1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 1999, 2010, Oracle and/or its affiliates. All rights reserved.
  23  *
  24  * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
  25  */
  26 
  27 /*
  28  * lofi (loopback file) driver - allows you to attach a file to a device,
  29  * which can then be accessed through that device. The simple model is that
  30  * you tell lofi to open a file, and then use the block device you get as
  31  * you would any block device. lofi translates access to the block device
  32  * into I/O on the underlying file. This is mostly useful for
  33  * mounting images of filesystems.
  34  *
  35  * lofi is controlled through /dev/lofictl - this is the only device exported
  36  * during attach, and is minor number 0. lofiadm communicates with lofi through
  37  * ioctls on this device. When a file is attached to lofi, block and character
  38  * devices are exported in /dev/lofi and /dev/rlofi. Currently, these devices
  39  * are identified by their minor number, and the minor number is also used
  40  * as the name in /dev/lofi. If we ever decide to support virtual disks,
  41  * we'll have to divide the minor number space to identify fdisk partitions
  42  * and slices, and the name will then be the minor number shifted down a
  43  * few bits. Minor devices are tracked with state structures handled with
  44  * ddi_soft_state(9F) for simplicity.
  45  *
  46  * A file attached to lofi is opened when attached and not closed until
  47  * explicitly detached from lofi. This seems more sensible than deferring
  48  * the open until the /dev/lofi device is opened, for a number of reasons.
  49  * One is that any failure is likely to be noticed by the person (or script)
  50  * running lofiadm. Another is that it would be a security problem if the
  51  * file was replaced by another one after being added but before being opened.
  52  *
  53  * The only hard part about lofi is the ioctls. In order to support things
  54  * like 'newfs' on a lofi device, it needs to support certain disk ioctls.
  55  * So it has to fake disk geometry and partition information. More may need
  56  * to be faked if your favorite utility doesn't work and you think it should
  57  * (fdformat doesn't work because it really wants to know the type of floppy
  58  * controller to talk to, and that didn't seem easy to fake. Or possibly even
  59  * necessary, since we have mkfs_pcfs now).
  60  *
  61  * Normally, a lofi device cannot be detached if it is open (i.e. busy).  To
  62  * support simulation of hotplug events, an optional force flag is provided.
  63  * If a lofi device is open when a force detach is requested, then the
  64  * underlying file is closed and any subsequent operations return EIO.  When the
  65  * device is closed for the last time, it will be cleaned up at that time.  In
  66  * addition, the DKIOCSTATE ioctl will return DKIO_DEV_GONE when the device is
  67  * detached but not removed.
  68  *
  69  * Known problems:
  70  *
  71  *      UFS logging. Mounting a UFS filesystem image "logging"
  72  *      works for basic copy testing but wedges during a build of ON through
  73  *      that image. Some deadlock in lufs holding the log mutex and then
  74  *      getting stuck on a buf. So for now, don't do that.
  75  *
  76  *      Direct I/O. Since the filesystem data is being cached in the buffer
  77  *      cache, _and_ again in the underlying filesystem, it's tempting to
  78  *      enable direct I/O on the underlying file. Don't, because that deadlocks.
  79  *      I think to fix the cache-twice problem we might need filesystem support.
  80  *
  81  * Interesting things to do:
  82  *
  83  *      Allow multiple files for each device. A poor-man's metadisk, basically.
  84  *
  85  *      Pass-through ioctls on block devices. You can (though it's not
  86  *      documented), give lofi a block device as a file name. Then we shouldn't
  87  *      need to fake a geometry, however, it may be relevant if you're replacing
  88  *      metadisk, or using lofi to get crypto.
  89  *      It makes sense to do lofiadm -c aes -a /dev/dsk/c0t0d0s4 /dev/lofi/1
  90  *      and then in /etc/vfstab have an entry for /dev/lofi/1 as /export/home.
  91  *      In fact this even makes sense if you have lofi "above" metadisk.
  92  *
  93  * Encryption:
  94  *      Each lofi device can have its own symmetric key and cipher.
  95  *      They are passed to us by lofiadm(1m) in the correct format for use
  96  *      with the misc/kcf crypto_* routines.
  97  *
  98  *      Each block has its own IV, that is calculated in lofi_blk_mech(), based
  99  *      on the "master" key held in the lsp and the block number of the buffer.
 100  */
 101 
 102 #include <sys/types.h>
 103 #include <netinet/in.h>
 104 #include <sys/sysmacros.h>
 105 #include <sys/uio.h>
 106 #include <sys/kmem.h>
 107 #include <sys/cred.h>
 108 #include <sys/mman.h>
 109 #include <sys/errno.h>
 110 #include <sys/aio_req.h>
 111 #include <sys/stat.h>
 112 #include <sys/file.h>
 113 #include <sys/modctl.h>
 114 #include <sys/conf.h>
 115 #include <sys/debug.h>
 116 #include <sys/vnode.h>
 117 #include <sys/lofi.h>
 118 #include <sys/fcntl.h>
 119 #include <sys/pathname.h>
 120 #include <sys/filio.h>
 121 #include <sys/fdio.h>
 122 #include <sys/open.h>
 123 #include <sys/disp.h>
 124 #include <vm/seg_map.h>
 125 #include <sys/ddi.h>
 126 #include <sys/sunddi.h>
 127 #include <sys/zmod.h>
 128 #include <sys/id_space.h>
 129 #include <sys/mkdev.h>
 130 #include <sys/crypto/common.h>
 131 #include <sys/crypto/api.h>
 132 #include <sys/rctl.h>
 133 #include <LzmaDec.h>
 134 
 135 /*
 136  * The basis for CRYOFF is derived from usr/src/uts/common/sys/fs/ufs_fs.h.
 137  * Crypto metadata, if it exists, is located at the end of the boot block
 138  * (BBOFF + BBSIZE, which is SBOFF).  The super block and everything after
 139  * is offset by the size of the crypto metadata which is handled by
 140  * lsp->ls_crypto_offset.
 141  */
 142 #define CRYOFF  ((off_t)8192)
 143 
 144 #define NBLOCKS_PROP_NAME       "Nblocks"
 145 #define SIZE_PROP_NAME          "Size"
 146 #define ZONE_PROP_NAME          "zone"
 147 
 148 #define SETUP_C_DATA(cd, buf, len)              \
 149         (cd).cd_format = CRYPTO_DATA_RAW;       \
 150         (cd).cd_offset = 0;                     \
 151         (cd).cd_miscdata = NULL;                \
 152         (cd).cd_length = (len);                 \
 153         (cd).cd_raw.iov_base = (buf);           \
 154         (cd).cd_raw.iov_len = (len);
 155 
 156 #define UIO_CHECK(uio)  \
 157         if (((uio)->uio_loffset % DEV_BSIZE) != 0 || \
 158             ((uio)->uio_resid % DEV_BSIZE) != 0) { \
 159                 return (EINVAL); \
 160         }
 161 
 162 static dev_info_t *lofi_dip = NULL;
 163 static void *lofi_statep = NULL;
 164 static kmutex_t lofi_lock;              /* state lock */
 165 static id_space_t *lofi_minor_id;
 166 static list_t lofi_list;
 167 static zone_key_t lofi_zone_key;
 168 
 169 /*
 170  * Because lofi_taskq_nthreads limits the actual swamping of the device, the
 171  * maxalloc parameter (lofi_taskq_maxalloc) should be tuned conservatively
 172  * high.  If we want to be assured that the underlying device is always busy,
 173  * we must be sure that the number of bytes enqueued when the number of
 174  * enqueued tasks exceeds maxalloc is sufficient to keep the device busy for
 175  * the duration of the sleep time in taskq_ent_alloc().  That is, lofi should
 176  * set maxalloc to be the maximum throughput (in bytes per second) of the
 177  * underlying device divided by the minimum I/O size.  We assume a realistic
 178  * maximum throughput of one hundred megabytes per second; we set maxalloc on
 179  * the lofi task queue to be 104857600 divided by DEV_BSIZE.
 180  */
 181 static int lofi_taskq_maxalloc = 104857600 / DEV_BSIZE;
 182 static int lofi_taskq_nthreads = 4;     /* # of taskq threads per device */
 183 
 184 const char lofi_crypto_magic[6] = LOFI_CRYPTO_MAGIC;
 185 
 186 /*
 187  * To avoid decompressing data in a compressed segment multiple times
 188  * when accessing small parts of a segment's data, we cache and reuse
 189  * the uncompressed segment's data.
 190  *
 191  * A single cached segment is sufficient to avoid lots of duplicate
 192  * segment decompress operations. A small cache size also reduces the
 193  * memory footprint.
 194  *
 195  * lofi_max_comp_cache is the maximum number of decompressed data segments
 196  * cached for each compressed lofi image. It can be set to 0 to disable
 197  * caching.
 198  */
 199 
 200 uint32_t lofi_max_comp_cache = 1;
 201 
 202 static int gzip_decompress(void *src, size_t srclen, void *dst,
 203         size_t *destlen, int level);
 204 
 205 static int lzma_decompress(void *src, size_t srclen, void *dst,
 206         size_t *dstlen, int level);
 207 
 208 lofi_compress_info_t lofi_compress_table[LOFI_COMPRESS_FUNCTIONS] = {
 209         {gzip_decompress,       NULL,   6,      "gzip"}, /* default */
 210         {gzip_decompress,       NULL,   6,      "gzip-6"},
 211         {gzip_decompress,       NULL,   9,      "gzip-9"},
 212         {lzma_decompress,       NULL,   0,      "lzma"}
 213 };
 214 
 215 /*ARGSUSED*/
 216 static void
 217 *SzAlloc(void *p, size_t size)
 218 {
 219         return (kmem_alloc(size, KM_SLEEP));
 220 }
 221 
 222 /*ARGSUSED*/
 223 static void
 224 SzFree(void *p, void *address, size_t size)
 225 {
 226         kmem_free(address, size);
 227 }
 228 
 229 static ISzAlloc g_Alloc = { SzAlloc, SzFree };
 230 
 231 /*
 232  * Free data referenced by the linked list of cached uncompressed
 233  * segments.
 234  */
 235 static void
 236 lofi_free_comp_cache(struct lofi_state *lsp)
 237 {
 238         struct lofi_comp_cache *lc;
 239 
 240         while ((lc = list_remove_head(&lsp->ls_comp_cache)) != NULL) {
 241                 kmem_free(lc->lc_data, lsp->ls_uncomp_seg_sz);
 242                 kmem_free(lc, sizeof (struct lofi_comp_cache));
 243                 lsp->ls_comp_cache_count--;
 244         }
 245         ASSERT(lsp->ls_comp_cache_count == 0);
 246 }
 247 
 248 static int
 249 is_opened(struct lofi_state *lsp)
 250 {
 251         ASSERT(MUTEX_HELD(&lofi_lock));
 252         return (lsp->ls_chr_open || lsp->ls_blk_open || lsp->ls_lyr_open_count);
 253 }
 254 
 255 static int
 256 mark_opened(struct lofi_state *lsp, int otyp)
 257 {
 258         ASSERT(MUTEX_HELD(&lofi_lock));
 259         switch (otyp) {
 260         case OTYP_CHR:
 261                 lsp->ls_chr_open = 1;
 262                 break;
 263         case OTYP_BLK:
 264                 lsp->ls_blk_open = 1;
 265                 break;
 266         case OTYP_LYR:
 267                 lsp->ls_lyr_open_count++;
 268                 break;
 269         default:
 270                 return (-1);
 271         }
 272         return (0);
 273 }
 274 
 275 static void
 276 mark_closed(struct lofi_state *lsp, int otyp)
 277 {
 278         ASSERT(MUTEX_HELD(&lofi_lock));
 279         switch (otyp) {
 280         case OTYP_CHR:
 281                 lsp->ls_chr_open = 0;
 282                 break;
 283         case OTYP_BLK:
 284                 lsp->ls_blk_open = 0;
 285                 break;
 286         case OTYP_LYR:
 287                 lsp->ls_lyr_open_count--;
 288                 break;
 289         default:
 290                 break;
 291         }
 292 }
 293 
 294 static void
 295 lofi_free_crypto(struct lofi_state *lsp)
 296 {
 297         ASSERT(MUTEX_HELD(&lofi_lock));
 298 
 299         if (lsp->ls_crypto_enabled) {
 300                 /*
 301                  * Clean up the crypto state so that it doesn't hang around
 302                  * in memory after we are done with it.
 303                  */
 304                 if (lsp->ls_key.ck_data != NULL) {
 305                         bzero(lsp->ls_key.ck_data,
 306                             CRYPTO_BITS2BYTES(lsp->ls_key.ck_length));
 307                         kmem_free(lsp->ls_key.ck_data,
 308                             CRYPTO_BITS2BYTES(lsp->ls_key.ck_length));
 309                         lsp->ls_key.ck_data = NULL;
 310                         lsp->ls_key.ck_length = 0;
 311                 }
 312 
 313                 if (lsp->ls_mech.cm_param != NULL) {
 314                         kmem_free(lsp->ls_mech.cm_param,
 315                             lsp->ls_mech.cm_param_len);
 316                         lsp->ls_mech.cm_param = NULL;
 317                         lsp->ls_mech.cm_param_len = 0;
 318                 }
 319 
 320                 if (lsp->ls_iv_mech.cm_param != NULL) {
 321                         kmem_free(lsp->ls_iv_mech.cm_param,
 322                             lsp->ls_iv_mech.cm_param_len);
 323                         lsp->ls_iv_mech.cm_param = NULL;
 324                         lsp->ls_iv_mech.cm_param_len = 0;
 325                 }
 326 
 327                 mutex_destroy(&lsp->ls_crypto_lock);
 328         }
 329 }
 330 
 331 static void
 332 lofi_destroy(struct lofi_state *lsp, cred_t *credp)
 333 {
 334         minor_t minor = getminor(lsp->ls_dev);
 335         int i;
 336 
 337         ASSERT(MUTEX_HELD(&lofi_lock));
 338 
 339         list_remove(&lofi_list, lsp);
 340 
 341         lofi_free_crypto(lsp);
 342 
 343         /*
 344          * Free pre-allocated compressed buffers
 345          */
 346         if (lsp->ls_comp_bufs != NULL) {
 347                 for (i = 0; i < lofi_taskq_nthreads; i++) {
 348                         if (lsp->ls_comp_bufs[i].bufsize > 0)
 349                                 kmem_free(lsp->ls_comp_bufs[i].buf,
 350                                     lsp->ls_comp_bufs[i].bufsize);
 351                 }
 352                 kmem_free(lsp->ls_comp_bufs,
 353                     sizeof (struct compbuf) * lofi_taskq_nthreads);
 354         }
 355 
 356         (void) VOP_CLOSE(lsp->ls_vp, lsp->ls_openflag,
 357             1, 0, credp, NULL);
 358         VN_RELE(lsp->ls_vp);
 359         if (lsp->ls_stacked_vp != lsp->ls_vp)
 360                 VN_RELE(lsp->ls_stacked_vp);
 361 
 362         taskq_destroy(lsp->ls_taskq);
 363 
 364         if (lsp->ls_kstat != NULL)
 365                 kstat_delete(lsp->ls_kstat);
 366 
 367         /*
 368          * Free cached decompressed segment data
 369          */
 370         lofi_free_comp_cache(lsp);
 371         list_destroy(&lsp->ls_comp_cache);
 372 
 373         if (lsp->ls_uncomp_seg_sz > 0) {
 374                 kmem_free(lsp->ls_comp_index_data, lsp->ls_comp_index_data_sz);
 375                 lsp->ls_uncomp_seg_sz = 0;
 376         }
 377 
 378         rctl_decr_lofi(lsp->ls_zone.zref_zone, 1);
 379         zone_rele_ref(&lsp->ls_zone, ZONE_REF_LOFI);
 380 
 381         mutex_destroy(&lsp->ls_comp_cache_lock);
 382         mutex_destroy(&lsp->ls_comp_bufs_lock);
 383         mutex_destroy(&lsp->ls_kstat_lock);
 384         mutex_destroy(&lsp->ls_vp_lock);
 385 
 386         ASSERT(ddi_get_soft_state(lofi_statep, minor) == lsp);
 387         ddi_soft_state_free(lofi_statep, minor);
 388         id_free(lofi_minor_id, minor);
 389 }
 390 
 391 static void
 392 lofi_free_dev(dev_t dev)
 393 {
 394         minor_t minor = getminor(dev);
 395         char namebuf[50];
 396 
 397         ASSERT(MUTEX_HELD(&lofi_lock));
 398 
 399         (void) ddi_prop_remove(dev, lofi_dip, ZONE_PROP_NAME);
 400         (void) ddi_prop_remove(dev, lofi_dip, SIZE_PROP_NAME);
 401         (void) ddi_prop_remove(dev, lofi_dip, NBLOCKS_PROP_NAME);
 402 
 403         (void) snprintf(namebuf, sizeof (namebuf), "%d", minor);
 404         ddi_remove_minor_node(lofi_dip, namebuf);
 405         (void) snprintf(namebuf, sizeof (namebuf), "%d,raw", minor);
 406         ddi_remove_minor_node(lofi_dip, namebuf);
 407 }
 408 
 409 /*ARGSUSED*/
 410 static void
 411 lofi_zone_shutdown(zoneid_t zoneid, void *arg)
 412 {
 413         struct lofi_state *lsp;
 414         struct lofi_state *next;
 415 
 416         mutex_enter(&lofi_lock);
 417 
 418         for (lsp = list_head(&lofi_list); lsp != NULL; lsp = next) {
 419 
 420                 /* lofi_destroy() frees lsp */
 421                 next = list_next(&lofi_list, lsp);
 422 
 423                 if (lsp->ls_zone.zref_zone->zone_id != zoneid)
 424                         continue;
 425 
 426                 /*
 427                  * No in-zone processes are running, but something has this
 428                  * open.  It's either a global zone process, or a lofi
 429                  * mount.  In either case we set ls_cleanup so the last
 430                  * user destroys the device.
 431                  */
 432                 if (is_opened(lsp)) {
 433                         lsp->ls_cleanup = 1;
 434                 } else {
 435                         lofi_free_dev(lsp->ls_dev);
 436                         lofi_destroy(lsp, kcred);
 437                 }
 438         }
 439 
 440         mutex_exit(&lofi_lock);
 441 }
 442 
 443 /*ARGSUSED*/
 444 static int
 445 lofi_open(dev_t *devp, int flag, int otyp, struct cred *credp)
 446 {
 447         minor_t minor;
 448         struct lofi_state *lsp;
 449 
 450         /*
 451          * lofiadm -a /dev/lofi/1 gets us here.
 452          */
 453         if (mutex_owner(&lofi_lock) == curthread)
 454                 return (EINVAL);
 455 
 456         mutex_enter(&lofi_lock);
 457 
 458         minor = getminor(*devp);
 459 
 460         /* master control device */
 461         if (minor == 0) {
 462                 mutex_exit(&lofi_lock);
 463                 return (0);
 464         }
 465 
 466         /* otherwise, the mapping should already exist */
 467         lsp = ddi_get_soft_state(lofi_statep, minor);
 468         if (lsp == NULL) {
 469                 mutex_exit(&lofi_lock);
 470                 return (EINVAL);
 471         }
 472 
 473         if (lsp->ls_vp == NULL) {
 474                 mutex_exit(&lofi_lock);
 475                 return (ENXIO);
 476         }
 477 
 478         if (lsp->ls_readonly && (flag & FWRITE)) {
 479                 mutex_exit(&lofi_lock);
 480                 return (EROFS);
 481         }
 482 
 483         if (mark_opened(lsp, otyp) == -1) {
 484                 mutex_exit(&lofi_lock);
 485                 return (EINVAL);
 486         }
 487 
 488         mutex_exit(&lofi_lock);
 489         return (0);
 490 }
 491 
 492 /*ARGSUSED*/
 493 static int
 494 lofi_close(dev_t dev, int flag, int otyp, struct cred *credp)
 495 {
 496         minor_t minor;
 497         struct lofi_state *lsp;
 498 
 499         mutex_enter(&lofi_lock);
 500         minor = getminor(dev);
 501         lsp = ddi_get_soft_state(lofi_statep, minor);
 502         if (lsp == NULL) {
 503                 mutex_exit(&lofi_lock);
 504                 return (EINVAL);
 505         }
 506 
 507         if (minor == 0) {
 508                 mutex_exit(&lofi_lock);
 509                 return (0);
 510         }
 511 
 512         mark_closed(lsp, otyp);
 513 
 514         /*
 515          * If we forcibly closed the underlying device (li_force), or
 516          * asked for cleanup (li_cleanup), finish up if we're the last
 517          * out of the door.
 518          */
 519         if (!is_opened(lsp) && (lsp->ls_cleanup || lsp->ls_vp == NULL)) {
 520                 lofi_free_dev(lsp->ls_dev);
 521                 lofi_destroy(lsp, credp);
 522         }
 523 
 524         mutex_exit(&lofi_lock);
 525         return (0);
 526 }
 527 
 528 /*
 529  * Sets the mechanism's initialization vector (IV) if one is needed.
 530  * The IV is computed from the data block number.  lsp->ls_mech is
 531  * altered so that:
 532  *      lsp->ls_mech.cm_param_len is set to the IV len.
 533  *      lsp->ls_mech.cm_param is set to the IV.
 534  */
 535 static int
 536 lofi_blk_mech(struct lofi_state *lsp, longlong_t lblkno)
 537 {
 538         int     ret;
 539         crypto_data_t cdata;
 540         char    *iv;
 541         size_t  iv_len;
 542         size_t  min;
 543         void    *data;
 544         size_t  datasz;
 545 
 546         ASSERT(MUTEX_HELD(&lsp->ls_crypto_lock));
 547 
 548         if (lsp == NULL)
 549                 return (CRYPTO_DEVICE_ERROR);
 550 
 551         /* lsp->ls_mech.cm_param{_len} has already been set for static iv */
 552         if (lsp->ls_iv_type == IVM_NONE) {
 553                 return (CRYPTO_SUCCESS);
 554         }
 555 
 556         /*
 557          * if kmem already alloced from previous call and it's the same size
 558          * we need now, just recycle it; allocate new kmem only if we have to
 559          */
 560         if (lsp->ls_mech.cm_param == NULL ||
 561             lsp->ls_mech.cm_param_len != lsp->ls_iv_len) {
 562                 iv_len = lsp->ls_iv_len;
 563                 iv = kmem_zalloc(iv_len, KM_SLEEP);
 564         } else {
 565                 iv_len = lsp->ls_mech.cm_param_len;
 566                 iv = lsp->ls_mech.cm_param;
 567                 bzero(iv, iv_len);
 568         }
 569 
 570         switch (lsp->ls_iv_type) {
 571         case IVM_ENC_BLKNO:
 572                 /* iv is not static, lblkno changes each time */
 573                 data = &lblkno;
 574                 datasz = sizeof (lblkno);
 575                 break;
 576         default:
 577                 data = 0;
 578                 datasz = 0;
 579                 break;
 580         }
 581 
 582         /*
 583          * write blkno into the iv buffer padded on the left in case
 584          * blkno ever grows bigger than its current longlong_t size
 585          * or a variation other than blkno is used for the iv data
 586          */
 587         min = MIN(datasz, iv_len);
 588         bcopy(data, iv + (iv_len - min), min);
 589 
 590         /* encrypt the data in-place to get the IV */
 591         SETUP_C_DATA(cdata, iv, iv_len);
 592 
 593         ret = crypto_encrypt(&lsp->ls_iv_mech, &cdata, &lsp->ls_key,
 594             NULL, NULL, NULL);
 595         if (ret != CRYPTO_SUCCESS) {
 596                 cmn_err(CE_WARN, "failed to create iv for block %lld: (0x%x)",
 597                     lblkno, ret);
 598                 if (lsp->ls_mech.cm_param != iv)
 599                         kmem_free(iv, iv_len);
 600 
 601                 return (ret);
 602         }
 603 
 604         /* clean up the iv from the last computation */
 605         if (lsp->ls_mech.cm_param != NULL && lsp->ls_mech.cm_param != iv)
 606                 kmem_free(lsp->ls_mech.cm_param, lsp->ls_mech.cm_param_len);
 607 
 608         lsp->ls_mech.cm_param_len = iv_len;
 609         lsp->ls_mech.cm_param = iv;
 610 
 611         return (CRYPTO_SUCCESS);
 612 }
 613 
 614 /*
 615  * Performs encryption and decryption of a chunk of data of size "len",
 616  * one DEV_BSIZE block at a time.  "len" is assumed to be a multiple of
 617  * DEV_BSIZE.
 618  */
 619 static int
 620 lofi_crypto(struct lofi_state *lsp, struct buf *bp, caddr_t plaintext,
 621     caddr_t ciphertext, size_t len, boolean_t op_encrypt)
 622 {
 623         crypto_data_t cdata;
 624         crypto_data_t wdata;
 625         int ret;
 626         longlong_t lblkno = bp->b_lblkno;
 627 
 628         mutex_enter(&lsp->ls_crypto_lock);
 629 
 630         /*
 631          * though we could encrypt/decrypt entire "len" chunk of data, we need
 632          * to break it into DEV_BSIZE pieces to capture blkno incrementing
 633          */
 634         SETUP_C_DATA(cdata, plaintext, len);
 635         cdata.cd_length = DEV_BSIZE;
 636         if (ciphertext != NULL) {               /* not in-place crypto */
 637                 SETUP_C_DATA(wdata, ciphertext, len);
 638                 wdata.cd_length = DEV_BSIZE;
 639         }
 640 
 641         do {
 642                 ret = lofi_blk_mech(lsp, lblkno);
 643                 if (ret != CRYPTO_SUCCESS)
 644                         continue;
 645 
 646                 if (op_encrypt) {
 647                         ret = crypto_encrypt(&lsp->ls_mech, &cdata,
 648                             &lsp->ls_key, NULL,
 649                             ((ciphertext != NULL) ? &wdata : NULL), NULL);
 650                 } else {
 651                         ret = crypto_decrypt(&lsp->ls_mech, &cdata,
 652                             &lsp->ls_key, NULL,
 653                             ((ciphertext != NULL) ? &wdata : NULL), NULL);
 654                 }
 655 
 656                 cdata.cd_offset += DEV_BSIZE;
 657                 if (ciphertext != NULL)
 658                         wdata.cd_offset += DEV_BSIZE;
 659                 lblkno++;
 660         } while (ret == CRYPTO_SUCCESS && cdata.cd_offset < len);
 661 
 662         mutex_exit(&lsp->ls_crypto_lock);
 663 
 664         if (ret != CRYPTO_SUCCESS) {
 665                 cmn_err(CE_WARN, "%s failed for block %lld:  (0x%x)",
 666                     op_encrypt ? "crypto_encrypt()" : "crypto_decrypt()",
 667                     lblkno, ret);
 668         }
 669 
 670         return (ret);
 671 }
 672 
 673 #define RDWR_RAW        1
 674 #define RDWR_BCOPY      2
 675 
 676 static int
 677 lofi_rdwr(caddr_t bufaddr, offset_t offset, struct buf *bp,
 678     struct lofi_state *lsp, size_t len, int method, caddr_t bcopy_locn)
 679 {
 680         ssize_t resid;
 681         int isread;
 682         int error;
 683 
 684         /*
 685          * Handles reads/writes for both plain and encrypted lofi
 686          * Note:  offset is already shifted by lsp->ls_crypto_offset
 687          * when it gets here.
 688          */
 689 
 690         isread = bp->b_flags & B_READ;
 691         if (isread) {
 692                 if (method == RDWR_BCOPY) {
 693                         /* DO NOT update bp->b_resid for bcopy */
 694                         bcopy(bcopy_locn, bufaddr, len);
 695                         error = 0;
 696                 } else {                /* RDWR_RAW */
 697                         error = vn_rdwr(UIO_READ, lsp->ls_vp, bufaddr, len,
 698                             offset, UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred,
 699                             &resid);
 700                         bp->b_resid = resid;
 701                 }
 702                 if (lsp->ls_crypto_enabled && error == 0) {
 703                         if (lofi_crypto(lsp, bp, bufaddr, NULL, len,
 704                             B_FALSE) != CRYPTO_SUCCESS) {
 705                                 /*
 706                                  * XXX: original code didn't set residual
 707                                  * back to len because no error was expected
 708                                  * from bcopy() if encryption is not enabled
 709                                  */
 710                                 if (method != RDWR_BCOPY)
 711                                         bp->b_resid = len;
 712                                 error = EIO;
 713                         }
 714                 }
 715                 return (error);
 716         } else {
 717                 void *iobuf = bufaddr;
 718 
 719                 if (lsp->ls_crypto_enabled) {
 720                         /* don't do in-place crypto to keep bufaddr intact */
 721                         iobuf = kmem_alloc(len, KM_SLEEP);
 722                         if (lofi_crypto(lsp, bp, bufaddr, iobuf, len,
 723                             B_TRUE) != CRYPTO_SUCCESS) {
 724                                 kmem_free(iobuf, len);
 725                                 if (method != RDWR_BCOPY)
 726                                         bp->b_resid = len;
 727                                 return (EIO);
 728                         }
 729                 }
 730                 if (method == RDWR_BCOPY) {
 731                         /* DO NOT update bp->b_resid for bcopy */
 732                         bcopy(iobuf, bcopy_locn, len);
 733                         error = 0;
 734                 } else {                /* RDWR_RAW */
 735                         error = vn_rdwr(UIO_WRITE, lsp->ls_vp, iobuf, len,
 736                             offset, UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred,
 737                             &resid);
 738                         bp->b_resid = resid;
 739                 }
 740                 if (lsp->ls_crypto_enabled) {
 741                         kmem_free(iobuf, len);
 742                 }
 743                 return (error);
 744         }
 745 }
 746 
 747 static int
 748 lofi_mapped_rdwr(caddr_t bufaddr, offset_t offset, struct buf *bp,
 749     struct lofi_state *lsp)
 750 {
 751         int error;
 752         offset_t alignedoffset, mapoffset;
 753         size_t  xfersize;
 754         int     isread;
 755         int     smflags;
 756         caddr_t mapaddr;
 757         size_t  len;
 758         enum seg_rw srw;
 759         int     save_error;
 760 
 761         /*
 762          * Note:  offset is already shifted by lsp->ls_crypto_offset
 763          * when it gets here.
 764          */
 765         if (lsp->ls_crypto_enabled)
 766                 ASSERT(lsp->ls_vp_comp_size == lsp->ls_vp_size);
 767 
 768         /*
 769          * segmap always gives us an 8K (MAXBSIZE) chunk, aligned on
 770          * an 8K boundary, but the buf transfer address may not be
 771          * aligned on more than a 512-byte boundary (we don't enforce
 772          * that even though we could). This matters since the initial
 773          * part of the transfer may not start at offset 0 within the
 774          * segmap'd chunk. So we have to compensate for that with
 775          * 'mapoffset'. Subsequent chunks always start off at the
 776          * beginning, and the last is capped by b_resid
 777          *
 778          * Visually, where "|" represents page map boundaries:
 779          *   alignedoffset (mapaddr begins at this segmap boundary)
 780          *    |   offset (from beginning of file)
 781          *    |    |       len
 782          *    v    v        v
 783          * ===|====X========|====...======|========X====|====
 784          *         /-------------...---------------/
 785          *              ^ bp->b_bcount/bp->b_resid at start
 786          *    /----/--------/----...------/--------/
 787          *      ^       ^       ^   ^           ^
 788          *      |       |       |   |           nth xfersize (<= MAXBSIZE)
 789          *      |       |       2nd thru n-1st xfersize (= MAXBSIZE)
 790          *      |       1st xfersize (<= MAXBSIZE)
 791          *    mapoffset (offset into 1st segmap, non-0 1st time, 0 thereafter)
 792          *
 793          * Notes: "alignedoffset" is "offset" rounded down to nearest
 794          * MAXBSIZE boundary.  "len" is next page boundary of size
 795          * PAGESIZE after "alignedoffset".
 796          */
 797         mapoffset = offset & MAXBOFFSET;
 798         alignedoffset = offset - mapoffset;
 799         bp->b_resid = bp->b_bcount;
 800         isread = bp->b_flags & B_READ;
 801         srw = isread ? S_READ : S_WRITE;
 802         do {
 803                 xfersize = MIN(lsp->ls_vp_comp_size - offset,
 804                     MIN(MAXBSIZE - mapoffset, bp->b_resid));
 805                 len = roundup(mapoffset + xfersize, PAGESIZE);
 806                 mapaddr = segmap_getmapflt(segkmap, lsp->ls_vp,
 807                     alignedoffset, MAXBSIZE, 1, srw);
 808                 /*
 809                  * Now fault in the pages. This lets us check
 810                  * for errors before we reference mapaddr and
 811                  * try to resolve the fault in bcopy (which would
 812                  * panic instead). And this can easily happen,
 813                  * particularly if you've lofi'd a file over NFS
 814                  * and someone deletes the file on the server.
 815                  */
 816                 error = segmap_fault(kas.a_hat, segkmap, mapaddr,
 817                     len, F_SOFTLOCK, srw);
 818                 if (error) {
 819                         (void) segmap_release(segkmap, mapaddr, 0);
 820                         if (FC_CODE(error) == FC_OBJERR)
 821                                 error = FC_ERRNO(error);
 822                         else
 823                                 error = EIO;
 824                         break;
 825                 }
 826                 /* error may be non-zero for encrypted lofi */
 827                 error = lofi_rdwr(bufaddr, 0, bp, lsp, xfersize,
 828                     RDWR_BCOPY, mapaddr + mapoffset);
 829                 if (error == 0) {
 830                         bp->b_resid -= xfersize;
 831                         bufaddr += xfersize;
 832                         offset += xfersize;
 833                 }
 834                 smflags = 0;
 835                 if (isread) {
 836                         smflags |= SM_FREE;
 837                         /*
 838                          * If we're reading an entire page starting
 839                          * at a page boundary, there's a good chance
 840                          * we won't need it again. Put it on the
 841                          * head of the freelist.
 842                          */
 843                         if (mapoffset == 0 && xfersize == MAXBSIZE)
 844                                 smflags |= SM_DONTNEED;
 845                 } else {
 846                         /*
 847                          * Write back good pages, it is okay to
 848                          * always release asynchronous here as we'll
 849                          * follow with VOP_FSYNC for B_SYNC buffers.
 850                          */
 851                         if (error == 0)
 852                                 smflags |= SM_WRITE | SM_ASYNC;
 853                 }
 854                 (void) segmap_fault(kas.a_hat, segkmap, mapaddr,
 855                     len, F_SOFTUNLOCK, srw);
 856                 save_error = segmap_release(segkmap, mapaddr, smflags);
 857                 if (error == 0)
 858                         error = save_error;
 859                 /* only the first map may start partial */
 860                 mapoffset = 0;
 861                 alignedoffset += MAXBSIZE;
 862         } while ((error == 0) && (bp->b_resid > 0) &&
 863             (offset < lsp->ls_vp_comp_size));
 864 
 865         return (error);
 866 }
 867 
 868 /*
 869  * Check if segment seg_index is present in the decompressed segment
 870  * data cache.
 871  *
 872  * Returns a pointer to the decompressed segment data cache entry if
 873  * found, and NULL when decompressed data for this segment is not yet
 874  * cached.
 875  */
 876 static struct lofi_comp_cache *
 877 lofi_find_comp_data(struct lofi_state *lsp, uint64_t seg_index)
 878 {
 879         struct lofi_comp_cache *lc;
 880 
 881         ASSERT(MUTEX_HELD(&lsp->ls_comp_cache_lock));
 882 
 883         for (lc = list_head(&lsp->ls_comp_cache); lc != NULL;
 884             lc = list_next(&lsp->ls_comp_cache, lc)) {
 885                 if (lc->lc_index == seg_index) {
 886                         /*
 887                          * Decompressed segment data was found in the
 888                          * cache.
 889                          *
 890                          * The cache uses an LRU replacement strategy;
 891                          * move the entry to head of list.
 892                          */
 893                         list_remove(&lsp->ls_comp_cache, lc);
 894                         list_insert_head(&lsp->ls_comp_cache, lc);
 895                         return (lc);
 896                 }
 897         }
 898         return (NULL);
 899 }
 900 
 901 /*
 902  * Add the data for a decompressed segment at segment index
 903  * seg_index to the cache of the decompressed segments.
 904  *
 905  * Returns a pointer to the cache element structure in case
 906  * the data was added to the cache; returns NULL when the data
 907  * wasn't cached.
 908  */
 909 static struct lofi_comp_cache *
 910 lofi_add_comp_data(struct lofi_state *lsp, uint64_t seg_index,
 911     uchar_t *data)
 912 {
 913         struct lofi_comp_cache *lc;
 914 
 915         ASSERT(MUTEX_HELD(&lsp->ls_comp_cache_lock));
 916 
 917         while (lsp->ls_comp_cache_count > lofi_max_comp_cache) {
 918                 lc = list_remove_tail(&lsp->ls_comp_cache);
 919                 ASSERT(lc != NULL);
 920                 kmem_free(lc->lc_data, lsp->ls_uncomp_seg_sz);
 921                 kmem_free(lc, sizeof (struct lofi_comp_cache));
 922                 lsp->ls_comp_cache_count--;
 923         }
 924 
 925         /*
 926          * Do not cache when disabled by tunable variable
 927          */
 928         if (lofi_max_comp_cache == 0)
 929                 return (NULL);
 930 
 931         /*
 932          * When the cache has not yet reached the maximum allowed
 933          * number of segments, allocate a new cache element.
 934          * Otherwise the cache is full; reuse the last list element
 935          * (LRU) for caching the decompressed segment data.
 936          *
 937          * The cache element for the new decompressed segment data is
 938          * added to the head of the list.
 939          */
 940         if (lsp->ls_comp_cache_count < lofi_max_comp_cache) {
 941                 lc = kmem_alloc(sizeof (struct lofi_comp_cache), KM_SLEEP);
 942                 lc->lc_data = NULL;
 943                 list_insert_head(&lsp->ls_comp_cache, lc);
 944                 lsp->ls_comp_cache_count++;
 945         } else {
 946                 lc = list_remove_tail(&lsp->ls_comp_cache);
 947                 if (lc == NULL)
 948                         return (NULL);
 949                 list_insert_head(&lsp->ls_comp_cache, lc);
 950         }
 951 
 952         /*
 953          * Free old uncompressed segment data when reusing a cache
 954          * entry.
 955          */
 956         if (lc->lc_data != NULL)
 957                 kmem_free(lc->lc_data, lsp->ls_uncomp_seg_sz);
 958 
 959         lc->lc_data = data;
 960         lc->lc_index = seg_index;
 961         return (lc);
 962 }
 963 
 964 
 965 /*ARGSUSED*/
 966 static int
 967 gzip_decompress(void *src, size_t srclen, void *dst,
 968     size_t *dstlen, int level)
 969 {
 970         ASSERT(*dstlen >= srclen);
 971 
 972         if (z_uncompress(dst, dstlen, src, srclen) != Z_OK)
 973                 return (-1);
 974         return (0);
 975 }
 976 
 977 #define LZMA_HEADER_SIZE        (LZMA_PROPS_SIZE + 8)
 978 /*ARGSUSED*/
 979 static int
 980 lzma_decompress(void *src, size_t srclen, void *dst,
 981         size_t *dstlen, int level)
 982 {
 983         size_t insizepure;
 984         void *actual_src;
 985         ELzmaStatus status;
 986 
 987         insizepure = srclen - LZMA_HEADER_SIZE;
 988         actual_src = (void *)((Byte *)src + LZMA_HEADER_SIZE);
 989 
 990         if (LzmaDecode((Byte *)dst, (size_t *)dstlen,
 991             (const Byte *)actual_src, &insizepure,
 992             (const Byte *)src, LZMA_PROPS_SIZE, LZMA_FINISH_ANY, &status,
 993             &g_Alloc) != SZ_OK) {
 994                 return (-1);
 995         }
 996         return (0);
 997 }
 998 
 999 /*
1000  * This is basically what strategy used to be before we found we
1001  * needed task queues.
1002  */
1003 static void
1004 lofi_strategy_task(void *arg)
1005 {
1006         struct buf *bp = (struct buf *)arg;
1007         int error;
1008         int syncflag = 0;
1009         struct lofi_state *lsp;
1010         offset_t offset;
1011         caddr_t bufaddr;
1012         size_t  len;
1013         size_t  xfersize;
1014         boolean_t bufinited = B_FALSE;
1015 
1016         lsp = ddi_get_soft_state(lofi_statep, getminor(bp->b_edev));
1017         if (lsp == NULL) {
1018                 error = ENXIO;
1019                 goto errout;
1020         }
1021         if (lsp->ls_kstat) {
1022                 mutex_enter(lsp->ls_kstat->ks_lock);
1023                 kstat_waitq_to_runq(KSTAT_IO_PTR(lsp->ls_kstat));
1024                 mutex_exit(lsp->ls_kstat->ks_lock);
1025         }
1026         bp_mapin(bp);
1027         bufaddr = bp->b_un.b_addr;
1028         offset = bp->b_lblkno * DEV_BSIZE;   /* offset within file */
1029         if (lsp->ls_crypto_enabled) {
1030                 /* encrypted data really begins after crypto header */
1031                 offset += lsp->ls_crypto_offset;
1032         }
1033         len = bp->b_bcount;
1034         bufinited = B_TRUE;
1035 
1036         if (lsp->ls_vp == NULL || lsp->ls_vp_closereq) {
1037                 error = EIO;
1038                 goto errout;
1039         }
1040 
1041         /*
1042          * If we're writing and the buffer was not B_ASYNC
1043          * we'll follow up with a VOP_FSYNC() to force any
1044          * asynchronous I/O to stable storage.
1045          */
1046         if (!(bp->b_flags & B_READ) && !(bp->b_flags & B_ASYNC))
1047                 syncflag = FSYNC;
1048 
1049         /*
1050          * We used to always use vn_rdwr here, but we cannot do that because
1051          * we might decide to read or write from the the underlying
1052          * file during this call, which would be a deadlock because
1053          * we have the rw_lock. So instead we page, unless it's not
1054          * mapable or it's a character device or it's an encrypted lofi.
1055          */
1056         if ((lsp->ls_vp->v_flag & VNOMAP) || (lsp->ls_vp->v_type == VCHR) ||
1057             lsp->ls_crypto_enabled) {
1058                 error = lofi_rdwr(bufaddr, offset, bp, lsp, len, RDWR_RAW,
1059                     NULL);
1060         } else if (lsp->ls_uncomp_seg_sz == 0) {
1061                 error = lofi_mapped_rdwr(bufaddr, offset, bp, lsp);
1062         } else {
1063                 uchar_t *compressed_seg = NULL, *cmpbuf;
1064                 uchar_t *uncompressed_seg = NULL;
1065                 lofi_compress_info_t *li;
1066                 size_t oblkcount;
1067                 ulong_t seglen;
1068                 uint64_t sblkno, eblkno, cmpbytes;
1069                 uint64_t uncompressed_seg_index;
1070                 struct lofi_comp_cache *lc;
1071                 offset_t sblkoff, eblkoff;
1072                 u_offset_t salign, ealign;
1073                 u_offset_t sdiff;
1074                 uint32_t comp_data_sz;
1075                 uint64_t i;
1076                 int j;
1077 
1078                 /*
1079                  * From here on we're dealing primarily with compressed files
1080                  */
1081                 ASSERT(!lsp->ls_crypto_enabled);
1082 
1083                 /*
1084                  * Compressed files can only be read from and
1085                  * not written to
1086                  */
1087                 if (!(bp->b_flags & B_READ)) {
1088                         bp->b_resid = bp->b_bcount;
1089                         error = EROFS;
1090                         goto done;
1091                 }
1092 
1093                 ASSERT(lsp->ls_comp_algorithm_index >= 0);
1094                 li = &lofi_compress_table[lsp->ls_comp_algorithm_index];
1095                 /*
1096                  * Compute starting and ending compressed segment numbers
1097                  * We use only bitwise operations avoiding division and
1098                  * modulus because we enforce the compression segment size
1099                  * to a power of 2
1100                  */
1101                 sblkno = offset >> lsp->ls_comp_seg_shift;
1102                 sblkoff = offset & (lsp->ls_uncomp_seg_sz - 1);
1103                 eblkno = (offset + bp->b_bcount) >> lsp->ls_comp_seg_shift;
1104                 eblkoff = (offset + bp->b_bcount) & (lsp->ls_uncomp_seg_sz - 1);
1105 
1106                 /*
1107                  * Check the decompressed segment cache.
1108                  *
1109                  * The cache is used only when the requested data
1110                  * is within a segment. Requests that cross
1111                  * segment boundaries bypass the cache.
1112                  */
1113                 if (sblkno == eblkno ||
1114                     (sblkno + 1 == eblkno && eblkoff == 0)) {
1115                         /*
1116                          * Request doesn't cross a segment boundary,
1117                          * now check the cache.
1118                          */
1119                         mutex_enter(&lsp->ls_comp_cache_lock);
1120                         lc = lofi_find_comp_data(lsp, sblkno);
1121                         if (lc != NULL) {
1122                                 /*
1123                                  * We've found the decompressed segment
1124                                  * data in the cache; reuse it.
1125                                  */
1126                                 bcopy(lc->lc_data + sblkoff, bufaddr,
1127                                     bp->b_bcount);
1128                                 mutex_exit(&lsp->ls_comp_cache_lock);
1129                                 bp->b_resid = 0;
1130                                 error = 0;
1131                                 goto done;
1132                         }
1133                         mutex_exit(&lsp->ls_comp_cache_lock);
1134                 }
1135 
1136                 /*
1137                  * Align start offset to block boundary for segmap
1138                  */
1139                 salign = lsp->ls_comp_seg_index[sblkno];
1140                 sdiff = salign & (DEV_BSIZE - 1);
1141                 salign -= sdiff;
1142                 if (eblkno >= (lsp->ls_comp_index_sz - 1)) {
1143                         /*
1144                          * We're dealing with the last segment of
1145                          * the compressed file -- the size of this
1146                          * segment *may not* be the same as the
1147                          * segment size for the file
1148                          */
1149                         eblkoff = (offset + bp->b_bcount) &
1150                             (lsp->ls_uncomp_last_seg_sz - 1);
1151                         ealign = lsp->ls_vp_comp_size;
1152                 } else {
1153                         ealign = lsp->ls_comp_seg_index[eblkno + 1];
1154                 }
1155 
1156                 /*
1157                  * Preserve original request paramaters
1158                  */
1159                 oblkcount = bp->b_bcount;
1160 
1161                 /*
1162                  * Assign the calculated parameters
1163                  */
1164                 comp_data_sz = ealign - salign;
1165                 bp->b_bcount = comp_data_sz;
1166 
1167                 /*
1168                  * Buffers to hold compressed segments are pre-allocated
1169                  * on a per-thread basis. Find a pre-allocated buffer
1170                  * that is not currently in use and mark it for use.
1171                  */
1172                 mutex_enter(&lsp->ls_comp_bufs_lock);
1173                 for (j = 0; j < lofi_taskq_nthreads; j++) {
1174                         if (lsp->ls_comp_bufs[j].inuse == 0) {
1175                                 lsp->ls_comp_bufs[j].inuse = 1;
1176                                 break;
1177                         }
1178                 }
1179 
1180                 mutex_exit(&lsp->ls_comp_bufs_lock);
1181                 ASSERT(j < lofi_taskq_nthreads);
1182 
1183                 /*
1184                  * If the pre-allocated buffer size does not match
1185                  * the size of the I/O request, re-allocate it with
1186                  * the appropriate size
1187                  */
1188                 if (lsp->ls_comp_bufs[j].bufsize < bp->b_bcount) {
1189                         if (lsp->ls_comp_bufs[j].bufsize > 0)
1190                                 kmem_free(lsp->ls_comp_bufs[j].buf,
1191                                     lsp->ls_comp_bufs[j].bufsize);
1192                         lsp->ls_comp_bufs[j].buf = kmem_alloc(bp->b_bcount,
1193                             KM_SLEEP);
1194                         lsp->ls_comp_bufs[j].bufsize = bp->b_bcount;
1195                 }
1196                 compressed_seg = lsp->ls_comp_bufs[j].buf;
1197 
1198                 /*
1199                  * Map in the calculated number of blocks
1200                  */
1201                 error = lofi_mapped_rdwr((caddr_t)compressed_seg, salign,
1202                     bp, lsp);
1203 
1204                 bp->b_bcount = oblkcount;
1205                 bp->b_resid = oblkcount;
1206                 if (error != 0)
1207                         goto done;
1208 
1209                 /*
1210                  * decompress compressed blocks start
1211                  */
1212                 cmpbuf = compressed_seg + sdiff;
1213                 for (i = sblkno; i <= eblkno; i++) {
1214                         ASSERT(i < lsp->ls_comp_index_sz - 1);
1215                         uchar_t *useg;
1216 
1217                         /*
1218                          * The last segment is special in that it is
1219                          * most likely not going to be the same
1220                          * (uncompressed) size as the other segments.
1221                          */
1222                         if (i == (lsp->ls_comp_index_sz - 2)) {
1223                                 seglen = lsp->ls_uncomp_last_seg_sz;
1224                         } else {
1225                                 seglen = lsp->ls_uncomp_seg_sz;
1226                         }
1227 
1228                         /*
1229                          * Each of the segment index entries contains
1230                          * the starting block number for that segment.
1231                          * The number of compressed bytes in a segment
1232                          * is thus the difference between the starting
1233                          * block number of this segment and the starting
1234                          * block number of the next segment.
1235                          */
1236                         cmpbytes = lsp->ls_comp_seg_index[i + 1] -
1237                             lsp->ls_comp_seg_index[i];
1238 
1239                         /*
1240                          * The first byte in a compressed segment is a flag
1241                          * that indicates whether this segment is compressed
1242                          * at all.
1243                          *
1244                          * The variable 'useg' is used (instead of
1245                          * uncompressed_seg) in this loop to keep a
1246                          * reference to the uncompressed segment.
1247                          *
1248                          * N.B. If 'useg' is replaced with uncompressed_seg,
1249                          * it leads to memory leaks and heap corruption in
1250                          * corner cases where compressed segments lie
1251                          * adjacent to uncompressed segments.
1252                          */
1253                         if (*cmpbuf == UNCOMPRESSED) {
1254                                 useg = cmpbuf + SEGHDR;
1255                         } else {
1256                                 if (uncompressed_seg == NULL)
1257                                         uncompressed_seg =
1258                                             kmem_alloc(lsp->ls_uncomp_seg_sz,
1259                                             KM_SLEEP);
1260                                 useg = uncompressed_seg;
1261                                 uncompressed_seg_index = i;
1262 
1263                                 if (li->l_decompress((cmpbuf + SEGHDR),
1264                                     (cmpbytes - SEGHDR), uncompressed_seg,
1265                                     &seglen, li->l_level) != 0) {
1266                                         error = EIO;
1267                                         goto done;
1268                                 }
1269                         }
1270 
1271                         /*
1272                          * Determine how much uncompressed data we
1273                          * have to copy and copy it
1274                          */
1275                         xfersize = lsp->ls_uncomp_seg_sz - sblkoff;
1276                         if (i == eblkno)
1277                                 xfersize -= (lsp->ls_uncomp_seg_sz - eblkoff);
1278 
1279                         bcopy((useg + sblkoff), bufaddr, xfersize);
1280 
1281                         cmpbuf += cmpbytes;
1282                         bufaddr += xfersize;
1283                         bp->b_resid -= xfersize;
1284                         sblkoff = 0;
1285 
1286                         if (bp->b_resid == 0)
1287                                 break;
1288                 } /* decompress compressed blocks ends */
1289 
1290                 /*
1291                  * Skip to done if there is no uncompressed data to cache
1292                  */
1293                 if (uncompressed_seg == NULL)
1294                         goto done;
1295 
1296                 /*
1297                  * Add the data for the last decompressed segment to
1298                  * the cache.
1299                  *
1300                  * In case the uncompressed segment data was added to (and
1301                  * is referenced by) the cache, make sure we don't free it
1302                  * here.
1303                  */
1304                 mutex_enter(&lsp->ls_comp_cache_lock);
1305                 if ((lc = lofi_add_comp_data(lsp, uncompressed_seg_index,
1306                     uncompressed_seg)) != NULL) {
1307                         uncompressed_seg = NULL;
1308                 }
1309                 mutex_exit(&lsp->ls_comp_cache_lock);
1310 
1311 done:
1312                 if (compressed_seg != NULL) {
1313                         mutex_enter(&lsp->ls_comp_bufs_lock);
1314                         lsp->ls_comp_bufs[j].inuse = 0;
1315                         mutex_exit(&lsp->ls_comp_bufs_lock);
1316                 }
1317                 if (uncompressed_seg != NULL)
1318                         kmem_free(uncompressed_seg, lsp->ls_uncomp_seg_sz);
1319         } /* end of handling compressed files */
1320 
1321         if ((error == 0) && (syncflag != 0))
1322                 error = VOP_FSYNC(lsp->ls_vp, syncflag, kcred, NULL);
1323 
1324 errout:
1325         if (bufinited && lsp->ls_kstat) {
1326                 size_t n_done = bp->b_bcount - bp->b_resid;
1327                 kstat_io_t *kioptr;
1328 
1329                 mutex_enter(lsp->ls_kstat->ks_lock);
1330                 kioptr = KSTAT_IO_PTR(lsp->ls_kstat);
1331                 if (bp->b_flags & B_READ) {
1332                         kioptr->nread += n_done;
1333                         kioptr->reads++;
1334                 } else {
1335                         kioptr->nwritten += n_done;
1336                         kioptr->writes++;
1337                 }
1338                 kstat_runq_exit(kioptr);
1339                 mutex_exit(lsp->ls_kstat->ks_lock);
1340         }
1341 
1342         mutex_enter(&lsp->ls_vp_lock);
1343         if (--lsp->ls_vp_iocount == 0)
1344                 cv_broadcast(&lsp->ls_vp_cv);
1345         mutex_exit(&lsp->ls_vp_lock);
1346 
1347         bioerror(bp, error);
1348         biodone(bp);
1349 }
1350 
1351 static int
1352 lofi_strategy(struct buf *bp)
1353 {
1354         struct lofi_state *lsp;
1355         offset_t        offset;
1356 
1357         /*
1358          * We cannot just do I/O here, because the current thread
1359          * _might_ end up back in here because the underlying filesystem
1360          * wants a buffer, which eventually gets into bio_recycle and
1361          * might call into lofi to write out a delayed-write buffer.
1362          * This is bad if the filesystem above lofi is the same as below.
1363          *
1364          * We could come up with a complex strategy using threads to
1365          * do the I/O asynchronously, or we could use task queues. task
1366          * queues were incredibly easy so they win.
1367          */
1368         lsp = ddi_get_soft_state(lofi_statep, getminor(bp->b_edev));
1369         if (lsp == NULL) {
1370                 bioerror(bp, ENXIO);
1371                 biodone(bp);
1372                 return (0);
1373         }
1374 
1375         mutex_enter(&lsp->ls_vp_lock);
1376         if (lsp->ls_vp == NULL || lsp->ls_vp_closereq) {
1377                 bioerror(bp, EIO);
1378                 biodone(bp);
1379                 mutex_exit(&lsp->ls_vp_lock);
1380                 return (0);
1381         }
1382 
1383         offset = bp->b_lblkno * DEV_BSIZE;   /* offset within file */
1384         if (lsp->ls_crypto_enabled) {
1385                 /* encrypted data really begins after crypto header */
1386                 offset += lsp->ls_crypto_offset;
1387         }
1388         if (offset == lsp->ls_vp_size) {
1389                 /* EOF */
1390                 if ((bp->b_flags & B_READ) != 0) {
1391                         bp->b_resid = bp->b_bcount;
1392                         bioerror(bp, 0);
1393                 } else {
1394                         /* writes should fail */
1395                         bioerror(bp, ENXIO);
1396                 }
1397                 biodone(bp);
1398                 mutex_exit(&lsp->ls_vp_lock);
1399                 return (0);
1400         }
1401         if (offset > lsp->ls_vp_size) {
1402                 bioerror(bp, ENXIO);
1403                 biodone(bp);
1404                 mutex_exit(&lsp->ls_vp_lock);
1405                 return (0);
1406         }
1407         lsp->ls_vp_iocount++;
1408         mutex_exit(&lsp->ls_vp_lock);
1409 
1410         if (lsp->ls_kstat) {
1411                 mutex_enter(lsp->ls_kstat->ks_lock);
1412                 kstat_waitq_enter(KSTAT_IO_PTR(lsp->ls_kstat));
1413                 mutex_exit(lsp->ls_kstat->ks_lock);
1414         }
1415         (void) taskq_dispatch(lsp->ls_taskq, lofi_strategy_task, bp, KM_SLEEP);
1416         return (0);
1417 }
1418 
1419 /*ARGSUSED2*/
1420 static int
1421 lofi_read(dev_t dev, struct uio *uio, struct cred *credp)
1422 {
1423         if (getminor(dev) == 0)
1424                 return (EINVAL);
1425         UIO_CHECK(uio);
1426         return (physio(lofi_strategy, NULL, dev, B_READ, minphys, uio));
1427 }
1428 
1429 /*ARGSUSED2*/
1430 static int
1431 lofi_write(dev_t dev, struct uio *uio, struct cred *credp)
1432 {
1433         if (getminor(dev) == 0)
1434                 return (EINVAL);
1435         UIO_CHECK(uio);
1436         return (physio(lofi_strategy, NULL, dev, B_WRITE, minphys, uio));
1437 }
1438 
1439 /*ARGSUSED2*/
1440 static int
1441 lofi_aread(dev_t dev, struct aio_req *aio, struct cred *credp)
1442 {
1443         if (getminor(dev) == 0)
1444                 return (EINVAL);
1445         UIO_CHECK(aio->aio_uio);
1446         return (aphysio(lofi_strategy, anocancel, dev, B_READ, minphys, aio));
1447 }
1448 
1449 /*ARGSUSED2*/
1450 static int
1451 lofi_awrite(dev_t dev, struct aio_req *aio, struct cred *credp)
1452 {
1453         if (getminor(dev) == 0)
1454                 return (EINVAL);
1455         UIO_CHECK(aio->aio_uio);
1456         return (aphysio(lofi_strategy, anocancel, dev, B_WRITE, minphys, aio));
1457 }
1458 
1459 /*ARGSUSED*/
1460 static int
1461 lofi_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
1462 {
1463         switch (infocmd) {
1464         case DDI_INFO_DEVT2DEVINFO:
1465                 *result = lofi_dip;
1466                 return (DDI_SUCCESS);
1467         case DDI_INFO_DEVT2INSTANCE:
1468                 *result = 0;
1469                 return (DDI_SUCCESS);
1470         }
1471         return (DDI_FAILURE);
1472 }
1473 
1474 static int
1475 lofi_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
1476 {
1477         int     error;
1478 
1479         if (cmd != DDI_ATTACH)
1480                 return (DDI_FAILURE);
1481 
1482         lofi_minor_id = id_space_create("lofi_minor_id", 1, L_MAXMIN32 + 1);
1483 
1484         if (!lofi_minor_id)
1485                 return (DDI_FAILURE);
1486 
1487         error = ddi_soft_state_zalloc(lofi_statep, 0);
1488         if (error == DDI_FAILURE) {
1489                 id_space_destroy(lofi_minor_id);
1490                 return (DDI_FAILURE);
1491         }
1492         error = ddi_create_minor_node(dip, LOFI_CTL_NODE, S_IFCHR, 0,
1493             DDI_PSEUDO, NULL);
1494         if (error == DDI_FAILURE) {
1495                 ddi_soft_state_free(lofi_statep, 0);
1496                 id_space_destroy(lofi_minor_id);
1497                 return (DDI_FAILURE);
1498         }
1499         /* driver handles kernel-issued IOCTLs */
1500         if (ddi_prop_create(DDI_DEV_T_NONE, dip, DDI_PROP_CANSLEEP,
1501             DDI_KERNEL_IOCTL, NULL, 0) != DDI_PROP_SUCCESS) {
1502                 ddi_remove_minor_node(dip, NULL);
1503                 ddi_soft_state_free(lofi_statep, 0);
1504                 id_space_destroy(lofi_minor_id);
1505                 return (DDI_FAILURE);
1506         }
1507 
1508         zone_key_create(&lofi_zone_key, NULL, lofi_zone_shutdown, NULL);
1509 
1510         lofi_dip = dip;
1511         ddi_report_dev(dip);
1512         return (DDI_SUCCESS);
1513 }
1514 
1515 static int
1516 lofi_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
1517 {
1518         if (cmd != DDI_DETACH)
1519                 return (DDI_FAILURE);
1520 
1521         mutex_enter(&lofi_lock);
1522 
1523         if (!list_is_empty(&lofi_list)) {
1524                 mutex_exit(&lofi_lock);
1525                 return (DDI_FAILURE);
1526         }
1527 
1528         lofi_dip = NULL;
1529         ddi_remove_minor_node(dip, NULL);
1530         ddi_prop_remove_all(dip);
1531 
1532         mutex_exit(&lofi_lock);
1533 
1534         if (zone_key_delete(lofi_zone_key) != 0)
1535                 cmn_err(CE_WARN, "failed to delete zone key");
1536 
1537         ddi_soft_state_free(lofi_statep, 0);
1538 
1539         id_space_destroy(lofi_minor_id);
1540 
1541         return (DDI_SUCCESS);
1542 }
1543 
1544 /*
1545  * With addition of encryption, be careful that encryption key is wiped before
1546  * kernel memory structures are freed, and also that key is not accidentally
1547  * passed out into userland structures.
1548  */
1549 static void
1550 free_lofi_ioctl(struct lofi_ioctl *klip)
1551 {
1552         /* Make sure this encryption key doesn't stick around */
1553         bzero(klip->li_key, sizeof (klip->li_key));
1554         kmem_free(klip, sizeof (struct lofi_ioctl));
1555 }
1556 
1557 /*
1558  * These two just simplify the rest of the ioctls that need to copyin/out
1559  * the lofi_ioctl structure.
1560  */
1561 int
1562 copy_in_lofi_ioctl(const struct lofi_ioctl *ulip, struct lofi_ioctl **klipp,
1563     int flag)
1564 {
1565         struct lofi_ioctl *klip;
1566         int     error;
1567 
1568         klip = *klipp = kmem_alloc(sizeof (struct lofi_ioctl), KM_SLEEP);
1569         error = ddi_copyin(ulip, klip, sizeof (struct lofi_ioctl), flag);
1570         if (error)
1571                 goto err;
1572 
1573         /* ensure NULL termination */
1574         klip->li_filename[MAXPATHLEN-1] = '\0';
1575         klip->li_algorithm[MAXALGLEN-1] = '\0';
1576         klip->li_cipher[CRYPTO_MAX_MECH_NAME-1] = '\0';
1577         klip->li_iv_cipher[CRYPTO_MAX_MECH_NAME-1] = '\0';
1578 
1579         if (klip->li_minor > L_MAXMIN32) {
1580                 error = EINVAL;
1581                 goto err;
1582         }
1583 
1584         return (0);
1585 
1586 err:
1587         free_lofi_ioctl(klip);
1588         return (error);
1589 }
1590 
1591 int
1592 copy_out_lofi_ioctl(const struct lofi_ioctl *klip, struct lofi_ioctl *ulip,
1593         int flag)
1594 {
1595         int     error;
1596 
1597         /*
1598          * NOTE: Do NOT copy the crypto_key_t "back" to userland.
1599          * This ensures that an attacker can't trivially find the
1600          * key for a mapping just by issuing the ioctl.
1601          *
1602          * It can still be found by poking around in kmem with mdb(1),
1603          * but there is no point in making it easy when the info isn't
1604          * of any use in this direction anyway.
1605          *
1606          * Either way we don't actually have the raw key stored in
1607          * a form that we can get it anyway, since we just used it
1608          * to create a ctx template and didn't keep "the original".
1609          */
1610         error = ddi_copyout(klip, ulip, sizeof (struct lofi_ioctl), flag);
1611         if (error)
1612                 return (EFAULT);
1613         return (0);
1614 }
1615 
1616 static int
1617 lofi_access(struct lofi_state *lsp)
1618 {
1619         ASSERT(MUTEX_HELD(&lofi_lock));
1620         if (INGLOBALZONE(curproc) || lsp->ls_zone.zref_zone == curzone)
1621                 return (0);
1622         return (EPERM);
1623 }
1624 
1625 /*
1626  * Find the lofi state for the given filename. We compare by vnode to
1627  * allow the global zone visibility into NGZ lofi nodes.
1628  */
1629 static int
1630 file_to_lofi_nocheck(char *filename, boolean_t readonly,
1631     struct lofi_state **lspp)
1632 {
1633         struct lofi_state *lsp;
1634         vnode_t *vp = NULL;
1635         int err = 0;
1636         int rdfiles = 0;
1637 
1638         ASSERT(MUTEX_HELD(&lofi_lock));
1639 
1640         if ((err = lookupname(filename, UIO_SYSSPACE, FOLLOW,
1641             NULLVPP, &vp)) != 0)
1642                 goto out;
1643 
1644         if (vp->v_type == VREG) {
1645                 vnode_t *realvp;
1646                 if (VOP_REALVP(vp, &realvp, NULL) == 0) {
1647                         VN_HOLD(realvp);
1648                         VN_RELE(vp);
1649                         vp = realvp;
1650                 }
1651         }
1652 
1653         for (lsp = list_head(&lofi_list); lsp != NULL;
1654             lsp = list_next(&lofi_list, lsp)) {
1655                 if (lsp->ls_vp == vp) {
1656                         if (lspp != NULL)
1657                                 *lspp = lsp;
1658                         if (lsp->ls_readonly) {
1659                                 rdfiles++;
1660                                 /* Skip if '-r' is specified */
1661                                 if (readonly)
1662                                         continue;
1663                         }
1664                         goto out;
1665                 }
1666         }
1667 
1668         err = ENOENT;
1669 
1670         /*
1671          * If a filename is given as an argument for lofi_unmap, we shouldn't
1672          * allow unmap if there are multiple read-only lofi devices associated
1673          * with this file.
1674          */
1675         if (lspp != NULL) {
1676                 if (rdfiles == 1)
1677                         err = 0;
1678                 else if (rdfiles > 1)
1679                         err = EBUSY;
1680         }
1681 
1682 out:
1683         if (vp != NULL)
1684                 VN_RELE(vp);
1685         return (err);
1686 }
1687 
1688 /*
1689  * Find the minor for the given filename, checking the zone can access
1690  * it.
1691  */
1692 static int
1693 file_to_lofi(char *filename, boolean_t readonly, struct lofi_state **lspp)
1694 {
1695         int err = 0;
1696 
1697         ASSERT(MUTEX_HELD(&lofi_lock));
1698 
1699         if ((err = file_to_lofi_nocheck(filename, readonly, lspp)) != 0)
1700                 return (err);
1701 
1702         if ((err = lofi_access(*lspp)) != 0)
1703                 return (err);
1704 
1705         return (0);
1706 }
1707 
1708 /*
1709  * Fakes up a disk geometry, and one big partition, based on the size
1710  * of the file. This is needed because we allow newfs'ing the device,
1711  * and newfs will do several disk ioctls to figure out the geometry and
1712  * partition information. It uses that information to determine the parameters
1713  * to pass to mkfs. Geometry is pretty much irrelevant these days, but we
1714  * have to support it.
1715  */
1716 static void
1717 fake_disk_geometry(struct lofi_state *lsp)
1718 {
1719         u_offset_t dsize = lsp->ls_vp_size - lsp->ls_crypto_offset;
1720 
1721         /* dk_geom - see dkio(7I) */
1722         /*
1723          * dkg_ncyl _could_ be set to one here (one big cylinder with gobs
1724          * of sectors), but that breaks programs like fdisk which want to
1725          * partition a disk by cylinder. With one cylinder, you can't create
1726          * an fdisk partition and put pcfs on it for testing (hard to pick
1727          * a number between one and one).
1728          *
1729          * The cheezy floppy test is an attempt to not have too few cylinders
1730          * for a small file, or so many on a big file that you waste space
1731          * for backup superblocks or cylinder group structures.
1732          */
1733         if (dsize < (2 * 1024 * 1024)) /* floppy? */
1734                 lsp->ls_dkg.dkg_ncyl = dsize / (100 * 1024);
1735         else
1736                 lsp->ls_dkg.dkg_ncyl = dsize / (300 * 1024);
1737         /* in case file file is < 100k */
1738         if (lsp->ls_dkg.dkg_ncyl == 0)
1739                 lsp->ls_dkg.dkg_ncyl = 1;
1740         lsp->ls_dkg.dkg_acyl = 0;
1741         lsp->ls_dkg.dkg_bcyl = 0;
1742         lsp->ls_dkg.dkg_nhead = 1;
1743         lsp->ls_dkg.dkg_obs1 = 0;
1744         lsp->ls_dkg.dkg_intrlv = 0;
1745         lsp->ls_dkg.dkg_obs2 = 0;
1746         lsp->ls_dkg.dkg_obs3 = 0;
1747         lsp->ls_dkg.dkg_apc = 0;
1748         lsp->ls_dkg.dkg_rpm = 7200;
1749         lsp->ls_dkg.dkg_pcyl = lsp->ls_dkg.dkg_ncyl + lsp->ls_dkg.dkg_acyl;
1750         lsp->ls_dkg.dkg_nsect = dsize / (DEV_BSIZE * lsp->ls_dkg.dkg_ncyl);
1751         lsp->ls_dkg.dkg_write_reinstruct = 0;
1752         lsp->ls_dkg.dkg_read_reinstruct = 0;
1753 
1754         /* vtoc - see dkio(7I) */
1755         bzero(&lsp->ls_vtoc, sizeof (struct vtoc));
1756         lsp->ls_vtoc.v_sanity = VTOC_SANE;
1757         lsp->ls_vtoc.v_version = V_VERSION;
1758         (void) strncpy(lsp->ls_vtoc.v_volume, LOFI_DRIVER_NAME,
1759             sizeof (lsp->ls_vtoc.v_volume));
1760         lsp->ls_vtoc.v_sectorsz = DEV_BSIZE;
1761         lsp->ls_vtoc.v_nparts = 1;
1762         lsp->ls_vtoc.v_part[0].p_tag = V_UNASSIGNED;
1763 
1764         /*
1765          * A compressed file is read-only, other files can
1766          * be read-write
1767          */
1768         if (lsp->ls_uncomp_seg_sz > 0) {
1769                 lsp->ls_vtoc.v_part[0].p_flag = V_UNMNT | V_RONLY;
1770         } else {
1771                 lsp->ls_vtoc.v_part[0].p_flag = V_UNMNT;
1772         }
1773         lsp->ls_vtoc.v_part[0].p_start = (daddr_t)0;
1774         /*
1775          * The partition size cannot just be the number of sectors, because
1776          * that might not end on a cylinder boundary. And if that's the case,
1777          * newfs/mkfs will print a scary warning. So just figure the size
1778          * based on the number of cylinders and sectors/cylinder.
1779          */
1780         lsp->ls_vtoc.v_part[0].p_size = lsp->ls_dkg.dkg_pcyl *
1781             lsp->ls_dkg.dkg_nsect * lsp->ls_dkg.dkg_nhead;
1782 
1783         /* dk_cinfo - see dkio(7I) */
1784         bzero(&lsp->ls_ci, sizeof (struct dk_cinfo));
1785         (void) strcpy(lsp->ls_ci.dki_cname, LOFI_DRIVER_NAME);
1786         lsp->ls_ci.dki_ctype = DKC_MD;
1787         lsp->ls_ci.dki_flags = 0;
1788         lsp->ls_ci.dki_cnum = 0;
1789         lsp->ls_ci.dki_addr = 0;
1790         lsp->ls_ci.dki_space = 0;
1791         lsp->ls_ci.dki_prio = 0;
1792         lsp->ls_ci.dki_vec = 0;
1793         (void) strcpy(lsp->ls_ci.dki_dname, LOFI_DRIVER_NAME);
1794         lsp->ls_ci.dki_unit = 0;
1795         lsp->ls_ci.dki_slave = 0;
1796         lsp->ls_ci.dki_partition = 0;
1797         /*
1798          * newfs uses this to set maxcontig. Must not be < 16, or it
1799          * will be 0 when newfs multiplies it by DEV_BSIZE and divides
1800          * it by the block size. Then tunefs doesn't work because
1801          * maxcontig is 0.
1802          */
1803         lsp->ls_ci.dki_maxtransfer = 16;
1804 }
1805 
1806 /*
1807  * map in a compressed file
1808  *
1809  * Read in the header and the index that follows.
1810  *
1811  * The header is as follows -
1812  *
1813  * Signature (name of the compression algorithm)
1814  * Compression segment size (a multiple of 512)
1815  * Number of index entries
1816  * Size of the last block
1817  * The array containing the index entries
1818  *
1819  * The header information is always stored in
1820  * network byte order on disk.
1821  */
1822 static int
1823 lofi_map_compressed_file(struct lofi_state *lsp, char *buf)
1824 {
1825         uint32_t index_sz, header_len, i;
1826         ssize_t resid;
1827         enum uio_rw rw;
1828         char *tbuf = buf;
1829         int error;
1830 
1831         /* The signature has already been read */
1832         tbuf += sizeof (lsp->ls_comp_algorithm);
1833         bcopy(tbuf, &(lsp->ls_uncomp_seg_sz), sizeof (lsp->ls_uncomp_seg_sz));
1834         lsp->ls_uncomp_seg_sz = ntohl(lsp->ls_uncomp_seg_sz);
1835 
1836         /*
1837          * The compressed segment size must be a power of 2
1838          */
1839         if (lsp->ls_uncomp_seg_sz < DEV_BSIZE ||
1840             !ISP2(lsp->ls_uncomp_seg_sz))
1841                 return (EINVAL);
1842 
1843         for (i = 0; !((lsp->ls_uncomp_seg_sz >> i) & 1); i++)
1844                 ;
1845 
1846         lsp->ls_comp_seg_shift = i;
1847 
1848         tbuf += sizeof (lsp->ls_uncomp_seg_sz);
1849         bcopy(tbuf, &(lsp->ls_comp_index_sz), sizeof (lsp->ls_comp_index_sz));
1850         lsp->ls_comp_index_sz = ntohl(lsp->ls_comp_index_sz);
1851 
1852         tbuf += sizeof (lsp->ls_comp_index_sz);
1853         bcopy(tbuf, &(lsp->ls_uncomp_last_seg_sz),
1854             sizeof (lsp->ls_uncomp_last_seg_sz));
1855         lsp->ls_uncomp_last_seg_sz = ntohl(lsp->ls_uncomp_last_seg_sz);
1856 
1857         /*
1858          * Compute the total size of the uncompressed data
1859          * for use in fake_disk_geometry and other calculations.
1860          * Disk geometry has to be faked with respect to the
1861          * actual uncompressed data size rather than the
1862          * compressed file size.
1863          */
1864         lsp->ls_vp_size =
1865             (u_offset_t)(lsp->ls_comp_index_sz - 2) * lsp->ls_uncomp_seg_sz
1866             + lsp->ls_uncomp_last_seg_sz;
1867 
1868         /*
1869          * Index size is rounded up to DEV_BSIZE for ease
1870          * of segmapping
1871          */
1872         index_sz = sizeof (*lsp->ls_comp_seg_index) * lsp->ls_comp_index_sz;
1873         header_len = sizeof (lsp->ls_comp_algorithm) +
1874             sizeof (lsp->ls_uncomp_seg_sz) +
1875             sizeof (lsp->ls_comp_index_sz) +
1876             sizeof (lsp->ls_uncomp_last_seg_sz);
1877         lsp->ls_comp_offbase = header_len + index_sz;
1878 
1879         index_sz += header_len;
1880         index_sz = roundup(index_sz, DEV_BSIZE);
1881 
1882         lsp->ls_comp_index_data = kmem_alloc(index_sz, KM_SLEEP);
1883         lsp->ls_comp_index_data_sz = index_sz;
1884 
1885         /*
1886          * Read in the index -- this has a side-effect
1887          * of reading in the header as well
1888          */
1889         rw = UIO_READ;
1890         error = vn_rdwr(rw, lsp->ls_vp, lsp->ls_comp_index_data, index_sz,
1891             0, UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred, &resid);
1892 
1893         if (error != 0)
1894                 return (error);
1895 
1896         /* Skip the header, this is where the index really begins */
1897         lsp->ls_comp_seg_index =
1898             /*LINTED*/
1899             (uint64_t *)(lsp->ls_comp_index_data + header_len);
1900 
1901         /*
1902          * Now recompute offsets in the index to account for
1903          * the header length
1904          */
1905         for (i = 0; i < lsp->ls_comp_index_sz; i++) {
1906                 lsp->ls_comp_seg_index[i] = lsp->ls_comp_offbase +
1907                     BE_64(lsp->ls_comp_seg_index[i]);
1908         }
1909 
1910         return (error);
1911 }
1912 
1913 static int
1914 lofi_init_crypto(struct lofi_state *lsp, struct lofi_ioctl *klip)
1915 {
1916         struct crypto_meta chead;
1917         char buf[DEV_BSIZE];
1918         ssize_t resid;
1919         char *marker;
1920         int error;
1921         int ret;
1922         int i;
1923 
1924         if (!klip->li_crypto_enabled)
1925                 return (0);
1926 
1927         /*
1928          * All current algorithms have a max of 448 bits.
1929          */
1930         if (klip->li_iv_len > CRYPTO_BITS2BYTES(512))
1931                 return (EINVAL);
1932 
1933         if (CRYPTO_BITS2BYTES(klip->li_key_len) > sizeof (klip->li_key))
1934                 return (EINVAL);
1935 
1936         lsp->ls_crypto_enabled = klip->li_crypto_enabled;
1937 
1938         mutex_init(&lsp->ls_crypto_lock, NULL, MUTEX_DRIVER, NULL);
1939 
1940         lsp->ls_mech.cm_type = crypto_mech2id(klip->li_cipher);
1941         if (lsp->ls_mech.cm_type == CRYPTO_MECH_INVALID) {
1942                 cmn_err(CE_WARN, "invalid cipher %s requested for %s",
1943                     klip->li_cipher, klip->li_filename);
1944                 return (EINVAL);
1945         }
1946 
1947         /* this is just initialization here */
1948         lsp->ls_mech.cm_param = NULL;
1949         lsp->ls_mech.cm_param_len = 0;
1950 
1951         lsp->ls_iv_type = klip->li_iv_type;
1952         lsp->ls_iv_mech.cm_type = crypto_mech2id(klip->li_iv_cipher);
1953         if (lsp->ls_iv_mech.cm_type == CRYPTO_MECH_INVALID) {
1954                 cmn_err(CE_WARN, "invalid iv cipher %s requested"
1955                     " for %s", klip->li_iv_cipher, klip->li_filename);
1956                 return (EINVAL);
1957         }
1958 
1959         /* iv mech must itself take a null iv */
1960         lsp->ls_iv_mech.cm_param = NULL;
1961         lsp->ls_iv_mech.cm_param_len = 0;
1962         lsp->ls_iv_len = klip->li_iv_len;
1963 
1964         /*
1965          * Create ctx using li_cipher & the raw li_key after checking
1966          * that it isn't a weak key.
1967          */
1968         lsp->ls_key.ck_format = CRYPTO_KEY_RAW;
1969         lsp->ls_key.ck_length = klip->li_key_len;
1970         lsp->ls_key.ck_data = kmem_alloc(
1971             CRYPTO_BITS2BYTES(lsp->ls_key.ck_length), KM_SLEEP);
1972         bcopy(klip->li_key, lsp->ls_key.ck_data,
1973             CRYPTO_BITS2BYTES(lsp->ls_key.ck_length));
1974 
1975         ret = crypto_key_check(&lsp->ls_mech, &lsp->ls_key);
1976         if (ret != CRYPTO_SUCCESS) {
1977                 cmn_err(CE_WARN, "weak key check failed for cipher "
1978                     "%s on file %s (0x%x)", klip->li_cipher,
1979                     klip->li_filename, ret);
1980                 return (EINVAL);
1981         }
1982 
1983         error = vn_rdwr(UIO_READ, lsp->ls_vp, buf, DEV_BSIZE,
1984             CRYOFF, UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred, &resid);
1985         if (error != 0)
1986                 return (error);
1987 
1988         /*
1989          * This is the case where the header in the lofi image is already
1990          * initialized to indicate it is encrypted.
1991          */
1992         if (strncmp(buf, lofi_crypto_magic, sizeof (lofi_crypto_magic)) == 0) {
1993                 /*
1994                  * The encryption header information is laid out this way:
1995                  *      6 bytes:        hex "CFLOFI"
1996                  *      2 bytes:        version = 0 ... for now
1997                  *      96 bytes:       reserved1 (not implemented yet)
1998                  *      4 bytes:        data_sector = 2 ... for now
1999                  *      more...         not implemented yet
2000                  */
2001 
2002                 marker = buf;
2003 
2004                 /* copy the magic */
2005                 bcopy(marker, lsp->ls_crypto.magic,
2006                     sizeof (lsp->ls_crypto.magic));
2007                 marker += sizeof (lsp->ls_crypto.magic);
2008 
2009                 /* read the encryption version number */
2010                 bcopy(marker, &(lsp->ls_crypto.version),
2011                     sizeof (lsp->ls_crypto.version));
2012                 lsp->ls_crypto.version = ntohs(lsp->ls_crypto.version);
2013                 marker += sizeof (lsp->ls_crypto.version);
2014 
2015                 /* read a chunk of reserved data */
2016                 bcopy(marker, lsp->ls_crypto.reserved1,
2017                     sizeof (lsp->ls_crypto.reserved1));
2018                 marker += sizeof (lsp->ls_crypto.reserved1);
2019 
2020                 /* read block number where encrypted data begins */
2021                 bcopy(marker, &(lsp->ls_crypto.data_sector),
2022                     sizeof (lsp->ls_crypto.data_sector));
2023                 lsp->ls_crypto.data_sector = ntohl(lsp->ls_crypto.data_sector);
2024                 marker += sizeof (lsp->ls_crypto.data_sector);
2025 
2026                 /* and ignore the rest until it is implemented */
2027 
2028                 lsp->ls_crypto_offset = lsp->ls_crypto.data_sector * DEV_BSIZE;
2029                 return (0);
2030         }
2031 
2032         /*
2033          * We've requested encryption, but no magic was found, so it must be
2034          * a new image.
2035          */
2036 
2037         for (i = 0; i < sizeof (struct crypto_meta); i++) {
2038                 if (buf[i] != '\0')
2039                         return (EINVAL);
2040         }
2041 
2042         marker = buf;
2043         bcopy(lofi_crypto_magic, marker, sizeof (lofi_crypto_magic));
2044         marker += sizeof (lofi_crypto_magic);
2045         chead.version = htons(LOFI_CRYPTO_VERSION);
2046         bcopy(&(chead.version), marker, sizeof (chead.version));
2047         marker += sizeof (chead.version);
2048         marker += sizeof (chead.reserved1);
2049         chead.data_sector = htonl(LOFI_CRYPTO_DATA_SECTOR);
2050         bcopy(&(chead.data_sector), marker, sizeof (chead.data_sector));
2051 
2052         /* write the header */
2053         error = vn_rdwr(UIO_WRITE, lsp->ls_vp, buf, DEV_BSIZE,
2054             CRYOFF, UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred, &resid);
2055         if (error != 0)
2056                 return (error);
2057 
2058         /* fix things up so it looks like we read this info */
2059         bcopy(lofi_crypto_magic, lsp->ls_crypto.magic,
2060             sizeof (lofi_crypto_magic));
2061         lsp->ls_crypto.version = LOFI_CRYPTO_VERSION;
2062         lsp->ls_crypto.data_sector = LOFI_CRYPTO_DATA_SECTOR;
2063         lsp->ls_crypto_offset = lsp->ls_crypto.data_sector * DEV_BSIZE;
2064         return (0);
2065 }
2066 
2067 /*
2068  * Check to see if the passed in signature is a valid one.  If it is
2069  * valid, return the index into lofi_compress_table.
2070  *
2071  * Return -1 if it is invalid
2072  */
2073 static int
2074 lofi_compress_select(const char *signature)
2075 {
2076         int i;
2077 
2078         for (i = 0; i < LOFI_COMPRESS_FUNCTIONS; i++) {
2079                 if (strcmp(lofi_compress_table[i].l_name, signature) == 0)
2080                         return (i);
2081         }
2082 
2083         return (-1);
2084 }
2085 
2086 static int
2087 lofi_init_compress(struct lofi_state *lsp)
2088 {
2089         char buf[DEV_BSIZE];
2090         int compress_index;
2091         ssize_t resid;
2092         int error;
2093 
2094         error = vn_rdwr(UIO_READ, lsp->ls_vp, buf, DEV_BSIZE, 0, UIO_SYSSPACE,
2095             0, RLIM64_INFINITY, kcred, &resid);
2096 
2097         if (error != 0)
2098                 return (error);
2099 
2100         if ((compress_index = lofi_compress_select(buf)) == -1)
2101                 return (0);
2102 
2103         /* compression and encryption are mutually exclusive */
2104         if (lsp->ls_crypto_enabled)
2105                 return (ENOTSUP);
2106 
2107         /* initialize compression info for compressed lofi */
2108         lsp->ls_comp_algorithm_index = compress_index;
2109         (void) strlcpy(lsp->ls_comp_algorithm,
2110             lofi_compress_table[compress_index].l_name,
2111             sizeof (lsp->ls_comp_algorithm));
2112 
2113         /* Finally setup per-thread pre-allocated buffers */
2114         lsp->ls_comp_bufs = kmem_zalloc(lofi_taskq_nthreads *
2115             sizeof (struct compbuf), KM_SLEEP);
2116 
2117         return (lofi_map_compressed_file(lsp, buf));
2118 }
2119 
2120 /*
2121  * map a file to a minor number. Return the minor number.
2122  */
2123 static int
2124 lofi_map_file(dev_t dev, struct lofi_ioctl *ulip, int pickminor,
2125     int *rvalp, struct cred *credp, int ioctl_flag)
2126 {
2127         minor_t minor = (minor_t)-1;
2128         struct lofi_state *lsp = NULL;
2129         struct lofi_ioctl *klip;
2130         int     error;
2131         struct vnode *vp = NULL;
2132         vattr_t vattr;
2133         int     flag;
2134         dev_t   newdev;
2135         char    namebuf[50];
2136 
2137         error = copy_in_lofi_ioctl(ulip, &klip, ioctl_flag);
2138         if (error != 0)
2139                 return (error);
2140 
2141         mutex_enter(&lofi_lock);
2142 
2143         mutex_enter(&curproc->p_lock);
2144         if ((error = rctl_incr_lofi(curproc, curproc->p_zone, 1)) != 0) {
2145                 mutex_exit(&curproc->p_lock);
2146                 mutex_exit(&lofi_lock);
2147                 free_lofi_ioctl(klip);
2148                 return (error);
2149         }
2150         mutex_exit(&curproc->p_lock);
2151 
2152         if (file_to_lofi_nocheck(klip->li_filename, klip->li_readonly,
2153             NULL) == 0) {
2154                 error = EBUSY;
2155                 goto err;
2156         }
2157 
2158         if (pickminor) {
2159                 minor = (minor_t)id_allocff_nosleep(lofi_minor_id);
2160                 if (minor == (minor_t)-1) {
2161                         error = EAGAIN;
2162                         goto err;
2163                 }
2164         } else {
2165                 if (ddi_get_soft_state(lofi_statep, klip->li_minor) != NULL) {
2166                         error = EEXIST;
2167                         goto err;
2168                 }
2169 
2170                 minor = (minor_t)
2171                     id_alloc_specific_nosleep(lofi_minor_id, klip->li_minor);
2172                 ASSERT(minor != (minor_t)-1);
2173         }
2174 
2175         flag = FREAD | FWRITE | FOFFMAX | FEXCL;
2176         error = vn_open(klip->li_filename, UIO_SYSSPACE, flag, 0, &vp, 0, 0);
2177         if (error) {
2178                 /* try read-only */
2179                 flag &= ~FWRITE;
2180                 error = vn_open(klip->li_filename, UIO_SYSSPACE, flag, 0,
2181                     &vp, 0, 0);
2182                 if (error)
2183                         goto err;
2184         }
2185 
2186         if (!V_ISLOFIABLE(vp->v_type)) {
2187                 error = EINVAL;
2188                 goto err;
2189         }
2190 
2191         vattr.va_mask = AT_SIZE;
2192         error = VOP_GETATTR(vp, &vattr, 0, credp, NULL);
2193         if (error)
2194                 goto err;
2195 
2196         /* the file needs to be a multiple of the block size */
2197         if ((vattr.va_size % DEV_BSIZE) != 0) {
2198                 error = EINVAL;
2199                 goto err;
2200         }
2201 
2202         /* lsp alloc+init */
2203 
2204         error = ddi_soft_state_zalloc(lofi_statep, minor);
2205         if (error == DDI_FAILURE) {
2206                 error = ENOMEM;
2207                 goto err;
2208         }
2209 
2210         lsp = ddi_get_soft_state(lofi_statep, minor);
2211         list_insert_tail(&lofi_list, lsp);
2212 
2213         newdev = makedevice(getmajor(dev), minor);
2214         lsp->ls_dev = newdev;
2215         zone_init_ref(&lsp->ls_zone);
2216         zone_hold_ref(curzone, &lsp->ls_zone, ZONE_REF_LOFI);
2217         lsp->ls_uncomp_seg_sz = 0;
2218         lsp->ls_comp_algorithm[0] = '\0';
2219         lsp->ls_crypto_offset = 0;
2220 
2221         cv_init(&lsp->ls_vp_cv, NULL, CV_DRIVER, NULL);
2222         mutex_init(&lsp->ls_comp_cache_lock, NULL, MUTEX_DRIVER, NULL);
2223         mutex_init(&lsp->ls_comp_bufs_lock, NULL, MUTEX_DRIVER, NULL);
2224         mutex_init(&lsp->ls_kstat_lock, NULL, MUTEX_DRIVER, NULL);
2225         mutex_init(&lsp->ls_vp_lock, NULL, MUTEX_DRIVER, NULL);
2226 
2227         (void) snprintf(namebuf, sizeof (namebuf), "%s_taskq_%d",
2228             LOFI_DRIVER_NAME, minor);
2229         lsp->ls_taskq = taskq_create_proc(namebuf, lofi_taskq_nthreads,
2230             minclsyspri, 1, lofi_taskq_maxalloc, curzone->zone_zsched, 0);
2231 
2232         list_create(&lsp->ls_comp_cache, sizeof (struct lofi_comp_cache),
2233             offsetof(struct lofi_comp_cache, lc_list));
2234 
2235         /*
2236          * save open mode so file can be closed properly and vnode counts
2237          * updated correctly.
2238          */
2239         lsp->ls_openflag = flag;
2240 
2241         lsp->ls_vp = vp;
2242         lsp->ls_stacked_vp = vp;
2243         /*
2244          * Try to handle stacked lofs vnodes.
2245          */
2246         if (vp->v_type == VREG) {
2247                 vnode_t *realvp;
2248 
2249                 if (VOP_REALVP(vp, &realvp, NULL) == 0) {
2250                         /*
2251                          * We need to use the realvp for uniqueness
2252                          * checking, but keep the stacked vp for
2253                          * LOFI_GET_FILENAME display.
2254                          */
2255                         VN_HOLD(realvp);
2256                         lsp->ls_vp = realvp;
2257                 }
2258         }
2259 
2260         lsp->ls_vp_size = vattr.va_size;
2261         lsp->ls_vp_comp_size = lsp->ls_vp_size;
2262 
2263         lsp->ls_kstat = kstat_create_zone(LOFI_DRIVER_NAME, minor,
2264             NULL, "disk", KSTAT_TYPE_IO, 1, 0, getzoneid());
2265 
2266         if (lsp->ls_kstat == NULL) {
2267                 error = ENOMEM;
2268                 goto err;
2269         }
2270 
2271         lsp->ls_kstat->ks_lock = &lsp->ls_kstat_lock;
2272         kstat_zone_add(lsp->ls_kstat, GLOBAL_ZONEID);
2273 
2274         lsp->ls_readonly = klip->li_readonly;
2275 
2276         if ((error = lofi_init_crypto(lsp, klip)) != 0)
2277                 goto err;
2278 
2279         if ((error = lofi_init_compress(lsp)) != 0)
2280                 goto err;
2281 
2282         fake_disk_geometry(lsp);
2283 
2284         /* create minor nodes */
2285 
2286         (void) snprintf(namebuf, sizeof (namebuf), "%d", minor);
2287         error = ddi_create_minor_node(lofi_dip, namebuf, S_IFBLK, minor,
2288             DDI_PSEUDO, NULL);
2289         if (error != DDI_SUCCESS) {
2290                 error = ENXIO;
2291                 goto err;
2292         }
2293 
2294         (void) snprintf(namebuf, sizeof (namebuf), "%d,raw", minor);
2295         error = ddi_create_minor_node(lofi_dip, namebuf, S_IFCHR, minor,
2296             DDI_PSEUDO, NULL);
2297         if (error != DDI_SUCCESS) {
2298                 /* remove block node */
2299                 (void) snprintf(namebuf, sizeof (namebuf), "%d", minor);
2300                 ddi_remove_minor_node(lofi_dip, namebuf);
2301                 error = ENXIO;
2302                 goto err;
2303         }
2304 
2305         /* create DDI properties */
2306 
2307         if ((ddi_prop_update_int64(newdev, lofi_dip, SIZE_PROP_NAME,
2308             lsp->ls_vp_size - lsp->ls_crypto_offset)) != DDI_PROP_SUCCESS) {
2309                 error = EINVAL;
2310                 goto nodeerr;
2311         }
2312 
2313         if ((ddi_prop_update_int64(newdev, lofi_dip, NBLOCKS_PROP_NAME,
2314             (lsp->ls_vp_size - lsp->ls_crypto_offset) / DEV_BSIZE))
2315             != DDI_PROP_SUCCESS) {
2316                 error = EINVAL;
2317                 goto nodeerr;
2318         }
2319 
2320         if (ddi_prop_update_string(newdev, lofi_dip, ZONE_PROP_NAME,
2321             (char *)curproc->p_zone->zone_name) != DDI_PROP_SUCCESS) {
2322                 error = EINVAL;
2323                 goto nodeerr;
2324         }
2325 
2326         kstat_install(lsp->ls_kstat);
2327 
2328         mutex_exit(&lofi_lock);
2329 
2330         if (rvalp)
2331                 *rvalp = (int)minor;
2332         klip->li_minor = minor;
2333         (void) copy_out_lofi_ioctl(klip, ulip, ioctl_flag);
2334         free_lofi_ioctl(klip);
2335         return (0);
2336 
2337 nodeerr:
2338         lofi_free_dev(newdev);
2339 err:
2340         if (lsp != NULL) {
2341                 lofi_destroy(lsp, credp);
2342         } else {
2343                 if (vp != NULL) {
2344                         (void) VOP_CLOSE(vp, flag, 1, 0, credp, NULL);
2345                         VN_RELE(vp);
2346                 }
2347 
2348                 if (minor != (minor_t)-1)
2349                         id_free(lofi_minor_id, minor);
2350 
2351                 rctl_decr_lofi(curproc->p_zone, 1);
2352         }
2353 
2354         mutex_exit(&lofi_lock);
2355         free_lofi_ioctl(klip);
2356         return (error);
2357 }
2358 
2359 /*
2360  * unmap a file.
2361  */
2362 static int
2363 lofi_unmap_file(struct lofi_ioctl *ulip, int byfilename,
2364     struct cred *credp, int ioctl_flag)
2365 {
2366         struct lofi_state *lsp;
2367         struct lofi_ioctl *klip;
2368         int err;
2369 
2370         err = copy_in_lofi_ioctl(ulip, &klip, ioctl_flag);
2371         if (err != 0)
2372                 return (err);
2373 
2374         mutex_enter(&lofi_lock);
2375         if (byfilename) {
2376                 if ((err = file_to_lofi(klip->li_filename, klip->li_readonly,
2377                     &lsp)) != 0) {
2378                         mutex_exit(&lofi_lock);
2379                         return (err);
2380                 }
2381         } else if (klip->li_minor == 0) {
2382                 mutex_exit(&lofi_lock);
2383                 free_lofi_ioctl(klip);
2384                 return (ENXIO);
2385         } else {
2386                 lsp = ddi_get_soft_state(lofi_statep, klip->li_minor);
2387         }
2388 
2389         if (lsp == NULL || lsp->ls_vp == NULL || lofi_access(lsp) != 0) {
2390                 mutex_exit(&lofi_lock);
2391                 free_lofi_ioctl(klip);
2392                 return (ENXIO);
2393         }
2394 
2395         klip->li_minor = getminor(lsp->ls_dev);
2396 
2397         /*
2398          * If it's still held open, we'll do one of three things:
2399          *
2400          * If no flag is set, just return EBUSY.
2401          *
2402          * If the 'cleanup' flag is set, unmap and remove the device when
2403          * the last user finishes.
2404          *
2405          * If the 'force' flag is set, then we forcibly close the underlying
2406          * file.  Subsequent operations will fail, and the DKIOCSTATE ioctl
2407          * will return DKIO_DEV_GONE.  When the device is last closed, the
2408          * device will be cleaned up appropriately.
2409          *
2410          * This is complicated by the fact that we may have outstanding
2411          * dispatched I/Os.  Rather than having a single mutex to serialize all
2412          * I/O, we keep a count of the number of outstanding I/O requests
2413          * (ls_vp_iocount), as well as a flag to indicate that no new I/Os
2414          * should be dispatched (ls_vp_closereq).
2415          *
2416          * We set the flag, wait for the number of outstanding I/Os to reach 0,
2417          * and then close the underlying vnode.
2418          */
2419         if (is_opened(lsp)) {
2420                 if (klip->li_force) {
2421                         mutex_enter(&lsp->ls_vp_lock);
2422                         lsp->ls_vp_closereq = B_TRUE;
2423                         /* wake up any threads waiting on dkiocstate */
2424                         cv_broadcast(&lsp->ls_vp_cv);
2425                         while (lsp->ls_vp_iocount > 0)
2426                                 cv_wait(&lsp->ls_vp_cv, &lsp->ls_vp_lock);
2427                         mutex_exit(&lsp->ls_vp_lock);
2428 
2429                         goto out;
2430                 } else if (klip->li_cleanup) {
2431                         lsp->ls_cleanup = 1;
2432                         mutex_exit(&lofi_lock);
2433                         free_lofi_ioctl(klip);
2434                         return (0);
2435                 }
2436 
2437                 mutex_exit(&lofi_lock);
2438                 free_lofi_ioctl(klip);
2439                 return (EBUSY);
2440         }
2441 
2442 out:
2443         lofi_free_dev(lsp->ls_dev);
2444         lofi_destroy(lsp, credp);
2445 
2446         mutex_exit(&lofi_lock);
2447         (void) copy_out_lofi_ioctl(klip, ulip, ioctl_flag);
2448         free_lofi_ioctl(klip);
2449         return (0);
2450 }
2451 
2452 /*
2453  * get the filename given the minor number, or the minor number given
2454  * the name.
2455  */
2456 /*ARGSUSED*/
2457 static int
2458 lofi_get_info(dev_t dev, struct lofi_ioctl *ulip, int which,
2459     struct cred *credp, int ioctl_flag)
2460 {
2461         struct lofi_ioctl *klip;
2462         struct lofi_state *lsp;
2463         int     error;
2464 
2465         error = copy_in_lofi_ioctl(ulip, &klip, ioctl_flag);
2466         if (error != 0)
2467                 return (error);
2468 
2469         switch (which) {
2470         case LOFI_GET_FILENAME:
2471                 if (klip->li_minor == 0) {
2472                         free_lofi_ioctl(klip);
2473                         return (EINVAL);
2474                 }
2475 
2476                 mutex_enter(&lofi_lock);
2477                 lsp = ddi_get_soft_state(lofi_statep, klip->li_minor);
2478                 if (lsp == NULL || lofi_access(lsp) != 0) {
2479                         mutex_exit(&lofi_lock);
2480                         free_lofi_ioctl(klip);
2481                         return (ENXIO);
2482                 }
2483 
2484                 /*
2485                  * This may fail if, for example, we're trying to look
2486                  * up a zoned NFS path from the global zone.
2487                  */
2488                 if (vnodetopath(NULL, lsp->ls_stacked_vp, klip->li_filename,
2489                     sizeof (klip->li_filename), CRED()) != 0) {
2490                         (void) strlcpy(klip->li_filename, "?",
2491                             sizeof (klip->li_filename));
2492                 }
2493 
2494                 klip->li_readonly = lsp->ls_readonly;
2495 
2496                 (void) strlcpy(klip->li_algorithm, lsp->ls_comp_algorithm,
2497                     sizeof (klip->li_algorithm));
2498                 klip->li_crypto_enabled = lsp->ls_crypto_enabled;
2499                 mutex_exit(&lofi_lock);
2500                 error = copy_out_lofi_ioctl(klip, ulip, ioctl_flag);
2501                 free_lofi_ioctl(klip);
2502                 return (error);
2503         case LOFI_GET_MINOR:
2504                 mutex_enter(&lofi_lock);
2505                 error = file_to_lofi(klip->li_filename,
2506                     klip->li_readonly, &lsp);
2507                 if (error == 0)
2508                         klip->li_minor = getminor(lsp->ls_dev);
2509                 mutex_exit(&lofi_lock);
2510 
2511                 if (error == 0)
2512                         error = copy_out_lofi_ioctl(klip, ulip, ioctl_flag);
2513 
2514                 free_lofi_ioctl(klip);
2515                 return (error);
2516         case LOFI_CHECK_COMPRESSED:
2517                 mutex_enter(&lofi_lock);
2518                 error = file_to_lofi(klip->li_filename,
2519                     klip->li_readonly, &lsp);
2520                 if (error != 0) {
2521                         mutex_exit(&lofi_lock);
2522                         free_lofi_ioctl(klip);
2523                         return (error);
2524                 }
2525 
2526                 klip->li_minor = getminor(lsp->ls_dev);
2527                 (void) strlcpy(klip->li_algorithm, lsp->ls_comp_algorithm,
2528                     sizeof (klip->li_algorithm));
2529 
2530                 mutex_exit(&lofi_lock);
2531                 error = copy_out_lofi_ioctl(klip, ulip, ioctl_flag);
2532                 free_lofi_ioctl(klip);
2533                 return (error);
2534         default:
2535                 free_lofi_ioctl(klip);
2536                 return (EINVAL);
2537         }
2538 }
2539 
2540 static int
2541 lofi_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *credp,
2542     int *rvalp)
2543 {
2544         int     error;
2545         enum dkio_state dkstate;
2546         struct lofi_state *lsp;
2547         minor_t minor;
2548 
2549         minor = getminor(dev);
2550         /* lofi ioctls only apply to the master device */
2551         if (minor == 0) {
2552                 struct lofi_ioctl *lip = (struct lofi_ioctl *)arg;
2553 
2554                 /*
2555                  * the query command only need read-access - i.e., normal
2556                  * users are allowed to do those on the ctl device as
2557                  * long as they can open it read-only.
2558                  */
2559                 switch (cmd) {
2560                 case LOFI_MAP_FILE:
2561                         if ((flag & FWRITE) == 0)
2562                                 return (EPERM);
2563                         return (lofi_map_file(dev, lip, 1, rvalp, credp, flag));
2564                 case LOFI_MAP_FILE_MINOR:
2565                         if ((flag & FWRITE) == 0)
2566                                 return (EPERM);
2567                         return (lofi_map_file(dev, lip, 0, rvalp, credp, flag));
2568                 case LOFI_UNMAP_FILE:
2569                         if ((flag & FWRITE) == 0)
2570                                 return (EPERM);
2571                         return (lofi_unmap_file(lip, 1, credp, flag));
2572                 case LOFI_UNMAP_FILE_MINOR:
2573                         if ((flag & FWRITE) == 0)
2574                                 return (EPERM);
2575                         return (lofi_unmap_file(lip, 0, credp, flag));
2576                 case LOFI_GET_FILENAME:
2577                         return (lofi_get_info(dev, lip, LOFI_GET_FILENAME,
2578                             credp, flag));
2579                 case LOFI_GET_MINOR:
2580                         return (lofi_get_info(dev, lip, LOFI_GET_MINOR,
2581                             credp, flag));
2582 
2583                 /*
2584                  * This API made limited sense when this value was fixed
2585                  * at LOFI_MAX_FILES.  However, its use to iterate
2586                  * across all possible devices in lofiadm means we don't
2587                  * want to return L_MAXMIN32, but the highest
2588                  * *allocated* minor.
2589                  */
2590                 case LOFI_GET_MAXMINOR:
2591                         minor = 0;
2592 
2593                         mutex_enter(&lofi_lock);
2594 
2595                         for (lsp = list_head(&lofi_list); lsp != NULL;
2596                             lsp = list_next(&lofi_list, lsp)) {
2597                                 if (lofi_access(lsp) != 0)
2598                                         continue;
2599 
2600                                 if (getminor(lsp->ls_dev) > minor)
2601                                         minor = getminor(lsp->ls_dev);
2602                         }
2603 
2604                         mutex_exit(&lofi_lock);
2605 
2606                         error = ddi_copyout(&minor, &lip->li_minor,
2607                             sizeof (minor), flag);
2608                         if (error)
2609                                 return (EFAULT);
2610                         return (0);
2611 
2612                 case LOFI_CHECK_COMPRESSED:
2613                         return (lofi_get_info(dev, lip, LOFI_CHECK_COMPRESSED,
2614                             credp, flag));
2615                 default:
2616                         return (EINVAL);
2617                 }
2618         }
2619 
2620         mutex_enter(&lofi_lock);
2621         lsp = ddi_get_soft_state(lofi_statep, minor);
2622         if (lsp == NULL || lsp->ls_vp_closereq) {
2623                 mutex_exit(&lofi_lock);
2624                 return (ENXIO);
2625         }
2626         mutex_exit(&lofi_lock);
2627 
2628         /*
2629          * We explicitly allow DKIOCSTATE, but all other ioctls should fail with
2630          * EIO as if the device was no longer present.
2631          */
2632         if (lsp->ls_vp == NULL && cmd != DKIOCSTATE)
2633                 return (EIO);
2634 
2635         /* these are for faking out utilities like newfs */
2636         switch (cmd) {
2637         case DKIOCGVTOC:
2638                 switch (ddi_model_convert_from(flag & FMODELS)) {
2639                 case DDI_MODEL_ILP32: {
2640                         struct vtoc32 vtoc32;
2641 
2642                         vtoctovtoc32(lsp->ls_vtoc, vtoc32);
2643                         if (ddi_copyout(&vtoc32, (void *)arg,
2644                             sizeof (struct vtoc32), flag))
2645                                 return (EFAULT);
2646                         break;
2647                         }
2648 
2649                 case DDI_MODEL_NONE:
2650                         if (ddi_copyout(&lsp->ls_vtoc, (void *)arg,
2651                             sizeof (struct vtoc), flag))
2652                                 return (EFAULT);
2653                         break;
2654                 }
2655                 return (0);
2656         case DKIOCINFO:
2657                 error = ddi_copyout(&lsp->ls_ci, (void *)arg,
2658                     sizeof (struct dk_cinfo), flag);
2659                 if (error)
2660                         return (EFAULT);
2661                 return (0);
2662         case DKIOCG_VIRTGEOM:
2663         case DKIOCG_PHYGEOM:
2664         case DKIOCGGEOM:
2665                 error = ddi_copyout(&lsp->ls_dkg, (void *)arg,
2666                     sizeof (struct dk_geom), flag);
2667                 if (error)
2668                         return (EFAULT);
2669                 return (0);
2670         case DKIOCSTATE:
2671                 /*
2672                  * Normally, lofi devices are always in the INSERTED state.  If
2673                  * a device is forcefully unmapped, then the device transitions
2674                  * to the DKIO_DEV_GONE state.
2675                  */
2676                 if (ddi_copyin((void *)arg, &dkstate, sizeof (dkstate),
2677                     flag) != 0)
2678                         return (EFAULT);
2679 
2680                 mutex_enter(&lsp->ls_vp_lock);
2681                 lsp->ls_vp_iocount++;
2682                 while (((dkstate == DKIO_INSERTED && lsp->ls_vp != NULL) ||
2683                     (dkstate == DKIO_DEV_GONE && lsp->ls_vp == NULL)) &&
2684                     !lsp->ls_vp_closereq) {
2685                         /*
2686                          * By virtue of having the device open, we know that
2687                          * 'lsp' will remain valid when we return.
2688                          */
2689                         if (!cv_wait_sig(&lsp->ls_vp_cv,
2690                             &lsp->ls_vp_lock)) {
2691                                 lsp->ls_vp_iocount--;
2692                                 cv_broadcast(&lsp->ls_vp_cv);
2693                                 mutex_exit(&lsp->ls_vp_lock);
2694                                 return (EINTR);
2695                         }
2696                 }
2697 
2698                 dkstate = (!lsp->ls_vp_closereq && lsp->ls_vp != NULL ?
2699                     DKIO_INSERTED : DKIO_DEV_GONE);
2700                 lsp->ls_vp_iocount--;
2701                 cv_broadcast(&lsp->ls_vp_cv);
2702                 mutex_exit(&lsp->ls_vp_lock);
2703 
2704                 if (ddi_copyout(&dkstate, (void *)arg,
2705                     sizeof (dkstate), flag) != 0)
2706                         return (EFAULT);
2707                 return (0);
2708         default:
2709                 return (ENOTTY);
2710         }
2711 }
2712 
2713 static struct cb_ops lofi_cb_ops = {
2714         lofi_open,              /* open */
2715         lofi_close,             /* close */
2716         lofi_strategy,          /* strategy */
2717         nodev,                  /* print */
2718         nodev,                  /* dump */
2719         lofi_read,              /* read */
2720         lofi_write,             /* write */
2721         lofi_ioctl,             /* ioctl */
2722         nodev,                  /* devmap */
2723         nodev,                  /* mmap */
2724         nodev,                  /* segmap */
2725         nochpoll,               /* poll */
2726         ddi_prop_op,            /* prop_op */
2727         0,                      /* streamtab  */
2728         D_64BIT | D_NEW | D_MP, /* Driver compatibility flag */
2729         CB_REV,
2730         lofi_aread,
2731         lofi_awrite
2732 };
2733 
2734 static struct dev_ops lofi_ops = {
2735         DEVO_REV,               /* devo_rev, */
2736         0,                      /* refcnt  */
2737         lofi_info,              /* info */
2738         nulldev,                /* identify */
2739         nulldev,                /* probe */
2740         lofi_attach,            /* attach */
2741         lofi_detach,            /* detach */
2742         nodev,                  /* reset */
2743         &lofi_cb_ops,               /* driver operations */
2744         NULL,                   /* no bus operations */
2745         NULL,                   /* power */
2746         ddi_quiesce_not_needed, /* quiesce */
2747 };
2748 
2749 static struct modldrv modldrv = {
2750         &mod_driverops,
2751         "loopback file driver",
2752         &lofi_ops,
2753 };
2754 
2755 static struct modlinkage modlinkage = {
2756         MODREV_1,
2757         &modldrv,
2758         NULL
2759 };
2760 
2761 int
2762 _init(void)
2763 {
2764         int error;
2765 
2766         list_create(&lofi_list, sizeof (struct lofi_state),
2767             offsetof(struct lofi_state, ls_list));
2768 
2769         error = ddi_soft_state_init(&lofi_statep,
2770             sizeof (struct lofi_state), 0);
2771         if (error)
2772                 return (error);
2773 
2774         mutex_init(&lofi_lock, NULL, MUTEX_DRIVER, NULL);
2775 
2776         error = mod_install(&modlinkage);
2777         if (error) {
2778                 mutex_destroy(&lofi_lock);
2779                 ddi_soft_state_fini(&lofi_statep);
2780                 list_destroy(&lofi_list);
2781         }
2782 
2783         return (error);
2784 }
2785 
2786 int
2787 _fini(void)
2788 {
2789         int     error;
2790 
2791         mutex_enter(&lofi_lock);
2792 
2793         if (!list_is_empty(&lofi_list)) {
2794                 mutex_exit(&lofi_lock);
2795                 return (EBUSY);
2796         }
2797 
2798         mutex_exit(&lofi_lock);
2799 
2800         error = mod_remove(&modlinkage);
2801         if (error)
2802                 return (error);
2803 
2804         mutex_destroy(&lofi_lock);
2805         ddi_soft_state_fini(&lofi_statep);
2806         list_destroy(&lofi_list);
2807 
2808         return (error);
2809 }
2810 
2811 int
2812 _info(struct modinfo *modinfop)
2813 {
2814         return (mod_info(&modlinkage, modinfop));
2815 }