fss.c revision 1.68
1 /* $NetBSD: fss.c,v 1.68 2010/04/13 07:58:54 hannken Exp $ */ 2 3 /*- 4 * Copyright (c) 2003 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Juergen Hannken-Illjes. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 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 the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * File system snapshot disk driver. 34 * 35 * Block/character interface to the snapshot of a mounted file system. 36 */ 37 38 #include <sys/cdefs.h> 39 __KERNEL_RCSID(0, "$NetBSD: fss.c,v 1.68 2010/04/13 07:58:54 hannken Exp $"); 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/namei.h> 44 #include <sys/proc.h> 45 #include <sys/errno.h> 46 #include <sys/malloc.h> 47 #include <sys/buf.h> 48 #include <sys/ioctl.h> 49 #include <sys/disklabel.h> 50 #include <sys/device.h> 51 #include <sys/disk.h> 52 #include <sys/stat.h> 53 #include <sys/mount.h> 54 #include <sys/vnode.h> 55 #include <sys/file.h> 56 #include <sys/uio.h> 57 #include <sys/conf.h> 58 #include <sys/kthread.h> 59 #include <sys/fstrans.h> 60 #include <sys/simplelock.h> 61 62 #include <miscfs/specfs/specdev.h> 63 64 #include <dev/fssvar.h> 65 66 #include <uvm/uvm.h> 67 68 void fssattach(int); 69 70 dev_type_open(fss_open); 71 dev_type_close(fss_close); 72 dev_type_read(fss_read); 73 dev_type_write(fss_write); 74 dev_type_ioctl(fss_ioctl); 75 dev_type_strategy(fss_strategy); 76 dev_type_dump(fss_dump); 77 dev_type_size(fss_size); 78 79 static void fss_unmount_hook(struct mount *); 80 static int fss_copy_on_write(void *, struct buf *, bool); 81 static inline void fss_error(struct fss_softc *, const char *); 82 static int fss_create_files(struct fss_softc *, struct fss_set *, 83 off_t *, struct lwp *); 84 static int fss_create_snapshot(struct fss_softc *, struct fss_set *, 85 struct lwp *); 86 static int fss_delete_snapshot(struct fss_softc *, struct lwp *); 87 static int fss_softc_alloc(struct fss_softc *); 88 static void fss_softc_free(struct fss_softc *); 89 static int fss_read_cluster(struct fss_softc *, u_int32_t); 90 static void fss_bs_thread(void *); 91 static int fss_bs_io(struct fss_softc *, fss_io_type, 92 u_int32_t, off_t, int, void *); 93 static u_int32_t *fss_bs_indir(struct fss_softc *, u_int32_t); 94 95 static kmutex_t fss_device_lock; /* Protect all units. */ 96 static int fss_num_attached = 0; /* Number of attached devices. */ 97 static struct vfs_hooks fss_vfs_hooks = { 98 .vh_unmount = fss_unmount_hook 99 }; 100 101 const struct bdevsw fss_bdevsw = { 102 fss_open, fss_close, fss_strategy, fss_ioctl, 103 fss_dump, fss_size, D_DISK | D_MPSAFE 104 }; 105 106 const struct cdevsw fss_cdevsw = { 107 fss_open, fss_close, fss_read, fss_write, fss_ioctl, 108 nostop, notty, nopoll, nommap, nokqfilter, D_DISK | D_MPSAFE 109 }; 110 111 static int fss_match(device_t, cfdata_t, void *); 112 static void fss_attach(device_t, device_t, void *); 113 static int fss_detach(device_t, int); 114 115 CFATTACH_DECL_NEW(fss, sizeof(struct fss_softc), 116 fss_match, fss_attach, fss_detach, NULL); 117 extern struct cfdriver fss_cd; 118 119 void 120 fssattach(int num) 121 { 122 123 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 124 if (config_cfattach_attach(fss_cd.cd_name, &fss_ca)) 125 aprint_error("%s: unable to register\n", fss_cd.cd_name); 126 } 127 128 static int 129 fss_match(device_t self, cfdata_t cfdata, void *aux) 130 { 131 return 1; 132 } 133 134 static void 135 fss_attach(device_t parent, device_t self, void *aux) 136 { 137 struct fss_softc *sc = device_private(self); 138 139 sc->sc_dev = self; 140 sc->sc_bdev = NODEV; 141 mutex_init(&sc->sc_slock, MUTEX_DEFAULT, IPL_NONE); 142 mutex_init(&sc->sc_lock, MUTEX_DEFAULT, IPL_NONE); 143 cv_init(&sc->sc_work_cv, "fssbs"); 144 cv_init(&sc->sc_cache_cv, "cowwait"); 145 bufq_alloc(&sc->sc_bufq, "fcfs", 0); 146 sc->sc_dkdev = malloc(sizeof(*sc->sc_dkdev), M_DEVBUF, M_WAITOK); 147 sc->sc_dkdev->dk_info = NULL; 148 disk_init(sc->sc_dkdev, device_xname(self), NULL); 149 if (!pmf_device_register(self, NULL, NULL)) 150 aprint_error_dev(self, "couldn't establish power handler\n"); 151 152 if (fss_num_attached++ == 0) 153 vfs_hooks_attach(&fss_vfs_hooks); 154 } 155 156 static int 157 fss_detach(device_t self, int flags) 158 { 159 struct fss_softc *sc = device_private(self); 160 161 if (sc->sc_flags & FSS_ACTIVE) 162 return EBUSY; 163 164 if (--fss_num_attached == 0) 165 vfs_hooks_detach(&fss_vfs_hooks); 166 167 pmf_device_deregister(self); 168 mutex_destroy(&sc->sc_slock); 169 mutex_destroy(&sc->sc_lock); 170 cv_destroy(&sc->sc_work_cv); 171 cv_destroy(&sc->sc_cache_cv); 172 bufq_drain(sc->sc_bufq); 173 bufq_free(sc->sc_bufq); 174 disk_destroy(sc->sc_dkdev); 175 free(sc->sc_dkdev, M_DEVBUF); 176 177 return 0; 178 } 179 180 int 181 fss_open(dev_t dev, int flags, int mode, struct lwp *l) 182 { 183 int mflag; 184 cfdata_t cf; 185 struct fss_softc *sc; 186 187 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 188 189 mutex_enter(&fss_device_lock); 190 191 sc = device_lookup_private(&fss_cd, minor(dev)); 192 if (sc == NULL) { 193 cf = malloc(sizeof(*cf), M_DEVBUF, M_WAITOK); 194 cf->cf_name = fss_cd.cd_name; 195 cf->cf_atname = fss_cd.cd_name; 196 cf->cf_unit = minor(dev); 197 cf->cf_fstate = FSTATE_STAR; 198 sc = device_private(config_attach_pseudo(cf)); 199 if (sc == NULL) 200 return ENOMEM; 201 } 202 203 mutex_enter(&sc->sc_slock); 204 205 sc->sc_flags |= mflag; 206 207 mutex_exit(&sc->sc_slock); 208 mutex_exit(&fss_device_lock); 209 210 return 0; 211 } 212 213 int 214 fss_close(dev_t dev, int flags, int mode, struct lwp *l) 215 { 216 int mflag, error; 217 cfdata_t cf; 218 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 219 220 mflag = (mode == S_IFCHR ? FSS_CDEV_OPEN : FSS_BDEV_OPEN); 221 error = 0; 222 223 restart: 224 mutex_enter(&sc->sc_slock); 225 if ((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) != mflag) { 226 sc->sc_flags &= ~mflag; 227 mutex_exit(&sc->sc_slock); 228 return 0; 229 } 230 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 231 (sc->sc_uflags & FSS_UNCONFIG_ON_CLOSE) != 0) { 232 sc->sc_uflags &= ~FSS_UNCONFIG_ON_CLOSE; 233 mutex_exit(&sc->sc_slock); 234 error = fss_ioctl(dev, FSSIOCCLR, NULL, FWRITE, l); 235 goto restart; 236 } 237 if ((sc->sc_flags & FSS_ACTIVE) != 0) { 238 mutex_exit(&sc->sc_slock); 239 return error; 240 } 241 if (! mutex_tryenter(&fss_device_lock)) { 242 mutex_exit(&sc->sc_slock); 243 goto restart; 244 } 245 246 KASSERT((sc->sc_flags & FSS_ACTIVE) == 0); 247 KASSERT((sc->sc_flags & (FSS_CDEV_OPEN|FSS_BDEV_OPEN)) == mflag); 248 mutex_exit(&sc->sc_slock); 249 cf = device_cfdata(sc->sc_dev); 250 error = config_detach(sc->sc_dev, DETACH_QUIET); 251 if (! error) 252 free(cf, M_DEVBUF); 253 mutex_exit(&fss_device_lock); 254 255 return error; 256 } 257 258 void 259 fss_strategy(struct buf *bp) 260 { 261 const bool write = ((bp->b_flags & B_READ) != B_READ); 262 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(bp->b_dev)); 263 264 mutex_enter(&sc->sc_slock); 265 266 if (write || !FSS_ISVALID(sc)) { 267 268 mutex_exit(&sc->sc_slock); 269 270 bp->b_error = (write ? EROFS : ENXIO); 271 bp->b_resid = bp->b_bcount; 272 biodone(bp); 273 return; 274 } 275 276 bp->b_rawblkno = bp->b_blkno; 277 bufq_put(sc->sc_bufq, bp); 278 cv_signal(&sc->sc_work_cv); 279 280 mutex_exit(&sc->sc_slock); 281 } 282 283 int 284 fss_read(dev_t dev, struct uio *uio, int flags) 285 { 286 return physio(fss_strategy, NULL, dev, B_READ, minphys, uio); 287 } 288 289 int 290 fss_write(dev_t dev, struct uio *uio, int flags) 291 { 292 return physio(fss_strategy, NULL, dev, B_WRITE, minphys, uio); 293 } 294 295 int 296 fss_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 297 { 298 int error; 299 struct fss_softc *sc = device_lookup_private(&fss_cd, minor(dev)); 300 struct fss_set *fss = (struct fss_set *)data; 301 struct fss_get *fsg = (struct fss_get *)data; 302 303 switch (cmd) { 304 case FSSIOCSET: 305 mutex_enter(&sc->sc_lock); 306 if ((flag & FWRITE) == 0) 307 error = EPERM; 308 else if ((sc->sc_flags & FSS_ACTIVE) != 0) 309 error = EBUSY; 310 else 311 error = fss_create_snapshot(sc, fss, l); 312 mutex_exit(&sc->sc_lock); 313 break; 314 315 case FSSIOCCLR: 316 mutex_enter(&sc->sc_lock); 317 if ((flag & FWRITE) == 0) 318 error = EPERM; 319 else if ((sc->sc_flags & FSS_ACTIVE) == 0) 320 error = ENXIO; 321 else 322 error = fss_delete_snapshot(sc, l); 323 mutex_exit(&sc->sc_lock); 324 break; 325 326 case FSSIOCGET: 327 mutex_enter(&sc->sc_lock); 328 switch (sc->sc_flags & (FSS_PERSISTENT | FSS_ACTIVE)) { 329 case FSS_ACTIVE: 330 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 331 fsg->fsg_csize = FSS_CLSIZE(sc); 332 fsg->fsg_time = sc->sc_time; 333 fsg->fsg_mount_size = sc->sc_clcount; 334 fsg->fsg_bs_size = sc->sc_clnext; 335 error = 0; 336 break; 337 case FSS_PERSISTENT | FSS_ACTIVE: 338 memcpy(fsg->fsg_mount, sc->sc_mntname, MNAMELEN); 339 fsg->fsg_csize = 0; 340 fsg->fsg_time = sc->sc_time; 341 fsg->fsg_mount_size = 0; 342 fsg->fsg_bs_size = 0; 343 error = 0; 344 break; 345 default: 346 error = ENXIO; 347 break; 348 } 349 mutex_exit(&sc->sc_lock); 350 break; 351 352 case FSSIOFSET: 353 mutex_enter(&sc->sc_slock); 354 sc->sc_uflags = *(int *)data; 355 mutex_exit(&sc->sc_slock); 356 error = 0; 357 break; 358 359 case FSSIOFGET: 360 mutex_enter(&sc->sc_slock); 361 *(int *)data = sc->sc_uflags; 362 mutex_exit(&sc->sc_slock); 363 error = 0; 364 break; 365 366 default: 367 error = EINVAL; 368 break; 369 } 370 371 return error; 372 } 373 374 int 375 fss_size(dev_t dev) 376 { 377 return -1; 378 } 379 380 int 381 fss_dump(dev_t dev, daddr_t blkno, void *va, 382 size_t size) 383 { 384 return EROFS; 385 } 386 387 /* 388 * An error occurred reading or writing the snapshot or backing store. 389 * If it is the first error log to console. 390 * The caller holds the mutex. 391 */ 392 static inline void 393 fss_error(struct fss_softc *sc, const char *msg) 394 { 395 396 if ((sc->sc_flags & (FSS_ACTIVE|FSS_ERROR)) == FSS_ACTIVE) 397 aprint_error_dev(sc->sc_dev, "snapshot invalid: %s\n", msg); 398 if ((sc->sc_flags & FSS_ACTIVE) == FSS_ACTIVE) 399 sc->sc_flags |= FSS_ERROR; 400 } 401 402 /* 403 * Allocate the variable sized parts of the softc and 404 * fork the kernel thread. 405 * 406 * The fields sc_clcount, sc_clshift, sc_cache_size and sc_indir_size 407 * must be initialized. 408 */ 409 static int 410 fss_softc_alloc(struct fss_softc *sc) 411 { 412 int i, error; 413 414 if ((sc->sc_flags & FSS_PERSISTENT) == 0) { 415 sc->sc_copied = 416 kmem_zalloc(howmany(sc->sc_clcount, NBBY), KM_SLEEP); 417 if (sc->sc_copied == NULL) 418 return(ENOMEM); 419 420 sc->sc_cache = kmem_alloc(sc->sc_cache_size * 421 sizeof(struct fss_cache), KM_SLEEP); 422 if (sc->sc_cache == NULL) 423 return(ENOMEM); 424 425 for (i = 0; i < sc->sc_cache_size; i++) { 426 sc->sc_cache[i].fc_type = FSS_CACHE_FREE; 427 sc->sc_cache[i].fc_data = 428 kmem_alloc(FSS_CLSIZE(sc), KM_SLEEP); 429 if (sc->sc_cache[i].fc_data == NULL) 430 return(ENOMEM); 431 cv_init(&sc->sc_cache[i].fc_state_cv, "cowwait1"); 432 } 433 434 sc->sc_indir_valid = 435 kmem_zalloc(howmany(sc->sc_indir_size, NBBY), KM_SLEEP); 436 if (sc->sc_indir_valid == NULL) 437 return(ENOMEM); 438 439 sc->sc_indir_data = kmem_zalloc(FSS_CLSIZE(sc), KM_SLEEP); 440 if (sc->sc_indir_data == NULL) 441 return(ENOMEM); 442 } else { 443 sc->sc_copied = NULL; 444 sc->sc_cache = NULL; 445 sc->sc_indir_valid = NULL; 446 sc->sc_indir_data = NULL; 447 } 448 449 sc->sc_flags |= FSS_BS_THREAD; 450 if ((error = kthread_create(PRI_BIO, 0, NULL, fss_bs_thread, sc, 451 &sc->sc_bs_lwp, device_xname(sc->sc_dev))) != 0) { 452 sc->sc_flags &= ~FSS_BS_THREAD; 453 return error; 454 } 455 456 disk_attach(sc->sc_dkdev); 457 458 return 0; 459 } 460 461 /* 462 * Free the variable sized parts of the softc. 463 */ 464 static void 465 fss_softc_free(struct fss_softc *sc) 466 { 467 int i; 468 469 if ((sc->sc_flags & FSS_BS_THREAD) != 0) { 470 mutex_enter(&sc->sc_slock); 471 sc->sc_flags &= ~FSS_BS_THREAD; 472 cv_signal(&sc->sc_work_cv); 473 while (sc->sc_bs_lwp != NULL) 474 kpause("fssdetach", false, 1, &sc->sc_slock); 475 mutex_exit(&sc->sc_slock); 476 477 disk_detach(sc->sc_dkdev); 478 } 479 480 if (sc->sc_copied != NULL) 481 kmem_free(sc->sc_copied, howmany(sc->sc_clcount, NBBY)); 482 sc->sc_copied = NULL; 483 484 if (sc->sc_cache != NULL) { 485 for (i = 0; i < sc->sc_cache_size; i++) 486 if (sc->sc_cache[i].fc_data != NULL) { 487 cv_destroy(&sc->sc_cache[i].fc_state_cv); 488 kmem_free(sc->sc_cache[i].fc_data, 489 FSS_CLSIZE(sc)); 490 } 491 kmem_free(sc->sc_cache, 492 sc->sc_cache_size*sizeof(struct fss_cache)); 493 } 494 sc->sc_cache = NULL; 495 496 if (sc->sc_indir_valid != NULL) 497 kmem_free(sc->sc_indir_valid, howmany(sc->sc_indir_size, NBBY)); 498 sc->sc_indir_valid = NULL; 499 500 if (sc->sc_indir_data != NULL) 501 kmem_free(sc->sc_indir_data, FSS_CLSIZE(sc)); 502 sc->sc_indir_data = NULL; 503 } 504 505 /* 506 * Set all active snapshots on this file system into ERROR state. 507 */ 508 static void 509 fss_unmount_hook(struct mount *mp) 510 { 511 int i; 512 struct fss_softc *sc; 513 514 mutex_enter(&fss_device_lock); 515 for (i = 0; i < fss_cd.cd_ndevs; i++) { 516 if ((sc = device_lookup_private(&fss_cd, i)) == NULL) 517 continue; 518 mutex_enter(&sc->sc_slock); 519 if ((sc->sc_flags & FSS_ACTIVE) != 0 && 520 sc->sc_mount == mp) 521 fss_error(sc, "forced unmount"); 522 mutex_exit(&sc->sc_slock); 523 } 524 mutex_exit(&fss_device_lock); 525 } 526 527 /* 528 * A buffer is written to the snapshotted block device. Copy to 529 * backing store if needed. 530 */ 531 static int 532 fss_copy_on_write(void *v, struct buf *bp, bool data_valid) 533 { 534 int error; 535 u_int32_t cl, ch, c; 536 struct fss_softc *sc = v; 537 538 mutex_enter(&sc->sc_slock); 539 if (!FSS_ISVALID(sc)) { 540 mutex_exit(&sc->sc_slock); 541 return 0; 542 } 543 544 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 545 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 546 error = 0; 547 if (curlwp == uvm.pagedaemon_lwp) { 548 for (c = cl; c <= ch; c++) 549 if (isclr(sc->sc_copied, c)) { 550 error = ENOMEM; 551 break; 552 } 553 } 554 mutex_exit(&sc->sc_slock); 555 556 if (error == 0) 557 for (c = cl; c <= ch; c++) { 558 error = fss_read_cluster(sc, c); 559 if (error) 560 break; 561 } 562 563 return error; 564 } 565 566 /* 567 * Lookup and open needed files. 568 * 569 * For file system internal snapshot initializes sc_mntname, sc_mount, 570 * sc_bs_vp and sc_time. 571 * 572 * Otherwise returns dev and size of the underlying block device. 573 * Initializes sc_mntname, sc_mount, sc_bdev, sc_bs_vp and sc_mount 574 */ 575 static int 576 fss_create_files(struct fss_softc *sc, struct fss_set *fss, 577 off_t *bsize, struct lwp *l) 578 { 579 int error, bits, fsbsize; 580 struct timespec ts; 581 struct partinfo dpart; 582 /* nd -> nd2 to reduce mistakes while updating only some namei calls */ 583 struct nameidata nd2; 584 struct vnode *vp; 585 586 /* 587 * Get the mounted file system. 588 */ 589 590 error = namei_simple_user(fss->fss_mount, 591 NSM_FOLLOW_NOEMULROOT, &vp); 592 if (error != 0) 593 return error; 594 595 if ((vp->v_vflag & VV_ROOT) != VV_ROOT) { 596 vrele(vp); 597 return EINVAL; 598 } 599 600 sc->sc_mount = vp->v_mount; 601 memcpy(sc->sc_mntname, sc->sc_mount->mnt_stat.f_mntonname, MNAMELEN); 602 603 vrele(vp); 604 605 /* 606 * Check for file system internal snapshot. 607 */ 608 609 error = namei_simple_user(fss->fss_bstore, 610 NSM_FOLLOW_NOEMULROOT, &vp); 611 if (error != 0) 612 return error; 613 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 614 if (error != 0) { 615 vrele(vp); 616 return error; 617 } 618 619 if (vp->v_type == VREG && vp->v_mount == sc->sc_mount) { 620 sc->sc_flags |= FSS_PERSISTENT; 621 sc->sc_bs_vp = vp; 622 623 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 624 bits = sizeof(sc->sc_bs_bshift)*NBBY; 625 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < bits; 626 sc->sc_bs_bshift++) 627 if (FSS_FSBSIZE(sc) == fsbsize) 628 break; 629 if (sc->sc_bs_bshift >= bits) { 630 VOP_UNLOCK(sc->sc_bs_vp, 0); 631 return EINVAL; 632 } 633 634 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 635 sc->sc_clshift = 0; 636 637 error = VFS_SNAPSHOT(sc->sc_mount, sc->sc_bs_vp, &ts); 638 TIMESPEC_TO_TIMEVAL(&sc->sc_time, &ts); 639 640 VOP_UNLOCK(sc->sc_bs_vp, 0); 641 642 return error; 643 } 644 vput(vp); 645 646 /* 647 * Get the block device it is mounted on. 648 */ 649 650 error = namei_simple_kernel(sc->sc_mount->mnt_stat.f_mntfromname, 651 NSM_FOLLOW_NOEMULROOT, &vp); 652 if (error != 0) 653 return error; 654 655 if (vp->v_type != VBLK) { 656 vrele(vp); 657 return EINVAL; 658 } 659 660 sc->sc_bdev = vp->v_rdev; 661 vrele(vp); 662 663 /* 664 * Get the block device size. 665 */ 666 667 error = bdev_ioctl(sc->sc_bdev, DIOCGPART, &dpart, FREAD, l); 668 if (error) 669 return error; 670 671 *bsize = (off_t)dpart.disklab->d_secsize*dpart.part->p_size; 672 673 /* 674 * Get the backing store 675 */ 676 677 NDINIT(&nd2, LOOKUP, FOLLOW, UIO_USERSPACE, fss->fss_bstore); 678 if ((error = vn_open(&nd2, FREAD|FWRITE, 0)) != 0) 679 return error; 680 VOP_UNLOCK(nd2.ni_vp, 0); 681 682 sc->sc_bs_vp = nd2.ni_vp; 683 684 if (nd2.ni_vp->v_type != VREG && nd2.ni_vp->v_type != VCHR) 685 return EINVAL; 686 687 if (sc->sc_bs_vp->v_type == VREG) { 688 fsbsize = sc->sc_bs_vp->v_mount->mnt_stat.f_iosize; 689 if (fsbsize & (fsbsize-1)) /* No power of two */ 690 return EINVAL; 691 for (sc->sc_bs_bshift = 1; sc->sc_bs_bshift < 32; 692 sc->sc_bs_bshift++) 693 if (FSS_FSBSIZE(sc) == fsbsize) 694 break; 695 if (sc->sc_bs_bshift >= 32) 696 return EINVAL; 697 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 698 } else { 699 sc->sc_bs_bshift = DEV_BSHIFT; 700 sc->sc_bs_bmask = FSS_FSBSIZE(sc)-1; 701 } 702 703 return 0; 704 } 705 706 /* 707 * Create a snapshot. 708 */ 709 static int 710 fss_create_snapshot(struct fss_softc *sc, struct fss_set *fss, struct lwp *l) 711 { 712 int len, error; 713 u_int32_t csize; 714 off_t bsize; 715 716 bsize = 0; /* XXX gcc */ 717 718 /* 719 * Open needed files. 720 */ 721 if ((error = fss_create_files(sc, fss, &bsize, l)) != 0) 722 goto bad; 723 724 if (sc->sc_flags & FSS_PERSISTENT) { 725 fss_softc_alloc(sc); 726 sc->sc_flags |= FSS_ACTIVE; 727 return 0; 728 } 729 730 /* 731 * Set cluster size. Must be a power of two and 732 * a multiple of backing store block size. 733 */ 734 if (fss->fss_csize <= 0) 735 csize = MAXPHYS; 736 else 737 csize = fss->fss_csize; 738 if (bsize/csize > FSS_CLUSTER_MAX) 739 csize = bsize/FSS_CLUSTER_MAX+1; 740 741 for (sc->sc_clshift = sc->sc_bs_bshift; sc->sc_clshift < 32; 742 sc->sc_clshift++) 743 if (FSS_CLSIZE(sc) >= csize) 744 break; 745 if (sc->sc_clshift >= 32) { 746 error = EINVAL; 747 goto bad; 748 } 749 sc->sc_clmask = FSS_CLSIZE(sc)-1; 750 751 /* 752 * Set number of cache slots. 753 */ 754 if (FSS_CLSIZE(sc) <= 8192) 755 sc->sc_cache_size = 32; 756 else if (FSS_CLSIZE(sc) <= 65536) 757 sc->sc_cache_size = 8; 758 else 759 sc->sc_cache_size = 4; 760 761 /* 762 * Set number of clusters and size of last cluster. 763 */ 764 sc->sc_clcount = FSS_BTOCL(sc, bsize-1)+1; 765 sc->sc_clresid = FSS_CLOFF(sc, bsize-1)+1; 766 767 /* 768 * Set size of indirect table. 769 */ 770 len = sc->sc_clcount*sizeof(u_int32_t); 771 sc->sc_indir_size = FSS_BTOCL(sc, len)+1; 772 sc->sc_clnext = sc->sc_indir_size; 773 sc->sc_indir_cur = 0; 774 775 if ((error = fss_softc_alloc(sc)) != 0) 776 goto bad; 777 778 /* 779 * Activate the snapshot. 780 */ 781 782 if ((error = vfs_suspend(sc->sc_mount, 0)) != 0) 783 goto bad; 784 785 microtime(&sc->sc_time); 786 787 if (error == 0) 788 error = fscow_establish(sc->sc_mount, 789 fss_copy_on_write, sc); 790 if (error == 0) 791 sc->sc_flags |= FSS_ACTIVE; 792 793 vfs_resume(sc->sc_mount); 794 795 if (error != 0) 796 goto bad; 797 798 aprint_debug_dev(sc->sc_dev, "%s snapshot active\n", sc->sc_mntname); 799 aprint_debug_dev(sc->sc_dev, 800 "%u clusters of %u, %u cache slots, %u indir clusters\n", 801 sc->sc_clcount, FSS_CLSIZE(sc), 802 sc->sc_cache_size, sc->sc_indir_size); 803 804 return 0; 805 806 bad: 807 fss_softc_free(sc); 808 if (sc->sc_bs_vp != NULL) { 809 if (sc->sc_flags & FSS_PERSISTENT) 810 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 811 else 812 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 813 } 814 sc->sc_bs_vp = NULL; 815 816 return error; 817 } 818 819 /* 820 * Delete a snapshot. 821 */ 822 static int 823 fss_delete_snapshot(struct fss_softc *sc, struct lwp *l) 824 { 825 826 if ((sc->sc_flags & FSS_PERSISTENT) == 0) 827 fscow_disestablish(sc->sc_mount, fss_copy_on_write, sc); 828 829 mutex_enter(&sc->sc_slock); 830 sc->sc_flags &= ~(FSS_ACTIVE|FSS_ERROR); 831 sc->sc_mount = NULL; 832 sc->sc_bdev = NODEV; 833 mutex_exit(&sc->sc_slock); 834 835 fss_softc_free(sc); 836 if (sc->sc_flags & FSS_PERSISTENT) 837 vn_close(sc->sc_bs_vp, FREAD, l->l_cred); 838 else 839 vn_close(sc->sc_bs_vp, FREAD|FWRITE, l->l_cred); 840 sc->sc_bs_vp = NULL; 841 sc->sc_flags &= ~FSS_PERSISTENT; 842 843 return 0; 844 } 845 846 /* 847 * Read a cluster from the snapshotted block device to the cache. 848 */ 849 static int 850 fss_read_cluster(struct fss_softc *sc, u_int32_t cl) 851 { 852 int error, todo, offset, len; 853 daddr_t dblk; 854 struct buf *bp, *mbp; 855 struct fss_cache *scp, *scl; 856 857 /* 858 * Get a free cache slot. 859 */ 860 scl = sc->sc_cache+sc->sc_cache_size; 861 862 mutex_enter(&sc->sc_slock); 863 864 restart: 865 if (isset(sc->sc_copied, cl) || !FSS_ISVALID(sc)) { 866 mutex_exit(&sc->sc_slock); 867 return 0; 868 } 869 870 for (scp = sc->sc_cache; scp < scl; scp++) 871 if (scp->fc_cluster == cl) { 872 if (scp->fc_type == FSS_CACHE_VALID) { 873 mutex_exit(&sc->sc_slock); 874 return 0; 875 } else if (scp->fc_type == FSS_CACHE_BUSY) { 876 cv_wait(&scp->fc_state_cv, &sc->sc_slock); 877 goto restart; 878 } 879 } 880 881 for (scp = sc->sc_cache; scp < scl; scp++) 882 if (scp->fc_type == FSS_CACHE_FREE) { 883 scp->fc_type = FSS_CACHE_BUSY; 884 scp->fc_cluster = cl; 885 break; 886 } 887 if (scp >= scl) { 888 cv_wait(&sc->sc_cache_cv, &sc->sc_slock); 889 goto restart; 890 } 891 892 mutex_exit(&sc->sc_slock); 893 894 /* 895 * Start the read. 896 */ 897 dblk = btodb(FSS_CLTOB(sc, cl)); 898 if (cl == sc->sc_clcount-1) { 899 todo = sc->sc_clresid; 900 memset((char *)scp->fc_data + todo, 0, FSS_CLSIZE(sc) - todo); 901 } else 902 todo = FSS_CLSIZE(sc); 903 offset = 0; 904 mbp = getiobuf(NULL, true); 905 mbp->b_bufsize = todo; 906 mbp->b_data = scp->fc_data; 907 mbp->b_resid = mbp->b_bcount = todo; 908 mbp->b_flags = B_READ; 909 mbp->b_cflags = BC_BUSY; 910 mbp->b_dev = sc->sc_bdev; 911 while (todo > 0) { 912 len = todo; 913 if (len > MAXPHYS) 914 len = MAXPHYS; 915 if (btodb(FSS_CLTOB(sc, cl)) == dblk && len == todo) 916 bp = mbp; 917 else { 918 bp = getiobuf(NULL, true); 919 nestiobuf_setup(mbp, bp, offset, len); 920 } 921 bp->b_lblkno = 0; 922 bp->b_blkno = dblk; 923 bdev_strategy(bp); 924 dblk += btodb(len); 925 offset += len; 926 todo -= len; 927 } 928 error = biowait(mbp); 929 putiobuf(mbp); 930 931 mutex_enter(&sc->sc_slock); 932 scp->fc_type = (error ? FSS_CACHE_FREE : FSS_CACHE_VALID); 933 cv_broadcast(&scp->fc_state_cv); 934 if (error == 0) { 935 setbit(sc->sc_copied, scp->fc_cluster); 936 cv_signal(&sc->sc_work_cv); 937 } 938 mutex_exit(&sc->sc_slock); 939 940 return error; 941 } 942 943 /* 944 * Read/write clusters from/to backing store. 945 * For persistent snapshots must be called with cl == 0. off is the 946 * offset into the snapshot. 947 */ 948 static int 949 fss_bs_io(struct fss_softc *sc, fss_io_type rw, 950 u_int32_t cl, off_t off, int len, void *data) 951 { 952 int error; 953 954 off += FSS_CLTOB(sc, cl); 955 956 vn_lock(sc->sc_bs_vp, LK_EXCLUSIVE|LK_RETRY); 957 958 error = vn_rdwr((rw == FSS_READ ? UIO_READ : UIO_WRITE), sc->sc_bs_vp, 959 data, len, off, UIO_SYSSPACE, IO_UNIT|IO_NODELOCKED, 960 sc->sc_bs_lwp->l_cred, NULL, NULL); 961 if (error == 0) { 962 mutex_enter(&sc->sc_bs_vp->v_interlock); 963 error = VOP_PUTPAGES(sc->sc_bs_vp, trunc_page(off), 964 round_page(off+len), PGO_CLEANIT|PGO_SYNCIO|PGO_FREE); 965 } 966 967 VOP_UNLOCK(sc->sc_bs_vp, 0); 968 969 return error; 970 } 971 972 /* 973 * Get a pointer to the indirect slot for this cluster. 974 */ 975 static u_int32_t * 976 fss_bs_indir(struct fss_softc *sc, u_int32_t cl) 977 { 978 u_int32_t icl; 979 int ioff; 980 981 icl = cl/(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 982 ioff = cl%(FSS_CLSIZE(sc)/sizeof(u_int32_t)); 983 984 if (sc->sc_indir_cur == icl) 985 return &sc->sc_indir_data[ioff]; 986 987 if (sc->sc_indir_dirty) { 988 if (fss_bs_io(sc, FSS_WRITE, sc->sc_indir_cur, 0, 989 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 990 return NULL; 991 setbit(sc->sc_indir_valid, sc->sc_indir_cur); 992 } 993 994 sc->sc_indir_dirty = 0; 995 sc->sc_indir_cur = icl; 996 997 if (isset(sc->sc_indir_valid, sc->sc_indir_cur)) { 998 if (fss_bs_io(sc, FSS_READ, sc->sc_indir_cur, 0, 999 FSS_CLSIZE(sc), (void *)sc->sc_indir_data) != 0) 1000 return NULL; 1001 } else 1002 memset(sc->sc_indir_data, 0, FSS_CLSIZE(sc)); 1003 1004 return &sc->sc_indir_data[ioff]; 1005 } 1006 1007 /* 1008 * The kernel thread (one for every active snapshot). 1009 * 1010 * After wakeup it cleans the cache and runs the I/O requests. 1011 */ 1012 static void 1013 fss_bs_thread(void *arg) 1014 { 1015 bool thread_idle, is_valid; 1016 int error, i, todo, len, crotor, is_read; 1017 long off; 1018 char *addr; 1019 u_int32_t c, cl, ch, *indirp; 1020 struct buf *bp, *nbp; 1021 struct fss_softc *sc; 1022 struct fss_cache *scp, *scl; 1023 1024 sc = arg; 1025 scl = sc->sc_cache+sc->sc_cache_size; 1026 crotor = 0; 1027 thread_idle = false; 1028 1029 mutex_enter(&sc->sc_slock); 1030 1031 for (;;) { 1032 if (thread_idle) 1033 cv_wait(&sc->sc_work_cv, &sc->sc_slock); 1034 thread_idle = true; 1035 if ((sc->sc_flags & FSS_BS_THREAD) == 0) { 1036 sc->sc_bs_lwp = NULL; 1037 mutex_exit(&sc->sc_slock); 1038 kthread_exit(0); 1039 } 1040 1041 /* 1042 * Process I/O requests (persistent) 1043 */ 1044 1045 if (sc->sc_flags & FSS_PERSISTENT) { 1046 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1047 continue; 1048 is_valid = FSS_ISVALID(sc); 1049 is_read = (bp->b_flags & B_READ); 1050 thread_idle = false; 1051 mutex_exit(&sc->sc_slock); 1052 1053 if (is_valid) { 1054 disk_busy(sc->sc_dkdev); 1055 error = fss_bs_io(sc, FSS_READ, 0, 1056 dbtob(bp->b_blkno), bp->b_bcount, 1057 bp->b_data); 1058 disk_unbusy(sc->sc_dkdev, 1059 (error ? 0 : bp->b_bcount), is_read); 1060 } else 1061 error = ENXIO; 1062 1063 bp->b_error = error; 1064 bp->b_resid = (error ? bp->b_bcount : 0); 1065 biodone(bp); 1066 1067 mutex_enter(&sc->sc_slock); 1068 continue; 1069 } 1070 1071 /* 1072 * Clean the cache 1073 */ 1074 for (i = 0; i < sc->sc_cache_size; i++) { 1075 crotor = (crotor + 1) % sc->sc_cache_size; 1076 scp = sc->sc_cache + crotor; 1077 if (scp->fc_type != FSS_CACHE_VALID) 1078 continue; 1079 mutex_exit(&sc->sc_slock); 1080 1081 thread_idle = false; 1082 indirp = fss_bs_indir(sc, scp->fc_cluster); 1083 if (indirp != NULL) { 1084 error = fss_bs_io(sc, FSS_WRITE, sc->sc_clnext, 1085 0, FSS_CLSIZE(sc), scp->fc_data); 1086 } else 1087 error = EIO; 1088 1089 mutex_enter(&sc->sc_slock); 1090 if (error == 0) { 1091 *indirp = sc->sc_clnext++; 1092 sc->sc_indir_dirty = 1; 1093 } else 1094 fss_error(sc, "write error on backing store"); 1095 1096 scp->fc_type = FSS_CACHE_FREE; 1097 cv_signal(&sc->sc_cache_cv); 1098 break; 1099 } 1100 1101 /* 1102 * Process I/O requests 1103 */ 1104 if ((bp = bufq_get(sc->sc_bufq)) == NULL) 1105 continue; 1106 is_valid = FSS_ISVALID(sc); 1107 is_read = (bp->b_flags & B_READ); 1108 thread_idle = false; 1109 1110 if (!is_valid) { 1111 mutex_exit(&sc->sc_slock); 1112 1113 bp->b_error = ENXIO; 1114 bp->b_resid = bp->b_bcount; 1115 biodone(bp); 1116 1117 mutex_enter(&sc->sc_slock); 1118 continue; 1119 } 1120 1121 disk_busy(sc->sc_dkdev); 1122 1123 /* 1124 * First read from the snapshotted block device unless 1125 * this request is completely covered by backing store. 1126 */ 1127 1128 cl = FSS_BTOCL(sc, dbtob(bp->b_blkno)); 1129 off = FSS_CLOFF(sc, dbtob(bp->b_blkno)); 1130 ch = FSS_BTOCL(sc, dbtob(bp->b_blkno)+bp->b_bcount-1); 1131 error = 0; 1132 bp->b_resid = 0; 1133 bp->b_error = 0; 1134 for (c = cl; c <= ch; c++) { 1135 if (isset(sc->sc_copied, c)) 1136 continue; 1137 mutex_exit(&sc->sc_slock); 1138 1139 /* Not on backing store, read from device. */ 1140 nbp = getiobuf(NULL, true); 1141 nbp->b_flags = B_READ; 1142 nbp->b_resid = nbp->b_bcount = bp->b_bcount; 1143 nbp->b_bufsize = bp->b_bcount; 1144 nbp->b_data = bp->b_data; 1145 nbp->b_blkno = bp->b_blkno; 1146 nbp->b_lblkno = 0; 1147 nbp->b_dev = sc->sc_bdev; 1148 SET(nbp->b_cflags, BC_BUSY); /* mark buffer busy */ 1149 1150 bdev_strategy(nbp); 1151 1152 error = biowait(nbp); 1153 if (error != 0) { 1154 bp->b_resid = bp->b_bcount; 1155 bp->b_error = nbp->b_error; 1156 disk_unbusy(sc->sc_dkdev, 0, is_read); 1157 biodone(bp); 1158 } 1159 putiobuf(nbp); 1160 1161 mutex_enter(&sc->sc_slock); 1162 break; 1163 } 1164 if (error) 1165 continue; 1166 1167 /* 1168 * Replace those parts that have been saved to backing store. 1169 */ 1170 1171 addr = bp->b_data; 1172 todo = bp->b_bcount; 1173 for (c = cl; c <= ch; c++, off = 0, todo -= len, addr += len) { 1174 len = FSS_CLSIZE(sc)-off; 1175 if (len > todo) 1176 len = todo; 1177 if (isclr(sc->sc_copied, c)) 1178 continue; 1179 mutex_exit(&sc->sc_slock); 1180 1181 indirp = fss_bs_indir(sc, c); 1182 if (indirp == NULL || *indirp == 0) { 1183 /* 1184 * Not on backing store. Either in cache 1185 * or hole in the snapshotted block device. 1186 */ 1187 1188 mutex_enter(&sc->sc_slock); 1189 for (scp = sc->sc_cache; scp < scl; scp++) 1190 if (scp->fc_type == FSS_CACHE_VALID && 1191 scp->fc_cluster == c) 1192 break; 1193 if (scp < scl) 1194 memcpy(addr, (char *)scp->fc_data+off, 1195 len); 1196 else 1197 memset(addr, 0, len); 1198 continue; 1199 } 1200 1201 /* 1202 * Read from backing store. 1203 */ 1204 error = 1205 fss_bs_io(sc, FSS_READ, *indirp, off, len, addr); 1206 1207 mutex_enter(&sc->sc_slock); 1208 if (error) { 1209 bp->b_resid = bp->b_bcount; 1210 bp->b_error = error; 1211 break; 1212 } 1213 } 1214 mutex_exit(&sc->sc_slock); 1215 1216 disk_unbusy(sc->sc_dkdev, (error ? 0 : bp->b_bcount), is_read); 1217 biodone(bp); 1218 1219 mutex_enter(&sc->sc_slock); 1220 } 1221 } 1222 1223 #ifdef _MODULE 1224 1225 #include <sys/module.h> 1226 1227 MODULE(MODULE_CLASS_DRIVER, fss, NULL); 1228 CFDRIVER_DECL(fss, DV_DISK, NULL); 1229 1230 static int 1231 fss_modcmd(modcmd_t cmd, void *arg) 1232 { 1233 int bmajor = -1, cmajor = -1, error = 0; 1234 1235 switch (cmd) { 1236 case MODULE_CMD_INIT: 1237 mutex_init(&fss_device_lock, MUTEX_DEFAULT, IPL_NONE); 1238 error = config_cfdriver_attach(&fss_cd); 1239 if (error) { 1240 mutex_destroy(&fss_device_lock); 1241 break; 1242 } 1243 error = config_cfattach_attach(fss_cd.cd_name, &fss_ca); 1244 if (error) { 1245 config_cfdriver_detach(&fss_cd); 1246 mutex_destroy(&fss_device_lock); 1247 break; 1248 } 1249 error = devsw_attach(fss_cd.cd_name, 1250 &fss_bdevsw, &bmajor, &fss_cdevsw, &cmajor); 1251 if (error == EEXIST) 1252 error = 0; 1253 if (error) { 1254 config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1255 config_cfdriver_detach(&fss_cd); 1256 mutex_destroy(&fss_device_lock); 1257 break; 1258 } 1259 break; 1260 1261 case MODULE_CMD_FINI: 1262 error = config_cfattach_detach(fss_cd.cd_name, &fss_ca); 1263 if (error) 1264 break; 1265 config_cfdriver_detach(&fss_cd); 1266 devsw_detach(&fss_bdevsw, &fss_cdevsw); 1267 mutex_destroy(&fss_device_lock); 1268 break; 1269 1270 default: 1271 error = ENOTTY; 1272 break; 1273 } 1274 1275 return error; 1276 } 1277 1278 #endif /* _MODULE */ 1279
Indexes created Wed Nov 11 00:26:22 UTC 2026