fs/xfs/xfs_vnodeops.c

   1 /*
   2  * Copyright (c) 2000-2004 Silicon Graphics, Inc.  All Rights Reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it would be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  11  *
  12  * Further, this software is distributed without any warranty that it is
  13  * free of the rightful claim of any third person regarding infringement
  14  * or the like.  Any license provided herein, whether implied or
  15  * otherwise, applies only to this software file.  Patent licenses, if
  16  * any, provided herein do not apply to combinations of this program with
  17  * other software, or any other product whatsoever.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, write the Free Software Foundation, Inc., 59
  21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
  22  *
  23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
  24  * Mountain View, CA  94043, or:
  25  *
  26  * http://www.sgi.com
  27  *
  28  * For further information regarding this notice, see:
  29  *
  30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
  31  */
  32
  33 #include "xfs.h"
  34 #include "xfs_macros.h"
  35 #include "xfs_types.h"
  36 #include "xfs_inum.h"
  37 #include "xfs_log.h"
  38 #include "xfs_trans.h"
  39 #include "xfs_sb.h"
  40 #include "xfs_ag.h"
  41 #include "xfs_dir.h"
  42 #include "xfs_dir2.h"
  43 #include "xfs_dmapi.h"
  44 #include "xfs_mount.h"
  45 #include "xfs_alloc_btree.h"
  46 #include "xfs_bmap_btree.h"
  47 #include "xfs_ialloc_btree.h"
  48 #include "xfs_itable.h"
  49 #include "xfs_btree.h"
  50 #include "xfs_ialloc.h"
  51 #include "xfs_alloc.h"
  52 #include "xfs_attr_sf.h"
  53 #include "xfs_dir_sf.h"
  54 #include "xfs_dir2_sf.h"
  55 #include "xfs_dinode.h"
  56 #include "xfs_inode_item.h"
  57 #include "xfs_inode.h"
  58 #include "xfs_bmap.h"
  59 #include "xfs_da_btree.h"
  60 #include "xfs_attr.h"
  61 #include "xfs_rw.h"
  62 #include "xfs_refcache.h"
  63 #include "xfs_error.h"
  64 #include "xfs_bit.h"
  65 #include "xfs_rtalloc.h"
  66 #include "xfs_quota.h"
  67 #include "xfs_utils.h"
  68 #include "xfs_trans_space.h"
  69 #include "xfs_dir_leaf.h"
  70 #include "xfs_mac.h"
  71 #include "xfs_log_priv.h"
  72
  73
  74 /*
  75  * The maximum pathlen is 1024 bytes. Since the minimum file system
  76  * blocksize is 512 bytes, we can get a max of 2 extents back from
  77  * bmapi.
  78  */
  79 #define SYMLINK_MAPS 2
  80
  81 /*
  82  * For xfs, we check that the file isn't too big to be opened by this kernel.
  83  * No other open action is required for regular files.  Devices are handled
  84  * through the specfs file system, pipes through fifofs.  Device and
  85  * fifo vnodes are "wrapped" by specfs and fifofs vnodes, respectively,
  86  * when a new vnode is first looked up or created.
  87  */
  88 STATIC int
  89 xfs_open(
  90         bhv_desc_t      *bdp,
  91         cred_t          *credp)
  92 {
  93         int             mode;
  94         vnode_t         *vp;
  95         xfs_inode_t     *ip;
  96
  97         vp = BHV_TO_VNODE(bdp);
  98         ip = XFS_BHVTOI(bdp);
  99
 100         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 101                 return XFS_ERROR(EIO);
 102
 103         /*
 104          * If it's a directory with any blocks, read-ahead block 0
 105          * as we're almost certain to have the next operation be a read there.
 106          */
 107         if (vp->v_type == VDIR && ip->i_d.di_nextents > 0) {
 108                 mode = xfs_ilock_map_shared(ip);
 109                 if (ip->i_d.di_nextents > 0)
 110                         (void)xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
 111                 xfs_iunlock(ip, mode);
 112         }
 113         return 0;
 114 }
 115
 116
 117 /*
 118  * xfs_getattr
 119  */
 120 STATIC int
 121 xfs_getattr(
 122         bhv_desc_t      *bdp,
 123         vattr_t         *vap,
 124         int             flags,
 125         cred_t          *credp)
 126 {
 127         xfs_inode_t     *ip;
 128         xfs_mount_t     *mp;
 129         vnode_t         *vp;
 130
 131         vp  = BHV_TO_VNODE(bdp);
 132         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
 133
 134         ip = XFS_BHVTOI(bdp);
 135         mp = ip->i_mount;
 136
 137         if (XFS_FORCED_SHUTDOWN(mp))
 138                 return XFS_ERROR(EIO);
 139
 140         if (!(flags & ATTR_LAZY))
 141                 xfs_ilock(ip, XFS_ILOCK_SHARED);
 142
 143         vap->va_size = ip->i_d.di_size;
 144         if (vap->va_mask == XFS_AT_SIZE)
 145                 goto all_done;
 146
 147         vap->va_nblocks =
 148                 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
 149         vap->va_nodeid = ip->i_ino;
 150 #if XFS_BIG_INUMS
 151         vap->va_nodeid += mp->m_inoadd;
 152 #endif
 153         vap->va_nlink = ip->i_d.di_nlink;
 154
 155         /*
 156          * Quick exit for non-stat callers
 157          */
 158         if ((vap->va_mask &
 159             ~(XFS_AT_SIZE|XFS_AT_FSID|XFS_AT_NODEID|
 160               XFS_AT_NLINK|XFS_AT_BLKSIZE)) == 0)
 161                 goto all_done;
 162
 163         /*
 164          * Copy from in-core inode.
 165          */
 166         vap->va_type = vp->v_type;
 167         vap->va_mode = ip->i_d.di_mode & MODEMASK;
 168         vap->va_uid = ip->i_d.di_uid;
 169         vap->va_gid = ip->i_d.di_gid;
 170         vap->va_xid = ip->i_d.di_xid;
 171         vap->va_projid = ip->i_d.di_projid;
 172
 173         /*
 174          * Check vnode type block/char vs. everything else.
 175          * Do it with bitmask because that's faster than looking
 176          * for multiple values individually.
 177          */
 178         if (((1 << vp->v_type) & ((1<<VBLK) | (1<<VCHR))) == 0) {
 179                 vap->va_rdev = 0;
 180
 181                 if (!(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
 182
 183 #if 0
 184                         /* Large block sizes confuse various
 185                          * user space programs, so letting the
 186                          * stripe size through is not a good
 187                          * idea for now.
 188                          */
 189                         vap->va_blocksize = mp->m_swidth ?
 190                                 /*
 191                                  * If the underlying volume is a stripe, then
 192                                  * return the stripe width in bytes as the
 193                                  * recommended I/O size.
 194                                  */
 195                                 (mp->m_swidth << mp->m_sb.sb_blocklog) :
 196                                 /*
 197                                  * Return the largest of the preferred buffer
 198                                  * sizes since doing small I/Os into larger
 199                                  * buffers causes buffers to be decommissioned.
 200                                  * The value returned is in bytes.
 201                                  */
 202                                 (1 << (int)MAX(mp->m_readio_log,
 203                                                mp->m_writeio_log));
 204
 205 #else
 206                         vap->va_blocksize =
 207                                 /*
 208                                  * Return the largest of the preferred buffer
 209                                  * sizes since doing small I/Os into larger
 210                                  * buffers causes buffers to be decommissioned.
 211                                  * The value returned is in bytes.
 212                                  */
 213                                 1 << (int)MAX(mp->m_readio_log,
 214                                                mp->m_writeio_log);
 215 #endif
 216                 } else {
 217
 218                         /*
 219                          * If the file blocks are being allocated from a
 220                          * realtime partition, then return the inode's
 221                          * realtime extent size or the realtime volume's
 222                          * extent size.
 223                          */
 224                         vap->va_blocksize = ip->i_d.di_extsize ?
 225                                 (ip->i_d.di_extsize << mp->m_sb.sb_blocklog) :
 226                                 (mp->m_sb.sb_rextsize << mp->m_sb.sb_blocklog);
 227                 }
 228         } else {
 229                 vap->va_rdev = ip->i_df.if_u2.if_rdev;
 230                 vap->va_blocksize = BLKDEV_IOSIZE;
 231         }
 232
 233         vap->va_atime.tv_sec = ip->i_d.di_atime.t_sec;
 234         vap->va_atime.tv_nsec = ip->i_d.di_atime.t_nsec;
 235         vap->va_mtime.tv_sec = ip->i_d.di_mtime.t_sec;
 236         vap->va_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec;
 237         vap->va_ctime.tv_sec = ip->i_d.di_ctime.t_sec;
 238         vap->va_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec;
 239
 240         /*
 241          * Exit for stat callers.  See if any of the rest of the fields
 242          * to be filled in are needed.
 243          */
 244         if ((vap->va_mask &
 245              (XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
 246               XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
 247                 goto all_done;
 248
 249         /*
 250          * Convert di_flags to xflags.
 251          */
 252         vap->va_xflags = xfs_ip2xflags(ip);
 253
 254         /*
 255          * Exit for inode revalidate.  See if any of the rest of
 256          * the fields to be filled in are needed.
 257          */
 258         if ((vap->va_mask &
 259              (XFS_AT_EXTSIZE|XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|
 260               XFS_AT_GENCOUNT|XFS_AT_VCODE)) == 0)
 261                 goto all_done;
 262
 263         vap->va_extsize = ip->i_d.di_extsize << mp->m_sb.sb_blocklog;
 264         vap->va_nextents =
 265                 (ip->i_df.if_flags & XFS_IFEXTENTS) ?
 266                         ip->i_df.if_bytes / sizeof(xfs_bmbt_rec_t) :
 267                         ip->i_d.di_nextents;
 268         if (ip->i_afp)
 269                 vap->va_anextents =
 270                         (ip->i_afp->if_flags & XFS_IFEXTENTS) ?
 271                                 ip->i_afp->if_bytes / sizeof(xfs_bmbt_rec_t) :
 272                                  ip->i_d.di_anextents;
 273         else
 274                 vap->va_anextents = 0;
 275         vap->va_gen = ip->i_d.di_gen;
 276
 277  all_done:
 278         if (!(flags & ATTR_LAZY))
 279                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
 280         return 0;
 281 }
 282
 283
 284 /*
 285  * xfs_setattr
 286  */
 287 int
 288 xfs_setattr(
 289         bhv_desc_t              *bdp,
 290         vattr_t                 *vap,
 291         int                     flags,
 292         cred_t                  *credp)
 293 {
 294         xfs_inode_t             *ip;
 295         xfs_trans_t             *tp;
 296         xfs_mount_t             *mp;
 297         int                     mask;
 298         int                     code;
 299         uint                    lock_flags;
 300         uint                    commit_flags=0;
 301         uid_t                   uid=0, iuid=0;
 302         gid_t                   gid=0, igid=0;
 303         xid_t                   xid=0, ixid=0;
 304         int                     timeflags = 0;
 305         vnode_t                 *vp;
 306         xfs_prid_t              projid=0, iprojid=0;
 307         int                     mandlock_before, mandlock_after;
 308         struct xfs_dquot        *udqp, *gdqp, *olddquot1, *olddquot2;
 309         int                     file_owner;
 310         int                     need_iolock = 1;
 311
 312         vp = BHV_TO_VNODE(bdp);
 313         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
 314
 315         if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
 316                 return XFS_ERROR(EROFS);
 317
 318         /*
 319          * Cannot set certain attributes.
 320          */
 321         mask = vap->va_mask;
 322         if (mask & XFS_AT_NOSET) {
 323                 return XFS_ERROR(EINVAL);
 324         }
 325
 326         ip = XFS_BHVTOI(bdp);
 327         mp = ip->i_mount;
 328
 329         if (XFS_FORCED_SHUTDOWN(mp))
 330                 return XFS_ERROR(EIO);
 331
 332         /*
 333          * Timestamps do not need to be logged and hence do not
 334          * need to be done within a transaction.
 335          */
 336         if (mask & XFS_AT_UPDTIMES) {
 337                 ASSERT((mask & ~XFS_AT_UPDTIMES) == 0);
 338                 timeflags = ((mask & XFS_AT_UPDATIME) ? XFS_ICHGTIME_ACC : 0) |
 339                             ((mask & XFS_AT_UPDCTIME) ? XFS_ICHGTIME_CHG : 0) |
 340                             ((mask & XFS_AT_UPDMTIME) ? XFS_ICHGTIME_MOD : 0);
 341                 xfs_ichgtime(ip, timeflags);
 342                 return 0;
 343         }
 344
 345         olddquot1 = olddquot2 = NULL;
 346         udqp = gdqp = NULL;
 347
 348         /*
 349          * If disk quotas is on, we make sure that the dquots do exist on disk,
 350          * before we start any other transactions. Trying to do this later
 351          * is messy. We don't care to take a readlock to look at the ids
 352          * in inode here, because we can't hold it across the trans_reserve.
 353          * If the IDs do change before we take the ilock, we're covered
 354          * because the i_*dquot fields will get updated anyway.
 355          */
 356         if (XFS_IS_QUOTA_ON(mp) && (mask & (XFS_AT_UID|XFS_AT_GID))) {
 357                 uint    qflags = 0;
 358
 359                 /* FIXME: handle xid? */
 360                 if (mask & XFS_AT_UID) {
 361                         uid = vap->va_uid;
 362                         qflags |= XFS_QMOPT_UQUOTA;
 363                 } else {
 364                         uid = ip->i_d.di_uid;
 365                 }
 366                 if (mask & XFS_AT_GID) {
 367                         gid = vap->va_gid;
 368                         qflags |= XFS_QMOPT_GQUOTA;
 369                 }  else {
 370                         gid = ip->i_d.di_gid;
 371                 }
 372                 /*
 373                  * We take a reference when we initialize udqp and gdqp,
 374                  * so it is important that we never blindly double trip on
 375                  * the same variable. See xfs_create() for an example.
 376                  */
 377                 ASSERT(udqp == NULL);
 378                 ASSERT(gdqp == NULL);
 379                 code = XFS_QM_DQVOPALLOC(mp, ip, uid,gid, qflags, &udqp, &gdqp);
 380                 if (code)
 381                         return (code);
 382         }
 383
 384         /*
 385          * For the other attributes, we acquire the inode lock and
 386          * first do an error checking pass.
 387          */
 388         tp = NULL;
 389         lock_flags = XFS_ILOCK_EXCL;
 390         ASSERT(flags & ATTR_NOLOCK ? flags & ATTR_DMI : 1);
 391         if (flags & ATTR_NOLOCK)
 392                 need_iolock = 0;
 393         if (!(mask & XFS_AT_SIZE)) {
 394                 if ((mask != (XFS_AT_CTIME|XFS_AT_ATIME|XFS_AT_MTIME)) ||
 395                     (mp->m_flags & XFS_MOUNT_WSYNC)) {
 396                         tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
 397                         commit_flags = 0;
 398                         if ((code = xfs_trans_reserve(tp, 0,
 399                                                      XFS_ICHANGE_LOG_RES(mp), 0,
 400                                                      0, 0))) {
 401                                 lock_flags = 0;
 402                                 goto error_return;
 403                         }
 404                 }
 405         } else {
 406                 if (DM_EVENT_ENABLED (vp->v_vfsp, ip, DM_EVENT_TRUNCATE) &&
 407                     !(flags & ATTR_DMI)) {
 408                         int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
 409                         code = XFS_SEND_DATA(mp, DM_EVENT_TRUNCATE, vp,
 410                                 vap->va_size, 0, dmflags, NULL);
 411                         if (code) {
 412                                 lock_flags = 0;
 413                                 goto error_return;
 414                         }
 415                 }
 416                 if (need_iolock)
 417                         lock_flags |= XFS_IOLOCK_EXCL;
 418         }
 419
 420         xfs_ilock(ip, lock_flags);
 421
 422         /* boolean: are we the file owner? */
 423         file_owner = (current_fsuid(credp) == ip->i_d.di_uid);
 424
 425         /*
 426          * Change various properties of a file.
 427          * Only the owner or users with CAP_FOWNER
 428          * capability may do these things.
 429          */
 430         if (mask &
 431             (XFS_AT_MODE|XFS_AT_XFLAGS|XFS_AT_EXTSIZE|XFS_AT_UID|
 432              XFS_AT_GID|XFS_AT_PROJID)) {
 433                 /* FIXME: handle xid? */
 434
 435                 /*
 436                  * CAP_FOWNER overrides the following restrictions:
 437                  *
 438                  * The user ID of the calling process must be equal
 439                  * to the file owner ID, except in cases where the
 440                  * CAP_FSETID capability is applicable.
 441                  */
 442                 if (!file_owner && !capable(CAP_FOWNER)) {
 443                         code = XFS_ERROR(EPERM);
 444                         goto error_return;
 445                 }
 446
 447                 /*
 448                  * CAP_FSETID overrides the following restrictions:
 449                  *
 450                  * The effective user ID of the calling process shall match
 451                  * the file owner when setting the set-user-ID and
 452                  * set-group-ID bits on that file.
 453                  *
 454                  * The effective group ID or one of the supplementary group
 455                  * IDs of the calling process shall match the group owner of
 456                  * the file when setting the set-group-ID bit on that file
 457                  */
 458                 if (mask & XFS_AT_MODE) {
 459                         mode_t m = 0;
 460
 461                         if ((vap->va_mode & S_ISUID) && !file_owner)
 462                                 m |= S_ISUID;
 463                         if ((vap->va_mode & S_ISGID) &&
 464                             !in_group_p((gid_t)ip->i_d.di_gid))
 465                                 m |= S_ISGID;
 466 #if 0
 467                         /* Linux allows this, Irix doesn't. */
 468                         if ((vap->va_mode & S_ISVTX) && vp->v_type != VDIR)
 469                                 m |= S_ISVTX;
 470 #endif
 471                         if (m && !capable(CAP_FSETID))
 472                                 vap->va_mode &= ~m;
 473                 }
 474         }
 475
 476         /*
 477          * Change file ownership.  Must be the owner or privileged.
 478          * If the system was configured with the "restricted_chown"
 479          * option, the owner is not permitted to give away the file,
 480          * and can change the group id only to a group of which he
 481          * or she is a member.
 482          */
 483         if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_XID|XFS_AT_PROJID)) {
 484                 /*
 485                  * These IDs could have changed since we last looked at them.
 486                  * But, we're assured that if the ownership did change
 487                  * while we didn't have the inode locked, inode's dquot(s)
 488                  * would have changed also.
 489                  */
 490                 iuid = ip->i_d.di_uid;
 491                 igid = ip->i_d.di_gid;
 492                 ixid = ip->i_d.di_xid;
 493                 iprojid = ip->i_d.di_projid;
 494                 uid = (mask & XFS_AT_UID) ? vap->va_uid : iuid;
 495                 gid = (mask & XFS_AT_GID) ? vap->va_gid : igid;
 496                 xid = (mask & XFS_AT_XID) ? vap->va_xid : ixid;
 497                 projid = (mask & XFS_AT_PROJID) ? (xfs_prid_t)vap->va_projid :
 498                          iprojid;
 499
 500                 /*
 501                  * CAP_CHOWN overrides the following restrictions:
 502                  *
 503                  * If _POSIX_CHOWN_RESTRICTED is defined, this capability
 504                  * shall override the restriction that a process cannot
 505                  * change the user ID of a file it owns and the restriction
 506                  * that the group ID supplied to the chown() function
 507                  * shall be equal to either the group ID or one of the
 508                  * supplementary group IDs of the calling process.
 509                  *
 510                  * XXX: How does restricted_chown affect projid?
 511                  */
 512                 if (restricted_chown &&
 513                     (iuid != uid || (igid != gid &&
 514                                      !in_group_p((gid_t)gid))) &&
 515                     !capable(CAP_CHOWN)) {
 516                         code = XFS_ERROR(EPERM);
 517                         goto error_return;
 518                 }
 519                 /*
 520                  * Do a quota reservation only if uid or gid is actually
 521                  * going to change.
 522                  */
 523                 if ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
 524                     (XFS_IS_GQUOTA_ON(mp) && igid != gid)) {
 525                         /* FIXME: handle xid? */
 526                         ASSERT(tp);
 527                         code = XFS_QM_DQVOPCHOWNRESV(mp, tp, ip, udqp, gdqp,
 528                                                 capable(CAP_FOWNER) ?
 529                                                 XFS_QMOPT_FORCE_RES : 0);
 530                         if (code)       /* out of quota */
 531                                 goto error_return;
 532                 }
 533         }
 534
 535         /*
 536          * Truncate file.  Must have write permission and not be a directory.
 537          */
 538         if (mask & XFS_AT_SIZE) {
 539                 /* Short circuit the truncate case for zero length files */
 540                 if ((vap->va_size == 0) &&
 541                    (ip->i_d.di_size == 0) && (ip->i_d.di_nextents == 0)) {
 542                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 543                         lock_flags &= ~XFS_ILOCK_EXCL;
 544                         if (mask & XFS_AT_CTIME)
 545                                 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
 546                         code = 0;
 547                         goto error_return;
 548                 }
 549
 550                 if (vp->v_type == VDIR) {
 551                         code = XFS_ERROR(EISDIR);
 552                         goto error_return;
 553                 } else if (vp->v_type != VREG) {
 554                         code = XFS_ERROR(EINVAL);
 555                         goto error_return;
 556                 }
 557                 /*
 558                  * Make sure that the dquots are attached to the inode.
 559                  */
 560                 if ((code = XFS_QM_DQATTACH(mp, ip, XFS_QMOPT_ILOCKED)))
 561                         goto error_return;
 562         }
 563
 564         /*
 565          * Change file access or modified times.
 566          */
 567         if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
 568                 if (!file_owner) {
 569                         if ((flags & ATTR_UTIME) &&
 570                             !capable(CAP_FOWNER)) {
 571                                 code = XFS_ERROR(EPERM);
 572                                 goto error_return;
 573                         }
 574                 }
 575         }
 576
 577         /*
 578          * Change extent size or realtime flag.
 579          */
 580         if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
 581                 /*
 582                  * Can't change extent size if any extents are allocated.
 583                  */
 584                 if ((ip->i_d.di_nextents || ip->i_delayed_blks) &&
 585                     (mask & XFS_AT_EXTSIZE) &&
 586                     ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
 587                      vap->va_extsize) ) {
 588                         code = XFS_ERROR(EINVAL);       /* EFBIG? */
 589                         goto error_return;
 590                 }
 591
 592                 /*
 593                  * Can't set extent size unless the file is marked, or
 594                  * about to be marked as a realtime file.
 595                  *
 596                  * This check will be removed when fixed size extents
 597                  * with buffered data writes is implemented.
 598                  *
 599                  */
 600                 if ((mask & XFS_AT_EXTSIZE)                     &&
 601                     ((ip->i_d.di_extsize << mp->m_sb.sb_blocklog) !=
 602                      vap->va_extsize) &&
 603                     (!((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
 604                        ((mask & XFS_AT_XFLAGS) &&
 605                         (vap->va_xflags & XFS_XFLAG_REALTIME))))) {
 606                         code = XFS_ERROR(EINVAL);
 607                         goto error_return;
 608                 }
 609
 610                 /*
 611                  * Can't change realtime flag if any extents are allocated.
 612                  */
 613                 if (ip->i_d.di_nextents && (mask & XFS_AT_XFLAGS) &&
 614                     (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) !=
 615                     (vap->va_xflags & XFS_XFLAG_REALTIME)) {
 616                         code = XFS_ERROR(EINVAL);       /* EFBIG? */
 617                         goto error_return;
 618                 }
 619                 /*
 620                  * Extent size must be a multiple of the appropriate block
 621                  * size, if set at all.
 622                  */
 623                 if ((mask & XFS_AT_EXTSIZE) && vap->va_extsize != 0) {
 624                         xfs_extlen_t    size;
 625
 626                         if ((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ||
 627                             ((mask & XFS_AT_XFLAGS) &&
 628                             (vap->va_xflags & XFS_XFLAG_REALTIME))) {
 629                                 size = mp->m_sb.sb_rextsize <<
 630                                        mp->m_sb.sb_blocklog;
 631                         } else {
 632                                 size = mp->m_sb.sb_blocksize;
 633                         }
 634                         if (vap->va_extsize % size) {
 635                                 code = XFS_ERROR(EINVAL);
 636                                 goto error_return;
 637                         }
 638                 }
 639                 /*
 640                  * If realtime flag is set then must have realtime data.
 641                  */
 642                 if ((mask & XFS_AT_XFLAGS) &&
 643                     (vap->va_xflags & XFS_XFLAG_REALTIME)) {
 644                         if ((mp->m_sb.sb_rblocks == 0) ||
 645                             (mp->m_sb.sb_rextsize == 0) ||
 646                             (ip->i_d.di_extsize % mp->m_sb.sb_rextsize)) {
 647                                 code = XFS_ERROR(EINVAL);
 648                                 goto error_return;
 649                         }
 650                 }
 651
 652                 /*
 653                  * Can't modify an immutable/append-only file unless
 654                  * we have appropriate permission.
 655                  */
 656                 if ((mask & XFS_AT_XFLAGS) &&
 657                     (ip->i_d.di_flags &
 658                                 (XFS_DIFLAG_IMMUTABLE|XFS_DIFLAG_APPEND) ||
 659                      (vap->va_xflags &
 660                                 (XFS_XFLAG_IMMUTABLE | XFS_XFLAG_APPEND))) &&
 661                     !capable(CAP_LINUX_IMMUTABLE)) {
 662                         code = XFS_ERROR(EPERM);
 663                         goto error_return;
 664                 }
 665         }
 666
 667         /*
 668          * Now we can make the changes.  Before we join the inode
 669          * to the transaction, if XFS_AT_SIZE is set then take care of
 670          * the part of the truncation that must be done without the
 671          * inode lock.  This needs to be done before joining the inode
 672          * to the transaction, because the inode cannot be unlocked
 673          * once it is a part of the transaction.
 674          */
 675         if (mask & XFS_AT_SIZE) {
 676                 code = 0;
 677                 if (vap->va_size > ip->i_d.di_size)
 678                         code = xfs_igrow_start(ip, vap->va_size, credp);
 679                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 680                 if (!code)
 681                         code = xfs_itruncate_data(ip, vap->va_size);
 682                 if (code) {
 683                         ASSERT(tp == NULL);
 684                         lock_flags &= ~XFS_ILOCK_EXCL;
 685                         ASSERT(lock_flags == XFS_IOLOCK_EXCL);
 686                         goto error_return;
 687                 }
 688                 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
 689                 if ((code = xfs_trans_reserve(tp, 0,
 690                                              XFS_ITRUNCATE_LOG_RES(mp), 0,
 691                                              XFS_TRANS_PERM_LOG_RES,
 692                                              XFS_ITRUNCATE_LOG_COUNT))) {
 693                         xfs_trans_cancel(tp, 0);
 694                         if (need_iolock)
 695                                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 696                         return code;
 697                 }
 698                 commit_flags = XFS_TRANS_RELEASE_LOG_RES;
 699                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 700         }
 701
 702         if (tp) {
 703                 xfs_trans_ijoin(tp, ip, lock_flags);
 704                 xfs_trans_ihold(tp, ip);
 705         }
 706
 707         /* determine whether mandatory locking mode changes */
 708         mandlock_before = MANDLOCK(vp, ip->i_d.di_mode);
 709
 710         /*
 711          * Truncate file.  Must have write permission and not be a directory.
 712          */
 713         if (mask & XFS_AT_SIZE) {
 714                 if (vap->va_size > ip->i_d.di_size) {
 715                         xfs_igrow_finish(tp, ip, vap->va_size,
 716                             !(flags & ATTR_DMI));
 717                 } else if ((vap->va_size <= ip->i_d.di_size) ||
 718                            ((vap->va_size == 0) && ip->i_d.di_nextents)) {
 719                         /*
 720                          * signal a sync transaction unless
 721                          * we're truncating an already unlinked
 722                          * file on a wsync filesystem
 723                          */
 724                         code = xfs_itruncate_finish(&tp, ip,
 725                                             (xfs_fsize_t)vap->va_size,
 726                                             XFS_DATA_FORK,
 727                                             ((ip->i_d.di_nlink != 0 ||
 728                                               !(mp->m_flags & XFS_MOUNT_WSYNC))
 729                                              ? 1 : 0));
 730                         if (code) {
 731                                 goto abort_return;
 732                         }
 733                 }
 734                 /*
 735                  * Have to do this even if the file's size doesn't change.
 736                  */
 737                 timeflags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
 738         }
 739
 740         /*
 741          * Change file access modes.
 742          */
 743         if (mask & XFS_AT_MODE) {
 744                 ip->i_d.di_mode &= S_IFMT;
 745                 ip->i_d.di_mode |= vap->va_mode & ~S_IFMT;
 746
 747                 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
 748                 timeflags |= XFS_ICHGTIME_CHG;
 749         }
 750
 751         /*
 752          * Change file ownership.  Must be the owner or privileged.
 753          * If the system was configured with the "restricted_chown"
 754          * option, the owner is not permitted to give away the file,
 755          * and can change the group id only to a group of which he
 756          * or she is a member.
 757          */
 758         if (mask & (XFS_AT_UID|XFS_AT_GID|XFS_AT_XID|XFS_AT_PROJID)) {
 759                 /*
 760                  * CAP_FSETID overrides the following restrictions:
 761                  *
 762                  * The set-user-ID and set-group-ID bits of a file will be
 763                  * cleared upon successful return from chown()
 764                  */
 765                 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
 766                     !capable(CAP_FSETID)) {
 767                         ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
 768                 }
 769
 770                 /*
 771                  * Change the ownerships and register quota modifications
 772                  * in the transaction.
 773                  */
 774                 if (ixid != xid) {
 775                         if (XFS_IS_GQUOTA_ON(mp)) {
 776                                 /* FIXME: handle xid quota? */
 777                         }
 778                         ip->i_d.di_xid = xid;
 779                 }
 780                 if (iuid != uid) {
 781                         if (XFS_IS_UQUOTA_ON(mp)) {
 782                                 ASSERT(mask & XFS_AT_UID);
 783                                 ASSERT(udqp);
 784                                 olddquot1 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
 785                                                         &ip->i_udquot, udqp);
 786                         }
 787                         ip->i_d.di_uid = uid;
 788                 }
 789                 if (igid != gid) {
 790                         if (XFS_IS_GQUOTA_ON(mp)) {
 791                                 ASSERT(mask & XFS_AT_GID);
 792                                 ASSERT(gdqp);
 793                                 olddquot2 = XFS_QM_DQVOPCHOWN(mp, tp, ip,
 794                                                         &ip->i_gdquot, gdqp);
 795                         }
 796                         ip->i_d.di_gid = gid;
 797                 }
 798                 if (iprojid != projid) {
 799                         ip->i_d.di_projid = projid;
 800                         /*
 801                          * We may have to rev the inode as well as
 802                          * the superblock version number since projids didn't
 803                          * exist before DINODE_VERSION_2 and SB_VERSION_NLINK.
 804                          */
 805                         if (ip->i_d.di_version == XFS_DINODE_VERSION_1)
 806                                 xfs_bump_ino_vers2(tp, ip);
 807                 }
 808
 809                 xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
 810                 timeflags |= XFS_ICHGTIME_CHG;
 811         }
 812
 813
 814         /*
 815          * Change file access or modified times.
 816          */
 817         if (mask & (XFS_AT_ATIME|XFS_AT_MTIME)) {
 818                 if (mask & XFS_AT_ATIME) {
 819                         ip->i_d.di_atime.t_sec = vap->va_atime.tv_sec;
 820                         ip->i_d.di_atime.t_nsec = vap->va_atime.tv_nsec;
 821                         ip->i_update_core = 1;
 822                         timeflags &= ~XFS_ICHGTIME_ACC;
 823                 }
 824                 if (mask & XFS_AT_MTIME) {
 825                         ip->i_d.di_mtime.t_sec = vap->va_mtime.tv_sec;
 826                         ip->i_d.di_mtime.t_nsec = vap->va_mtime.tv_nsec;
 827                         timeflags &= ~XFS_ICHGTIME_MOD;
 828                         timeflags |= XFS_ICHGTIME_CHG;
 829                 }
 830                 if (tp && (flags & ATTR_UTIME))
 831                         xfs_trans_log_inode (tp, ip, XFS_ILOG_CORE);
 832         }
 833
 834         /*
 835          * Change XFS-added attributes.
 836          */
 837         if (mask & (XFS_AT_EXTSIZE|XFS_AT_XFLAGS)) {
 838                 if (mask & XFS_AT_EXTSIZE) {
 839                         /*
 840                          * Converting bytes to fs blocks.
 841                          */
 842                         ip->i_d.di_extsize = vap->va_extsize >>
 843                                 mp->m_sb.sb_blocklog;
 844                 }
 845                 if (mask & XFS_AT_XFLAGS) {
 846                         uint    di_flags;
 847
 848                         /* can't set PREALLOC this way, just preserve it */
 849                         di_flags = (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC);
 850                         if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
 851                                 di_flags |= XFS_DIFLAG_IMMUTABLE;
 852                         if (vap->va_xflags & XFS_XFLAG_IUNLINK)
 853                                 di_flags |= XFS_DIFLAG_IUNLINK;
 854                         if (vap->va_xflags & XFS_XFLAG_BARRIER)
 855                                 di_flags |= XFS_DIFLAG_BARRIER;
 856                         if (vap->va_xflags & XFS_XFLAG_APPEND)
 857                                 di_flags |= XFS_DIFLAG_APPEND;
 858                         if (vap->va_xflags & XFS_XFLAG_SYNC)
 859                                 di_flags |= XFS_DIFLAG_SYNC;
 860                         if (vap->va_xflags & XFS_XFLAG_NOATIME)
 861                                 di_flags |= XFS_DIFLAG_NOATIME;
 862                         if (vap->va_xflags & XFS_XFLAG_NODUMP)
 863                                 di_flags |= XFS_DIFLAG_NODUMP;
 864                         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
 865                                 if (vap->va_xflags & XFS_XFLAG_RTINHERIT)
 866                                         di_flags |= XFS_DIFLAG_RTINHERIT;
 867                                 if (vap->va_xflags & XFS_XFLAG_NOSYMLINKS)
 868                                         di_flags |= XFS_DIFLAG_NOSYMLINKS;
 869                         } else {
 870                                 if (vap->va_xflags & XFS_XFLAG_REALTIME) {
 871                                         di_flags |= XFS_DIFLAG_REALTIME;
 872                                         ip->i_iocore.io_flags |= XFS_IOCORE_RT;
 873                                 } else {
 874                                         ip->i_iocore.io_flags &= ~XFS_IOCORE_RT;
 875                                 }
 876                         }
 877                         ip->i_d.di_flags = di_flags;
 878                 }
 879                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 880                 timeflags |= XFS_ICHGTIME_CHG;
 881         }
 882
 883         /*
 884          * Change file inode change time only if XFS_AT_CTIME set
 885          * AND we have been called by a DMI function.
 886          */
 887
 888         if ( (flags & ATTR_DMI) && (mask & XFS_AT_CTIME) ) {
 889                 ip->i_d.di_ctime.t_sec = vap->va_ctime.tv_sec;
 890                 ip->i_d.di_ctime.t_nsec = vap->va_ctime.tv_nsec;
 891                 ip->i_update_core = 1;
 892                 timeflags &= ~XFS_ICHGTIME_CHG;
 893         }
 894
 895         /*
 896          * Send out timestamp changes that need to be set to the
 897          * current time.  Not done when called by a DMI function.
 898          */
 899         if (timeflags && !(flags & ATTR_DMI))
 900                 xfs_ichgtime(ip, timeflags);
 901
 902         XFS_STATS_INC(xs_ig_attrchg);
 903
 904         /*
 905          * If this is a synchronous mount, make sure that the
 906          * transaction goes to disk before returning to the user.
 907          * This is slightly sub-optimal in that truncates require
 908          * two sync transactions instead of one for wsync filesytems.
 909          * One for the truncate and one for the timestamps since we
 910          * don't want to change the timestamps unless we're sure the
 911          * truncate worked.  Truncates are less than 1% of the laddis
 912          * mix so this probably isn't worth the trouble to optimize.
 913          */
 914         code = 0;
 915         if (tp) {
 916                 if (mp->m_flags & XFS_MOUNT_WSYNC)
 917                         xfs_trans_set_sync(tp);
 918
 919                 code = xfs_trans_commit(tp, commit_flags, NULL);
 920         }
 921
 922         /*
 923          * If the (regular) file's mandatory locking mode changed, then
 924          * notify the vnode.  We do this under the inode lock to prevent
 925          * racing calls to vop_vnode_change.
 926          */
 927         mandlock_after = MANDLOCK(vp, ip->i_d.di_mode);
 928         if (mandlock_before != mandlock_after) {
 929                 VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_ENF_LOCKING,
 930                                  mandlock_after);
 931         }
 932
 933         xfs_iunlock(ip, lock_flags);
 934
 935         /*
 936          * Release any dquot(s) the inode had kept before chown.
 937          */
 938         XFS_QM_DQRELE(mp, olddquot1);
 939         XFS_QM_DQRELE(mp, olddquot2);
 940         XFS_QM_DQRELE(mp, udqp);
 941         XFS_QM_DQRELE(mp, gdqp);
 942
 943         if (code) {
 944                 return code;
 945         }
 946
 947         if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_ATTRIBUTE) &&
 948             !(flags & ATTR_DMI)) {
 949                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_ATTRIBUTE, vp, DM_RIGHT_NULL,
 950                                         NULL, DM_RIGHT_NULL, NULL, NULL,
 951                                         0, 0, AT_DELAY_FLAG(flags));
 952         }
 953         return 0;
 954
 955  abort_return:
 956         commit_flags |= XFS_TRANS_ABORT;
 957         /* FALLTHROUGH */
 958  error_return:
 959         XFS_QM_DQRELE(mp, udqp);
 960         XFS_QM_DQRELE(mp, gdqp);
 961         if (tp) {
 962                 xfs_trans_cancel(tp, commit_flags);
 963         }
 964         if (lock_flags != 0) {
 965                 xfs_iunlock(ip, lock_flags);
 966         }
 967         return code;
 968 }
 969
 970
 971 /*
 972  * xfs_access
 973  * Null conversion from vnode mode bits to inode mode bits, as in efs.
 974  */
 975 STATIC int
 976 xfs_access(
 977         bhv_desc_t      *bdp,
 978         int             mode,
 979         cred_t          *credp)
 980 {
 981         xfs_inode_t     *ip;
 982         int             error;
 983
 984         vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
 985                                                (inst_t *)__return_address);
 986
 987         ip = XFS_BHVTOI(bdp);
 988         xfs_ilock(ip, XFS_ILOCK_SHARED);
 989         error = xfs_iaccess(ip, mode, credp);
 990         xfs_iunlock(ip, XFS_ILOCK_SHARED);
 991         return error;
 992 }
 993
 994
 995 /*
 996  * xfs_readlink
 997  *
 998  */
 999 STATIC int
1000 xfs_readlink(
1001         bhv_desc_t      *bdp,
1002         uio_t           *uiop,
1003         int             ioflags,
1004         cred_t          *credp)
1005 {
1006         xfs_inode_t     *ip;
1007         int             count;
1008         xfs_off_t       offset;
1009         int             pathlen;
1010         vnode_t         *vp;
1011         int             error = 0;
1012         xfs_mount_t     *mp;
1013         int             nmaps;
1014         xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1015         xfs_daddr_t     d;
1016         int             byte_cnt;
1017         int             n;
1018         xfs_buf_t       *bp;
1019
1020         vp = BHV_TO_VNODE(bdp);
1021         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1022
1023         ip = XFS_BHVTOI(bdp);
1024         mp = ip->i_mount;
1025
1026         if (XFS_FORCED_SHUTDOWN(mp))
1027                 return XFS_ERROR(EIO);
1028
1029         xfs_ilock(ip, XFS_ILOCK_SHARED);
1030
1031         ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFLNK);
1032
1033         offset = uiop->uio_offset;
1034         count = uiop->uio_resid;
1035
1036         if (offset < 0) {
1037                 error = XFS_ERROR(EINVAL);
1038                 goto error_return;
1039         }
1040         if (count <= 0) {
1041                 error = 0;
1042                 goto error_return;
1043         }
1044
1045         if (!(ioflags & IO_INVIS)) {
1046                 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
1047         }
1048
1049         /*
1050          * See if the symlink is stored inline.
1051          */
1052         pathlen = (int)ip->i_d.di_size;
1053
1054         if (ip->i_df.if_flags & XFS_IFINLINE) {
1055                 error = uio_read(ip->i_df.if_u1.if_data, pathlen, uiop);
1056         }
1057         else {
1058                 /*
1059                  * Symlink not inline.  Call bmap to get it in.
1060                  */
1061                 nmaps = SYMLINK_MAPS;
1062
1063                 error = xfs_bmapi(NULL, ip, 0, XFS_B_TO_FSB(mp, pathlen),
1064                                   0, NULL, 0, mval, &nmaps, NULL);
1065
1066                 if (error) {
1067                         goto error_return;
1068                 }
1069
1070                 for (n = 0; n < nmaps; n++) {
1071                         d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
1072                         byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
1073                         bp = xfs_buf_read(mp->m_ddev_targp, d,
1074                                       BTOBB(byte_cnt), 0);
1075                         error = XFS_BUF_GETERROR(bp);
1076                         if (error) {
1077                                 xfs_ioerror_alert("xfs_readlink",
1078                                           ip->i_mount, bp, XFS_BUF_ADDR(bp));
1079                                 xfs_buf_relse(bp);
1080                                 goto error_return;
1081                         }
1082                         if (pathlen < byte_cnt)
1083                                 byte_cnt = pathlen;
1084                         pathlen -= byte_cnt;
1085
1086                         error = uio_read(XFS_BUF_PTR(bp), byte_cnt, uiop);
1087                         xfs_buf_relse (bp);
1088                 }
1089
1090         }
1091
1092
1093 error_return:
1094
1095         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1096
1097         return error;
1098 }
1099
1100
1101 /*
1102  * xfs_fsync
1103  *
1104  * This is called to sync the inode and its data out to disk.
1105  * We need to hold the I/O lock while flushing the data, and
1106  * the inode lock while flushing the inode.  The inode lock CANNOT
1107  * be held while flushing the data, so acquire after we're done
1108  * with that.
1109  */
1110 STATIC int
1111 xfs_fsync(
1112         bhv_desc_t      *bdp,
1113         int             flag,
1114         cred_t          *credp,
1115         xfs_off_t       start,
1116         xfs_off_t       stop)
1117 {
1118         xfs_inode_t     *ip;
1119         xfs_trans_t     *tp;
1120         int             error;
1121
1122         vn_trace_entry(BHV_TO_VNODE(bdp),
1123                         __FUNCTION__, (inst_t *)__return_address);
1124
1125         ip = XFS_BHVTOI(bdp);
1126
1127         ASSERT(start >= 0 && stop >= -1);
1128
1129         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
1130                 return XFS_ERROR(EIO);
1131
1132         /*
1133          * We always need to make sure that the required inode state
1134          * is safe on disk.  The vnode might be clean but because
1135          * of committed transactions that haven't hit the disk yet.
1136          * Likewise, there could be unflushed non-transactional
1137          * changes to the inode core that have to go to disk.
1138          *
1139          * The following code depends on one assumption:  that
1140          * any transaction that changes an inode logs the core
1141          * because it has to change some field in the inode core
1142          * (typically nextents or nblocks).  That assumption
1143          * implies that any transactions against an inode will
1144          * catch any non-transactional updates.  If inode-altering
1145          * transactions exist that violate this assumption, the
1146          * code breaks.  Right now, it figures that if the involved
1147          * update_* field is clear and the inode is unpinned, the
1148          * inode is clean.  Either it's been flushed or it's been
1149          * committed and the commit has hit the disk unpinning the inode.
1150          * (Note that xfs_inode_item_format() called at commit clears
1151          * the update_* fields.)
1152          */
1153         xfs_ilock(ip, XFS_ILOCK_SHARED);
1154
1155         /* If we are flushing data then we care about update_size
1156          * being set, otherwise we care about update_core
1157          */
1158         if ((flag & FSYNC_DATA) ?
1159                         (ip->i_update_size == 0) :
1160                         (ip->i_update_core == 0)) {
1161                 /*
1162                  * Timestamps/size haven't changed since last inode
1163                  * flush or inode transaction commit.  That means
1164                  * either nothing got written or a transaction
1165                  * committed which caught the updates.  If the
1166                  * latter happened and the transaction hasn't
1167                  * hit the disk yet, the inode will be still
1168                  * be pinned.  If it is, force the log.
1169                  */
1170
1171                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1172
1173                 if (xfs_ipincount(ip)) {
1174                         xfs_log_force(ip->i_mount, (xfs_lsn_t)0,
1175                                       XFS_LOG_FORCE |
1176                                       ((flag & FSYNC_WAIT)
1177                                        ? XFS_LOG_SYNC : 0));
1178                 }
1179                 error = 0;
1180         } else  {
1181                 /*
1182                  * Kick off a transaction to log the inode
1183                  * core to get the updates.  Make it
1184                  * sync if FSYNC_WAIT is passed in (which
1185                  * is done by everybody but specfs).  The
1186                  * sync transaction will also force the log.
1187                  */
1188                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1189                 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_FSYNC_TS);
1190                 if ((error = xfs_trans_reserve(tp, 0,
1191                                 XFS_FSYNC_TS_LOG_RES(ip->i_mount),
1192                                 0, 0, 0)))  {
1193                         xfs_trans_cancel(tp, 0);
1194                         return error;
1195                 }
1196                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1197
1198                 /*
1199                  * Note - it's possible that we might have pushed
1200                  * ourselves out of the way during trans_reserve
1201                  * which would flush the inode.  But there's no
1202                  * guarantee that the inode buffer has actually
1203                  * gone out yet (it's delwri).  Plus the buffer
1204                  * could be pinned anyway if it's part of an
1205                  * inode in another recent transaction.  So we
1206                  * play it safe and fire off the transaction anyway.
1207                  */
1208                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1209                 xfs_trans_ihold(tp, ip);
1210                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1211                 if (flag & FSYNC_WAIT)
1212                         xfs_trans_set_sync(tp);
1213                 error = xfs_trans_commit(tp, 0, NULL);
1214
1215                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1216         }
1217         return error;
1218 }
1219
1220 /*
1221  * This is called by xfs_inactive to free any blocks beyond eof,
1222  * when the link count isn't zero.
1223  */
1224 STATIC int
1225 xfs_inactive_free_eofblocks(
1226         xfs_mount_t     *mp,
1227         xfs_inode_t     *ip)
1228 {
1229         xfs_trans_t     *tp;
1230         int             error;
1231         xfs_fileoff_t   end_fsb;
1232         xfs_fileoff_t   last_fsb;
1233         xfs_filblks_t   map_len;
1234         int             nimaps;
1235         xfs_bmbt_irec_t imap;
1236
1237         /*
1238          * Figure out if there are any blocks beyond the end
1239          * of the file.  If not, then there is nothing to do.
1240          */
1241         end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)ip->i_d.di_size));
1242         last_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
1243         map_len = last_fsb - end_fsb;
1244         if (map_len <= 0)
1245                 return (0);
1246
1247         nimaps = 1;
1248         xfs_ilock(ip, XFS_ILOCK_SHARED);
1249         error = xfs_bmapi(NULL, ip, end_fsb, map_len, 0,
1250                           NULL, 0, &imap, &nimaps, NULL);
1251         xfs_iunlock(ip, XFS_ILOCK_SHARED);
1252
1253         if (!error && (nimaps != 0) &&
1254             (imap.br_startblock != HOLESTARTBLOCK)) {
1255                 /*
1256                  * Attach the dquots to the inode up front.
1257                  */
1258                 if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1259                         return (error);
1260
1261                 /*
1262                  * There are blocks after the end of file.
1263                  * Free them up now by truncating the file to
1264                  * its current size.
1265                  */
1266                 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1267
1268                 /*
1269                  * Do the xfs_itruncate_start() call before
1270                  * reserving any log space because
1271                  * itruncate_start will call into the buffer
1272                  * cache and we can't
1273                  * do that within a transaction.
1274                  */
1275                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1276                 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE,
1277                                     ip->i_d.di_size);
1278
1279                 error = xfs_trans_reserve(tp, 0,
1280                                           XFS_ITRUNCATE_LOG_RES(mp),
1281                                           0, XFS_TRANS_PERM_LOG_RES,
1282                                           XFS_ITRUNCATE_LOG_COUNT);
1283                 if (error) {
1284                         ASSERT(XFS_FORCED_SHUTDOWN(mp));
1285                         xfs_trans_cancel(tp, 0);
1286                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1287                         return (error);
1288                 }
1289
1290                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1291                 xfs_trans_ijoin(tp, ip,
1292                                 XFS_IOLOCK_EXCL |
1293                                 XFS_ILOCK_EXCL);
1294                 xfs_trans_ihold(tp, ip);
1295
1296                 error = xfs_itruncate_finish(&tp, ip,
1297                                              ip->i_d.di_size,
1298                                              XFS_DATA_FORK,
1299                                              0);
1300                 /*
1301                  * If we get an error at this point we
1302                  * simply don't bother truncating the file.
1303                  */
1304                 if (error) {
1305                         xfs_trans_cancel(tp,
1306                                          (XFS_TRANS_RELEASE_LOG_RES |
1307                                           XFS_TRANS_ABORT));
1308                 } else {
1309                         error = xfs_trans_commit(tp,
1310                                                 XFS_TRANS_RELEASE_LOG_RES,
1311                                                 NULL);
1312                 }
1313                 xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1314         }
1315         return (error);
1316 }
1317
1318 /*
1319  * Free a symlink that has blocks associated with it.
1320  */
1321 STATIC int
1322 xfs_inactive_symlink_rmt(
1323         xfs_inode_t     *ip,
1324         xfs_trans_t     **tpp)
1325 {
1326         xfs_buf_t       *bp;
1327         int             committed;
1328         int             done;
1329         int             error;
1330         xfs_fsblock_t   first_block;
1331         xfs_bmap_free_t free_list;
1332         int             i;
1333         xfs_mount_t     *mp;
1334         xfs_bmbt_irec_t mval[SYMLINK_MAPS];
1335         int             nmaps;
1336         xfs_trans_t     *ntp;
1337         int             size;
1338         xfs_trans_t     *tp;
1339
1340         tp = *tpp;
1341         mp = ip->i_mount;
1342         ASSERT(ip->i_d.di_size > XFS_IFORK_DSIZE(ip));
1343         /*
1344          * We're freeing a symlink that has some
1345          * blocks allocated to it.  Free the
1346          * blocks here.  We know that we've got
1347          * either 1 or 2 extents and that we can
1348          * free them all in one bunmapi call.
1349          */
1350         ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2);
1351         if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1352                         XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1353                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1354                 xfs_trans_cancel(tp, 0);
1355                 *tpp = NULL;
1356                 return error;
1357         }
1358         /*
1359          * Lock the inode, fix the size, and join it to the transaction.
1360          * Hold it so in the normal path, we still have it locked for
1361          * the second transaction.  In the error paths we need it
1362          * held so the cancel won't rele it, see below.
1363          */
1364         xfs_ilock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1365         size = (int)ip->i_d.di_size;
1366         ip->i_d.di_size = 0;
1367         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1368         xfs_trans_ihold(tp, ip);
1369         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1370         /*
1371          * Find the block(s) so we can inval and unmap them.
1372          */
1373         done = 0;
1374         XFS_BMAP_INIT(&free_list, &first_block);
1375         nmaps = sizeof(mval) / sizeof(mval[0]);
1376         if ((error = xfs_bmapi(tp, ip, 0, XFS_B_TO_FSB(mp, size),
1377                         XFS_BMAPI_METADATA, &first_block, 0, mval, &nmaps,
1378                         &free_list)))
1379                 goto error0;
1380         /*
1381          * Invalidate the block(s).
1382          */
1383         for (i = 0; i < nmaps; i++) {
1384                 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp,
1385                         XFS_FSB_TO_DADDR(mp, mval[i].br_startblock),
1386                         XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0);
1387                 xfs_trans_binval(tp, bp);
1388         }
1389         /*
1390          * Unmap the dead block(s) to the free_list.
1391          */
1392         if ((error = xfs_bunmapi(tp, ip, 0, size, XFS_BMAPI_METADATA, nmaps,
1393                         &first_block, &free_list, &done)))
1394                 goto error1;
1395         ASSERT(done);
1396         /*
1397          * Commit the first transaction.  This logs the EFI and the inode.
1398          */
1399         if ((error = xfs_bmap_finish(&tp, &free_list, first_block, &committed)))
1400                 goto error1;
1401         /*
1402          * The transaction must have been committed, since there were
1403          * actually extents freed by xfs_bunmapi.  See xfs_bmap_finish.
1404          * The new tp has the extent freeing and EFDs.
1405          */
1406         ASSERT(committed);
1407         /*
1408          * The first xact was committed, so add the inode to the new one.
1409          * Mark it dirty so it will be logged and moved forward in the log as
1410          * part of every commit.
1411          */
1412         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1413         xfs_trans_ihold(tp, ip);
1414         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1415         /*
1416          * Get a new, empty transaction to return to our caller.
1417          */
1418         ntp = xfs_trans_dup(tp);
1419         /*
1420          * Commit the transaction containing extent freeing and EFD's.
1421          * If we get an error on the commit here or on the reserve below,
1422          * we need to unlock the inode since the new transaction doesn't
1423          * have the inode attached.
1424          */
1425         error = xfs_trans_commit(tp, 0, NULL);
1426         tp = ntp;
1427         if (error) {
1428                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1429                 goto error0;
1430         }
1431         /*
1432          * Remove the memory for extent descriptions (just bookkeeping).
1433          */
1434         if (ip->i_df.if_bytes)
1435                 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK);
1436         ASSERT(ip->i_df.if_bytes == 0);
1437         /*
1438          * Put an itruncate log reservation in the new transaction
1439          * for our caller.
1440          */
1441         if ((error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
1442                         XFS_TRANS_PERM_LOG_RES, XFS_ITRUNCATE_LOG_COUNT))) {
1443                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1444                 goto error0;
1445         }
1446         /*
1447          * Return with the inode locked but not joined to the transaction.
1448          */
1449         *tpp = tp;
1450         return 0;
1451
1452  error1:
1453         xfs_bmap_cancel(&free_list);
1454  error0:
1455         /*
1456          * Have to come here with the inode locked and either
1457          * (held and in the transaction) or (not in the transaction).
1458          * If the inode isn't held then cancel would iput it, but
1459          * that's wrong since this is inactive and the vnode ref
1460          * count is 0 already.
1461          * Cancel won't do anything to the inode if held, but it still
1462          * needs to be locked until the cancel is done, if it was
1463          * joined to the transaction.
1464          */
1465         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1466         xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1467         *tpp = NULL;
1468         return error;
1469
1470 }
1471
1472 STATIC int
1473 xfs_inactive_symlink_local(
1474         xfs_inode_t     *ip,
1475         xfs_trans_t     **tpp)
1476 {
1477         int             error;
1478
1479         ASSERT(ip->i_d.di_size <= XFS_IFORK_DSIZE(ip));
1480         /*
1481          * We're freeing a symlink which fit into
1482          * the inode.  Just free the memory used
1483          * to hold the old symlink.
1484          */
1485         error = xfs_trans_reserve(*tpp, 0,
1486                                   XFS_ITRUNCATE_LOG_RES(ip->i_mount),
1487                                   0, XFS_TRANS_PERM_LOG_RES,
1488                                   XFS_ITRUNCATE_LOG_COUNT);
1489
1490         if (error) {
1491                 xfs_trans_cancel(*tpp, 0);
1492                 *tpp = NULL;
1493                 return (error);
1494         }
1495         xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1496
1497         /*
1498          * Zero length symlinks _can_ exist.
1499          */
1500         if (ip->i_df.if_bytes > 0) {
1501                 xfs_idata_realloc(ip,
1502                                   -(ip->i_df.if_bytes),
1503                                   XFS_DATA_FORK);
1504                 ASSERT(ip->i_df.if_bytes == 0);
1505         }
1506         return (0);
1507 }
1508
1509 /*
1510  *
1511  */
1512 STATIC int
1513 xfs_inactive_attrs(
1514         xfs_inode_t     *ip,
1515         xfs_trans_t     **tpp)
1516 {
1517         xfs_trans_t     *tp;
1518         int             error;
1519         xfs_mount_t     *mp;
1520
1521         ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE));
1522         tp = *tpp;
1523         mp = ip->i_mount;
1524         ASSERT(ip->i_d.di_forkoff != 0);
1525         xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1526         xfs_iunlock(ip, XFS_ILOCK_EXCL);
1527
1528         error = xfs_attr_inactive(ip);
1529         if (error) {
1530                 *tpp = NULL;
1531                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1532                 return (error); /* goto out*/
1533         }
1534
1535         tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1536         error = xfs_trans_reserve(tp, 0,
1537                                   XFS_IFREE_LOG_RES(mp),
1538                                   0, XFS_TRANS_PERM_LOG_RES,
1539                                   XFS_INACTIVE_LOG_COUNT);
1540         if (error) {
1541                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1542                 xfs_trans_cancel(tp, 0);
1543                 *tpp = NULL;
1544                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1545                 return (error);
1546         }
1547
1548         xfs_ilock(ip, XFS_ILOCK_EXCL);
1549         xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1550         xfs_trans_ihold(tp, ip);
1551         xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1552
1553         ASSERT(ip->i_d.di_anextents == 0);
1554
1555         *tpp = tp;
1556         return (0);
1557 }
1558
1559 STATIC int
1560 xfs_release(
1561         bhv_desc_t      *bdp)
1562 {
1563         xfs_inode_t     *ip;
1564         vnode_t         *vp;
1565         xfs_mount_t     *mp;
1566         int             error;
1567
1568         vp = BHV_TO_VNODE(bdp);
1569         ip = XFS_BHVTOI(bdp);
1570
1571         if ((vp->v_type != VREG) || (ip->i_d.di_mode == 0)) {
1572                 return 0;
1573         }
1574
1575         /* If this is a read-only mount, don't do this (would generate I/O) */
1576         if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1577                 return 0;
1578
1579 #ifdef HAVE_REFCACHE
1580         /* If we are in the NFS reference cache then don't do this now */
1581         if (ip->i_refcache)
1582                 return 0;
1583 #endif
1584
1585         mp = ip->i_mount;
1586
1587         if (ip->i_d.di_nlink != 0) {
1588                 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1589                      ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1590                      (ip->i_df.if_flags & XFS_IFEXTENTS))  &&
1591                     (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)))) {
1592                         if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1593                                 return (error);
1594                         /* Update linux inode block count after free above */
1595                         LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1596                                 ip->i_d.di_nblocks + ip->i_delayed_blks);
1597                 }
1598         }
1599
1600         return 0;
1601 }
1602
1603 /*
1604  * xfs_inactive
1605  *
1606  * This is called when the vnode reference count for the vnode
1607  * goes to zero.  If the file has been unlinked, then it must
1608  * now be truncated.  Also, we clear all of the read-ahead state
1609  * kept for the inode here since the file is now closed.
1610  */
1611 STATIC int
1612 xfs_inactive(
1613         bhv_desc_t      *bdp,
1614         cred_t          *credp)
1615 {
1616         xfs_inode_t     *ip;
1617         vnode_t         *vp;
1618         xfs_bmap_free_t free_list;
1619         xfs_fsblock_t   first_block;
1620         int             committed;
1621         xfs_trans_t     *tp;
1622         xfs_mount_t     *mp;
1623         int             error;
1624         int             truncate;
1625
1626         vp = BHV_TO_VNODE(bdp);
1627         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
1628
1629         ip = XFS_BHVTOI(bdp);
1630
1631         /*
1632          * If the inode is already free, then there can be nothing
1633          * to clean up here.
1634          */
1635         if (ip->i_d.di_mode == 0 || VN_BAD(vp)) {
1636                 ASSERT(ip->i_df.if_real_bytes == 0);
1637                 ASSERT(ip->i_df.if_broot_bytes == 0);
1638                 return VN_INACTIVE_CACHE;
1639         }
1640
1641         /*
1642          * Only do a truncate if it's a regular file with
1643          * some actual space in it.  It's OK to look at the
1644          * inode's fields without the lock because we're the
1645          * only one with a reference to the inode.
1646          */
1647         truncate = ((ip->i_d.di_nlink == 0) &&
1648             ((ip->i_d.di_size != 0) || (ip->i_d.di_nextents > 0)) &&
1649             ((ip->i_d.di_mode & S_IFMT) == S_IFREG));
1650
1651         mp = ip->i_mount;
1652
1653         if (ip->i_d.di_nlink == 0 &&
1654             DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_DESTROY)) {
1655                 (void) XFS_SEND_DESTROY(mp, vp, DM_RIGHT_NULL);
1656         }
1657
1658         error = 0;
1659
1660         /* If this is a read-only mount, don't do this (would generate I/O) */
1661         if (vp->v_vfsp->vfs_flag & VFS_RDONLY)
1662                 goto out;
1663
1664         if (ip->i_d.di_nlink != 0) {
1665                 if ((((ip->i_d.di_mode & S_IFMT) == S_IFREG) &&
1666                      ((ip->i_d.di_size > 0) || (VN_CACHED(vp) > 0)) &&
1667                      (ip->i_df.if_flags & XFS_IFEXTENTS))  &&
1668                     (!(ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)) ||
1669                      (ip->i_delayed_blks != 0))) {
1670                         if ((error = xfs_inactive_free_eofblocks(mp, ip)))
1671                                 return (VN_INACTIVE_CACHE);
1672                         /* Update linux inode block count after free above */
1673                         LINVFS_GET_IP(vp)->i_blocks = XFS_FSB_TO_BB(mp,
1674                                 ip->i_d.di_nblocks + ip->i_delayed_blks);
1675                 }
1676                 goto out;
1677         }
1678
1679         ASSERT(ip->i_d.di_nlink == 0);
1680
1681         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
1682                 return (VN_INACTIVE_CACHE);
1683
1684         tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
1685         if (truncate) {
1686                 /*
1687                  * Do the xfs_itruncate_start() call before
1688                  * reserving any log space because itruncate_start
1689                  * will call into the buffer cache and we can't
1690                  * do that within a transaction.
1691                  */
1692                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
1693
1694                 xfs_itruncate_start(ip, XFS_ITRUNC_DEFINITE, 0);
1695
1696                 error = xfs_trans_reserve(tp, 0,
1697                                           XFS_ITRUNCATE_LOG_RES(mp),
1698                                           0, XFS_TRANS_PERM_LOG_RES,
1699                                           XFS_ITRUNCATE_LOG_COUNT);
1700                 if (error) {
1701                         /* Don't call itruncate_cleanup */
1702                         ASSERT(XFS_FORCED_SHUTDOWN(mp));
1703                         xfs_trans_cancel(tp, 0);
1704                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1705                         return (VN_INACTIVE_CACHE);
1706                 }
1707
1708                 xfs_ilock(ip, XFS_ILOCK_EXCL);
1709                 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1710                 xfs_trans_ihold(tp, ip);
1711
1712                 /*
1713                  * normally, we have to run xfs_itruncate_finish sync.
1714                  * But if filesystem is wsync and we're in the inactive
1715                  * path, then we know that nlink == 0, and that the
1716                  * xaction that made nlink == 0 is permanently committed
1717                  * since xfs_remove runs as a synchronous transaction.
1718                  */
1719                 error = xfs_itruncate_finish(&tp, ip, 0, XFS_DATA_FORK,
1720                                 (!(mp->m_flags & XFS_MOUNT_WSYNC) ? 1 : 0));
1721
1722                 if (error) {
1723                         xfs_trans_cancel(tp,
1724                                 XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1725                         xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1726                         return (VN_INACTIVE_CACHE);
1727                 }
1728         } else if ((ip->i_d.di_mode & S_IFMT) == S_IFLNK) {
1729
1730                 /*
1731                  * If we get an error while cleaning up a
1732                  * symlink we bail out.
1733                  */
1734                 error = (ip->i_d.di_size > XFS_IFORK_DSIZE(ip)) ?
1735                         xfs_inactive_symlink_rmt(ip, &tp) :
1736                         xfs_inactive_symlink_local(ip, &tp);
1737
1738                 if (error) {
1739                         ASSERT(tp == NULL);
1740                         return (VN_INACTIVE_CACHE);
1741                 }
1742
1743                 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1744                 xfs_trans_ihold(tp, ip);
1745         } else {
1746                 error = xfs_trans_reserve(tp, 0,
1747                                           XFS_IFREE_LOG_RES(mp),
1748                                           0, XFS_TRANS_PERM_LOG_RES,
1749                                           XFS_INACTIVE_LOG_COUNT);
1750                 if (error) {
1751                         ASSERT(XFS_FORCED_SHUTDOWN(mp));
1752                         xfs_trans_cancel(tp, 0);
1753                         return (VN_INACTIVE_CACHE);
1754                 }
1755
1756                 xfs_ilock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1757                 xfs_trans_ijoin(tp, ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1758                 xfs_trans_ihold(tp, ip);
1759         }
1760
1761         /*
1762          * If there are attributes associated with the file
1763          * then blow them away now.  The code calls a routine
1764          * that recursively deconstructs the attribute fork.
1765          * We need to just commit the current transaction
1766          * because we can't use it for xfs_attr_inactive().
1767          */
1768         if (ip->i_d.di_anextents > 0) {
1769                 error = xfs_inactive_attrs(ip, &tp);
1770                 /*
1771                  * If we got an error, the transaction is already
1772                  * cancelled, and the inode is unlocked. Just get out.
1773                  */
1774                  if (error)
1775                          return (VN_INACTIVE_CACHE);
1776         } else if (ip->i_afp) {
1777                 xfs_idestroy_fork(ip, XFS_ATTR_FORK);
1778         }
1779
1780         /*
1781          * Free the inode.
1782          */
1783         XFS_BMAP_INIT(&free_list, &first_block);
1784         error = xfs_ifree(tp, ip, &free_list);
1785         if (error) {
1786                 /*
1787                  * If we fail to free the inode, shut down.  The cancel
1788                  * might do that, we need to make sure.  Otherwise the
1789                  * inode might be lost for a long time or forever.
1790                  */
1791                 if (!XFS_FORCED_SHUTDOWN(mp)) {
1792                         cmn_err(CE_NOTE,
1793                 "xfs_inactive:  xfs_ifree() returned an error = %d on %s",
1794                                 error, mp->m_fsname);
1795                         xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1796                 }
1797                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1798         } else {
1799                 /*
1800                  * Credit the quota account(s). The inode is gone.
1801                  */
1802                 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
1803
1804                 /*
1805                  * Just ignore errors at this point.  There is
1806                  * nothing we can do except to try to keep going.
1807                  */
1808                 (void) xfs_bmap_finish(&tp,  &free_list, first_block,
1809                                        &committed);
1810                 (void) xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
1811         }
1812         /*
1813          * Release the dquots held by inode, if any.
1814          */
1815         XFS_QM_DQDETACH(mp, ip);
1816
1817         xfs_iunlock(ip, XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL);
1818
1819  out:
1820         return VN_INACTIVE_CACHE;
1821 }
1822
1823
1824 /*
1825  * xfs_lookup
1826  */
1827 STATIC int
1828 xfs_lookup(
1829         bhv_desc_t              *dir_bdp,
1830         vname_t                 *dentry,
1831         vnode_t                 **vpp,
1832         int                     flags,
1833         vnode_t                 *rdir,
1834         cred_t                  *credp)
1835 {
1836         xfs_inode_t             *dp, *ip;
1837         xfs_ino_t               e_inum;
1838         int                     error;
1839         uint                    lock_mode;
1840         vnode_t                 *dir_vp;
1841
1842         dir_vp = BHV_TO_VNODE(dir_bdp);
1843         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1844
1845         dp = XFS_BHVTOI(dir_bdp);
1846
1847         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
1848                 return XFS_ERROR(EIO);
1849
1850         lock_mode = xfs_ilock_map_shared(dp);
1851         error = xfs_dir_lookup_int(dir_bdp, lock_mode, dentry, &e_inum, &ip);
1852         if (!error) {
1853                 *vpp = XFS_ITOV(ip);
1854                 ITRACE(ip);
1855         }
1856         xfs_iunlock_map_shared(dp, lock_mode);
1857         return error;
1858 }
1859
1860
1861 /*
1862  * xfs_create (create a new file).
1863  */
1864 STATIC int
1865 xfs_create(
1866         bhv_desc_t              *dir_bdp,
1867         vname_t                 *dentry,
1868         vattr_t                 *vap,
1869         vnode_t                 **vpp,
1870         cred_t                  *credp)
1871 {
1872         char                    *name = VNAME(dentry);
1873         vnode_t                 *dir_vp;
1874         xfs_inode_t             *dp, *ip;
1875         vnode_t                 *vp=NULL;
1876         xfs_trans_t             *tp;
1877         xfs_mount_t             *mp;
1878         xfs_dev_t               rdev;
1879         int                     error;
1880         xfs_bmap_free_t         free_list;
1881         xfs_fsblock_t           first_block;
1882         boolean_t               dp_joined_to_trans;
1883         int                     dm_event_sent = 0;
1884         uint                    cancel_flags;
1885         int                     committed;
1886         xfs_prid_t              prid;
1887         struct xfs_dquot        *udqp, *gdqp;
1888         uint                    resblks;
1889         int                     dm_di_mode;
1890         int                     namelen;
1891
1892         ASSERT(!*vpp);
1893         dir_vp = BHV_TO_VNODE(dir_bdp);
1894         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
1895
1896         dp = XFS_BHVTOI(dir_bdp);
1897         mp = dp->i_mount;
1898
1899         dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
1900         namelen = VNAMELEN(dentry);
1901
1902         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
1903                 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
1904                                 dir_vp, DM_RIGHT_NULL, NULL,
1905                                 DM_RIGHT_NULL, name, NULL,
1906                                 dm_di_mode, 0, 0);
1907
1908                 if (error)
1909                         return error;
1910                 dm_event_sent = 1;
1911         }
1912
1913         if (XFS_FORCED_SHUTDOWN(mp))
1914                 return XFS_ERROR(EIO);
1915
1916         /* Return through std_return after this point. */
1917
1918         udqp = gdqp = NULL;
1919         if (vap->va_mask & XFS_AT_PROJID)
1920                 prid = (xfs_prid_t)vap->va_projid;
1921         else
1922                 prid = (xfs_prid_t)dfltprid;
1923
1924         /*
1925          * Make sure that we have allocated dquot(s) on disk.
1926          */
1927         error = XFS_QM_DQVOPALLOC(mp, dp,
1928                         current_fsuid(credp), current_fsgid(credp),
1929                         XFS_QMOPT_QUOTALL|XFS_QMOPT_INHERIT, &udqp, &gdqp);
1930         if (error)
1931                 goto std_return;
1932
1933         ip = NULL;
1934         dp_joined_to_trans = B_FALSE;
1935
1936         tp = xfs_trans_alloc(mp, XFS_TRANS_CREATE);
1937         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1938         resblks = XFS_CREATE_SPACE_RES(mp, namelen);
1939         /*
1940          * Initially assume that the file does not exist and
1941          * reserve the resources for that case.  If that is not
1942          * the case we'll drop the one we have and get a more
1943          * appropriate transaction later.
1944          */
1945         error = xfs_trans_reserve(tp, resblks, XFS_CREATE_LOG_RES(mp), 0,
1946                         XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1947         if (error == ENOSPC) {
1948                 resblks = 0;
1949                 error = xfs_trans_reserve(tp, 0, XFS_CREATE_LOG_RES(mp), 0,
1950                                 XFS_TRANS_PERM_LOG_RES, XFS_CREATE_LOG_COUNT);
1951         }
1952         if (error) {
1953                 cancel_flags = 0;
1954                 dp = NULL;
1955                 goto error_return;
1956         }
1957
1958         xfs_ilock(dp, XFS_ILOCK_EXCL);
1959
1960         XFS_BMAP_INIT(&free_list, &first_block);
1961
1962         ASSERT(ip == NULL);
1963
1964         /*
1965          * Reserve disk quota and the inode.
1966          */
1967         error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
1968         if (error)
1969                 goto error_return;
1970
1971         if (resblks == 0 &&
1972             (error = XFS_DIR_CANENTER(mp, tp, dp, name, namelen)))
1973                 goto error_return;
1974         rdev = (vap->va_mask & XFS_AT_RDEV) ? vap->va_rdev : 0;
1975         error = xfs_dir_ialloc(&tp, dp,
1976                         MAKEIMODE(vap->va_type,vap->va_mode), 1,
1977                         rdev, credp, prid, resblks > 0,
1978                         &ip, &committed);
1979         if (error) {
1980                 if (error == ENOSPC)
1981                         goto error_return;
1982                 goto abort_return;
1983         }
1984         ITRACE(ip);
1985
1986         /*
1987          * At this point, we've gotten a newly allocated inode.
1988          * It is locked (and joined to the transaction).
1989          */
1990
1991         ASSERT(ismrlocked (&ip->i_lock, MR_UPDATE));
1992
1993         /*
1994          * Now we join the directory inode to the transaction.
1995          * We do not do it earlier because xfs_dir_ialloc
1996          * might commit the previous transaction (and release
1997          * all the locks).
1998          */
1999
2000         VN_HOLD(dir_vp);
2001         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2002         dp_joined_to_trans = B_TRUE;
2003
2004         error = XFS_DIR_CREATENAME(mp, tp, dp, name, namelen, ip->i_ino,
2005                 &first_block, &free_list,
2006                 resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2007         if (error) {
2008                 ASSERT(error != ENOSPC);
2009                 goto abort_return;
2010         }
2011         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2012         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2013
2014         /*
2015          * If this is a synchronous mount, make sure that the
2016          * create transaction goes to disk before returning to
2017          * the user.
2018          */
2019         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2020                 xfs_trans_set_sync(tp);
2021         }
2022
2023         dp->i_gen++;
2024
2025         /*
2026          * Attach the dquot(s) to the inodes and modify them incore.
2027          * These ids of the inode couldn't have changed since the new
2028          * inode has been locked ever since it was created.
2029          */
2030         XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
2031
2032         /*
2033          * xfs_trans_commit normally decrements the vnode ref count
2034          * when it unlocks the inode. Since we want to return the
2035          * vnode to the caller, we bump the vnode ref count now.
2036          */
2037         IHOLD(ip);
2038         vp = XFS_ITOV(ip);
2039
2040         error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2041         if (error) {
2042                 xfs_bmap_cancel(&free_list);
2043                 goto abort_rele;
2044         }
2045
2046         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2047         if (error) {
2048                 IRELE(ip);
2049                 tp = NULL;
2050                 goto error_return;
2051         }
2052
2053         XFS_QM_DQRELE(mp, udqp);
2054         XFS_QM_DQRELE(mp, gdqp);
2055
2056         /*
2057          * Propogate the fact that the vnode changed after the
2058          * xfs_inode locks have been released.
2059          */
2060         VOP_VNODE_CHANGE(vp, VCHANGE_FLAGS_TRUNCATED, 3);
2061
2062         *vpp = vp;
2063
2064         /* Fallthrough to std_return with error = 0  */
2065
2066 std_return:
2067         if ( (*vpp || (error != 0 && dm_event_sent != 0)) &&
2068                         DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2069                                                         DM_EVENT_POSTCREATE)) {
2070                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2071                         dir_vp, DM_RIGHT_NULL,
2072                         *vpp ? vp:NULL,
2073                         DM_RIGHT_NULL, name, NULL,
2074                         dm_di_mode, error, 0);
2075         }
2076         return error;
2077
2078  abort_return:
2079         cancel_flags |= XFS_TRANS_ABORT;
2080         /* FALLTHROUGH */
2081  error_return:
2082
2083         if (tp != NULL)
2084                 xfs_trans_cancel(tp, cancel_flags);
2085
2086         if (!dp_joined_to_trans && (dp != NULL))
2087                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2088         XFS_QM_DQRELE(mp, udqp);
2089         XFS_QM_DQRELE(mp, gdqp);
2090
2091         goto std_return;
2092
2093  abort_rele:
2094         /*
2095          * Wait until after the current transaction is aborted to
2096          * release the inode.  This prevents recursive transactions
2097          * and deadlocks from xfs_inactive.
2098          */
2099         cancel_flags |= XFS_TRANS_ABORT;
2100         xfs_trans_cancel(tp, cancel_flags);
2101         IRELE(ip);
2102
2103         XFS_QM_DQRELE(mp, udqp);
2104         XFS_QM_DQRELE(mp, gdqp);
2105
2106         goto std_return;
2107 }
2108
2109 #ifdef DEBUG
2110 /*
2111  * Some counters to see if (and how often) we are hitting some deadlock
2112  * prevention code paths.
2113  */
2114
2115 int xfs_rm_locks;
2116 int xfs_rm_lock_delays;
2117 int xfs_rm_attempts;
2118 #endif
2119
2120 /*
2121  * The following routine will lock the inodes associated with the
2122  * directory and the named entry in the directory. The locks are
2123  * acquired in increasing inode number.
2124  *
2125  * If the entry is "..", then only the directory is locked. The
2126  * vnode ref count will still include that from the .. entry in
2127  * this case.
2128  *
2129  * There is a deadlock we need to worry about. If the locked directory is
2130  * in the AIL, it might be blocking up the log. The next inode we lock
2131  * could be already locked by another thread waiting for log space (e.g
2132  * a permanent log reservation with a long running transaction (see
2133  * xfs_itruncate_finish)). To solve this, we must check if the directory
2134  * is in the ail and use lock_nowait. If we can't lock, we need to
2135  * drop the inode lock on the directory and try again. xfs_iunlock will
2136  * potentially push the tail if we were holding up the log.
2137  */
2138 STATIC int
2139 xfs_lock_dir_and_entry(
2140         xfs_inode_t     *dp,
2141         vname_t         *dentry,
2142         xfs_inode_t     *ip)    /* inode of entry 'name' */
2143 {
2144         int             attempts;
2145         xfs_ino_t       e_inum;
2146         xfs_inode_t     *ips[2];
2147         xfs_log_item_t  *lp;
2148
2149 #ifdef DEBUG
2150         xfs_rm_locks++;
2151 #endif
2152         attempts = 0;
2153
2154 again:
2155         xfs_ilock(dp, XFS_ILOCK_EXCL);
2156
2157         e_inum = ip->i_ino;
2158
2159         ITRACE(ip);
2160
2161         /*
2162          * We want to lock in increasing inum. Since we've already
2163          * acquired the lock on the directory, we may need to release
2164          * if if the inum of the entry turns out to be less.
2165          */
2166         if (e_inum > dp->i_ino) {
2167                 /*
2168                  * We are already in the right order, so just
2169                  * lock on the inode of the entry.
2170                  * We need to use nowait if dp is in the AIL.
2171                  */
2172
2173                 lp = (xfs_log_item_t *)dp->i_itemp;
2174                 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2175                         if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
2176                                 attempts++;
2177 #ifdef DEBUG
2178                                 xfs_rm_attempts++;
2179 #endif
2180
2181                                 /*
2182                                  * Unlock dp and try again.
2183                                  * xfs_iunlock will try to push the tail
2184                                  * if the inode is in the AIL.
2185                                  */
2186
2187                                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2188
2189                                 if ((attempts % 5) == 0) {
2190                                         delay(1); /* Don't just spin the CPU */
2191 #ifdef DEBUG
2192                                         xfs_rm_lock_delays++;
2193 #endif
2194                                 }
2195                                 goto again;
2196                         }
2197                 } else {
2198                         xfs_ilock(ip, XFS_ILOCK_EXCL);
2199                 }
2200         } else if (e_inum < dp->i_ino) {
2201                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2202
2203                 ips[0] = ip;
2204                 ips[1] = dp;
2205                 xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2206         }
2207         /* else  e_inum == dp->i_ino */
2208         /*     This can happen if we're asked to lock /x/..
2209          *     the entry is "..", which is also the parent directory.
2210          */
2211
2212         return 0;
2213 }
2214
2215 #ifdef DEBUG
2216 int xfs_locked_n;
2217 int xfs_small_retries;
2218 int xfs_middle_retries;
2219 int xfs_lots_retries;
2220 int xfs_lock_delays;
2221 #endif
2222
2223 /*
2224  * The following routine will lock n inodes in exclusive mode.
2225  * We assume the caller calls us with the inodes in i_ino order.
2226  *
2227  * We need to detect deadlock where an inode that we lock
2228  * is in the AIL and we start waiting for another inode that is locked
2229  * by a thread in a long running transaction (such as truncate). This can
2230  * result in deadlock since the long running trans might need to wait
2231  * for the inode we just locked in order to push the tail and free space
2232  * in the log.
2233  */
2234 void
2235 xfs_lock_inodes(
2236         xfs_inode_t     **ips,
2237         int             inodes,
2238         int             first_locked,
2239         uint            lock_mode)
2240 {
2241         int             attempts = 0, i, j, try_lock;
2242         xfs_log_item_t  *lp;
2243
2244         ASSERT(ips && (inodes >= 2)); /* we need at least two */
2245
2246         if (first_locked) {
2247                 try_lock = 1;
2248                 i = 1;
2249         } else {
2250                 try_lock = 0;
2251                 i = 0;
2252         }
2253
2254 again:
2255         for (; i < inodes; i++) {
2256                 ASSERT(ips[i]);
2257
2258                 if (i && (ips[i] == ips[i-1]))  /* Already locked */
2259                         continue;
2260
2261                 /*
2262                  * If try_lock is not set yet, make sure all locked inodes
2263                  * are not in the AIL.
2264                  * If any are, set try_lock to be used later.
2265                  */
2266
2267                 if (!try_lock) {
2268                         for (j = (i - 1); j >= 0 && !try_lock; j--) {
2269                                 lp = (xfs_log_item_t *)ips[j]->i_itemp;
2270                                 if (lp && (lp->li_flags & XFS_LI_IN_AIL)) {
2271                                         try_lock++;
2272                                 }
2273                         }
2274                 }
2275
2276                 /*
2277                  * If any of the previous locks we have locked is in the AIL,
2278                  * we must TRY to get the second and subsequent locks. If
2279                  * we can't get any, we must release all we have
2280                  * and try again.
2281                  */
2282
2283                 if (try_lock) {
2284                         /* try_lock must be 0 if i is 0. */
2285                         /*
2286                          * try_lock means we have an inode locked
2287                          * that is in the AIL.
2288                          */
2289                         ASSERT(i != 0);
2290                         if (!xfs_ilock_nowait(ips[i], lock_mode)) {
2291                                 attempts++;
2292
2293                                 /*
2294                                  * Unlock all previous guys and try again.
2295                                  * xfs_iunlock will try to push the tail
2296                                  * if the inode is in the AIL.
2297                                  */
2298
2299                                 for(j = i - 1; j >= 0; j--) {
2300
2301                                         /*
2302                                          * Check to see if we've already
2303                                          * unlocked this one.
2304                                          * Not the first one going back,
2305                                          * and the inode ptr is the same.
2306                                          */
2307                                         if ((j != (i - 1)) && ips[j] ==
2308                                                                 ips[j+1])
2309                                                 continue;
2310
2311                                         xfs_iunlock(ips[j], lock_mode);
2312                                 }
2313
2314                                 if ((attempts % 5) == 0) {
2315                                         delay(1); /* Don't just spin the CPU */
2316 #ifdef DEBUG
2317                                         xfs_lock_delays++;
2318 #endif
2319                                 }
2320                                 i = 0;
2321                                 try_lock = 0;
2322                                 goto again;
2323                         }
2324                 } else {
2325                         xfs_ilock(ips[i], lock_mode);
2326                 }
2327         }
2328
2329 #ifdef DEBUG
2330         if (attempts) {
2331                 if (attempts < 5) xfs_small_retries++;
2332                 else if (attempts < 100) xfs_middle_retries++;
2333                 else xfs_lots_retries++;
2334         } else {
2335                 xfs_locked_n++;
2336         }
2337 #endif
2338 }
2339
2340 #ifdef  DEBUG
2341 #define REMOVE_DEBUG_TRACE(x)   {remove_which_error_return = (x);}
2342 int remove_which_error_return = 0;
2343 #else /* ! DEBUG */
2344 #define REMOVE_DEBUG_TRACE(x)
2345 #endif  /* ! DEBUG */
2346
2347
2348 /*
2349  * xfs_remove
2350  *
2351  */
2352 STATIC int
2353 xfs_remove(
2354         bhv_desc_t              *dir_bdp,
2355         vname_t                 *dentry,
2356         cred_t                  *credp)
2357 {
2358         vnode_t                 *dir_vp;
2359         char                    *name = VNAME(dentry);
2360         xfs_inode_t             *dp, *ip;
2361         xfs_trans_t             *tp = NULL;
2362         xfs_mount_t             *mp;
2363         int                     error = 0;
2364         xfs_bmap_free_t         free_list;
2365         xfs_fsblock_t           first_block;
2366         int                     cancel_flags;
2367         int                     committed;
2368         int                     dm_di_mode = 0;
2369         int                     link_zero;
2370         uint                    resblks;
2371         int                     namelen;
2372
2373         dir_vp = BHV_TO_VNODE(dir_bdp);
2374         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2375
2376         dp = XFS_BHVTOI(dir_bdp);
2377         mp = dp->i_mount;
2378
2379         if (XFS_FORCED_SHUTDOWN(mp))
2380                 return XFS_ERROR(EIO);
2381
2382         namelen = VNAMELEN(dentry);
2383
2384         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
2385                 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE, dir_vp,
2386                                         DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
2387                                         name, NULL, 0, 0, 0);
2388                 if (error)
2389                         return error;
2390         }
2391
2392         /* From this point on, return through std_return */
2393         ip = NULL;
2394
2395         /*
2396          * We need to get a reference to ip before we get our log
2397          * reservation. The reason for this is that we cannot call
2398          * xfs_iget for an inode for which we do not have a reference
2399          * once we've acquired a log reservation. This is because the
2400          * inode we are trying to get might be in xfs_inactive going
2401          * for a log reservation. Since we'll have to wait for the
2402          * inactive code to complete before returning from xfs_iget,
2403          * we need to make sure that we don't have log space reserved
2404          * when we call xfs_iget.  Instead we get an unlocked referece
2405          * to the inode before getting our log reservation.
2406          */
2407         error = xfs_get_dir_entry(dentry, &ip);
2408         if (error) {
2409                 REMOVE_DEBUG_TRACE(__LINE__);
2410                 goto std_return;
2411         }
2412
2413         dm_di_mode = ip->i_d.di_mode;
2414
2415         vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2416
2417         ITRACE(ip);
2418
2419         error = XFS_QM_DQATTACH(mp, dp, 0);
2420         if (!error && dp != ip)
2421                 error = XFS_QM_DQATTACH(mp, ip, 0);
2422         if (error) {
2423                 REMOVE_DEBUG_TRACE(__LINE__);
2424                 IRELE(ip);
2425                 goto std_return;
2426         }
2427
2428         tp = xfs_trans_alloc(mp, XFS_TRANS_REMOVE);
2429         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2430         /*
2431          * We try to get the real space reservation first,
2432          * allowing for directory btree deletion(s) implying
2433          * possible bmap insert(s).  If we can't get the space
2434          * reservation then we use 0 instead, and avoid the bmap
2435          * btree insert(s) in the directory code by, if the bmap
2436          * insert tries to happen, instead trimming the LAST
2437          * block from the directory.
2438          */
2439         resblks = XFS_REMOVE_SPACE_RES(mp);
2440         error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
2441                         XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2442         if (error == ENOSPC) {
2443                 resblks = 0;
2444                 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
2445                                 XFS_TRANS_PERM_LOG_RES, XFS_REMOVE_LOG_COUNT);
2446         }
2447         if (error) {
2448                 ASSERT(error != ENOSPC);
2449                 REMOVE_DEBUG_TRACE(__LINE__);
2450                 xfs_trans_cancel(tp, 0);
2451                 IRELE(ip);
2452                 return error;
2453         }
2454
2455         error = xfs_lock_dir_and_entry(dp, dentry, ip);
2456         if (error) {
2457                 REMOVE_DEBUG_TRACE(__LINE__);
2458                 xfs_trans_cancel(tp, cancel_flags);
2459                 IRELE(ip);
2460                 goto std_return;
2461         }
2462
2463         /*
2464          * At this point, we've gotten both the directory and the entry
2465          * inodes locked.
2466          */
2467         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2468         if (dp != ip) {
2469                 /*
2470                  * Increment vnode ref count only in this case since
2471                  * there's an extra vnode reference in the case where
2472                  * dp == ip.
2473                  */
2474                 IHOLD(dp);
2475                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
2476         }
2477
2478         /*
2479          * Entry must exist since we did a lookup in xfs_lock_dir_and_entry.
2480          */
2481         XFS_BMAP_INIT(&free_list, &first_block);
2482         error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, ip->i_ino,
2483                 &first_block, &free_list, 0);
2484         if (error) {
2485                 ASSERT(error != ENOENT);
2486                 REMOVE_DEBUG_TRACE(__LINE__);
2487                 goto error1;
2488         }
2489         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2490
2491         dp->i_gen++;
2492         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
2493
2494         error = xfs_droplink(tp, ip);
2495         if (error) {
2496                 REMOVE_DEBUG_TRACE(__LINE__);
2497                 goto error1;
2498         }
2499
2500         /* Determine if this is the last link while
2501          * we are in the transaction.
2502          */
2503         link_zero = (ip)->i_d.di_nlink==0;
2504
2505         /*
2506          * Take an extra ref on the inode so that it doesn't
2507          * go to xfs_inactive() from within the commit.
2508          */
2509         IHOLD(ip);
2510
2511         /*
2512          * If this is a synchronous mount, make sure that the
2513          * remove transaction goes to disk before returning to
2514          * the user.
2515          */
2516         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2517                 xfs_trans_set_sync(tp);
2518         }
2519
2520         error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2521         if (error) {
2522                 REMOVE_DEBUG_TRACE(__LINE__);
2523                 goto error_rele;
2524         }
2525
2526         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2527         if (error) {
2528                 IRELE(ip);
2529                 goto std_return;
2530         }
2531
2532         /*
2533          * Before we drop our extra reference to the inode, purge it
2534          * from the refcache if it is there.  By waiting until afterwards
2535          * to do the IRELE, we ensure that we won't go inactive in the
2536          * xfs_refcache_purge_ip routine (although that would be OK).
2537          */
2538         xfs_refcache_purge_ip(ip);
2539
2540         vn_trace_exit(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
2541
2542         /*
2543          * Let interposed file systems know about removed links.
2544          */
2545         VOP_LINK_REMOVED(XFS_ITOV(ip), dir_vp, link_zero);
2546
2547         IRELE(ip);
2548
2549 /*      Fall through to std_return with error = 0 */
2550  std_return:
2551         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp,
2552                                                 DM_EVENT_POSTREMOVE)) {
2553                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
2554                                 dir_vp, DM_RIGHT_NULL,
2555                                 NULL, DM_RIGHT_NULL,
2556                                 name, NULL, dm_di_mode, error, 0);
2557         }
2558         return error;
2559
2560  error1:
2561         xfs_bmap_cancel(&free_list);
2562         cancel_flags |= XFS_TRANS_ABORT;
2563         xfs_trans_cancel(tp, cancel_flags);
2564         goto std_return;
2565
2566  error_rele:
2567         /*
2568          * In this case make sure to not release the inode until after
2569          * the current transaction is aborted.  Releasing it beforehand
2570          * can cause us to go to xfs_inactive and start a recursive
2571          * transaction which can easily deadlock with the current one.
2572          */
2573         xfs_bmap_cancel(&free_list);
2574         cancel_flags |= XFS_TRANS_ABORT;
2575         xfs_trans_cancel(tp, cancel_flags);
2576
2577         /*
2578          * Before we drop our extra reference to the inode, purge it
2579          * from the refcache if it is there.  By waiting until afterwards
2580          * to do the IRELE, we ensure that we won't go inactive in the
2581          * xfs_refcache_purge_ip routine (although that would be OK).
2582          */
2583         xfs_refcache_purge_ip(ip);
2584
2585         IRELE(ip);
2586
2587         goto std_return;
2588 }
2589
2590
2591 /*
2592  * xfs_link
2593  *
2594  */
2595 STATIC int
2596 xfs_link(
2597         bhv_desc_t              *target_dir_bdp,
2598         vnode_t                 *src_vp,
2599         vname_t                 *dentry,
2600         cred_t                  *credp)
2601 {
2602         xfs_inode_t             *tdp, *sip;
2603         xfs_trans_t             *tp;
2604         xfs_mount_t             *mp;
2605         xfs_inode_t             *ips[2];
2606         int                     error;
2607         xfs_bmap_free_t         free_list;
2608         xfs_fsblock_t           first_block;
2609         int                     cancel_flags;
2610         int                     committed;
2611         vnode_t                 *target_dir_vp;
2612         bhv_desc_t              *src_bdp;
2613         int                     resblks;
2614         char                    *target_name = VNAME(dentry);
2615         int                     target_namelen;
2616
2617         target_dir_vp = BHV_TO_VNODE(target_dir_bdp);
2618         vn_trace_entry(target_dir_vp, __FUNCTION__, (inst_t *)__return_address);
2619         vn_trace_entry(src_vp, __FUNCTION__, (inst_t *)__return_address);
2620
2621         target_namelen = VNAMELEN(dentry);
2622         if (src_vp->v_type == VDIR)
2623                 return XFS_ERROR(EPERM);
2624
2625         /*
2626          * For now, manually find the XFS behavior descriptor for
2627          * the source vnode.  If it doesn't exist then something
2628          * is wrong and we should just return an error.
2629          * Eventually we need to figure out how link is going to
2630          * work in the face of stacked vnodes.
2631          */
2632         src_bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(src_vp), &xfs_vnodeops);
2633         if (src_bdp == NULL) {
2634                 return XFS_ERROR(EXDEV);
2635         }
2636         sip = XFS_BHVTOI(src_bdp);
2637         tdp = XFS_BHVTOI(target_dir_bdp);
2638         mp = tdp->i_mount;
2639         if (XFS_FORCED_SHUTDOWN(mp))
2640                 return XFS_ERROR(EIO);
2641
2642         if (DM_EVENT_ENABLED(src_vp->v_vfsp, tdp, DM_EVENT_LINK)) {
2643                 error = XFS_SEND_NAMESP(mp, DM_EVENT_LINK,
2644                                         target_dir_vp, DM_RIGHT_NULL,
2645                                         src_vp, DM_RIGHT_NULL,
2646                                         target_name, NULL, 0, 0, 0);
2647                 if (error)
2648                         return error;
2649         }
2650
2651         /* Return through std_return after this point. */
2652
2653         error = XFS_QM_DQATTACH(mp, sip, 0);
2654         if (!error && sip != tdp)
2655                 error = XFS_QM_DQATTACH(mp, tdp, 0);
2656         if (error)
2657                 goto std_return;
2658
2659         tp = xfs_trans_alloc(mp, XFS_TRANS_LINK);
2660         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2661         resblks = XFS_LINK_SPACE_RES(mp, target_namelen);
2662         error = xfs_trans_reserve(tp, resblks, XFS_LINK_LOG_RES(mp), 0,
2663                         XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2664         if (error == ENOSPC) {
2665                 resblks = 0;
2666                 error = xfs_trans_reserve(tp, 0, XFS_LINK_LOG_RES(mp), 0,
2667                                 XFS_TRANS_PERM_LOG_RES, XFS_LINK_LOG_COUNT);
2668         }
2669         if (error) {
2670                 cancel_flags = 0;
2671                 goto error_return;
2672         }
2673
2674         if (sip->i_ino < tdp->i_ino) {
2675                 ips[0] = sip;
2676                 ips[1] = tdp;
2677         } else {
2678                 ips[0] = tdp;
2679                 ips[1] = sip;
2680         }
2681
2682         xfs_lock_inodes(ips, 2, 0, XFS_ILOCK_EXCL);
2683
2684         /*
2685          * Increment vnode ref counts since xfs_trans_commit &
2686          * xfs_trans_cancel will both unlock the inodes and
2687          * decrement the associated ref counts.
2688          */
2689         VN_HOLD(src_vp);
2690         VN_HOLD(target_dir_vp);
2691         xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
2692         xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
2693
2694         /*
2695          * If the source has too many links, we can't make any more to it.
2696          */
2697         if (sip->i_d.di_nlink >= XFS_MAXLINK) {
2698                 error = XFS_ERROR(EMLINK);
2699                 goto error_return;
2700         }
2701
2702         if (resblks == 0 &&
2703             (error = XFS_DIR_CANENTER(mp, tp, tdp, target_name,
2704                         target_namelen)))
2705                 goto error_return;
2706
2707         XFS_BMAP_INIT(&free_list, &first_block);
2708
2709         error = XFS_DIR_CREATENAME(mp, tp, tdp, target_name, target_namelen,
2710                                    sip->i_ino, &first_block, &free_list,
2711                                    resblks);
2712         if (error)
2713                 goto abort_return;
2714         xfs_ichgtime(tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2715         tdp->i_gen++;
2716         xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
2717
2718         error = xfs_bumplink(tp, sip);
2719         if (error) {
2720                 goto abort_return;
2721         }
2722
2723         /*
2724          * If this is a synchronous mount, make sure that the
2725          * link transaction goes to disk before returning to
2726          * the user.
2727          */
2728         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2729                 xfs_trans_set_sync(tp);
2730         }
2731
2732         error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
2733         if (error) {
2734                 xfs_bmap_cancel(&free_list);
2735                 goto abort_return;
2736         }
2737
2738         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2739         if (error) {
2740                 goto std_return;
2741         }
2742
2743         /* Fall through to std_return with error = 0. */
2744 std_return:
2745         if (DM_EVENT_ENABLED(src_vp->v_vfsp, sip,
2746                                                 DM_EVENT_POSTLINK)) {
2747                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTLINK,
2748                                 target_dir_vp, DM_RIGHT_NULL,
2749                                 src_vp, DM_RIGHT_NULL,
2750                                 target_name, NULL, 0, error, 0);
2751         }
2752         return error;
2753
2754  abort_return:
2755         cancel_flags |= XFS_TRANS_ABORT;
2756         /* FALLTHROUGH */
2757  error_return:
2758         xfs_trans_cancel(tp, cancel_flags);
2759
2760         goto std_return;
2761 }
2762 /*
2763  * xfs_mkdir
2764  *
2765  */
2766 STATIC int
2767 xfs_mkdir(
2768         bhv_desc_t              *dir_bdp,
2769         vname_t                 *dentry,
2770         vattr_t                 *vap,
2771         vnode_t                 **vpp,
2772         cred_t                  *credp)
2773 {
2774         char                    *dir_name = VNAME(dentry);
2775         xfs_inode_t             *dp;
2776         xfs_inode_t             *cdp;   /* inode of created dir */
2777         vnode_t                 *cvp;   /* vnode of created dir */
2778         xfs_trans_t             *tp;
2779         xfs_mount_t             *mp;
2780         int                     cancel_flags;
2781         int                     error;
2782         int                     committed;
2783         xfs_bmap_free_t         free_list;
2784         xfs_fsblock_t           first_block;
2785         vnode_t                 *dir_vp;
2786         boolean_t               dp_joined_to_trans;
2787         boolean_t               created = B_FALSE;
2788         int                     dm_event_sent = 0;
2789         xfs_prid_t              prid;
2790         struct xfs_dquot        *udqp, *gdqp;
2791         uint                    resblks;
2792         int                     dm_di_mode;
2793         int                     dir_namelen;
2794
2795         dir_vp = BHV_TO_VNODE(dir_bdp);
2796         dp = XFS_BHVTOI(dir_bdp);
2797         mp = dp->i_mount;
2798
2799         if (XFS_FORCED_SHUTDOWN(mp))
2800                 return XFS_ERROR(EIO);
2801
2802         dir_namelen = VNAMELEN(dentry);
2803
2804         tp = NULL;
2805         dp_joined_to_trans = B_FALSE;
2806         dm_di_mode = vap->va_mode|VTTOIF(vap->va_type);
2807
2808         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_CREATE)) {
2809                 error = XFS_SEND_NAMESP(mp, DM_EVENT_CREATE,
2810                                         dir_vp, DM_RIGHT_NULL, NULL,
2811                                         DM_RIGHT_NULL, dir_name, NULL,
2812                                         dm_di_mode, 0, 0);
2813                 if (error)
2814                         return error;
2815                 dm_event_sent = 1;
2816         }
2817
2818         /* Return through std_return after this point. */
2819
2820         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
2821
2822         mp = dp->i_mount;
2823         udqp = gdqp = NULL;
2824         if (vap->va_mask & XFS_AT_PROJID)
2825                 prid = (xfs_prid_t)vap->va_projid;
2826         else
2827                 prid = (xfs_prid_t)dfltprid;
2828
2829         /*
2830          * Make sure that we have allocated dquot(s) on disk.
2831          */
2832         error = XFS_QM_DQVOPALLOC(mp, dp,
2833                         current_fsuid(credp), current_fsgid(credp),
2834                         XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
2835         if (error)
2836                 goto std_return;
2837
2838         tp = xfs_trans_alloc(mp, XFS_TRANS_MKDIR);
2839         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
2840         resblks = XFS_MKDIR_SPACE_RES(mp, dir_namelen);
2841         error = xfs_trans_reserve(tp, resblks, XFS_MKDIR_LOG_RES(mp), 0,
2842                                   XFS_TRANS_PERM_LOG_RES, XFS_MKDIR_LOG_COUNT);
2843         if (error == ENOSPC) {
2844                 resblks = 0;
2845                 error = xfs_trans_reserve(tp, 0, XFS_MKDIR_LOG_RES(mp), 0,
2846                                           XFS_TRANS_PERM_LOG_RES,
2847                                           XFS_MKDIR_LOG_COUNT);
2848         }
2849         if (error) {
2850                 cancel_flags = 0;
2851                 dp = NULL;
2852                 goto error_return;
2853         }
2854
2855         xfs_ilock(dp, XFS_ILOCK_EXCL);
2856
2857         /*
2858          * Check for directory link count overflow.
2859          */
2860         if (dp->i_d.di_nlink >= XFS_MAXLINK) {
2861                 error = XFS_ERROR(EMLINK);
2862                 goto error_return;
2863         }
2864
2865         /*
2866          * Reserve disk quota and the inode.
2867          */
2868         error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
2869         if (error)
2870                 goto error_return;
2871
2872         if (resblks == 0 &&
2873             (error = XFS_DIR_CANENTER(mp, tp, dp, dir_name, dir_namelen)))
2874                 goto error_return;
2875         /*
2876          * create the directory inode.
2877          */
2878         error = xfs_dir_ialloc(&tp, dp,
2879                         MAKEIMODE(vap->va_type,vap->va_mode), 2,
2880                         0, credp, prid, resblks > 0,
2881                 &cdp, NULL);
2882         if (error) {
2883                 if (error == ENOSPC)
2884                         goto error_return;
2885                 goto abort_return;
2886         }
2887         ITRACE(cdp);
2888
2889         /*
2890          * Now we add the directory inode to the transaction.
2891          * We waited until now since xfs_dir_ialloc might start
2892          * a new transaction.  Had we joined the transaction
2893          * earlier, the locks might have gotten released.
2894          */
2895         VN_HOLD(dir_vp);
2896         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
2897         dp_joined_to_trans = B_TRUE;
2898
2899         XFS_BMAP_INIT(&free_list, &first_block);
2900
2901         error = XFS_DIR_CREATENAME(mp, tp, dp, dir_name, dir_namelen,
2902                         cdp->i_ino, &first_block, &free_list,
2903                         resblks ? resblks - XFS_IALLOC_SPACE_RES(mp) : 0);
2904         if (error) {
2905                 ASSERT(error != ENOSPC);
2906                 goto error1;
2907         }
2908         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
2909
2910         /*
2911          * Bump the in memory version number of the parent directory
2912          * so that other processes accessing it will recognize that
2913          * the directory has changed.
2914          */
2915         dp->i_gen++;
2916
2917         error = XFS_DIR_INIT(mp, tp, cdp, dp);
2918         if (error) {
2919                 goto error2;
2920         }
2921
2922         cdp->i_gen = 1;
2923         error = xfs_bumplink(tp, dp);
2924         if (error) {
2925                 goto error2;
2926         }
2927
2928         cvp = XFS_ITOV(cdp);
2929
2930         created = B_TRUE;
2931
2932         *vpp = cvp;
2933         IHOLD(cdp);
2934
2935         /*
2936          * Attach the dquots to the new inode and modify the icount incore.
2937          */
2938         XFS_QM_DQVOPCREATE(mp, tp, cdp, udqp, gdqp);
2939
2940         /*
2941          * If this is a synchronous mount, make sure that the
2942          * mkdir transaction goes to disk before returning to
2943          * the user.
2944          */
2945         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
2946                 xfs_trans_set_sync(tp);
2947         }
2948
2949         error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
2950         if (error) {
2951                 IRELE(cdp);
2952                 goto error2;
2953         }
2954
2955         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
2956         XFS_QM_DQRELE(mp, udqp);
2957         XFS_QM_DQRELE(mp, gdqp);
2958         if (error) {
2959                 IRELE(cdp);
2960         }
2961
2962         /* Fall through to std_return with error = 0 or errno from
2963          * xfs_trans_commit. */
2964
2965 std_return:
2966         if ( (created || (error != 0 && dm_event_sent != 0)) &&
2967                         DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
2968                                                 DM_EVENT_POSTCREATE)) {
2969                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTCREATE,
2970                                         dir_vp, DM_RIGHT_NULL,
2971                                         created ? XFS_ITOV(cdp):NULL,
2972                                         DM_RIGHT_NULL,
2973                                         dir_name, NULL,
2974                                         dm_di_mode, error, 0);
2975         }
2976         return error;
2977
2978  error2:
2979  error1:
2980         xfs_bmap_cancel(&free_list);
2981  abort_return:
2982         cancel_flags |= XFS_TRANS_ABORT;
2983  error_return:
2984         xfs_trans_cancel(tp, cancel_flags);
2985         XFS_QM_DQRELE(mp, udqp);
2986         XFS_QM_DQRELE(mp, gdqp);
2987
2988         if (!dp_joined_to_trans && (dp != NULL)) {
2989                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
2990         }
2991
2992         goto std_return;
2993 }
2994
2995
2996 /*
2997  * xfs_rmdir
2998  *
2999  */
3000 STATIC int
3001 xfs_rmdir(
3002         bhv_desc_t              *dir_bdp,
3003         vname_t                 *dentry,
3004         cred_t                  *credp)
3005 {
3006         char                    *name = VNAME(dentry);
3007         xfs_inode_t             *dp;
3008         xfs_inode_t             *cdp;   /* child directory */
3009         xfs_trans_t             *tp;
3010         xfs_mount_t             *mp;
3011         int                     error;
3012         xfs_bmap_free_t         free_list;
3013         xfs_fsblock_t           first_block;
3014         int                     cancel_flags;
3015         int                     committed;
3016         vnode_t                 *dir_vp;
3017         int                     dm_di_mode = 0;
3018         int                     last_cdp_link;
3019         int                     namelen;
3020         uint                    resblks;
3021
3022         dir_vp = BHV_TO_VNODE(dir_bdp);
3023         dp = XFS_BHVTOI(dir_bdp);
3024         mp = dp->i_mount;
3025
3026         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3027
3028         if (XFS_FORCED_SHUTDOWN(XFS_BHVTOI(dir_bdp)->i_mount))
3029                 return XFS_ERROR(EIO);
3030         namelen = VNAMELEN(dentry);
3031
3032         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_REMOVE)) {
3033                 error = XFS_SEND_NAMESP(mp, DM_EVENT_REMOVE,
3034                                         dir_vp, DM_RIGHT_NULL,
3035                                         NULL, DM_RIGHT_NULL,
3036                                         name, NULL, 0, 0, 0);
3037                 if (error)
3038                         return XFS_ERROR(error);
3039         }
3040
3041         /* Return through std_return after this point. */
3042
3043         cdp = NULL;
3044
3045         /*
3046          * We need to get a reference to cdp before we get our log
3047          * reservation.  The reason for this is that we cannot call
3048          * xfs_iget for an inode for which we do not have a reference
3049          * once we've acquired a log reservation.  This is because the
3050          * inode we are trying to get might be in xfs_inactive going
3051          * for a log reservation.  Since we'll have to wait for the
3052          * inactive code to complete before returning from xfs_iget,
3053          * we need to make sure that we don't have log space reserved
3054          * when we call xfs_iget.  Instead we get an unlocked referece
3055          * to the inode before getting our log reservation.
3056          */
3057         error = xfs_get_dir_entry(dentry, &cdp);
3058         if (error) {
3059                 REMOVE_DEBUG_TRACE(__LINE__);
3060                 goto std_return;
3061         }
3062         mp = dp->i_mount;
3063         dm_di_mode = cdp->i_d.di_mode;
3064
3065         /*
3066          * Get the dquots for the inodes.
3067          */
3068         error = XFS_QM_DQATTACH(mp, dp, 0);
3069         if (!error && dp != cdp)
3070                 error = XFS_QM_DQATTACH(mp, cdp, 0);
3071         if (error) {
3072                 IRELE(cdp);
3073                 REMOVE_DEBUG_TRACE(__LINE__);
3074                 goto std_return;
3075         }
3076
3077         tp = xfs_trans_alloc(mp, XFS_TRANS_RMDIR);
3078         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3079         /*
3080          * We try to get the real space reservation first,
3081          * allowing for directory btree deletion(s) implying
3082          * possible bmap insert(s).  If we can't get the space
3083          * reservation then we use 0 instead, and avoid the bmap
3084          * btree insert(s) in the directory code by, if the bmap
3085          * insert tries to happen, instead trimming the LAST
3086          * block from the directory.
3087          */
3088         resblks = XFS_REMOVE_SPACE_RES(mp);
3089         error = xfs_trans_reserve(tp, resblks, XFS_REMOVE_LOG_RES(mp), 0,
3090                         XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3091         if (error == ENOSPC) {
3092                 resblks = 0;
3093                 error = xfs_trans_reserve(tp, 0, XFS_REMOVE_LOG_RES(mp), 0,
3094                                 XFS_TRANS_PERM_LOG_RES, XFS_DEFAULT_LOG_COUNT);
3095         }
3096         if (error) {
3097                 ASSERT(error != ENOSPC);
3098                 cancel_flags = 0;
3099                 IRELE(cdp);
3100                 goto error_return;
3101         }
3102         XFS_BMAP_INIT(&free_list, &first_block);
3103
3104         /*
3105          * Now lock the child directory inode and the parent directory
3106          * inode in the proper order.  This will take care of validating
3107          * that the directory entry for the child directory inode has
3108          * not changed while we were obtaining a log reservation.
3109          */
3110         error = xfs_lock_dir_and_entry(dp, dentry, cdp);
3111         if (error) {
3112                 xfs_trans_cancel(tp, cancel_flags);
3113                 IRELE(cdp);
3114                 goto std_return;
3115         }
3116
3117         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3118         if (dp != cdp) {
3119                 /*
3120                  * Only increment the parent directory vnode count if
3121                  * we didn't bump it in looking up cdp.  The only time
3122                  * we don't bump it is when we're looking up ".".
3123                  */
3124                 VN_HOLD(dir_vp);
3125         }
3126
3127         ITRACE(cdp);
3128         xfs_trans_ijoin(tp, cdp, XFS_ILOCK_EXCL);
3129
3130         ASSERT(cdp->i_d.di_nlink >= 2);
3131         if (cdp->i_d.di_nlink != 2) {
3132                 error = XFS_ERROR(ENOTEMPTY);
3133                 goto error_return;
3134         }
3135         if (!XFS_DIR_ISEMPTY(mp, cdp)) {
3136                 error = XFS_ERROR(ENOTEMPTY);
3137                 goto error_return;
3138         }
3139
3140         error = XFS_DIR_REMOVENAME(mp, tp, dp, name, namelen, cdp->i_ino,
3141                 &first_block, &free_list, resblks);
3142         if (error) {
3143                 goto error1;
3144         }
3145
3146         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3147
3148         /*
3149          * Bump the in memory generation count on the parent
3150          * directory so that other can know that it has changed.
3151          */
3152         dp->i_gen++;
3153
3154         /*
3155          * Drop the link from cdp's "..".
3156          */
3157         error = xfs_droplink(tp, dp);
3158         if (error) {
3159                 goto error1;
3160         }
3161
3162         /*
3163          * Drop the link from dp to cdp.
3164          */
3165         error = xfs_droplink(tp, cdp);
3166         if (error) {
3167                 goto error1;
3168         }
3169
3170         /*
3171          * Drop the "." link from cdp to self.
3172          */
3173         error = xfs_droplink(tp, cdp);
3174         if (error) {
3175                 goto error1;
3176         }
3177
3178         /* Determine these before committing transaction */
3179         last_cdp_link = (cdp)->i_d.di_nlink==0;
3180
3181         /*
3182          * Take an extra ref on the child vnode so that it
3183          * does not go to xfs_inactive() from within the commit.
3184          */
3185         IHOLD(cdp);
3186
3187         /*
3188          * If this is a synchronous mount, make sure that the
3189          * rmdir transaction goes to disk before returning to
3190          * the user.
3191          */
3192         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3193                 xfs_trans_set_sync(tp);
3194         }
3195
3196         error = xfs_bmap_finish (&tp, &free_list, first_block, &committed);
3197         if (error) {
3198                 xfs_bmap_cancel(&free_list);
3199                 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES |
3200                                  XFS_TRANS_ABORT));
3201                 IRELE(cdp);
3202                 goto std_return;
3203         }
3204
3205         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3206         if (error) {
3207                 IRELE(cdp);
3208                 goto std_return;
3209         }
3210
3211
3212         /*
3213          * Let interposed file systems know about removed links.
3214          */
3215         VOP_LINK_REMOVED(XFS_ITOV(cdp), dir_vp, last_cdp_link);
3216
3217         IRELE(cdp);
3218
3219         /* Fall through to std_return with error = 0 or the errno
3220          * from xfs_trans_commit. */
3221 std_return:
3222         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_POSTREMOVE)) {
3223                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTREMOVE,
3224                                         dir_vp, DM_RIGHT_NULL,
3225                                         NULL, DM_RIGHT_NULL,
3226                                         name, NULL, dm_di_mode,
3227                                         error, 0);
3228         }
3229         return error;
3230
3231  error1:
3232         xfs_bmap_cancel(&free_list);
3233         cancel_flags |= XFS_TRANS_ABORT;
3234  error_return:
3235         xfs_trans_cancel(tp, cancel_flags);
3236         goto std_return;
3237 }
3238
3239
3240 /*
3241  * xfs_readdir
3242  *
3243  * Read dp's entries starting at uiop->uio_offset and translate them into
3244  * bufsize bytes worth of struct dirents starting at bufbase.
3245  */
3246 STATIC int
3247 xfs_readdir(
3248         bhv_desc_t      *dir_bdp,
3249         uio_t           *uiop,
3250         cred_t          *credp,
3251         int             *eofp)
3252 {
3253         xfs_inode_t     *dp;
3254         xfs_trans_t     *tp = NULL;
3255         int             error = 0;
3256         uint            lock_mode;
3257         xfs_off_t       start_offset;
3258
3259         vn_trace_entry(BHV_TO_VNODE(dir_bdp), __FUNCTION__,
3260                                                (inst_t *)__return_address);
3261         dp = XFS_BHVTOI(dir_bdp);
3262
3263         if (XFS_FORCED_SHUTDOWN(dp->i_mount)) {
3264                 return XFS_ERROR(EIO);
3265         }
3266
3267         lock_mode = xfs_ilock_map_shared(dp);
3268         start_offset = uiop->uio_offset;
3269         error = XFS_DIR_GETDENTS(dp->i_mount, tp, dp, uiop, eofp);
3270         if (start_offset != uiop->uio_offset) {
3271                 xfs_ichgtime(dp, XFS_ICHGTIME_ACC);
3272         }
3273         xfs_iunlock_map_shared(dp, lock_mode);
3274         return error;
3275 }
3276
3277
3278 /*
3279  * xfs_symlink
3280  *
3281  */
3282 STATIC int
3283 xfs_symlink(
3284         bhv_desc_t              *dir_bdp,
3285         vname_t                 *dentry,
3286         vattr_t                 *vap,
3287         char                    *target_path,
3288         vnode_t                 **vpp,
3289         cred_t                  *credp)
3290 {
3291         xfs_trans_t             *tp;
3292         xfs_mount_t             *mp;
3293         xfs_inode_t             *dp;
3294         xfs_inode_t             *ip;
3295         int                     error;
3296         int                     pathlen;
3297         xfs_bmap_free_t         free_list;
3298         xfs_fsblock_t           first_block;
3299         boolean_t               dp_joined_to_trans;
3300         vnode_t                 *dir_vp;
3301         uint                    cancel_flags;
3302         int                     committed;
3303         xfs_fileoff_t           first_fsb;
3304         xfs_filblks_t           fs_blocks;
3305         int                     nmaps;
3306         xfs_bmbt_irec_t         mval[SYMLINK_MAPS];
3307         xfs_daddr_t             d;
3308         char                    *cur_chunk;
3309         int                     byte_cnt;
3310         int                     n;
3311         xfs_buf_t               *bp;
3312         xfs_prid_t              prid;
3313         struct xfs_dquot        *udqp, *gdqp;
3314         uint                    resblks;
3315         char                    *link_name = VNAME(dentry);
3316         int                     link_namelen;
3317
3318         *vpp = NULL;
3319         dir_vp = BHV_TO_VNODE(dir_bdp);
3320         dp = XFS_BHVTOI(dir_bdp);
3321         dp_joined_to_trans = B_FALSE;
3322         error = 0;
3323         ip = NULL;
3324         tp = NULL;
3325
3326         vn_trace_entry(dir_vp, __FUNCTION__, (inst_t *)__return_address);
3327
3328         mp = dp->i_mount;
3329
3330         if (XFS_FORCED_SHUTDOWN(mp))
3331                 return XFS_ERROR(EIO);
3332
3333         link_namelen = VNAMELEN(dentry);
3334
3335         /*
3336          * Check component lengths of the target path name.
3337          */
3338         pathlen = strlen(target_path);
3339         if (pathlen >= MAXPATHLEN)      /* total string too long */
3340                 return XFS_ERROR(ENAMETOOLONG);
3341         if (pathlen >= MAXNAMELEN) {    /* is any component too long? */
3342                 int len, total;
3343                 char *path;
3344
3345                 for(total = 0, path = target_path; total < pathlen;) {
3346                         /*
3347                          * Skip any slashes.
3348                          */
3349                         while(*path == '/') {
3350                                 total++;
3351                                 path++;
3352                         }
3353
3354                         /*
3355                          * Count up to the next slash or end of path.
3356                          * Error out if the component is bigger than MAXNAMELEN.
3357                          */
3358                         for(len = 0; *path != '/' && total < pathlen;total++, path++) {
3359                                 if (++len >= MAXNAMELEN) {
3360                                         error = ENAMETOOLONG;
3361                                         return error;
3362                                 }
3363                         }
3364                 }
3365         }
3366
3367         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, dp, DM_EVENT_SYMLINK)) {
3368                 error = XFS_SEND_NAMESP(mp, DM_EVENT_SYMLINK, dir_vp,
3369                                         DM_RIGHT_NULL, NULL, DM_RIGHT_NULL,
3370                                         link_name, target_path, 0, 0, 0);
3371                 if (error)
3372                         return error;
3373         }
3374
3375         /* Return through std_return after this point. */
3376
3377         udqp = gdqp = NULL;
3378         if (vap->va_mask & XFS_AT_PROJID)
3379                 prid = (xfs_prid_t)vap->va_projid;
3380         else
3381                 prid = (xfs_prid_t)dfltprid;
3382
3383         /*
3384          * Make sure that we have allocated dquot(s) on disk.
3385          */
3386         error = XFS_QM_DQVOPALLOC(mp, dp,
3387                         current_fsuid(credp), current_fsgid(credp),
3388                         XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, &udqp, &gdqp);
3389         if (error)
3390                 goto std_return;
3391
3392         tp = xfs_trans_alloc(mp, XFS_TRANS_SYMLINK);
3393         cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
3394         /*
3395          * The symlink will fit into the inode data fork?
3396          * There can't be any attributes so we get the whole variable part.
3397          */
3398         if (pathlen <= XFS_LITINO(mp))
3399                 fs_blocks = 0;
3400         else
3401                 fs_blocks = XFS_B_TO_FSB(mp, pathlen);
3402         resblks = XFS_SYMLINK_SPACE_RES(mp, link_namelen, fs_blocks);
3403         error = xfs_trans_reserve(tp, resblks, XFS_SYMLINK_LOG_RES(mp), 0,
3404                         XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3405         if (error == ENOSPC && fs_blocks == 0) {
3406                 resblks = 0;
3407                 error = xfs_trans_reserve(tp, 0, XFS_SYMLINK_LOG_RES(mp), 0,
3408                                 XFS_TRANS_PERM_LOG_RES, XFS_SYMLINK_LOG_COUNT);
3409         }
3410         if (error) {
3411                 cancel_flags = 0;
3412                 dp = NULL;
3413                 goto error_return;
3414         }
3415
3416         xfs_ilock(dp, XFS_ILOCK_EXCL);
3417
3418         /*
3419          * Check whether the directory allows new symlinks or not.
3420          */
3421         if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
3422                 error = XFS_ERROR(EPERM);
3423                 goto error_return;
3424         }
3425
3426         /*
3427          * Reserve disk quota : blocks and inode.
3428          */
3429         error = XFS_TRANS_RESERVE_QUOTA(mp, tp, udqp, gdqp, resblks, 1, 0);
3430         if (error)
3431                 goto error_return;
3432
3433         /*
3434          * Check for ability to enter directory entry, if no space reserved.
3435          */
3436         if (resblks == 0 &&
3437             (error = XFS_DIR_CANENTER(mp, tp, dp, link_name, link_namelen)))
3438                 goto error_return;
3439         /*
3440          * Initialize the bmap freelist prior to calling either
3441          * bmapi or the directory create code.
3442          */
3443         XFS_BMAP_INIT(&free_list, &first_block);
3444
3445         /*
3446          * Allocate an inode for the symlink.
3447          */
3448         error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (vap->va_mode&~S_IFMT),
3449                                1, 0, credp, prid, resblks > 0, &ip, NULL);
3450         if (error) {
3451                 if (error == ENOSPC)
3452                         goto error_return;
3453                 goto error1;
3454         }
3455         ITRACE(ip);
3456
3457         VN_HOLD(dir_vp);
3458         xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
3459         dp_joined_to_trans = B_TRUE;
3460
3461         /*
3462          * Also attach the dquot(s) to it, if applicable.
3463          */
3464         XFS_QM_DQVOPCREATE(mp, tp, ip, udqp, gdqp);
3465
3466         if (resblks)
3467                 resblks -= XFS_IALLOC_SPACE_RES(mp);
3468         /*
3469          * If the symlink will fit into the inode, write it inline.
3470          */
3471         if (pathlen <= XFS_IFORK_DSIZE(ip)) {
3472                 xfs_idata_realloc(ip, pathlen, XFS_DATA_FORK);
3473                 memcpy(ip->i_df.if_u1.if_data, target_path, pathlen);
3474                 ip->i_d.di_size = pathlen;
3475
3476                 /*
3477                  * The inode was initially created in extent format.
3478                  */
3479                 ip->i_df.if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
3480                 ip->i_df.if_flags |= XFS_IFINLINE;
3481
3482                 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL;
3483                 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
3484
3485         } else {
3486                 first_fsb = 0;
3487                 nmaps = SYMLINK_MAPS;
3488
3489                 error = xfs_bmapi(tp, ip, first_fsb, fs_blocks,
3490                                   XFS_BMAPI_WRITE | XFS_BMAPI_METADATA,
3491                                   &first_block, resblks, mval, &nmaps,
3492                                   &free_list);
3493                 if (error) {
3494                         goto error1;
3495                 }
3496
3497                 if (resblks)
3498                         resblks -= fs_blocks;
3499                 ip->i_d.di_size = pathlen;
3500                 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3501
3502                 cur_chunk = target_path;
3503                 for (n = 0; n < nmaps; n++) {
3504                         d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock);
3505                         byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount);
3506                         bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
3507                                                BTOBB(byte_cnt), 0);
3508                         ASSERT(bp && !XFS_BUF_GETERROR(bp));
3509                         if (pathlen < byte_cnt) {
3510                                 byte_cnt = pathlen;
3511                         }
3512                         pathlen -= byte_cnt;
3513
3514                         memcpy(XFS_BUF_PTR(bp), cur_chunk, byte_cnt);
3515                         cur_chunk += byte_cnt;
3516
3517                         xfs_trans_log_buf(tp, bp, 0, byte_cnt - 1);
3518                 }
3519         }
3520
3521         /*
3522          * Create the directory entry for the symlink.
3523          */
3524         error = XFS_DIR_CREATENAME(mp, tp, dp, link_name, link_namelen,
3525                         ip->i_ino, &first_block, &free_list, resblks);
3526         if (error) {
3527                 goto error1;
3528         }
3529         xfs_ichgtime(dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
3530         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
3531
3532         /*
3533          * Bump the in memory version number of the parent directory
3534          * so that other processes accessing it will recognize that
3535          * the directory has changed.
3536          */
3537         dp->i_gen++;
3538
3539         /*
3540          * If this is a synchronous mount, make sure that the
3541          * symlink transaction goes to disk before returning to
3542          * the user.
3543          */
3544         if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) {
3545                 xfs_trans_set_sync(tp);
3546         }
3547
3548         /*
3549          * xfs_trans_commit normally decrements the vnode ref count
3550          * when it unlocks the inode. Since we want to return the
3551          * vnode to the caller, we bump the vnode ref count now.
3552          */
3553         IHOLD(ip);
3554
3555         error = xfs_bmap_finish(&tp, &free_list, first_block, &committed);
3556         if (error) {
3557                 goto error2;
3558         }
3559         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
3560         XFS_QM_DQRELE(mp, udqp);
3561         XFS_QM_DQRELE(mp, gdqp);
3562
3563         /* Fall through to std_return with error = 0 or errno from
3564          * xfs_trans_commit     */
3565 std_return:
3566         if (DM_EVENT_ENABLED(dir_vp->v_vfsp, XFS_BHVTOI(dir_bdp),
3567                              DM_EVENT_POSTSYMLINK)) {
3568                 (void) XFS_SEND_NAMESP(mp, DM_EVENT_POSTSYMLINK,
3569                                         dir_vp, DM_RIGHT_NULL,
3570                                         error ? NULL : XFS_ITOV(ip),
3571                                         DM_RIGHT_NULL, link_name, target_path,
3572                                         0, error, 0);
3573         }
3574
3575         if (!error) {
3576                 vnode_t *vp;
3577
3578                 ASSERT(ip);
3579                 vp = XFS_ITOV(ip);
3580                 *vpp = vp;
3581         }
3582         return error;
3583
3584  error2:
3585         IRELE(ip);
3586  error1:
3587         xfs_bmap_cancel(&free_list);
3588         cancel_flags |= XFS_TRANS_ABORT;
3589  error_return:
3590         xfs_trans_cancel(tp, cancel_flags);
3591         XFS_QM_DQRELE(mp, udqp);
3592         XFS_QM_DQRELE(mp, gdqp);
3593
3594         if (!dp_joined_to_trans && (dp != NULL)) {
3595                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
3596         }
3597
3598         goto std_return;
3599 }
3600
3601
3602 /*
3603  * xfs_fid2
3604  *
3605  * A fid routine that takes a pointer to a previously allocated
3606  * fid structure (like xfs_fast_fid) but uses a 64 bit inode number.
3607  */
3608 STATIC int
3609 xfs_fid2(
3610         bhv_desc_t      *bdp,
3611         fid_t           *fidp)
3612 {
3613         xfs_inode_t     *ip;
3614         xfs_fid2_t      *xfid;
3615
3616         vn_trace_entry(BHV_TO_VNODE(bdp), __FUNCTION__,
3617                                        (inst_t *)__return_address);
3618         ASSERT(sizeof(fid_t) >= sizeof(xfs_fid2_t));
3619
3620         xfid = (xfs_fid2_t *)fidp;
3621         ip = XFS_BHVTOI(bdp);
3622         xfid->fid_len = sizeof(xfs_fid2_t) - sizeof(xfid->fid_len);
3623         xfid->fid_pad = 0;
3624         /*
3625          * use memcpy because the inode is a long long and there's no
3626          * assurance that xfid->fid_ino is properly aligned.
3627          */
3628         memcpy(&xfid->fid_ino, &ip->i_ino, sizeof(xfid->fid_ino));
3629         xfid->fid_gen = ip->i_d.di_gen;
3630
3631         return 0;
3632 }
3633
3634
3635 /*
3636  * xfs_rwlock
3637  */
3638 int
3639 xfs_rwlock(
3640         bhv_desc_t      *bdp,
3641         vrwlock_t       locktype)
3642 {
3643         xfs_inode_t     *ip;
3644         vnode_t         *vp;
3645
3646         vp = BHV_TO_VNODE(bdp);
3647         if (vp->v_type == VDIR)
3648                 return 1;
3649         ip = XFS_BHVTOI(bdp);
3650         if (locktype == VRWLOCK_WRITE) {
3651                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
3652         } else if (locktype == VRWLOCK_TRY_READ) {
3653                 return (xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED));
3654         } else if (locktype == VRWLOCK_TRY_WRITE) {
3655                 return (xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL));
3656         } else {
3657                 ASSERT((locktype == VRWLOCK_READ) ||
3658                        (locktype == VRWLOCK_WRITE_DIRECT));
3659                 xfs_ilock(ip, XFS_IOLOCK_SHARED);
3660         }
3661
3662         return 1;
3663 }
3664
3665
3666 /*
3667  * xfs_rwunlock
3668  */
3669 void
3670 xfs_rwunlock(
3671         bhv_desc_t      *bdp,
3672         vrwlock_t       locktype)
3673 {
3674         xfs_inode_t     *ip;
3675         vnode_t         *vp;
3676
3677         vp = BHV_TO_VNODE(bdp);
3678         if (vp->v_type == VDIR)
3679                 return;
3680         ip = XFS_BHVTOI(bdp);
3681         if (locktype == VRWLOCK_WRITE) {
3682                 /*
3683                  * In the write case, we may have added a new entry to
3684                  * the reference cache.  This might store a pointer to
3685                  * an inode to be released in this inode.  If it is there,
3686                  * clear the pointer and release the inode after unlocking
3687                  * this one.
3688                  */
3689                 xfs_refcache_iunlock(ip, XFS_IOLOCK_EXCL);
3690         } else {
3691                 ASSERT((locktype == VRWLOCK_READ) ||
3692                        (locktype == VRWLOCK_WRITE_DIRECT));
3693                 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
3694         }
3695         return;
3696 }
3697
3698 STATIC int
3699 xfs_inode_flush(
3700         bhv_desc_t      *bdp,
3701         int             flags)
3702 {
3703         xfs_inode_t     *ip;
3704         xfs_mount_t     *mp;
3705         xfs_inode_log_item_t *iip;
3706         int             error = 0;
3707
3708         ip = XFS_BHVTOI(bdp);
3709         mp = ip->i_mount;
3710         iip = ip->i_itemp;
3711
3712         if (XFS_FORCED_SHUTDOWN(mp))
3713                 return XFS_ERROR(EIO);
3714
3715         /*
3716          * Bypass inodes which have already been cleaned by
3717          * the inode flush clustering code inside xfs_iflush
3718          */
3719         if ((ip->i_update_core == 0) &&
3720             ((iip == NULL) || !(iip->ili_format.ilf_fields & XFS_ILOG_ALL)))
3721                 return 0;
3722
3723         if (flags & FLUSH_LOG) {
3724                 if (iip && iip->ili_last_lsn) {
3725                         xlog_t          *log = mp->m_log;
3726                         xfs_lsn_t       sync_lsn;
3727                         int             s, log_flags = XFS_LOG_FORCE;
3728
3729                         s = GRANT_LOCK(log);
3730                         sync_lsn = log->l_last_sync_lsn;
3731                         GRANT_UNLOCK(log, s);
3732
3733                         if ((XFS_LSN_CMP(iip->ili_last_lsn, sync_lsn) <= 0))
3734                                 return 0;
3735
3736                         if (flags & FLUSH_SYNC)
3737                                 log_flags |= XFS_LOG_SYNC;
3738                         return xfs_log_force(mp, iip->ili_last_lsn, log_flags);
3739                 }
3740         }
3741
3742         /*
3743          * We make this non-blocking if the inode is contended,
3744          * return EAGAIN to indicate to the caller that they
3745          * did not succeed. This prevents the flush path from
3746          * blocking on inodes inside another operation right
3747          * now, they get caught later by xfs_sync.
3748          */
3749         if (flags & FLUSH_INODE) {
3750                 int     flush_flags;
3751
3752                 if (xfs_ipincount(ip))
3753                         return EAGAIN;
3754
3755                 if (flags & FLUSH_SYNC) {
3756                         xfs_ilock(ip, XFS_ILOCK_SHARED);
3757                         xfs_iflock(ip);
3758                 } else if (xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
3759                         if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip)) {
3760                                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3761                                 return EAGAIN;
3762                         }
3763                 } else {
3764                         return EAGAIN;
3765                 }
3766
3767                 if (flags & FLUSH_SYNC)
3768                         flush_flags = XFS_IFLUSH_SYNC;
3769                 else
3770                         flush_flags = XFS_IFLUSH_ASYNC;
3771
3772                 error = xfs_iflush(ip, flush_flags);
3773                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
3774         }
3775
3776         return error;
3777 }
3778
3779
3780 int
3781 xfs_set_dmattrs (
3782         bhv_desc_t      *bdp,
3783         u_int           evmask,
3784         u_int16_t       state,
3785         cred_t          *credp)
3786 {
3787         xfs_inode_t     *ip;
3788         xfs_trans_t     *tp;
3789         xfs_mount_t     *mp;
3790         int             error;
3791
3792         if (!capable(CAP_SYS_ADMIN))
3793                 return XFS_ERROR(EPERM);
3794
3795         ip = XFS_BHVTOI(bdp);
3796         mp = ip->i_mount;
3797
3798         if (XFS_FORCED_SHUTDOWN(mp))
3799                 return XFS_ERROR(EIO);
3800
3801         tp = xfs_trans_alloc(mp, XFS_TRANS_SET_DMATTRS);
3802         error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES (mp), 0, 0, 0);
3803         if (error) {
3804                 xfs_trans_cancel(tp, 0);
3805                 return error;
3806         }
3807         xfs_ilock(ip, XFS_ILOCK_EXCL);
3808         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3809
3810         ip->i_iocore.io_dmevmask = ip->i_d.di_dmevmask = evmask;
3811         ip->i_iocore.io_dmstate  = ip->i_d.di_dmstate  = state;
3812
3813         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3814         IHOLD(ip);
3815         error = xfs_trans_commit(tp, 0, NULL);
3816
3817         return error;
3818 }
3819
3820
3821 /*
3822  * xfs_reclaim
3823  */
3824 STATIC int
3825 xfs_reclaim(
3826         bhv_desc_t      *bdp)
3827 {
3828         xfs_inode_t     *ip;
3829         vnode_t         *vp;
3830
3831         vp = BHV_TO_VNODE(bdp);
3832         ip = XFS_BHVTOI(bdp);
3833
3834         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
3835
3836         ASSERT(!VN_MAPPED(vp));
3837
3838         /* bad inode, get out here ASAP */
3839         if (VN_BAD(vp)) {
3840                 xfs_ireclaim(ip);
3841                 return 0;
3842         }
3843
3844         if ((ip->i_d.di_mode & S_IFMT) == S_IFREG) {
3845                 if (ip->i_d.di_size > 0) {
3846                         /*
3847                          * Flush and invalidate any data left around that is
3848                          * a part of this file.
3849                          *
3850                          * Get the inode's i/o lock so that buffers are pushed
3851                          * out while holding the proper lock.  We can't hold
3852                          * the inode lock here since flushing out buffers may
3853                          * cause us to try to get the lock in xfs_strategy().
3854                          *
3855                          * We don't have to call remapf() here, because there
3856                          * cannot be any mapped file references to this vnode
3857                          * since it is being reclaimed.
3858                          */
3859                         xfs_ilock(ip, XFS_IOLOCK_EXCL);
3860
3861                         /*
3862                          * If we hit an IO error, we need to make sure that the
3863                          * buffer and page caches of file data for
3864                          * the file are tossed away. We don't want to use
3865                          * VOP_FLUSHINVAL_PAGES here because we don't want dirty
3866                          * pages to stay attached to the vnode, but be
3867                          * marked P_BAD. pdflush/vnode_pagebad
3868                          * hates that.
3869                          */
3870                         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3871                                 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_NONE);
3872                         } else {
3873                                 VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3874                         }
3875
3876                         ASSERT(VN_CACHED(vp) == 0);
3877                         ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) ||
3878                                ip->i_delayed_blks == 0);
3879                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
3880                 } else if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3881                         /*
3882                          * di_size field may not be quite accurate if we're
3883                          * shutting down.
3884                          */
3885                         VOP_TOSS_PAGES(vp, 0, -1, FI_NONE);
3886                         ASSERT(VN_CACHED(vp) == 0);
3887                 }
3888         }
3889
3890         /* If we have nothing to flush with this inode then complete the
3891          * teardown now, otherwise break the link between the xfs inode
3892          * and the linux inode and clean up the xfs inode later. This
3893          * avoids flushing the inode to disk during the delete operation
3894          * itself.
3895          */
3896         if (!ip->i_update_core && (ip->i_itemp == NULL)) {
3897                 xfs_ilock(ip, XFS_ILOCK_EXCL);
3898                 xfs_iflock(ip);
3899                 return xfs_finish_reclaim(ip, 1, XFS_IFLUSH_DELWRI_ELSE_SYNC);
3900         } else {
3901                 xfs_mount_t     *mp = ip->i_mount;
3902
3903                 /* Protect sync from us */
3904                 XFS_MOUNT_ILOCK(mp);
3905                 vn_bhv_remove(VN_BHV_HEAD(vp), XFS_ITOBHV(ip));
3906                 list_add_tail(&ip->i_reclaim, &mp->m_del_inodes);
3907                 ip->i_flags |= XFS_IRECLAIMABLE;
3908                 XFS_MOUNT_IUNLOCK(mp);
3909         }
3910         return 0;
3911 }
3912
3913 int
3914 xfs_finish_reclaim(
3915         xfs_inode_t     *ip,
3916         int             locked,
3917         int             sync_mode)
3918 {
3919         xfs_ihash_t     *ih = ip->i_hash;
3920         vnode_t         *vp = XFS_ITOV_NULL(ip);
3921         int             error;
3922
3923         if (vp && VN_BAD(vp))
3924                 goto reclaim;
3925
3926         /* The hash lock here protects a thread in xfs_iget_core from
3927          * racing with us on linking the inode back with a vnode.
3928          * Once we have the XFS_IRECLAIM flag set it will not touch
3929          * us.
3930          */
3931         write_lock(&ih->ih_lock);
3932         if ((ip->i_flags & XFS_IRECLAIM) ||
3933             (!(ip->i_flags & XFS_IRECLAIMABLE) && vp == NULL)) {
3934                 write_unlock(&ih->ih_lock);
3935                 if (locked) {
3936                         xfs_ifunlock(ip);
3937                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3938                 }
3939                 return(1);
3940         }
3941         ip->i_flags |= XFS_IRECLAIM;
3942         write_unlock(&ih->ih_lock);
3943
3944         /*
3945          * If the inode is still dirty, then flush it out.  If the inode
3946          * is not in the AIL, then it will be OK to flush it delwri as
3947          * long as xfs_iflush() does not keep any references to the inode.
3948          * We leave that decision up to xfs_iflush() since it has the
3949          * knowledge of whether it's OK to simply do a delwri flush of
3950          * the inode or whether we need to wait until the inode is
3951          * pulled from the AIL.
3952          * We get the flush lock regardless, though, just to make sure
3953          * we don't free it while it is being flushed.
3954          */
3955         if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
3956                 if (!locked) {
3957                         xfs_ilock(ip, XFS_ILOCK_EXCL);
3958                         xfs_iflock(ip);
3959                 }
3960
3961                 if (ip->i_update_core ||
3962                     ((ip->i_itemp != NULL) &&
3963                      (ip->i_itemp->ili_format.ilf_fields != 0))) {
3964                         error = xfs_iflush(ip, sync_mode);
3965                         /*
3966                          * If we hit an error, typically because of filesystem
3967                          * shutdown, we don't need to let vn_reclaim to know
3968                          * because we're gonna reclaim the inode anyway.
3969                          */
3970                         if (error) {
3971                                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3972                                 goto reclaim;
3973                         }
3974                         xfs_iflock(ip); /* synchronize with xfs_iflush_done */
3975                 }
3976
3977                 ASSERT(ip->i_update_core == 0);
3978                 ASSERT(ip->i_itemp == NULL ||
3979                        ip->i_itemp->ili_format.ilf_fields == 0);
3980                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3981         } else if (locked) {
3982                 /*
3983                  * We are not interested in doing an iflush if we're
3984                  * in the process of shutting down the filesystem forcibly.
3985                  * So, just reclaim the inode.
3986                  */
3987                 xfs_ifunlock(ip);
3988                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3989         }
3990
3991  reclaim:
3992         xfs_ireclaim(ip);
3993         return 0;
3994 }
3995
3996 int
3997 xfs_finish_reclaim_all(xfs_mount_t *mp, int noblock)
3998 {
3999         int             purged;
4000         xfs_inode_t     *ip, *n;
4001         int             done = 0;
4002
4003         while (!done) {
4004                 purged = 0;
4005                 XFS_MOUNT_ILOCK(mp);
4006                 list_for_each_entry_safe(ip, n, &mp->m_del_inodes, i_reclaim) {
4007                         if (noblock) {
4008                                 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0)
4009                                         continue;
4010                                 if (xfs_ipincount(ip) ||
4011                                     !xfs_iflock_nowait(ip)) {
4012                                         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4013                                         continue;
4014                                 }
4015                         }
4016                         XFS_MOUNT_IUNLOCK(mp);
4017                         xfs_finish_reclaim(ip, noblock,
4018                                 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
4019                         purged = 1;
4020                         break;
4021                 }
4022
4023                 done = !purged;
4024         }
4025
4026         XFS_MOUNT_IUNLOCK(mp);
4027         return 0;
4028 }
4029
4030 /*
4031  * xfs_alloc_file_space()
4032  *      This routine allocates disk space for the given file.
4033  *
4034  *      If alloc_type == 0, this request is for an ALLOCSP type
4035  *      request which will change the file size.  In this case, no
4036  *      DMAPI event will be generated by the call.  A TRUNCATE event
4037  *      will be generated later by xfs_setattr.
4038  *
4039  *      If alloc_type != 0, this request is for a RESVSP type
4040  *      request, and a DMAPI DM_EVENT_WRITE will be generated if the
4041  *      lower block boundary byte address is less than the file's
4042  *      length.
4043  *
4044  * RETURNS:
4045  *       0 on success
4046  *      errno on error
4047  *
4048  */
4049 int
4050 xfs_alloc_file_space(
4051         xfs_inode_t             *ip,
4052         xfs_off_t               offset,
4053         xfs_off_t               len,
4054         int                     alloc_type,
4055         int                     attr_flags)
4056 {
4057         xfs_filblks_t           allocated_fsb;
4058         xfs_filblks_t           allocatesize_fsb;
4059         int                     committed;
4060         xfs_off_t               count;
4061         xfs_filblks_t           datablocks;
4062         int                     error;
4063         xfs_fsblock_t           firstfsb;
4064         xfs_bmap_free_t         free_list;
4065         xfs_bmbt_irec_t         *imapp;
4066         xfs_bmbt_irec_t         imaps[1];
4067         xfs_mount_t             *mp;
4068         int                     numrtextents;
4069         int                     reccount;
4070         uint                    resblks;
4071         int                     rt;
4072         int                     rtextsize;
4073         xfs_fileoff_t           startoffset_fsb;
4074         xfs_trans_t             *tp;
4075         int                     xfs_bmapi_flags;
4076
4077         vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4078         mp = ip->i_mount;
4079
4080         if (XFS_FORCED_SHUTDOWN(mp))
4081                 return XFS_ERROR(EIO);
4082
4083         /*
4084          * determine if this is a realtime file
4085          */
4086         if ((rt = XFS_IS_REALTIME_INODE(ip)) != 0) {
4087                 if (ip->i_d.di_extsize)
4088                         rtextsize = ip->i_d.di_extsize;
4089                 else
4090                         rtextsize = mp->m_sb.sb_rextsize;
4091         } else
4092                 rtextsize = 0;
4093
4094         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4095                 return error;
4096
4097         if (len <= 0)
4098                 return XFS_ERROR(EINVAL);
4099
4100         count = len;
4101         error = 0;
4102         imapp = &imaps[0];
4103         reccount = 1;
4104         xfs_bmapi_flags = XFS_BMAPI_WRITE | (alloc_type ? XFS_BMAPI_PREALLOC : 0);
4105         startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
4106         allocatesize_fsb = XFS_B_TO_FSB(mp, count);
4107
4108         /*      Generate a DMAPI event if needed.       */
4109         if (alloc_type != 0 && offset < ip->i_d.di_size &&
4110                         (attr_flags&ATTR_DMI) == 0  &&
4111                         DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4112                 xfs_off_t           end_dmi_offset;
4113
4114                 end_dmi_offset = offset+len;
4115                 if (end_dmi_offset > ip->i_d.di_size)
4116                         end_dmi_offset = ip->i_d.di_size;
4117                 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4118                         offset, end_dmi_offset - offset,
4119                         0, NULL);
4120                 if (error)
4121                         return(error);
4122         }
4123
4124         /*
4125          * allocate file space until done or until there is an error
4126          */
4127 retry:
4128         while (allocatesize_fsb && !error) {
4129                 /*
4130                  * determine if reserving space on
4131                  * the data or realtime partition.
4132                  */
4133                 if (rt) {
4134                         xfs_fileoff_t s, e;
4135
4136                         s = startoffset_fsb;
4137                         do_div(s, rtextsize);
4138                         s *= rtextsize;
4139                         e = roundup_64(startoffset_fsb + allocatesize_fsb,
4140                                 rtextsize);
4141                         numrtextents = (int)(e - s) / mp->m_sb.sb_rextsize;
4142                         datablocks = 0;
4143                 } else {
4144                         datablocks = allocatesize_fsb;
4145                         numrtextents = 0;
4146                 }
4147
4148                 /*
4149                  * allocate and setup the transaction
4150                  */
4151                 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4152                 resblks = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
4153                 error = xfs_trans_reserve(tp,
4154                                           resblks,
4155                                           XFS_WRITE_LOG_RES(mp),
4156                                           numrtextents,
4157                                           XFS_TRANS_PERM_LOG_RES,
4158                                           XFS_WRITE_LOG_COUNT);
4159
4160                 /*
4161                  * check for running out of space
4162                  */
4163                 if (error) {
4164                         /*
4165                          * Free the transaction structure.
4166                          */
4167                         ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4168                         xfs_trans_cancel(tp, 0);
4169                         break;
4170                 }
4171                 xfs_ilock(ip, XFS_ILOCK_EXCL);
4172                 error = XFS_TRANS_RESERVE_QUOTA_BYDQUOTS(mp, tp,
4173                                 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4174                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4175                 if (error)
4176                         goto error1;
4177
4178                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4179                 xfs_trans_ihold(tp, ip);
4180
4181                 /*
4182                  * issue the bmapi() call to allocate the blocks
4183                  */
4184                 XFS_BMAP_INIT(&free_list, &firstfsb);
4185                 error = xfs_bmapi(tp, ip, startoffset_fsb,
4186                                   allocatesize_fsb, xfs_bmapi_flags,
4187                                   &firstfsb, 0, imapp, &reccount,
4188                                   &free_list);
4189                 if (error) {
4190                         goto error0;
4191                 }
4192
4193                 /*
4194                  * complete the transaction
4195                  */
4196                 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4197                 if (error) {
4198                         goto error0;
4199                 }
4200
4201                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4202                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4203                 if (error) {
4204                         break;
4205                 }
4206
4207                 allocated_fsb = imapp->br_blockcount;
4208
4209                 if (reccount == 0) {
4210                         error = XFS_ERROR(ENOSPC);
4211                         break;
4212                 }
4213
4214                 startoffset_fsb += allocated_fsb;
4215                 allocatesize_fsb -= allocated_fsb;
4216         }
4217 dmapi_enospc_check:
4218         if (error == ENOSPC && (attr_flags&ATTR_DMI) == 0 &&
4219             DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_NOSPACE)) {
4220
4221                 error = XFS_SEND_NAMESP(mp, DM_EVENT_NOSPACE,
4222                                 XFS_ITOV(ip), DM_RIGHT_NULL,
4223                                 XFS_ITOV(ip), DM_RIGHT_NULL,
4224                                 NULL, NULL, 0, 0, 0); /* Delay flag intentionally unused */
4225                 if (error == 0)
4226                         goto retry;     /* Maybe DMAPI app. has made space */
4227                 /* else fall through with error from XFS_SEND_DATA */
4228         }
4229
4230         return error;
4231
4232  error0:
4233         xfs_bmap_cancel(&free_list);
4234  error1:
4235         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4236         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4237         goto dmapi_enospc_check;
4238 }
4239
4240 /*
4241  * Zero file bytes between startoff and endoff inclusive.
4242  * The iolock is held exclusive and no blocks are buffered.
4243  */
4244 STATIC int
4245 xfs_zero_remaining_bytes(
4246         xfs_inode_t             *ip,
4247         xfs_off_t               startoff,
4248         xfs_off_t               endoff)
4249 {
4250         xfs_bmbt_irec_t         imap;
4251         xfs_fileoff_t           offset_fsb;
4252         xfs_off_t               lastoffset;
4253         xfs_off_t               offset;
4254         xfs_buf_t               *bp;
4255         xfs_mount_t             *mp = ip->i_mount;
4256         int                     nimap;
4257         int                     error = 0;
4258
4259         bp = xfs_buf_get_noaddr(mp->m_sb.sb_blocksize,
4260                                 ip->i_d.di_flags & XFS_DIFLAG_REALTIME ?
4261                                 mp->m_rtdev_targp : mp->m_ddev_targp);
4262
4263         for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
4264                 offset_fsb = XFS_B_TO_FSBT(mp, offset);
4265                 nimap = 1;
4266                 error = xfs_bmapi(NULL, ip, offset_fsb, 1, 0, NULL, 0, &imap,
4267                         &nimap, NULL);
4268                 if (error || nimap < 1)
4269                         break;
4270                 ASSERT(imap.br_blockcount >= 1);
4271                 ASSERT(imap.br_startoff == offset_fsb);
4272                 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
4273                 if (lastoffset > endoff)
4274                         lastoffset = endoff;
4275                 if (imap.br_startblock == HOLESTARTBLOCK)
4276                         continue;
4277                 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4278                 if (imap.br_state == XFS_EXT_UNWRITTEN)
4279                         continue;
4280                 XFS_BUF_UNDONE(bp);
4281                 XFS_BUF_UNWRITE(bp);
4282                 XFS_BUF_READ(bp);
4283                 XFS_BUF_SET_ADDR(bp, XFS_FSB_TO_DB(ip, imap.br_startblock));
4284                 xfsbdstrat(mp, bp);
4285                 if ((error = xfs_iowait(bp))) {
4286                         xfs_ioerror_alert("xfs_zero_remaining_bytes(read)",
4287                                           mp, bp, XFS_BUF_ADDR(bp));
4288                         break;
4289                 }
4290                 memset(XFS_BUF_PTR(bp) +
4291                         (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
4292                       0, lastoffset - offset + 1);
4293                 XFS_BUF_UNDONE(bp);
4294                 XFS_BUF_UNREAD(bp);
4295                 XFS_BUF_WRITE(bp);
4296                 xfsbdstrat(mp, bp);
4297                 if ((error = xfs_iowait(bp))) {
4298                         xfs_ioerror_alert("xfs_zero_remaining_bytes(write)",
4299                                           mp, bp, XFS_BUF_ADDR(bp));
4300                         break;
4301                 }
4302         }
4303         xfs_buf_free(bp);
4304         return error;
4305 }
4306
4307 /*
4308  * xfs_free_file_space()
4309  *      This routine frees disk space for the given file.
4310  *
4311  *      This routine is only called by xfs_change_file_space
4312  *      for an UNRESVSP type call.
4313  *
4314  * RETURNS:
4315  *       0 on success
4316  *      errno on error
4317  *
4318  */
4319 STATIC int
4320 xfs_free_file_space(
4321         xfs_inode_t             *ip,
4322         xfs_off_t               offset,
4323         xfs_off_t               len,
4324         int                     attr_flags)
4325 {
4326         int                     committed;
4327         int                     done;
4328         xfs_off_t               end_dmi_offset;
4329         xfs_fileoff_t           endoffset_fsb;
4330         int                     error;
4331         xfs_fsblock_t           firstfsb;
4332         xfs_bmap_free_t         free_list;
4333         xfs_off_t               ilen;
4334         xfs_bmbt_irec_t         imap;
4335         xfs_off_t               ioffset;
4336         xfs_extlen_t            mod=0;
4337         xfs_mount_t             *mp;
4338         int                     nimap;
4339         uint                    resblks;
4340         int                     rounding;
4341         int                     rt;
4342         xfs_fileoff_t           startoffset_fsb;
4343         xfs_trans_t             *tp;
4344         int                     need_iolock = 1;
4345
4346         vn_trace_entry(XFS_ITOV(ip), __FUNCTION__, (inst_t *)__return_address);
4347         mp = ip->i_mount;
4348
4349         if ((error = XFS_QM_DQATTACH(mp, ip, 0)))
4350                 return error;
4351
4352         error = 0;
4353         if (len <= 0)   /* if nothing being freed */
4354                 return error;
4355         rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
4356         startoffset_fsb = XFS_B_TO_FSB(mp, offset);
4357         end_dmi_offset = offset + len;
4358         endoffset_fsb = XFS_B_TO_FSBT(mp, end_dmi_offset);
4359
4360         if (offset < ip->i_d.di_size &&
4361             (attr_flags & ATTR_DMI) == 0 &&
4362             DM_EVENT_ENABLED(XFS_MTOVFS(mp), ip, DM_EVENT_WRITE)) {
4363                 if (end_dmi_offset > ip->i_d.di_size)
4364                         end_dmi_offset = ip->i_d.di_size;
4365                 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, XFS_ITOV(ip),
4366                                 offset, end_dmi_offset - offset,
4367                                 AT_DELAY_FLAG(attr_flags), NULL);
4368                 if (error)
4369                         return(error);
4370         }
4371
4372         ASSERT(attr_flags & ATTR_NOLOCK ? attr_flags & ATTR_DMI : 1);
4373         if (attr_flags & ATTR_NOLOCK)
4374                 need_iolock = 0;
4375         if (need_iolock)
4376                 xfs_ilock(ip, XFS_IOLOCK_EXCL);
4377
4378         rounding = MAX((__uint8_t)(1 << mp->m_sb.sb_blocklog),
4379                         (__uint8_t)NBPP);
4380         ilen = len + (offset & (rounding - 1));
4381         ioffset = offset & ~(rounding - 1);
4382         if (ilen & (rounding - 1))
4383                 ilen = (ilen + rounding) & ~(rounding - 1);
4384         xfs_inval_cached_pages(XFS_ITOV(ip), &(ip->i_iocore), ioffset, 0, 0);
4385         /*
4386          * Need to zero the stuff we're not freeing, on disk.
4387          * If its a realtime file & can't use unwritten extents then we
4388          * actually need to zero the extent edges.  Otherwise xfs_bunmapi
4389          * will take care of it for us.
4390          */
4391         if (rt && !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4392                 nimap = 1;
4393                 error = xfs_bmapi(NULL, ip, startoffset_fsb, 1, 0, NULL, 0,
4394                         &imap, &nimap, NULL);
4395                 if (error)
4396                         goto out_unlock_iolock;
4397                 ASSERT(nimap == 0 || nimap == 1);
4398                 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4399                         xfs_daddr_t     block;
4400
4401                         ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4402                         block = imap.br_startblock;
4403                         mod = do_div(block, mp->m_sb.sb_rextsize);
4404                         if (mod)
4405                                 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
4406                 }
4407                 nimap = 1;
4408                 error = xfs_bmapi(NULL, ip, endoffset_fsb - 1, 1, 0, NULL, 0,
4409                         &imap, &nimap, NULL);
4410                 if (error)
4411                         goto out_unlock_iolock;
4412                 ASSERT(nimap == 0 || nimap == 1);
4413                 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
4414                         ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
4415                         mod++;
4416                         if (mod && (mod != mp->m_sb.sb_rextsize))
4417                                 endoffset_fsb -= mod;
4418                 }
4419         }
4420         if ((done = (endoffset_fsb <= startoffset_fsb)))
4421                 /*
4422                  * One contiguous piece to clear
4423                  */
4424                 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
4425         else {
4426                 /*
4427                  * Some full blocks, possibly two pieces to clear
4428                  */
4429                 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
4430                         error = xfs_zero_remaining_bytes(ip, offset,
4431                                 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
4432                 if (!error &&
4433                     XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
4434                         error = xfs_zero_remaining_bytes(ip,
4435                                 XFS_FSB_TO_B(mp, endoffset_fsb),
4436                                 offset + len - 1);
4437         }
4438
4439         /*
4440          * free file space until done or until there is an error
4441          */
4442         resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
4443         while (!error && !done) {
4444
4445                 /*
4446                  * allocate and setup the transaction
4447                  */
4448                 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
4449                 error = xfs_trans_reserve(tp,
4450                                           resblks,
4451                                           XFS_WRITE_LOG_RES(mp),
4452                                           0,
4453                                           XFS_TRANS_PERM_LOG_RES,
4454                                           XFS_WRITE_LOG_COUNT);
4455
4456                 /*
4457                  * check for running out of space
4458                  */
4459                 if (error) {
4460                         /*
4461                          * Free the transaction structure.
4462                          */
4463                         ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
4464                         xfs_trans_cancel(tp, 0);
4465                         break;
4466                 }
4467                 xfs_ilock(ip, XFS_ILOCK_EXCL);
4468                 error = XFS_TRANS_RESERVE_QUOTA(mp, tp,
4469                                 ip->i_udquot, ip->i_gdquot, resblks, 0, rt ?
4470                                 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4471                 if (error)
4472                         goto error1;
4473
4474                 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4475                 xfs_trans_ihold(tp, ip);
4476
4477                 /*
4478                  * issue the bunmapi() call to free the blocks
4479                  */
4480                 XFS_BMAP_INIT(&free_list, &firstfsb);
4481                 error = xfs_bunmapi(tp, ip, startoffset_fsb,
4482                                   endoffset_fsb - startoffset_fsb,
4483                                   0, 2, &firstfsb, &free_list, &done);
4484                 if (error) {
4485                         goto error0;
4486                 }
4487
4488                 /*
4489                  * complete the transaction
4490                  */
4491                 error = xfs_bmap_finish(&tp, &free_list, firstfsb, &committed);
4492                 if (error) {
4493                         goto error0;
4494                 }
4495
4496                 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES, NULL);
4497                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
4498         }
4499
4500  out_unlock_iolock:
4501         if (need_iolock)
4502                 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
4503         return error;
4504
4505  error0:
4506         xfs_bmap_cancel(&free_list);
4507  error1:
4508         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
4509         xfs_iunlock(ip, need_iolock ? (XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL) :
4510                     XFS_ILOCK_EXCL);
4511         return error;
4512 }
4513
4514 /*
4515  * xfs_change_file_space()
4516  *      This routine allocates or frees disk space for the given file.
4517  *      The user specified parameters are checked for alignment and size
4518  *      limitations.
4519  *
4520  * RETURNS:
4521  *       0 on success
4522  *      errno on error
4523  *
4524  */
4525 int
4526 xfs_change_file_space(
4527         bhv_desc_t      *bdp,
4528         int             cmd,
4529         xfs_flock64_t   *bf,
4530         xfs_off_t       offset,
4531         cred_t          *credp,
4532         int             attr_flags)
4533 {
4534         int             clrprealloc;
4535         int             error;
4536         xfs_fsize_t     fsize;
4537         xfs_inode_t     *ip;
4538         xfs_mount_t     *mp;
4539         int             setprealloc;
4540         xfs_off_t       startoffset;
4541         xfs_off_t       llen;
4542         xfs_trans_t     *tp;
4543         vattr_t         va;
4544         vnode_t         *vp;
4545
4546         vp = BHV_TO_VNODE(bdp);
4547         vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
4548
4549         ip = XFS_BHVTOI(bdp);
4550         mp = ip->i_mount;
4551
4552         /*
4553          * must be a regular file and have write permission
4554          */
4555         if (vp->v_type != VREG)
4556                 return XFS_ERROR(EINVAL);
4557
4558         xfs_ilock(ip, XFS_ILOCK_SHARED);
4559
4560         if ((error = xfs_iaccess(ip, S_IWUSR, credp))) {
4561                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
4562                 return error;
4563         }
4564
4565         xfs_iunlock(ip, XFS_ILOCK_SHARED);
4566
4567         switch (bf->l_whence) {
4568         case 0: /*SEEK_SET*/
4569                 break;
4570         case 1: /*SEEK_CUR*/
4571                 bf->l_start += offset;
4572                 break;
4573         case 2: /*SEEK_END*/
4574                 bf->l_start += ip->i_d.di_size;
4575                 break;
4576         default:
4577                 return XFS_ERROR(EINVAL);
4578         }
4579
4580         llen = bf->l_len > 0 ? bf->l_len - 1 : bf->l_len;
4581
4582         if (   (bf->l_start < 0)
4583             || (bf->l_start > XFS_MAXIOFFSET(mp))
4584             || (bf->l_start + llen < 0)
4585             || (bf->l_start + llen > XFS_MAXIOFFSET(mp)))
4586                 return XFS_ERROR(EINVAL);
4587
4588         bf->l_whence = 0;
4589
4590         startoffset = bf->l_start;
4591         fsize = ip->i_d.di_size;
4592
4593         /*
4594          * XFS_IOC_RESVSP and XFS_IOC_UNRESVSP will reserve or unreserve
4595          * file space.
4596          * These calls do NOT zero the data space allocated to the file,
4597          * nor do they change the file size.
4598          *
4599          * XFS_IOC_ALLOCSP and XFS_IOC_FREESP will allocate and free file
4600          * space.
4601          * These calls cause the new file data to be zeroed and the file
4602          * size to be changed.
4603          */
4604         setprealloc = clrprealloc = 0;
4605
4606         switch (cmd) {
4607         case XFS_IOC_RESVSP:
4608         case XFS_IOC_RESVSP64:
4609                 error = xfs_alloc_file_space(ip, startoffset, bf->l_len,
4610                                                                 1, attr_flags);
4611                 if (error)
4612                         return error;
4613                 setprealloc = 1;
4614                 break;
4615
4616         case XFS_IOC_UNRESVSP:
4617         case XFS_IOC_UNRESVSP64:
4618                 if ((error = xfs_free_file_space(ip, startoffset, bf->l_len,
4619                                                                 attr_flags)))
4620                         return error;
4621                 break;
4622
4623         case XFS_IOC_ALLOCSP:
4624         case XFS_IOC_ALLOCSP64:
4625         case XFS_IOC_FREESP:
4626         case XFS_IOC_FREESP64:
4627                 if (startoffset > fsize) {
4628                         error = xfs_alloc_file_space(ip, fsize,
4629                                         startoffset - fsize, 0, attr_flags);
4630                         if (error)
4631                                 break;
4632                 }
4633
4634                 va.va_mask = XFS_AT_SIZE;
4635                 va.va_size = startoffset;
4636
4637                 error = xfs_setattr(bdp, &va, attr_flags, credp);
4638
4639                 if (error)
4640                         return error;
4641
4642                 clrprealloc = 1;
4643                 break;
4644
4645         default:
4646                 ASSERT(0);
4647                 return XFS_ERROR(EINVAL);
4648         }
4649
4650         /*
4651          * update the inode timestamp, mode, and prealloc flag bits
4652          */
4653         tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
4654
4655         if ((error = xfs_trans_reserve(tp, 0, XFS_WRITEID_LOG_RES(mp),
4656                                       0, 0, 0))) {
4657                 /* ASSERT(0); */
4658                 xfs_trans_cancel(tp, 0);
4659                 return error;
4660         }
4661
4662         xfs_ilock(ip, XFS_ILOCK_EXCL);
4663
4664         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
4665         xfs_trans_ihold(tp, ip);
4666
4667         if ((attr_flags & ATTR_DMI) == 0) {
4668                 ip->i_d.di_mode &= ~S_ISUID;
4669
4670                 /*
4671                  * Note that we don't have to worry about mandatory
4672                  * file locking being disabled here because we only
4673                  * clear the S_ISGID bit if the Group execute bit is
4674                  * on, but if it was on then mandatory locking wouldn't
4675                  * have been enabled.
4676                  */
4677                 if (ip->i_d.di_mode & S_IXGRP)
4678                         ip->i_d.di_mode &= ~S_ISGID;
4679
4680                 xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
4681         }
4682         if (setprealloc)
4683                 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
4684         else if (clrprealloc)
4685                 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
4686
4687         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
4688         xfs_trans_set_sync(tp);
4689
4690         error = xfs_trans_commit(tp, 0, NULL);
4691
4692         xfs_iunlock(ip, XFS_ILOCK_EXCL);
4693
4694         return error;
4695 }
4696
4697 vnodeops_t xfs_vnodeops = {
4698         BHV_IDENTITY_INIT(VN_BHV_XFS,VNODE_POSITION_XFS),
4699         .vop_open               = xfs_open,
4700         .vop_read               = xfs_read,
4701 #ifdef HAVE_SENDFILE
4702         .vop_sendfile           = xfs_sendfile,
4703 #endif
4704         .vop_write              = xfs_write,
4705         .vop_ioctl              = xfs_ioctl,
4706         .vop_getattr            = xfs_getattr,
4707         .vop_setattr            = xfs_setattr,
4708         .vop_access             = xfs_access,
4709         .vop_lookup             = xfs_lookup,
4710         .vop_create             = xfs_create,
4711         .vop_remove             = xfs_remove,
4712         .vop_link               = xfs_link,
4713         .vop_rename             = xfs_rename,
4714         .vop_mkdir              = xfs_mkdir,
4715         .vop_rmdir              = xfs_rmdir,
4716         .vop_readdir            = xfs_readdir,
4717         .vop_symlink            = xfs_symlink,
4718         .vop_readlink           = xfs_readlink,
4719         .vop_fsync              = xfs_fsync,
4720         .vop_inactive           = xfs_inactive,
4721         .vop_fid2               = xfs_fid2,
4722         .vop_rwlock             = xfs_rwlock,
4723         .vop_rwunlock           = xfs_rwunlock,
4724         .vop_bmap               = xfs_bmap,
4725         .vop_reclaim            = xfs_reclaim,
4726         .vop_attr_get           = xfs_attr_get,
4727         .vop_attr_set           = xfs_attr_set,
4728         .vop_attr_remove        = xfs_attr_remove,
4729         .vop_attr_list          = xfs_attr_list,
4730         .vop_link_removed       = (vop_link_removed_t)fs_noval,
4731         .vop_vnode_change       = (vop_vnode_change_t)fs_noval,
4732         .vop_tosspages          = fs_tosspages,
4733         .vop_flushinval_pages   = fs_flushinval_pages,
4734         .vop_flush_pages        = fs_flush_pages,
4735         .vop_release            = xfs_release,
4736         .vop_iflush             = xfs_inode_flush,
4737 };