_ncookies) { 526 ncookies = uio->uio_resid / 16 + 1; 527 if (ncookies > 1024) 528 ncookies = 1024; 529 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK); 530 } else { 531 ncookies = -1; 532 cookies = NULL; 533 } 534 cookie_index = 0; 535 536 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED); 537 538 /* 539 * Handle artificial entries. To ensure that only positive 64 bit 540 * quantities are returned to userland we always strip off bit 63. 541 * The hash code is designed such that codes 0x0000-0x7FFF are not 542 * used, allowing us to use these codes for articial entries. 543 * 544 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not 545 * allow '..' to cross the mount point into (e.g.) the super-root. 546 */ 547 if (saveoff == 0) { 548 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK; 549 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, "."); 550 if (r) 551 goto done; 552 if (cookies) 553 cookies[cookie_index] = saveoff; 554 ++saveoff; 555 ++cookie_index; 556 if (cookie_index == ncookies) 557 goto done; 558 } 559 560 if (saveoff == 1) { 561 /* 562 * Be careful with lockorder when accessing ".." 563 * 564 * (ip is the current dir. xip is the parent dir). 565 */ 566 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK; 567 if (ip != ip->pmp->iroot) 568 inum = ip->meta.iparent & HAMMER2_DIRHASH_USERMSK; 569 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, ".."); 570 if (r) 571 goto done; 572 if (cookies) 573 cookies[cookie_index] = saveoff; 574 ++saveoff; 575 ++cookie_index; 576 if (cookie_index == ncookies) 577 goto done; 578 } 579 580 lkey = saveoff | HAMMER2_DIRHASH_VISIBLE; 581 if (hammer2_debug & 0x0020) 582 kprintf("readdir: lkey %016jx\n", lkey); 583 if (error) 584 goto done; 585 586 /* 587 * Use XOP for cluster scan. 588 * 589 * parent is the inode cluster, already locked for us. Don't 590 * double lock shared locks as this will screw up upgrades. 591 */ 592 xop = hammer2_xop_alloc(ip, 0); 593 xop->lkey = lkey; 594 hammer2_xop_start(&xop->head, hammer2_xop_readdir); 595 596 for (;;) { 597 const hammer2_inode_data_t *ripdata; 598 599 error = hammer2_xop_collect(&xop->head, 0); 600 if (error) 601 break; 602 if (cookie_index == ncookies) 603 break; 604 if (hammer2_debug & 0x0020) 605 kprintf("cluster chain %p %p\n", 606 xop->head.cluster.focus, 607 (xop->head.cluster.focus ? 608 xop->head.cluster.focus->data : (void *)-1)); 609 ripdata = &hammer2_cluster_rdata(&xop->head.cluster)->ipdata; 610 hammer2_cluster_bref(&xop->head.cluster, &bref); 611 if (bref.type == HAMMER2_BREF_TYPE_INODE) { 612 dtype = hammer2_get_dtype(ripdata); 613 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK; 614 r = vop_write_dirent(&error, uio, 615 ripdata->meta.inum & 616 HAMMER2_DIRHASH_USERMSK, 617 dtype, 618 ripdata->meta.name_len, 619 ripdata->filename); 620 if (r) 621 break; 622 if (cookies) 623 cookies[cookie_index] = saveoff; 624 ++cookie_index; 625 } else { 626 /* XXX chain error */ 627 kprintf("bad chain type readdir %d\n", bref.type); 628 } 629 } 630 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 631 if (error == ENOENT) { 632 error = 0; 633 eofflag = 1; 634 saveoff = (hammer2_key_t)-1; 635 } else { 636 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK; 637 } 638 done: 639 hammer2_inode_unlock(ip); 640 if (ap->a_eofflag) 641 *ap->a_eofflag = eofflag; 642 if (hammer2_debug & 0x0020) 643 kprintf("readdir: done at %016jx\n", saveoff); 644 uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE; 645 if (error && cookie_index == 0) { 646 if (cookies) { 647 kfree(cookies, M_TEMP); 648 *ap->a_ncookies = 0; 649 *ap->a_cookies = NULL; 650 } 651 } else { 652 if (cookies) { 653 *ap->a_ncookies = cookie_index; 654 *ap->a_cookies = cookies; 655 } 656 } 657 LOCKSTOP; 658 return (error); 659 } 660 661 /* 662 * hammer2_vop_readlink { vp, uio, cred } 663 */ 664 static 665 int 666 hammer2_vop_readlink(struct vop_readlink_args *ap) 667 { 668 struct vnode *vp; 669 hammer2_inode_t *ip; 670 int error; 671 672 vp = ap->a_vp; 673 if (vp->v_type != VLNK) 674 return (EINVAL); 675 ip = VTOI(vp); 676 677 error = hammer2_read_file(ip, ap->a_uio, 0); 678 return (error); 679 } 680 681 static 682 int 683 hammer2_vop_read(struct vop_read_args *ap) 684 { 685 struct vnode *vp; 686 hammer2_inode_t *ip; 687 struct uio *uio; 688 int error; 689 int seqcount; 690 int bigread; 691 692 /* 693 * Read operations supported on this vnode? 694 */ 695 vp = ap->a_vp; 696 if (vp->v_type != VREG) 697 return (EINVAL); 698 699 /* 700 * Misc 701 */ 702 ip = VTOI(vp); 703 uio = ap->a_uio; 704 error = 0; 705 706 seqcount = ap->a_ioflag >> 16; 707 bigread = (uio->uio_resid > 100 * 1024 * 1024); 708 709 error = hammer2_read_file(ip, uio, seqcount); 710 return (error); 711 } 712 713 static 714 int 715 hammer2_vop_write(struct vop_write_args *ap) 716 { 717 hammer2_inode_t *ip; 718 thread_t td; 719 struct vnode *vp; 720 struct uio *uio; 721 int error; 722 int seqcount; 723 724 /* 725 * Read operations supported on this vnode? 726 */ 727 vp = ap->a_vp; 728 if (vp->v_type != VREG) 729 return (EINVAL); 730 731 /* 732 * Misc 733 */ 734 ip = VTOI(vp); 735 uio = ap->a_uio; 736 error = 0; 737 if (ip->pmp->ronly) { 738 return (EROFS); 739 } 740 741 seqcount = ap->a_ioflag >> 16; 742 743 /* 744 * Check resource limit 745 */ 746 if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc && 747 uio->uio_offset + uio->uio_resid > 748 td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 749 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ); 750 return (EFBIG); 751 } 752 753 /* 754 * The transaction interlocks against flushes initiations 755 * (note: but will run concurrently with the actual flush). 756 */ 757 hammer2_trans_init(ip->pmp, 0); 758 error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount); 759 hammer2_trans_done(ip->pmp); 760 761 return (error); 762 } 763 764 /* 765 * Perform read operations on a file or symlink given an UNLOCKED 766 * inode and uio. 767 * 768 * The passed ip is not locked. 769 */ 770 static 771 int 772 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount) 773 { 774 hammer2_off_t size; 775 struct buf *bp; 776 int error; 777 778 error = 0; 779 780 /* 781 * UIO read loop. 782 * 783 * WARNING! Assumes that the kernel interlocks size changes at the 784 * vnode level. 785 */ 786 hammer2_mtx_sh(&ip->lock); 787 hammer2_mtx_sh(&ip->truncate_lock); 788 size = ip->meta.size; 789 hammer2_mtx_unlock(&ip->lock); 790 791 while (uio->uio_resid > 0 && uio->uio_offset < size) { 792 hammer2_key_t lbase; 793 hammer2_key_t leof; 794 int lblksize; 795 int loff; 796 int n; 797 798 lblksize = hammer2_calc_logical(ip, uio->uio_offset, 799 &lbase, &leof); 800 801 #if 1 802 error = cluster_read(ip->vp, leof, lbase, lblksize, 803 uio->uio_resid, seqcount * BKVASIZE, 804 &bp); 805 #else 806 if (uio->uio_segflg == UIO_NOCOPY) { 807 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0); 808 if (bp->b_flags & B_CACHE) { 809 int i; 810 int j = 0; 811 if (bp->b_xio.xio_npages != 16) 812 kprintf("NPAGES BAD\n"); 813 for (i = 0; i < bp->b_xio.xio_npages; ++i) { 814 vm_page_t m; 815 m = bp->b_xio.xio_pages[i]; 816 if (m == NULL || m->valid == 0) { 817 kprintf("bp %016jx %016jx pg %d inv", 818 lbase, leof, i); 819 if (m) 820 kprintf("m->object %p/%p", m->object, ip->vp->v_object); 821 kprintf("\n"); 822 j = 1; 823 } 824 } 825 if (j) 826 kprintf("b_flags %08x, b_error %d\n", bp->b_flags, bp->b_error); 827 } 828 bqrelse(bp); 829 } 830 error = bread(ip->vp, lbase, lblksize, &bp); 831 #endif 832 if (error) 833 break; 834 loff = (int)(uio->uio_offset - lbase); 835 n = lblksize - loff; 836 if (n > uio->uio_resid) 837 n = uio->uio_resid; 838 if (n > size - uio->uio_offset) 839 n = (int)(size - uio->uio_offset); 840 bp->b_flags |= B_AGE; 841 uiomovebp(bp, (char *)bp->b_data + loff, n, uio); 842 bqrelse(bp); 843 } 844 hammer2_mtx_unlock(&ip->truncate_lock); 845 846 return (error); 847 } 848 849 /* 850 * Write to the file represented by the inode via the logical buffer cache. 851 * The inode may represent a regular file or a symlink. 852 * 853 * The inode must not be locked. 854 */ 855 static 856 int 857 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio, 858 int ioflag, int seqcount) 859 { 860 hammer2_key_t old_eof; 861 hammer2_key_t new_eof; 862 struct buf *bp; 863 int kflags; 864 int error; 865 int modified; 866 867 /* 868 * Setup if append 869 * 870 * WARNING! Assumes that the kernel interlocks size changes at the 871 * vnode level. 872 */ 873 hammer2_mtx_ex(&ip->lock); 874 hammer2_mtx_sh(&ip->truncate_lock); 875 if (ioflag & IO_APPEND) 876 uio->uio_offset = ip->meta.size; 877 old_eof = ip->meta.size; 878 879 /* 880 * Extend the file if necessary. If the write fails at some point 881 * we will truncate it back down to cover as much as we were able 882 * to write. 883 * 884 * Doing this now makes it easier to calculate buffer sizes in 885 * the loop. 886 */ 887 kflags = 0; 888 error = 0; 889 modified = 0; 890 891 if (uio->uio_offset + uio->uio_resid > old_eof) { 892 new_eof = uio->uio_offset + uio->uio_resid; 893 modified = 1; 894 hammer2_extend_file(ip, new_eof); 895 kflags |= NOTE_EXTEND; 896 } else { 897 new_eof = old_eof; 898 } 899 hammer2_mtx_unlock(&ip->lock); 900 901 /* 902 * UIO write loop 903 */ 904 while (uio->uio_resid > 0) { 905 hammer2_key_t lbase; 906 int trivial; 907 int endofblk; 908 int lblksize; 909 int loff; 910 int n; 911 912 /* 913 * Don't allow the buffer build to blow out the buffer 914 * cache. 915 */ 916 if ((ioflag & IO_RECURSE) == 0) 917 bwillwrite(HAMMER2_PBUFSIZE); 918 919 /* 920 * This nominally tells us how much we can cluster and 921 * what the logical buffer size needs to be. Currently 922 * we don't try to cluster the write and just handle one 923 * block at a time. 924 */ 925 lblksize = hammer2_calc_logical(ip, uio->uio_offset, 926 &lbase, NULL); 927 loff = (int)(uio->uio_offset - lbase); 928 929 KKASSERT(lblksize <= 65536); 930 931 /* 932 * Calculate bytes to copy this transfer and whether the 933 * copy completely covers the buffer or not. 934 */ 935 trivial = 0; 936 n = lblksize - loff; 937 if (n > uio->uio_resid) { 938 n = uio->uio_resid; 939 if (loff == lbase && uio->uio_offset + n == new_eof) 940 trivial = 1; 941 endofblk = 0; 942 } else { 943 if (loff == 0) 944 trivial = 1; 945 endofblk = 1; 946 } 947 if (lbase >= new_eof) 948 trivial = 1; 949 950 /* 951 * Get the buffer 952 */ 953 if (uio->uio_segflg == UIO_NOCOPY) { 954 /* 955 * Issuing a write with the same data backing the 956 * buffer. Instantiate the buffer to collect the 957 * backing vm pages, then read-in any missing bits. 958 * 959 * This case is used by vop_stdputpages(). 960 */ 961 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0); 962 if ((bp->b_flags & B_CACHE) == 0) { 963 bqrelse(bp); 964 error = bread(ip->vp, lbase, lblksize, &bp); 965 } 966 } else if (trivial) { 967 /* 968 * Even though we are entirely overwriting the buffer 969 * we may still have to zero it out to avoid a 970 * mmap/write visibility issue. 971 */ 972 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0); 973 if ((bp->b_flags & B_CACHE) == 0) 974 vfs_bio_clrbuf(bp); 975 } else { 976 /* 977 * Partial overwrite, read in any missing bits then 978 * replace the portion being written. 979 * 980 * (The strategy code will detect zero-fill physical 981 * blocks for this case). 982 */ 983 error = bread(ip->vp, lbase, lblksize, &bp); 984 if (error == 0) 985 bheavy(bp); 986 } 987 988 if (error) { 989 brelse(bp); 990 break; 991 } 992 993 /* 994 * Ok, copy the data in 995 */ 996 error = uiomovebp(bp, bp->b_data + loff, n, uio); 997 kflags |= NOTE_WRITE; 998 modified = 1; 999 if (error) { 1000 brelse(bp); 1001 break; 1002 } 1003 1004 /* 1005 * WARNING: Pageout daemon will issue UIO_NOCOPY writes 1006 * with IO_SYNC or IO_ASYNC set. These writes 1007 * must be handled as the pageout daemon expects. 1008 * 1009 * NOTE! H2 relies on cluster_write() here because it 1010 * cannot preallocate disk blocks at the logical 1011 * level due to not knowing what the compression 1012 * size will be at this time. 1013 * 1014 * We must use cluster_write() here and we depend 1015 * on the write-behind feature to flush buffers 1016 * appropriately. If we let the buffer daemons do 1017 * it the block allocations will be all over the 1018 * map. 1019 */ 1020 if (ioflag & IO_SYNC) { 1021 bwrite(bp); 1022 } else if ((ioflag & IO_DIRECT) && endofblk) { 1023 bawrite(bp); 1024 } else if (ioflag & IO_ASYNC) { 1025 bawrite(bp); 1026 } else if (ip->vp->v_mount->mnt_flag & MNT_NOCLUSTERW) { 1027 bdwrite(bp); 1028 } else { 1029 #if 1 1030 bp->b_flags |= B_CLUSTEROK; 1031 cluster_write(bp, new_eof, lblksize, seqcount); 1032 #else 1033 bdwrite(bp); 1034 #endif 1035 } 1036 } 1037 1038 /* 1039 * Cleanup. If we extended the file EOF but failed to write through 1040 * the entire write is a failure and we have to back-up. 1041 */ 1042 if (error && new_eof != old_eof) { 1043 hammer2_mtx_unlock(&ip->truncate_lock); 1044 hammer2_mtx_ex(&ip->lock); 1045 hammer2_mtx_ex(&ip->truncate_lock); 1046 hammer2_truncate_file(ip, old_eof); 1047 if (ip->flags & HAMMER2_INODE_MODIFIED) 1048 hammer2_inode_chain_sync(ip); 1049 hammer2_mtx_unlock(&ip->lock); 1050 } else if (modified) { 1051 hammer2_mtx_ex(&ip->lock); 1052 hammer2_inode_modify(ip); 1053 hammer2_update_time(&ip->meta.mtime); 1054 if (ip->flags & HAMMER2_INODE_MODIFIED) 1055 hammer2_inode_chain_sync(ip); 1056 hammer2_mtx_unlock(&ip->lock); 1057 hammer2_knote(ip->vp, kflags); 1058 } 1059 hammer2_trans_assert_strategy(ip->pmp); 1060 hammer2_mtx_unlock(&ip->truncate_lock); 1061 1062 return error; 1063 } 1064 1065 /* 1066 * Truncate the size of a file. The inode must not be locked. 1067 * 1068 * We must unconditionally set HAMMER2_INODE_RESIZED to properly 1069 * ensure that any on-media data beyond the new file EOF has been destroyed. 1070 * 1071 * WARNING: nvtruncbuf() can only be safely called without the inode lock 1072 * held due to the way our write thread works. If the truncation 1073 * occurs in the middle of a buffer, nvtruncbuf() is responsible 1074 * for dirtying that buffer and zeroing out trailing bytes. 1075 * 1076 * WARNING! Assumes that the kernel interlocks size changes at the 1077 * vnode level. 1078 * 1079 * WARNING! Caller assumes responsibility for removing dead blocks 1080 * if INODE_RESIZED is set. 1081 */ 1082 static 1083 void 1084 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize) 1085 { 1086 hammer2_key_t lbase; 1087 int nblksize; 1088 1089 LOCKSTART; 1090 hammer2_mtx_unlock(&ip->lock); 1091 if (ip->vp) { 1092 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL); 1093 nvtruncbuf(ip->vp, nsize, 1094 nblksize, (int)nsize & (nblksize - 1), 1095 0); 1096 } 1097 hammer2_mtx_ex(&ip->lock); 1098 KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0); 1099 ip->osize = ip->meta.size; 1100 ip->meta.size = nsize; 1101 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED); 1102 hammer2_inode_modify(ip); 1103 LOCKSTOP; 1104 } 1105 1106 /* 1107 * Extend the size of a file. The inode must not be locked. 1108 * 1109 * Even though the file size is changing, we do not have to set the 1110 * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES 1111 * boundary. When this occurs a hammer2_inode_chain_sync() is required 1112 * to prepare the inode cluster's indirect block table, otherwise 1113 * async execution of the strategy code will implode on us. 1114 * 1115 * WARNING! Assumes that the kernel interlocks size changes at the 1116 * vnode level. 1117 * 1118 * WARNING! Caller assumes responsibility for transitioning out 1119 * of the inode DIRECTDATA mode if INODE_RESIZED is set. 1120 */ 1121 static 1122 void 1123 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize) 1124 { 1125 hammer2_key_t lbase; 1126 hammer2_key_t osize; 1127 int oblksize; 1128 int nblksize; 1129 1130 LOCKSTART; 1131 1132 KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0); 1133 hammer2_inode_modify(ip); 1134 osize = ip->meta.size; 1135 ip->osize = osize; 1136 ip->meta.size = nsize; 1137 1138 if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) { 1139 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED); 1140 hammer2_inode_chain_sync(ip); 1141 } 1142 1143 hammer2_mtx_unlock(&ip->lock); 1144 if (ip->vp) { 1145 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL); 1146 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL); 1147 nvextendbuf(ip->vp, 1148 osize, nsize, 1149 oblksize, nblksize, 1150 -1, -1, 0); 1151 } 1152 hammer2_mtx_ex(&ip->lock); 1153 1154 LOCKSTOP; 1155 } 1156 1157 static 1158 int 1159 hammer2_vop_nresolve(struct vop_nresolve_args *ap) 1160 { 1161 hammer2_xop_nresolve_t *xop; 1162 hammer2_inode_t *ip; 1163 hammer2_inode_t *dip; 1164 struct namecache *ncp; 1165 struct vnode *vp; 1166 int error; 1167 1168 LOCKSTART; 1169 dip = VTOI(ap->a_dvp); 1170 xop = hammer2_xop_alloc(dip, 0); 1171 1172 ncp = ap->a_nch->ncp; 1173 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen); 1174 1175 /* 1176 * Note: In DragonFly the kernel handles '.' and '..'. 1177 */ 1178 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1179 hammer2_xop_start(&xop->head, hammer2_xop_nresolve); 1180 1181 error = hammer2_xop_collect(&xop->head, 0); 1182 if (error) { 1183 ip = NULL; 1184 } else { 1185 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1); 1186 } 1187 hammer2_inode_unlock(dip); 1188 1189 /* 1190 * Acquire the related vnode 1191 * 1192 * NOTE: For error processing, only ENOENT resolves the namecache 1193 * entry to NULL, otherwise we just return the error and 1194 * leave the namecache unresolved. 1195 * 1196 * NOTE: multiple hammer2_inode structures can be aliased to the 1197 * same chain element, for example for hardlinks. This 1198 * use case does not 'reattach' inode associations that 1199 * might already exist, but always allocates a new one. 1200 * 1201 * WARNING: inode structure is locked exclusively via inode_get 1202 * but chain was locked shared. inode_unlock() 1203 * will handle it properly. 1204 */ 1205 if (ip) { 1206 vp = hammer2_igetv(ip, &error); 1207 if (error == 0) { 1208 vn_unlock(vp); 1209 cache_setvp(ap->a_nch, vp); 1210 } else if (error == ENOENT) { 1211 cache_setvp(ap->a_nch, NULL); 1212 } 1213 hammer2_inode_unlock(ip); 1214 1215 /* 1216 * The vp should not be released until after we've disposed 1217 * of our locks, because it might cause vop_inactive() to 1218 * be called. 1219 */ 1220 if (vp) 1221 vrele(vp); 1222 } else { 1223 error = ENOENT; 1224 cache_setvp(ap->a_nch, NULL); 1225 } 1226 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 1227 KASSERT(error || ap->a_nch->ncp->nc_vp != NULL, 1228 ("resolve error %d/%p ap %p\n", 1229 error, ap->a_nch->ncp->nc_vp, ap)); 1230 LOCKSTOP; 1231 1232 return error; 1233 } 1234 1235 static 1236 int 1237 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap) 1238 { 1239 hammer2_inode_t *dip; 1240 hammer2_tid_t inum; 1241 int error; 1242 1243 LOCKSTART; 1244 dip = VTOI(ap->a_dvp); 1245 inum = dip->meta.iparent; 1246 *ap->a_vpp = NULL; 1247 1248 if (inum) { 1249 error = hammer2_vfs_vget(ap->a_dvp->v_mount, NULL, 1250 inum, ap->a_vpp); 1251 } else { 1252 error = ENOENT; 1253 } 1254 LOCKSTOP; 1255 return error; 1256 } 1257 1258 static 1259 int 1260 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap) 1261 { 1262 hammer2_inode_t *dip; 1263 hammer2_inode_t *nip; 1264 struct namecache *ncp; 1265 const uint8_t *name; 1266 size_t name_len; 1267 hammer2_tid_t inum; 1268 int error; 1269 1270 LOCKSTART; 1271 dip = VTOI(ap->a_dvp); 1272 if (dip->pmp->ronly) { 1273 LOCKSTOP; 1274 return (EROFS); 1275 } 1276 1277 ncp = ap->a_nch->ncp; 1278 name = ncp->nc_name; 1279 name_len = ncp->nc_nlen; 1280 1281 hammer2_pfs_memory_wait(dip->pmp); 1282 hammer2_trans_init(dip->pmp, 0); 1283 1284 inum = hammer2_trans_newinum(dip->pmp); 1285 1286 /* 1287 * Create the actual inode as a hidden file in the iroot, then 1288 * create the directory entry as a hardlink to it. The creation 1289 * of the actual inode sets its nlinks to 1 which is the value 1290 * we desire. 1291 */ 1292 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred, 1293 NULL, 0, inum, 1294 inum, 0, 0, 1295 0, &error); 1296 if (error == 0) { 1297 hammer2_inode_create(dip, dip, NULL, NULL, 1298 name, name_len, 0, 1299 nip->meta.inum, 1300 HAMMER2_OBJTYPE_HARDLINK, nip->meta.type, 1301 0, &error); 1302 } 1303 1304 if (error) { 1305 KKASSERT(nip == NULL); 1306 *ap->a_vpp = NULL; 1307 } else { 1308 *ap->a_vpp = hammer2_igetv(nip, &error); 1309 hammer2_inode_unlock(nip); 1310 } 1311 1312 /* 1313 * Update dip's mtime 1314 */ 1315 if (error == 0) { 1316 uint64_t mtime; 1317 1318 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1319 hammer2_update_time(&mtime); 1320 hammer2_inode_modify(dip); 1321 dip->meta.mtime = mtime; 1322 hammer2_inode_unlock(dip); 1323 } 1324 1325 hammer2_trans_done(dip->pmp); 1326 1327 if (error == 0) { 1328 cache_setunresolved(ap->a_nch); 1329 cache_setvp(ap->a_nch, *ap->a_vpp); 1330 } 1331 LOCKSTOP; 1332 return error; 1333 } 1334 1335 static 1336 int 1337 hammer2_vop_open(struct vop_open_args *ap) 1338 { 1339 return vop_stdopen(ap); 1340 } 1341 1342 /* 1343 * hammer2_vop_advlock { vp, id, op, fl, flags } 1344 */ 1345 static 1346 int 1347 hammer2_vop_advlock(struct vop_advlock_args *ap) 1348 { 1349 hammer2_inode_t *ip = VTOI(ap->a_vp); 1350 hammer2_off_t size; 1351 1352 size = ip->meta.size; 1353 return (lf_advlock(ap, &ip->advlock, size)); 1354 } 1355 1356 static 1357 int 1358 hammer2_vop_close(struct vop_close_args *ap) 1359 { 1360 return vop_stdclose(ap); 1361 } 1362 1363 /* 1364 * hammer2_vop_nlink { nch, dvp, vp, cred } 1365 * 1366 * Create a hardlink from (vp) to {dvp, nch}. 1367 */ 1368 static 1369 int 1370 hammer2_vop_nlink(struct vop_nlink_args *ap) 1371 { 1372 hammer2_inode_t *tdip; /* target directory to create link in */ 1373 hammer2_inode_t *ip; /* inode we are hardlinking to */ 1374 struct namecache *ncp; 1375 const uint8_t *name; 1376 size_t name_len; 1377 int error; 1378 1379 LOCKSTART; 1380 tdip = VTOI(ap->a_dvp); 1381 if (tdip->pmp->ronly) { 1382 LOCKSTOP; 1383 return (EROFS); 1384 } 1385 1386 ncp = ap->a_nch->ncp; 1387 name = ncp->nc_name; 1388 name_len = ncp->nc_nlen; 1389 1390 /* 1391 * ip represents the file being hardlinked. The file could be a 1392 * normal file or a hardlink target if it has already been hardlinked. 1393 * (with the new semantics, it will almost always be a hardlink 1394 * target). 1395 * 1396 * Bump nlinks and potentially also create or move the hardlink 1397 * target in the parent directory common to (ip) and (tdip). The 1398 * consolidation code can modify ip->cluster. The returned cluster 1399 * is locked. 1400 */ 1401 ip = VTOI(ap->a_vp); 1402 hammer2_pfs_memory_wait(ip->pmp); 1403 hammer2_trans_init(ip->pmp, 0); 1404 1405 /* 1406 * Target should be an indexed inode or there's no way we will ever 1407 * be able to find it! 1408 */ 1409 KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0); 1410 1411 error = 0; 1412 1413 /* 1414 * Can return NULL and error == EXDEV if the common parent 1415 * crosses a directory with the xlink flag set. 1416 */ 1417 hammer2_inode_lock(tdip, 0); 1418 hammer2_inode_lock(ip, 0); 1419 1420 /* 1421 * Create the hardlink target and bump nlinks. 1422 */ 1423 if (error == 0) { 1424 hammer2_inode_create(tdip, tdip, NULL, NULL, 1425 name, name_len, 0, 1426 ip->meta.inum, 1427 HAMMER2_OBJTYPE_HARDLINK, ip->meta.type, 1428 0, &error); 1429 hammer2_inode_modify(ip); 1430 ++ip->meta.nlinks; 1431 } 1432 if (error == 0) { 1433 /* 1434 * Update dip's mtime 1435 */ 1436 uint64_t mtime; 1437 1438 hammer2_update_time(&mtime); 1439 hammer2_inode_modify(tdip); 1440 tdip->meta.mtime = mtime; 1441 1442 cache_setunresolved(ap->a_nch); 1443 cache_setvp(ap->a_nch, ap->a_vp); 1444 } 1445 hammer2_inode_unlock(ip); 1446 hammer2_inode_unlock(tdip); 1447 1448 hammer2_trans_done(ip->pmp); 1449 1450 LOCKSTOP; 1451 return error; 1452 } 1453 1454 /* 1455 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap } 1456 * 1457 * The operating system has already ensured that the directory entry 1458 * does not exist and done all appropriate namespace locking. 1459 */ 1460 static 1461 int 1462 hammer2_vop_ncreate(struct vop_ncreate_args *ap) 1463 { 1464 hammer2_inode_t *dip; 1465 hammer2_inode_t *nip; 1466 struct namecache *ncp; 1467 const uint8_t *name; 1468 size_t name_len; 1469 hammer2_tid_t inum; 1470 int error; 1471 1472 LOCKSTART; 1473 dip = VTOI(ap->a_dvp); 1474 if (dip->pmp->ronly) { 1475 LOCKSTOP; 1476 return (EROFS); 1477 } 1478 1479 ncp = ap->a_nch->ncp; 1480 name = ncp->nc_name; 1481 name_len = ncp->nc_nlen; 1482 hammer2_pfs_memory_wait(dip->pmp); 1483 hammer2_trans_init(dip->pmp, 0); 1484 1485 inum = hammer2_trans_newinum(dip->pmp); 1486 1487 /* 1488 * Create the actual inode as a hidden file in the iroot, then 1489 * create the directory entry as a hardlink to it. The creation 1490 * of the actual inode sets its nlinks to 1 which is the value 1491 * we desire. 1492 */ 1493 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred, 1494 NULL, 0, inum, 1495 inum, 0, 0, 1496 0, &error); 1497 1498 if (error == 0) { 1499 hammer2_inode_create(dip, dip, NULL, NULL, 1500 name, name_len, 0, 1501 nip->meta.inum, 1502 HAMMER2_OBJTYPE_HARDLINK, nip->meta.type, 1503 0, &error); 1504 } 1505 if (error) { 1506 KKASSERT(nip == NULL); 1507 *ap->a_vpp = NULL; 1508 } else { 1509 *ap->a_vpp = hammer2_igetv(nip, &error); 1510 hammer2_inode_unlock(nip); 1511 } 1512 1513 /* 1514 * Update dip's mtime 1515 */ 1516 if (error == 0) { 1517 uint64_t mtime; 1518 1519 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1520 hammer2_update_time(&mtime); 1521 hammer2_inode_modify(dip); 1522 dip->meta.mtime = mtime; 1523 hammer2_inode_unlock(dip); 1524 } 1525 1526 hammer2_trans_done(dip->pmp); 1527 1528 if (error == 0) { 1529 cache_setunresolved(ap->a_nch); 1530 cache_setvp(ap->a_nch, *ap->a_vpp); 1531 } 1532 LOCKSTOP; 1533 return error; 1534 } 1535 1536 /* 1537 * Make a device node (typically a fifo) 1538 */ 1539 static 1540 int 1541 hammer2_vop_nmknod(struct vop_nmknod_args *ap) 1542 { 1543 hammer2_inode_t *dip; 1544 hammer2_inode_t *nip; 1545 struct namecache *ncp; 1546 const uint8_t *name; 1547 size_t name_len; 1548 hammer2_tid_t inum; 1549 int error; 1550 1551 LOCKSTART; 1552 dip = VTOI(ap->a_dvp); 1553 if (dip->pmp->ronly) { 1554 LOCKSTOP; 1555 return (EROFS); 1556 } 1557 1558 ncp = ap->a_nch->ncp; 1559 name = ncp->nc_name; 1560 name_len = ncp->nc_nlen; 1561 hammer2_pfs_memory_wait(dip->pmp); 1562 hammer2_trans_init(dip->pmp, 0); 1563 1564 /* 1565 * The device node is entered as the directory entry itself and not 1566 * as a hardlink to an inode. Since one cannot obtain a 1567 * file handle on the filesystem entry representing the device, we 1568 * do not have to worry about indexing its inode. 1569 */ 1570 inum = hammer2_trans_newinum(dip->pmp); 1571 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred, 1572 NULL, 0, inum, 1573 inum, 0, 0, 1574 0, &error); 1575 if (error == 0) { 1576 hammer2_inode_create(dip, dip, NULL, NULL, 1577 name, name_len, 0, 1578 nip->meta.inum, 1579 HAMMER2_OBJTYPE_HARDLINK, nip->meta.type, 1580 0, &error); 1581 } 1582 1583 1584 if (error) { 1585 KKASSERT(nip == NULL); 1586 *ap->a_vpp = NULL; 1587 } else { 1588 *ap->a_vpp = hammer2_igetv(nip, &error); 1589 hammer2_inode_unlock(nip); 1590 } 1591 1592 /* 1593 * Update dip's mtime 1594 */ 1595 if (error == 0) { 1596 uint64_t mtime; 1597 1598 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1599 hammer2_update_time(&mtime); 1600 hammer2_inode_modify(dip); 1601 dip->meta.mtime = mtime; 1602 hammer2_inode_unlock(dip); 1603 } 1604 1605 hammer2_trans_done(dip->pmp); 1606 1607 if (error == 0) { 1608 cache_setunresolved(ap->a_nch); 1609 cache_setvp(ap->a_nch, *ap->a_vpp); 1610 } 1611 LOCKSTOP; 1612 return error; 1613 } 1614 1615 /* 1616 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target } 1617 */ 1618 static 1619 int 1620 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap) 1621 { 1622 hammer2_inode_t *dip; 1623 hammer2_inode_t *nip; 1624 struct namecache *ncp; 1625 const uint8_t *name; 1626 size_t name_len; 1627 hammer2_tid_t inum; 1628 int error; 1629 1630 dip = VTOI(ap->a_dvp); 1631 if (dip->pmp->ronly) 1632 return (EROFS); 1633 1634 ncp = ap->a_nch->ncp; 1635 name = ncp->nc_name; 1636 name_len = ncp->nc_nlen; 1637 hammer2_pfs_memory_wait(dip->pmp); 1638 hammer2_trans_init(dip->pmp, 0); 1639 1640 ap->a_vap->va_type = VLNK; /* enforce type */ 1641 1642 /* 1643 * The softlink is entered into the directory itself and not 1644 * as a hardlink to an inode. Since one cannot obtain a 1645 * file handle on the softlink itself we do not have to worry 1646 * about indexing its inode. 1647 */ 1648 inum = hammer2_trans_newinum(dip->pmp); 1649 1650 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred, 1651 NULL, 0, inum, 1652 inum, 0, 0, 1653 0, &error); 1654 if (error == 0) { 1655 hammer2_inode_create(dip, dip, NULL, NULL, 1656 name, name_len, 0, 1657 nip->meta.inum, 1658 HAMMER2_OBJTYPE_HARDLINK, nip->meta.type, 1659 0, &error); 1660 } 1661 1662 1663 if (error) { 1664 KKASSERT(nip == NULL); 1665 *ap->a_vpp = NULL; 1666 hammer2_trans_done(dip->pmp); 1667 return error; 1668 } 1669 *ap->a_vpp = hammer2_igetv(nip, &error); 1670 1671 /* 1672 * Build the softlink (~like file data) and finalize the namecache. 1673 */ 1674 if (error == 0) { 1675 size_t bytes; 1676 struct uio auio; 1677 struct iovec aiov; 1678 1679 bytes = strlen(ap->a_target); 1680 1681 hammer2_inode_unlock(nip); 1682 bzero(&auio, sizeof(auio)); 1683 bzero(&aiov, sizeof(aiov)); 1684 auio.uio_iov = &aiov; 1685 auio.uio_segflg = UIO_SYSSPACE; 1686 auio.uio_rw = UIO_WRITE; 1687 auio.uio_resid = bytes; 1688 auio.uio_iovcnt = 1; 1689 auio.uio_td = curthread; 1690 aiov.iov_base = ap->a_target; 1691 aiov.iov_len = bytes; 1692 error = hammer2_write_file(nip, &auio, IO_APPEND, 0); 1693 /* XXX handle error */ 1694 error = 0; 1695 } else { 1696 hammer2_inode_unlock(nip); 1697 } 1698 1699 /* 1700 * Update dip's mtime 1701 */ 1702 if (error == 0) { 1703 uint64_t mtime; 1704 1705 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1706 hammer2_update_time(&mtime); 1707 hammer2_inode_modify(dip); 1708 dip->meta.mtime = mtime; 1709 hammer2_inode_unlock(dip); 1710 } 1711 1712 hammer2_trans_done(dip->pmp); 1713 1714 /* 1715 * Finalize namecache 1716 */ 1717 if (error == 0) { 1718 cache_setunresolved(ap->a_nch); 1719 cache_setvp(ap->a_nch, *ap->a_vpp); 1720 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */ 1721 } 1722 return error; 1723 } 1724 1725 /* 1726 * hammer2_vop_nremove { nch, dvp, cred } 1727 */ 1728 static 1729 int 1730 hammer2_vop_nremove(struct vop_nremove_args *ap) 1731 { 1732 hammer2_xop_unlink_t *xop; 1733 hammer2_inode_t *dip; 1734 hammer2_inode_t *ip; 1735 struct namecache *ncp; 1736 int error; 1737 int isopen; 1738 1739 LOCKSTART; 1740 dip = VTOI(ap->a_dvp); 1741 if (dip->pmp->ronly) { 1742 LOCKSTOP; 1743 return(EROFS); 1744 } 1745 1746 ncp = ap->a_nch->ncp; 1747 1748 hammer2_pfs_memory_wait(dip->pmp); 1749 hammer2_trans_init(dip->pmp, 0); 1750 hammer2_inode_lock(dip, 0); 1751 1752 /* 1753 * The unlink XOP unlinks the path from the directory and 1754 * locates and returns the cluster associated with the real inode. 1755 * We have to handle nlinks here on the frontend. 1756 */ 1757 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING); 1758 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen); 1759 isopen = cache_isopen(ap->a_nch); 1760 xop->isdir = 0; 1761 xop->dopermanent = isopen ? 0 : HAMMER2_DELETE_PERMANENT; 1762 hammer2_xop_start(&xop->head, hammer2_xop_unlink); 1763 1764 /* 1765 * Collect the real inode and adjust nlinks, destroy the real 1766 * inode if nlinks transitions to 0 and it was the real inode 1767 * (else it has already been removed). 1768 */ 1769 error = hammer2_xop_collect(&xop->head, 0); 1770 hammer2_inode_unlock(dip); 1771 1772 if (error == 0) { 1773 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1); 1774 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 1775 if (ip) { 1776 hammer2_inode_unlink_finisher(ip, isopen); 1777 hammer2_inode_unlock(ip); 1778 } 1779 } else { 1780 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 1781 } 1782 1783 /* 1784 * Update dip's mtime 1785 */ 1786 if (error == 0) { 1787 uint64_t mtime; 1788 1789 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1790 hammer2_update_time(&mtime); 1791 hammer2_inode_modify(dip); 1792 dip->meta.mtime = mtime; 1793 hammer2_inode_unlock(dip); 1794 } 1795 1796 hammer2_inode_run_sideq(dip->pmp); 1797 hammer2_trans_done(dip->pmp); 1798 if (error == 0) 1799 cache_unlink(ap->a_nch); 1800 LOCKSTOP; 1801 return (error); 1802 } 1803 1804 /* 1805 * hammer2_vop_nrmdir { nch, dvp, cred } 1806 */ 1807 static 1808 int 1809 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap) 1810 { 1811 hammer2_xop_unlink_t *xop; 1812 hammer2_inode_t *dip; 1813 hammer2_inode_t *ip; 1814 struct namecache *ncp; 1815 int isopen; 1816 int error; 1817 1818 LOCKSTART; 1819 dip = VTOI(ap->a_dvp); 1820 if (dip->pmp->ronly) { 1821 LOCKSTOP; 1822 return(EROFS); 1823 } 1824 1825 hammer2_pfs_memory_wait(dip->pmp); 1826 hammer2_trans_init(dip->pmp, 0); 1827 hammer2_inode_lock(dip, 0); 1828 1829 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING); 1830 1831 ncp = ap->a_nch->ncp; 1832 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen); 1833 isopen = cache_isopen(ap->a_nch); 1834 xop->isdir = 1; 1835 xop->dopermanent = isopen ? 0 : HAMMER2_DELETE_PERMANENT; 1836 hammer2_xop_start(&xop->head, hammer2_xop_unlink); 1837 1838 /* 1839 * Collect the real inode and adjust nlinks, destroy the real 1840 * inode if nlinks transitions to 0 and it was the real inode 1841 * (else it has already been removed). 1842 */ 1843 error = hammer2_xop_collect(&xop->head, 0); 1844 hammer2_inode_unlock(dip); 1845 1846 if (error == 0) { 1847 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1); 1848 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 1849 if (ip) { 1850 hammer2_inode_unlink_finisher(ip, isopen); 1851 hammer2_inode_unlock(ip); 1852 } 1853 } else { 1854 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 1855 } 1856 1857 /* 1858 * Update dip's mtime 1859 */ 1860 if (error == 0) { 1861 uint64_t mtime; 1862 1863 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1864 hammer2_update_time(&mtime); 1865 hammer2_inode_modify(dip); 1866 dip->meta.mtime = mtime; 1867 hammer2_inode_unlock(dip); 1868 } 1869 1870 hammer2_inode_run_sideq(dip->pmp); 1871 hammer2_trans_done(dip->pmp); 1872 if (error == 0) 1873 cache_unlink(ap->a_nch); 1874 LOCKSTOP; 1875 return (error); 1876 } 1877 1878 /* 1879 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred } 1880 */ 1881 static 1882 int 1883 hammer2_vop_nrename(struct vop_nrename_args *ap) 1884 { 1885 struct namecache *fncp; 1886 struct namecache *tncp; 1887 hammer2_inode_t *fdip; 1888 hammer2_inode_t *tdip; 1889 hammer2_inode_t *ip; 1890 const uint8_t *fname; 1891 size_t fname_len; 1892 const uint8_t *tname; 1893 size_t tname_len; 1894 int error; 1895 int tnch_error; 1896 int update_tdip; 1897 int update_fdip; 1898 hammer2_key_t tlhc; 1899 1900 if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount) 1901 return(EXDEV); 1902 if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount) 1903 return(EXDEV); 1904 1905 fdip = VTOI(ap->a_fdvp); /* source directory */ 1906 tdip = VTOI(ap->a_tdvp); /* target directory */ 1907 1908 if (fdip->pmp->ronly) 1909 return(EROFS); 1910 1911 LOCKSTART; 1912 fncp = ap->a_fnch->ncp; /* entry name in source */ 1913 fname = fncp->nc_name; 1914 fname_len = fncp->nc_nlen; 1915 1916 tncp = ap->a_tnch->ncp; /* entry name in target */ 1917 tname = tncp->nc_name; 1918 tname_len = tncp->nc_nlen; 1919 1920 hammer2_pfs_memory_wait(tdip->pmp); 1921 hammer2_trans_init(tdip->pmp, 0); 1922 1923 update_tdip = 0; 1924 update_fdip = 0; 1925 1926 /* 1927 * ip is the inode being renamed. If this is a hardlink then 1928 * ip represents the actual file and not the hardlink marker. 1929 */ 1930 ip = VTOI(fncp->nc_vp); 1931 1932 KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0); 1933 1934 /* 1935 * Can return NULL and error == EXDEV if the common parent 1936 * crosses a directory with the xlink flag set. 1937 */ 1938 error = 0; 1939 hammer2_inode_lock(fdip, 0); 1940 hammer2_inode_lock(tdip, 0); 1941 hammer2_inode_ref(ip); /* extra ref */ 1942 1943 hammer2_inode_lock(ip, 0); 1944 1945 /* 1946 * Delete the target namespace. 1947 */ 1948 { 1949 hammer2_xop_unlink_t *xop2; 1950 hammer2_inode_t *tip; 1951 int isopen; 1952 1953 /* 1954 * The unlink XOP unlinks the path from the directory and 1955 * locates and returns the cluster associated with the real 1956 * inode. We have to handle nlinks here on the frontend. 1957 */ 1958 xop2 = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING); 1959 hammer2_xop_setname(&xop2->head, tname, tname_len); 1960 isopen = cache_isopen(ap->a_tnch); 1961 xop2->isdir = -1; 1962 xop2->dopermanent = isopen ? 0 : HAMMER2_DELETE_PERMANENT; 1963 hammer2_xop_start(&xop2->head, hammer2_xop_unlink); 1964 1965 /* 1966 * Collect the real inode and adjust nlinks, destroy the real 1967 * inode if nlinks transitions to 0 and it was the real inode 1968 * (else it has already been removed). 1969 */ 1970 tnch_error = hammer2_xop_collect(&xop2->head, 0); 1971 /* hammer2_inode_unlock(tdip); */ 1972 1973 if (tnch_error == 0) { 1974 tip = hammer2_inode_get(tdip->pmp, NULL, 1975 &xop2->head.cluster, -1); 1976 hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP); 1977 if (tip) { 1978 hammer2_inode_unlink_finisher(tip, isopen); 1979 hammer2_inode_unlock(tip); 1980 } 1981 } else { 1982 hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP); 1983 } 1984 /* hammer2_inode_lock(tdip, 0); */ 1985 1986 if (tnch_error && tnch_error != ENOENT) { 1987 error = tnch_error; 1988 goto done2; 1989 } 1990 update_tdip = 1; 1991 } 1992 1993 /* 1994 * Resolve the collision space for (tdip, tname, tname_len) 1995 * 1996 * tdip must be held exclusively locked to prevent races. 1997 */ 1998 { 1999 hammer2_xop_scanlhc_t *sxop; 2000 hammer2_tid_t lhcbase; 2001 2002 tlhc = hammer2_dirhash(tname, tname_len); 2003 lhcbase = tlhc; 2004 sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING); 2005 sxop->lhc = tlhc; 2006 hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc); 2007 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) { 2008 if (tlhc != sxop->head.cluster.focus->bref.key) 2009 break; 2010 ++tlhc; 2011 } 2012 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP); 2013 2014 if (error) { 2015 if (error != ENOENT) 2016 goto done2; 2017 ++tlhc; 2018 error = 0; 2019 } 2020 if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) { 2021 error = ENOSPC; 2022 goto done2; 2023 } 2024 } 2025 2026 /* 2027 * Everything is setup, do the rename. 2028 * 2029 * We have to synchronize ip->meta to the underlying operation. 2030 * 2031 * NOTE: To avoid deadlocks we cannot lock (ip) while we are 2032 * unlinking elements from their directories. Locking 2033 * the nlinks field does not lock the whole inode. 2034 */ 2035 /* hammer2_inode_lock(ip, 0); */ 2036 if (error == 0) { 2037 hammer2_xop_nrename_t *xop4; 2038 2039 xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING); 2040 xop4->lhc = tlhc; 2041 xop4->ip_key = ip->meta.name_key; 2042 hammer2_xop_setip2(&xop4->head, ip); 2043 hammer2_xop_setip3(&xop4->head, tdip); 2044 hammer2_xop_setname(&xop4->head, fname, fname_len); 2045 hammer2_xop_setname2(&xop4->head, tname, tname_len); 2046 hammer2_xop_start(&xop4->head, hammer2_xop_nrename); 2047 2048 error = hammer2_xop_collect(&xop4->head, 0); 2049 hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP); 2050 2051 if (error == ENOENT) 2052 error = 0; 2053 if (error == 0 && 2054 (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) { 2055 hammer2_inode_modify(ip); 2056 ip->meta.name_len = tname_len; 2057 ip->meta.name_key = tlhc; 2058 2059 } 2060 update_fdip = 1; 2061 update_fdip = 1; 2062 } 2063 2064 done2: 2065 /* 2066 * Update directory mtimes to represent the something changed. 2067 */ 2068 if (update_fdip || update_tdip) { 2069 uint64_t mtime; 2070 2071 hammer2_update_time(&mtime); 2072 if (update_fdip) { 2073 hammer2_inode_modify(fdip); 2074 fdip->meta.mtime = mtime; 2075 } 2076 if (update_tdip) { 2077 hammer2_inode_modify(tdip); 2078 tdip->meta.mtime = mtime; 2079 } 2080 } 2081 hammer2_inode_unlock(ip); 2082 hammer2_inode_unlock(tdip); 2083 hammer2_inode_unlock(fdip); 2084 hammer2_inode_drop(ip); 2085 hammer2_inode_run_sideq(fdip->pmp); 2086 2087 hammer2_trans_done(tdip->pmp); 2088 2089 /* 2090 * Issue the namecache update after unlocking all the internal 2091 * hammer structures, otherwise we might deadlock. 2092 */ 2093 if (tnch_error == 0) { 2094 cache_unlink(ap->a_tnch); 2095 cache_setunresolved(ap->a_tnch); 2096 } 2097 if (error == 0) 2098 cache_rename(ap->a_fnch, ap->a_tnch); 2099 2100 LOCKSTOP; 2101 return (error); 2102 } 2103 2104 /* 2105 * hammer2_vop_ioctl { vp, command, data, fflag, cred } 2106 */ 2107 static 2108 int 2109 hammer2_vop_ioctl(struct vop_ioctl_args *ap) 2110 { 2111 hammer2_inode_t *ip; 2112 int error; 2113 2114 LOCKSTART; 2115 ip = VTOI(ap->a_vp); 2116 2117 error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data, 2118 ap->a_fflag, ap->a_cred); 2119 LOCKSTOP; 2120 return (error); 2121 } 2122 2123 static 2124 int 2125 hammer2_vop_mountctl(struct vop_mountctl_args *ap) 2126 { 2127 struct mount *mp; 2128 hammer2_pfs_t *pmp; 2129 int rc; 2130 2131 LOCKSTART; 2132 switch (ap->a_op) { 2133 case (MOUNTCTL_SET_EXPORT): 2134 mp = ap->a_head.a_ops->head.vv_mount; 2135 pmp = MPTOPMP(mp); 2136 2137 if (ap->a_ctllen != sizeof(struct export_args)) 2138 rc = (EINVAL); 2139 else 2140 rc = vfs_export(mp, &pmp->export, 2141 (const struct export_args *)ap->a_ctl); 2142 break; 2143 default: 2144 rc = vop_stdmountctl(ap); 2145 break; 2146 } 2147 LOCKSTOP; 2148 return (rc); 2149 } 2150 2151 /* 2152 * KQFILTER 2153 */ 2154 static void filt_hammer2detach(struct knote *kn); 2155 static int filt_hammer2read(struct knote *kn, long hint); 2156 static int filt_hammer2write(struct knote *kn, long hint); 2157 static int filt_hammer2vnode(struct knote *kn, long hint); 2158 2159 static struct filterops hammer2read_filtops = 2160 { FILTEROP_ISFD | FILTEROP_MPSAFE, 2161 NULL, filt_hammer2detach, filt_hammer2read }; 2162 static struct filterops hammer2write_filtops = 2163 { FILTEROP_ISFD | FILTEROP_MPSAFE, 2164 NULL, filt_hammer2detach, filt_hammer2write }; 2165 static struct filterops hammer2vnode_filtops = 2166 { FILTEROP_ISFD | FILTEROP_MPSAFE, 2167 NULL, filt_hammer2detach, filt_hammer2vnode }; 2168 2169 static 2170 int 2171 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap) 2172 { 2173 struct vnode *vp = ap->a_vp; 2174 struct knote *kn = ap->a_kn; 2175 2176 switch (kn->kn_filter) { 2177 case EVFILT_READ: 2178 kn->kn_fop = &hammer2read_filtops; 2179 break; 2180 case EVFILT_WRITE: 2181 kn->kn_fop = &hammer2write_filtops; 2182 break; 2183 case EVFILT_VNODE: 2184 kn->kn_fop = &hammer2vnode_filtops; 2185 break; 2186 default: 2187 return (EOPNOTSUPP); 2188 } 2189 2190 kn->kn_hook = (caddr_t)vp; 2191 2192 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2193 2194 return(0); 2195 } 2196 2197 static void 2198 filt_hammer2detach(struct knote *kn) 2199 { 2200 struct vnode *vp = (void *)kn->kn_hook; 2201 2202 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2203 } 2204 2205 static int 2206 filt_hammer2read(struct knote *kn, long hint) 2207 { 2208 struct vnode *vp = (void *)kn->kn_hook; 2209 hammer2_inode_t *ip = VTOI(vp); 2210 off_t off; 2211 2212 if (hint == NOTE_REVOKE) { 2213 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2214 return(1); 2215 } 2216 off = ip->meta.size - kn->kn_fp->f_offset; 2217 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 2218 if (kn->kn_sfflags & NOTE_OLDAPI) 2219 return(1); 2220 return (kn->kn_data != 0); 2221 } 2222 2223 2224 static int 2225 filt_hammer2write(struct knote *kn, long hint) 2226 { 2227 if (hint == NOTE_REVOKE) 2228 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2229 kn->kn_data = 0; 2230 return (1); 2231 } 2232 2233 static int 2234 filt_hammer2vnode(struct knote *kn, long hint) 2235 { 2236 if (kn->kn_sfflags & hint) 2237 kn->kn_fflags |= hint; 2238 if (hint == NOTE_REVOKE) { 2239 kn->kn_flags |= (EV_EOF | EV_NODATA); 2240 return (1); 2241 } 2242 return (kn->kn_fflags != 0); 2243 } 2244 2245 /* 2246 * FIFO VOPS 2247 */ 2248 static 2249 int 2250 hammer2_vop_markatime(struct vop_markatime_args *ap) 2251 { 2252 hammer2_inode_t *ip; 2253 struct vnode *vp; 2254 2255 vp = ap->a_vp; 2256 ip = VTOI(vp); 2257 2258 if (ip->pmp->ronly) 2259 return(EROFS); 2260 return(0); 2261 } 2262 2263 static 2264 int 2265 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap) 2266 { 2267 int error; 2268 2269 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 2270 if (error) 2271 error = hammer2_vop_kqfilter(ap); 2272 return(error); 2273 } 2274 2275 /* 2276 * VOPS vector 2277 */ 2278 struct vop_ops hammer2_vnode_vops = { 2279 .vop_default = vop_defaultop, 2280 .vop_fsync = hammer2_vop_fsync, 2281 .vop_getpages = vop_stdgetpages, 2282 .vop_putpages = vop_stdputpages, 2283 .vop_access = hammer2_vop_access, 2284 .vop_advlock = hammer2_vop_advlock, 2285 .vop_close = hammer2_vop_close, 2286 .vop_nlink = hammer2_vop_nlink, 2287 .vop_ncreate = hammer2_vop_ncreate, 2288 .vop_nsymlink = hammer2_vop_nsymlink, 2289 .vop_nremove = hammer2_vop_nremove, 2290 .vop_nrmdir = hammer2_vop_nrmdir, 2291 .vop_nrename = hammer2_vop_nrename, 2292 .vop_getattr = hammer2_vop_getattr, 2293 .vop_setattr = hammer2_vop_setattr, 2294 .vop_readdir = hammer2_vop_readdir, 2295 .vop_readlink = hammer2_vop_readlink, 2296 .vop_getpages = vop_stdgetpages, 2297 .vop_putpages = vop_stdputpages, 2298 .vop_read = hammer2_vop_read, 2299 .vop_write = hammer2_vop_write, 2300 .vop_open = hammer2_vop_open, 2301 .vop_inactive = hammer2_vop_inactive, 2302 .vop_reclaim = hammer2_vop_reclaim, 2303 .vop_nresolve = hammer2_vop_nresolve, 2304 .vop_nlookupdotdot = hammer2_vop_nlookupdotdot, 2305 .vop_nmkdir = hammer2_vop_nmkdir, 2306 .vop_nmknod = hammer2_vop_nmknod, 2307 .vop_ioctl = hammer2_vop_ioctl, 2308 .vop_mountctl = hammer2_vop_mountctl, 2309 .vop_bmap = hammer2_vop_bmap, 2310 .vop_strategy = hammer2_vop_strategy, 2311 .vop_kqfilter = hammer2_vop_kqfilter 2312 }; 2313 2314 struct vop_ops hammer2_spec_vops = { 2315 .vop_default = vop_defaultop, 2316 .vop_fsync = hammer2_vop_fsync, 2317 .vop_read = vop_stdnoread, 2318 .vop_write = vop_stdnowrite, 2319 .vop_access = hammer2_vop_access, 2320 .vop_close = hammer2_vop_close, 2321 .vop_markatime = hammer2_vop_markatime, 2322 .vop_getattr = hammer2_vop_getattr, 2323 .vop_inactive = hammer2_vop_inactive, 2324 .vop_reclaim = hammer2_vop_reclaim, 2325 .vop_setattr = hammer2_vop_setattr 2326 }; 2327 2328 struct vop_ops hammer2_fifo_vops = { 2329 .vop_default = fifo_vnoperate, 2330 .vop_fsync = hammer2_vop_fsync, 2331 #if 0 2332 .vop_read = hammer2_vop_fiforead, 2333 .vop_write = hammer2_vop_fifowrite, 2334 #endif 2335 .vop_access = hammer2_vop_access, 2336 #if 0 2337 .vop_close = hammer2_vop_fifoclose, 2338 #endif 2339 .vop_markatime = hammer2_vop_markatime, 2340 .vop_getattr = hammer2_vop_getattr, 2341 .vop_inactive = hammer2_vop_inactive, 2342 .vop_reclaim = hammer2_vop_reclaim, 2343 .vop_setattr = hammer2_vop_setattr, 2344 .vop_kqfilter = hammer2_vop_fifokqfilter 2345 }; 2346 2347