module/zfs/zfs_log.c

   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) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2015, 2018 by Delphix. All rights reserved.
  24  */
  25
  26
  27 #include <sys/types.h>
  28 #include <sys/param.h>
  29 #include <sys/sysmacros.h>
  30 #include <sys/cmn_err.h>
  31 #include <sys/kmem.h>
  32 #include <sys/thread.h>
  33 #include <sys/file.h>
  34 #include <sys/vfs.h>
  35 #include <sys/zfs_znode.h>
  36 #include <sys/zfs_dir.h>
  37 #include <sys/zil.h>
  38 #include <sys/zil_impl.h>
  39 #include <sys/byteorder.h>
  40 #include <sys/policy.h>
  41 #include <sys/stat.h>
  42 #include <sys/acl.h>
  43 #include <sys/dmu.h>
  44 #include <sys/dbuf.h>
  45 #include <sys/spa.h>
  46 #include <sys/zfs_fuid.h>
  47 #include <sys/dsl_dataset.h>
  48
  49 /*
  50  * These zfs_log_* functions must be called within a dmu tx, in one
  51  * of 2 contexts depending on zilog->z_replay:
  52  *
  53  * Non replay mode
  54  * ---------------
  55  * We need to record the transaction so that if it is committed to
  56  * the Intent Log then it can be replayed.  An intent log transaction
  57  * structure (itx_t) is allocated and all the information necessary to
  58  * possibly replay the transaction is saved in it. The itx is then assigned
  59  * a sequence number and inserted in the in-memory list anchored in the zilog.
  60  *
  61  * Replay mode
  62  * -----------
  63  * We need to mark the intent log record as replayed in the log header.
  64  * This is done in the same transaction as the replay so that they
  65  * commit atomically.
  66  */
  67
  68 int
  69 zfs_log_create_txtype(zil_create_t type, vsecattr_t *vsecp, vattr_t *vap)
  70 {
  71         int isxvattr = (vap->va_mask & ATTR_XVATTR);
  72         switch (type) {
  73         case Z_FILE:
  74                 if (vsecp == NULL && !isxvattr)
  75                         return (TX_CREATE);
  76                 if (vsecp && isxvattr)
  77                         return (TX_CREATE_ACL_ATTR);
  78                 if (vsecp)
  79                         return (TX_CREATE_ACL);
  80                 else
  81                         return (TX_CREATE_ATTR);
  82         case Z_DIR:
  83                 if (vsecp == NULL && !isxvattr)
  84                         return (TX_MKDIR);
  85                 if (vsecp && isxvattr)
  86                         return (TX_MKDIR_ACL_ATTR);
  87                 if (vsecp)
  88                         return (TX_MKDIR_ACL);
  89                 else
  90                         return (TX_MKDIR_ATTR);
  91         case Z_XATTRDIR:
  92                 return (TX_MKXATTR);
  93         }
  94         ASSERT(0);
  95         return (TX_MAX_TYPE);
  96 }
  97
  98 /*
  99  * build up the log data necessary for logging xvattr_t
 100  * First lr_attr_t is initialized.  following the lr_attr_t
 101  * is the mapsize and attribute bitmap copied from the xvattr_t.
 102  * Following the bitmap and bitmapsize two 64 bit words are reserved
 103  * for the create time which may be set.  Following the create time
 104  * records a single 64 bit integer which has the bits to set on
 105  * replay for the xvattr.
 106  */
 107 static void
 108 zfs_log_xvattr(lr_attr_t *lrattr, xvattr_t *xvap)
 109 {
 110         uint32_t        *bitmap;
 111         uint64_t        *attrs;
 112         uint64_t        *crtime;
 113         xoptattr_t      *xoap;
 114         void            *scanstamp;
 115         int             i;
 116
 117         xoap = xva_getxoptattr(xvap);
 118         ASSERT(xoap);
 119
 120         lrattr->lr_attr_masksize = xvap->xva_mapsize;
 121         bitmap = &lrattr->lr_attr_bitmap;
 122         for (i = 0; i != xvap->xva_mapsize; i++, bitmap++) {
 123                 *bitmap = xvap->xva_reqattrmap[i];
 124         }
 125
 126         /* Now pack the attributes up in a single uint64_t */
 127         attrs = (uint64_t *)bitmap;
 128         *attrs = 0;
 129         crtime = attrs + 1;
 130         memset(crtime, 0, 2 * sizeof (uint64_t));
 131         scanstamp = (caddr_t)(crtime + 2);
 132         memset(scanstamp, 0, AV_SCANSTAMP_SZ);
 133         if (XVA_ISSET_REQ(xvap, XAT_READONLY))
 134                 *attrs |= (xoap->xoa_readonly == 0) ? 0 :
 135                     XAT0_READONLY;
 136         if (XVA_ISSET_REQ(xvap, XAT_HIDDEN))
 137                 *attrs |= (xoap->xoa_hidden == 0) ? 0 :
 138                     XAT0_HIDDEN;
 139         if (XVA_ISSET_REQ(xvap, XAT_SYSTEM))
 140                 *attrs |= (xoap->xoa_system == 0) ? 0 :
 141                     XAT0_SYSTEM;
 142         if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE))
 143                 *attrs |= (xoap->xoa_archive == 0) ? 0 :
 144                     XAT0_ARCHIVE;
 145         if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE))
 146                 *attrs |= (xoap->xoa_immutable == 0) ? 0 :
 147                     XAT0_IMMUTABLE;
 148         if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK))
 149                 *attrs |= (xoap->xoa_nounlink == 0) ? 0 :
 150                     XAT0_NOUNLINK;
 151         if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY))
 152                 *attrs |= (xoap->xoa_appendonly == 0) ? 0 :
 153                     XAT0_APPENDONLY;
 154         if (XVA_ISSET_REQ(xvap, XAT_OPAQUE))
 155                 *attrs |= (xoap->xoa_opaque == 0) ? 0 :
 156                     XAT0_APPENDONLY;
 157         if (XVA_ISSET_REQ(xvap, XAT_NODUMP))
 158                 *attrs |= (xoap->xoa_nodump == 0) ? 0 :
 159                     XAT0_NODUMP;
 160         if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED))
 161                 *attrs |= (xoap->xoa_av_quarantined == 0) ? 0 :
 162                     XAT0_AV_QUARANTINED;
 163         if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED))
 164                 *attrs |= (xoap->xoa_av_modified == 0) ? 0 :
 165                     XAT0_AV_MODIFIED;
 166         if (XVA_ISSET_REQ(xvap, XAT_CREATETIME))
 167                 ZFS_TIME_ENCODE(&xoap->xoa_createtime, crtime);
 168         if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) {
 169                 ASSERT(!XVA_ISSET_REQ(xvap, XAT_PROJID));
 170
 171                 memcpy(scanstamp, xoap->xoa_av_scanstamp, AV_SCANSTAMP_SZ);
 172         } else if (XVA_ISSET_REQ(xvap, XAT_PROJID)) {
 173                 /*
 174                  * XAT_PROJID and XAT_AV_SCANSTAMP will never be valid
 175                  * at the same time, so we can share the same space.
 176                  */
 177                 memcpy(scanstamp, &xoap->xoa_projid, sizeof (uint64_t));
 178         }
 179         if (XVA_ISSET_REQ(xvap, XAT_REPARSE))
 180                 *attrs |= (xoap->xoa_reparse == 0) ? 0 :
 181                     XAT0_REPARSE;
 182         if (XVA_ISSET_REQ(xvap, XAT_OFFLINE))
 183                 *attrs |= (xoap->xoa_offline == 0) ? 0 :
 184                     XAT0_OFFLINE;
 185         if (XVA_ISSET_REQ(xvap, XAT_SPARSE))
 186                 *attrs |= (xoap->xoa_sparse == 0) ? 0 :
 187                     XAT0_SPARSE;
 188         if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT))
 189                 *attrs |= (xoap->xoa_projinherit == 0) ? 0 :
 190                     XAT0_PROJINHERIT;
 191 }
 192
 193 static void *
 194 zfs_log_fuid_ids(zfs_fuid_info_t *fuidp, void *start)
 195 {
 196         zfs_fuid_t *zfuid;
 197         uint64_t *fuidloc = start;
 198
 199         /* First copy in the ACE FUIDs */
 200         for (zfuid = list_head(&fuidp->z_fuids); zfuid;
 201             zfuid = list_next(&fuidp->z_fuids, zfuid)) {
 202                 *fuidloc++ = zfuid->z_logfuid;
 203         }
 204         return (fuidloc);
 205 }
 206
 207
 208 static void *
 209 zfs_log_fuid_domains(zfs_fuid_info_t *fuidp, void *start)
 210 {
 211         zfs_fuid_domain_t *zdomain;
 212
 213         /* now copy in the domain info, if any */
 214         if (fuidp->z_domain_str_sz != 0) {
 215                 for (zdomain = list_head(&fuidp->z_domains); zdomain;
 216                     zdomain = list_next(&fuidp->z_domains, zdomain)) {
 217                         memcpy(start, zdomain->z_domain,
 218                             strlen(zdomain->z_domain) + 1);
 219                         start = (caddr_t)start +
 220                             strlen(zdomain->z_domain) + 1;
 221                 }
 222         }
 223         return (start);
 224 }
 225
 226 /*
 227  * If zp is an xattr node, check whether the xattr owner is unlinked.
 228  * We don't want to log anything if the owner is unlinked.
 229  */
 230 static int
 231 zfs_xattr_owner_unlinked(znode_t *zp)
 232 {
 233         int unlinked = 0;
 234         znode_t *dzp;
 235 #ifdef __FreeBSD__
 236         znode_t *tzp = zp;
 237
 238         /*
 239          * zrele drops the vnode lock which violates the VOP locking contract
 240          * on FreeBSD. See comment at the top of zfs_replay.c for more detail.
 241          */
 242         /*
 243          * if zp is XATTR node, keep walking up via z_xattr_parent until we
 244          * get the owner
 245          */
 246         while (tzp->z_pflags & ZFS_XATTR) {
 247                 ASSERT3U(zp->z_xattr_parent, !=, 0);
 248                 if (zfs_zget(ZTOZSB(tzp), tzp->z_xattr_parent, &dzp) != 0) {
 249                         unlinked = 1;
 250                         break;
 251                 }
 252
 253                 if (tzp != zp)
 254                         zrele(tzp);
 255                 tzp = dzp;
 256                 unlinked = tzp->z_unlinked;
 257         }
 258         if (tzp != zp)
 259                 zrele(tzp);
 260 #else
 261         zhold(zp);
 262         /*
 263          * if zp is XATTR node, keep walking up via z_xattr_parent until we
 264          * get the owner
 265          */
 266         while (zp->z_pflags & ZFS_XATTR) {
 267                 ASSERT3U(zp->z_xattr_parent, !=, 0);
 268                 if (zfs_zget(ZTOZSB(zp), zp->z_xattr_parent, &dzp) != 0) {
 269                         unlinked = 1;
 270                         break;
 271                 }
 272
 273                 zrele(zp);
 274                 zp = dzp;
 275                 unlinked = zp->z_unlinked;
 276         }
 277         zrele(zp);
 278 #endif
 279         return (unlinked);
 280 }
 281
 282 /*
 283  * Handles TX_CREATE, TX_CREATE_ATTR, TX_MKDIR, TX_MKDIR_ATTR and
 284  * TK_MKXATTR transactions.
 285  *
 286  * TX_CREATE and TX_MKDIR are standard creates, but they may have FUID
 287  * domain information appended prior to the name.  In this case the
 288  * uid/gid in the log record will be a log centric FUID.
 289  *
 290  * TX_CREATE_ACL_ATTR and TX_MKDIR_ACL_ATTR handle special creates that
 291  * may contain attributes, ACL and optional fuid information.
 292  *
 293  * TX_CREATE_ACL and TX_MKDIR_ACL handle special creates that specify
 294  * and ACL and normal users/groups in the ACEs.
 295  *
 296  * There may be an optional xvattr attribute information similar
 297  * to zfs_log_setattr.
 298  *
 299  * Also, after the file name "domain" strings may be appended.
 300  */
 301 void
 302 zfs_log_create(zilog_t *zilog, dmu_tx_t *tx, uint64_t txtype,
 303     znode_t *dzp, znode_t *zp, const char *name, vsecattr_t *vsecp,
 304     zfs_fuid_info_t *fuidp, vattr_t *vap)
 305 {
 306         itx_t *itx;
 307         lr_create_t *lr;
 308         lr_acl_create_t *lracl;
 309         size_t aclsize = 0;
 310         size_t xvatsize = 0;
 311         size_t txsize;
 312         xvattr_t *xvap = (xvattr_t *)vap;
 313         void *end;
 314         size_t lrsize;
 315         size_t namesize = strlen(name) + 1;
 316         size_t fuidsz = 0;
 317
 318         if (zil_replaying(zilog, tx) || zfs_xattr_owner_unlinked(dzp))
 319                 return;
 320
 321         /*
 322          * If we have FUIDs present then add in space for
 323          * domains and ACE fuid's if any.
 324          */
 325         if (fuidp) {
 326                 fuidsz += fuidp->z_domain_str_sz;
 327                 fuidsz += fuidp->z_fuid_cnt * sizeof (uint64_t);
 328         }
 329
 330         if (vap->va_mask & ATTR_XVATTR)
 331                 xvatsize = ZIL_XVAT_SIZE(xvap->xva_mapsize);
 332
 333         if ((int)txtype == TX_CREATE_ATTR || (int)txtype == TX_MKDIR_ATTR ||
 334             (int)txtype == TX_CREATE || (int)txtype == TX_MKDIR ||
 335             (int)txtype == TX_MKXATTR) {
 336                 txsize = sizeof (*lr) + namesize + fuidsz + xvatsize;
 337                 lrsize = sizeof (*lr);
 338         } else {
 339                 txsize =
 340                     sizeof (lr_acl_create_t) + namesize + fuidsz +
 341                     ZIL_ACE_LENGTH(aclsize) + xvatsize;
 342                 lrsize = sizeof (lr_acl_create_t);
 343         }
 344
 345         itx = zil_itx_create(txtype, txsize);
 346
 347         lr = (lr_create_t *)&itx->itx_lr;
 348         lr->lr_doid = dzp->z_id;
 349         lr->lr_foid = zp->z_id;
 350         /* Store dnode slot count in 8 bits above object id. */
 351         LR_FOID_SET_SLOTS(lr->lr_foid, zp->z_dnodesize >> DNODE_SHIFT);
 352         lr->lr_mode = zp->z_mode;
 353         if (!IS_EPHEMERAL(KUID_TO_SUID(ZTOUID(zp)))) {
 354                 lr->lr_uid = (uint64_t)KUID_TO_SUID(ZTOUID(zp));
 355         } else {
 356                 lr->lr_uid = fuidp->z_fuid_owner;
 357         }
 358         if (!IS_EPHEMERAL(KGID_TO_SGID(ZTOGID(zp)))) {
 359                 lr->lr_gid = (uint64_t)KGID_TO_SGID(ZTOGID(zp));
 360         } else {
 361                 lr->lr_gid = fuidp->z_fuid_group;
 362         }
 363         (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(ZTOZSB(zp)), &lr->lr_gen,
 364             sizeof (uint64_t));
 365         (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(ZTOZSB(zp)),
 366             lr->lr_crtime, sizeof (uint64_t) * 2);
 367
 368         if (sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(ZTOZSB(zp)), &lr->lr_rdev,
 369             sizeof (lr->lr_rdev)) != 0)
 370                 lr->lr_rdev = 0;
 371
 372         /*
 373          * Fill in xvattr info if any
 374          */
 375         if (vap->va_mask & ATTR_XVATTR) {
 376                 zfs_log_xvattr((lr_attr_t *)((caddr_t)lr + lrsize), xvap);
 377                 end = (caddr_t)lr + lrsize + xvatsize;
 378         } else {
 379                 end = (caddr_t)lr + lrsize;
 380         }
 381
 382         /* Now fill in any ACL info */
 383
 384         if (vsecp) {
 385                 lracl = (lr_acl_create_t *)&itx->itx_lr;
 386                 lracl->lr_aclcnt = vsecp->vsa_aclcnt;
 387                 lracl->lr_acl_bytes = aclsize;
 388                 lracl->lr_domcnt = fuidp ? fuidp->z_domain_cnt : 0;
 389                 lracl->lr_fuidcnt  = fuidp ? fuidp->z_fuid_cnt : 0;
 390                 if (vsecp->vsa_aclflags & VSA_ACE_ACLFLAGS)
 391                         lracl->lr_acl_flags = (uint64_t)vsecp->vsa_aclflags;
 392                 else
 393                         lracl->lr_acl_flags = 0;
 394
 395                 memcpy(end, vsecp->vsa_aclentp, aclsize);
 396                 end = (caddr_t)end + ZIL_ACE_LENGTH(aclsize);
 397         }
 398
 399         /* drop in FUID info */
 400         if (fuidp) {
 401                 end = zfs_log_fuid_ids(fuidp, end);
 402                 end = zfs_log_fuid_domains(fuidp, end);
 403         }
 404         /*
 405          * Now place file name in log record
 406          */
 407         memcpy(end, name, namesize);
 408
 409         zil_itx_assign(zilog, itx, tx);
 410 }
 411
 412 /*
 413  * Handles both TX_REMOVE and TX_RMDIR transactions.
 414  */
 415 void
 416 zfs_log_remove(zilog_t *zilog, dmu_tx_t *tx, uint64_t txtype,
 417     znode_t *dzp, const char *name, uint64_t foid, boolean_t unlinked)
 418 {
 419         itx_t *itx;
 420         lr_remove_t *lr;
 421         size_t namesize = strlen(name) + 1;
 422
 423         if (zil_replaying(zilog, tx) || zfs_xattr_owner_unlinked(dzp))
 424                 return;
 425
 426         itx = zil_itx_create(txtype, sizeof (*lr) + namesize);
 427         lr = (lr_remove_t *)&itx->itx_lr;
 428         lr->lr_doid = dzp->z_id;
 429         memcpy(lr + 1, name, namesize);
 430
 431         itx->itx_oid = foid;
 432
 433         /*
 434          * Object ids can be re-instantiated in the next txg so
 435          * remove any async transactions to avoid future leaks.
 436          * This can happen if a fsync occurs on the re-instantiated
 437          * object for a WR_INDIRECT or WR_NEED_COPY write, which gets
 438          * the new file data and flushes a write record for the old object.
 439          */
 440         if (unlinked) {
 441                 ASSERT((txtype & ~TX_CI) == TX_REMOVE);
 442                 zil_remove_async(zilog, foid);
 443         }
 444         zil_itx_assign(zilog, itx, tx);
 445 }
 446
 447 /*
 448  * Handles TX_LINK transactions.
 449  */
 450 void
 451 zfs_log_link(zilog_t *zilog, dmu_tx_t *tx, uint64_t txtype,
 452     znode_t *dzp, znode_t *zp, const char *name)
 453 {
 454         itx_t *itx;
 455         lr_link_t *lr;
 456         size_t namesize = strlen(name) + 1;
 457
 458         if (zil_replaying(zilog, tx))
 459                 return;
 460
 461         itx = zil_itx_create(txtype, sizeof (*lr) + namesize);
 462         lr = (lr_link_t *)&itx->itx_lr;
 463         lr->lr_doid = dzp->z_id;
 464         lr->lr_link_obj = zp->z_id;
 465         memcpy(lr + 1, name, namesize);
 466
 467         zil_itx_assign(zilog, itx, tx);
 468 }
 469
 470 /*
 471  * Handles TX_SYMLINK transactions.
 472  */
 473 void
 474 zfs_log_symlink(zilog_t *zilog, dmu_tx_t *tx, uint64_t txtype,
 475     znode_t *dzp, znode_t *zp, const char *name, const char *link)
 476 {
 477         itx_t *itx;
 478         lr_create_t *lr;
 479         size_t namesize = strlen(name) + 1;
 480         size_t linksize = strlen(link) + 1;
 481
 482         if (zil_replaying(zilog, tx))
 483                 return;
 484
 485         itx = zil_itx_create(txtype, sizeof (*lr) + namesize + linksize);
 486         lr = (lr_create_t *)&itx->itx_lr;
 487         lr->lr_doid = dzp->z_id;
 488         lr->lr_foid = zp->z_id;
 489         lr->lr_uid = KUID_TO_SUID(ZTOUID(zp));
 490         lr->lr_gid = KGID_TO_SGID(ZTOGID(zp));
 491         lr->lr_mode = zp->z_mode;
 492         (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(ZTOZSB(zp)), &lr->lr_gen,
 493             sizeof (uint64_t));
 494         (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(ZTOZSB(zp)),
 495             lr->lr_crtime, sizeof (uint64_t) * 2);
 496         memcpy((char *)(lr + 1), name, namesize);
 497         memcpy((char *)(lr + 1) + namesize, link, linksize);
 498
 499         zil_itx_assign(zilog, itx, tx);
 500 }
 501
 502 /*
 503  * Handles TX_RENAME transactions.
 504  */
 505 void
 506 zfs_log_rename(zilog_t *zilog, dmu_tx_t *tx, uint64_t txtype, znode_t *sdzp,
 507     const char *sname, znode_t *tdzp, const char *dname, znode_t *szp)
 508 {
 509         itx_t *itx;
 510         lr_rename_t *lr;
 511         size_t snamesize = strlen(sname) + 1;
 512         size_t dnamesize = strlen(dname) + 1;
 513
 514         if (zil_replaying(zilog, tx))
 515                 return;
 516
 517         itx = zil_itx_create(txtype, sizeof (*lr) + snamesize + dnamesize);
 518         lr = (lr_rename_t *)&itx->itx_lr;
 519         lr->lr_sdoid = sdzp->z_id;
 520         lr->lr_tdoid = tdzp->z_id;
 521         memcpy((char *)(lr + 1), sname, snamesize);
 522         memcpy((char *)(lr + 1) + snamesize, dname, dnamesize);
 523         itx->itx_oid = szp->z_id;
 524
 525         zil_itx_assign(zilog, itx, tx);
 526 }
 527
 528 /*
 529  * zfs_log_write() handles TX_WRITE transactions. The specified callback is
 530  * called as soon as the write is on stable storage (be it via a DMU sync or a
 531  * ZIL commit).
 532  */
 533 static long zfs_immediate_write_sz = 32768;
 534
 535 void
 536 zfs_log_write(zilog_t *zilog, dmu_tx_t *tx, int txtype,
 537     znode_t *zp, offset_t off, ssize_t resid, int ioflag,
 538     zil_callback_t callback, void *callback_data)
 539 {
 540         dmu_buf_impl_t *db = (dmu_buf_impl_t *)sa_get_db(zp->z_sa_hdl);
 541         uint32_t blocksize = zp->z_blksz;
 542         itx_wr_state_t write_state;
 543         uintptr_t fsync_cnt;
 544         uint64_t gen = 0;
 545         ssize_t size = resid;
 546
 547         if (zil_replaying(zilog, tx) || zp->z_unlinked ||
 548             zfs_xattr_owner_unlinked(zp)) {
 549                 if (callback != NULL)
 550                         callback(callback_data);
 551                 return;
 552         }
 553
 554         if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
 555                 write_state = WR_INDIRECT;
 556         else if (!spa_has_slogs(zilog->zl_spa) &&
 557             resid >= zfs_immediate_write_sz)
 558                 write_state = WR_INDIRECT;
 559         else if (ioflag & (O_SYNC | O_DSYNC))
 560                 write_state = WR_COPIED;
 561         else
 562                 write_state = WR_NEED_COPY;
 563
 564         if ((fsync_cnt = (uintptr_t)tsd_get(zfs_fsyncer_key)) != 0) {
 565                 (void) tsd_set(zfs_fsyncer_key, (void *)(fsync_cnt - 1));
 566         }
 567
 568         (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(ZTOZSB(zp)), &gen,
 569             sizeof (gen));
 570
 571         while (resid) {
 572                 itx_t *itx;
 573                 lr_write_t *lr;
 574                 itx_wr_state_t wr_state = write_state;
 575                 ssize_t len = resid;
 576
 577                 /*
 578                  * A WR_COPIED record must fit entirely in one log block.
 579                  * Large writes can use WR_NEED_COPY, which the ZIL will
 580                  * split into multiple records across several log blocks
 581                  * if necessary.
 582                  */
 583                 if (wr_state == WR_COPIED &&
 584                     resid > zil_max_copied_data(zilog))
 585                         wr_state = WR_NEED_COPY;
 586                 else if (wr_state == WR_INDIRECT)
 587                         len = MIN(blocksize - P2PHASE(off, blocksize), resid);
 588
 589                 itx = zil_itx_create(txtype, sizeof (*lr) +
 590                     (wr_state == WR_COPIED ? len : 0));
 591                 lr = (lr_write_t *)&itx->itx_lr;
 592
 593                 /*
 594                  * For WR_COPIED records, copy the data into the lr_write_t.
 595                  */
 596                 if (wr_state == WR_COPIED) {
 597                         int err;
 598                         DB_DNODE_ENTER(db);
 599                         err = dmu_read_by_dnode(DB_DNODE(db), off, len, lr + 1,
 600                             DMU_READ_NO_PREFETCH);
 601                         if (err != 0) {
 602                                 zil_itx_destroy(itx);
 603                                 itx = zil_itx_create(txtype, sizeof (*lr));
 604                                 lr = (lr_write_t *)&itx->itx_lr;
 605                                 wr_state = WR_NEED_COPY;
 606                         }
 607                         DB_DNODE_EXIT(db);
 608                 }
 609
 610                 itx->itx_wr_state = wr_state;
 611                 lr->lr_foid = zp->z_id;
 612                 lr->lr_offset = off;
 613                 lr->lr_length = len;
 614                 lr->lr_blkoff = 0;
 615                 BP_ZERO(&lr->lr_blkptr);
 616
 617                 itx->itx_private = ZTOZSB(zp);
 618                 itx->itx_gen = gen;
 619
 620                 if (!(ioflag & (O_SYNC | O_DSYNC)) && (zp->z_sync_cnt == 0) &&
 621                     (fsync_cnt == 0))
 622                         itx->itx_sync = B_FALSE;
 623
 624                 itx->itx_callback = callback;
 625                 itx->itx_callback_data = callback_data;
 626                 zil_itx_assign(zilog, itx, tx);
 627
 628                 off += len;
 629                 resid -= len;
 630         }
 631
 632         if (write_state == WR_COPIED || write_state == WR_NEED_COPY) {
 633                 dsl_pool_wrlog_count(zilog->zl_dmu_pool, size, tx->tx_txg);
 634         }
 635 }
 636
 637 /*
 638  * Handles TX_TRUNCATE transactions.
 639  */
 640 void
 641 zfs_log_truncate(zilog_t *zilog, dmu_tx_t *tx, int txtype,
 642     znode_t *zp, uint64_t off, uint64_t len)
 643 {
 644         itx_t *itx;
 645         lr_truncate_t *lr;
 646
 647         if (zil_replaying(zilog, tx) || zp->z_unlinked ||
 648             zfs_xattr_owner_unlinked(zp))
 649                 return;
 650
 651         itx = zil_itx_create(txtype, sizeof (*lr));
 652         lr = (lr_truncate_t *)&itx->itx_lr;
 653         lr->lr_foid = zp->z_id;
 654         lr->lr_offset = off;
 655         lr->lr_length = len;
 656
 657         itx->itx_sync = (zp->z_sync_cnt != 0);
 658         zil_itx_assign(zilog, itx, tx);
 659 }
 660
 661 /*
 662  * Handles TX_SETATTR transactions.
 663  */
 664 void
 665 zfs_log_setattr(zilog_t *zilog, dmu_tx_t *tx, int txtype,
 666     znode_t *zp, vattr_t *vap, uint_t mask_applied, zfs_fuid_info_t *fuidp)
 667 {
 668         itx_t           *itx;
 669         lr_setattr_t    *lr;
 670         xvattr_t        *xvap = (xvattr_t *)vap;
 671         size_t          recsize = sizeof (lr_setattr_t);
 672         void            *start;
 673
 674         if (zil_replaying(zilog, tx) || zp->z_unlinked)
 675                 return;
 676
 677         /*
 678          * If XVATTR set, then log record size needs to allow
 679          * for lr_attr_t + xvattr mask, mapsize and create time
 680          * plus actual attribute values
 681          */
 682         if (vap->va_mask & ATTR_XVATTR)
 683                 recsize = sizeof (*lr) + ZIL_XVAT_SIZE(xvap->xva_mapsize);
 684
 685         if (fuidp)
 686                 recsize += fuidp->z_domain_str_sz;
 687
 688         itx = zil_itx_create(txtype, recsize);
 689         lr = (lr_setattr_t *)&itx->itx_lr;
 690         lr->lr_foid = zp->z_id;
 691         lr->lr_mask = (uint64_t)mask_applied;
 692         lr->lr_mode = (uint64_t)vap->va_mode;
 693         if ((mask_applied & ATTR_UID) && IS_EPHEMERAL(vap->va_uid))
 694                 lr->lr_uid = fuidp->z_fuid_owner;
 695         else
 696                 lr->lr_uid = (uint64_t)vap->va_uid;
 697
 698         if ((mask_applied & ATTR_GID) && IS_EPHEMERAL(vap->va_gid))
 699                 lr->lr_gid = fuidp->z_fuid_group;
 700         else
 701                 lr->lr_gid = (uint64_t)vap->va_gid;
 702
 703         lr->lr_size = (uint64_t)vap->va_size;
 704         ZFS_TIME_ENCODE(&vap->va_atime, lr->lr_atime);
 705         ZFS_TIME_ENCODE(&vap->va_mtime, lr->lr_mtime);
 706         start = (lr_setattr_t *)(lr + 1);
 707         if (vap->va_mask & ATTR_XVATTR) {
 708                 zfs_log_xvattr((lr_attr_t *)start, xvap);
 709                 start = (caddr_t)start + ZIL_XVAT_SIZE(xvap->xva_mapsize);
 710         }
 711
 712         /*
 713          * Now stick on domain information if any on end
 714          */
 715
 716         if (fuidp)
 717                 (void) zfs_log_fuid_domains(fuidp, start);
 718
 719         itx->itx_sync = (zp->z_sync_cnt != 0);
 720         zil_itx_assign(zilog, itx, tx);
 721 }
 722
 723 /*
 724  * Handles TX_SETSAXATTR transactions.
 725  */
 726 void
 727 zfs_log_setsaxattr(zilog_t *zilog, dmu_tx_t *tx, int txtype,
 728     znode_t *zp, const char *name, const void *value, size_t size)
 729 {
 730         itx_t           *itx;
 731         lr_setsaxattr_t *lr;
 732         size_t          recsize = sizeof (lr_setsaxattr_t);
 733         void            *xattrstart;
 734         int             namelen;
 735
 736         if (zil_replaying(zilog, tx) || zp->z_unlinked)
 737                 return;
 738
 739         namelen = strlen(name) + 1;
 740         recsize += (namelen + size);
 741         itx = zil_itx_create(txtype, recsize);
 742         lr = (lr_setsaxattr_t *)&itx->itx_lr;
 743         lr->lr_foid = zp->z_id;
 744         xattrstart = (char *)(lr + 1);
 745         memcpy(xattrstart, name, namelen);
 746         if (value != NULL) {
 747                 memcpy((char *)xattrstart + namelen, value, size);
 748                 lr->lr_size = size;
 749         } else {
 750                 lr->lr_size = 0;
 751         }
 752
 753         itx->itx_sync = (zp->z_sync_cnt != 0);
 754         zil_itx_assign(zilog, itx, tx);
 755 }
 756
 757 /*
 758  * Handles TX_ACL transactions.
 759  */
 760 void
 761 zfs_log_acl(zilog_t *zilog, dmu_tx_t *tx, znode_t *zp,
 762     vsecattr_t *vsecp, zfs_fuid_info_t *fuidp)
 763 {
 764         itx_t *itx;
 765         lr_acl_v0_t *lrv0;
 766         lr_acl_t *lr;
 767         int txtype;
 768         int lrsize;
 769         size_t txsize;
 770         size_t aclbytes = vsecp->vsa_aclentsz;
 771
 772         if (zil_replaying(zilog, tx) || zp->z_unlinked)
 773                 return;
 774
 775         txtype = (ZTOZSB(zp)->z_version < ZPL_VERSION_FUID) ?
 776             TX_ACL_V0 : TX_ACL;
 777
 778         if (txtype == TX_ACL)
 779                 lrsize = sizeof (*lr);
 780         else
 781                 lrsize = sizeof (*lrv0);
 782
 783         txsize = lrsize +
 784             ((txtype == TX_ACL) ? ZIL_ACE_LENGTH(aclbytes) : aclbytes) +
 785             (fuidp ? fuidp->z_domain_str_sz : 0) +
 786             sizeof (uint64_t) * (fuidp ? fuidp->z_fuid_cnt : 0);
 787
 788         itx = zil_itx_create(txtype, txsize);
 789
 790         lr = (lr_acl_t *)&itx->itx_lr;
 791         lr->lr_foid = zp->z_id;
 792         if (txtype == TX_ACL) {
 793                 lr->lr_acl_bytes = aclbytes;
 794                 lr->lr_domcnt = fuidp ? fuidp->z_domain_cnt : 0;
 795                 lr->lr_fuidcnt = fuidp ? fuidp->z_fuid_cnt : 0;
 796                 if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS)
 797                         lr->lr_acl_flags = (uint64_t)vsecp->vsa_aclflags;
 798                 else
 799                         lr->lr_acl_flags = 0;
 800         }
 801         lr->lr_aclcnt = (uint64_t)vsecp->vsa_aclcnt;
 802
 803         if (txtype == TX_ACL_V0) {
 804                 lrv0 = (lr_acl_v0_t *)lr;
 805                 memcpy(lrv0 + 1, vsecp->vsa_aclentp, aclbytes);
 806         } else {
 807                 void *start = (ace_t *)(lr + 1);
 808
 809                 memcpy(start, vsecp->vsa_aclentp, aclbytes);
 810
 811                 start = (caddr_t)start + ZIL_ACE_LENGTH(aclbytes);
 812
 813                 if (fuidp) {
 814                         start = zfs_log_fuid_ids(fuidp, start);
 815                         (void) zfs_log_fuid_domains(fuidp, start);
 816                 }
 817         }
 818
 819         itx->itx_sync = (zp->z_sync_cnt != 0);
 820         zil_itx_assign(zilog, itx, tx);
 821 }
 822
 823 ZFS_MODULE_PARAM(zfs, zfs_, immediate_write_sz, LONG, ZMOD_RW,
 824         "Largest data block to write to zil");