1 /* 2 * Copyright (c) 2011-2018 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 vp = ap->a_vp; 90 ip = VTOI(vp); 91 92 /* 93 * Degenerate case 94 */ 95 if (ip == NULL) { 96 vrecycle(vp); 97 return (0); 98 } 99 100 /* 101 * Check for deleted inodes and recycle immediately on the last 102 * release. Be sure to destroy any left-over buffer cache buffers 103 * so we do not waste time trying to flush them. 104 * 105 * Note that deleting the file block chains under the inode chain 106 * would just be a waste of energy, so don't do it. 107 * 108 * WARNING: nvtruncbuf() can only be safely called without the inode 109 * lock held due to the way our write thread works. 110 */ 111 if (ip->flags & HAMMER2_INODE_ISUNLINKED) { 112 hammer2_key_t lbase; 113 int nblksize; 114 115 /* 116 * Detect updates to the embedded data which may be 117 * synchronized by the strategy code. Simply mark the 118 * inode modified so it gets picked up by our normal flush. 119 */ 120 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL); 121 nvtruncbuf(vp, 0, nblksize, 0, 0); 122 vrecycle(vp); 123 } 124 return (0); 125 } 126 127 /* 128 * Reclaim a vnode so that it can be reused; after the inode is 129 * disassociated, the filesystem must manage it alone. 130 */ 131 static 132 int 133 hammer2_vop_reclaim(struct vop_reclaim_args *ap) 134 { 135 hammer2_inode_t *ip; 136 hammer2_pfs_t *pmp; 137 struct vnode *vp; 138 139 vp = ap->a_vp; 140 ip = VTOI(vp); 141 if (ip == NULL) { 142 return(0); 143 } 144 pmp = ip->pmp; 145 146 /* 147 * The final close of a deleted file or directory marks it for 148 * destruction. The DELETED flag allows the flusher to shortcut 149 * any modified blocks still unflushed (that is, just ignore them). 150 * 151 * HAMMER2 usually does not try to optimize the freemap by returning 152 * deleted blocks to it as it does not usually know how many snapshots 153 * might be referencing portions of the file/dir. 154 */ 155 vp->v_data = NULL; 156 ip->vp = NULL; 157 158 /* 159 * NOTE! We do not attempt to flush chains here, flushing is 160 * really fragile and could also deadlock. 161 */ 162 vclrisdirty(vp); 163 164 /* 165 * A modified inode may require chain synchronization. This 166 * synchronization is usually handled by VOP_SYNC / VOP_FSYNC 167 * when vfsync() is called. However, that requires a vnode. 168 * 169 * When the vnode is disassociated we must keep track of any modified 170 * inode via the sideq so that it is properly flushed. We cannot 171 * safely synchronize the inode from inside the reclaim due to 172 * potentially deep locks held as-of when the reclaim occurs. 173 * Interactions and potential deadlocks abound. 174 */ 175 if ((ip->flags & (HAMMER2_INODE_ISUNLINKED | 176 HAMMER2_INODE_MODIFIED | 177 HAMMER2_INODE_RESIZED | 178 HAMMER2_INODE_DIRTYDATA)) && 179 (ip->flags & HAMMER2_INODE_ISDELETED) == 0) { 180 hammer2_inode_sideq_t *ipul; 181 182 ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO); 183 ipul->ip = ip; 184 185 hammer2_spin_ex(&pmp->list_spin); 186 if ((ip->flags & HAMMER2_INODE_ONSIDEQ) == 0) { 187 /* ref -> sideq */ 188 atomic_set_int(&ip->flags, HAMMER2_INODE_ONSIDEQ); 189 TAILQ_INSERT_TAIL(&pmp->sideq, ipul, entry); 190 ++pmp->sideq_count; 191 hammer2_spin_unex(&pmp->list_spin); 192 } else { 193 hammer2_spin_unex(&pmp->list_spin); 194 kfree(ipul, pmp->minode); 195 hammer2_inode_drop(ip); /* vp ref */ 196 } 197 /* retain ref from vp for ipul */ 198 } else { 199 hammer2_inode_drop(ip); /* vp ref */ 200 } 201 202 /* 203 * XXX handle background sync when ip dirty, kernel will no longer 204 * notify us regarding this inode because there is no longer a 205 * vnode attached to it. 206 */ 207 208 return (0); 209 } 210 211 /* 212 * Currently this function synchronizes the front-end inode state to the 213 * backend chain topology, then flushes the inode's chain and sub-topology 214 * to backend media. This function does not flush the root topology down to 215 * the inode. 216 */ 217 static 218 int 219 hammer2_vop_fsync(struct vop_fsync_args *ap) 220 { 221 hammer2_inode_t *ip; 222 struct vnode *vp; 223 int error1; 224 int error2; 225 226 vp = ap->a_vp; 227 ip = VTOI(vp); 228 error1 = 0; 229 230 hammer2_trans_init(ip->pmp, 0); 231 232 /* 233 * Flush dirty buffers in the file's logical buffer cache. 234 * It is best to wait for the strategy code to commit the 235 * buffers to the device's backing buffer cache before 236 * then trying to flush the inode. 237 * 238 * This should be quick, but certain inode modifications cached 239 * entirely in the hammer2_inode structure may not trigger a 240 * buffer read until the flush so the fsync can wind up also 241 * doing scattered reads. 242 */ 243 vfsync(vp, ap->a_waitfor, 1, NULL, NULL); 244 bio_track_wait(&vp->v_track_write, 0, 0); 245 246 /* 247 * Flush any inode changes 248 */ 249 hammer2_inode_lock(ip, 0); 250 if (ip->flags & (HAMMER2_INODE_RESIZED|HAMMER2_INODE_MODIFIED)) 251 error1 = hammer2_inode_chain_sync(ip); 252 253 /* 254 * Flush dirty chains related to the inode. 255 * 256 * NOTE! XXX We do not currently flush to the volume root, ultimately 257 * we will want to have a shortcut for the flushed inode stored 258 * in the volume root for recovery purposes. 259 */ 260 error2 = hammer2_inode_chain_flush(ip); 261 if (error2) 262 error1 = error2; 263 264 /* 265 * We may be able to clear the vnode dirty flag. The 266 * hammer2_pfs_moderate() code depends on this usually working. 267 */ 268 if ((ip->flags & (HAMMER2_INODE_MODIFIED | 269 HAMMER2_INODE_RESIZED | 270 HAMMER2_INODE_DIRTYDATA)) == 0 && 271 RB_EMPTY(&vp->v_rbdirty_tree) && 272 !bio_track_active(&vp->v_track_write)) { 273 vclrisdirty(vp); 274 } 275 hammer2_inode_unlock(ip); 276 hammer2_trans_done(ip->pmp); 277 278 return (error1); 279 } 280 281 static 282 int 283 hammer2_vop_access(struct vop_access_args *ap) 284 { 285 hammer2_inode_t *ip = VTOI(ap->a_vp); 286 uid_t uid; 287 gid_t gid; 288 int error; 289 290 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED); 291 uid = hammer2_to_unix_xid(&ip->meta.uid); 292 gid = hammer2_to_unix_xid(&ip->meta.gid); 293 error = vop_helper_access(ap, uid, gid, ip->meta.mode, ip->meta.uflags); 294 hammer2_inode_unlock(ip); 295 296 return (error); 297 } 298 299 static 300 int 301 hammer2_vop_getattr(struct vop_getattr_args *ap) 302 { 303 hammer2_pfs_t *pmp; 304 hammer2_inode_t *ip; 305 struct vnode *vp; 306 struct vattr *vap; 307 hammer2_chain_t *chain; 308 int i; 309 310 vp = ap->a_vp; 311 vap = ap->a_vap; 312 313 ip = VTOI(vp); 314 pmp = ip->pmp; 315 316 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED); 317 318 vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0]; 319 vap->va_fileid = ip->meta.inum; 320 vap->va_mode = ip->meta.mode; 321 vap->va_nlink = ip->meta.nlinks; 322 vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid); 323 vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid); 324 vap->va_rmajor = 0; 325 vap->va_rminor = 0; 326 vap->va_size = ip->meta.size; /* protected by shared lock */ 327 vap->va_blocksize = HAMMER2_PBUFSIZE; 328 vap->va_flags = ip->meta.uflags; 329 hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime); 330 hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime); 331 hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime); 332 vap->va_gen = 1; 333 vap->va_bytes = 0; 334 if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY) { 335 /* 336 * Can't really calculate directory use sans the files under 337 * it, just assume one block for now. 338 */ 339 vap->va_bytes += HAMMER2_INODE_BYTES; 340 } else { 341 for (i = 0; i < ip->cluster.nchains; ++i) { 342 if ((chain = ip->cluster.array[i].chain) != NULL) { 343 if (vap->va_bytes < 344 chain->bref.embed.stats.data_count) { 345 vap->va_bytes = 346 chain->bref.embed.stats.data_count; 347 } 348 } 349 } 350 } 351 vap->va_type = hammer2_get_vtype(ip->meta.type); 352 vap->va_filerev = 0; 353 vap->va_uid_uuid = ip->meta.uid; 354 vap->va_gid_uuid = ip->meta.gid; 355 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID | 356 VA_FSID_UUID_VALID; 357 358 hammer2_inode_unlock(ip); 359 360 return (0); 361 } 362 363 static 364 int 365 hammer2_vop_setattr(struct vop_setattr_args *ap) 366 { 367 hammer2_inode_t *ip; 368 struct vnode *vp; 369 struct vattr *vap; 370 int error; 371 int kflags = 0; 372 uint64_t ctime; 373 374 vp = ap->a_vp; 375 vap = ap->a_vap; 376 hammer2_update_time(&ctime); 377 378 ip = VTOI(vp); 379 380 if (ip->pmp->ronly) 381 return (EROFS); 382 if (hammer2_vfs_enospace(ip, 0, ap->a_cred) > 1) 383 return (ENOSPC); 384 385 hammer2_pfs_memory_wait(ip, 0); 386 hammer2_trans_init(ip->pmp, 0); 387 hammer2_inode_lock(ip, 0); 388 error = 0; 389 390 if (vap->va_flags != VNOVAL) { 391 uint32_t flags; 392 393 flags = ip->meta.uflags; 394 error = vop_helper_setattr_flags(&flags, vap->va_flags, 395 hammer2_to_unix_xid(&ip->meta.uid), 396 ap->a_cred); 397 if (error == 0) { 398 if (ip->meta.uflags != flags) { 399 hammer2_inode_modify(ip); 400 ip->meta.uflags = flags; 401 ip->meta.ctime = ctime; 402 kflags |= NOTE_ATTRIB; 403 } 404 if (ip->meta.uflags & (IMMUTABLE | APPEND)) { 405 error = 0; 406 goto done; 407 } 408 } 409 goto done; 410 } 411 if (ip->meta.uflags & (IMMUTABLE | APPEND)) { 412 error = EPERM; 413 goto done; 414 } 415 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { 416 mode_t cur_mode = ip->meta.mode; 417 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid); 418 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid); 419 uuid_t uuid_uid; 420 uuid_t uuid_gid; 421 422 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid, 423 ap->a_cred, 424 &cur_uid, &cur_gid, &cur_mode); 425 if (error == 0) { 426 hammer2_guid_to_uuid(&uuid_uid, cur_uid); 427 hammer2_guid_to_uuid(&uuid_gid, cur_gid); 428 if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) || 429 bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) || 430 ip->meta.mode != cur_mode 431 ) { 432 hammer2_inode_modify(ip); 433 ip->meta.uid = uuid_uid; 434 ip->meta.gid = uuid_gid; 435 ip->meta.mode = cur_mode; 436 ip->meta.ctime = ctime; 437 } 438 kflags |= NOTE_ATTRIB; 439 } 440 } 441 442 /* 443 * Resize the file 444 */ 445 if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) { 446 switch(vp->v_type) { 447 case VREG: 448 if (vap->va_size == ip->meta.size) 449 break; 450 if (vap->va_size < ip->meta.size) { 451 hammer2_mtx_ex(&ip->truncate_lock); 452 hammer2_truncate_file(ip, vap->va_size); 453 hammer2_mtx_unlock(&ip->truncate_lock); 454 kflags |= NOTE_WRITE; 455 } else { 456 hammer2_extend_file(ip, vap->va_size); 457 kflags |= NOTE_WRITE | NOTE_EXTEND; 458 } 459 hammer2_inode_modify(ip); 460 ip->meta.mtime = ctime; 461 vclrflags(vp, VLASTWRITETS); 462 break; 463 default: 464 error = EINVAL; 465 goto done; 466 } 467 } 468 #if 0 469 /* atime not supported */ 470 if (vap->va_atime.tv_sec != VNOVAL) { 471 hammer2_inode_modify(ip); 472 ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime); 473 kflags |= NOTE_ATTRIB; 474 } 475 #endif 476 if (vap->va_mode != (mode_t)VNOVAL) { 477 mode_t cur_mode = ip->meta.mode; 478 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid); 479 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid); 480 481 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred, 482 cur_uid, cur_gid, &cur_mode); 483 if (error == 0 && ip->meta.mode != cur_mode) { 484 hammer2_inode_modify(ip); 485 ip->meta.mode = cur_mode; 486 ip->meta.ctime = ctime; 487 kflags |= NOTE_ATTRIB; 488 } 489 } 490 491 if (vap->va_mtime.tv_sec != VNOVAL) { 492 hammer2_inode_modify(ip); 493 ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime); 494 kflags |= NOTE_ATTRIB; 495 vclrflags(vp, VLASTWRITETS); 496 } 497 498 done: 499 /* 500 * If a truncation occurred we must call chain_sync() now in order 501 * to trim the related data chains, otherwise a later expansion can 502 * cause havoc. 503 * 504 * If an extend occured that changed the DIRECTDATA state, we must 505 * call inode_fsync now in order to prepare the inode's indirect 506 * block table. 507 * 508 * WARNING! This means we are making an adjustment to the inode's 509 * chain outside of sync/fsync, and not just to inode->meta, which 510 * may result in some consistency issues if a crash were to occur 511 * at just the wrong time. 512 */ 513 if (ip->flags & HAMMER2_INODE_RESIZED) 514 hammer2_inode_chain_sync(ip); 515 516 /* 517 * Cleanup. 518 */ 519 hammer2_inode_unlock(ip); 520 hammer2_trans_done(ip->pmp); 521 hammer2_knote(ip->vp, kflags); 522 523 return (error); 524 } 525 526 static 527 int 528 hammer2_vop_readdir(struct vop_readdir_args *ap) 529 { 530 hammer2_xop_readdir_t *xop; 531 hammer2_blockref_t bref; 532 hammer2_inode_t *ip; 533 hammer2_tid_t inum; 534 hammer2_key_t lkey; 535 struct uio *uio; 536 off_t *cookies; 537 off_t saveoff; 538 int cookie_index; 539 int ncookies; 540 int error; 541 int eofflag; 542 int r; 543 544 ip = VTOI(ap->a_vp); 545 uio = ap->a_uio; 546 saveoff = uio->uio_offset; 547 eofflag = 0; 548 error = 0; 549 550 /* 551 * Setup cookies directory entry cookies if requested 552 */ 553 if (ap->a_ncookies) { 554 ncookies = uio->uio_resid / 16 + 1; 555 if (ncookies > 1024) 556 ncookies = 1024; 557 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK); 558 } else { 559 ncookies = -1; 560 cookies = NULL; 561 } 562 cookie_index = 0; 563 564 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED); 565 566 /* 567 * Handle artificial entries. To ensure that only positive 64 bit 568 * quantities are returned to userland we always strip off bit 63. 569 * The hash code is designed such that codes 0x0000-0x7FFF are not 570 * used, allowing us to use these codes for articial entries. 571 * 572 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not 573 * allow '..' to cross the mount point into (e.g.) the super-root. 574 */ 575 if (saveoff == 0) { 576 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK; 577 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, "."); 578 if (r) 579 goto done; 580 if (cookies) 581 cookies[cookie_index] = saveoff; 582 ++saveoff; 583 ++cookie_index; 584 if (cookie_index == ncookies) 585 goto done; 586 } 587 588 if (saveoff == 1) { 589 /* 590 * Be careful with lockorder when accessing ".." 591 * 592 * (ip is the current dir. xip is the parent dir). 593 */ 594 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK; 595 if (ip != ip->pmp->iroot) 596 inum = ip->meta.iparent & HAMMER2_DIRHASH_USERMSK; 597 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, ".."); 598 if (r) 599 goto done; 600 if (cookies) 601 cookies[cookie_index] = saveoff; 602 ++saveoff; 603 ++cookie_index; 604 if (cookie_index == ncookies) 605 goto done; 606 } 607 608 lkey = saveoff | HAMMER2_DIRHASH_VISIBLE; 609 if (hammer2_debug & 0x0020) 610 kprintf("readdir: lkey %016jx\n", lkey); 611 if (error) 612 goto done; 613 614 /* 615 * Use XOP for cluster scan. 616 * 617 * parent is the inode cluster, already locked for us. Don't 618 * double lock shared locks as this will screw up upgrades. 619 */ 620 xop = hammer2_xop_alloc(ip, 0); 621 xop->lkey = lkey; 622 hammer2_xop_start(&xop->head, hammer2_xop_readdir); 623 624 for (;;) { 625 const hammer2_inode_data_t *ripdata; 626 const char *dname; 627 int dtype; 628 629 error = hammer2_xop_collect(&xop->head, 0); 630 error = hammer2_error_to_errno(error); 631 if (error) { 632 break; 633 } 634 if (cookie_index == ncookies) 635 break; 636 if (hammer2_debug & 0x0020) 637 kprintf("cluster chain %p %p\n", 638 xop->head.cluster.focus, 639 (xop->head.cluster.focus ? 640 xop->head.cluster.focus->data : (void *)-1)); 641 hammer2_cluster_bref(&xop->head.cluster, &bref); 642 643 if (bref.type == HAMMER2_BREF_TYPE_INODE) { 644 ripdata = 645 &hammer2_cluster_rdata(&xop->head.cluster)->ipdata; 646 dtype = hammer2_get_dtype(ripdata->meta.type); 647 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK; 648 r = vop_write_dirent(&error, uio, 649 ripdata->meta.inum & 650 HAMMER2_DIRHASH_USERMSK, 651 dtype, 652 ripdata->meta.name_len, 653 ripdata->filename); 654 if (r) 655 break; 656 if (cookies) 657 cookies[cookie_index] = saveoff; 658 ++cookie_index; 659 } else if (bref.type == HAMMER2_BREF_TYPE_DIRENT) { 660 dtype = hammer2_get_dtype(bref.embed.dirent.type); 661 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK; 662 if (bref.embed.dirent.namlen <= 663 sizeof(bref.check.buf)) { 664 dname = bref.check.buf; 665 } else { 666 dname = 667 hammer2_cluster_rdata(&xop->head.cluster)->buf; 668 } 669 r = vop_write_dirent(&error, uio, 670 bref.embed.dirent.inum, 671 dtype, 672 bref.embed.dirent.namlen, 673 dname); 674 if (r) 675 break; 676 if (cookies) 677 cookies[cookie_index] = saveoff; 678 ++cookie_index; 679 } else { 680 /* XXX chain error */ 681 kprintf("bad chain type readdir %d\n", bref.type); 682 } 683 } 684 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 685 if (error == ENOENT) { 686 error = 0; 687 eofflag = 1; 688 saveoff = (hammer2_key_t)-1; 689 } else { 690 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK; 691 } 692 done: 693 hammer2_inode_unlock(ip); 694 if (ap->a_eofflag) 695 *ap->a_eofflag = eofflag; 696 if (hammer2_debug & 0x0020) 697 kprintf("readdir: done at %016jx\n", saveoff); 698 uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE; 699 if (error && cookie_index == 0) { 700 if (cookies) { 701 kfree(cookies, M_TEMP); 702 *ap->a_ncookies = 0; 703 *ap->a_cookies = NULL; 704 } 705 } else { 706 if (cookies) { 707 *ap->a_ncookies = cookie_index; 708 *ap->a_cookies = cookies; 709 } 710 } 711 return (error); 712 } 713 714 /* 715 * hammer2_vop_readlink { vp, uio, cred } 716 */ 717 static 718 int 719 hammer2_vop_readlink(struct vop_readlink_args *ap) 720 { 721 struct vnode *vp; 722 hammer2_inode_t *ip; 723 int error; 724 725 vp = ap->a_vp; 726 if (vp->v_type != VLNK) 727 return (EINVAL); 728 ip = VTOI(vp); 729 730 error = hammer2_read_file(ip, ap->a_uio, 0); 731 return (error); 732 } 733 734 static 735 int 736 hammer2_vop_read(struct vop_read_args *ap) 737 { 738 struct vnode *vp; 739 hammer2_inode_t *ip; 740 struct uio *uio; 741 int error; 742 int seqcount; 743 int bigread; 744 745 /* 746 * Read operations supported on this vnode? 747 */ 748 vp = ap->a_vp; 749 if (vp->v_type != VREG) 750 return (EINVAL); 751 752 /* 753 * Misc 754 */ 755 ip = VTOI(vp); 756 uio = ap->a_uio; 757 error = 0; 758 759 seqcount = ap->a_ioflag >> 16; 760 bigread = (uio->uio_resid > 100 * 1024 * 1024); 761 762 error = hammer2_read_file(ip, uio, seqcount); 763 return (error); 764 } 765 766 static 767 int 768 hammer2_vop_write(struct vop_write_args *ap) 769 { 770 hammer2_inode_t *ip; 771 thread_t td; 772 struct vnode *vp; 773 struct uio *uio; 774 int error; 775 int seqcount; 776 int ioflag; 777 778 /* 779 * Read operations supported on this vnode? 780 */ 781 vp = ap->a_vp; 782 if (vp->v_type != VREG) 783 return (EINVAL); 784 785 /* 786 * Misc 787 */ 788 ip = VTOI(vp); 789 ioflag = ap->a_ioflag; 790 uio = ap->a_uio; 791 error = 0; 792 if (ip->pmp->ronly) 793 return (EROFS); 794 switch (hammer2_vfs_enospace(ip, uio->uio_resid, ap->a_cred)) { 795 case 2: 796 return (ENOSPC); 797 case 1: 798 ioflag |= IO_DIRECT; /* semi-synchronous */ 799 /* fall through */ 800 default: 801 break; 802 } 803 804 seqcount = ioflag >> 16; 805 806 /* 807 * Check resource limit 808 */ 809 if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc && 810 uio->uio_offset + uio->uio_resid > 811 td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) { 812 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ); 813 return (EFBIG); 814 } 815 816 /* 817 * The transaction interlocks against flush initiations 818 * (note: but will run concurrently with the actual flush). 819 * 820 * To avoid deadlocking against the VM system, we must flag any 821 * transaction related to the buffer cache or other direct 822 * VM page manipulation. 823 */ 824 if (uio->uio_segflg == UIO_NOCOPY) { 825 hammer2_trans_init(ip->pmp, HAMMER2_TRANS_BUFCACHE); 826 } else { 827 hammer2_pfs_memory_wait(ip, 0); 828 hammer2_trans_init(ip->pmp, 0); 829 } 830 error = hammer2_write_file(ip, uio, ioflag, seqcount); 831 hammer2_trans_done(ip->pmp); 832 833 return (error); 834 } 835 836 /* 837 * Perform read operations on a file or symlink given an UNLOCKED 838 * inode and uio. 839 * 840 * The passed ip is not locked. 841 */ 842 static 843 int 844 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount) 845 { 846 hammer2_off_t size; 847 struct buf *bp; 848 int error; 849 850 error = 0; 851 852 /* 853 * UIO read loop. 854 * 855 * WARNING! Assumes that the kernel interlocks size changes at the 856 * vnode level. 857 */ 858 hammer2_mtx_sh(&ip->lock); 859 hammer2_mtx_sh(&ip->truncate_lock); 860 size = ip->meta.size; 861 hammer2_mtx_unlock(&ip->lock); 862 863 while (uio->uio_resid > 0 && uio->uio_offset < size) { 864 hammer2_key_t lbase; 865 hammer2_key_t leof; 866 int lblksize; 867 int loff; 868 int n; 869 870 lblksize = hammer2_calc_logical(ip, uio->uio_offset, 871 &lbase, &leof); 872 873 #if 1 874 bp = NULL; 875 error = cluster_readx(ip->vp, leof, lbase, lblksize, 876 B_NOTMETA | B_KVABIO, 877 uio->uio_resid, 878 seqcount * MAXBSIZE, 879 &bp); 880 #else 881 if (uio->uio_segflg == UIO_NOCOPY) { 882 bp = getblk(ip->vp, lbase, lblksize, 883 GETBLK_BHEAVY | GETBLK_KVABIO, 0); 884 if (bp->b_flags & B_CACHE) { 885 int i; 886 int j = 0; 887 if (bp->b_xio.xio_npages != 16) 888 kprintf("NPAGES BAD\n"); 889 for (i = 0; i < bp->b_xio.xio_npages; ++i) { 890 vm_page_t m; 891 m = bp->b_xio.xio_pages[i]; 892 if (m == NULL || m->valid == 0) { 893 kprintf("bp %016jx %016jx pg %d inv", 894 lbase, leof, i); 895 if (m) 896 kprintf("m->object %p/%p", m->object, ip->vp->v_object); 897 kprintf("\n"); 898 j = 1; 899 } 900 } 901 if (j) 902 kprintf("b_flags %08x, b_error %d\n", bp->b_flags, bp->b_error); 903 } 904 bqrelse(bp); 905 } 906 error = bread_kvabio(ip->vp, lbase, lblksize, &bp); 907 #endif 908 if (error) { 909 brelse(bp); 910 break; 911 } 912 bkvasync(bp); 913 loff = (int)(uio->uio_offset - lbase); 914 n = lblksize - loff; 915 if (n > uio->uio_resid) 916 n = uio->uio_resid; 917 if (n > size - uio->uio_offset) 918 n = (int)(size - uio->uio_offset); 919 bp->b_flags |= B_AGE; 920 uiomovebp(bp, (char *)bp->b_data + loff, n, uio); 921 bqrelse(bp); 922 } 923 hammer2_mtx_unlock(&ip->truncate_lock); 924 925 return (error); 926 } 927 928 /* 929 * Write to the file represented by the inode via the logical buffer cache. 930 * The inode may represent a regular file or a symlink. 931 * 932 * The inode must not be locked. 933 */ 934 static 935 int 936 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio, 937 int ioflag, int seqcount) 938 { 939 hammer2_key_t old_eof; 940 hammer2_key_t new_eof; 941 struct buf *bp; 942 int kflags; 943 int error; 944 int modified; 945 946 /* 947 * Setup if append 948 * 949 * WARNING! Assumes that the kernel interlocks size changes at the 950 * vnode level. 951 */ 952 hammer2_mtx_ex(&ip->lock); 953 hammer2_mtx_sh(&ip->truncate_lock); 954 if (ioflag & IO_APPEND) 955 uio->uio_offset = ip->meta.size; 956 old_eof = ip->meta.size; 957 958 /* 959 * Extend the file if necessary. If the write fails at some point 960 * we will truncate it back down to cover as much as we were able 961 * to write. 962 * 963 * Doing this now makes it easier to calculate buffer sizes in 964 * the loop. 965 */ 966 kflags = 0; 967 error = 0; 968 modified = 0; 969 970 if (uio->uio_offset + uio->uio_resid > old_eof) { 971 new_eof = uio->uio_offset + uio->uio_resid; 972 modified = 1; 973 hammer2_extend_file(ip, new_eof); 974 kflags |= NOTE_EXTEND; 975 } else { 976 new_eof = old_eof; 977 } 978 hammer2_mtx_unlock(&ip->lock); 979 980 /* 981 * UIO write loop 982 */ 983 while (uio->uio_resid > 0) { 984 hammer2_key_t lbase; 985 int trivial; 986 int endofblk; 987 int lblksize; 988 int loff; 989 int n; 990 991 /* 992 * Don't allow the buffer build to blow out the buffer 993 * cache. 994 */ 995 if ((ioflag & IO_RECURSE) == 0) 996 bwillwrite(HAMMER2_PBUFSIZE); 997 998 /* 999 * This nominally tells us how much we can cluster and 1000 * what the logical buffer size needs to be. Currently 1001 * we don't try to cluster the write and just handle one 1002 * block at a time. 1003 */ 1004 lblksize = hammer2_calc_logical(ip, uio->uio_offset, 1005 &lbase, NULL); 1006 loff = (int)(uio->uio_offset - lbase); 1007 1008 KKASSERT(lblksize <= 65536); 1009 1010 /* 1011 * Calculate bytes to copy this transfer and whether the 1012 * copy completely covers the buffer or not. 1013 */ 1014 trivial = 0; 1015 n = lblksize - loff; 1016 if (n > uio->uio_resid) { 1017 n = uio->uio_resid; 1018 if (loff == lbase && uio->uio_offset + n == new_eof) 1019 trivial = 1; 1020 endofblk = 0; 1021 } else { 1022 if (loff == 0) 1023 trivial = 1; 1024 endofblk = 1; 1025 } 1026 if (lbase >= new_eof) 1027 trivial = 1; 1028 1029 /* 1030 * Get the buffer 1031 */ 1032 if (uio->uio_segflg == UIO_NOCOPY) { 1033 /* 1034 * Issuing a write with the same data backing the 1035 * buffer. Instantiate the buffer to collect the 1036 * backing vm pages, then read-in any missing bits. 1037 * 1038 * This case is used by vop_stdputpages(). 1039 */ 1040 bp = getblk(ip->vp, lbase, lblksize, 1041 GETBLK_BHEAVY | GETBLK_KVABIO, 0); 1042 if ((bp->b_flags & B_CACHE) == 0) { 1043 bqrelse(bp); 1044 error = bread_kvabio(ip->vp, lbase, 1045 lblksize, &bp); 1046 } 1047 } else if (trivial) { 1048 /* 1049 * Even though we are entirely overwriting the buffer 1050 * we may still have to zero it out to avoid a 1051 * mmap/write visibility issue. 1052 */ 1053 bp = getblk(ip->vp, lbase, lblksize, 1054 GETBLK_BHEAVY | GETBLK_KVABIO, 0); 1055 if ((bp->b_flags & B_CACHE) == 0) 1056 vfs_bio_clrbuf(bp); 1057 } else { 1058 /* 1059 * Partial overwrite, read in any missing bits then 1060 * replace the portion being written. 1061 * 1062 * (The strategy code will detect zero-fill physical 1063 * blocks for this case). 1064 */ 1065 error = bread_kvabio(ip->vp, lbase, lblksize, &bp); 1066 if (error == 0) 1067 bheavy(bp); 1068 } 1069 1070 if (error) { 1071 brelse(bp); 1072 break; 1073 } 1074 1075 /* 1076 * Ok, copy the data in 1077 */ 1078 bkvasync(bp); 1079 error = uiomovebp(bp, bp->b_data + loff, n, uio); 1080 kflags |= NOTE_WRITE; 1081 modified = 1; 1082 if (error) { 1083 brelse(bp); 1084 break; 1085 } 1086 1087 /* 1088 * WARNING: Pageout daemon will issue UIO_NOCOPY writes 1089 * with IO_SYNC or IO_ASYNC set. These writes 1090 * must be handled as the pageout daemon expects. 1091 * 1092 * NOTE! H2 relies on cluster_write() here because it 1093 * cannot preallocate disk blocks at the logical 1094 * level due to not knowing what the compression 1095 * size will be at this time. 1096 * 1097 * We must use cluster_write() here and we depend 1098 * on the write-behind feature to flush buffers 1099 * appropriately. If we let the buffer daemons do 1100 * it the block allocations will be all over the 1101 * map. 1102 */ 1103 if (ioflag & IO_SYNC) { 1104 bwrite(bp); 1105 } else if ((ioflag & IO_DIRECT) && endofblk) { 1106 bawrite(bp); 1107 } else if (ioflag & IO_ASYNC) { 1108 bawrite(bp); 1109 } else if (ip->vp->v_mount->mnt_flag & MNT_NOCLUSTERW) { 1110 bdwrite(bp); 1111 } else { 1112 #if 1 1113 bp->b_flags |= B_CLUSTEROK; 1114 cluster_write(bp, new_eof, lblksize, seqcount); 1115 #else 1116 bp->b_flags |= B_CLUSTEROK; 1117 bdwrite(bp); 1118 #endif 1119 } 1120 } 1121 1122 /* 1123 * Cleanup. If we extended the file EOF but failed to write through 1124 * the entire write is a failure and we have to back-up. 1125 */ 1126 if (error && new_eof != old_eof) { 1127 hammer2_mtx_unlock(&ip->truncate_lock); 1128 hammer2_mtx_ex(&ip->lock); 1129 hammer2_mtx_ex(&ip->truncate_lock); 1130 hammer2_truncate_file(ip, old_eof); 1131 if (ip->flags & HAMMER2_INODE_MODIFIED) 1132 hammer2_inode_chain_sync(ip); 1133 hammer2_mtx_unlock(&ip->lock); 1134 } else if (modified) { 1135 struct vnode *vp = ip->vp; 1136 1137 hammer2_mtx_ex(&ip->lock); 1138 hammer2_inode_modify(ip); 1139 if (uio->uio_segflg == UIO_NOCOPY) { 1140 if (vp->v_flag & VLASTWRITETS) { 1141 ip->meta.mtime = 1142 (unsigned long)vp->v_lastwrite_ts.tv_sec * 1143 1000000 + 1144 vp->v_lastwrite_ts.tv_nsec / 1000; 1145 } 1146 } else { 1147 hammer2_update_time(&ip->meta.mtime); 1148 vclrflags(vp, VLASTWRITETS); 1149 } 1150 1151 #if 0 1152 /* 1153 * REMOVED - handled by hammer2_extend_file(). Do not issue 1154 * a chain_sync() outside of a sync/fsync except for DIRECTDATA 1155 * state changes. 1156 * 1157 * Under normal conditions we only issue a chain_sync if 1158 * the inode's DIRECTDATA state changed. 1159 */ 1160 if (ip->flags & HAMMER2_INODE_RESIZED) 1161 hammer2_inode_chain_sync(ip); 1162 #endif 1163 hammer2_mtx_unlock(&ip->lock); 1164 hammer2_knote(ip->vp, kflags); 1165 } 1166 hammer2_trans_assert_strategy(ip->pmp); 1167 hammer2_mtx_unlock(&ip->truncate_lock); 1168 1169 return error; 1170 } 1171 1172 /* 1173 * Truncate the size of a file. The inode must not be locked. 1174 * 1175 * We must unconditionally set HAMMER2_INODE_RESIZED to properly 1176 * ensure that any on-media data beyond the new file EOF has been destroyed. 1177 * 1178 * WARNING: nvtruncbuf() can only be safely called without the inode lock 1179 * held due to the way our write thread works. If the truncation 1180 * occurs in the middle of a buffer, nvtruncbuf() is responsible 1181 * for dirtying that buffer and zeroing out trailing bytes. 1182 * 1183 * WARNING! Assumes that the kernel interlocks size changes at the 1184 * vnode level. 1185 * 1186 * WARNING! Caller assumes responsibility for removing dead blocks 1187 * if INODE_RESIZED is set. 1188 */ 1189 static 1190 void 1191 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize) 1192 { 1193 hammer2_key_t lbase; 1194 int nblksize; 1195 1196 hammer2_mtx_unlock(&ip->lock); 1197 if (ip->vp) { 1198 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL); 1199 nvtruncbuf(ip->vp, nsize, 1200 nblksize, (int)nsize & (nblksize - 1), 1201 0); 1202 } 1203 hammer2_mtx_ex(&ip->lock); 1204 KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0); 1205 ip->osize = ip->meta.size; 1206 ip->meta.size = nsize; 1207 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED); 1208 hammer2_inode_modify(ip); 1209 } 1210 1211 /* 1212 * Extend the size of a file. The inode must not be locked. 1213 * 1214 * Even though the file size is changing, we do not have to set the 1215 * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES 1216 * boundary. When this occurs a hammer2_inode_chain_sync() is required 1217 * to prepare the inode cluster's indirect block table, otherwise 1218 * async execution of the strategy code will implode on us. 1219 * 1220 * WARNING! Assumes that the kernel interlocks size changes at the 1221 * vnode level. 1222 * 1223 * WARNING! Caller assumes responsibility for transitioning out 1224 * of the inode DIRECTDATA mode if INODE_RESIZED is set. 1225 */ 1226 static 1227 void 1228 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize) 1229 { 1230 hammer2_key_t lbase; 1231 hammer2_key_t osize; 1232 int oblksize; 1233 int nblksize; 1234 1235 KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0); 1236 hammer2_inode_modify(ip); 1237 osize = ip->meta.size; 1238 ip->osize = osize; 1239 ip->meta.size = nsize; 1240 1241 /* 1242 * We must issue a chain_sync() when the DIRECTDATA state changes 1243 * to prevent confusion between the flush code and the in-memory 1244 * state. This is not perfect because we are doing it outside of 1245 * a sync/fsync operation, so it might not be fully synchronized 1246 * with the meta-data topology flush. 1247 */ 1248 if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) { 1249 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED); 1250 hammer2_inode_chain_sync(ip); 1251 } 1252 1253 hammer2_mtx_unlock(&ip->lock); 1254 if (ip->vp) { 1255 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL); 1256 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL); 1257 nvextendbuf(ip->vp, 1258 osize, nsize, 1259 oblksize, nblksize, 1260 -1, -1, 0); 1261 } 1262 hammer2_mtx_ex(&ip->lock); 1263 } 1264 1265 static 1266 int 1267 hammer2_vop_nresolve(struct vop_nresolve_args *ap) 1268 { 1269 hammer2_xop_nresolve_t *xop; 1270 hammer2_inode_t *ip; 1271 hammer2_inode_t *dip; 1272 struct namecache *ncp; 1273 struct vnode *vp; 1274 int error; 1275 1276 dip = VTOI(ap->a_dvp); 1277 xop = hammer2_xop_alloc(dip, 0); 1278 1279 ncp = ap->a_nch->ncp; 1280 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen); 1281 1282 /* 1283 * Note: In DragonFly the kernel handles '.' and '..'. 1284 */ 1285 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1286 hammer2_xop_start(&xop->head, hammer2_xop_nresolve); 1287 1288 error = hammer2_xop_collect(&xop->head, 0); 1289 error = hammer2_error_to_errno(error); 1290 if (error) { 1291 ip = NULL; 1292 } else { 1293 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1); 1294 } 1295 hammer2_inode_unlock(dip); 1296 1297 /* 1298 * Acquire the related vnode 1299 * 1300 * NOTE: For error processing, only ENOENT resolves the namecache 1301 * entry to NULL, otherwise we just return the error and 1302 * leave the namecache unresolved. 1303 * 1304 * NOTE: multiple hammer2_inode structures can be aliased to the 1305 * same chain element, for example for hardlinks. This 1306 * use case does not 'reattach' inode associations that 1307 * might already exist, but always allocates a new one. 1308 * 1309 * WARNING: inode structure is locked exclusively via inode_get 1310 * but chain was locked shared. inode_unlock() 1311 * will handle it properly. 1312 */ 1313 if (ip) { 1314 vp = hammer2_igetv(ip, &error); /* error set to UNIX error */ 1315 if (error == 0) { 1316 vn_unlock(vp); 1317 cache_setvp(ap->a_nch, vp); 1318 } else if (error == ENOENT) { 1319 cache_setvp(ap->a_nch, NULL); 1320 } 1321 hammer2_inode_unlock(ip); 1322 1323 /* 1324 * The vp should not be released until after we've disposed 1325 * of our locks, because it might cause vop_inactive() to 1326 * be called. 1327 */ 1328 if (vp) 1329 vrele(vp); 1330 } else { 1331 error = ENOENT; 1332 cache_setvp(ap->a_nch, NULL); 1333 } 1334 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 1335 KASSERT(error || ap->a_nch->ncp->nc_vp != NULL, 1336 ("resolve error %d/%p ap %p\n", 1337 error, ap->a_nch->ncp->nc_vp, ap)); 1338 1339 return error; 1340 } 1341 1342 static 1343 int 1344 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap) 1345 { 1346 hammer2_inode_t *dip; 1347 hammer2_tid_t inum; 1348 int error; 1349 1350 dip = VTOI(ap->a_dvp); 1351 inum = dip->meta.iparent; 1352 *ap->a_vpp = NULL; 1353 1354 if (inum) { 1355 error = hammer2_vfs_vget(ap->a_dvp->v_mount, NULL, 1356 inum, ap->a_vpp); 1357 } else { 1358 error = ENOENT; 1359 } 1360 return error; 1361 } 1362 1363 static 1364 int 1365 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap) 1366 { 1367 hammer2_inode_t *dip; 1368 hammer2_inode_t *nip; 1369 struct namecache *ncp; 1370 const uint8_t *name; 1371 size_t name_len; 1372 hammer2_tid_t inum; 1373 int error; 1374 1375 dip = VTOI(ap->a_dvp); 1376 if (dip->pmp->ronly) 1377 return (EROFS); 1378 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1) 1379 return (ENOSPC); 1380 1381 ncp = ap->a_nch->ncp; 1382 name = ncp->nc_name; 1383 name_len = ncp->nc_nlen; 1384 1385 hammer2_pfs_memory_wait(dip, 1); 1386 hammer2_trans_init(dip->pmp, 0); 1387 1388 inum = hammer2_trans_newinum(dip->pmp); 1389 1390 /* 1391 * Create the actual inode as a hidden file in the iroot, then 1392 * create the directory entry. The creation of the actual inode 1393 * sets its nlinks to 1 which is the value we desire. 1394 */ 1395 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred, 1396 NULL, 0, inum, 1397 inum, 0, 0, 1398 0, &error); 1399 if (error) { 1400 error = hammer2_error_to_errno(error); 1401 } else { 1402 error = hammer2_dirent_create(dip, name, name_len, 1403 nip->meta.inum, nip->meta.type); 1404 /* returns UNIX error code */ 1405 } 1406 if (error) { 1407 if (nip) { 1408 hammer2_inode_unlink_finisher(nip, 0); 1409 hammer2_inode_unlock(nip); 1410 nip = NULL; 1411 } 1412 *ap->a_vpp = NULL; 1413 } else { 1414 *ap->a_vpp = hammer2_igetv(nip, &error); 1415 hammer2_inode_unlock(nip); 1416 } 1417 1418 /* 1419 * Update dip's mtime 1420 * 1421 * We can use a shared inode lock and allow the meta.mtime update 1422 * SMP race. hammer2_inode_modify() is MPSAFE w/a shared lock. 1423 */ 1424 if (error == 0) { 1425 uint64_t mtime; 1426 1427 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1428 hammer2_update_time(&mtime); 1429 hammer2_inode_modify(dip); 1430 dip->meta.mtime = mtime; 1431 hammer2_inode_unlock(dip); 1432 } 1433 1434 hammer2_trans_done(dip->pmp); 1435 1436 if (error == 0) { 1437 cache_setunresolved(ap->a_nch); 1438 cache_setvp(ap->a_nch, *ap->a_vpp); 1439 hammer2_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK); 1440 } 1441 return error; 1442 } 1443 1444 static 1445 int 1446 hammer2_vop_open(struct vop_open_args *ap) 1447 { 1448 return vop_stdopen(ap); 1449 } 1450 1451 /* 1452 * hammer2_vop_advlock { vp, id, op, fl, flags } 1453 */ 1454 static 1455 int 1456 hammer2_vop_advlock(struct vop_advlock_args *ap) 1457 { 1458 hammer2_inode_t *ip = VTOI(ap->a_vp); 1459 hammer2_off_t size; 1460 1461 size = ip->meta.size; 1462 return (lf_advlock(ap, &ip->advlock, size)); 1463 } 1464 1465 static 1466 int 1467 hammer2_vop_close(struct vop_close_args *ap) 1468 { 1469 return vop_stdclose(ap); 1470 } 1471 1472 /* 1473 * hammer2_vop_nlink { nch, dvp, vp, cred } 1474 * 1475 * Create a hardlink from (vp) to {dvp, nch}. 1476 */ 1477 static 1478 int 1479 hammer2_vop_nlink(struct vop_nlink_args *ap) 1480 { 1481 hammer2_inode_t *tdip; /* target directory to create link in */ 1482 hammer2_inode_t *ip; /* inode we are hardlinking to */ 1483 struct namecache *ncp; 1484 const uint8_t *name; 1485 size_t name_len; 1486 int error; 1487 1488 if (ap->a_dvp->v_mount != ap->a_vp->v_mount) 1489 return(EXDEV); 1490 1491 tdip = VTOI(ap->a_dvp); 1492 if (tdip->pmp->ronly) 1493 return (EROFS); 1494 if (hammer2_vfs_enospace(tdip, 0, ap->a_cred) > 1) 1495 return (ENOSPC); 1496 1497 ncp = ap->a_nch->ncp; 1498 name = ncp->nc_name; 1499 name_len = ncp->nc_nlen; 1500 1501 /* 1502 * ip represents the file being hardlinked. The file could be a 1503 * normal file or a hardlink target if it has already been hardlinked. 1504 * (with the new semantics, it will almost always be a hardlink 1505 * target). 1506 * 1507 * Bump nlinks and potentially also create or move the hardlink 1508 * target in the parent directory common to (ip) and (tdip). The 1509 * consolidation code can modify ip->cluster. The returned cluster 1510 * is locked. 1511 */ 1512 ip = VTOI(ap->a_vp); 1513 KASSERT(ip->pmp, ("ip->pmp is NULL %p %p", ip, ip->pmp)); 1514 hammer2_pfs_memory_wait(ip, 0); 1515 hammer2_trans_init(ip->pmp, 0); 1516 1517 /* 1518 * Target should be an indexed inode or there's no way we will ever 1519 * be able to find it! 1520 */ 1521 KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0); 1522 1523 error = 0; 1524 1525 /* 1526 * Can return NULL and error == EXDEV if the common parent 1527 * crosses a directory with the xlink flag set. 1528 */ 1529 hammer2_inode_lock(tdip, 0); 1530 hammer2_inode_lock(ip, 0); 1531 1532 /* 1533 * Create the directory entry and bump nlinks. 1534 */ 1535 if (error == 0) { 1536 error = hammer2_dirent_create(tdip, name, name_len, 1537 ip->meta.inum, ip->meta.type); 1538 hammer2_inode_modify(ip); 1539 ++ip->meta.nlinks; 1540 } 1541 if (error == 0) { 1542 /* 1543 * Update dip's mtime 1544 */ 1545 uint64_t mtime; 1546 1547 hammer2_update_time(&mtime); 1548 hammer2_inode_modify(tdip); 1549 tdip->meta.mtime = mtime; 1550 1551 cache_setunresolved(ap->a_nch); 1552 cache_setvp(ap->a_nch, ap->a_vp); 1553 } 1554 hammer2_inode_unlock(ip); 1555 hammer2_inode_unlock(tdip); 1556 1557 hammer2_trans_done(ip->pmp); 1558 hammer2_knote(ap->a_vp, NOTE_LINK); 1559 hammer2_knote(ap->a_dvp, NOTE_WRITE); 1560 1561 return error; 1562 } 1563 1564 /* 1565 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap } 1566 * 1567 * The operating system has already ensured that the directory entry 1568 * does not exist and done all appropriate namespace locking. 1569 */ 1570 static 1571 int 1572 hammer2_vop_ncreate(struct vop_ncreate_args *ap) 1573 { 1574 hammer2_inode_t *dip; 1575 hammer2_inode_t *nip; 1576 struct namecache *ncp; 1577 const uint8_t *name; 1578 size_t name_len; 1579 hammer2_tid_t inum; 1580 int error; 1581 1582 dip = VTOI(ap->a_dvp); 1583 if (dip->pmp->ronly) 1584 return (EROFS); 1585 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1) 1586 return (ENOSPC); 1587 1588 ncp = ap->a_nch->ncp; 1589 name = ncp->nc_name; 1590 name_len = ncp->nc_nlen; 1591 hammer2_pfs_memory_wait(dip, 1); 1592 hammer2_trans_init(dip->pmp, 0); 1593 1594 inum = hammer2_trans_newinum(dip->pmp); 1595 1596 /* 1597 * Create the actual inode as a hidden file in the iroot, then 1598 * create the directory entry. The creation of the actual inode 1599 * sets its nlinks to 1 which is the value we desire. 1600 */ 1601 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred, 1602 NULL, 0, inum, 1603 inum, 0, 0, 1604 0, &error); 1605 1606 if (error) { 1607 error = hammer2_error_to_errno(error); 1608 } else { 1609 error = hammer2_dirent_create(dip, name, name_len, 1610 nip->meta.inum, nip->meta.type); 1611 } 1612 if (error) { 1613 if (nip) { 1614 hammer2_inode_unlink_finisher(nip, 0); 1615 hammer2_inode_unlock(nip); 1616 nip = NULL; 1617 } 1618 *ap->a_vpp = NULL; 1619 } else { 1620 *ap->a_vpp = hammer2_igetv(nip, &error); 1621 hammer2_inode_unlock(nip); 1622 } 1623 1624 /* 1625 * Update dip's mtime 1626 */ 1627 if (error == 0) { 1628 uint64_t mtime; 1629 1630 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1631 hammer2_update_time(&mtime); 1632 hammer2_inode_modify(dip); 1633 dip->meta.mtime = mtime; 1634 hammer2_inode_unlock(dip); 1635 } 1636 1637 hammer2_trans_done(dip->pmp); 1638 1639 if (error == 0) { 1640 cache_setunresolved(ap->a_nch); 1641 cache_setvp(ap->a_nch, *ap->a_vpp); 1642 hammer2_knote(ap->a_dvp, NOTE_WRITE); 1643 } 1644 return error; 1645 } 1646 1647 /* 1648 * Make a device node (typically a fifo) 1649 */ 1650 static 1651 int 1652 hammer2_vop_nmknod(struct vop_nmknod_args *ap) 1653 { 1654 hammer2_inode_t *dip; 1655 hammer2_inode_t *nip; 1656 struct namecache *ncp; 1657 const uint8_t *name; 1658 size_t name_len; 1659 hammer2_tid_t inum; 1660 int error; 1661 1662 dip = VTOI(ap->a_dvp); 1663 if (dip->pmp->ronly) 1664 return (EROFS); 1665 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1) 1666 return (ENOSPC); 1667 1668 ncp = ap->a_nch->ncp; 1669 name = ncp->nc_name; 1670 name_len = ncp->nc_nlen; 1671 hammer2_pfs_memory_wait(dip, 1); 1672 hammer2_trans_init(dip->pmp, 0); 1673 1674 /* 1675 * Create the device inode and then create the directory entry. 1676 */ 1677 inum = hammer2_trans_newinum(dip->pmp); 1678 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred, 1679 NULL, 0, inum, 1680 inum, 0, 0, 1681 0, &error); 1682 if (error == 0) { 1683 error = hammer2_dirent_create(dip, name, name_len, 1684 nip->meta.inum, nip->meta.type); 1685 } 1686 if (error) { 1687 if (nip) { 1688 hammer2_inode_unlink_finisher(nip, 0); 1689 hammer2_inode_unlock(nip); 1690 nip = NULL; 1691 } 1692 *ap->a_vpp = NULL; 1693 } else { 1694 *ap->a_vpp = hammer2_igetv(nip, &error); 1695 hammer2_inode_unlock(nip); 1696 } 1697 1698 /* 1699 * Update dip's mtime 1700 */ 1701 if (error == 0) { 1702 uint64_t mtime; 1703 1704 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1705 hammer2_update_time(&mtime); 1706 hammer2_inode_modify(dip); 1707 dip->meta.mtime = mtime; 1708 hammer2_inode_unlock(dip); 1709 } 1710 1711 hammer2_trans_done(dip->pmp); 1712 1713 if (error == 0) { 1714 cache_setunresolved(ap->a_nch); 1715 cache_setvp(ap->a_nch, *ap->a_vpp); 1716 hammer2_knote(ap->a_dvp, NOTE_WRITE); 1717 } 1718 return error; 1719 } 1720 1721 /* 1722 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target } 1723 */ 1724 static 1725 int 1726 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap) 1727 { 1728 hammer2_inode_t *dip; 1729 hammer2_inode_t *nip; 1730 struct namecache *ncp; 1731 const uint8_t *name; 1732 size_t name_len; 1733 hammer2_tid_t inum; 1734 int error; 1735 1736 dip = VTOI(ap->a_dvp); 1737 if (dip->pmp->ronly) 1738 return (EROFS); 1739 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1) 1740 return (ENOSPC); 1741 1742 ncp = ap->a_nch->ncp; 1743 name = ncp->nc_name; 1744 name_len = ncp->nc_nlen; 1745 hammer2_pfs_memory_wait(dip, 1); 1746 hammer2_trans_init(dip->pmp, 0); 1747 1748 ap->a_vap->va_type = VLNK; /* enforce type */ 1749 1750 /* 1751 * Create the softlink as an inode and then create the directory 1752 * entry. 1753 */ 1754 inum = hammer2_trans_newinum(dip->pmp); 1755 1756 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred, 1757 NULL, 0, inum, 1758 inum, 0, 0, 1759 0, &error); 1760 if (error == 0) { 1761 error = hammer2_dirent_create(dip, name, name_len, 1762 nip->meta.inum, nip->meta.type); 1763 } 1764 if (error) { 1765 if (nip) { 1766 hammer2_inode_unlink_finisher(nip, 0); 1767 hammer2_inode_unlock(nip); 1768 nip = NULL; 1769 } 1770 *ap->a_vpp = NULL; 1771 hammer2_trans_done(dip->pmp); 1772 return error; 1773 } 1774 *ap->a_vpp = hammer2_igetv(nip, &error); 1775 1776 /* 1777 * Build the softlink (~like file data) and finalize the namecache. 1778 */ 1779 if (error == 0) { 1780 size_t bytes; 1781 struct uio auio; 1782 struct iovec aiov; 1783 1784 bytes = strlen(ap->a_target); 1785 1786 hammer2_inode_unlock(nip); 1787 bzero(&auio, sizeof(auio)); 1788 bzero(&aiov, sizeof(aiov)); 1789 auio.uio_iov = &aiov; 1790 auio.uio_segflg = UIO_SYSSPACE; 1791 auio.uio_rw = UIO_WRITE; 1792 auio.uio_resid = bytes; 1793 auio.uio_iovcnt = 1; 1794 auio.uio_td = curthread; 1795 aiov.iov_base = ap->a_target; 1796 aiov.iov_len = bytes; 1797 error = hammer2_write_file(nip, &auio, IO_APPEND, 0); 1798 /* XXX handle error */ 1799 error = 0; 1800 } else { 1801 hammer2_inode_unlock(nip); 1802 } 1803 1804 /* 1805 * Update dip's mtime 1806 */ 1807 if (error == 0) { 1808 uint64_t mtime; 1809 1810 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1811 hammer2_update_time(&mtime); 1812 hammer2_inode_modify(dip); 1813 dip->meta.mtime = mtime; 1814 hammer2_inode_unlock(dip); 1815 } 1816 1817 hammer2_trans_done(dip->pmp); 1818 1819 /* 1820 * Finalize namecache 1821 */ 1822 if (error == 0) { 1823 cache_setunresolved(ap->a_nch); 1824 cache_setvp(ap->a_nch, *ap->a_vpp); 1825 hammer2_knote(ap->a_dvp, NOTE_WRITE); 1826 } 1827 return error; 1828 } 1829 1830 /* 1831 * hammer2_vop_nremove { nch, dvp, cred } 1832 */ 1833 static 1834 int 1835 hammer2_vop_nremove(struct vop_nremove_args *ap) 1836 { 1837 hammer2_xop_unlink_t *xop; 1838 hammer2_inode_t *dip; 1839 hammer2_inode_t *ip; 1840 struct namecache *ncp; 1841 int error; 1842 int isopen; 1843 1844 dip = VTOI(ap->a_dvp); 1845 if (dip->pmp->ronly) 1846 return (EROFS); 1847 #if 0 1848 /* allow removals, except user to also bulkfree */ 1849 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1) 1850 return (ENOSPC); 1851 #endif 1852 1853 ncp = ap->a_nch->ncp; 1854 1855 hammer2_pfs_memory_wait(dip, 1); 1856 hammer2_trans_init(dip->pmp, 0); 1857 hammer2_inode_lock(dip, 0); 1858 1859 /* 1860 * The unlink XOP unlinks the path from the directory and 1861 * locates and returns the cluster associated with the real inode. 1862 * We have to handle nlinks here on the frontend. 1863 */ 1864 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING); 1865 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen); 1866 1867 /* 1868 * The namecache entry is locked so nobody can use this namespace. 1869 * Calculate isopen to determine if this namespace has an open vp 1870 * associated with it and resolve the vp only if it does. 1871 * 1872 * We try to avoid resolving the vnode if nobody has it open, but 1873 * note that the test is via this namespace only. 1874 */ 1875 isopen = cache_isopen(ap->a_nch); 1876 xop->isdir = 0; 1877 xop->dopermanent = 0; 1878 hammer2_xop_start(&xop->head, hammer2_xop_unlink); 1879 1880 /* 1881 * Collect the real inode and adjust nlinks, destroy the real 1882 * inode if nlinks transitions to 0 and it was the real inode 1883 * (else it has already been removed). 1884 */ 1885 error = hammer2_xop_collect(&xop->head, 0); 1886 error = hammer2_error_to_errno(error); 1887 hammer2_inode_unlock(dip); 1888 1889 if (error == 0) { 1890 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1); 1891 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 1892 if (ip) { 1893 hammer2_inode_unlink_finisher(ip, isopen); 1894 hammer2_inode_unlock(ip); 1895 } 1896 } else { 1897 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 1898 } 1899 1900 /* 1901 * Update dip's mtime 1902 */ 1903 if (error == 0) { 1904 uint64_t mtime; 1905 1906 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1907 hammer2_update_time(&mtime); 1908 hammer2_inode_modify(dip); 1909 dip->meta.mtime = mtime; 1910 hammer2_inode_unlock(dip); 1911 } 1912 1913 hammer2_inode_run_sideq(dip->pmp, 0); 1914 hammer2_trans_done(dip->pmp); 1915 if (error == 0) { 1916 cache_unlink(ap->a_nch); 1917 hammer2_knote(ap->a_dvp, NOTE_WRITE); 1918 } 1919 return (error); 1920 } 1921 1922 /* 1923 * hammer2_vop_nrmdir { nch, dvp, cred } 1924 */ 1925 static 1926 int 1927 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap) 1928 { 1929 hammer2_xop_unlink_t *xop; 1930 hammer2_inode_t *dip; 1931 hammer2_inode_t *ip; 1932 struct namecache *ncp; 1933 int isopen; 1934 int error; 1935 1936 dip = VTOI(ap->a_dvp); 1937 if (dip->pmp->ronly) 1938 return (EROFS); 1939 #if 0 1940 /* allow removals, except user to also bulkfree */ 1941 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1) 1942 return (ENOSPC); 1943 #endif 1944 1945 hammer2_pfs_memory_wait(dip, 1); 1946 hammer2_trans_init(dip->pmp, 0); 1947 hammer2_inode_lock(dip, 0); 1948 1949 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING); 1950 1951 ncp = ap->a_nch->ncp; 1952 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen); 1953 isopen = cache_isopen(ap->a_nch); 1954 xop->isdir = 1; 1955 xop->dopermanent = 0; 1956 hammer2_xop_start(&xop->head, hammer2_xop_unlink); 1957 1958 /* 1959 * Collect the real inode and adjust nlinks, destroy the real 1960 * inode if nlinks transitions to 0 and it was the real inode 1961 * (else it has already been removed). 1962 */ 1963 error = hammer2_xop_collect(&xop->head, 0); 1964 error = hammer2_error_to_errno(error); 1965 hammer2_inode_unlock(dip); 1966 1967 if (error == 0) { 1968 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1); 1969 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 1970 if (ip) { 1971 hammer2_inode_unlink_finisher(ip, isopen); 1972 hammer2_inode_unlock(ip); 1973 } 1974 } else { 1975 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP); 1976 } 1977 1978 /* 1979 * Update dip's mtime 1980 */ 1981 if (error == 0) { 1982 uint64_t mtime; 1983 1984 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED); 1985 hammer2_update_time(&mtime); 1986 hammer2_inode_modify(dip); 1987 dip->meta.mtime = mtime; 1988 hammer2_inode_unlock(dip); 1989 } 1990 1991 hammer2_inode_run_sideq(dip->pmp, 0); 1992 hammer2_trans_done(dip->pmp); 1993 if (error == 0) { 1994 cache_unlink(ap->a_nch); 1995 hammer2_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK); 1996 } 1997 return (error); 1998 } 1999 2000 /* 2001 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred } 2002 */ 2003 static 2004 int 2005 hammer2_vop_nrename(struct vop_nrename_args *ap) 2006 { 2007 struct namecache *fncp; 2008 struct namecache *tncp; 2009 hammer2_inode_t *fdip; /* source directory */ 2010 hammer2_inode_t *tdip; /* target directory */ 2011 hammer2_inode_t *ip; /* file being renamed */ 2012 hammer2_inode_t *tip; /* replaced target during rename or NULL */ 2013 const uint8_t *fname; 2014 size_t fname_len; 2015 const uint8_t *tname; 2016 size_t tname_len; 2017 int error; 2018 int update_tdip; 2019 int update_fdip; 2020 hammer2_key_t tlhc; 2021 2022 if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount) 2023 return(EXDEV); 2024 if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount) 2025 return(EXDEV); 2026 2027 fdip = VTOI(ap->a_fdvp); /* source directory */ 2028 tdip = VTOI(ap->a_tdvp); /* target directory */ 2029 2030 if (fdip->pmp->ronly) 2031 return (EROFS); 2032 if (hammer2_vfs_enospace(fdip, 0, ap->a_cred) > 1) 2033 return (ENOSPC); 2034 2035 fncp = ap->a_fnch->ncp; /* entry name in source */ 2036 fname = fncp->nc_name; 2037 fname_len = fncp->nc_nlen; 2038 2039 tncp = ap->a_tnch->ncp; /* entry name in target */ 2040 tname = tncp->nc_name; 2041 tname_len = tncp->nc_nlen; 2042 2043 hammer2_pfs_memory_wait(tdip, 0); 2044 hammer2_trans_init(tdip->pmp, 0); 2045 2046 update_tdip = 0; 2047 update_fdip = 0; 2048 2049 ip = VTOI(fncp->nc_vp); 2050 hammer2_inode_ref(ip); /* extra ref */ 2051 2052 /* 2053 * Lookup the target name to determine if a directory entry 2054 * is being overwritten. We only hold related inode locks 2055 * temporarily, the operating system is expected to protect 2056 * against rename races. 2057 */ 2058 tip = tncp->nc_vp ? VTOI(tncp->nc_vp) : NULL; 2059 if (tip) 2060 hammer2_inode_ref(tip); /* extra ref */ 2061 2062 /* 2063 * Can return NULL and error == EXDEV if the common parent 2064 * crosses a directory with the xlink flag set. 2065 * 2066 * For now try to avoid deadlocks with a simple pointer address 2067 * test. (tip) can be NULL. 2068 */ 2069 error = 0; 2070 if (fdip <= tdip) { 2071 hammer2_inode_lock(fdip, 0); 2072 hammer2_inode_lock(tdip, 0); 2073 } else { 2074 hammer2_inode_lock(tdip, 0); 2075 hammer2_inode_lock(fdip, 0); 2076 } 2077 if (tip) { 2078 if (ip <= tip) { 2079 hammer2_inode_lock(ip, 0); 2080 hammer2_inode_lock(tip, 0); 2081 } else { 2082 hammer2_inode_lock(tip, 0); 2083 hammer2_inode_lock(ip, 0); 2084 } 2085 } else { 2086 hammer2_inode_lock(ip, 0); 2087 } 2088 2089 #if 0 2090 /* 2091 * Delete the target namespace. 2092 * 2093 * REMOVED - NOW FOLDED INTO XOP_NRENAME OPERATION 2094 */ 2095 { 2096 hammer2_xop_unlink_t *xop2; 2097 hammer2_inode_t *tip; 2098 int isopen; 2099 2100 /* 2101 * The unlink XOP unlinks the path from the directory and 2102 * locates and returns the cluster associated with the real 2103 * inode. We have to handle nlinks here on the frontend. 2104 */ 2105 xop2 = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING); 2106 hammer2_xop_setname(&xop2->head, tname, tname_len); 2107 isopen = cache_isopen(ap->a_tnch); 2108 xop2->isdir = -1; 2109 xop2->dopermanent = 0; 2110 hammer2_xop_start(&xop2->head, hammer2_xop_unlink); 2111 2112 /* 2113 * Collect the real inode and adjust nlinks, destroy the real 2114 * inode if nlinks transitions to 0 and it was the real inode 2115 * (else it has already been removed). 2116 */ 2117 tnch_error = hammer2_xop_collect(&xop2->head, 0); 2118 tnch_error = hammer2_error_to_errno(tnch_error); 2119 /* hammer2_inode_unlock(tdip); */ 2120 2121 if (tnch_error == 0) { 2122 tip = hammer2_inode_get(tdip->pmp, NULL, 2123 &xop2->head.cluster, -1); 2124 hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP); 2125 if (tip) { 2126 hammer2_inode_unlink_finisher(tip, isopen); 2127 hammer2_inode_unlock(tip); 2128 } 2129 } else { 2130 hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP); 2131 } 2132 /* hammer2_inode_lock(tdip, 0); */ 2133 2134 if (tnch_error && tnch_error != ENOENT) { 2135 error = tnch_error; 2136 goto done2; 2137 } 2138 update_tdip = 1; 2139 } 2140 #endif 2141 2142 /* 2143 * Resolve the collision space for (tdip, tname, tname_len) 2144 * 2145 * tdip must be held exclusively locked to prevent races since 2146 * multiple filenames can end up in the same collision space. 2147 */ 2148 { 2149 hammer2_xop_scanlhc_t *sxop; 2150 hammer2_tid_t lhcbase; 2151 2152 tlhc = hammer2_dirhash(tname, tname_len); 2153 lhcbase = tlhc; 2154 sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING); 2155 sxop->lhc = tlhc; 2156 hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc); 2157 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) { 2158 if (tlhc != sxop->head.cluster.focus->bref.key) 2159 break; 2160 ++tlhc; 2161 } 2162 error = hammer2_error_to_errno(error); 2163 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP); 2164 2165 if (error) { 2166 if (error != ENOENT) 2167 goto done2; 2168 ++tlhc; 2169 error = 0; 2170 } 2171 if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) { 2172 error = ENOSPC; 2173 goto done2; 2174 } 2175 } 2176 2177 /* 2178 * Ready to go, issue the rename to the backend. Note that meta-data 2179 * updates to the related inodes occur separately from the rename 2180 * operation. 2181 * 2182 * NOTE: While it is not necessary to update ip->meta.name*, doing 2183 * so aids catastrophic recovery and debugging. 2184 */ 2185 if (error == 0) { 2186 hammer2_xop_nrename_t *xop4; 2187 2188 xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING); 2189 xop4->lhc = tlhc; 2190 xop4->ip_key = ip->meta.name_key; 2191 hammer2_xop_setip2(&xop4->head, ip); 2192 hammer2_xop_setip3(&xop4->head, tdip); 2193 hammer2_xop_setname(&xop4->head, fname, fname_len); 2194 hammer2_xop_setname2(&xop4->head, tname, tname_len); 2195 hammer2_xop_start(&xop4->head, hammer2_xop_nrename); 2196 2197 error = hammer2_xop_collect(&xop4->head, 0); 2198 error = hammer2_error_to_errno(error); 2199 hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP); 2200 2201 if (error == ENOENT) 2202 error = 0; 2203 2204 /* 2205 * Update inode meta-data. 2206 * 2207 * WARNING! The in-memory inode (ip) structure does not 2208 * maintain a copy of the inode's filename buffer. 2209 */ 2210 if (error == 0 && 2211 (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) { 2212 hammer2_inode_modify(ip); 2213 ip->meta.name_len = tname_len; 2214 ip->meta.name_key = tlhc; 2215 } 2216 if (error == 0) { 2217 hammer2_inode_modify(ip); 2218 ip->meta.iparent = tdip->meta.inum; 2219 } 2220 update_fdip = 1; 2221 update_tdip = 1; 2222 } 2223 2224 done2: 2225 /* 2226 * If no error, the backend has replaced the target directory entry. 2227 * We must adjust nlinks on the original replace target if it exists. 2228 */ 2229 if (error == 0 && tip) { 2230 int isopen; 2231 2232 isopen = cache_isopen(ap->a_tnch); 2233 hammer2_inode_unlink_finisher(tip, isopen); 2234 } 2235 2236 /* 2237 * Update directory mtimes to represent the something changed. 2238 */ 2239 if (update_fdip || update_tdip) { 2240 uint64_t mtime; 2241 2242 hammer2_update_time(&mtime); 2243 if (update_fdip) { 2244 hammer2_inode_modify(fdip); 2245 fdip->meta.mtime = mtime; 2246 } 2247 if (update_tdip) { 2248 hammer2_inode_modify(tdip); 2249 tdip->meta.mtime = mtime; 2250 } 2251 } 2252 if (tip) { 2253 hammer2_inode_unlock(tip); 2254 hammer2_inode_drop(tip); 2255 } 2256 hammer2_inode_unlock(ip); 2257 hammer2_inode_unlock(tdip); 2258 hammer2_inode_unlock(fdip); 2259 hammer2_inode_drop(ip); 2260 hammer2_inode_run_sideq(fdip->pmp, 0); 2261 2262 hammer2_trans_done(tdip->pmp); 2263 2264 /* 2265 * Issue the namecache update after unlocking all the internal 2266 * hammer2 structures, otherwise we might deadlock. 2267 * 2268 * WARNING! The target namespace must be updated atomically, 2269 * and we depend on cache_rename() to handle that for 2270 * us. Do not do a separate cache_unlink() because 2271 * that leaves a small window of opportunity for other 2272 * threads to allocate the target namespace before we 2273 * manage to complete our rename. 2274 * 2275 * WARNING! cache_rename() (and cache_unlink()) will properly 2276 * set VREF_FINALIZE on any attached vnode. Do not 2277 * call cache_setunresolved() manually before-hand as 2278 * this will prevent the flag from being set later via 2279 * cache_rename(). If VREF_FINALIZE is not properly set 2280 * and the inode is no longer in the topology, related 2281 * chains can remain dirty indefinitely. 2282 */ 2283 if (error == 0 && tip) { 2284 /*cache_unlink(ap->a_tnch); see above */ 2285 /*cache_setunresolved(ap->a_tnch); see above */ 2286 } 2287 if (error == 0) { 2288 cache_rename(ap->a_fnch, ap->a_tnch); 2289 hammer2_knote(ap->a_fdvp, NOTE_WRITE); 2290 hammer2_knote(ap->a_tdvp, NOTE_WRITE); 2291 hammer2_knote(fncp->nc_vp, NOTE_RENAME); 2292 } 2293 2294 return (error); 2295 } 2296 2297 /* 2298 * hammer2_vop_ioctl { vp, command, data, fflag, cred } 2299 */ 2300 static 2301 int 2302 hammer2_vop_ioctl(struct vop_ioctl_args *ap) 2303 { 2304 hammer2_inode_t *ip; 2305 int error; 2306 2307 ip = VTOI(ap->a_vp); 2308 2309 error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data, 2310 ap->a_fflag, ap->a_cred); 2311 return (error); 2312 } 2313 2314 static 2315 int 2316 hammer2_vop_mountctl(struct vop_mountctl_args *ap) 2317 { 2318 struct mount *mp; 2319 hammer2_pfs_t *pmp; 2320 int rc; 2321 2322 switch (ap->a_op) { 2323 case (MOUNTCTL_SET_EXPORT): 2324 mp = ap->a_head.a_ops->head.vv_mount; 2325 pmp = MPTOPMP(mp); 2326 2327 if (ap->a_ctllen != sizeof(struct export_args)) 2328 rc = (EINVAL); 2329 else 2330 rc = vfs_export(mp, &pmp->export, 2331 (const struct export_args *)ap->a_ctl); 2332 break; 2333 default: 2334 rc = vop_stdmountctl(ap); 2335 break; 2336 } 2337 return (rc); 2338 } 2339 2340 /* 2341 * KQFILTER 2342 */ 2343 static void filt_hammer2detach(struct knote *kn); 2344 static int filt_hammer2read(struct knote *kn, long hint); 2345 static int filt_hammer2write(struct knote *kn, long hint); 2346 static int filt_hammer2vnode(struct knote *kn, long hint); 2347 2348 static struct filterops hammer2read_filtops = 2349 { FILTEROP_ISFD | FILTEROP_MPSAFE, 2350 NULL, filt_hammer2detach, filt_hammer2read }; 2351 static struct filterops hammer2write_filtops = 2352 { FILTEROP_ISFD | FILTEROP_MPSAFE, 2353 NULL, filt_hammer2detach, filt_hammer2write }; 2354 static struct filterops hammer2vnode_filtops = 2355 { FILTEROP_ISFD | FILTEROP_MPSAFE, 2356 NULL, filt_hammer2detach, filt_hammer2vnode }; 2357 2358 static 2359 int 2360 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap) 2361 { 2362 struct vnode *vp = ap->a_vp; 2363 struct knote *kn = ap->a_kn; 2364 2365 switch (kn->kn_filter) { 2366 case EVFILT_READ: 2367 kn->kn_fop = &hammer2read_filtops; 2368 break; 2369 case EVFILT_WRITE: 2370 kn->kn_fop = &hammer2write_filtops; 2371 break; 2372 case EVFILT_VNODE: 2373 kn->kn_fop = &hammer2vnode_filtops; 2374 break; 2375 default: 2376 return (EOPNOTSUPP); 2377 } 2378 2379 kn->kn_hook = (caddr_t)vp; 2380 2381 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2382 2383 return(0); 2384 } 2385 2386 static void 2387 filt_hammer2detach(struct knote *kn) 2388 { 2389 struct vnode *vp = (void *)kn->kn_hook; 2390 2391 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 2392 } 2393 2394 static int 2395 filt_hammer2read(struct knote *kn, long hint) 2396 { 2397 struct vnode *vp = (void *)kn->kn_hook; 2398 hammer2_inode_t *ip = VTOI(vp); 2399 off_t off; 2400 2401 if (hint == NOTE_REVOKE) { 2402 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2403 return(1); 2404 } 2405 off = ip->meta.size - kn->kn_fp->f_offset; 2406 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 2407 if (kn->kn_sfflags & NOTE_OLDAPI) 2408 return(1); 2409 return (kn->kn_data != 0); 2410 } 2411 2412 2413 static int 2414 filt_hammer2write(struct knote *kn, long hint) 2415 { 2416 if (hint == NOTE_REVOKE) 2417 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 2418 kn->kn_data = 0; 2419 return (1); 2420 } 2421 2422 static int 2423 filt_hammer2vnode(struct knote *kn, long hint) 2424 { 2425 if (kn->kn_sfflags & hint) 2426 kn->kn_fflags |= hint; 2427 if (hint == NOTE_REVOKE) { 2428 kn->kn_flags |= (EV_EOF | EV_NODATA); 2429 return (1); 2430 } 2431 return (kn->kn_fflags != 0); 2432 } 2433 2434 /* 2435 * FIFO VOPS 2436 */ 2437 static 2438 int 2439 hammer2_vop_markatime(struct vop_markatime_args *ap) 2440 { 2441 hammer2_inode_t *ip; 2442 struct vnode *vp; 2443 2444 vp = ap->a_vp; 2445 ip = VTOI(vp); 2446 2447 if (ip->pmp->ronly) 2448 return (EROFS); 2449 return(0); 2450 } 2451 2452 static 2453 int 2454 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap) 2455 { 2456 int error; 2457 2458 error = VOCALL(&fifo_vnode_vops, &ap->a_head); 2459 if (error) 2460 error = hammer2_vop_kqfilter(ap); 2461 return(error); 2462 } 2463 2464 /* 2465 * VOPS vector 2466 */ 2467 struct vop_ops hammer2_vnode_vops = { 2468 .vop_default = vop_defaultop, 2469 .vop_fsync = hammer2_vop_fsync, 2470 .vop_getpages = vop_stdgetpages, 2471 .vop_putpages = vop_stdputpages, 2472 .vop_access = hammer2_vop_access, 2473 .vop_advlock = hammer2_vop_advlock, 2474 .vop_close = hammer2_vop_close, 2475 .vop_nlink = hammer2_vop_nlink, 2476 .vop_ncreate = hammer2_vop_ncreate, 2477 .vop_nsymlink = hammer2_vop_nsymlink, 2478 .vop_nremove = hammer2_vop_nremove, 2479 .vop_nrmdir = hammer2_vop_nrmdir, 2480 .vop_nrename = hammer2_vop_nrename, 2481 .vop_getattr = hammer2_vop_getattr, 2482 .vop_setattr = hammer2_vop_setattr, 2483 .vop_readdir = hammer2_vop_readdir, 2484 .vop_readlink = hammer2_vop_readlink, 2485 .vop_read = hammer2_vop_read, 2486 .vop_write = hammer2_vop_write, 2487 .vop_open = hammer2_vop_open, 2488 .vop_inactive = hammer2_vop_inactive, 2489 .vop_reclaim = hammer2_vop_reclaim, 2490 .vop_nresolve = hammer2_vop_nresolve, 2491 .vop_nlookupdotdot = hammer2_vop_nlookupdotdot, 2492 .vop_nmkdir = hammer2_vop_nmkdir, 2493 .vop_nmknod = hammer2_vop_nmknod, 2494 .vop_ioctl = hammer2_vop_ioctl, 2495 .vop_mountctl = hammer2_vop_mountctl, 2496 .vop_bmap = hammer2_vop_bmap, 2497 .vop_strategy = hammer2_vop_strategy, 2498 .vop_kqfilter = hammer2_vop_kqfilter 2499 }; 2500 2501 struct vop_ops hammer2_spec_vops = { 2502 .vop_default = vop_defaultop, 2503 .vop_fsync = hammer2_vop_fsync, 2504 .vop_read = vop_stdnoread, 2505 .vop_write = vop_stdnowrite, 2506 .vop_access = hammer2_vop_access, 2507 .vop_close = hammer2_vop_close, 2508 .vop_markatime = hammer2_vop_markatime, 2509 .vop_getattr = hammer2_vop_getattr, 2510 .vop_inactive = hammer2_vop_inactive, 2511 .vop_reclaim = hammer2_vop_reclaim, 2512 .vop_setattr = hammer2_vop_setattr 2513 }; 2514 2515 struct vop_ops hammer2_fifo_vops = { 2516 .vop_default = fifo_vnoperate, 2517 .vop_fsync = hammer2_vop_fsync, 2518 #if 0 2519 .vop_read = hammer2_vop_fiforead, 2520 .vop_write = hammer2_vop_fifowrite, 2521 #endif 2522 .vop_access = hammer2_vop_access, 2523 #if 0 2524 .vop_close = hammer2_vop_fifoclose, 2525 #endif 2526 .vop_markatime = hammer2_vop_markatime, 2527 .vop_getattr = hammer2_vop_getattr, 2528 .vop_inactive = hammer2_vop_inactive, 2529 .vop_reclaim = hammer2_vop_reclaim, 2530 .vop_setattr = hammer2_vop_setattr, 2531 .vop_kqfilter = hammer2_vop_fifokqfilter 2532 }; 2533 2534