Print this page
    
OS-5148 ftruncate at offset should emit proper events
Reviewed by: Bryan Cantrill <bryan@joyent.com>
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
OS-3294 add support for inotify
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
Reviewed by: Robert Mustacchi <rm@joyent.com>
    
      
        | Split | 
	Close | 
      
      | Expand all | 
      | Collapse all | 
    
    
          --- old/usr/src/uts/common/fs/nfs/nfs_vnops.c
          +++ new/usr/src/uts/common/fs/nfs/nfs_vnops.c
   1    1  /*
   2    2   * CDDL HEADER START
   3    3   *
   4    4   * The contents of this file are subject to the terms of the
   5    5   * Common Development and Distribution License (the "License").
   6    6   * You may not use this file except in compliance with the License.
   7    7   *
   8    8   * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9    9   * or http://www.opensolaris.org/os/licensing.
  10   10   * See the License for the specific language governing permissions
  11   11   * and limitations under the License.
  12   12   *
  13   13   * When distributing Covered Code, include this CDDL HEADER in each
  14   14   * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15   15   * If applicable, add the following below this CDDL HEADER, with the
  16   16   * fields enclosed by brackets "[]" replaced with your own identifying
  17   17   * information: Portions Copyright [yyyy] [name of copyright owner]
  18   18   *
  
    | 
      ↓ open down ↓ | 
    18 lines elided | 
    
      ↑ open up ↑ | 
  
  19   19   * CDDL HEADER END
  20   20   */
  21   21  /*
  22   22   * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
  23   23   *
  24   24   *      Copyright (c) 1983,1984,1985,1986,1987,1988,1989 AT&T.
  25   25   *      All rights reserved.
  26   26   */
  27   27  
  28   28  /*
  29      - * Copyright (c) 2013, Joyent, Inc. All rights reserved.
       29 + * Copyright (c) 2014, Joyent, Inc. All rights reserved.
  30   30   * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  31   31   */
  32   32  
  33   33  #include <sys/param.h>
  34   34  #include <sys/types.h>
  35   35  #include <sys/systm.h>
  36   36  #include <sys/cred.h>
  37   37  #include <sys/time.h>
  38   38  #include <sys/vnode.h>
  39   39  #include <sys/vfs.h>
  40   40  #include <sys/vfs_opreg.h>
  41   41  #include <sys/file.h>
  42   42  #include <sys/filio.h>
  43   43  #include <sys/uio.h>
  44   44  #include <sys/buf.h>
  45   45  #include <sys/mman.h>
  46   46  #include <sys/pathname.h>
  47   47  #include <sys/dirent.h>
  48   48  #include <sys/debug.h>
  49   49  #include <sys/vmsystm.h>
  50   50  #include <sys/fcntl.h>
  51   51  #include <sys/flock.h>
  52   52  #include <sys/swap.h>
  53   53  #include <sys/errno.h>
  54   54  #include <sys/strsubr.h>
  55   55  #include <sys/sysmacros.h>
  56   56  #include <sys/kmem.h>
  57   57  #include <sys/cmn_err.h>
  58   58  #include <sys/pathconf.h>
  59   59  #include <sys/utsname.h>
  60   60  #include <sys/dnlc.h>
  61   61  #include <sys/acl.h>
  62   62  #include <sys/atomic.h>
  63   63  #include <sys/policy.h>
  64   64  #include <sys/sdt.h>
  65   65  
  66   66  #include <rpc/types.h>
  67   67  #include <rpc/auth.h>
  68   68  #include <rpc/clnt.h>
  69   69  
  70   70  #include <nfs/nfs.h>
  71   71  #include <nfs/nfs_clnt.h>
  72   72  #include <nfs/rnode.h>
  73   73  #include <nfs/nfs_acl.h>
  74   74  #include <nfs/lm.h>
  75   75  
  76   76  #include <vm/hat.h>
  77   77  #include <vm/as.h>
  78   78  #include <vm/page.h>
  79   79  #include <vm/pvn.h>
  80   80  #include <vm/seg.h>
  81   81  #include <vm/seg_map.h>
  82   82  #include <vm/seg_kpm.h>
  83   83  #include <vm/seg_vn.h>
  84   84  
  85   85  #include <fs/fs_subr.h>
  86   86  
  87   87  #include <sys/ddi.h>
  88   88  
  89   89  static int      nfs_rdwrlbn(vnode_t *, page_t *, u_offset_t, size_t, int,
  90   90                          cred_t *);
  91   91  static int      nfswrite(vnode_t *, caddr_t, uint_t, int, cred_t *);
  92   92  static int      nfsread(vnode_t *, caddr_t, uint_t, int, size_t *, cred_t *);
  93   93  static int      nfssetattr(vnode_t *, struct vattr *, int, cred_t *);
  94   94  static int      nfslookup_dnlc(vnode_t *, char *, vnode_t **, cred_t *);
  95   95  static int      nfslookup_otw(vnode_t *, char *, vnode_t **, cred_t *, int);
  96   96  static int      nfsrename(vnode_t *, char *, vnode_t *, char *, cred_t *,
  97   97                          caller_context_t *);
  98   98  static int      nfsreaddir(vnode_t *, rddir_cache *, cred_t *);
  99   99  static int      nfs_bio(struct buf *, cred_t *);
 100  100  static int      nfs_getapage(vnode_t *, u_offset_t, size_t, uint_t *,
 101  101                          page_t *[], size_t, struct seg *, caddr_t,
 102  102                          enum seg_rw, cred_t *);
 103  103  static void     nfs_readahead(vnode_t *, u_offset_t, caddr_t, struct seg *,
 104  104                          cred_t *);
 105  105  static int      nfs_sync_putapage(vnode_t *, page_t *, u_offset_t, size_t,
 106  106                          int, cred_t *);
 107  107  static int      nfs_sync_pageio(vnode_t *, page_t *, u_offset_t, size_t,
 108  108                          int, cred_t *);
 109  109  static void     nfs_delmap_callback(struct as *, void *, uint_t);
 110  110  
 111  111  /*
 112  112   * Error flags used to pass information about certain special errors
 113  113   * which need to be handled specially.
 114  114   */
 115  115  #define NFS_EOF                 -98
 116  116  
 117  117  /*
 118  118   * These are the vnode ops routines which implement the vnode interface to
 119  119   * the networked file system.  These routines just take their parameters,
 120  120   * make them look networkish by putting the right info into interface structs,
 121  121   * and then calling the appropriate remote routine(s) to do the work.
 122  122   *
 123  123   * Note on directory name lookup cacheing:  If we detect a stale fhandle,
 124  124   * we purge the directory cache relative to that vnode.  This way, the
 125  125   * user won't get burned by the cache repeatedly.  See <nfs/rnode.h> for
 126  126   * more details on rnode locking.
 127  127   */
 128  128  
 129  129  static int      nfs_open(vnode_t **, int, cred_t *, caller_context_t *);
 130  130  static int      nfs_close(vnode_t *, int, int, offset_t, cred_t *,
 131  131                          caller_context_t *);
 132  132  static int      nfs_read(vnode_t *, struct uio *, int, cred_t *,
 133  133                          caller_context_t *);
 134  134  static int      nfs_write(vnode_t *, struct uio *, int, cred_t *,
 135  135                          caller_context_t *);
 136  136  static int      nfs_ioctl(vnode_t *, int, intptr_t, int, cred_t *, int *,
 137  137                          caller_context_t *);
 138  138  static int      nfs_getattr(vnode_t *, struct vattr *, int, cred_t *,
 139  139                          caller_context_t *);
 140  140  static int      nfs_setattr(vnode_t *, struct vattr *, int, cred_t *,
 141  141                          caller_context_t *);
 142  142  static int      nfs_access(vnode_t *, int, int, cred_t *, caller_context_t *);
 143  143  static int      nfs_accessx(void *, int, cred_t *);
 144  144  static int      nfs_readlink(vnode_t *, struct uio *, cred_t *,
 145  145                          caller_context_t *);
 146  146  static int      nfs_fsync(vnode_t *, int, cred_t *, caller_context_t *);
 147  147  static void     nfs_inactive(vnode_t *, cred_t *, caller_context_t *);
 148  148  static int      nfs_lookup(vnode_t *, char *, vnode_t **, struct pathname *,
 149  149                          int, vnode_t *, cred_t *, caller_context_t *,
 150  150                          int *, pathname_t *);
 151  151  static int      nfs_create(vnode_t *, char *, struct vattr *, enum vcexcl,
 152  152                          int, vnode_t **, cred_t *, int, caller_context_t *,
 153  153                          vsecattr_t *);
 154  154  static int      nfs_remove(vnode_t *, char *, cred_t *, caller_context_t *,
 155  155                          int);
 156  156  static int      nfs_link(vnode_t *, vnode_t *, char *, cred_t *,
 157  157                          caller_context_t *, int);
 158  158  static int      nfs_rename(vnode_t *, char *, vnode_t *, char *, cred_t *,
 159  159                          caller_context_t *, int);
 160  160  static int      nfs_mkdir(vnode_t *, char *, struct vattr *, vnode_t **,
 161  161                          cred_t *, caller_context_t *, int, vsecattr_t *);
 162  162  static int      nfs_rmdir(vnode_t *, char *, vnode_t *, cred_t *,
 163  163                          caller_context_t *, int);
 164  164  static int      nfs_symlink(vnode_t *, char *, struct vattr *, char *,
 165  165                          cred_t *, caller_context_t *, int);
 166  166  static int      nfs_readdir(vnode_t *, struct uio *, cred_t *, int *,
 167  167                          caller_context_t *, int);
 168  168  static int      nfs_fid(vnode_t *, fid_t *, caller_context_t *);
 169  169  static int      nfs_rwlock(vnode_t *, int, caller_context_t *);
 170  170  static void     nfs_rwunlock(vnode_t *, int, caller_context_t *);
 171  171  static int      nfs_seek(vnode_t *, offset_t, offset_t *, caller_context_t *);
 172  172  static int      nfs_getpage(vnode_t *, offset_t, size_t, uint_t *,
 173  173                          page_t *[], size_t, struct seg *, caddr_t,
 174  174                          enum seg_rw, cred_t *, caller_context_t *);
 175  175  static int      nfs_putpage(vnode_t *, offset_t, size_t, int, cred_t *,
 176  176                          caller_context_t *);
 177  177  static int      nfs_map(vnode_t *, offset_t, struct as *, caddr_t *, size_t,
 178  178                          uchar_t, uchar_t, uint_t, cred_t *, caller_context_t *);
 179  179  static int      nfs_addmap(vnode_t *, offset_t, struct as *, caddr_t, size_t,
 180  180                          uchar_t, uchar_t, uint_t, cred_t *, caller_context_t *);
 181  181  static int      nfs_frlock(vnode_t *, int, struct flock64 *, int, offset_t,
 182  182                          struct flk_callback *, cred_t *, caller_context_t *);
 183  183  static int      nfs_space(vnode_t *, int, struct flock64 *, int, offset_t,
 184  184                          cred_t *, caller_context_t *);
 185  185  static int      nfs_realvp(vnode_t *, vnode_t **, caller_context_t *);
 186  186  static int      nfs_delmap(vnode_t *, offset_t, struct as *, caddr_t, size_t,
 187  187                          uint_t, uint_t, uint_t, cred_t *, caller_context_t *);
 188  188  static int      nfs_pathconf(vnode_t *, int, ulong_t *, cred_t *,
 189  189                          caller_context_t *);
 190  190  static int      nfs_pageio(vnode_t *, page_t *, u_offset_t, size_t, int,
 191  191                          cred_t *, caller_context_t *);
 192  192  static int      nfs_setsecattr(vnode_t *, vsecattr_t *, int, cred_t *,
 193  193                          caller_context_t *);
 194  194  static int      nfs_getsecattr(vnode_t *, vsecattr_t *, int, cred_t *,
 195  195                          caller_context_t *);
 196  196  static int      nfs_shrlock(vnode_t *, int, struct shrlock *, int, cred_t *,
 197  197                          caller_context_t *);
 198  198  
 199  199  struct vnodeops *nfs_vnodeops;
 200  200  
 201  201  const fs_operation_def_t nfs_vnodeops_template[] = {
 202  202          VOPNAME_OPEN,           { .vop_open = nfs_open },
 203  203          VOPNAME_CLOSE,          { .vop_close = nfs_close },
 204  204          VOPNAME_READ,           { .vop_read = nfs_read },
 205  205          VOPNAME_WRITE,          { .vop_write = nfs_write },
 206  206          VOPNAME_IOCTL,          { .vop_ioctl = nfs_ioctl },
 207  207          VOPNAME_GETATTR,        { .vop_getattr = nfs_getattr },
 208  208          VOPNAME_SETATTR,        { .vop_setattr = nfs_setattr },
 209  209          VOPNAME_ACCESS,         { .vop_access = nfs_access },
 210  210          VOPNAME_LOOKUP,         { .vop_lookup = nfs_lookup },
 211  211          VOPNAME_CREATE,         { .vop_create = nfs_create },
 212  212          VOPNAME_REMOVE,         { .vop_remove = nfs_remove },
 213  213          VOPNAME_LINK,           { .vop_link = nfs_link },
 214  214          VOPNAME_RENAME,         { .vop_rename = nfs_rename },
 215  215          VOPNAME_MKDIR,          { .vop_mkdir = nfs_mkdir },
 216  216          VOPNAME_RMDIR,          { .vop_rmdir = nfs_rmdir },
 217  217          VOPNAME_READDIR,        { .vop_readdir = nfs_readdir },
 218  218          VOPNAME_SYMLINK,        { .vop_symlink = nfs_symlink },
 219  219          VOPNAME_READLINK,       { .vop_readlink = nfs_readlink },
 220  220          VOPNAME_FSYNC,          { .vop_fsync = nfs_fsync },
 221  221          VOPNAME_INACTIVE,       { .vop_inactive = nfs_inactive },
 222  222          VOPNAME_FID,            { .vop_fid = nfs_fid },
 223  223          VOPNAME_RWLOCK,         { .vop_rwlock = nfs_rwlock },
 224  224          VOPNAME_RWUNLOCK,       { .vop_rwunlock = nfs_rwunlock },
 225  225          VOPNAME_SEEK,           { .vop_seek = nfs_seek },
 226  226          VOPNAME_FRLOCK,         { .vop_frlock = nfs_frlock },
 227  227          VOPNAME_SPACE,          { .vop_space = nfs_space },
 228  228          VOPNAME_REALVP,         { .vop_realvp = nfs_realvp },
 229  229          VOPNAME_GETPAGE,        { .vop_getpage = nfs_getpage },
 230  230          VOPNAME_PUTPAGE,        { .vop_putpage = nfs_putpage },
 231  231          VOPNAME_MAP,            { .vop_map = nfs_map },
 232  232          VOPNAME_ADDMAP,         { .vop_addmap = nfs_addmap },
 233  233          VOPNAME_DELMAP,         { .vop_delmap = nfs_delmap },
 234  234          VOPNAME_DUMP,           { .vop_dump = nfs_dump },
 235  235          VOPNAME_PATHCONF,       { .vop_pathconf = nfs_pathconf },
 236  236          VOPNAME_PAGEIO,         { .vop_pageio = nfs_pageio },
 237  237          VOPNAME_SETSECATTR,     { .vop_setsecattr = nfs_setsecattr },
 238  238          VOPNAME_GETSECATTR,     { .vop_getsecattr = nfs_getsecattr },
 239  239          VOPNAME_SHRLOCK,        { .vop_shrlock = nfs_shrlock },
 240  240          VOPNAME_VNEVENT,        { .vop_vnevent = fs_vnevent_support },
 241  241          NULL,                   NULL
 242  242  };
 243  243  
 244  244  /*
 245  245   * XXX:  This is referenced in modstubs.s
 246  246   */
 247  247  struct vnodeops *
 248  248  nfs_getvnodeops(void)
 249  249  {
 250  250          return (nfs_vnodeops);
 251  251  }
 252  252  
 253  253  /* ARGSUSED */
 254  254  static int
 255  255  nfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
 256  256  {
 257  257          int error;
 258  258          struct vattr va;
 259  259          rnode_t *rp;
 260  260          vnode_t *vp;
 261  261  
 262  262          vp = *vpp;
 263  263          rp = VTOR(vp);
 264  264          if (nfs_zone() != VTOMI(vp)->mi_zone)
 265  265                  return (EIO);
 266  266          mutex_enter(&rp->r_statelock);
 267  267          if (rp->r_cred == NULL) {
 268  268                  crhold(cr);
 269  269                  rp->r_cred = cr;
 270  270          }
 271  271          mutex_exit(&rp->r_statelock);
 272  272  
 273  273          /*
 274  274           * If there is no cached data or if close-to-open
 275  275           * consistency checking is turned off, we can avoid
 276  276           * the over the wire getattr.  Otherwise, if the
 277  277           * file system is mounted readonly, then just verify
 278  278           * the caches are up to date using the normal mechanism.
 279  279           * Else, if the file is not mmap'd, then just mark
 280  280           * the attributes as timed out.  They will be refreshed
 281  281           * and the caches validated prior to being used.
 282  282           * Else, the file system is mounted writeable so
 283  283           * force an over the wire GETATTR in order to ensure
 284  284           * that all cached data is valid.
 285  285           */
 286  286          if (vp->v_count > 1 ||
 287  287              ((vn_has_cached_data(vp) || HAVE_RDDIR_CACHE(rp)) &&
 288  288              !(VTOMI(vp)->mi_flags & MI_NOCTO))) {
 289  289                  if (vn_is_readonly(vp))
 290  290                          error = nfs_validate_caches(vp, cr);
 291  291                  else if (rp->r_mapcnt == 0 && vp->v_count == 1) {
 292  292                          PURGE_ATTRCACHE(vp);
 293  293                          error = 0;
 294  294                  } else {
 295  295                          va.va_mask = AT_ALL;
 296  296                          error = nfs_getattr_otw(vp, &va, cr);
 297  297                  }
 298  298          } else
 299  299                  error = 0;
 300  300  
 301  301          return (error);
 302  302  }
 303  303  
 304  304  /* ARGSUSED */
 305  305  static int
 306  306  nfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
 307  307          caller_context_t *ct)
 308  308  {
 309  309          rnode_t *rp;
 310  310          int error;
 311  311          struct vattr va;
 312  312  
 313  313          /*
 314  314           * zone_enter(2) prevents processes from changing zones with NFS files
 315  315           * open; if we happen to get here from the wrong zone we can't do
 316  316           * anything over the wire.
 317  317           */
 318  318          if (VTOMI(vp)->mi_zone != nfs_zone()) {
 319  319                  /*
 320  320                   * We could attempt to clean up locks, except we're sure
 321  321                   * that the current process didn't acquire any locks on
 322  322                   * the file: any attempt to lock a file belong to another zone
 323  323                   * will fail, and one can't lock an NFS file and then change
 324  324                   * zones, as that fails too.
 325  325                   *
 326  326                   * Returning an error here is the sane thing to do.  A
 327  327                   * subsequent call to VN_RELE() which translates to a
 328  328                   * nfs_inactive() will clean up state: if the zone of the
 329  329                   * vnode's origin is still alive and kicking, an async worker
 330  330                   * thread will handle the request (from the correct zone), and
 331  331                   * everything (minus the final nfs_getattr_otw() call) should
 332  332                   * be OK. If the zone is going away nfs_async_inactive() will
 333  333                   * throw away cached pages inline.
 334  334                   */
 335  335                  return (EIO);
 336  336          }
 337  337  
 338  338          /*
 339  339           * If we are using local locking for this filesystem, then
 340  340           * release all of the SYSV style record locks.  Otherwise,
 341  341           * we are doing network locking and we need to release all
 342  342           * of the network locks.  All of the locks held by this
 343  343           * process on this file are released no matter what the
 344  344           * incoming reference count is.
 345  345           */
 346  346          if (VTOMI(vp)->mi_flags & MI_LLOCK) {
 347  347                  cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
 348  348                  cleanshares(vp, ttoproc(curthread)->p_pid);
 349  349          } else
 350  350                  nfs_lockrelease(vp, flag, offset, cr);
 351  351  
 352  352          if (count > 1)
 353  353                  return (0);
 354  354  
 355  355          /*
 356  356           * If the file has been `unlinked', then purge the
 357  357           * DNLC so that this vnode will get reycled quicker
 358  358           * and the .nfs* file on the server will get removed.
 359  359           */
 360  360          rp = VTOR(vp);
 361  361          if (rp->r_unldvp != NULL)
 362  362                  dnlc_purge_vp(vp);
 363  363  
 364  364          /*
 365  365           * If the file was open for write and there are pages,
 366  366           * then if the file system was mounted using the "no-close-
 367  367           *      to-open" semantics, then start an asynchronous flush
 368  368           *      of the all of the pages in the file.
 369  369           * else the file system was not mounted using the "no-close-
 370  370           *      to-open" semantics, then do a synchronous flush and
 371  371           *      commit of all of the dirty and uncommitted pages.
 372  372           *
 373  373           * The asynchronous flush of the pages in the "nocto" path
 374  374           * mostly just associates a cred pointer with the rnode so
 375  375           * writes which happen later will have a better chance of
 376  376           * working.  It also starts the data being written to the
 377  377           * server, but without unnecessarily delaying the application.
 378  378           */
 379  379          if ((flag & FWRITE) && vn_has_cached_data(vp)) {
 380  380                  if ((VTOMI(vp)->mi_flags & MI_NOCTO)) {
 381  381                          error = nfs_putpage(vp, (offset_t)0, 0, B_ASYNC,
 382  382                              cr, ct);
 383  383                          if (error == EAGAIN)
 384  384                                  error = 0;
 385  385                  } else
 386  386                          error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
 387  387                  if (!error) {
 388  388                          mutex_enter(&rp->r_statelock);
 389  389                          error = rp->r_error;
 390  390                          rp->r_error = 0;
 391  391                          mutex_exit(&rp->r_statelock);
 392  392                  }
 393  393          } else {
 394  394                  mutex_enter(&rp->r_statelock);
 395  395                  error = rp->r_error;
 396  396                  rp->r_error = 0;
 397  397                  mutex_exit(&rp->r_statelock);
 398  398          }
 399  399  
 400  400          /*
 401  401           * If RWRITEATTR is set, then issue an over the wire GETATTR to
 402  402           * refresh the attribute cache with a set of attributes which
 403  403           * weren't returned from a WRITE.  This will enable the close-
 404  404           * to-open processing to work.
 405  405           */
 406  406          if (rp->r_flags & RWRITEATTR)
 407  407                  (void) nfs_getattr_otw(vp, &va, cr);
 408  408  
 409  409          return (error);
 410  410  }
 411  411  
 412  412  /* ARGSUSED */
 413  413  static int
 414  414  nfs_read(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
 415  415          caller_context_t *ct)
 416  416  {
 417  417          rnode_t *rp;
 418  418          u_offset_t off;
 419  419          offset_t diff;
 420  420          int on;
 421  421          size_t n;
 422  422          caddr_t base;
 423  423          uint_t flags;
 424  424          int error;
 425  425          mntinfo_t *mi;
 426  426  
 427  427          rp = VTOR(vp);
 428  428          mi = VTOMI(vp);
 429  429  
 430  430          if (nfs_zone() != mi->mi_zone)
 431  431                  return (EIO);
 432  432  
 433  433          ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER));
 434  434  
 435  435          if (vp->v_type != VREG)
 436  436                  return (EISDIR);
 437  437  
 438  438          if (uiop->uio_resid == 0)
 439  439                  return (0);
 440  440  
 441  441          if (uiop->uio_loffset > MAXOFF32_T)
 442  442                  return (EFBIG);
 443  443  
 444  444          if (uiop->uio_loffset < 0 ||
 445  445              uiop->uio_loffset + uiop->uio_resid > MAXOFF32_T)
 446  446                  return (EINVAL);
 447  447  
 448  448          /*
 449  449           * Bypass VM if caching has been disabled (e.g., locking) or if
 450  450           * using client-side direct I/O and the file is not mmap'd and
 451  451           * there are no cached pages.
 452  452           */
 453  453          if ((vp->v_flag & VNOCACHE) ||
 454  454              (((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) &&
 455  455              rp->r_mapcnt == 0 && rp->r_inmap == 0 &&
 456  456              !vn_has_cached_data(vp))) {
 457  457                  size_t bufsize;
 458  458                  size_t resid = 0;
 459  459  
 460  460                  /*
 461  461                   * Let's try to do read in as large a chunk as we can
 462  462                   * (Filesystem (NFS client) bsize if possible/needed).
 463  463                   * For V3, this is 32K and for V2, this is 8K.
 464  464                   */
 465  465                  bufsize = MIN(uiop->uio_resid, VTOMI(vp)->mi_curread);
 466  466                  base = kmem_alloc(bufsize, KM_SLEEP);
 467  467                  do {
 468  468                          n = MIN(uiop->uio_resid, bufsize);
 469  469                          error = nfsread(vp, base, uiop->uio_offset, n,
 470  470                              &resid, cr);
 471  471                          if (!error) {
 472  472                                  n -= resid;
 473  473                                  error = uiomove(base, n, UIO_READ, uiop);
 474  474                          }
 475  475                  } while (!error && uiop->uio_resid > 0 && n > 0);
 476  476                  kmem_free(base, bufsize);
 477  477                  return (error);
 478  478          }
 479  479  
 480  480          error = 0;
 481  481  
 482  482          do {
 483  483                  off = uiop->uio_loffset & MAXBMASK; /* mapping offset */
 484  484                  on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */
 485  485                  n = MIN(MAXBSIZE - on, uiop->uio_resid);
 486  486  
 487  487                  error = nfs_validate_caches(vp, cr);
 488  488                  if (error)
 489  489                          break;
 490  490  
 491  491                  mutex_enter(&rp->r_statelock);
 492  492                  while (rp->r_flags & RINCACHEPURGE) {
 493  493                          if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) {
 494  494                                  mutex_exit(&rp->r_statelock);
 495  495                                  return (EINTR);
 496  496                          }
 497  497                  }
 498  498                  diff = rp->r_size - uiop->uio_loffset;
 499  499                  mutex_exit(&rp->r_statelock);
 500  500                  if (diff <= 0)
 501  501                          break;
 502  502                  if (diff < n)
 503  503                          n = (size_t)diff;
 504  504  
 505  505                  if (vpm_enable) {
 506  506                          /*
 507  507                           * Copy data.
 508  508                           */
 509  509                          error = vpm_data_copy(vp, off + on, n, uiop,
 510  510                              1, NULL, 0, S_READ);
 511  511                  } else {
 512  512                          base = segmap_getmapflt(segkmap, vp, off + on, n,
 513  513                              1, S_READ);
 514  514                          error = uiomove(base + on, n, UIO_READ, uiop);
 515  515                  }
 516  516  
 517  517                  if (!error) {
 518  518                          /*
 519  519                           * If read a whole block or read to eof,
 520  520                           * won't need this buffer again soon.
 521  521                           */
 522  522                          mutex_enter(&rp->r_statelock);
 523  523                          if (n + on == MAXBSIZE ||
 524  524                              uiop->uio_loffset == rp->r_size)
 525  525                                  flags = SM_DONTNEED;
 526  526                          else
 527  527                                  flags = 0;
 528  528                          mutex_exit(&rp->r_statelock);
 529  529                          if (vpm_enable) {
 530  530                                  error = vpm_sync_pages(vp, off, n, flags);
 531  531                          } else {
 532  532                                  error = segmap_release(segkmap, base, flags);
 533  533                          }
 534  534                  } else {
 535  535                          if (vpm_enable) {
 536  536                                  (void) vpm_sync_pages(vp, off, n, 0);
 537  537                          } else {
 538  538                                  (void) segmap_release(segkmap, base, 0);
 539  539                          }
 540  540                  }
 541  541          } while (!error && uiop->uio_resid > 0);
 542  542  
 543  543          return (error);
 544  544  }
 545  545  
 546  546  /* ARGSUSED */
 547  547  static int
 548  548  nfs_write(vnode_t *vp, struct uio *uiop, int ioflag, cred_t *cr,
 549  549          caller_context_t *ct)
 550  550  {
 551  551          rnode_t *rp;
 552  552          u_offset_t off;
 553  553          caddr_t base;
 554  554          uint_t flags;
 555  555          int remainder;
 556  556          size_t n;
 557  557          int on;
 558  558          int error;
 559  559          int resid;
 560  560          offset_t offset;
 561  561          rlim_t limit;
 562  562          mntinfo_t *mi;
 563  563  
 564  564          rp = VTOR(vp);
 565  565  
 566  566          mi = VTOMI(vp);
 567  567          if (nfs_zone() != mi->mi_zone)
 568  568                  return (EIO);
 569  569          if (vp->v_type != VREG)
 570  570                  return (EISDIR);
 571  571  
 572  572          if (uiop->uio_resid == 0)
 573  573                  return (0);
 574  574  
 575  575          if (ioflag & FAPPEND) {
 576  576                  struct vattr va;
 577  577  
 578  578                  /*
 579  579                   * Must serialize if appending.
 580  580                   */
 581  581                  if (nfs_rw_lock_held(&rp->r_rwlock, RW_READER)) {
 582  582                          nfs_rw_exit(&rp->r_rwlock);
 583  583                          if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER,
 584  584                              INTR(vp)))
 585  585                                  return (EINTR);
 586  586                  }
 587  587  
 588  588                  va.va_mask = AT_SIZE;
 589  589                  error = nfsgetattr(vp, &va, cr);
 590  590                  if (error)
 591  591                          return (error);
 592  592                  uiop->uio_loffset = va.va_size;
 593  593          }
 594  594  
 595  595          if (uiop->uio_loffset > MAXOFF32_T)
 596  596                  return (EFBIG);
 597  597  
 598  598          offset = uiop->uio_loffset + uiop->uio_resid;
 599  599  
 600  600          if (uiop->uio_loffset < 0 || offset > MAXOFF32_T)
 601  601                  return (EINVAL);
 602  602  
 603  603          if (uiop->uio_llimit > (rlim64_t)MAXOFF32_T) {
 604  604                  limit = MAXOFF32_T;
 605  605          } else {
 606  606                  limit = (rlim_t)uiop->uio_llimit;
 607  607          }
 608  608  
 609  609          /*
 610  610           * Check to make sure that the process will not exceed
 611  611           * its limit on file size.  It is okay to write up to
 612  612           * the limit, but not beyond.  Thus, the write which
 613  613           * reaches the limit will be short and the next write
 614  614           * will return an error.
 615  615           */
 616  616          remainder = 0;
 617  617          if (offset > limit) {
 618  618                  remainder = offset - limit;
 619  619                  uiop->uio_resid = limit - uiop->uio_offset;
 620  620                  if (uiop->uio_resid <= 0) {
 621  621                          proc_t *p = ttoproc(curthread);
 622  622  
 623  623                          uiop->uio_resid += remainder;
 624  624                          mutex_enter(&p->p_lock);
 625  625                          (void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE],
 626  626                              p->p_rctls, p, RCA_UNSAFE_SIGINFO);
 627  627                          mutex_exit(&p->p_lock);
 628  628                          return (EFBIG);
 629  629                  }
 630  630          }
 631  631  
 632  632          if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR(vp)))
 633  633                  return (EINTR);
 634  634  
 635  635          /*
 636  636           * Bypass VM if caching has been disabled (e.g., locking) or if
 637  637           * using client-side direct I/O and the file is not mmap'd and
 638  638           * there are no cached pages.
 639  639           */
 640  640          if ((vp->v_flag & VNOCACHE) ||
 641  641              (((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO)) &&
 642  642              rp->r_mapcnt == 0 && rp->r_inmap == 0 &&
 643  643              !vn_has_cached_data(vp))) {
 644  644                  size_t bufsize;
 645  645                  int count;
 646  646                  uint_t org_offset;
 647  647  
 648  648  nfs_fwrite:
 649  649                  if (rp->r_flags & RSTALE) {
 650  650                          resid = uiop->uio_resid;
 651  651                          offset = uiop->uio_loffset;
 652  652                          error = rp->r_error;
 653  653                          /*
 654  654                           * A close may have cleared r_error, if so,
 655  655                           * propagate ESTALE error return properly
 656  656                           */
 657  657                          if (error == 0)
 658  658                                  error = ESTALE;
 659  659                          goto bottom;
 660  660                  }
 661  661                  bufsize = MIN(uiop->uio_resid, mi->mi_curwrite);
 662  662                  base = kmem_alloc(bufsize, KM_SLEEP);
 663  663                  do {
 664  664                          resid = uiop->uio_resid;
 665  665                          offset = uiop->uio_loffset;
 666  666                          count = MIN(uiop->uio_resid, bufsize);
 667  667                          org_offset = uiop->uio_offset;
 668  668                          error = uiomove(base, count, UIO_WRITE, uiop);
 669  669                          if (!error) {
 670  670                                  error = nfswrite(vp, base, org_offset,
 671  671                                      count, cr);
 672  672                          }
 673  673                  } while (!error && uiop->uio_resid > 0);
 674  674                  kmem_free(base, bufsize);
 675  675                  goto bottom;
 676  676          }
 677  677  
 678  678          do {
 679  679                  off = uiop->uio_loffset & MAXBMASK; /* mapping offset */
 680  680                  on = uiop->uio_loffset & MAXBOFFSET; /* Relative offset */
 681  681                  n = MIN(MAXBSIZE - on, uiop->uio_resid);
 682  682  
 683  683                  resid = uiop->uio_resid;
 684  684                  offset = uiop->uio_loffset;
 685  685  
 686  686                  if (rp->r_flags & RSTALE) {
 687  687                          error = rp->r_error;
 688  688                          /*
 689  689                           * A close may have cleared r_error, if so,
 690  690                           * propagate ESTALE error return properly
 691  691                           */
 692  692                          if (error == 0)
 693  693                                  error = ESTALE;
 694  694                          break;
 695  695                  }
 696  696  
 697  697                  /*
 698  698                   * Don't create dirty pages faster than they
 699  699                   * can be cleaned so that the system doesn't
 700  700                   * get imbalanced.  If the async queue is
 701  701                   * maxed out, then wait for it to drain before
 702  702                   * creating more dirty pages.  Also, wait for
 703  703                   * any threads doing pagewalks in the vop_getattr
 704  704                   * entry points so that they don't block for
 705  705                   * long periods.
 706  706                   */
 707  707                  mutex_enter(&rp->r_statelock);
 708  708                  while ((mi->mi_max_threads != 0 &&
 709  709                      rp->r_awcount > 2 * mi->mi_max_threads) ||
 710  710                      rp->r_gcount > 0) {
 711  711                          if (INTR(vp)) {
 712  712                                  klwp_t *lwp = ttolwp(curthread);
 713  713  
 714  714                                  if (lwp != NULL)
 715  715                                          lwp->lwp_nostop++;
 716  716                                  if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) {
 717  717                                          mutex_exit(&rp->r_statelock);
 718  718                                          if (lwp != NULL)
 719  719                                                  lwp->lwp_nostop--;
 720  720                                          error = EINTR;
 721  721                                          goto bottom;
 722  722                                  }
 723  723                                  if (lwp != NULL)
 724  724                                          lwp->lwp_nostop--;
 725  725                          } else
 726  726                                  cv_wait(&rp->r_cv, &rp->r_statelock);
 727  727                  }
 728  728                  mutex_exit(&rp->r_statelock);
 729  729  
 730  730                  /*
 731  731                   * Touch the page and fault it in if it is not in core
 732  732                   * before segmap_getmapflt or vpm_data_copy can lock it.
 733  733                   * This is to avoid the deadlock if the buffer is mapped
 734  734                   * to the same file through mmap which we want to write.
 735  735                   */
 736  736                  uio_prefaultpages((long)n, uiop);
 737  737  
 738  738                  if (vpm_enable) {
 739  739                          /*
 740  740                           * It will use kpm mappings, so no need to
 741  741                           * pass an address.
 742  742                           */
 743  743                          error = writerp(rp, NULL, n, uiop, 0);
 744  744                  } else  {
 745  745                          if (segmap_kpm) {
 746  746                                  int pon = uiop->uio_loffset & PAGEOFFSET;
 747  747                                  size_t pn = MIN(PAGESIZE - pon,
 748  748                                      uiop->uio_resid);
 749  749                                  int pagecreate;
 750  750  
 751  751                                  mutex_enter(&rp->r_statelock);
 752  752                                  pagecreate = (pon == 0) && (pn == PAGESIZE ||
 753  753                                      uiop->uio_loffset + pn >= rp->r_size);
 754  754                                  mutex_exit(&rp->r_statelock);
 755  755  
 756  756                                  base = segmap_getmapflt(segkmap, vp, off + on,
 757  757                                      pn, !pagecreate, S_WRITE);
 758  758  
 759  759                                  error = writerp(rp, base + pon, n, uiop,
 760  760                                      pagecreate);
 761  761  
 762  762                          } else {
 763  763                                  base = segmap_getmapflt(segkmap, vp, off + on,
 764  764                                      n, 0, S_READ);
 765  765                                  error = writerp(rp, base + on, n, uiop, 0);
 766  766                          }
 767  767                  }
 768  768  
 769  769                  if (!error) {
 770  770                          if (mi->mi_flags & MI_NOAC)
 771  771                                  flags = SM_WRITE;
 772  772                          else if (n + on == MAXBSIZE || IS_SWAPVP(vp)) {
 773  773                                  /*
 774  774                                   * Have written a whole block.
 775  775                                   * Start an asynchronous write
 776  776                                   * and mark the buffer to
 777  777                                   * indicate that it won't be
 778  778                                   * needed again soon.
 779  779                                   */
 780  780                                  flags = SM_WRITE | SM_ASYNC | SM_DONTNEED;
 781  781                          } else
 782  782                                  flags = 0;
 783  783                          if ((ioflag & (FSYNC|FDSYNC)) ||
 784  784                              (rp->r_flags & ROUTOFSPACE)) {
 785  785                                  flags &= ~SM_ASYNC;
 786  786                                  flags |= SM_WRITE;
 787  787                          }
 788  788                          if (vpm_enable) {
 789  789                                  error = vpm_sync_pages(vp, off, n, flags);
 790  790                          } else {
 791  791                                  error = segmap_release(segkmap, base, flags);
 792  792                          }
 793  793                  } else {
 794  794                          if (vpm_enable) {
 795  795                                  (void) vpm_sync_pages(vp, off, n, 0);
 796  796                          } else {
 797  797                                  (void) segmap_release(segkmap, base, 0);
 798  798                          }
 799  799                          /*
 800  800                           * In the event that we got an access error while
 801  801                           * faulting in a page for a write-only file just
 802  802                           * force a write.
 803  803                           */
 804  804                          if (error == EACCES)
 805  805                                  goto nfs_fwrite;
 806  806                  }
 807  807          } while (!error && uiop->uio_resid > 0);
 808  808  
 809  809  bottom:
 810  810          if (error) {
 811  811                  uiop->uio_resid = resid + remainder;
 812  812                  uiop->uio_loffset = offset;
 813  813          } else
 814  814                  uiop->uio_resid += remainder;
 815  815  
 816  816          nfs_rw_exit(&rp->r_lkserlock);
 817  817  
 818  818          return (error);
 819  819  }
 820  820  
 821  821  /*
 822  822   * Flags are composed of {B_ASYNC, B_INVAL, B_FREE, B_DONTNEED}
 823  823   */
 824  824  static int
 825  825  nfs_rdwrlbn(vnode_t *vp, page_t *pp, u_offset_t off, size_t len,
 826  826          int flags, cred_t *cr)
 827  827  {
 828  828          struct buf *bp;
 829  829          int error;
 830  830  
 831  831          ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
 832  832          bp = pageio_setup(pp, len, vp, flags);
 833  833          ASSERT(bp != NULL);
 834  834  
 835  835          /*
 836  836           * pageio_setup should have set b_addr to 0.  This
 837  837           * is correct since we want to do I/O on a page
 838  838           * boundary.  bp_mapin will use this addr to calculate
 839  839           * an offset, and then set b_addr to the kernel virtual
 840  840           * address it allocated for us.
 841  841           */
 842  842          ASSERT(bp->b_un.b_addr == 0);
 843  843  
 844  844          bp->b_edev = 0;
 845  845          bp->b_dev = 0;
 846  846          bp->b_lblkno = lbtodb(off);
 847  847          bp->b_file = vp;
 848  848          bp->b_offset = (offset_t)off;
 849  849          bp_mapin(bp);
 850  850  
 851  851          error = nfs_bio(bp, cr);
 852  852  
 853  853          bp_mapout(bp);
 854  854          pageio_done(bp);
 855  855  
 856  856          return (error);
 857  857  }
 858  858  
 859  859  /*
 860  860   * Write to file.  Writes to remote server in largest size
 861  861   * chunks that the server can handle.  Write is synchronous.
 862  862   */
 863  863  static int
 864  864  nfswrite(vnode_t *vp, caddr_t base, uint_t offset, int count, cred_t *cr)
 865  865  {
 866  866          rnode_t *rp;
 867  867          mntinfo_t *mi;
 868  868          struct nfswriteargs wa;
 869  869          struct nfsattrstat ns;
 870  870          int error;
 871  871          int tsize;
 872  872          int douprintf;
 873  873  
 874  874          douprintf = 1;
 875  875  
 876  876          rp = VTOR(vp);
 877  877          mi = VTOMI(vp);
 878  878  
 879  879          ASSERT(nfs_zone() == mi->mi_zone);
 880  880  
 881  881          wa.wa_args = &wa.wa_args_buf;
 882  882          wa.wa_fhandle = *VTOFH(vp);
 883  883  
 884  884          do {
 885  885                  tsize = MIN(mi->mi_curwrite, count);
 886  886                  wa.wa_data = base;
 887  887                  wa.wa_begoff = offset;
 888  888                  wa.wa_totcount = tsize;
 889  889                  wa.wa_count = tsize;
 890  890                  wa.wa_offset = offset;
 891  891  
 892  892                  if (mi->mi_io_kstats) {
 893  893                          mutex_enter(&mi->mi_lock);
 894  894                          kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
 895  895                          mutex_exit(&mi->mi_lock);
 896  896                  }
 897  897                  wa.wa_mblk = NULL;
 898  898                  do {
 899  899                          error = rfs2call(mi, RFS_WRITE,
 900  900                              xdr_writeargs, (caddr_t)&wa,
 901  901                              xdr_attrstat, (caddr_t)&ns, cr,
 902  902                              &douprintf, &ns.ns_status, 0, NULL);
 903  903                  } while (error == ENFS_TRYAGAIN);
 904  904                  if (mi->mi_io_kstats) {
 905  905                          mutex_enter(&mi->mi_lock);
 906  906                          kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
 907  907                          mutex_exit(&mi->mi_lock);
 908  908                  }
 909  909  
 910  910                  if (!error) {
 911  911                          error = geterrno(ns.ns_status);
 912  912                          /*
 913  913                           * Can't check for stale fhandle and purge caches
 914  914                           * here because pages are held by nfs_getpage.
 915  915                           * Just mark the attribute cache as timed out
 916  916                           * and set RWRITEATTR to indicate that the file
 917  917                           * was modified with a WRITE operation.
 918  918                           */
 919  919                          if (!error) {
 920  920                                  count -= tsize;
 921  921                                  base += tsize;
 922  922                                  offset += tsize;
 923  923                                  if (mi->mi_io_kstats) {
 924  924                                          mutex_enter(&mi->mi_lock);
 925  925                                          KSTAT_IO_PTR(mi->mi_io_kstats)->
 926  926                                              writes++;
 927  927                                          KSTAT_IO_PTR(mi->mi_io_kstats)->
 928  928                                              nwritten += tsize;
 929  929                                          mutex_exit(&mi->mi_lock);
 930  930                                  }
 931  931                                  lwp_stat_update(LWP_STAT_OUBLK, 1);
 932  932                                  mutex_enter(&rp->r_statelock);
 933  933                                  PURGE_ATTRCACHE_LOCKED(rp);
 934  934                                  rp->r_flags |= RWRITEATTR;
 935  935                                  mutex_exit(&rp->r_statelock);
 936  936                          }
 937  937                  }
 938  938          } while (!error && count);
 939  939  
 940  940          return (error);
 941  941  }
 942  942  
 943  943  /*
 944  944   * Read from a file.  Reads data in largest chunks our interface can handle.
 945  945   */
 946  946  static int
 947  947  nfsread(vnode_t *vp, caddr_t base, uint_t offset,
 948  948      int count, size_t *residp, cred_t *cr)
 949  949  {
 950  950          mntinfo_t *mi;
 951  951          struct nfsreadargs ra;
 952  952          struct nfsrdresult rr;
 953  953          int tsize;
 954  954          int error;
 955  955          int douprintf;
 956  956          failinfo_t fi;
 957  957          rnode_t *rp;
 958  958          struct vattr va;
 959  959          hrtime_t t;
 960  960  
 961  961          rp = VTOR(vp);
 962  962          mi = VTOMI(vp);
 963  963  
 964  964          ASSERT(nfs_zone() == mi->mi_zone);
 965  965  
 966  966          douprintf = 1;
 967  967  
 968  968          ra.ra_fhandle = *VTOFH(vp);
 969  969  
 970  970          fi.vp = vp;
 971  971          fi.fhp = (caddr_t)&ra.ra_fhandle;
 972  972          fi.copyproc = nfscopyfh;
 973  973          fi.lookupproc = nfslookup;
 974  974          fi.xattrdirproc = acl_getxattrdir2;
 975  975  
 976  976          do {
 977  977                  if (mi->mi_io_kstats) {
 978  978                          mutex_enter(&mi->mi_lock);
 979  979                          kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
 980  980                          mutex_exit(&mi->mi_lock);
 981  981                  }
 982  982  
 983  983                  do {
 984  984                          tsize = MIN(mi->mi_curread, count);
 985  985                          rr.rr_data = base;
 986  986                          ra.ra_offset = offset;
 987  987                          ra.ra_totcount = tsize;
 988  988                          ra.ra_count = tsize;
 989  989                          ra.ra_data = base;
 990  990                          t = gethrtime();
 991  991                          error = rfs2call(mi, RFS_READ,
 992  992                              xdr_readargs, (caddr_t)&ra,
 993  993                              xdr_rdresult, (caddr_t)&rr, cr,
 994  994                              &douprintf, &rr.rr_status, 0, &fi);
 995  995                  } while (error == ENFS_TRYAGAIN);
 996  996  
 997  997                  if (mi->mi_io_kstats) {
 998  998                          mutex_enter(&mi->mi_lock);
 999  999                          kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
1000 1000                          mutex_exit(&mi->mi_lock);
1001 1001                  }
1002 1002  
1003 1003                  if (!error) {
1004 1004                          error = geterrno(rr.rr_status);
1005 1005                          if (!error) {
1006 1006                                  count -= rr.rr_count;
1007 1007                                  base += rr.rr_count;
1008 1008                                  offset += rr.rr_count;
1009 1009                                  if (mi->mi_io_kstats) {
1010 1010                                          mutex_enter(&mi->mi_lock);
1011 1011                                          KSTAT_IO_PTR(mi->mi_io_kstats)->reads++;
1012 1012                                          KSTAT_IO_PTR(mi->mi_io_kstats)->nread +=
1013 1013                                              rr.rr_count;
1014 1014                                          mutex_exit(&mi->mi_lock);
1015 1015                                  }
1016 1016                                  lwp_stat_update(LWP_STAT_INBLK, 1);
1017 1017                          }
1018 1018                  }
1019 1019          } while (!error && count && rr.rr_count == tsize);
1020 1020  
1021 1021          *residp = count;
1022 1022  
1023 1023          if (!error) {
1024 1024                  /*
1025 1025                   * Since no error occurred, we have the current
1026 1026                   * attributes and we need to do a cache check and then
1027 1027                   * potentially update the cached attributes.  We can't
1028 1028                   * use the normal attribute check and cache mechanisms
1029 1029                   * because they might cause a cache flush which would
1030 1030                   * deadlock.  Instead, we just check the cache to see
1031 1031                   * if the attributes have changed.  If it is, then we
1032 1032                   * just mark the attributes as out of date.  The next
1033 1033                   * time that the attributes are checked, they will be
1034 1034                   * out of date, new attributes will be fetched, and
1035 1035                   * the page cache will be flushed.  If the attributes
1036 1036                   * weren't changed, then we just update the cached
1037 1037                   * attributes with these attributes.
1038 1038                   */
1039 1039                  /*
1040 1040                   * If NFS_ACL is supported on the server, then the
1041 1041                   * attributes returned by server may have minimal
1042 1042                   * permissions sometimes denying access to users having
1043 1043                   * proper access.  To get the proper attributes, mark
1044 1044                   * the attributes as expired so that they will be
1045 1045                   * regotten via the NFS_ACL GETATTR2 procedure.
1046 1046                   */
1047 1047                  error = nattr_to_vattr(vp, &rr.rr_attr, &va);
1048 1048                  mutex_enter(&rp->r_statelock);
1049 1049                  if (error || !CACHE_VALID(rp, va.va_mtime, va.va_size) ||
1050 1050                      (mi->mi_flags & MI_ACL)) {
1051 1051                          mutex_exit(&rp->r_statelock);
1052 1052                          PURGE_ATTRCACHE(vp);
1053 1053                  } else {
1054 1054                          if (rp->r_mtime <= t) {
1055 1055                                  nfs_attrcache_va(vp, &va);
1056 1056                          }
1057 1057                          mutex_exit(&rp->r_statelock);
1058 1058                  }
1059 1059          }
1060 1060  
1061 1061          return (error);
1062 1062  }
1063 1063  
1064 1064  /* ARGSUSED */
1065 1065  static int
1066 1066  nfs_ioctl(vnode_t *vp, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp,
1067 1067          caller_context_t *ct)
1068 1068  {
1069 1069  
1070 1070          if (nfs_zone() != VTOMI(vp)->mi_zone)
1071 1071                  return (EIO);
1072 1072          switch (cmd) {
1073 1073                  case _FIODIRECTIO:
1074 1074                          return (nfs_directio(vp, (int)arg, cr));
1075 1075                  default:
1076 1076                          return (ENOTTY);
1077 1077          }
1078 1078  }
1079 1079  
1080 1080  /* ARGSUSED */
1081 1081  static int
1082 1082  nfs_getattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
1083 1083          caller_context_t *ct)
1084 1084  {
1085 1085          int error;
1086 1086          rnode_t *rp;
1087 1087  
1088 1088          if (nfs_zone() != VTOMI(vp)->mi_zone)
1089 1089                  return (EIO);
1090 1090          /*
1091 1091           * If it has been specified that the return value will
1092 1092           * just be used as a hint, and we are only being asked
1093 1093           * for size, fsid or rdevid, then return the client's
1094 1094           * notion of these values without checking to make sure
1095 1095           * that the attribute cache is up to date.
1096 1096           * The whole point is to avoid an over the wire GETATTR
1097 1097           * call.
1098 1098           */
1099 1099          rp = VTOR(vp);
1100 1100          if (flags & ATTR_HINT) {
1101 1101                  if (vap->va_mask ==
1102 1102                      (vap->va_mask & (AT_SIZE | AT_FSID | AT_RDEV))) {
1103 1103                          mutex_enter(&rp->r_statelock);
1104 1104                          if (vap->va_mask | AT_SIZE)
1105 1105                                  vap->va_size = rp->r_size;
1106 1106                          if (vap->va_mask | AT_FSID)
1107 1107                                  vap->va_fsid = rp->r_attr.va_fsid;
1108 1108                          if (vap->va_mask | AT_RDEV)
1109 1109                                  vap->va_rdev = rp->r_attr.va_rdev;
1110 1110                          mutex_exit(&rp->r_statelock);
1111 1111                          return (0);
1112 1112                  }
1113 1113          }
1114 1114  
1115 1115          /*
1116 1116           * Only need to flush pages if asking for the mtime
1117 1117           * and if there any dirty pages or any outstanding
1118 1118           * asynchronous (write) requests for this file.
1119 1119           */
1120 1120          if (vap->va_mask & AT_MTIME) {
1121 1121                  if (vn_has_cached_data(vp) &&
1122 1122                      ((rp->r_flags & RDIRTY) || rp->r_awcount > 0)) {
1123 1123                          mutex_enter(&rp->r_statelock);
1124 1124                          rp->r_gcount++;
1125 1125                          mutex_exit(&rp->r_statelock);
1126 1126                          error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
1127 1127                          mutex_enter(&rp->r_statelock);
1128 1128                          if (error && (error == ENOSPC || error == EDQUOT)) {
1129 1129                                  if (!rp->r_error)
1130 1130                                          rp->r_error = error;
1131 1131                          }
1132 1132                          if (--rp->r_gcount == 0)
1133 1133                                  cv_broadcast(&rp->r_cv);
1134 1134                          mutex_exit(&rp->r_statelock);
1135 1135                  }
1136 1136          }
1137 1137  
1138 1138          return (nfsgetattr(vp, vap, cr));
1139 1139  }
1140 1140  
1141 1141  /*ARGSUSED4*/
1142 1142  static int
1143 1143  nfs_setattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr,
1144 1144                  caller_context_t *ct)
1145 1145  {
1146 1146          int error;
1147 1147          uint_t mask;
1148 1148          struct vattr va;
1149 1149  
1150 1150          mask = vap->va_mask;
1151 1151  
1152 1152          if (mask & AT_NOSET)
1153 1153                  return (EINVAL);
1154 1154  
1155 1155          if ((mask & AT_SIZE) &&
1156 1156              vap->va_type == VREG &&
1157 1157              vap->va_size > MAXOFF32_T)
1158 1158                  return (EFBIG);
1159 1159  
1160 1160          if (nfs_zone() != VTOMI(vp)->mi_zone)
1161 1161                  return (EIO);
1162 1162  
1163 1163          va.va_mask = AT_UID | AT_MODE;
1164 1164  
1165 1165          error = nfsgetattr(vp, &va, cr);
1166 1166          if (error)
  
    | 
      ↓ open down ↓ | 
    1127 lines elided | 
    
      ↑ open up ↑ | 
  
1167 1167                  return (error);
1168 1168  
1169 1169          error = secpolicy_vnode_setattr(cr, vp, vap, &va, flags, nfs_accessx,
1170 1170              vp);
1171 1171  
1172 1172          if (error)
1173 1173                  return (error);
1174 1174  
1175 1175          error = nfssetattr(vp, vap, flags, cr);
1176 1176  
1177      -        if (error == 0 && (mask & AT_SIZE) && vap->va_size == 0)
1178      -                vnevent_truncate(vp, ct);
     1177 +        if (error == 0 && (mask & AT_SIZE)) {
     1178 +                if (vap->va_size == 0) {
     1179 +                        vnevent_truncate(vp, ct);
     1180 +                } else {
     1181 +                        vnevent_resize(vp, ct);
     1182 +                }
     1183 +        }
1179 1184  
1180 1185          return (error);
1181 1186  }
1182 1187  
1183 1188  static int
1184 1189  nfssetattr(vnode_t *vp, struct vattr *vap, int flags, cred_t *cr)
1185 1190  {
1186 1191          int error;
1187 1192          uint_t mask;
1188 1193          struct nfssaargs args;
1189 1194          struct nfsattrstat ns;
1190 1195          int douprintf;
1191 1196          rnode_t *rp;
1192 1197          struct vattr va;
1193 1198          mode_t omode;
1194 1199          mntinfo_t *mi;
1195 1200          vsecattr_t *vsp;
1196 1201          hrtime_t t;
1197 1202  
1198 1203          mask = vap->va_mask;
1199 1204  
1200 1205          ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
1201 1206  
1202 1207          rp = VTOR(vp);
1203 1208  
1204 1209          /*
1205 1210           * Only need to flush pages if there are any pages and
1206 1211           * if the file is marked as dirty in some fashion.  The
1207 1212           * file must be flushed so that we can accurately
1208 1213           * determine the size of the file and the cached data
1209 1214           * after the SETATTR returns.  A file is considered to
1210 1215           * be dirty if it is either marked with RDIRTY, has
1211 1216           * outstanding i/o's active, or is mmap'd.  In this
1212 1217           * last case, we can't tell whether there are dirty
1213 1218           * pages, so we flush just to be sure.
1214 1219           */
1215 1220          if (vn_has_cached_data(vp) &&
1216 1221              ((rp->r_flags & RDIRTY) ||
1217 1222              rp->r_count > 0 ||
1218 1223              rp->r_mapcnt > 0)) {
1219 1224                  ASSERT(vp->v_type != VCHR);
1220 1225                  error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, NULL);
1221 1226                  if (error && (error == ENOSPC || error == EDQUOT)) {
1222 1227                          mutex_enter(&rp->r_statelock);
1223 1228                          if (!rp->r_error)
1224 1229                                  rp->r_error = error;
1225 1230                          mutex_exit(&rp->r_statelock);
1226 1231                  }
1227 1232          }
1228 1233  
1229 1234          /*
1230 1235           * If the system call was utime(2) or utimes(2) and the
1231 1236           * application did not specify the times, then set the
1232 1237           * mtime nanosecond field to 1 billion.  This will get
1233 1238           * translated from 1 billion nanoseconds to 1 million
1234 1239           * microseconds in the over the wire request.  The
1235 1240           * server will use 1 million in the microsecond field
1236 1241           * to tell whether both the mtime and atime should be
1237 1242           * set to the server's current time.
1238 1243           *
1239 1244           * This is an overload of the protocol and should be
1240 1245           * documented in the NFS Version 2 protocol specification.
1241 1246           */
1242 1247          if ((mask & AT_MTIME) && !(flags & ATTR_UTIME)) {
1243 1248                  vap->va_mtime.tv_nsec = 1000000000;
1244 1249                  if (NFS_TIME_T_OK(vap->va_mtime.tv_sec) &&
1245 1250                      NFS_TIME_T_OK(vap->va_atime.tv_sec)) {
1246 1251                          error = vattr_to_sattr(vap, &args.saa_sa);
1247 1252                  } else {
1248 1253                          /*
1249 1254                           * Use server times. vap time values will not be used.
1250 1255                           * To ensure no time overflow, make sure vap has
1251 1256                           * valid values, but retain the original values.
1252 1257                           */
1253 1258                          timestruc_t     mtime = vap->va_mtime;
1254 1259                          timestruc_t     atime = vap->va_atime;
1255 1260                          time_t          now;
1256 1261  
1257 1262                          now = gethrestime_sec();
1258 1263                          if (NFS_TIME_T_OK(now)) {
1259 1264                                  /* Just in case server does not know of this */
1260 1265                                  vap->va_mtime.tv_sec = now;
1261 1266                                  vap->va_atime.tv_sec = now;
1262 1267                          } else {
1263 1268                                  vap->va_mtime.tv_sec = 0;
1264 1269                                  vap->va_atime.tv_sec = 0;
1265 1270                          }
1266 1271                          error = vattr_to_sattr(vap, &args.saa_sa);
1267 1272                          /* set vap times back on */
1268 1273                          vap->va_mtime = mtime;
1269 1274                          vap->va_atime = atime;
1270 1275                  }
1271 1276          } else {
1272 1277                  /* Either do not set times or use the client specified times */
1273 1278                  error = vattr_to_sattr(vap, &args.saa_sa);
1274 1279          }
1275 1280          if (error) {
1276 1281                  /* req time field(s) overflow - return immediately */
1277 1282                  return (error);
1278 1283          }
1279 1284          args.saa_fh = *VTOFH(vp);
1280 1285  
1281 1286          va.va_mask = AT_MODE;
1282 1287          error = nfsgetattr(vp, &va, cr);
1283 1288          if (error)
1284 1289                  return (error);
1285 1290          omode = va.va_mode;
1286 1291  
1287 1292          mi = VTOMI(vp);
1288 1293  
1289 1294          douprintf = 1;
1290 1295  
1291 1296          t = gethrtime();
1292 1297  
1293 1298          error = rfs2call(mi, RFS_SETATTR,
1294 1299              xdr_saargs, (caddr_t)&args,
1295 1300              xdr_attrstat, (caddr_t)&ns, cr,
1296 1301              &douprintf, &ns.ns_status, 0, NULL);
1297 1302  
1298 1303          /*
1299 1304           * Purge the access cache and ACL cache if changing either the
1300 1305           * owner of the file, the group owner, or the mode.  These may
1301 1306           * change the access permissions of the file, so purge old
1302 1307           * information and start over again.
1303 1308           */
1304 1309          if ((mask & (AT_UID | AT_GID | AT_MODE)) && (mi->mi_flags & MI_ACL)) {
1305 1310                  (void) nfs_access_purge_rp(rp);
1306 1311                  if (rp->r_secattr != NULL) {
1307 1312                          mutex_enter(&rp->r_statelock);
1308 1313                          vsp = rp->r_secattr;
1309 1314                          rp->r_secattr = NULL;
1310 1315                          mutex_exit(&rp->r_statelock);
1311 1316                          if (vsp != NULL)
1312 1317                                  nfs_acl_free(vsp);
1313 1318                  }
1314 1319          }
1315 1320  
1316 1321          if (!error) {
1317 1322                  error = geterrno(ns.ns_status);
1318 1323                  if (!error) {
1319 1324                          /*
1320 1325                           * If changing the size of the file, invalidate
1321 1326                           * any local cached data which is no longer part
1322 1327                           * of the file.  We also possibly invalidate the
1323 1328                           * last page in the file.  We could use
1324 1329                           * pvn_vpzero(), but this would mark the page as
1325 1330                           * modified and require it to be written back to
1326 1331                           * the server for no particularly good reason.
1327 1332                           * This way, if we access it, then we bring it
1328 1333                           * back in.  A read should be cheaper than a
1329 1334                           * write.
1330 1335                           */
1331 1336                          if (mask & AT_SIZE) {
1332 1337                                  nfs_invalidate_pages(vp,
1333 1338                                      (vap->va_size & PAGEMASK), cr);
1334 1339                          }
1335 1340                          (void) nfs_cache_fattr(vp, &ns.ns_attr, &va, t, cr);
1336 1341                          /*
1337 1342                           * If NFS_ACL is supported on the server, then the
1338 1343                           * attributes returned by server may have minimal
1339 1344                           * permissions sometimes denying access to users having
1340 1345                           * proper access.  To get the proper attributes, mark
1341 1346                           * the attributes as expired so that they will be
1342 1347                           * regotten via the NFS_ACL GETATTR2 procedure.
1343 1348                           */
1344 1349                          if (mi->mi_flags & MI_ACL) {
1345 1350                                  PURGE_ATTRCACHE(vp);
1346 1351                          }
1347 1352                          /*
1348 1353                           * This next check attempts to deal with NFS
1349 1354                           * servers which can not handle increasing
1350 1355                           * the size of the file via setattr.  Most
1351 1356                           * of these servers do not return an error,
1352 1357                           * but do not change the size of the file.
1353 1358                           * Hence, this check and then attempt to set
1354 1359                           * the file size by writing 1 byte at the
1355 1360                           * offset of the end of the file that we need.
1356 1361                           */
1357 1362                          if ((mask & AT_SIZE) &&
1358 1363                              ns.ns_attr.na_size < (uint32_t)vap->va_size) {
1359 1364                                  char zb = '\0';
1360 1365  
1361 1366                                  error = nfswrite(vp, &zb,
1362 1367                                      vap->va_size - sizeof (zb),
1363 1368                                      sizeof (zb), cr);
1364 1369                          }
1365 1370                          /*
1366 1371                           * Some servers will change the mode to clear the setuid
1367 1372                           * and setgid bits when changing the uid or gid.  The
1368 1373                           * client needs to compensate appropriately.
1369 1374                           */
1370 1375                          if (mask & (AT_UID | AT_GID)) {
1371 1376                                  int terror;
1372 1377  
1373 1378                                  va.va_mask = AT_MODE;
1374 1379                                  terror = nfsgetattr(vp, &va, cr);
1375 1380                                  if (!terror &&
1376 1381                                      (((mask & AT_MODE) &&
1377 1382                                      va.va_mode != vap->va_mode) ||
1378 1383                                      (!(mask & AT_MODE) &&
1379 1384                                      va.va_mode != omode))) {
1380 1385                                          va.va_mask = AT_MODE;
1381 1386                                          if (mask & AT_MODE)
1382 1387                                                  va.va_mode = vap->va_mode;
1383 1388                                          else
1384 1389                                                  va.va_mode = omode;
1385 1390                                          (void) nfssetattr(vp, &va, 0, cr);
1386 1391                                  }
1387 1392                          }
1388 1393                  } else {
1389 1394                          PURGE_ATTRCACHE(vp);
1390 1395                          PURGE_STALE_FH(error, vp, cr);
1391 1396                  }
1392 1397          } else {
1393 1398                  PURGE_ATTRCACHE(vp);
1394 1399          }
1395 1400  
1396 1401          return (error);
1397 1402  }
1398 1403  
1399 1404  static int
1400 1405  nfs_accessx(void *vp, int mode, cred_t *cr)
1401 1406  {
1402 1407          ASSERT(nfs_zone() == VTOMI((vnode_t *)vp)->mi_zone);
1403 1408          return (nfs_access(vp, mode, 0, cr, NULL));
1404 1409  }
1405 1410  
1406 1411  /* ARGSUSED */
1407 1412  static int
1408 1413  nfs_access(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct)
1409 1414  {
1410 1415          struct vattr va;
1411 1416          int error;
1412 1417          mntinfo_t *mi;
1413 1418          int shift = 0;
1414 1419  
1415 1420          mi = VTOMI(vp);
1416 1421  
1417 1422          if (nfs_zone() != mi->mi_zone)
1418 1423                  return (EIO);
1419 1424          if (mi->mi_flags & MI_ACL) {
1420 1425                  error = acl_access2(vp, mode, flags, cr);
1421 1426                  if (mi->mi_flags & MI_ACL)
1422 1427                          return (error);
1423 1428          }
1424 1429  
1425 1430          va.va_mask = AT_MODE | AT_UID | AT_GID;
1426 1431          error = nfsgetattr(vp, &va, cr);
1427 1432          if (error)
1428 1433                  return (error);
1429 1434  
1430 1435          /*
1431 1436           * Disallow write attempts on read-only
1432 1437           * file systems, unless the file is a
1433 1438           * device node.
1434 1439           */
1435 1440          if ((mode & VWRITE) && vn_is_readonly(vp) && !IS_DEVVP(vp))
1436 1441                  return (EROFS);
1437 1442  
1438 1443          /*
1439 1444           * Disallow attempts to access mandatory lock files.
1440 1445           */
1441 1446          if ((mode & (VWRITE | VREAD | VEXEC)) &&
1442 1447              MANDLOCK(vp, va.va_mode))
1443 1448                  return (EACCES);
1444 1449  
1445 1450          /*
1446 1451           * Access check is based on only
1447 1452           * one of owner, group, public.
1448 1453           * If not owner, then check group.
1449 1454           * If not a member of the group,
1450 1455           * then check public access.
1451 1456           */
1452 1457          if (crgetuid(cr) != va.va_uid) {
1453 1458                  shift += 3;
1454 1459                  if (!groupmember(va.va_gid, cr))
1455 1460                          shift += 3;
1456 1461          }
1457 1462  
1458 1463          return (secpolicy_vnode_access2(cr, vp, va.va_uid,
1459 1464              va.va_mode << shift, mode));
1460 1465  }
1461 1466  
1462 1467  static int nfs_do_symlink_cache = 1;
1463 1468  
1464 1469  /* ARGSUSED */
1465 1470  static int
1466 1471  nfs_readlink(vnode_t *vp, struct uio *uiop, cred_t *cr, caller_context_t *ct)
1467 1472  {
1468 1473          int error;
1469 1474          struct nfsrdlnres rl;
1470 1475          rnode_t *rp;
1471 1476          int douprintf;
1472 1477          failinfo_t fi;
1473 1478  
1474 1479          /*
1475 1480           * We want to be consistent with UFS semantics so we will return
1476 1481           * EINVAL instead of ENXIO. This violates the XNFS spec and
1477 1482           * the RFC 1094, which are wrong any way. BUGID 1138002.
1478 1483           */
1479 1484          if (vp->v_type != VLNK)
1480 1485                  return (EINVAL);
1481 1486  
1482 1487          if (nfs_zone() != VTOMI(vp)->mi_zone)
1483 1488                  return (EIO);
1484 1489  
1485 1490          rp = VTOR(vp);
1486 1491          if (nfs_do_symlink_cache && rp->r_symlink.contents != NULL) {
1487 1492                  error = nfs_validate_caches(vp, cr);
1488 1493                  if (error)
1489 1494                          return (error);
1490 1495                  mutex_enter(&rp->r_statelock);
1491 1496                  if (rp->r_symlink.contents != NULL) {
1492 1497                          error = uiomove(rp->r_symlink.contents,
1493 1498                              rp->r_symlink.len, UIO_READ, uiop);
1494 1499                          mutex_exit(&rp->r_statelock);
1495 1500                          return (error);
1496 1501                  }
1497 1502                  mutex_exit(&rp->r_statelock);
1498 1503          }
1499 1504  
1500 1505  
1501 1506          rl.rl_data = kmem_alloc(NFS_MAXPATHLEN, KM_SLEEP);
1502 1507  
1503 1508          fi.vp = vp;
1504 1509          fi.fhp = NULL;          /* no need to update, filehandle not copied */
1505 1510          fi.copyproc = nfscopyfh;
1506 1511          fi.lookupproc = nfslookup;
1507 1512          fi.xattrdirproc = acl_getxattrdir2;
1508 1513  
1509 1514          douprintf = 1;
1510 1515  
1511 1516          error = rfs2call(VTOMI(vp), RFS_READLINK,
1512 1517              xdr_readlink, (caddr_t)VTOFH(vp),
1513 1518              xdr_rdlnres, (caddr_t)&rl, cr,
1514 1519              &douprintf, &rl.rl_status, 0, &fi);
1515 1520  
1516 1521          if (error) {
1517 1522  
1518 1523                  kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN);
1519 1524                  return (error);
1520 1525          }
1521 1526  
1522 1527          error = geterrno(rl.rl_status);
1523 1528          if (!error) {
1524 1529                  error = uiomove(rl.rl_data, (int)rl.rl_count, UIO_READ, uiop);
1525 1530                  if (nfs_do_symlink_cache && rp->r_symlink.contents == NULL) {
1526 1531                          mutex_enter(&rp->r_statelock);
1527 1532                          if (rp->r_symlink.contents == NULL) {
1528 1533                                  rp->r_symlink.contents = rl.rl_data;
1529 1534                                  rp->r_symlink.len = (int)rl.rl_count;
1530 1535                                  rp->r_symlink.size = NFS_MAXPATHLEN;
1531 1536                                  mutex_exit(&rp->r_statelock);
1532 1537                          } else {
1533 1538                                  mutex_exit(&rp->r_statelock);
1534 1539  
1535 1540                                  kmem_free((void *)rl.rl_data,
1536 1541                                      NFS_MAXPATHLEN);
1537 1542                          }
1538 1543                  } else {
1539 1544  
1540 1545                          kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN);
1541 1546                  }
1542 1547          } else {
1543 1548                  PURGE_STALE_FH(error, vp, cr);
1544 1549  
1545 1550                  kmem_free((void *)rl.rl_data, NFS_MAXPATHLEN);
1546 1551          }
1547 1552  
1548 1553          /*
1549 1554           * Conform to UFS semantics (see comment above)
1550 1555           */
1551 1556          return (error == ENXIO ? EINVAL : error);
1552 1557  }
1553 1558  
1554 1559  /*
1555 1560   * Flush local dirty pages to stable storage on the server.
1556 1561   *
1557 1562   * If FNODSYNC is specified, then there is nothing to do because
1558 1563   * metadata changes are not cached on the client before being
1559 1564   * sent to the server.
1560 1565   */
1561 1566  /* ARGSUSED */
1562 1567  static int
1563 1568  nfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
1564 1569  {
1565 1570          int error;
1566 1571  
1567 1572          if ((syncflag & FNODSYNC) || IS_SWAPVP(vp))
1568 1573                  return (0);
1569 1574  
1570 1575          if (nfs_zone() != VTOMI(vp)->mi_zone)
1571 1576                  return (EIO);
1572 1577  
1573 1578          error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
1574 1579          if (!error)
1575 1580                  error = VTOR(vp)->r_error;
1576 1581          return (error);
1577 1582  }
1578 1583  
1579 1584  
1580 1585  /*
1581 1586   * Weirdness: if the file was removed or the target of a rename
1582 1587   * operation while it was open, it got renamed instead.  Here we
1583 1588   * remove the renamed file.
1584 1589   */
1585 1590  /* ARGSUSED */
1586 1591  static void
1587 1592  nfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
1588 1593  {
1589 1594          rnode_t *rp;
1590 1595  
1591 1596          ASSERT(vp != DNLC_NO_VNODE);
1592 1597  
1593 1598          /*
1594 1599           * If this is coming from the wrong zone, we let someone in the right
1595 1600           * zone take care of it asynchronously.  We can get here due to
1596 1601           * VN_RELE() being called from pageout() or fsflush().  This call may
1597 1602           * potentially turn into an expensive no-op if, for instance, v_count
1598 1603           * gets incremented in the meantime, but it's still correct.
1599 1604           */
1600 1605          if (nfs_zone() != VTOMI(vp)->mi_zone) {
1601 1606                  nfs_async_inactive(vp, cr, nfs_inactive);
1602 1607                  return;
1603 1608          }
1604 1609  
1605 1610          rp = VTOR(vp);
1606 1611  redo:
1607 1612          if (rp->r_unldvp != NULL) {
1608 1613                  /*
1609 1614                   * Save the vnode pointer for the directory where the
1610 1615                   * unlinked-open file got renamed, then set it to NULL
1611 1616                   * to prevent another thread from getting here before
1612 1617                   * we're done with the remove.  While we have the
1613 1618                   * statelock, make local copies of the pertinent rnode
1614 1619                   * fields.  If we weren't to do this in an atomic way, the
1615 1620                   * the unl* fields could become inconsistent with respect
1616 1621                   * to each other due to a race condition between this
1617 1622                   * code and nfs_remove().  See bug report 1034328.
1618 1623                   */
1619 1624                  mutex_enter(&rp->r_statelock);
1620 1625                  if (rp->r_unldvp != NULL) {
1621 1626                          vnode_t *unldvp;
1622 1627                          char *unlname;
1623 1628                          cred_t *unlcred;
1624 1629                          struct nfsdiropargs da;
1625 1630                          enum nfsstat status;
1626 1631                          int douprintf;
1627 1632                          int error;
1628 1633  
1629 1634                          unldvp = rp->r_unldvp;
1630 1635                          rp->r_unldvp = NULL;
1631 1636                          unlname = rp->r_unlname;
1632 1637                          rp->r_unlname = NULL;
1633 1638                          unlcred = rp->r_unlcred;
1634 1639                          rp->r_unlcred = NULL;
1635 1640                          mutex_exit(&rp->r_statelock);
1636 1641  
1637 1642                          /*
1638 1643                           * If there are any dirty pages left, then flush
1639 1644                           * them.  This is unfortunate because they just
1640 1645                           * may get thrown away during the remove operation,
1641 1646                           * but we have to do this for correctness.
1642 1647                           */
1643 1648                          if (vn_has_cached_data(vp) &&
1644 1649                              ((rp->r_flags & RDIRTY) || rp->r_count > 0)) {
1645 1650                                  ASSERT(vp->v_type != VCHR);
1646 1651                                  error = nfs_putpage(vp, (offset_t)0, 0, 0,
1647 1652                                      cr, ct);
1648 1653                                  if (error) {
1649 1654                                          mutex_enter(&rp->r_statelock);
1650 1655                                          if (!rp->r_error)
1651 1656                                                  rp->r_error = error;
1652 1657                                          mutex_exit(&rp->r_statelock);
1653 1658                                  }
1654 1659                          }
1655 1660  
1656 1661                          /*
1657 1662                           * Do the remove operation on the renamed file
1658 1663                           */
1659 1664                          setdiropargs(&da, unlname, unldvp);
1660 1665  
1661 1666                          douprintf = 1;
1662 1667  
1663 1668                          (void) rfs2call(VTOMI(unldvp), RFS_REMOVE,
1664 1669                              xdr_diropargs, (caddr_t)&da,
1665 1670                              xdr_enum, (caddr_t)&status, unlcred,
1666 1671                              &douprintf, &status, 0, NULL);
1667 1672  
1668 1673                          if (HAVE_RDDIR_CACHE(VTOR(unldvp)))
1669 1674                                  nfs_purge_rddir_cache(unldvp);
1670 1675                          PURGE_ATTRCACHE(unldvp);
1671 1676  
1672 1677                          /*
1673 1678                           * Release stuff held for the remove
1674 1679                           */
1675 1680                          VN_RELE(unldvp);
1676 1681                          kmem_free(unlname, MAXNAMELEN);
1677 1682                          crfree(unlcred);
1678 1683                          goto redo;
1679 1684                  }
1680 1685                  mutex_exit(&rp->r_statelock);
1681 1686          }
1682 1687  
1683 1688          rp_addfree(rp, cr);
1684 1689  }
1685 1690  
1686 1691  /*
1687 1692   * Remote file system operations having to do with directory manipulation.
1688 1693   */
1689 1694  
1690 1695  /* ARGSUSED */
1691 1696  static int
1692 1697  nfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1693 1698          int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
1694 1699          int *direntflags, pathname_t *realpnp)
1695 1700  {
1696 1701          int error;
1697 1702          vnode_t *vp;
1698 1703          vnode_t *avp = NULL;
1699 1704          rnode_t *drp;
1700 1705  
1701 1706          if (nfs_zone() != VTOMI(dvp)->mi_zone)
1702 1707                  return (EPERM);
1703 1708  
1704 1709          drp = VTOR(dvp);
1705 1710  
1706 1711          /*
1707 1712           * Are we looking up extended attributes?  If so, "dvp" is
1708 1713           * the file or directory for which we want attributes, and
1709 1714           * we need a lookup of the hidden attribute directory
1710 1715           * before we lookup the rest of the path.
1711 1716           */
1712 1717          if (flags & LOOKUP_XATTR) {
1713 1718                  bool_t cflag = ((flags & CREATE_XATTR_DIR) != 0);
1714 1719                  mntinfo_t *mi;
1715 1720  
1716 1721                  mi = VTOMI(dvp);
1717 1722                  if (!(mi->mi_flags & MI_EXTATTR))
1718 1723                          return (EINVAL);
1719 1724  
1720 1725                  if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR(dvp)))
1721 1726                          return (EINTR);
1722 1727  
1723 1728                  (void) nfslookup_dnlc(dvp, XATTR_DIR_NAME, &avp, cr);
1724 1729                  if (avp == NULL)
1725 1730                          error = acl_getxattrdir2(dvp, &avp, cflag, cr, 0);
1726 1731                  else
1727 1732                          error = 0;
1728 1733  
1729 1734                  nfs_rw_exit(&drp->r_rwlock);
1730 1735  
1731 1736                  if (error) {
1732 1737                          if (mi->mi_flags & MI_EXTATTR)
1733 1738                                  return (error);
1734 1739                          return (EINVAL);
1735 1740                  }
1736 1741                  dvp = avp;
1737 1742                  drp = VTOR(dvp);
1738 1743          }
1739 1744  
1740 1745          if (nfs_rw_enter_sig(&drp->r_rwlock, RW_READER, INTR(dvp))) {
1741 1746                  error = EINTR;
1742 1747                  goto out;
1743 1748          }
1744 1749  
1745 1750          error = nfslookup(dvp, nm, vpp, pnp, flags, rdir, cr, 0);
1746 1751  
1747 1752          nfs_rw_exit(&drp->r_rwlock);
1748 1753  
1749 1754          /*
1750 1755           * If vnode is a device, create special vnode.
1751 1756           */
1752 1757          if (!error && IS_DEVVP(*vpp)) {
1753 1758                  vp = *vpp;
1754 1759                  *vpp = specvp(vp, vp->v_rdev, vp->v_type, cr);
1755 1760                  VN_RELE(vp);
1756 1761          }
1757 1762  
1758 1763  out:
1759 1764          if (avp != NULL)
1760 1765                  VN_RELE(avp);
1761 1766  
1762 1767          return (error);
1763 1768  }
1764 1769  
1765 1770  static int nfs_lookup_neg_cache = 1;
1766 1771  
1767 1772  #ifdef DEBUG
1768 1773  static int nfs_lookup_dnlc_hits = 0;
1769 1774  static int nfs_lookup_dnlc_misses = 0;
1770 1775  static int nfs_lookup_dnlc_neg_hits = 0;
1771 1776  static int nfs_lookup_dnlc_disappears = 0;
1772 1777  static int nfs_lookup_dnlc_lookups = 0;
1773 1778  #endif
1774 1779  
1775 1780  /* ARGSUSED */
1776 1781  int
1777 1782  nfslookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
1778 1783          int flags, vnode_t *rdir, cred_t *cr, int rfscall_flags)
1779 1784  {
1780 1785          int error;
1781 1786  
1782 1787          ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone);
1783 1788  
1784 1789          /*
1785 1790           * If lookup is for "", just return dvp.  Don't need
1786 1791           * to send it over the wire, look it up in the dnlc,
1787 1792           * or perform any access checks.
1788 1793           */
1789 1794          if (*nm == '\0') {
1790 1795                  VN_HOLD(dvp);
1791 1796                  *vpp = dvp;
1792 1797                  return (0);
1793 1798          }
1794 1799  
1795 1800          /*
1796 1801           * Can't do lookups in non-directories.
1797 1802           */
1798 1803          if (dvp->v_type != VDIR)
1799 1804                  return (ENOTDIR);
1800 1805  
1801 1806          /*
1802 1807           * If we're called with RFSCALL_SOFT, it's important that
1803 1808           * the only rfscall is one we make directly; if we permit
1804 1809           * an access call because we're looking up "." or validating
1805 1810           * a dnlc hit, we'll deadlock because that rfscall will not
1806 1811           * have the RFSCALL_SOFT set.
1807 1812           */
1808 1813          if (rfscall_flags & RFSCALL_SOFT)
1809 1814                  goto callit;
1810 1815  
1811 1816          /*
1812 1817           * If lookup is for ".", just return dvp.  Don't need
1813 1818           * to send it over the wire or look it up in the dnlc,
1814 1819           * just need to check access.
1815 1820           */
1816 1821          if (strcmp(nm, ".") == 0) {
1817 1822                  error = nfs_access(dvp, VEXEC, 0, cr, NULL);
1818 1823                  if (error)
1819 1824                          return (error);
1820 1825                  VN_HOLD(dvp);
1821 1826                  *vpp = dvp;
1822 1827                  return (0);
1823 1828          }
1824 1829  
1825 1830          /*
1826 1831           * Lookup this name in the DNLC.  If there was a valid entry,
1827 1832           * then return the results of the lookup.
1828 1833           */
1829 1834          error = nfslookup_dnlc(dvp, nm, vpp, cr);
1830 1835          if (error || *vpp != NULL)
1831 1836                  return (error);
1832 1837  
1833 1838  callit:
1834 1839          error = nfslookup_otw(dvp, nm, vpp, cr, rfscall_flags);
1835 1840  
1836 1841          return (error);
1837 1842  }
1838 1843  
1839 1844  static int
1840 1845  nfslookup_dnlc(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr)
1841 1846  {
1842 1847          int error;
1843 1848          vnode_t *vp;
1844 1849  
1845 1850          ASSERT(*nm != '\0');
1846 1851          ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone);
1847 1852  
1848 1853          /*
1849 1854           * Lookup this name in the DNLC.  If successful, then validate
1850 1855           * the caches and then recheck the DNLC.  The DNLC is rechecked
1851 1856           * just in case this entry got invalidated during the call
1852 1857           * to nfs_validate_caches.
1853 1858           *
1854 1859           * An assumption is being made that it is safe to say that a
1855 1860           * file exists which may not on the server.  Any operations to
1856 1861           * the server will fail with ESTALE.
1857 1862           */
1858 1863  #ifdef DEBUG
1859 1864          nfs_lookup_dnlc_lookups++;
1860 1865  #endif
1861 1866          vp = dnlc_lookup(dvp, nm);
1862 1867          if (vp != NULL) {
1863 1868                  VN_RELE(vp);
1864 1869                  if (vp == DNLC_NO_VNODE && !vn_is_readonly(dvp)) {
1865 1870                          PURGE_ATTRCACHE(dvp);
1866 1871                  }
1867 1872                  error = nfs_validate_caches(dvp, cr);
1868 1873                  if (error)
1869 1874                          return (error);
1870 1875                  vp = dnlc_lookup(dvp, nm);
1871 1876                  if (vp != NULL) {
1872 1877                          error = nfs_access(dvp, VEXEC, 0, cr, NULL);
1873 1878                          if (error) {
1874 1879                                  VN_RELE(vp);
1875 1880                                  return (error);
1876 1881                          }
1877 1882                          if (vp == DNLC_NO_VNODE) {
1878 1883                                  VN_RELE(vp);
1879 1884  #ifdef DEBUG
1880 1885                                  nfs_lookup_dnlc_neg_hits++;
1881 1886  #endif
1882 1887                                  return (ENOENT);
1883 1888                          }
1884 1889                          *vpp = vp;
1885 1890  #ifdef DEBUG
1886 1891                          nfs_lookup_dnlc_hits++;
1887 1892  #endif
1888 1893                          return (0);
1889 1894                  }
1890 1895  #ifdef DEBUG
1891 1896                  nfs_lookup_dnlc_disappears++;
1892 1897  #endif
1893 1898          }
1894 1899  #ifdef DEBUG
1895 1900          else
1896 1901                  nfs_lookup_dnlc_misses++;
1897 1902  #endif
1898 1903  
1899 1904          *vpp = NULL;
1900 1905  
1901 1906          return (0);
1902 1907  }
1903 1908  
1904 1909  static int
1905 1910  nfslookup_otw(vnode_t *dvp, char *nm, vnode_t **vpp, cred_t *cr,
1906 1911          int rfscall_flags)
1907 1912  {
1908 1913          int error;
1909 1914          struct nfsdiropargs da;
1910 1915          struct nfsdiropres dr;
1911 1916          int douprintf;
1912 1917          failinfo_t fi;
1913 1918          hrtime_t t;
1914 1919  
1915 1920          ASSERT(*nm != '\0');
1916 1921          ASSERT(dvp->v_type == VDIR);
1917 1922          ASSERT(nfs_zone() == VTOMI(dvp)->mi_zone);
1918 1923  
1919 1924          setdiropargs(&da, nm, dvp);
1920 1925  
1921 1926          fi.vp = dvp;
1922 1927          fi.fhp = NULL;          /* no need to update, filehandle not copied */
1923 1928          fi.copyproc = nfscopyfh;
1924 1929          fi.lookupproc = nfslookup;
1925 1930          fi.xattrdirproc = acl_getxattrdir2;
1926 1931  
1927 1932          douprintf = 1;
1928 1933  
1929 1934          t = gethrtime();
1930 1935  
1931 1936          error = rfs2call(VTOMI(dvp), RFS_LOOKUP,
1932 1937              xdr_diropargs, (caddr_t)&da,
1933 1938              xdr_diropres, (caddr_t)&dr, cr,
1934 1939              &douprintf, &dr.dr_status, rfscall_flags, &fi);
1935 1940  
1936 1941          if (!error) {
1937 1942                  error = geterrno(dr.dr_status);
1938 1943                  if (!error) {
1939 1944                          *vpp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr,
1940 1945                              dvp->v_vfsp, t, cr, VTOR(dvp)->r_path, nm);
1941 1946                          /*
1942 1947                           * If NFS_ACL is supported on the server, then the
1943 1948                           * attributes returned by server may have minimal
1944 1949                           * permissions sometimes denying access to users having
1945 1950                           * proper access.  To get the proper attributes, mark
1946 1951                           * the attributes as expired so that they will be
1947 1952                           * regotten via the NFS_ACL GETATTR2 procedure.
1948 1953                           */
1949 1954                          if (VTOMI(*vpp)->mi_flags & MI_ACL) {
1950 1955                                  PURGE_ATTRCACHE(*vpp);
1951 1956                          }
1952 1957                          if (!(rfscall_flags & RFSCALL_SOFT))
1953 1958                                  dnlc_update(dvp, nm, *vpp);
1954 1959                  } else {
1955 1960                          PURGE_STALE_FH(error, dvp, cr);
1956 1961                          if (error == ENOENT && nfs_lookup_neg_cache)
1957 1962                                  dnlc_enter(dvp, nm, DNLC_NO_VNODE);
1958 1963                  }
1959 1964          }
1960 1965  
1961 1966          return (error);
1962 1967  }
1963 1968  
1964 1969  /* ARGSUSED */
1965 1970  static int
1966 1971  nfs_create(vnode_t *dvp, char *nm, struct vattr *va, enum vcexcl exclusive,
1967 1972          int mode, vnode_t **vpp, cred_t *cr, int lfaware, caller_context_t *ct,
1968 1973          vsecattr_t *vsecp)
1969 1974  {
1970 1975          int error;
1971 1976          struct nfscreatargs args;
1972 1977          struct nfsdiropres dr;
1973 1978          int douprintf;
1974 1979          vnode_t *vp;
1975 1980          rnode_t *rp;
1976 1981          struct vattr vattr;
1977 1982          rnode_t *drp;
1978 1983          vnode_t *tempvp;
1979 1984          hrtime_t t;
1980 1985  
1981 1986          drp = VTOR(dvp);
1982 1987  
1983 1988          if (nfs_zone() != VTOMI(dvp)->mi_zone)
1984 1989                  return (EPERM);
1985 1990          if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
1986 1991                  return (EINTR);
1987 1992  
1988 1993          /*
1989 1994           * We make a copy of the attributes because the caller does not
1990 1995           * expect us to change what va points to.
1991 1996           */
1992 1997          vattr = *va;
1993 1998  
1994 1999          /*
1995 2000           * If the pathname is "", just use dvp.  Don't need
1996 2001           * to send it over the wire, look it up in the dnlc,
1997 2002           * or perform any access checks.
1998 2003           */
1999 2004          if (*nm == '\0') {
2000 2005                  error = 0;
2001 2006                  VN_HOLD(dvp);
2002 2007                  vp = dvp;
2003 2008          /*
2004 2009           * If the pathname is ".", just use dvp.  Don't need
2005 2010           * to send it over the wire or look it up in the dnlc,
2006 2011           * just need to check access.
2007 2012           */
2008 2013          } else if (strcmp(nm, ".") == 0) {
2009 2014                  error = nfs_access(dvp, VEXEC, 0, cr, ct);
2010 2015                  if (error) {
2011 2016                          nfs_rw_exit(&drp->r_rwlock);
2012 2017                          return (error);
2013 2018                  }
2014 2019                  VN_HOLD(dvp);
2015 2020                  vp = dvp;
2016 2021          /*
2017 2022           * We need to go over the wire, just to be sure whether the
2018 2023           * file exists or not.  Using the DNLC can be dangerous in
2019 2024           * this case when making a decision regarding existence.
2020 2025           */
2021 2026          } else {
2022 2027                  error = nfslookup_otw(dvp, nm, &vp, cr, 0);
2023 2028          }
2024 2029          if (!error) {
2025 2030                  if (exclusive == EXCL)
2026 2031                          error = EEXIST;
2027 2032                  else if (vp->v_type == VDIR && (mode & VWRITE))
2028 2033                          error = EISDIR;
2029 2034                  else {
2030 2035                          /*
2031 2036                           * If vnode is a device, create special vnode.
2032 2037                           */
2033 2038                          if (IS_DEVVP(vp)) {
2034 2039                                  tempvp = vp;
2035 2040                                  vp = specvp(vp, vp->v_rdev, vp->v_type, cr);
2036 2041                                  VN_RELE(tempvp);
2037 2042                          }
2038 2043                          if (!(error = VOP_ACCESS(vp, mode, 0, cr, ct))) {
2039 2044                                  if ((vattr.va_mask & AT_SIZE) &&
2040 2045                                      vp->v_type == VREG) {
2041 2046                                          vattr.va_mask = AT_SIZE;
2042 2047                                          error = nfssetattr(vp, &vattr, 0, cr);
2043 2048  
2044 2049                                          if (!error) {
2045 2050                                                  /*
2046 2051                                                   * Existing file was truncated;
2047 2052                                                   * emit a create event.
2048 2053                                                   */
2049 2054                                                  vnevent_create(vp, ct);
2050 2055                                          }
2051 2056                                  }
2052 2057                          }
2053 2058                  }
2054 2059                  nfs_rw_exit(&drp->r_rwlock);
2055 2060                  if (error) {
2056 2061                          VN_RELE(vp);
2057 2062                  } else {
2058 2063                          *vpp = vp;
2059 2064                  }
2060 2065                  return (error);
2061 2066          }
2062 2067  
2063 2068          ASSERT(vattr.va_mask & AT_TYPE);
2064 2069          if (vattr.va_type == VREG) {
2065 2070                  ASSERT(vattr.va_mask & AT_MODE);
2066 2071                  if (MANDMODE(vattr.va_mode)) {
2067 2072                          nfs_rw_exit(&drp->r_rwlock);
2068 2073                          return (EACCES);
2069 2074                  }
2070 2075          }
2071 2076  
2072 2077          dnlc_remove(dvp, nm);
2073 2078  
2074 2079          setdiropargs(&args.ca_da, nm, dvp);
2075 2080  
2076 2081          /*
2077 2082           * Decide what the group-id of the created file should be.
2078 2083           * Set it in attribute list as advisory...then do a setattr
2079 2084           * if the server didn't get it right the first time.
2080 2085           */
2081 2086          error = setdirgid(dvp, &vattr.va_gid, cr);
2082 2087          if (error) {
2083 2088                  nfs_rw_exit(&drp->r_rwlock);
2084 2089                  return (error);
2085 2090          }
2086 2091          vattr.va_mask |= AT_GID;
2087 2092  
2088 2093          /*
2089 2094           * This is a completely gross hack to make mknod
2090 2095           * work over the wire until we can wack the protocol
2091 2096           */
2092 2097  #define IFCHR           0020000         /* character special */
2093 2098  #define IFBLK           0060000         /* block special */
2094 2099  #define IFSOCK          0140000         /* socket */
2095 2100  
2096 2101          /*
2097 2102           * dev_t is uint_t in 5.x and short in 4.x. Both 4.x
2098 2103           * supports 8 bit majors. 5.x supports 14 bit majors. 5.x supports 18
2099 2104           * bits in the minor number where 4.x supports 8 bits.  If the 5.x
2100 2105           * minor/major numbers <= 8 bits long, compress the device
2101 2106           * number before sending it. Otherwise, the 4.x server will not
2102 2107           * create the device with the correct device number and nothing can be
2103 2108           * done about this.
2104 2109           */
2105 2110          if (vattr.va_type == VCHR || vattr.va_type == VBLK) {
2106 2111                  dev_t d = vattr.va_rdev;
2107 2112                  dev32_t dev32;
2108 2113  
2109 2114                  if (vattr.va_type == VCHR)
2110 2115                          vattr.va_mode |= IFCHR;
2111 2116                  else
2112 2117                          vattr.va_mode |= IFBLK;
2113 2118  
2114 2119                  (void) cmpldev(&dev32, d);
2115 2120                  if (dev32 & ~((SO4_MAXMAJ << L_BITSMINOR32) | SO4_MAXMIN))
2116 2121                          vattr.va_size = (u_offset_t)dev32;
2117 2122                  else
2118 2123                          vattr.va_size = (u_offset_t)nfsv2_cmpdev(d);
2119 2124  
2120 2125                  vattr.va_mask |= AT_MODE|AT_SIZE;
2121 2126          } else if (vattr.va_type == VFIFO) {
2122 2127                  vattr.va_mode |= IFCHR;         /* xtra kludge for namedpipe */
2123 2128                  vattr.va_size = (u_offset_t)NFS_FIFO_DEV;       /* blech */
2124 2129                  vattr.va_mask |= AT_MODE|AT_SIZE;
2125 2130          } else if (vattr.va_type == VSOCK) {
2126 2131                  vattr.va_mode |= IFSOCK;
2127 2132                  /*
2128 2133                   * To avoid triggering bugs in the servers set AT_SIZE
2129 2134                   * (all other RFS_CREATE calls set this).
2130 2135                   */
2131 2136                  vattr.va_size = 0;
2132 2137                  vattr.va_mask |= AT_MODE|AT_SIZE;
2133 2138          }
2134 2139  
2135 2140          args.ca_sa = &args.ca_sa_buf;
2136 2141          error = vattr_to_sattr(&vattr, args.ca_sa);
2137 2142          if (error) {
2138 2143                  /* req time field(s) overflow - return immediately */
2139 2144                  nfs_rw_exit(&drp->r_rwlock);
2140 2145                  return (error);
2141 2146          }
2142 2147  
2143 2148          douprintf = 1;
2144 2149  
2145 2150          t = gethrtime();
2146 2151  
2147 2152          error = rfs2call(VTOMI(dvp), RFS_CREATE,
2148 2153              xdr_creatargs, (caddr_t)&args,
2149 2154              xdr_diropres, (caddr_t)&dr, cr,
2150 2155              &douprintf, &dr.dr_status, 0, NULL);
2151 2156  
2152 2157          PURGE_ATTRCACHE(dvp);   /* mod time changed */
2153 2158  
2154 2159          if (!error) {
2155 2160                  error = geterrno(dr.dr_status);
2156 2161                  if (!error) {
2157 2162                          if (HAVE_RDDIR_CACHE(drp))
2158 2163                                  nfs_purge_rddir_cache(dvp);
2159 2164                          vp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr,
2160 2165                              dvp->v_vfsp, t, cr, NULL, NULL);
2161 2166                          /*
2162 2167                           * If NFS_ACL is supported on the server, then the
2163 2168                           * attributes returned by server may have minimal
2164 2169                           * permissions sometimes denying access to users having
2165 2170                           * proper access.  To get the proper attributes, mark
2166 2171                           * the attributes as expired so that they will be
2167 2172                           * regotten via the NFS_ACL GETATTR2 procedure.
2168 2173                           */
2169 2174                          if (VTOMI(vp)->mi_flags & MI_ACL) {
2170 2175                                  PURGE_ATTRCACHE(vp);
2171 2176                          }
2172 2177                          dnlc_update(dvp, nm, vp);
2173 2178                          rp = VTOR(vp);
2174 2179                          if (vattr.va_size == 0) {
2175 2180                                  mutex_enter(&rp->r_statelock);
2176 2181                                  rp->r_size = 0;
2177 2182                                  mutex_exit(&rp->r_statelock);
2178 2183                                  if (vn_has_cached_data(vp)) {
2179 2184                                          ASSERT(vp->v_type != VCHR);
2180 2185                                          nfs_invalidate_pages(vp,
2181 2186                                              (u_offset_t)0, cr);
2182 2187                                  }
2183 2188                          }
2184 2189  
2185 2190                          /*
2186 2191                           * Make sure the gid was set correctly.
2187 2192                           * If not, try to set it (but don't lose
2188 2193                           * any sleep over it).
2189 2194                           */
2190 2195                          if (vattr.va_gid != rp->r_attr.va_gid) {
2191 2196                                  vattr.va_mask = AT_GID;
2192 2197                                  (void) nfssetattr(vp, &vattr, 0, cr);
2193 2198                          }
2194 2199  
2195 2200                          /*
2196 2201                           * If vnode is a device create special vnode
2197 2202                           */
2198 2203                          if (IS_DEVVP(vp)) {
2199 2204                                  *vpp = specvp(vp, vp->v_rdev, vp->v_type, cr);
2200 2205                                  VN_RELE(vp);
2201 2206                          } else
2202 2207                                  *vpp = vp;
2203 2208                  } else {
2204 2209                          PURGE_STALE_FH(error, dvp, cr);
2205 2210                  }
2206 2211          }
2207 2212  
2208 2213          nfs_rw_exit(&drp->r_rwlock);
2209 2214  
2210 2215          return (error);
2211 2216  }
2212 2217  
2213 2218  /*
2214 2219   * Weirdness: if the vnode to be removed is open
2215 2220   * we rename it instead of removing it and nfs_inactive
2216 2221   * will remove the new name.
2217 2222   */
2218 2223  /* ARGSUSED */
2219 2224  static int
2220 2225  nfs_remove(vnode_t *dvp, char *nm, cred_t *cr, caller_context_t *ct, int flags)
2221 2226  {
2222 2227          int error;
2223 2228          struct nfsdiropargs da;
2224 2229          enum nfsstat status;
2225 2230          vnode_t *vp;
2226 2231          char *tmpname;
2227 2232          int douprintf;
2228 2233          rnode_t *rp;
2229 2234          rnode_t *drp;
2230 2235  
2231 2236          if (nfs_zone() != VTOMI(dvp)->mi_zone)
2232 2237                  return (EPERM);
2233 2238          drp = VTOR(dvp);
2234 2239          if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2235 2240                  return (EINTR);
2236 2241  
2237 2242          error = nfslookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0);
2238 2243          if (error) {
2239 2244                  nfs_rw_exit(&drp->r_rwlock);
2240 2245                  return (error);
2241 2246          }
2242 2247  
2243 2248          if (vp->v_type == VDIR && secpolicy_fs_linkdir(cr, dvp->v_vfsp)) {
2244 2249                  VN_RELE(vp);
2245 2250                  nfs_rw_exit(&drp->r_rwlock);
2246 2251                  return (EPERM);
2247 2252          }
2248 2253  
2249 2254          /*
2250 2255           * First just remove the entry from the name cache, as it
2251 2256           * is most likely the only entry for this vp.
2252 2257           */
2253 2258          dnlc_remove(dvp, nm);
2254 2259  
2255 2260          /*
2256 2261           * If the file has a v_count > 1 then there may be more than one
2257 2262           * entry in the name cache due multiple links or an open file,
2258 2263           * but we don't have the real reference count so flush all
2259 2264           * possible entries.
2260 2265           */
2261 2266          if (vp->v_count > 1)
2262 2267                  dnlc_purge_vp(vp);
2263 2268  
2264 2269          /*
2265 2270           * Now we have the real reference count on the vnode
2266 2271           */
2267 2272          rp = VTOR(vp);
2268 2273          mutex_enter(&rp->r_statelock);
2269 2274          if (vp->v_count > 1 &&
2270 2275              (rp->r_unldvp == NULL || strcmp(nm, rp->r_unlname) == 0)) {
2271 2276                  mutex_exit(&rp->r_statelock);
2272 2277                  tmpname = newname();
2273 2278                  error = nfsrename(dvp, nm, dvp, tmpname, cr, ct);
2274 2279                  if (error)
2275 2280                          kmem_free(tmpname, MAXNAMELEN);
2276 2281                  else {
2277 2282                          mutex_enter(&rp->r_statelock);
2278 2283                          if (rp->r_unldvp == NULL) {
2279 2284                                  VN_HOLD(dvp);
2280 2285                                  rp->r_unldvp = dvp;
2281 2286                                  if (rp->r_unlcred != NULL)
2282 2287                                          crfree(rp->r_unlcred);
2283 2288                                  crhold(cr);
2284 2289                                  rp->r_unlcred = cr;
2285 2290                                  rp->r_unlname = tmpname;
2286 2291                          } else {
2287 2292                                  kmem_free(rp->r_unlname, MAXNAMELEN);
2288 2293                                  rp->r_unlname = tmpname;
2289 2294                          }
2290 2295                          mutex_exit(&rp->r_statelock);
2291 2296                  }
2292 2297          } else {
2293 2298                  mutex_exit(&rp->r_statelock);
2294 2299                  /*
2295 2300                   * We need to flush any dirty pages which happen to
2296 2301                   * be hanging around before removing the file.  This
2297 2302                   * shouldn't happen very often and mostly on file
2298 2303                   * systems mounted "nocto".
2299 2304                   */
2300 2305                  if (vn_has_cached_data(vp) &&
2301 2306                      ((rp->r_flags & RDIRTY) || rp->r_count > 0)) {
2302 2307                          error = nfs_putpage(vp, (offset_t)0, 0, 0, cr, ct);
2303 2308                          if (error && (error == ENOSPC || error == EDQUOT)) {
2304 2309                                  mutex_enter(&rp->r_statelock);
2305 2310                                  if (!rp->r_error)
2306 2311                                          rp->r_error = error;
2307 2312                                  mutex_exit(&rp->r_statelock);
2308 2313                          }
2309 2314                  }
2310 2315  
2311 2316                  setdiropargs(&da, nm, dvp);
2312 2317  
2313 2318                  douprintf = 1;
2314 2319  
2315 2320                  error = rfs2call(VTOMI(dvp), RFS_REMOVE,
2316 2321                      xdr_diropargs, (caddr_t)&da,
2317 2322                      xdr_enum, (caddr_t)&status, cr,
2318 2323                      &douprintf, &status, 0, NULL);
2319 2324  
2320 2325                  /*
2321 2326                   * The xattr dir may be gone after last attr is removed,
2322 2327                   * so flush it from dnlc.
2323 2328                   */
2324 2329                  if (dvp->v_flag & V_XATTRDIR)
2325 2330                          dnlc_purge_vp(dvp);
2326 2331  
2327 2332                  PURGE_ATTRCACHE(dvp);   /* mod time changed */
2328 2333                  PURGE_ATTRCACHE(vp);    /* link count changed */
2329 2334  
2330 2335                  if (!error) {
2331 2336                          error = geterrno(status);
2332 2337                          if (!error) {
2333 2338                                  if (HAVE_RDDIR_CACHE(drp))
2334 2339                                          nfs_purge_rddir_cache(dvp);
2335 2340                          } else {
2336 2341                                  PURGE_STALE_FH(error, dvp, cr);
2337 2342                          }
2338 2343                  }
2339 2344          }
2340 2345  
2341 2346          if (error == 0) {
2342 2347                  vnevent_remove(vp, dvp, nm, ct);
2343 2348          }
2344 2349          VN_RELE(vp);
2345 2350  
2346 2351          nfs_rw_exit(&drp->r_rwlock);
2347 2352  
2348 2353          return (error);
2349 2354  }
2350 2355  
2351 2356  /* ARGSUSED */
2352 2357  static int
2353 2358  nfs_link(vnode_t *tdvp, vnode_t *svp, char *tnm, cred_t *cr,
2354 2359          caller_context_t *ct, int flags)
2355 2360  {
2356 2361          int error;
2357 2362          struct nfslinkargs args;
2358 2363          enum nfsstat status;
2359 2364          vnode_t *realvp;
2360 2365          int douprintf;
2361 2366          rnode_t *tdrp;
2362 2367  
2363 2368          if (nfs_zone() != VTOMI(tdvp)->mi_zone)
2364 2369                  return (EPERM);
2365 2370          if (VOP_REALVP(svp, &realvp, ct) == 0)
2366 2371                  svp = realvp;
2367 2372  
2368 2373          args.la_from = VTOFH(svp);
2369 2374          setdiropargs(&args.la_to, tnm, tdvp);
2370 2375  
2371 2376          tdrp = VTOR(tdvp);
2372 2377          if (nfs_rw_enter_sig(&tdrp->r_rwlock, RW_WRITER, INTR(tdvp)))
2373 2378                  return (EINTR);
2374 2379  
2375 2380          dnlc_remove(tdvp, tnm);
2376 2381  
2377 2382          douprintf = 1;
2378 2383  
2379 2384          error = rfs2call(VTOMI(svp), RFS_LINK,
2380 2385              xdr_linkargs, (caddr_t)&args,
2381 2386              xdr_enum, (caddr_t)&status, cr,
2382 2387              &douprintf, &status, 0, NULL);
2383 2388  
2384 2389          PURGE_ATTRCACHE(tdvp);  /* mod time changed */
2385 2390          PURGE_ATTRCACHE(svp);   /* link count changed */
2386 2391  
2387 2392          if (!error) {
2388 2393                  error = geterrno(status);
2389 2394                  if (!error) {
2390 2395                          if (HAVE_RDDIR_CACHE(tdrp))
2391 2396                                  nfs_purge_rddir_cache(tdvp);
2392 2397                  }
2393 2398          }
2394 2399  
2395 2400          nfs_rw_exit(&tdrp->r_rwlock);
2396 2401  
2397 2402          if (!error) {
2398 2403                  /*
2399 2404                   * Notify the source file of this link operation.
2400 2405                   */
2401 2406                  vnevent_link(svp, ct);
2402 2407          }
2403 2408          return (error);
2404 2409  }
2405 2410  
2406 2411  /* ARGSUSED */
2407 2412  static int
2408 2413  nfs_rename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr,
2409 2414          caller_context_t *ct, int flags)
2410 2415  {
2411 2416          vnode_t *realvp;
2412 2417  
2413 2418          if (nfs_zone() != VTOMI(odvp)->mi_zone)
2414 2419                  return (EPERM);
2415 2420          if (VOP_REALVP(ndvp, &realvp, ct) == 0)
2416 2421                  ndvp = realvp;
2417 2422  
2418 2423          return (nfsrename(odvp, onm, ndvp, nnm, cr, ct));
2419 2424  }
2420 2425  
2421 2426  /*
2422 2427   * nfsrename does the real work of renaming in NFS Version 2.
2423 2428   */
2424 2429  static int
2425 2430  nfsrename(vnode_t *odvp, char *onm, vnode_t *ndvp, char *nnm, cred_t *cr,
2426 2431      caller_context_t *ct)
2427 2432  {
2428 2433          int error;
2429 2434          enum nfsstat status;
2430 2435          struct nfsrnmargs args;
2431 2436          int douprintf;
2432 2437          vnode_t *nvp = NULL;
2433 2438          vnode_t *ovp = NULL;
2434 2439          char *tmpname;
2435 2440          rnode_t *rp;
2436 2441          rnode_t *odrp;
2437 2442          rnode_t *ndrp;
2438 2443  
2439 2444          ASSERT(nfs_zone() == VTOMI(odvp)->mi_zone);
2440 2445          if (strcmp(onm, ".") == 0 || strcmp(onm, "..") == 0 ||
2441 2446              strcmp(nnm, ".") == 0 || strcmp(nnm, "..") == 0)
2442 2447                  return (EINVAL);
2443 2448  
2444 2449          odrp = VTOR(odvp);
2445 2450          ndrp = VTOR(ndvp);
2446 2451          if ((intptr_t)odrp < (intptr_t)ndrp) {
2447 2452                  if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR(odvp)))
2448 2453                          return (EINTR);
2449 2454                  if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR(ndvp))) {
2450 2455                          nfs_rw_exit(&odrp->r_rwlock);
2451 2456                          return (EINTR);
2452 2457                  }
2453 2458          } else {
2454 2459                  if (nfs_rw_enter_sig(&ndrp->r_rwlock, RW_WRITER, INTR(ndvp)))
2455 2460                          return (EINTR);
2456 2461                  if (nfs_rw_enter_sig(&odrp->r_rwlock, RW_WRITER, INTR(odvp))) {
2457 2462                          nfs_rw_exit(&ndrp->r_rwlock);
2458 2463                          return (EINTR);
2459 2464                  }
2460 2465          }
2461 2466  
2462 2467          /*
2463 2468           * Lookup the target file.  If it exists, it needs to be
2464 2469           * checked to see whether it is a mount point and whether
2465 2470           * it is active (open).
2466 2471           */
2467 2472          error = nfslookup(ndvp, nnm, &nvp, NULL, 0, NULL, cr, 0);
2468 2473          if (!error) {
2469 2474                  /*
2470 2475                   * If this file has been mounted on, then just
2471 2476                   * return busy because renaming to it would remove
2472 2477                   * the mounted file system from the name space.
2473 2478                   */
2474 2479                  if (vn_mountedvfs(nvp) != NULL) {
2475 2480                          VN_RELE(nvp);
2476 2481                          nfs_rw_exit(&odrp->r_rwlock);
2477 2482                          nfs_rw_exit(&ndrp->r_rwlock);
2478 2483                          return (EBUSY);
2479 2484                  }
2480 2485  
2481 2486                  /*
2482 2487                   * Purge the name cache of all references to this vnode
2483 2488                   * so that we can check the reference count to infer
2484 2489                   * whether it is active or not.
2485 2490                   */
2486 2491                  /*
2487 2492                   * First just remove the entry from the name cache, as it
2488 2493                   * is most likely the only entry for this vp.
2489 2494                   */
2490 2495                  dnlc_remove(ndvp, nnm);
2491 2496                  /*
2492 2497                   * If the file has a v_count > 1 then there may be more
2493 2498                   * than one entry in the name cache due multiple links
2494 2499                   * or an open file, but we don't have the real reference
2495 2500                   * count so flush all possible entries.
2496 2501                   */
2497 2502                  if (nvp->v_count > 1)
2498 2503                          dnlc_purge_vp(nvp);
2499 2504  
2500 2505                  /*
2501 2506                   * If the vnode is active and is not a directory,
2502 2507                   * arrange to rename it to a
2503 2508                   * temporary file so that it will continue to be
2504 2509                   * accessible.  This implements the "unlink-open-file"
2505 2510                   * semantics for the target of a rename operation.
2506 2511                   * Before doing this though, make sure that the
2507 2512                   * source and target files are not already the same.
2508 2513                   */
2509 2514                  if (nvp->v_count > 1 && nvp->v_type != VDIR) {
2510 2515                          /*
2511 2516                           * Lookup the source name.
2512 2517                           */
2513 2518                          error = nfslookup(odvp, onm, &ovp, NULL, 0, NULL,
2514 2519                              cr, 0);
2515 2520  
2516 2521                          /*
2517 2522                           * The source name *should* already exist.
2518 2523                           */
2519 2524                          if (error) {
2520 2525                                  VN_RELE(nvp);
2521 2526                                  nfs_rw_exit(&odrp->r_rwlock);
2522 2527                                  nfs_rw_exit(&ndrp->r_rwlock);
2523 2528                                  return (error);
2524 2529                          }
2525 2530  
2526 2531                          /*
2527 2532                           * Compare the two vnodes.  If they are the same,
2528 2533                           * just release all held vnodes and return success.
2529 2534                           */
2530 2535                          if (ovp == nvp) {
2531 2536                                  VN_RELE(ovp);
2532 2537                                  VN_RELE(nvp);
2533 2538                                  nfs_rw_exit(&odrp->r_rwlock);
2534 2539                                  nfs_rw_exit(&ndrp->r_rwlock);
2535 2540                                  return (0);
2536 2541                          }
2537 2542  
2538 2543                          /*
2539 2544                           * Can't mix and match directories and non-
2540 2545                           * directories in rename operations.  We already
2541 2546                           * know that the target is not a directory.  If
2542 2547                           * the source is a directory, return an error.
2543 2548                           */
2544 2549                          if (ovp->v_type == VDIR) {
2545 2550                                  VN_RELE(ovp);
2546 2551                                  VN_RELE(nvp);
2547 2552                                  nfs_rw_exit(&odrp->r_rwlock);
2548 2553                                  nfs_rw_exit(&ndrp->r_rwlock);
2549 2554                                  return (ENOTDIR);
2550 2555                          }
2551 2556  
2552 2557                          /*
2553 2558                           * The target file exists, is not the same as
2554 2559                           * the source file, and is active.  Link it
2555 2560                           * to a temporary filename to avoid having
2556 2561                           * the server removing the file completely.
2557 2562                           */
2558 2563                          tmpname = newname();
2559 2564                          error = nfs_link(ndvp, nvp, tmpname, cr, NULL, 0);
2560 2565                          if (error == EOPNOTSUPP) {
2561 2566                                  error = nfs_rename(ndvp, nnm, ndvp, tmpname,
2562 2567                                      cr, NULL, 0);
2563 2568                          }
2564 2569                          if (error) {
2565 2570                                  kmem_free(tmpname, MAXNAMELEN);
2566 2571                                  VN_RELE(ovp);
2567 2572                                  VN_RELE(nvp);
2568 2573                                  nfs_rw_exit(&odrp->r_rwlock);
2569 2574                                  nfs_rw_exit(&ndrp->r_rwlock);
2570 2575                                  return (error);
2571 2576                          }
2572 2577                          rp = VTOR(nvp);
2573 2578                          mutex_enter(&rp->r_statelock);
2574 2579                          if (rp->r_unldvp == NULL) {
2575 2580                                  VN_HOLD(ndvp);
2576 2581                                  rp->r_unldvp = ndvp;
2577 2582                                  if (rp->r_unlcred != NULL)
2578 2583                                          crfree(rp->r_unlcred);
2579 2584                                  crhold(cr);
2580 2585                                  rp->r_unlcred = cr;
2581 2586                                  rp->r_unlname = tmpname;
2582 2587                          } else {
2583 2588                                  kmem_free(rp->r_unlname, MAXNAMELEN);
2584 2589                                  rp->r_unlname = tmpname;
2585 2590                          }
2586 2591                          mutex_exit(&rp->r_statelock);
2587 2592                  }
2588 2593          }
2589 2594  
2590 2595          if (ovp == NULL) {
2591 2596                  /*
2592 2597                   * When renaming directories to be a subdirectory of a
2593 2598                   * different parent, the dnlc entry for ".." will no
2594 2599                   * longer be valid, so it must be removed.
2595 2600                   *
2596 2601                   * We do a lookup here to determine whether we are renaming
2597 2602                   * a directory and we need to check if we are renaming
2598 2603                   * an unlinked file.  This might have already been done
2599 2604                   * in previous code, so we check ovp == NULL to avoid
2600 2605                   * doing it twice.
2601 2606                   */
2602 2607  
2603 2608                  error = nfslookup(odvp, onm, &ovp, NULL, 0, NULL, cr, 0);
2604 2609  
2605 2610                  /*
2606 2611                   * The source name *should* already exist.
2607 2612                   */
2608 2613                  if (error) {
2609 2614                          nfs_rw_exit(&odrp->r_rwlock);
2610 2615                          nfs_rw_exit(&ndrp->r_rwlock);
2611 2616                          if (nvp) {
2612 2617                                  VN_RELE(nvp);
2613 2618                          }
2614 2619                          return (error);
2615 2620                  }
2616 2621                  ASSERT(ovp != NULL);
2617 2622          }
2618 2623  
2619 2624          dnlc_remove(odvp, onm);
2620 2625          dnlc_remove(ndvp, nnm);
2621 2626  
2622 2627          setdiropargs(&args.rna_from, onm, odvp);
2623 2628          setdiropargs(&args.rna_to, nnm, ndvp);
2624 2629  
2625 2630          douprintf = 1;
2626 2631  
2627 2632          error = rfs2call(VTOMI(odvp), RFS_RENAME,
2628 2633              xdr_rnmargs, (caddr_t)&args,
2629 2634              xdr_enum, (caddr_t)&status, cr,
2630 2635              &douprintf, &status, 0, NULL);
2631 2636  
2632 2637          PURGE_ATTRCACHE(odvp);  /* mod time changed */
2633 2638          PURGE_ATTRCACHE(ndvp);  /* mod time changed */
2634 2639  
2635 2640          if (!error) {
2636 2641                  error = geterrno(status);
2637 2642                  if (!error) {
2638 2643                          if (HAVE_RDDIR_CACHE(odrp))
2639 2644                                  nfs_purge_rddir_cache(odvp);
2640 2645                          if (HAVE_RDDIR_CACHE(ndrp))
2641 2646                                  nfs_purge_rddir_cache(ndvp);
2642 2647                          /*
2643 2648                           * when renaming directories to be a subdirectory of a
2644 2649                           * different parent, the dnlc entry for ".." will no
2645 2650                           * longer be valid, so it must be removed
2646 2651                           */
2647 2652                          rp = VTOR(ovp);
2648 2653                          if (ndvp != odvp) {
2649 2654                                  if (ovp->v_type == VDIR) {
2650 2655                                          dnlc_remove(ovp, "..");
2651 2656                                          if (HAVE_RDDIR_CACHE(rp))
2652 2657                                                  nfs_purge_rddir_cache(ovp);
2653 2658                                  }
2654 2659                          }
2655 2660  
2656 2661                          /*
2657 2662                           * If we are renaming the unlinked file, update the
2658 2663                           * r_unldvp and r_unlname as needed.
2659 2664                           */
2660 2665                          mutex_enter(&rp->r_statelock);
2661 2666                          if (rp->r_unldvp != NULL) {
2662 2667                                  if (strcmp(rp->r_unlname, onm) == 0) {
2663 2668                                          (void) strncpy(rp->r_unlname,
2664 2669                                              nnm, MAXNAMELEN);
2665 2670                                          rp->r_unlname[MAXNAMELEN - 1] = '\0';
2666 2671  
2667 2672                                          if (ndvp != rp->r_unldvp) {
2668 2673                                                  VN_RELE(rp->r_unldvp);
2669 2674                                                  rp->r_unldvp = ndvp;
2670 2675                                                  VN_HOLD(ndvp);
2671 2676                                          }
2672 2677                                  }
2673 2678                          }
2674 2679                          mutex_exit(&rp->r_statelock);
2675 2680                  } else {
2676 2681                          /*
2677 2682                           * System V defines rename to return EEXIST, not
2678 2683                           * ENOTEMPTY if the target directory is not empty.
2679 2684                           * Over the wire, the error is NFSERR_ENOTEMPTY
2680 2685                           * which geterrno maps to ENOTEMPTY.
  
    | 
      ↓ open down ↓ | 
    1492 lines elided | 
    
      ↑ open up ↑ | 
  
2681 2686                           */
2682 2687                          if (error == ENOTEMPTY)
2683 2688                                  error = EEXIST;
2684 2689                  }
2685 2690          }
2686 2691  
2687 2692          if (error == 0) {
2688 2693                  if (nvp)
2689 2694                          vnevent_rename_dest(nvp, ndvp, nnm, ct);
2690 2695  
2691      -                if (odvp != ndvp)
2692      -                        vnevent_rename_dest_dir(ndvp, ct);
2693      -
2694 2696                  ASSERT(ovp != NULL);
2695 2697                  vnevent_rename_src(ovp, odvp, onm, ct);
     2698 +                vnevent_rename_dest_dir(ndvp, ovp, nnm, ct);
2696 2699          }
2697 2700  
2698 2701          if (nvp) {
2699 2702                  VN_RELE(nvp);
2700 2703          }
2701 2704          VN_RELE(ovp);
2702 2705  
2703 2706          nfs_rw_exit(&odrp->r_rwlock);
2704 2707          nfs_rw_exit(&ndrp->r_rwlock);
2705 2708  
2706 2709          return (error);
2707 2710  }
2708 2711  
2709 2712  /* ARGSUSED */
2710 2713  static int
2711 2714  nfs_mkdir(vnode_t *dvp, char *nm, struct vattr *va, vnode_t **vpp, cred_t *cr,
2712 2715          caller_context_t *ct, int flags, vsecattr_t *vsecp)
2713 2716  {
2714 2717          int error;
2715 2718          struct nfscreatargs args;
2716 2719          struct nfsdiropres dr;
2717 2720          int douprintf;
2718 2721          rnode_t *drp;
2719 2722          hrtime_t t;
2720 2723  
2721 2724          if (nfs_zone() != VTOMI(dvp)->mi_zone)
2722 2725                  return (EPERM);
2723 2726  
2724 2727          setdiropargs(&args.ca_da, nm, dvp);
2725 2728  
2726 2729          /*
2727 2730           * Decide what the group-id and set-gid bit of the created directory
2728 2731           * should be.  May have to do a setattr to get the gid right.
2729 2732           */
2730 2733          error = setdirgid(dvp, &va->va_gid, cr);
2731 2734          if (error)
2732 2735                  return (error);
2733 2736          error = setdirmode(dvp, &va->va_mode, cr);
2734 2737          if (error)
2735 2738                  return (error);
2736 2739          va->va_mask |= AT_MODE|AT_GID;
2737 2740  
2738 2741          args.ca_sa = &args.ca_sa_buf;
2739 2742          error = vattr_to_sattr(va, args.ca_sa);
2740 2743          if (error) {
2741 2744                  /* req time field(s) overflow - return immediately */
2742 2745                  return (error);
2743 2746          }
2744 2747  
2745 2748          drp = VTOR(dvp);
2746 2749          if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2747 2750                  return (EINTR);
2748 2751  
2749 2752          dnlc_remove(dvp, nm);
2750 2753  
2751 2754          douprintf = 1;
2752 2755  
2753 2756          t = gethrtime();
2754 2757  
2755 2758          error = rfs2call(VTOMI(dvp), RFS_MKDIR,
2756 2759              xdr_creatargs, (caddr_t)&args,
2757 2760              xdr_diropres, (caddr_t)&dr, cr,
2758 2761              &douprintf, &dr.dr_status, 0, NULL);
2759 2762  
2760 2763          PURGE_ATTRCACHE(dvp);   /* mod time changed */
2761 2764  
2762 2765          if (!error) {
2763 2766                  error = geterrno(dr.dr_status);
2764 2767                  if (!error) {
2765 2768                          if (HAVE_RDDIR_CACHE(drp))
2766 2769                                  nfs_purge_rddir_cache(dvp);
2767 2770                          /*
2768 2771                           * The attributes returned by RFS_MKDIR can not
2769 2772                           * be depended upon, so mark the attribute cache
2770 2773                           * as purged.  A subsequent GETATTR will get the
2771 2774                           * correct attributes from the server.
2772 2775                           */
2773 2776                          *vpp = makenfsnode(&dr.dr_fhandle, &dr.dr_attr,
2774 2777                              dvp->v_vfsp, t, cr, NULL, NULL);
2775 2778                          PURGE_ATTRCACHE(*vpp);
2776 2779                          dnlc_update(dvp, nm, *vpp);
2777 2780  
2778 2781                          /*
2779 2782                           * Make sure the gid was set correctly.
2780 2783                           * If not, try to set it (but don't lose
2781 2784                           * any sleep over it).
2782 2785                           */
2783 2786                          if (va->va_gid != VTOR(*vpp)->r_attr.va_gid) {
2784 2787                                  va->va_mask = AT_GID;
2785 2788                                  (void) nfssetattr(*vpp, va, 0, cr);
2786 2789                          }
2787 2790                  } else {
2788 2791                          PURGE_STALE_FH(error, dvp, cr);
2789 2792                  }
2790 2793          }
2791 2794  
2792 2795          nfs_rw_exit(&drp->r_rwlock);
2793 2796  
2794 2797          return (error);
2795 2798  }
2796 2799  
2797 2800  /* ARGSUSED */
2798 2801  static int
2799 2802  nfs_rmdir(vnode_t *dvp, char *nm, vnode_t *cdir, cred_t *cr,
2800 2803          caller_context_t *ct, int flags)
2801 2804  {
2802 2805          int error;
2803 2806          enum nfsstat status;
2804 2807          struct nfsdiropargs da;
2805 2808          vnode_t *vp;
2806 2809          int douprintf;
2807 2810          rnode_t *drp;
2808 2811  
2809 2812          if (nfs_zone() != VTOMI(dvp)->mi_zone)
2810 2813                  return (EPERM);
2811 2814          drp = VTOR(dvp);
2812 2815          if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2813 2816                  return (EINTR);
2814 2817  
2815 2818          /*
2816 2819           * Attempt to prevent a rmdir(".") from succeeding.
2817 2820           */
2818 2821          error = nfslookup(dvp, nm, &vp, NULL, 0, NULL, cr, 0);
2819 2822          if (error) {
2820 2823                  nfs_rw_exit(&drp->r_rwlock);
2821 2824                  return (error);
2822 2825          }
2823 2826  
2824 2827          if (vp == cdir) {
2825 2828                  VN_RELE(vp);
2826 2829                  nfs_rw_exit(&drp->r_rwlock);
2827 2830                  return (EINVAL);
2828 2831          }
2829 2832  
2830 2833          setdiropargs(&da, nm, dvp);
2831 2834  
2832 2835          /*
2833 2836           * First just remove the entry from the name cache, as it
2834 2837           * is most likely an entry for this vp.
2835 2838           */
2836 2839          dnlc_remove(dvp, nm);
2837 2840  
2838 2841          /*
2839 2842           * If there vnode reference count is greater than one, then
2840 2843           * there may be additional references in the DNLC which will
2841 2844           * need to be purged.  First, trying removing the entry for
2842 2845           * the parent directory and see if that removes the additional
2843 2846           * reference(s).  If that doesn't do it, then use dnlc_purge_vp
2844 2847           * to completely remove any references to the directory which
2845 2848           * might still exist in the DNLC.
2846 2849           */
2847 2850          if (vp->v_count > 1) {
2848 2851                  dnlc_remove(vp, "..");
2849 2852                  if (vp->v_count > 1)
2850 2853                          dnlc_purge_vp(vp);
2851 2854          }
2852 2855  
2853 2856          douprintf = 1;
2854 2857  
2855 2858          error = rfs2call(VTOMI(dvp), RFS_RMDIR,
2856 2859              xdr_diropargs, (caddr_t)&da,
2857 2860              xdr_enum, (caddr_t)&status, cr,
2858 2861              &douprintf, &status, 0, NULL);
2859 2862  
2860 2863          PURGE_ATTRCACHE(dvp);   /* mod time changed */
2861 2864  
2862 2865          if (error) {
2863 2866                  VN_RELE(vp);
2864 2867                  nfs_rw_exit(&drp->r_rwlock);
2865 2868                  return (error);
2866 2869          }
2867 2870  
2868 2871          error = geterrno(status);
2869 2872          if (!error) {
2870 2873                  if (HAVE_RDDIR_CACHE(drp))
2871 2874                          nfs_purge_rddir_cache(dvp);
2872 2875                  if (HAVE_RDDIR_CACHE(VTOR(vp)))
2873 2876                          nfs_purge_rddir_cache(vp);
2874 2877          } else {
2875 2878                  PURGE_STALE_FH(error, dvp, cr);
2876 2879                  /*
2877 2880                   * System V defines rmdir to return EEXIST, not
2878 2881                   * ENOTEMPTY if the directory is not empty.  Over
2879 2882                   * the wire, the error is NFSERR_ENOTEMPTY which
2880 2883                   * geterrno maps to ENOTEMPTY.
2881 2884                   */
2882 2885                  if (error == ENOTEMPTY)
2883 2886                          error = EEXIST;
2884 2887          }
2885 2888  
2886 2889          if (error == 0) {
2887 2890                  vnevent_rmdir(vp, dvp, nm, ct);
2888 2891          }
2889 2892          VN_RELE(vp);
2890 2893  
2891 2894          nfs_rw_exit(&drp->r_rwlock);
2892 2895  
2893 2896          return (error);
2894 2897  }
2895 2898  
2896 2899  /* ARGSUSED */
2897 2900  static int
2898 2901  nfs_symlink(vnode_t *dvp, char *lnm, struct vattr *tva, char *tnm, cred_t *cr,
2899 2902          caller_context_t *ct, int flags)
2900 2903  {
2901 2904          int error;
2902 2905          struct nfsslargs args;
2903 2906          enum nfsstat status;
2904 2907          int douprintf;
2905 2908          rnode_t *drp;
2906 2909  
2907 2910          if (nfs_zone() != VTOMI(dvp)->mi_zone)
2908 2911                  return (EPERM);
2909 2912          setdiropargs(&args.sla_from, lnm, dvp);
2910 2913          args.sla_sa = &args.sla_sa_buf;
2911 2914          error = vattr_to_sattr(tva, args.sla_sa);
2912 2915          if (error) {
2913 2916                  /* req time field(s) overflow - return immediately */
2914 2917                  return (error);
2915 2918          }
2916 2919          args.sla_tnm = tnm;
2917 2920  
2918 2921          drp = VTOR(dvp);
2919 2922          if (nfs_rw_enter_sig(&drp->r_rwlock, RW_WRITER, INTR(dvp)))
2920 2923                  return (EINTR);
2921 2924  
2922 2925          dnlc_remove(dvp, lnm);
2923 2926  
2924 2927          douprintf = 1;
2925 2928  
2926 2929          error = rfs2call(VTOMI(dvp), RFS_SYMLINK,
2927 2930              xdr_slargs, (caddr_t)&args,
2928 2931              xdr_enum, (caddr_t)&status, cr,
2929 2932              &douprintf, &status, 0, NULL);
2930 2933  
2931 2934          PURGE_ATTRCACHE(dvp);   /* mod time changed */
2932 2935  
2933 2936          if (!error) {
2934 2937                  error = geterrno(status);
2935 2938                  if (!error) {
2936 2939                          if (HAVE_RDDIR_CACHE(drp))
2937 2940                                  nfs_purge_rddir_cache(dvp);
2938 2941                  } else {
2939 2942                          PURGE_STALE_FH(error, dvp, cr);
2940 2943                  }
2941 2944          }
2942 2945  
2943 2946          nfs_rw_exit(&drp->r_rwlock);
2944 2947  
2945 2948          return (error);
2946 2949  }
2947 2950  
2948 2951  #ifdef DEBUG
2949 2952  static int nfs_readdir_cache_hits = 0;
2950 2953  static int nfs_readdir_cache_shorts = 0;
2951 2954  static int nfs_readdir_cache_waits = 0;
2952 2955  static int nfs_readdir_cache_misses = 0;
2953 2956  static int nfs_readdir_readahead = 0;
2954 2957  #endif
2955 2958  
2956 2959  static int nfs_shrinkreaddir = 0;
2957 2960  
2958 2961  /*
2959 2962   * Read directory entries.
2960 2963   * There are some weird things to look out for here.  The uio_offset
2961 2964   * field is either 0 or it is the offset returned from a previous
2962 2965   * readdir.  It is an opaque value used by the server to find the
2963 2966   * correct directory block to read. The count field is the number
2964 2967   * of blocks to read on the server.  This is advisory only, the server
2965 2968   * may return only one block's worth of entries.  Entries may be compressed
2966 2969   * on the server.
2967 2970   */
2968 2971  /* ARGSUSED */
2969 2972  static int
2970 2973  nfs_readdir(vnode_t *vp, struct uio *uiop, cred_t *cr, int *eofp,
2971 2974          caller_context_t *ct, int flags)
2972 2975  {
2973 2976          int error;
2974 2977          size_t count;
2975 2978          rnode_t *rp;
2976 2979          rddir_cache *rdc;
2977 2980          rddir_cache *nrdc;
2978 2981          rddir_cache *rrdc;
2979 2982  #ifdef DEBUG
2980 2983          int missed;
2981 2984  #endif
2982 2985          rddir_cache srdc;
2983 2986          avl_index_t where;
2984 2987  
2985 2988          rp = VTOR(vp);
2986 2989  
2987 2990          ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_READER));
2988 2991          if (nfs_zone() != VTOMI(vp)->mi_zone)
2989 2992                  return (EIO);
2990 2993          /*
2991 2994           * Make sure that the directory cache is valid.
2992 2995           */
2993 2996          if (HAVE_RDDIR_CACHE(rp)) {
2994 2997                  if (nfs_disable_rddir_cache) {
2995 2998                          /*
2996 2999                           * Setting nfs_disable_rddir_cache in /etc/system
2997 3000                           * allows interoperability with servers that do not
2998 3001                           * properly update the attributes of directories.
2999 3002                           * Any cached information gets purged before an
3000 3003                           * access is made to it.
3001 3004                           */
3002 3005                          nfs_purge_rddir_cache(vp);
3003 3006                  } else {
3004 3007                          error = nfs_validate_caches(vp, cr);
3005 3008                          if (error)
3006 3009                                  return (error);
3007 3010                  }
3008 3011          }
3009 3012  
3010 3013          /*
3011 3014           * UGLINESS: SunOS 3.2 servers apparently cannot always handle an
3012 3015           * RFS_READDIR request with rda_count set to more than 0x400. So
3013 3016           * we reduce the request size here purely for compatibility.
3014 3017           *
3015 3018           * In general, this is no longer required.  However, if a server
3016 3019           * is discovered which can not handle requests larger than 1024,
3017 3020           * nfs_shrinkreaddir can be set to 1 to enable this backwards
3018 3021           * compatibility.
3019 3022           *
3020 3023           * In any case, the request size is limited to NFS_MAXDATA bytes.
3021 3024           */
3022 3025          count = MIN(uiop->uio_iov->iov_len,
3023 3026              nfs_shrinkreaddir ? 0x400 : NFS_MAXDATA);
3024 3027  
3025 3028          nrdc = NULL;
3026 3029  #ifdef DEBUG
3027 3030          missed = 0;
3028 3031  #endif
3029 3032  top:
3030 3033          /*
3031 3034           * Short circuit last readdir which always returns 0 bytes.
3032 3035           * This can be done after the directory has been read through
3033 3036           * completely at least once.  This will set r_direof which
3034 3037           * can be used to find the value of the last cookie.
3035 3038           */
3036 3039          mutex_enter(&rp->r_statelock);
3037 3040          if (rp->r_direof != NULL &&
3038 3041              uiop->uio_offset == rp->r_direof->nfs_ncookie) {
3039 3042                  mutex_exit(&rp->r_statelock);
3040 3043  #ifdef DEBUG
3041 3044                  nfs_readdir_cache_shorts++;
3042 3045  #endif
3043 3046                  if (eofp)
3044 3047                          *eofp = 1;
3045 3048                  if (nrdc != NULL)
3046 3049                          rddir_cache_rele(nrdc);
3047 3050                  return (0);
3048 3051          }
3049 3052          /*
3050 3053           * Look for a cache entry.  Cache entries are identified
3051 3054           * by the NFS cookie value and the byte count requested.
3052 3055           */
3053 3056          srdc.nfs_cookie = uiop->uio_offset;
3054 3057          srdc.buflen = count;
3055 3058          rdc = avl_find(&rp->r_dir, &srdc, &where);
3056 3059          if (rdc != NULL) {
3057 3060                  rddir_cache_hold(rdc);
3058 3061                  /*
3059 3062                   * If the cache entry is in the process of being
3060 3063                   * filled in, wait until this completes.  The
3061 3064                   * RDDIRWAIT bit is set to indicate that someone
3062 3065                   * is waiting and then the thread currently
3063 3066                   * filling the entry is done, it should do a
3064 3067                   * cv_broadcast to wakeup all of the threads
3065 3068                   * waiting for it to finish.
3066 3069                   */
3067 3070                  if (rdc->flags & RDDIR) {
3068 3071                          nfs_rw_exit(&rp->r_rwlock);
3069 3072                          rdc->flags |= RDDIRWAIT;
3070 3073  #ifdef DEBUG
3071 3074                          nfs_readdir_cache_waits++;
3072 3075  #endif
3073 3076                          if (!cv_wait_sig(&rdc->cv, &rp->r_statelock)) {
3074 3077                                  /*
3075 3078                                   * We got interrupted, probably
3076 3079                                   * the user typed ^C or an alarm
3077 3080                                   * fired.  We free the new entry
3078 3081                                   * if we allocated one.
3079 3082                                   */
3080 3083                                  mutex_exit(&rp->r_statelock);
3081 3084                                  (void) nfs_rw_enter_sig(&rp->r_rwlock,
3082 3085                                      RW_READER, FALSE);
3083 3086                                  rddir_cache_rele(rdc);
3084 3087                                  if (nrdc != NULL)
3085 3088                                          rddir_cache_rele(nrdc);
3086 3089                                  return (EINTR);
3087 3090                          }
3088 3091                          mutex_exit(&rp->r_statelock);
3089 3092                          (void) nfs_rw_enter_sig(&rp->r_rwlock,
3090 3093                              RW_READER, FALSE);
3091 3094                          rddir_cache_rele(rdc);
3092 3095                          goto top;
3093 3096                  }
3094 3097                  /*
3095 3098                   * Check to see if a readdir is required to
3096 3099                   * fill the entry.  If so, mark this entry
3097 3100                   * as being filled, remove our reference,
3098 3101                   * and branch to the code to fill the entry.
3099 3102                   */
3100 3103                  if (rdc->flags & RDDIRREQ) {
3101 3104                          rdc->flags &= ~RDDIRREQ;
3102 3105                          rdc->flags |= RDDIR;
3103 3106                          if (nrdc != NULL)
3104 3107                                  rddir_cache_rele(nrdc);
3105 3108                          nrdc = rdc;
3106 3109                          mutex_exit(&rp->r_statelock);
3107 3110                          goto bottom;
3108 3111                  }
3109 3112  #ifdef DEBUG
3110 3113                  if (!missed)
3111 3114                          nfs_readdir_cache_hits++;
3112 3115  #endif
3113 3116                  /*
3114 3117                   * If an error occurred while attempting
3115 3118                   * to fill the cache entry, just return it.
3116 3119                   */
3117 3120                  if (rdc->error) {
3118 3121                          error = rdc->error;
3119 3122                          mutex_exit(&rp->r_statelock);
3120 3123                          rddir_cache_rele(rdc);
3121 3124                          if (nrdc != NULL)
3122 3125                                  rddir_cache_rele(nrdc);
3123 3126                          return (error);
3124 3127                  }
3125 3128  
3126 3129                  /*
3127 3130                   * The cache entry is complete and good,
3128 3131                   * copyout the dirent structs to the calling
3129 3132                   * thread.
3130 3133                   */
3131 3134                  error = uiomove(rdc->entries, rdc->entlen, UIO_READ, uiop);
3132 3135  
3133 3136                  /*
3134 3137                   * If no error occurred during the copyout,
3135 3138                   * update the offset in the uio struct to
3136 3139                   * contain the value of the next cookie
3137 3140                   * and set the eof value appropriately.
3138 3141                   */
3139 3142                  if (!error) {
3140 3143                          uiop->uio_offset = rdc->nfs_ncookie;
3141 3144                          if (eofp)
3142 3145                                  *eofp = rdc->eof;
3143 3146                  }
3144 3147  
3145 3148                  /*
3146 3149                   * Decide whether to do readahead.  Don't if
3147 3150                   * have already read to the end of directory.
3148 3151                   */
3149 3152                  if (rdc->eof) {
3150 3153                          rp->r_direof = rdc;
3151 3154                          mutex_exit(&rp->r_statelock);
3152 3155                          rddir_cache_rele(rdc);
3153 3156                          if (nrdc != NULL)
3154 3157                                  rddir_cache_rele(nrdc);
3155 3158                          return (error);
3156 3159                  }
3157 3160  
3158 3161                  /*
3159 3162                   * Check to see whether we found an entry
3160 3163                   * for the readahead.  If so, we don't need
3161 3164                   * to do anything further, so free the new
3162 3165                   * entry if one was allocated.  Otherwise,
3163 3166                   * allocate a new entry, add it to the cache,
3164 3167                   * and then initiate an asynchronous readdir
3165 3168                   * operation to fill it.
3166 3169                   */
3167 3170                  srdc.nfs_cookie = rdc->nfs_ncookie;
3168 3171                  srdc.buflen = count;
3169 3172                  rrdc = avl_find(&rp->r_dir, &srdc, &where);
3170 3173                  if (rrdc != NULL) {
3171 3174                          if (nrdc != NULL)
3172 3175                                  rddir_cache_rele(nrdc);
3173 3176                  } else {
3174 3177                          if (nrdc != NULL)
3175 3178                                  rrdc = nrdc;
3176 3179                          else {
3177 3180                                  rrdc = rddir_cache_alloc(KM_NOSLEEP);
3178 3181                          }
3179 3182                          if (rrdc != NULL) {
3180 3183                                  rrdc->nfs_cookie = rdc->nfs_ncookie;
3181 3184                                  rrdc->buflen = count;
3182 3185                                  avl_insert(&rp->r_dir, rrdc, where);
3183 3186                                  rddir_cache_hold(rrdc);
3184 3187                                  mutex_exit(&rp->r_statelock);
3185 3188                                  rddir_cache_rele(rdc);
3186 3189  #ifdef DEBUG
3187 3190                                  nfs_readdir_readahead++;
3188 3191  #endif
3189 3192                                  nfs_async_readdir(vp, rrdc, cr, nfsreaddir);
3190 3193                                  return (error);
3191 3194                          }
3192 3195                  }
3193 3196  
3194 3197                  mutex_exit(&rp->r_statelock);
3195 3198                  rddir_cache_rele(rdc);
3196 3199                  return (error);
3197 3200          }
3198 3201  
3199 3202          /*
3200 3203           * Didn't find an entry in the cache.  Construct a new empty
3201 3204           * entry and link it into the cache.  Other processes attempting
3202 3205           * to access this entry will need to wait until it is filled in.
3203 3206           *
3204 3207           * Since kmem_alloc may block, another pass through the cache
3205 3208           * will need to be taken to make sure that another process
3206 3209           * hasn't already added an entry to the cache for this request.
3207 3210           */
3208 3211          if (nrdc == NULL) {
3209 3212                  mutex_exit(&rp->r_statelock);
3210 3213                  nrdc = rddir_cache_alloc(KM_SLEEP);
3211 3214                  nrdc->nfs_cookie = uiop->uio_offset;
3212 3215                  nrdc->buflen = count;
3213 3216                  goto top;
3214 3217          }
3215 3218  
3216 3219          /*
3217 3220           * Add this entry to the cache.
3218 3221           */
3219 3222          avl_insert(&rp->r_dir, nrdc, where);
3220 3223          rddir_cache_hold(nrdc);
3221 3224          mutex_exit(&rp->r_statelock);
3222 3225  
3223 3226  bottom:
3224 3227  #ifdef DEBUG
3225 3228          missed = 1;
3226 3229          nfs_readdir_cache_misses++;
3227 3230  #endif
3228 3231          /*
3229 3232           * Do the readdir.
3230 3233           */
3231 3234          error = nfsreaddir(vp, nrdc, cr);
3232 3235  
3233 3236          /*
3234 3237           * If this operation failed, just return the error which occurred.
3235 3238           */
3236 3239          if (error != 0)
3237 3240                  return (error);
3238 3241  
3239 3242          /*
3240 3243           * Since the RPC operation will have taken sometime and blocked
3241 3244           * this process, another pass through the cache will need to be
3242 3245           * taken to find the correct cache entry.  It is possible that
3243 3246           * the correct cache entry will not be there (although one was
3244 3247           * added) because the directory changed during the RPC operation
3245 3248           * and the readdir cache was flushed.  In this case, just start
3246 3249           * over.  It is hoped that this will not happen too often... :-)
3247 3250           */
3248 3251          nrdc = NULL;
3249 3252          goto top;
3250 3253          /* NOTREACHED */
3251 3254  }
3252 3255  
3253 3256  static int
3254 3257  nfsreaddir(vnode_t *vp, rddir_cache *rdc, cred_t *cr)
3255 3258  {
3256 3259          int error;
3257 3260          struct nfsrddirargs rda;
3258 3261          struct nfsrddirres rd;
3259 3262          rnode_t *rp;
3260 3263          mntinfo_t *mi;
3261 3264          uint_t count;
3262 3265          int douprintf;
3263 3266          failinfo_t fi, *fip;
3264 3267  
3265 3268          ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
3266 3269          count = rdc->buflen;
3267 3270  
3268 3271          rp = VTOR(vp);
3269 3272          mi = VTOMI(vp);
3270 3273  
3271 3274          rda.rda_fh = *VTOFH(vp);
3272 3275          rda.rda_offset = rdc->nfs_cookie;
3273 3276  
3274 3277          /*
3275 3278           * NFS client failover support
3276 3279           * suppress failover unless we have a zero cookie
3277 3280           */
3278 3281          if (rdc->nfs_cookie == (off_t)0) {
3279 3282                  fi.vp = vp;
3280 3283                  fi.fhp = (caddr_t)&rda.rda_fh;
3281 3284                  fi.copyproc = nfscopyfh;
3282 3285                  fi.lookupproc = nfslookup;
3283 3286                  fi.xattrdirproc = acl_getxattrdir2;
3284 3287                  fip = &fi;
3285 3288          } else {
3286 3289                  fip = NULL;
3287 3290          }
3288 3291  
3289 3292          rd.rd_entries = kmem_alloc(rdc->buflen, KM_SLEEP);
3290 3293          rd.rd_size = count;
3291 3294          rd.rd_offset = rda.rda_offset;
3292 3295  
3293 3296          douprintf = 1;
3294 3297  
3295 3298          if (mi->mi_io_kstats) {
3296 3299                  mutex_enter(&mi->mi_lock);
3297 3300                  kstat_runq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
3298 3301                  mutex_exit(&mi->mi_lock);
3299 3302          }
3300 3303  
3301 3304          do {
3302 3305                  rda.rda_count = MIN(count, mi->mi_curread);
3303 3306                  error = rfs2call(mi, RFS_READDIR,
3304 3307                      xdr_rddirargs, (caddr_t)&rda,
3305 3308                      xdr_getrddirres, (caddr_t)&rd, cr,
3306 3309                      &douprintf, &rd.rd_status, 0, fip);
3307 3310          } while (error == ENFS_TRYAGAIN);
3308 3311  
3309 3312          if (mi->mi_io_kstats) {
3310 3313                  mutex_enter(&mi->mi_lock);
3311 3314                  kstat_runq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
3312 3315                  mutex_exit(&mi->mi_lock);
3313 3316          }
3314 3317  
3315 3318          /*
3316 3319           * Since we are actually doing a READDIR RPC, we must have
3317 3320           * exclusive access to the cache entry being filled.  Thus,
3318 3321           * it is safe to update all fields except for the flags
3319 3322           * field.  The r_statelock in the rnode must be held to
3320 3323           * prevent two different threads from simultaneously
3321 3324           * attempting to update the flags field.  This can happen
3322 3325           * if we are turning off RDDIR and the other thread is
3323 3326           * trying to set RDDIRWAIT.
3324 3327           */
3325 3328          ASSERT(rdc->flags & RDDIR);
3326 3329          if (!error) {
3327 3330                  error = geterrno(rd.rd_status);
3328 3331                  if (!error) {
3329 3332                          rdc->nfs_ncookie = rd.rd_offset;
3330 3333                          rdc->eof = rd.rd_eof ? 1 : 0;
3331 3334                          rdc->entlen = rd.rd_size;
3332 3335                          ASSERT(rdc->entlen <= rdc->buflen);
3333 3336  #ifdef DEBUG
3334 3337                          rdc->entries = rddir_cache_buf_alloc(rdc->buflen,
3335 3338                              KM_SLEEP);
3336 3339  #else
3337 3340                          rdc->entries = kmem_alloc(rdc->buflen, KM_SLEEP);
3338 3341  #endif
3339 3342                          bcopy(rd.rd_entries, rdc->entries, rdc->entlen);
3340 3343                          rdc->error = 0;
3341 3344                          if (mi->mi_io_kstats) {
3342 3345                                  mutex_enter(&mi->mi_lock);
3343 3346                                  KSTAT_IO_PTR(mi->mi_io_kstats)->reads++;
3344 3347                                  KSTAT_IO_PTR(mi->mi_io_kstats)->nread +=
3345 3348                                      rd.rd_size;
3346 3349                                  mutex_exit(&mi->mi_lock);
3347 3350                          }
3348 3351                  } else {
3349 3352                          PURGE_STALE_FH(error, vp, cr);
3350 3353                  }
3351 3354          }
3352 3355          if (error) {
3353 3356                  rdc->entries = NULL;
3354 3357                  rdc->error = error;
3355 3358          }
3356 3359          kmem_free(rd.rd_entries, rdc->buflen);
3357 3360  
3358 3361          mutex_enter(&rp->r_statelock);
3359 3362          rdc->flags &= ~RDDIR;
3360 3363          if (rdc->flags & RDDIRWAIT) {
3361 3364                  rdc->flags &= ~RDDIRWAIT;
3362 3365                  cv_broadcast(&rdc->cv);
3363 3366          }
3364 3367          if (error)
3365 3368                  rdc->flags |= RDDIRREQ;
3366 3369          mutex_exit(&rp->r_statelock);
3367 3370  
3368 3371          rddir_cache_rele(rdc);
3369 3372  
3370 3373          return (error);
3371 3374  }
3372 3375  
3373 3376  #ifdef DEBUG
3374 3377  static int nfs_bio_do_stop = 0;
3375 3378  #endif
3376 3379  
3377 3380  static int
3378 3381  nfs_bio(struct buf *bp, cred_t *cr)
3379 3382  {
3380 3383          rnode_t *rp = VTOR(bp->b_vp);
3381 3384          int count;
3382 3385          int error;
3383 3386          cred_t *cred;
3384 3387          uint_t offset;
3385 3388  
3386 3389          DTRACE_IO1(start, struct buf *, bp);
3387 3390  
3388 3391          ASSERT(nfs_zone() == VTOMI(bp->b_vp)->mi_zone);
3389 3392          offset = dbtob(bp->b_blkno);
3390 3393  
3391 3394          if (bp->b_flags & B_READ) {
3392 3395                  mutex_enter(&rp->r_statelock);
3393 3396                  if (rp->r_cred != NULL) {
3394 3397                          cred = rp->r_cred;
3395 3398                          crhold(cred);
3396 3399                  } else {
3397 3400                          rp->r_cred = cr;
3398 3401                          crhold(cr);
3399 3402                          cred = cr;
3400 3403                          crhold(cred);
3401 3404                  }
3402 3405                  mutex_exit(&rp->r_statelock);
3403 3406          read_again:
3404 3407                  error = bp->b_error = nfsread(bp->b_vp, bp->b_un.b_addr,
3405 3408                      offset, bp->b_bcount, &bp->b_resid, cred);
3406 3409  
3407 3410                  crfree(cred);
3408 3411                  if (!error) {
3409 3412                          if (bp->b_resid) {
3410 3413                                  /*
3411 3414                                   * Didn't get it all because we hit EOF,
3412 3415                                   * zero all the memory beyond the EOF.
3413 3416                                   */
3414 3417                                  /* bzero(rdaddr + */
3415 3418                                  bzero(bp->b_un.b_addr +
3416 3419                                      bp->b_bcount - bp->b_resid, bp->b_resid);
3417 3420                          }
3418 3421                          mutex_enter(&rp->r_statelock);
3419 3422                          if (bp->b_resid == bp->b_bcount &&
3420 3423                              offset >= rp->r_size) {
3421 3424                                  /*
3422 3425                                   * We didn't read anything at all as we are
3423 3426                                   * past EOF.  Return an error indicator back
3424 3427                                   * but don't destroy the pages (yet).
3425 3428                                   */
3426 3429                                  error = NFS_EOF;
3427 3430                          }
3428 3431                          mutex_exit(&rp->r_statelock);
3429 3432                  } else if (error == EACCES) {
3430 3433                          mutex_enter(&rp->r_statelock);
3431 3434                          if (cred != cr) {
3432 3435                                  if (rp->r_cred != NULL)
3433 3436                                          crfree(rp->r_cred);
3434 3437                                  rp->r_cred = cr;
3435 3438                                  crhold(cr);
3436 3439                                  cred = cr;
3437 3440                                  crhold(cred);
3438 3441                                  mutex_exit(&rp->r_statelock);
3439 3442                                  goto read_again;
3440 3443                          }
3441 3444                          mutex_exit(&rp->r_statelock);
3442 3445                  }
3443 3446          } else {
3444 3447                  if (!(rp->r_flags & RSTALE)) {
3445 3448                          mutex_enter(&rp->r_statelock);
3446 3449                          if (rp->r_cred != NULL) {
3447 3450                                  cred = rp->r_cred;
3448 3451                                  crhold(cred);
3449 3452                          } else {
3450 3453                                  rp->r_cred = cr;
3451 3454                                  crhold(cr);
3452 3455                                  cred = cr;
3453 3456                                  crhold(cred);
3454 3457                          }
3455 3458                          mutex_exit(&rp->r_statelock);
3456 3459                  write_again:
3457 3460                          mutex_enter(&rp->r_statelock);
3458 3461                          count = MIN(bp->b_bcount, rp->r_size - offset);
3459 3462                          mutex_exit(&rp->r_statelock);
3460 3463                          if (count < 0)
3461 3464                                  cmn_err(CE_PANIC, "nfs_bio: write count < 0");
3462 3465  #ifdef DEBUG
3463 3466                          if (count == 0) {
3464 3467                                  zcmn_err(getzoneid(), CE_WARN,
3465 3468                                      "nfs_bio: zero length write at %d",
3466 3469                                      offset);
3467 3470                                  nfs_printfhandle(&rp->r_fh);
3468 3471                                  if (nfs_bio_do_stop)
3469 3472                                          debug_enter("nfs_bio");
3470 3473                          }
3471 3474  #endif
3472 3475                          error = nfswrite(bp->b_vp, bp->b_un.b_addr, offset,
3473 3476                              count, cred);
3474 3477                          if (error == EACCES) {
3475 3478                                  mutex_enter(&rp->r_statelock);
3476 3479                                  if (cred != cr) {
3477 3480                                          if (rp->r_cred != NULL)
3478 3481                                                  crfree(rp->r_cred);
3479 3482                                          rp->r_cred = cr;
3480 3483                                          crhold(cr);
3481 3484                                          crfree(cred);
3482 3485                                          cred = cr;
3483 3486                                          crhold(cred);
3484 3487                                          mutex_exit(&rp->r_statelock);
3485 3488                                          goto write_again;
3486 3489                                  }
3487 3490                                  mutex_exit(&rp->r_statelock);
3488 3491                          }
3489 3492                          bp->b_error = error;
3490 3493                          if (error && error != EINTR) {
3491 3494                                  /*
3492 3495                                   * Don't print EDQUOT errors on the console.
3493 3496                                   * Don't print asynchronous EACCES errors.
3494 3497                                   * Don't print EFBIG errors.
3495 3498                                   * Print all other write errors.
3496 3499                                   */
3497 3500                                  if (error != EDQUOT && error != EFBIG &&
3498 3501                                      (error != EACCES ||
3499 3502                                      !(bp->b_flags & B_ASYNC)))
3500 3503                                          nfs_write_error(bp->b_vp, error, cred);
3501 3504                                  /*
3502 3505                                   * Update r_error and r_flags as appropriate.
3503 3506                                   * If the error was ESTALE, then mark the
3504 3507                                   * rnode as not being writeable and save
3505 3508                                   * the error status.  Otherwise, save any
3506 3509                                   * errors which occur from asynchronous
3507 3510                                   * page invalidations.  Any errors occurring
3508 3511                                   * from other operations should be saved
3509 3512                                   * by the caller.
3510 3513                                   */
3511 3514                                  mutex_enter(&rp->r_statelock);
3512 3515                                  if (error == ESTALE) {
3513 3516                                          rp->r_flags |= RSTALE;
3514 3517                                          if (!rp->r_error)
3515 3518                                                  rp->r_error = error;
3516 3519                                  } else if (!rp->r_error &&
3517 3520                                      (bp->b_flags &
3518 3521                                      (B_INVAL|B_FORCE|B_ASYNC)) ==
3519 3522                                      (B_INVAL|B_FORCE|B_ASYNC)) {
3520 3523                                          rp->r_error = error;
3521 3524                                  }
3522 3525                                  mutex_exit(&rp->r_statelock);
3523 3526                          }
3524 3527                          crfree(cred);
3525 3528                  } else {
3526 3529                          error = rp->r_error;
3527 3530                          /*
3528 3531                           * A close may have cleared r_error, if so,
3529 3532                           * propagate ESTALE error return properly
3530 3533                           */
3531 3534                          if (error == 0)
3532 3535                                  error = ESTALE;
3533 3536                  }
3534 3537          }
3535 3538  
3536 3539          if (error != 0 && error != NFS_EOF)
3537 3540                  bp->b_flags |= B_ERROR;
3538 3541  
3539 3542          DTRACE_IO1(done, struct buf *, bp);
3540 3543  
3541 3544          return (error);
3542 3545  }
3543 3546  
3544 3547  /* ARGSUSED */
3545 3548  static int
3546 3549  nfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
3547 3550  {
3548 3551          struct nfs_fid *fp;
3549 3552          rnode_t *rp;
3550 3553  
3551 3554          rp = VTOR(vp);
3552 3555  
3553 3556          if (fidp->fid_len < (sizeof (struct nfs_fid) - sizeof (short))) {
3554 3557                  fidp->fid_len = sizeof (struct nfs_fid) - sizeof (short);
3555 3558                  return (ENOSPC);
3556 3559          }
3557 3560          fp = (struct nfs_fid *)fidp;
3558 3561          fp->nf_pad = 0;
3559 3562          fp->nf_len = sizeof (struct nfs_fid) - sizeof (short);
3560 3563          bcopy(rp->r_fh.fh_buf, fp->nf_data, NFS_FHSIZE);
3561 3564          return (0);
3562 3565  }
3563 3566  
3564 3567  /* ARGSUSED2 */
3565 3568  static int
3566 3569  nfs_rwlock(vnode_t *vp, int write_lock, caller_context_t *ctp)
3567 3570  {
3568 3571          rnode_t *rp = VTOR(vp);
3569 3572  
3570 3573          if (!write_lock) {
3571 3574                  (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE);
3572 3575                  return (V_WRITELOCK_FALSE);
3573 3576          }
3574 3577  
3575 3578          if ((rp->r_flags & RDIRECTIO) || (VTOMI(vp)->mi_flags & MI_DIRECTIO)) {
3576 3579                  (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_READER, FALSE);
3577 3580                  if (rp->r_mapcnt == 0 && !vn_has_cached_data(vp))
3578 3581                          return (V_WRITELOCK_FALSE);
3579 3582                  nfs_rw_exit(&rp->r_rwlock);
3580 3583          }
3581 3584  
3582 3585          (void) nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, FALSE);
3583 3586          return (V_WRITELOCK_TRUE);
3584 3587  }
3585 3588  
3586 3589  /* ARGSUSED */
3587 3590  static void
3588 3591  nfs_rwunlock(vnode_t *vp, int write_lock, caller_context_t *ctp)
3589 3592  {
3590 3593          rnode_t *rp = VTOR(vp);
3591 3594  
3592 3595          nfs_rw_exit(&rp->r_rwlock);
3593 3596  }
3594 3597  
3595 3598  /* ARGSUSED */
3596 3599  static int
3597 3600  nfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, caller_context_t *ct)
3598 3601  {
3599 3602  
3600 3603          /*
3601 3604           * Because we stuff the readdir cookie into the offset field
3602 3605           * someone may attempt to do an lseek with the cookie which
3603 3606           * we want to succeed.
3604 3607           */
3605 3608          if (vp->v_type == VDIR)
3606 3609                  return (0);
3607 3610          if (*noffp < 0 || *noffp > MAXOFF32_T)
3608 3611                  return (EINVAL);
3609 3612          return (0);
3610 3613  }
3611 3614  
3612 3615  /*
3613 3616   * number of NFS_MAXDATA blocks to read ahead
3614 3617   * optimized for 100 base-T.
3615 3618   */
3616 3619  static int nfs_nra = 4;
3617 3620  
3618 3621  #ifdef DEBUG
3619 3622  static int nfs_lostpage = 0;    /* number of times we lost original page */
3620 3623  #endif
3621 3624  
3622 3625  /*
3623 3626   * Return all the pages from [off..off+len) in file
3624 3627   */
3625 3628  /* ARGSUSED */
3626 3629  static int
3627 3630  nfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp,
3628 3631          page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3629 3632          enum seg_rw rw, cred_t *cr, caller_context_t *ct)
3630 3633  {
3631 3634          rnode_t *rp;
3632 3635          int error;
3633 3636          mntinfo_t *mi;
3634 3637  
3635 3638          if (vp->v_flag & VNOMAP)
3636 3639                  return (ENOSYS);
3637 3640  
3638 3641          ASSERT(off <= MAXOFF32_T);
3639 3642          if (nfs_zone() != VTOMI(vp)->mi_zone)
3640 3643                  return (EIO);
3641 3644          if (protp != NULL)
3642 3645                  *protp = PROT_ALL;
3643 3646  
3644 3647          /*
3645 3648           * Now valididate that the caches are up to date.
3646 3649           */
3647 3650          error = nfs_validate_caches(vp, cr);
3648 3651          if (error)
3649 3652                  return (error);
3650 3653  
3651 3654          rp = VTOR(vp);
3652 3655          mi = VTOMI(vp);
3653 3656  retry:
3654 3657          mutex_enter(&rp->r_statelock);
3655 3658  
3656 3659          /*
3657 3660           * Don't create dirty pages faster than they
3658 3661           * can be cleaned so that the system doesn't
3659 3662           * get imbalanced.  If the async queue is
3660 3663           * maxed out, then wait for it to drain before
3661 3664           * creating more dirty pages.  Also, wait for
3662 3665           * any threads doing pagewalks in the vop_getattr
3663 3666           * entry points so that they don't block for
3664 3667           * long periods.
3665 3668           */
3666 3669          if (rw == S_CREATE) {
3667 3670                  while ((mi->mi_max_threads != 0 &&
3668 3671                      rp->r_awcount > 2 * mi->mi_max_threads) ||
3669 3672                      rp->r_gcount > 0)
3670 3673                          cv_wait(&rp->r_cv, &rp->r_statelock);
3671 3674          }
3672 3675  
3673 3676          /*
3674 3677           * If we are getting called as a side effect of an nfs_write()
3675 3678           * operation the local file size might not be extended yet.
3676 3679           * In this case we want to be able to return pages of zeroes.
3677 3680           */
3678 3681          if (off + len > rp->r_size + PAGEOFFSET && seg != segkmap) {
3679 3682                  mutex_exit(&rp->r_statelock);
3680 3683                  return (EFAULT);                /* beyond EOF */
3681 3684          }
3682 3685  
3683 3686          mutex_exit(&rp->r_statelock);
3684 3687  
3685 3688          error = pvn_getpages(nfs_getapage, vp, off, len, protp, pl, plsz,
3686 3689              seg, addr, rw, cr);
3687 3690  
3688 3691          switch (error) {
3689 3692          case NFS_EOF:
3690 3693                  nfs_purge_caches(vp, NFS_NOPURGE_DNLC, cr);
3691 3694                  goto retry;
3692 3695          case ESTALE:
3693 3696                  PURGE_STALE_FH(error, vp, cr);
3694 3697          }
3695 3698  
3696 3699          return (error);
3697 3700  }
3698 3701  
3699 3702  /*
3700 3703   * Called from pvn_getpages to get a particular page.
3701 3704   */
3702 3705  /* ARGSUSED */
3703 3706  static int
3704 3707  nfs_getapage(vnode_t *vp, u_offset_t off, size_t len, uint_t *protp,
3705 3708          page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr,
3706 3709          enum seg_rw rw, cred_t *cr)
3707 3710  {
3708 3711          rnode_t *rp;
3709 3712          uint_t bsize;
3710 3713          struct buf *bp;
3711 3714          page_t *pp;
3712 3715          u_offset_t lbn;
3713 3716          u_offset_t io_off;
3714 3717          u_offset_t blkoff;
3715 3718          u_offset_t rablkoff;
3716 3719          size_t io_len;
3717 3720          uint_t blksize;
3718 3721          int error;
3719 3722          int readahead;
3720 3723          int readahead_issued = 0;
3721 3724          int ra_window; /* readahead window */
3722 3725          page_t *pagefound;
3723 3726  
3724 3727          if (nfs_zone() != VTOMI(vp)->mi_zone)
3725 3728                  return (EIO);
3726 3729          rp = VTOR(vp);
3727 3730          bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
3728 3731  
3729 3732  reread:
3730 3733          bp = NULL;
3731 3734          pp = NULL;
3732 3735          pagefound = NULL;
3733 3736  
3734 3737          if (pl != NULL)
3735 3738                  pl[0] = NULL;
3736 3739  
3737 3740          error = 0;
3738 3741          lbn = off / bsize;
3739 3742          blkoff = lbn * bsize;
3740 3743  
3741 3744          /*
3742 3745           * Queueing up the readahead before doing the synchronous read
3743 3746           * results in a significant increase in read throughput because
3744 3747           * of the increased parallelism between the async threads and
3745 3748           * the process context.
3746 3749           */
3747 3750          if ((off & ((vp->v_vfsp->vfs_bsize) - 1)) == 0 &&
3748 3751              rw != S_CREATE &&
3749 3752              !(vp->v_flag & VNOCACHE)) {
3750 3753                  mutex_enter(&rp->r_statelock);
3751 3754  
3752 3755                  /*
3753 3756                   * Calculate the number of readaheads to do.
3754 3757                   * a) No readaheads at offset = 0.
3755 3758                   * b) Do maximum(nfs_nra) readaheads when the readahead
3756 3759                   *    window is closed.
3757 3760                   * c) Do readaheads between 1 to (nfs_nra - 1) depending
3758 3761                   *    upon how far the readahead window is open or close.
3759 3762                   * d) No readaheads if rp->r_nextr is not within the scope
3760 3763                   *    of the readahead window (random i/o).
3761 3764                   */
3762 3765  
3763 3766                  if (off == 0)
3764 3767                          readahead = 0;
3765 3768                  else if (blkoff == rp->r_nextr)
3766 3769                          readahead = nfs_nra;
3767 3770                  else if (rp->r_nextr > blkoff &&
3768 3771                      ((ra_window = (rp->r_nextr - blkoff) / bsize)
3769 3772                      <= (nfs_nra - 1)))
3770 3773                          readahead = nfs_nra - ra_window;
3771 3774                  else
3772 3775                          readahead = 0;
3773 3776  
3774 3777                  rablkoff = rp->r_nextr;
3775 3778                  while (readahead > 0 && rablkoff + bsize < rp->r_size) {
3776 3779                          mutex_exit(&rp->r_statelock);
3777 3780                          if (nfs_async_readahead(vp, rablkoff + bsize,
3778 3781                              addr + (rablkoff + bsize - off), seg, cr,
3779 3782                              nfs_readahead) < 0) {
3780 3783                                  mutex_enter(&rp->r_statelock);
3781 3784                                  break;
3782 3785                          }
3783 3786                          readahead--;
3784 3787                          rablkoff += bsize;
3785 3788                          /*
3786 3789                           * Indicate that we did a readahead so
3787 3790                           * readahead offset is not updated
3788 3791                           * by the synchronous read below.
3789 3792                           */
3790 3793                          readahead_issued = 1;
3791 3794                          mutex_enter(&rp->r_statelock);
3792 3795                          /*
3793 3796                           * set readahead offset to
3794 3797                           * offset of last async readahead
3795 3798                           * request.
3796 3799                           */
3797 3800                          rp->r_nextr = rablkoff;
3798 3801                  }
3799 3802                  mutex_exit(&rp->r_statelock);
3800 3803          }
3801 3804  
3802 3805  again:
3803 3806          if ((pagefound = page_exists(vp, off)) == NULL) {
3804 3807                  if (pl == NULL) {
3805 3808                          (void) nfs_async_readahead(vp, blkoff, addr, seg, cr,
3806 3809                              nfs_readahead);
3807 3810                  } else if (rw == S_CREATE) {
3808 3811                          /*
3809 3812                           * Block for this page is not allocated, or the offset
3810 3813                           * is beyond the current allocation size, or we're
3811 3814                           * allocating a swap slot and the page was not found,
3812 3815                           * so allocate it and return a zero page.
3813 3816                           */
3814 3817                          if ((pp = page_create_va(vp, off,
3815 3818                              PAGESIZE, PG_WAIT, seg, addr)) == NULL)
3816 3819                                  cmn_err(CE_PANIC, "nfs_getapage: page_create");
3817 3820                          io_len = PAGESIZE;
3818 3821                          mutex_enter(&rp->r_statelock);
3819 3822                          rp->r_nextr = off + PAGESIZE;
3820 3823                          mutex_exit(&rp->r_statelock);
3821 3824                  } else {
3822 3825                          /*
3823 3826                           * Need to go to server to get a BLOCK, exception to
3824 3827                           * that being while reading at offset = 0 or doing
3825 3828                           * random i/o, in that case read only a PAGE.
3826 3829                           */
3827 3830                          mutex_enter(&rp->r_statelock);
3828 3831                          if (blkoff < rp->r_size &&
3829 3832                              blkoff + bsize >= rp->r_size) {
3830 3833                                  /*
3831 3834                                   * If only a block or less is left in
3832 3835                                   * the file, read all that is remaining.
3833 3836                                   */
3834 3837                                  if (rp->r_size <= off) {
3835 3838                                          /*
3836 3839                                           * Trying to access beyond EOF,
3837 3840                                           * set up to get at least one page.
3838 3841                                           */
3839 3842                                          blksize = off + PAGESIZE - blkoff;
3840 3843                                  } else
3841 3844                                          blksize = rp->r_size - blkoff;
3842 3845                          } else if ((off == 0) ||
3843 3846                              (off != rp->r_nextr && !readahead_issued)) {
3844 3847                                  blksize = PAGESIZE;
3845 3848                                  blkoff = off; /* block = page here */
3846 3849                          } else
3847 3850                                  blksize = bsize;
3848 3851                          mutex_exit(&rp->r_statelock);
3849 3852  
3850 3853                          pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
3851 3854                              &io_len, blkoff, blksize, 0);
3852 3855  
3853 3856                          /*
3854 3857                           * Some other thread has entered the page,
3855 3858                           * so just use it.
3856 3859                           */
3857 3860                          if (pp == NULL)
3858 3861                                  goto again;
3859 3862  
3860 3863                          /*
3861 3864                           * Now round the request size up to page boundaries.
3862 3865                           * This ensures that the entire page will be
3863 3866                           * initialized to zeroes if EOF is encountered.
3864 3867                           */
3865 3868                          io_len = ptob(btopr(io_len));
3866 3869  
3867 3870                          bp = pageio_setup(pp, io_len, vp, B_READ);
3868 3871                          ASSERT(bp != NULL);
3869 3872  
3870 3873                          /*
3871 3874                           * pageio_setup should have set b_addr to 0.  This
3872 3875                           * is correct since we want to do I/O on a page
3873 3876                           * boundary.  bp_mapin will use this addr to calculate
3874 3877                           * an offset, and then set b_addr to the kernel virtual
3875 3878                           * address it allocated for us.
3876 3879                           */
3877 3880                          ASSERT(bp->b_un.b_addr == 0);
3878 3881  
3879 3882                          bp->b_edev = 0;
3880 3883                          bp->b_dev = 0;
3881 3884                          bp->b_lblkno = lbtodb(io_off);
3882 3885                          bp->b_file = vp;
3883 3886                          bp->b_offset = (offset_t)off;
3884 3887                          bp_mapin(bp);
3885 3888  
3886 3889                          /*
3887 3890                           * If doing a write beyond what we believe is EOF,
3888 3891                           * don't bother trying to read the pages from the
3889 3892                           * server, we'll just zero the pages here.  We
3890 3893                           * don't check that the rw flag is S_WRITE here
3891 3894                           * because some implementations may attempt a
3892 3895                           * read access to the buffer before copying data.
3893 3896                           */
3894 3897                          mutex_enter(&rp->r_statelock);
3895 3898                          if (io_off >= rp->r_size && seg == segkmap) {
3896 3899                                  mutex_exit(&rp->r_statelock);
3897 3900                                  bzero(bp->b_un.b_addr, io_len);
3898 3901                          } else {
3899 3902                                  mutex_exit(&rp->r_statelock);
3900 3903                                  error = nfs_bio(bp, cr);
3901 3904                          }
3902 3905  
3903 3906                          /*
3904 3907                           * Unmap the buffer before freeing it.
3905 3908                           */
3906 3909                          bp_mapout(bp);
3907 3910                          pageio_done(bp);
3908 3911  
3909 3912                          if (error == NFS_EOF) {
3910 3913                                  /*
3911 3914                                   * If doing a write system call just return
3912 3915                                   * zeroed pages, else user tried to get pages
3913 3916                                   * beyond EOF, return error.  We don't check
3914 3917                                   * that the rw flag is S_WRITE here because
3915 3918                                   * some implementations may attempt a read
3916 3919                                   * access to the buffer before copying data.
3917 3920                                   */
3918 3921                                  if (seg == segkmap)
3919 3922                                          error = 0;
3920 3923                                  else
3921 3924                                          error = EFAULT;
3922 3925                          }
3923 3926  
3924 3927                          if (!readahead_issued && !error) {
3925 3928                                  mutex_enter(&rp->r_statelock);
3926 3929                                  rp->r_nextr = io_off + io_len;
3927 3930                                  mutex_exit(&rp->r_statelock);
3928 3931                          }
3929 3932                  }
3930 3933          }
3931 3934  
3932 3935  out:
3933 3936          if (pl == NULL)
3934 3937                  return (error);
3935 3938  
3936 3939          if (error) {
3937 3940                  if (pp != NULL)
3938 3941                          pvn_read_done(pp, B_ERROR);
3939 3942                  return (error);
3940 3943          }
3941 3944  
3942 3945          if (pagefound) {
3943 3946                  se_t se = (rw == S_CREATE ? SE_EXCL : SE_SHARED);
3944 3947  
3945 3948                  /*
3946 3949                   * Page exists in the cache, acquire the appropriate lock.
3947 3950                   * If this fails, start all over again.
3948 3951                   */
3949 3952                  if ((pp = page_lookup(vp, off, se)) == NULL) {
3950 3953  #ifdef DEBUG
3951 3954                          nfs_lostpage++;
3952 3955  #endif
3953 3956                          goto reread;
3954 3957                  }
3955 3958                  pl[0] = pp;
3956 3959                  pl[1] = NULL;
3957 3960                  return (0);
3958 3961          }
3959 3962  
3960 3963          if (pp != NULL)
3961 3964                  pvn_plist_init(pp, pl, plsz, off, io_len, rw);
3962 3965  
3963 3966          return (error);
3964 3967  }
3965 3968  
3966 3969  static void
3967 3970  nfs_readahead(vnode_t *vp, u_offset_t blkoff, caddr_t addr, struct seg *seg,
3968 3971          cred_t *cr)
3969 3972  {
3970 3973          int error;
3971 3974          page_t *pp;
3972 3975          u_offset_t io_off;
3973 3976          size_t io_len;
3974 3977          struct buf *bp;
3975 3978          uint_t bsize, blksize;
3976 3979          rnode_t *rp = VTOR(vp);
3977 3980  
3978 3981          ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
3979 3982  
3980 3983          bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
3981 3984  
3982 3985          mutex_enter(&rp->r_statelock);
3983 3986          if (blkoff < rp->r_size && blkoff + bsize > rp->r_size) {
3984 3987                  /*
3985 3988                   * If less than a block left in file read less
3986 3989                   * than a block.
3987 3990                   */
3988 3991                  blksize = rp->r_size - blkoff;
3989 3992          } else
3990 3993                  blksize = bsize;
3991 3994          mutex_exit(&rp->r_statelock);
3992 3995  
3993 3996          pp = pvn_read_kluster(vp, blkoff, segkmap, addr,
3994 3997              &io_off, &io_len, blkoff, blksize, 1);
3995 3998          /*
3996 3999           * The isra flag passed to the kluster function is 1, we may have
3997 4000           * gotten a return value of NULL for a variety of reasons (# of free
3998 4001           * pages < minfree, someone entered the page on the vnode etc). In all
3999 4002           * cases, we want to punt on the readahead.
4000 4003           */
4001 4004          if (pp == NULL)
4002 4005                  return;
4003 4006  
4004 4007          /*
4005 4008           * Now round the request size up to page boundaries.
4006 4009           * This ensures that the entire page will be
4007 4010           * initialized to zeroes if EOF is encountered.
4008 4011           */
4009 4012          io_len = ptob(btopr(io_len));
4010 4013  
4011 4014          bp = pageio_setup(pp, io_len, vp, B_READ);
4012 4015          ASSERT(bp != NULL);
4013 4016  
4014 4017          /*
4015 4018           * pageio_setup should have set b_addr to 0.  This is correct since
4016 4019           * we want to do I/O on a page boundary. bp_mapin() will use this addr
4017 4020           * to calculate an offset, and then set b_addr to the kernel virtual
4018 4021           * address it allocated for us.
4019 4022           */
4020 4023          ASSERT(bp->b_un.b_addr == 0);
4021 4024  
4022 4025          bp->b_edev = 0;
4023 4026          bp->b_dev = 0;
4024 4027          bp->b_lblkno = lbtodb(io_off);
4025 4028          bp->b_file = vp;
4026 4029          bp->b_offset = (offset_t)blkoff;
4027 4030          bp_mapin(bp);
4028 4031  
4029 4032          /*
4030 4033           * If doing a write beyond what we believe is EOF, don't bother trying
4031 4034           * to read the pages from the server, we'll just zero the pages here.
4032 4035           * We don't check that the rw flag is S_WRITE here because some
4033 4036           * implementations may attempt a read access to the buffer before
4034 4037           * copying data.
4035 4038           */
4036 4039          mutex_enter(&rp->r_statelock);
4037 4040          if (io_off >= rp->r_size && seg == segkmap) {
4038 4041                  mutex_exit(&rp->r_statelock);
4039 4042                  bzero(bp->b_un.b_addr, io_len);
4040 4043                  error = 0;
4041 4044          } else {
4042 4045                  mutex_exit(&rp->r_statelock);
4043 4046                  error = nfs_bio(bp, cr);
4044 4047                  if (error == NFS_EOF)
4045 4048                          error = 0;
4046 4049          }
4047 4050  
4048 4051          /*
4049 4052           * Unmap the buffer before freeing it.
4050 4053           */
4051 4054          bp_mapout(bp);
4052 4055          pageio_done(bp);
4053 4056  
4054 4057          pvn_read_done(pp, error ? B_READ | B_ERROR : B_READ);
4055 4058  
4056 4059          /*
4057 4060           * In case of error set readahead offset
4058 4061           * to the lowest offset.
4059 4062           * pvn_read_done() calls VN_DISPOSE to destroy the pages
4060 4063           */
4061 4064          if (error && rp->r_nextr > io_off) {
4062 4065                  mutex_enter(&rp->r_statelock);
4063 4066                  if (rp->r_nextr > io_off)
4064 4067                          rp->r_nextr = io_off;
4065 4068                  mutex_exit(&rp->r_statelock);
4066 4069          }
4067 4070  }
4068 4071  
4069 4072  /*
4070 4073   * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE}
4071 4074   * If len == 0, do from off to EOF.
4072 4075   *
4073 4076   * The normal cases should be len == 0 && off == 0 (entire vp list),
4074 4077   * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
4075 4078   * (from pageout).
4076 4079   */
4077 4080  /* ARGSUSED */
4078 4081  static int
4079 4082  nfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr,
4080 4083          caller_context_t *ct)
4081 4084  {
4082 4085          int error;
4083 4086          rnode_t *rp;
4084 4087  
4085 4088          ASSERT(cr != NULL);
4086 4089  
4087 4090          /*
4088 4091           * XXX - Why should this check be made here?
4089 4092           */
4090 4093          if (vp->v_flag & VNOMAP)
4091 4094                  return (ENOSYS);
4092 4095  
4093 4096          if (len == 0 && !(flags & B_INVAL) && vn_is_readonly(vp))
4094 4097                  return (0);
4095 4098  
4096 4099          if (!(flags & B_ASYNC) && nfs_zone() != VTOMI(vp)->mi_zone)
4097 4100                  return (EIO);
4098 4101          ASSERT(off <= MAXOFF32_T);
4099 4102  
4100 4103          rp = VTOR(vp);
4101 4104          mutex_enter(&rp->r_statelock);
4102 4105          rp->r_count++;
4103 4106          mutex_exit(&rp->r_statelock);
4104 4107          error = nfs_putpages(vp, off, len, flags, cr);
4105 4108          mutex_enter(&rp->r_statelock);
4106 4109          rp->r_count--;
4107 4110          cv_broadcast(&rp->r_cv);
4108 4111          mutex_exit(&rp->r_statelock);
4109 4112  
4110 4113          return (error);
4111 4114  }
4112 4115  
4113 4116  /*
4114 4117   * Write out a single page, possibly klustering adjacent dirty pages.
4115 4118   */
4116 4119  int
4117 4120  nfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
4118 4121          int flags, cred_t *cr)
4119 4122  {
4120 4123          u_offset_t io_off;
4121 4124          u_offset_t lbn_off;
4122 4125          u_offset_t lbn;
4123 4126          size_t io_len;
4124 4127          uint_t bsize;
4125 4128          int error;
4126 4129          rnode_t *rp;
4127 4130  
4128 4131          ASSERT(!vn_is_readonly(vp));
4129 4132          ASSERT(pp != NULL);
4130 4133          ASSERT(cr != NULL);
4131 4134          ASSERT((flags & B_ASYNC) || nfs_zone() == VTOMI(vp)->mi_zone);
4132 4135  
4133 4136          rp = VTOR(vp);
4134 4137          ASSERT(rp->r_count > 0);
4135 4138  
4136 4139          ASSERT(pp->p_offset <= MAXOFF32_T);
4137 4140  
4138 4141          bsize = MAX(vp->v_vfsp->vfs_bsize, PAGESIZE);
4139 4142          lbn = pp->p_offset / bsize;
4140 4143          lbn_off = lbn * bsize;
4141 4144  
4142 4145          /*
4143 4146           * Find a kluster that fits in one block, or in
4144 4147           * one page if pages are bigger than blocks.  If
4145 4148           * there is less file space allocated than a whole
4146 4149           * page, we'll shorten the i/o request below.
4147 4150           */
4148 4151          pp = pvn_write_kluster(vp, pp, &io_off, &io_len, lbn_off,
4149 4152              roundup(bsize, PAGESIZE), flags);
4150 4153  
4151 4154          /*
4152 4155           * pvn_write_kluster shouldn't have returned a page with offset
4153 4156           * behind the original page we were given.  Verify that.
4154 4157           */
4155 4158          ASSERT((pp->p_offset / bsize) >= lbn);
4156 4159  
4157 4160          /*
4158 4161           * Now pp will have the list of kept dirty pages marked for
4159 4162           * write back.  It will also handle invalidation and freeing
4160 4163           * of pages that are not dirty.  Check for page length rounding
4161 4164           * problems.
4162 4165           */
4163 4166          if (io_off + io_len > lbn_off + bsize) {
4164 4167                  ASSERT((io_off + io_len) - (lbn_off + bsize) < PAGESIZE);
4165 4168                  io_len = lbn_off + bsize - io_off;
4166 4169          }
4167 4170          /*
4168 4171           * The RMODINPROGRESS flag makes sure that nfs(3)_bio() sees a
4169 4172           * consistent value of r_size. RMODINPROGRESS is set in writerp().
4170 4173           * When RMODINPROGRESS is set it indicates that a uiomove() is in
4171 4174           * progress and the r_size has not been made consistent with the
4172 4175           * new size of the file. When the uiomove() completes the r_size is
4173 4176           * updated and the RMODINPROGRESS flag is cleared.
4174 4177           *
4175 4178           * The RMODINPROGRESS flag makes sure that nfs(3)_bio() sees a
4176 4179           * consistent value of r_size. Without this handshaking, it is
4177 4180           * possible that nfs(3)_bio() picks  up the old value of r_size
4178 4181           * before the uiomove() in writerp() completes. This will result
4179 4182           * in the write through nfs(3)_bio() being dropped.
4180 4183           *
4181 4184           * More precisely, there is a window between the time the uiomove()
4182 4185           * completes and the time the r_size is updated. If a VOP_PUTPAGE()
4183 4186           * operation intervenes in this window, the page will be picked up,
4184 4187           * because it is dirty (it will be unlocked, unless it was
4185 4188           * pagecreate'd). When the page is picked up as dirty, the dirty
4186 4189           * bit is reset (pvn_getdirty()). In nfs(3)write(), r_size is
4187 4190           * checked. This will still be the old size. Therefore the page will
4188 4191           * not be written out. When segmap_release() calls VOP_PUTPAGE(),
4189 4192           * the page will be found to be clean and the write will be dropped.
4190 4193           */
4191 4194          if (rp->r_flags & RMODINPROGRESS) {
4192 4195                  mutex_enter(&rp->r_statelock);
4193 4196                  if ((rp->r_flags & RMODINPROGRESS) &&
4194 4197                      rp->r_modaddr + MAXBSIZE > io_off &&
4195 4198                      rp->r_modaddr < io_off + io_len) {
4196 4199                          page_t *plist;
4197 4200                          /*
4198 4201                           * A write is in progress for this region of the file.
4199 4202                           * If we did not detect RMODINPROGRESS here then this
4200 4203                           * path through nfs_putapage() would eventually go to
4201 4204                           * nfs(3)_bio() and may not write out all of the data
4202 4205                           * in the pages. We end up losing data. So we decide
4203 4206                           * to set the modified bit on each page in the page
4204 4207                           * list and mark the rnode with RDIRTY. This write
4205 4208                           * will be restarted at some later time.
4206 4209                           */
4207 4210                          plist = pp;
4208 4211                          while (plist != NULL) {
4209 4212                                  pp = plist;
4210 4213                                  page_sub(&plist, pp);
4211 4214                                  hat_setmod(pp);
4212 4215                                  page_io_unlock(pp);
4213 4216                                  page_unlock(pp);
4214 4217                          }
4215 4218                          rp->r_flags |= RDIRTY;
4216 4219                          mutex_exit(&rp->r_statelock);
4217 4220                          if (offp)
4218 4221                                  *offp = io_off;
4219 4222                          if (lenp)
4220 4223                                  *lenp = io_len;
4221 4224                          return (0);
4222 4225                  }
4223 4226                  mutex_exit(&rp->r_statelock);
4224 4227          }
4225 4228  
4226 4229          if (flags & B_ASYNC) {
4227 4230                  error = nfs_async_putapage(vp, pp, io_off, io_len, flags, cr,
4228 4231                      nfs_sync_putapage);
4229 4232          } else
4230 4233                  error = nfs_sync_putapage(vp, pp, io_off, io_len, flags, cr);
4231 4234  
4232 4235          if (offp)
4233 4236                  *offp = io_off;
4234 4237          if (lenp)
4235 4238                  *lenp = io_len;
4236 4239          return (error);
4237 4240  }
4238 4241  
4239 4242  static int
4240 4243  nfs_sync_putapage(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
4241 4244          int flags, cred_t *cr)
4242 4245  {
4243 4246          int error;
4244 4247          rnode_t *rp;
4245 4248  
4246 4249          flags |= B_WRITE;
4247 4250  
4248 4251          ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
4249 4252          error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
4250 4253  
4251 4254          rp = VTOR(vp);
4252 4255  
4253 4256          if ((error == ENOSPC || error == EDQUOT || error == EACCES) &&
4254 4257              (flags & (B_INVAL|B_FORCE)) != (B_INVAL|B_FORCE)) {
4255 4258                  if (!(rp->r_flags & ROUTOFSPACE)) {
4256 4259                          mutex_enter(&rp->r_statelock);
4257 4260                          rp->r_flags |= ROUTOFSPACE;
4258 4261                          mutex_exit(&rp->r_statelock);
4259 4262                  }
4260 4263                  flags |= B_ERROR;
4261 4264                  pvn_write_done(pp, flags);
4262 4265                  /*
4263 4266                   * If this was not an async thread, then try again to
4264 4267                   * write out the pages, but this time, also destroy
4265 4268                   * them whether or not the write is successful.  This
4266 4269                   * will prevent memory from filling up with these
4267 4270                   * pages and destroying them is the only alternative
4268 4271                   * if they can't be written out.
4269 4272                   *
4270 4273                   * Don't do this if this is an async thread because
4271 4274                   * when the pages are unlocked in pvn_write_done,
4272 4275                   * some other thread could have come along, locked
4273 4276                   * them, and queued for an async thread.  It would be
4274 4277                   * possible for all of the async threads to be tied
4275 4278                   * up waiting to lock the pages again and they would
4276 4279                   * all already be locked and waiting for an async
4277 4280                   * thread to handle them.  Deadlock.
4278 4281                   */
4279 4282                  if (!(flags & B_ASYNC)) {
4280 4283                          error = nfs_putpage(vp, io_off, io_len,
4281 4284                              B_INVAL | B_FORCE, cr, NULL);
4282 4285                  }
4283 4286          } else {
4284 4287                  if (error)
4285 4288                          flags |= B_ERROR;
4286 4289                  else if (rp->r_flags & ROUTOFSPACE) {
4287 4290                          mutex_enter(&rp->r_statelock);
4288 4291                          rp->r_flags &= ~ROUTOFSPACE;
4289 4292                          mutex_exit(&rp->r_statelock);
4290 4293                  }
4291 4294                  pvn_write_done(pp, flags);
4292 4295          }
4293 4296  
4294 4297          return (error);
4295 4298  }
4296 4299  
4297 4300  /* ARGSUSED */
4298 4301  static int
4299 4302  nfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp,
4300 4303          size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4301 4304          caller_context_t *ct)
4302 4305  {
4303 4306          struct segvn_crargs vn_a;
4304 4307          int error;
4305 4308          rnode_t *rp;
4306 4309          struct vattr va;
4307 4310  
4308 4311          if (nfs_zone() != VTOMI(vp)->mi_zone)
4309 4312                  return (EIO);
4310 4313  
4311 4314          if (vp->v_flag & VNOMAP)
4312 4315                  return (ENOSYS);
4313 4316  
4314 4317          if (off > MAXOFF32_T)
4315 4318                  return (EFBIG);
4316 4319  
4317 4320          if (off < 0 || off + len < 0)
4318 4321                  return (ENXIO);
4319 4322  
4320 4323          if (vp->v_type != VREG)
4321 4324                  return (ENODEV);
4322 4325  
4323 4326          /*
4324 4327           * If there is cached data and if close-to-open consistency
4325 4328           * checking is not turned off and if the file system is not
4326 4329           * mounted readonly, then force an over the wire getattr.
4327 4330           * Otherwise, just invoke nfsgetattr to get a copy of the
4328 4331           * attributes.  The attribute cache will be used unless it
4329 4332           * is timed out and if it is, then an over the wire getattr
4330 4333           * will be issued.
4331 4334           */
4332 4335          va.va_mask = AT_ALL;
4333 4336          if (vn_has_cached_data(vp) &&
4334 4337              !(VTOMI(vp)->mi_flags & MI_NOCTO) && !vn_is_readonly(vp))
4335 4338                  error = nfs_getattr_otw(vp, &va, cr);
4336 4339          else
4337 4340                  error = nfsgetattr(vp, &va, cr);
4338 4341          if (error)
4339 4342                  return (error);
4340 4343  
4341 4344          /*
4342 4345           * Check to see if the vnode is currently marked as not cachable.
4343 4346           * This means portions of the file are locked (through VOP_FRLOCK).
4344 4347           * In this case the map request must be refused.  We use
4345 4348           * rp->r_lkserlock to avoid a race with concurrent lock requests.
4346 4349           */
4347 4350          rp = VTOR(vp);
4348 4351  
4349 4352          /*
4350 4353           * Atomically increment r_inmap after acquiring r_rwlock. The
4351 4354           * idea here is to acquire r_rwlock to block read/write and
4352 4355           * not to protect r_inmap. r_inmap will inform nfs_read/write()
4353 4356           * that we are in nfs_map(). Now, r_rwlock is acquired in order
4354 4357           * and we can prevent the deadlock that would have occurred
4355 4358           * when nfs_addmap() would have acquired it out of order.
4356 4359           *
4357 4360           * Since we are not protecting r_inmap by any lock, we do not
4358 4361           * hold any lock when we decrement it. We atomically decrement
4359 4362           * r_inmap after we release r_lkserlock.
4360 4363           */
4361 4364  
4362 4365          if (nfs_rw_enter_sig(&rp->r_rwlock, RW_WRITER, INTR(vp)))
4363 4366                  return (EINTR);
4364 4367          atomic_inc_uint(&rp->r_inmap);
4365 4368          nfs_rw_exit(&rp->r_rwlock);
4366 4369  
4367 4370          if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_READER, INTR(vp))) {
4368 4371                  atomic_dec_uint(&rp->r_inmap);
4369 4372                  return (EINTR);
4370 4373          }
4371 4374          if (vp->v_flag & VNOCACHE) {
4372 4375                  error = EAGAIN;
4373 4376                  goto done;
4374 4377          }
4375 4378  
4376 4379          /*
4377 4380           * Don't allow concurrent locks and mapping if mandatory locking is
4378 4381           * enabled.
4379 4382           */
4380 4383          if ((flk_has_remote_locks(vp) || lm_has_sleep(vp)) &&
4381 4384              MANDLOCK(vp, va.va_mode)) {
4382 4385                  error = EAGAIN;
4383 4386                  goto done;
4384 4387          }
4385 4388  
4386 4389          as_rangelock(as);
4387 4390          error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
4388 4391          if (error != 0) {
4389 4392                  as_rangeunlock(as);
4390 4393                  goto done;
4391 4394          }
4392 4395  
4393 4396          vn_a.vp = vp;
4394 4397          vn_a.offset = off;
4395 4398          vn_a.type = (flags & MAP_TYPE);
4396 4399          vn_a.prot = (uchar_t)prot;
4397 4400          vn_a.maxprot = (uchar_t)maxprot;
4398 4401          vn_a.flags = (flags & ~MAP_TYPE);
4399 4402          vn_a.cred = cr;
4400 4403          vn_a.amp = NULL;
4401 4404          vn_a.szc = 0;
4402 4405          vn_a.lgrp_mem_policy_flags = 0;
4403 4406  
4404 4407          error = as_map(as, *addrp, len, segvn_create, &vn_a);
4405 4408          as_rangeunlock(as);
4406 4409  
4407 4410  done:
4408 4411          nfs_rw_exit(&rp->r_lkserlock);
4409 4412          atomic_dec_uint(&rp->r_inmap);
4410 4413          return (error);
4411 4414  }
4412 4415  
4413 4416  /* ARGSUSED */
4414 4417  static int
4415 4418  nfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4416 4419          size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr,
4417 4420          caller_context_t *ct)
4418 4421  {
4419 4422          rnode_t *rp;
4420 4423  
4421 4424          if (vp->v_flag & VNOMAP)
4422 4425                  return (ENOSYS);
4423 4426          if (nfs_zone() != VTOMI(vp)->mi_zone)
4424 4427                  return (EIO);
4425 4428  
4426 4429          rp = VTOR(vp);
4427 4430          atomic_add_long((ulong_t *)&rp->r_mapcnt, btopr(len));
4428 4431  
4429 4432          return (0);
4430 4433  }
4431 4434  
4432 4435  /* ARGSUSED */
4433 4436  static int
4434 4437  nfs_frlock(vnode_t *vp, int cmd, struct flock64 *bfp, int flag, offset_t offset,
4435 4438          struct flk_callback *flk_cbp, cred_t *cr, caller_context_t *ct)
4436 4439  {
4437 4440          netobj lm_fh;
4438 4441          int rc;
4439 4442          u_offset_t start, end;
4440 4443          rnode_t *rp;
4441 4444          int error = 0, intr = INTR(vp);
4442 4445  
4443 4446          /* check for valid cmd parameter */
4444 4447          if (cmd != F_GETLK && cmd != F_SETLK && cmd != F_SETLKW)
4445 4448                  return (EINVAL);
4446 4449          if (nfs_zone() != VTOMI(vp)->mi_zone)
4447 4450                  return (EIO);
4448 4451  
4449 4452          /* Verify l_type. */
4450 4453          switch (bfp->l_type) {
4451 4454          case F_RDLCK:
4452 4455                  if (cmd != F_GETLK && !(flag & FREAD))
4453 4456                          return (EBADF);
4454 4457                  break;
4455 4458          case F_WRLCK:
4456 4459                  if (cmd != F_GETLK && !(flag & FWRITE))
4457 4460                          return (EBADF);
4458 4461                  break;
4459 4462          case F_UNLCK:
4460 4463                  intr = 0;
4461 4464                  break;
4462 4465  
4463 4466          default:
4464 4467                  return (EINVAL);
4465 4468          }
4466 4469  
4467 4470          /* check the validity of the lock range */
4468 4471          if (rc = flk_convert_lock_data(vp, bfp, &start, &end, offset))
4469 4472                  return (rc);
4470 4473          if (rc = flk_check_lock_data(start, end, MAXOFF32_T))
4471 4474                  return (rc);
4472 4475  
4473 4476          /*
4474 4477           * If the filesystem is mounted using local locking, pass the
4475 4478           * request off to the local locking code.
4476 4479           */
4477 4480          if (VTOMI(vp)->mi_flags & MI_LLOCK) {
4478 4481                  if (offset > MAXOFF32_T)
4479 4482                          return (EFBIG);
4480 4483                  if (cmd == F_SETLK || cmd == F_SETLKW) {
4481 4484                          /*
4482 4485                           * For complete safety, we should be holding
4483 4486                           * r_lkserlock.  However, we can't call
4484 4487                           * lm_safelock and then fs_frlock while
4485 4488                           * holding r_lkserlock, so just invoke
4486 4489                           * lm_safelock and expect that this will
4487 4490                           * catch enough of the cases.
4488 4491                           */
4489 4492                          if (!lm_safelock(vp, bfp, cr))
4490 4493                                  return (EAGAIN);
4491 4494                  }
4492 4495                  return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4493 4496          }
4494 4497  
4495 4498          rp = VTOR(vp);
4496 4499  
4497 4500          /*
4498 4501           * Check whether the given lock request can proceed, given the
4499 4502           * current file mappings.
4500 4503           */
4501 4504          if (nfs_rw_enter_sig(&rp->r_lkserlock, RW_WRITER, intr))
4502 4505                  return (EINTR);
4503 4506          if (cmd == F_SETLK || cmd == F_SETLKW) {
4504 4507                  if (!lm_safelock(vp, bfp, cr)) {
4505 4508                          rc = EAGAIN;
4506 4509                          goto done;
4507 4510                  }
4508 4511          }
4509 4512  
4510 4513          /*
4511 4514           * Flush the cache after waiting for async I/O to finish.  For new
4512 4515           * locks, this is so that the process gets the latest bits from the
4513 4516           * server.  For unlocks, this is so that other clients see the
4514 4517           * latest bits once the file has been unlocked.  If currently dirty
4515 4518           * pages can't be flushed, then don't allow a lock to be set.  But
4516 4519           * allow unlocks to succeed, to avoid having orphan locks on the
4517 4520           * server.
4518 4521           */
4519 4522          if (cmd != F_GETLK) {
4520 4523                  mutex_enter(&rp->r_statelock);
4521 4524                  while (rp->r_count > 0) {
4522 4525                          if (intr) {
4523 4526                                  klwp_t *lwp = ttolwp(curthread);
4524 4527  
4525 4528                                  if (lwp != NULL)
4526 4529                                          lwp->lwp_nostop++;
4527 4530                                  if (cv_wait_sig(&rp->r_cv, &rp->r_statelock)
4528 4531                                      == 0) {
4529 4532                                          if (lwp != NULL)
4530 4533                                                  lwp->lwp_nostop--;
4531 4534                                          rc = EINTR;
4532 4535                                          break;
4533 4536                                  }
4534 4537                                  if (lwp != NULL)
4535 4538                                          lwp->lwp_nostop--;
4536 4539                          } else
4537 4540                          cv_wait(&rp->r_cv, &rp->r_statelock);
4538 4541                  }
4539 4542                  mutex_exit(&rp->r_statelock);
4540 4543                  if (rc != 0)
4541 4544                          goto done;
4542 4545                  error = nfs_putpage(vp, (offset_t)0, 0, B_INVAL, cr, ct);
4543 4546                  if (error) {
4544 4547                          if (error == ENOSPC || error == EDQUOT) {
4545 4548                                  mutex_enter(&rp->r_statelock);
4546 4549                                  if (!rp->r_error)
4547 4550                                          rp->r_error = error;
4548 4551                                  mutex_exit(&rp->r_statelock);
4549 4552                          }
4550 4553                          if (bfp->l_type != F_UNLCK) {
4551 4554                                  rc = ENOLCK;
4552 4555                                  goto done;
4553 4556                          }
4554 4557                  }
4555 4558          }
4556 4559  
4557 4560          lm_fh.n_len = sizeof (fhandle_t);
4558 4561          lm_fh.n_bytes = (char *)VTOFH(vp);
4559 4562  
4560 4563          /*
4561 4564           * Call the lock manager to do the real work of contacting
4562 4565           * the server and obtaining the lock.
4563 4566           */
4564 4567          rc = lm_frlock(vp, cmd, bfp, flag, offset, cr, &lm_fh, flk_cbp);
4565 4568  
4566 4569          if (rc == 0)
4567 4570                  nfs_lockcompletion(vp, cmd);
4568 4571  
4569 4572  done:
4570 4573          nfs_rw_exit(&rp->r_lkserlock);
4571 4574          return (rc);
4572 4575  }
4573 4576  
4574 4577  /*
4575 4578   * Free storage space associated with the specified vnode.  The portion
4576 4579   * to be freed is specified by bfp->l_start and bfp->l_len (already
4577 4580   * normalized to a "whence" of 0).
4578 4581   *
4579 4582   * This is an experimental facility whose continued existence is not
4580 4583   * guaranteed.  Currently, we only support the special case
4581 4584   * of l_len == 0, meaning free to end of file.
4582 4585   */
4583 4586  /* ARGSUSED */
4584 4587  static int
4585 4588  nfs_space(vnode_t *vp, int cmd, struct flock64 *bfp, int flag,
4586 4589          offset_t offset, cred_t *cr, caller_context_t *ct)
4587 4590  {
4588 4591          int error;
4589 4592  
4590 4593          ASSERT(vp->v_type == VREG);
4591 4594          if (cmd != F_FREESP)
4592 4595                  return (EINVAL);
4593 4596  
4594 4597          if (offset > MAXOFF32_T)
4595 4598                  return (EFBIG);
4596 4599  
4597 4600          if ((bfp->l_start > MAXOFF32_T) || (bfp->l_end > MAXOFF32_T) ||
4598 4601              (bfp->l_len > MAXOFF32_T))
4599 4602                  return (EFBIG);
4600 4603  
4601 4604          if (nfs_zone() != VTOMI(vp)->mi_zone)
4602 4605                  return (EIO);
4603 4606  
4604 4607          error = convoff(vp, bfp, 0, offset);
4605 4608          if (!error) {
4606 4609                  ASSERT(bfp->l_start >= 0);
4607 4610                  if (bfp->l_len == 0) {
4608 4611                          struct vattr va;
4609 4612  
4610 4613                          /*
4611 4614                           * ftruncate should not change the ctime and
4612 4615                           * mtime if we truncate the file to its
  
    | 
      ↓ open down ↓ | 
    1907 lines elided | 
    
      ↑ open up ↑ | 
  
4613 4616                           * previous size.
4614 4617                           */
4615 4618                          va.va_mask = AT_SIZE;
4616 4619                          error = nfsgetattr(vp, &va, cr);
4617 4620                          if (error || va.va_size == bfp->l_start)
4618 4621                                  return (error);
4619 4622                          va.va_mask = AT_SIZE;
4620 4623                          va.va_size = bfp->l_start;
4621 4624                          error = nfssetattr(vp, &va, 0, cr);
4622 4625  
4623      -                        if (error == 0 && bfp->l_start == 0)
4624      -                                vnevent_truncate(vp, ct);
     4626 +                        if (error == 0) {
     4627 +                                if (bfp->l_start == 0) {
     4628 +                                        vnevent_truncate(vp, ct);
     4629 +                                } else {
     4630 +                                        vnevent_resize(vp, ct);
     4631 +                                }
     4632 +                        }
4625 4633                  } else
4626 4634                          error = EINVAL;
4627 4635          }
4628 4636  
4629 4637          return (error);
4630 4638  }
4631 4639  
4632 4640  /* ARGSUSED */
4633 4641  static int
4634 4642  nfs_realvp(vnode_t *vp, vnode_t **vpp, caller_context_t *ct)
4635 4643  {
4636 4644  
4637 4645          return (EINVAL);
4638 4646  }
4639 4647  
4640 4648  /*
4641 4649   * Setup and add an address space callback to do the work of the delmap call.
4642 4650   * The callback will (and must be) deleted in the actual callback function.
4643 4651   *
4644 4652   * This is done in order to take care of the problem that we have with holding
4645 4653   * the address space's a_lock for a long period of time (e.g. if the NFS server
4646 4654   * is down).  Callbacks will be executed in the address space code while the
4647 4655   * a_lock is not held.  Holding the address space's a_lock causes things such
4648 4656   * as ps and fork to hang because they are trying to acquire this lock as well.
4649 4657   */
4650 4658  /* ARGSUSED */
4651 4659  static int
4652 4660  nfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr,
4653 4661          size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr,
4654 4662          caller_context_t *ct)
4655 4663  {
4656 4664          int                     caller_found;
4657 4665          int                     error;
4658 4666          rnode_t                 *rp;
4659 4667          nfs_delmap_args_t       *dmapp;
4660 4668          nfs_delmapcall_t        *delmap_call;
4661 4669  
4662 4670          if (vp->v_flag & VNOMAP)
4663 4671                  return (ENOSYS);
4664 4672          /*
4665 4673           * A process may not change zones if it has NFS pages mmap'ed
4666 4674           * in, so we can't legitimately get here from the wrong zone.
4667 4675           */
4668 4676          ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
4669 4677  
4670 4678          rp = VTOR(vp);
4671 4679  
4672 4680          /*
4673 4681           * The way that the address space of this process deletes its mapping
4674 4682           * of this file is via the following call chains:
4675 4683           * - as_free()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs_delmap()
4676 4684           * - as_unmap()->SEGOP_UNMAP()/segvn_unmap()->VOP_DELMAP()/nfs_delmap()
4677 4685           *
4678 4686           * With the use of address space callbacks we are allowed to drop the
4679 4687           * address space lock, a_lock, while executing the NFS operations that
4680 4688           * need to go over the wire.  Returning EAGAIN to the caller of this
4681 4689           * function is what drives the execution of the callback that we add
4682 4690           * below.  The callback will be executed by the address space code
4683 4691           * after dropping the a_lock.  When the callback is finished, since
4684 4692           * we dropped the a_lock, it must be re-acquired and segvn_unmap()
4685 4693           * is called again on the same segment to finish the rest of the work
4686 4694           * that needs to happen during unmapping.
4687 4695           *
4688 4696           * This action of calling back into the segment driver causes
4689 4697           * nfs_delmap() to get called again, but since the callback was
4690 4698           * already executed at this point, it already did the work and there
4691 4699           * is nothing left for us to do.
4692 4700           *
4693 4701           * To Summarize:
4694 4702           * - The first time nfs_delmap is called by the current thread is when
4695 4703           * we add the caller associated with this delmap to the delmap caller
4696 4704           * list, add the callback, and return EAGAIN.
4697 4705           * - The second time in this call chain when nfs_delmap is called we
4698 4706           * will find this caller in the delmap caller list and realize there
4699 4707           * is no more work to do thus removing this caller from the list and
4700 4708           * returning the error that was set in the callback execution.
4701 4709           */
4702 4710          caller_found = nfs_find_and_delete_delmapcall(rp, &error);
4703 4711          if (caller_found) {
4704 4712                  /*
4705 4713                   * 'error' is from the actual delmap operations.  To avoid
4706 4714                   * hangs, we need to handle the return of EAGAIN differently
4707 4715                   * since this is what drives the callback execution.
4708 4716                   * In this case, we don't want to return EAGAIN and do the
4709 4717                   * callback execution because there are none to execute.
4710 4718                   */
4711 4719                  if (error == EAGAIN)
4712 4720                          return (0);
4713 4721                  else
4714 4722                          return (error);
4715 4723          }
4716 4724  
4717 4725          /* current caller was not in the list */
4718 4726          delmap_call = nfs_init_delmapcall();
4719 4727  
4720 4728          mutex_enter(&rp->r_statelock);
4721 4729          list_insert_tail(&rp->r_indelmap, delmap_call);
4722 4730          mutex_exit(&rp->r_statelock);
4723 4731  
4724 4732          dmapp = kmem_alloc(sizeof (nfs_delmap_args_t), KM_SLEEP);
4725 4733  
4726 4734          dmapp->vp = vp;
4727 4735          dmapp->off = off;
4728 4736          dmapp->addr = addr;
4729 4737          dmapp->len = len;
4730 4738          dmapp->prot = prot;
4731 4739          dmapp->maxprot = maxprot;
4732 4740          dmapp->flags = flags;
4733 4741          dmapp->cr = cr;
4734 4742          dmapp->caller = delmap_call;
4735 4743  
4736 4744          error = as_add_callback(as, nfs_delmap_callback, dmapp,
4737 4745              AS_UNMAP_EVENT, addr, len, KM_SLEEP);
4738 4746  
4739 4747          return (error ? error : EAGAIN);
4740 4748  }
4741 4749  
4742 4750  /*
4743 4751   * Remove some pages from an mmap'd vnode.  Just update the
4744 4752   * count of pages.  If doing close-to-open, then flush all
4745 4753   * of the pages associated with this file.  Otherwise, start
4746 4754   * an asynchronous page flush to write out any dirty pages.
4747 4755   * This will also associate a credential with the rnode which
4748 4756   * can be used to write the pages.
4749 4757   */
4750 4758  /* ARGSUSED */
4751 4759  static void
4752 4760  nfs_delmap_callback(struct as *as, void *arg, uint_t event)
4753 4761  {
4754 4762          int                     error;
4755 4763          rnode_t                 *rp;
4756 4764          mntinfo_t               *mi;
4757 4765          nfs_delmap_args_t       *dmapp = (nfs_delmap_args_t *)arg;
4758 4766  
4759 4767          rp = VTOR(dmapp->vp);
4760 4768          mi = VTOMI(dmapp->vp);
4761 4769  
4762 4770          atomic_add_long((ulong_t *)&rp->r_mapcnt, -btopr(dmapp->len));
4763 4771          ASSERT(rp->r_mapcnt >= 0);
4764 4772  
4765 4773          /*
4766 4774           * Initiate a page flush if there are pages, the file system
4767 4775           * was not mounted readonly, the segment was mapped shared, and
4768 4776           * the pages themselves were writeable.
4769 4777           */
4770 4778          if (vn_has_cached_data(dmapp->vp) && !vn_is_readonly(dmapp->vp) &&
4771 4779              dmapp->flags == MAP_SHARED && (dmapp->maxprot & PROT_WRITE)) {
4772 4780                  mutex_enter(&rp->r_statelock);
4773 4781                  rp->r_flags |= RDIRTY;
4774 4782                  mutex_exit(&rp->r_statelock);
4775 4783                  /*
4776 4784                   * If this is a cross-zone access a sync putpage won't work, so
4777 4785                   * the best we can do is try an async putpage.  That seems
4778 4786                   * better than something more draconian such as discarding the
4779 4787                   * dirty pages.
4780 4788                   */
4781 4789                  if ((mi->mi_flags & MI_NOCTO) ||
4782 4790                      nfs_zone() != mi->mi_zone)
4783 4791                          error = nfs_putpage(dmapp->vp, dmapp->off, dmapp->len,
4784 4792                              B_ASYNC, dmapp->cr, NULL);
4785 4793                  else
4786 4794                          error = nfs_putpage(dmapp->vp, dmapp->off, dmapp->len,
4787 4795                              0, dmapp->cr, NULL);
4788 4796                  if (!error) {
4789 4797                          mutex_enter(&rp->r_statelock);
4790 4798                          error = rp->r_error;
4791 4799                          rp->r_error = 0;
4792 4800                          mutex_exit(&rp->r_statelock);
4793 4801                  }
4794 4802          } else
4795 4803                  error = 0;
4796 4804  
4797 4805          if ((rp->r_flags & RDIRECTIO) || (mi->mi_flags & MI_DIRECTIO))
4798 4806                  (void) nfs_putpage(dmapp->vp, dmapp->off, dmapp->len,
4799 4807                      B_INVAL, dmapp->cr, NULL);
4800 4808  
4801 4809          dmapp->caller->error = error;
4802 4810          (void) as_delete_callback(as, arg);
4803 4811          kmem_free(dmapp, sizeof (nfs_delmap_args_t));
4804 4812  }
4805 4813  
4806 4814  /* ARGSUSED */
4807 4815  static int
4808 4816  nfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
4809 4817          caller_context_t *ct)
4810 4818  {
4811 4819          int error = 0;
4812 4820  
4813 4821          if (nfs_zone() != VTOMI(vp)->mi_zone)
4814 4822                  return (EIO);
4815 4823          /*
4816 4824           * This looks a little weird because it's written in a general
4817 4825           * manner but we make little use of cases.  If cntl() ever gets
4818 4826           * widely used, the outer switch will make more sense.
4819 4827           */
4820 4828  
4821 4829          switch (cmd) {
4822 4830  
4823 4831          /*
4824 4832           * Large file spec - need to base answer new query with
4825 4833           * hardcoded constant based on the protocol.
4826 4834           */
4827 4835          case _PC_FILESIZEBITS:
4828 4836                  *valp = 32;
4829 4837                  return (0);
4830 4838  
4831 4839          case _PC_LINK_MAX:
4832 4840          case _PC_NAME_MAX:
4833 4841          case _PC_PATH_MAX:
4834 4842          case _PC_SYMLINK_MAX:
4835 4843          case _PC_CHOWN_RESTRICTED:
4836 4844          case _PC_NO_TRUNC: {
4837 4845                  mntinfo_t *mi;
4838 4846                  struct pathcnf *pc;
4839 4847  
4840 4848                  if ((mi = VTOMI(vp)) == NULL || (pc = mi->mi_pathconf) == NULL)
4841 4849                          return (EINVAL);
4842 4850                  error = _PC_ISSET(cmd, pc->pc_mask);    /* error or bool */
4843 4851                  switch (cmd) {
4844 4852                  case _PC_LINK_MAX:
4845 4853                          *valp = pc->pc_link_max;
4846 4854                          break;
4847 4855                  case _PC_NAME_MAX:
4848 4856                          *valp = pc->pc_name_max;
4849 4857                          break;
4850 4858                  case _PC_PATH_MAX:
4851 4859                  case _PC_SYMLINK_MAX:
4852 4860                          *valp = pc->pc_path_max;
4853 4861                          break;
4854 4862                  case _PC_CHOWN_RESTRICTED:
4855 4863                          /*
4856 4864                           * if we got here, error is really a boolean which
4857 4865                           * indicates whether cmd is set or not.
4858 4866                           */
4859 4867                          *valp = error ? 1 : 0;  /* see above */
4860 4868                          error = 0;
4861 4869                          break;
4862 4870                  case _PC_NO_TRUNC:
4863 4871                          /*
4864 4872                           * if we got here, error is really a boolean which
4865 4873                           * indicates whether cmd is set or not.
4866 4874                           */
4867 4875                          *valp = error ? 1 : 0;  /* see above */
4868 4876                          error = 0;
4869 4877                          break;
4870 4878                  }
4871 4879                  return (error ? EINVAL : 0);
4872 4880                  }
4873 4881  
4874 4882          case _PC_XATTR_EXISTS:
4875 4883                  *valp = 0;
4876 4884                  if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
4877 4885                          vnode_t *avp;
4878 4886                          rnode_t *rp;
4879 4887                          mntinfo_t *mi = VTOMI(vp);
4880 4888  
4881 4889                          if (!(mi->mi_flags & MI_EXTATTR))
4882 4890                                  return (0);
4883 4891  
4884 4892                          rp = VTOR(vp);
4885 4893                          if (nfs_rw_enter_sig(&rp->r_rwlock, RW_READER,
4886 4894                              INTR(vp)))
4887 4895                                  return (EINTR);
4888 4896  
4889 4897                          error = nfslookup_dnlc(vp, XATTR_DIR_NAME, &avp, cr);
4890 4898                          if (error || avp == NULL)
4891 4899                                  error = acl_getxattrdir2(vp, &avp, 0, cr, 0);
4892 4900  
4893 4901                          nfs_rw_exit(&rp->r_rwlock);
4894 4902  
4895 4903                          if (error == 0 && avp != NULL) {
4896 4904                                  error = do_xattr_exists_check(avp, valp, cr);
4897 4905                                  VN_RELE(avp);
4898 4906                          }
4899 4907                  }
4900 4908                  return (error ? EINVAL : 0);
4901 4909  
4902 4910          case _PC_ACL_ENABLED:
4903 4911                  *valp = _ACL_ACLENT_ENABLED;
4904 4912                  return (0);
4905 4913  
4906 4914          default:
4907 4915                  return (EINVAL);
4908 4916          }
4909 4917  }
4910 4918  
4911 4919  /*
4912 4920   * Called by async thread to do synchronous pageio. Do the i/o, wait
4913 4921   * for it to complete, and cleanup the page list when done.
4914 4922   */
4915 4923  static int
4916 4924  nfs_sync_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
4917 4925          int flags, cred_t *cr)
4918 4926  {
4919 4927          int error;
4920 4928  
4921 4929          ASSERT(nfs_zone() == VTOMI(vp)->mi_zone);
4922 4930          error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
4923 4931          if (flags & B_READ)
4924 4932                  pvn_read_done(pp, (error ? B_ERROR : 0) | flags);
4925 4933          else
4926 4934                  pvn_write_done(pp, (error ? B_ERROR : 0) | flags);
4927 4935          return (error);
4928 4936  }
4929 4937  
4930 4938  /* ARGSUSED */
4931 4939  static int
4932 4940  nfs_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
4933 4941          int flags, cred_t *cr, caller_context_t *ct)
4934 4942  {
4935 4943          int error;
4936 4944          rnode_t *rp;
4937 4945  
4938 4946          if (pp == NULL)
4939 4947                  return (EINVAL);
4940 4948  
4941 4949          if (io_off > MAXOFF32_T)
4942 4950                  return (EFBIG);
4943 4951          if (nfs_zone() != VTOMI(vp)->mi_zone)
4944 4952                  return (EIO);
4945 4953          rp = VTOR(vp);
4946 4954          mutex_enter(&rp->r_statelock);
4947 4955          rp->r_count++;
4948 4956          mutex_exit(&rp->r_statelock);
4949 4957  
4950 4958          if (flags & B_ASYNC) {
4951 4959                  error = nfs_async_pageio(vp, pp, io_off, io_len, flags, cr,
4952 4960                      nfs_sync_pageio);
4953 4961          } else
4954 4962                  error = nfs_rdwrlbn(vp, pp, io_off, io_len, flags, cr);
4955 4963          mutex_enter(&rp->r_statelock);
4956 4964          rp->r_count--;
4957 4965          cv_broadcast(&rp->r_cv);
4958 4966          mutex_exit(&rp->r_statelock);
4959 4967          return (error);
4960 4968  }
4961 4969  
4962 4970  /* ARGSUSED */
4963 4971  static int
4964 4972  nfs_setsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr,
4965 4973          caller_context_t *ct)
4966 4974  {
4967 4975          int error;
4968 4976          mntinfo_t *mi;
4969 4977  
4970 4978          mi = VTOMI(vp);
4971 4979  
4972 4980          if (nfs_zone() != mi->mi_zone)
4973 4981                  return (EIO);
4974 4982          if (mi->mi_flags & MI_ACL) {
4975 4983                  error = acl_setacl2(vp, vsecattr, flag, cr);
4976 4984                  if (mi->mi_flags & MI_ACL)
4977 4985                          return (error);
4978 4986          }
4979 4987  
4980 4988          return (ENOSYS);
4981 4989  }
4982 4990  
4983 4991  /* ARGSUSED */
4984 4992  static int
4985 4993  nfs_getsecattr(vnode_t *vp, vsecattr_t *vsecattr, int flag, cred_t *cr,
4986 4994          caller_context_t *ct)
4987 4995  {
4988 4996          int error;
4989 4997          mntinfo_t *mi;
4990 4998  
4991 4999          mi = VTOMI(vp);
4992 5000  
4993 5001          if (nfs_zone() != mi->mi_zone)
4994 5002                  return (EIO);
4995 5003          if (mi->mi_flags & MI_ACL) {
4996 5004                  error = acl_getacl2(vp, vsecattr, flag, cr);
4997 5005                  if (mi->mi_flags & MI_ACL)
4998 5006                          return (error);
4999 5007          }
5000 5008  
5001 5009          return (fs_fab_acl(vp, vsecattr, flag, cr, ct));
5002 5010  }
5003 5011  
5004 5012  /* ARGSUSED */
5005 5013  static int
5006 5014  nfs_shrlock(vnode_t *vp, int cmd, struct shrlock *shr, int flag, cred_t *cr,
5007 5015          caller_context_t *ct)
5008 5016  {
5009 5017          int error;
5010 5018          struct shrlock nshr;
5011 5019          struct nfs_owner nfs_owner;
5012 5020          netobj lm_fh;
5013 5021  
5014 5022          if (nfs_zone() != VTOMI(vp)->mi_zone)
5015 5023                  return (EIO);
5016 5024  
5017 5025          /*
5018 5026           * check for valid cmd parameter
5019 5027           */
5020 5028          if (cmd != F_SHARE && cmd != F_UNSHARE && cmd != F_HASREMOTELOCKS)
5021 5029                  return (EINVAL);
5022 5030  
5023 5031          /*
5024 5032           * Check access permissions
5025 5033           */
5026 5034          if (cmd == F_SHARE &&
5027 5035              (((shr->s_access & F_RDACC) && !(flag & FREAD)) ||
5028 5036              ((shr->s_access & F_WRACC) && !(flag & FWRITE))))
5029 5037                  return (EBADF);
5030 5038  
5031 5039          /*
5032 5040           * If the filesystem is mounted using local locking, pass the
5033 5041           * request off to the local share code.
5034 5042           */
5035 5043          if (VTOMI(vp)->mi_flags & MI_LLOCK)
5036 5044                  return (fs_shrlock(vp, cmd, shr, flag, cr, ct));
5037 5045  
5038 5046          switch (cmd) {
5039 5047          case F_SHARE:
5040 5048          case F_UNSHARE:
5041 5049                  lm_fh.n_len = sizeof (fhandle_t);
5042 5050                  lm_fh.n_bytes = (char *)VTOFH(vp);
5043 5051  
5044 5052                  /*
5045 5053                   * If passed an owner that is too large to fit in an
5046 5054                   * nfs_owner it is likely a recursive call from the
5047 5055                   * lock manager client and pass it straight through.  If
5048 5056                   * it is not a nfs_owner then simply return an error.
5049 5057                   */
5050 5058                  if (shr->s_own_len > sizeof (nfs_owner.lowner)) {
5051 5059                          if (((struct nfs_owner *)shr->s_owner)->magic !=
5052 5060                              NFS_OWNER_MAGIC)
5053 5061                                  return (EINVAL);
5054 5062  
5055 5063                          if (error = lm_shrlock(vp, cmd, shr, flag, &lm_fh)) {
5056 5064                                  error = set_errno(error);
5057 5065                          }
5058 5066                          return (error);
5059 5067                  }
5060 5068                  /*
5061 5069                   * Remote share reservations owner is a combination of
5062 5070                   * a magic number, hostname, and the local owner
5063 5071                   */
5064 5072                  bzero(&nfs_owner, sizeof (nfs_owner));
5065 5073                  nfs_owner.magic = NFS_OWNER_MAGIC;
5066 5074                  (void) strncpy(nfs_owner.hname, uts_nodename(),
5067 5075                      sizeof (nfs_owner.hname));
5068 5076                  bcopy(shr->s_owner, nfs_owner.lowner, shr->s_own_len);
5069 5077                  nshr.s_access = shr->s_access;
5070 5078                  nshr.s_deny = shr->s_deny;
5071 5079                  nshr.s_sysid = 0;
5072 5080                  nshr.s_pid = ttoproc(curthread)->p_pid;
5073 5081                  nshr.s_own_len = sizeof (nfs_owner);
5074 5082                  nshr.s_owner = (caddr_t)&nfs_owner;
5075 5083  
5076 5084                  if (error = lm_shrlock(vp, cmd, &nshr, flag, &lm_fh)) {
5077 5085                          error = set_errno(error);
5078 5086                  }
5079 5087  
5080 5088                  break;
5081 5089  
5082 5090          case F_HASREMOTELOCKS:
5083 5091                  /*
5084 5092                   * NFS client can't store remote locks itself
5085 5093                   */
5086 5094                  shr->s_access = 0;
5087 5095                  error = 0;
5088 5096                  break;
5089 5097  
5090 5098          default:
5091 5099                  error = EINVAL;
5092 5100                  break;
5093 5101          }
5094 5102  
5095 5103          return (error);
5096 5104  }
  
    | 
      ↓ open down ↓ | 
    462 lines elided | 
    
      ↑ open up ↑ | 
  
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX