fs/xfs/xfs_iops.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_acl.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_alloc.h"
  28 #include "xfs_quota.h"
  29 #include "xfs_mount.h"
  30 #include "xfs_bmap_btree.h"
  31 #include "xfs_dinode.h"
  32 #include "xfs_inode.h"
  33 #include "xfs_bmap.h"
  34 #include "xfs_rtalloc.h"
  35 #include "xfs_error.h"
  36 #include "xfs_itable.h"
  37 #include "xfs_rw.h"
  38 #include "xfs_attr.h"
  39 #include "xfs_buf_item.h"
  40 #include "xfs_utils.h"
  41 #include "xfs_vnodeops.h"
  42 #include "xfs_inode_item.h"
  43 #include "xfs_trace.h"
  44
  45 #include <linux/capability.h>
  46 #include <linux/xattr.h>
  47 #include <linux/namei.h>
  48 #include <linux/posix_acl.h>
  49 #include <linux/security.h>
  50 #include <linux/fiemap.h>
  51 #include <linux/slab.h>
  52
  53 static int
  54 xfs_initxattrs(
  55         struct inode            *inode,
  56         const struct xattr      *xattr_array,
  57         void                    *fs_info)
  58 {
  59         const struct xattr      *xattr;
  60         struct xfs_inode        *ip = XFS_I(inode);
  61         int                     error = 0;
  62
  63         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
  64                 error = xfs_attr_set(ip, xattr->name, xattr->value,
  65                                      xattr->value_len, ATTR_SECURE);
  66                 if (error < 0)
  67                         break;
  68         }
  69         return error;
  70 }
  71
  72 /*
  73  * Hook in SELinux.  This is not quite correct yet, what we really need
  74  * here (as we do for default ACLs) is a mechanism by which creation of
  75  * these attrs can be journalled at inode creation time (along with the
  76  * inode, of course, such that log replay can't cause these to be lost).
  77  */
  78
  79 STATIC int
  80 xfs_init_security(
  81         struct inode    *inode,
  82         struct inode    *dir,
  83         const struct qstr *qstr)
  84 {
  85         return security_inode_init_security(inode, dir, qstr,
  86                                             &xfs_initxattrs, NULL);
  87 }
  88
  89 static void
  90 xfs_dentry_to_name(
  91         struct xfs_name *namep,
  92         struct dentry   *dentry)
  93 {
  94         namep->name = dentry->d_name.name;
  95         namep->len = dentry->d_name.len;
  96 }
  97
  98 STATIC void
  99 xfs_cleanup_inode(
 100         struct inode    *dir,
 101         struct inode    *inode,
 102         struct dentry   *dentry)
 103 {
 104         struct xfs_name teardown;
 105
 106         /* Oh, the horror.
 107          * If we can't add the ACL or we fail in
 108          * xfs_init_security we must back out.
 109          * ENOSPC can hit here, among other things.
 110          */
 111         xfs_dentry_to_name(&teardown, dentry);
 112
 113         xfs_remove(XFS_I(dir), &teardown, XFS_I(inode));
 114         iput(inode);
 115 }
 116
 117 STATIC int
 118 xfs_vn_mknod(
 119         struct inode    *dir,
 120         struct dentry   *dentry,
 121         umode_t         mode,
 122         dev_t           rdev)
 123 {
 124         struct inode    *inode;
 125         struct xfs_inode *ip = NULL;
 126         struct posix_acl *default_acl = NULL;
 127         struct xfs_name name;
 128         int             error;
 129
 130         /*
 131          * Irix uses Missed'em'V split, but doesn't want to see
 132          * the upper 5 bits of (14bit) major.
 133          */
 134         if (S_ISCHR(mode) || S_ISBLK(mode)) {
 135                 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff))
 136                         return -EINVAL;
 137                 rdev = sysv_encode_dev(rdev);
 138         } else {
 139                 rdev = 0;
 140         }
 141
 142         if (IS_POSIXACL(dir)) {
 143                 default_acl = xfs_get_acl(dir, ACL_TYPE_DEFAULT);
 144                 if (IS_ERR(default_acl))
 145                         return PTR_ERR(default_acl);
 146
 147                 if (!default_acl)
 148                         mode &= ~current_umask();
 149         }
 150
 151         xfs_dentry_to_name(&name, dentry);
 152         error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip);
 153         if (unlikely(error))
 154                 goto out_free_acl;
 155
 156         inode = VFS_I(ip);
 157
 158         error = xfs_init_security(inode, dir, &dentry->d_name);
 159         if (unlikely(error))
 160                 goto out_cleanup_inode;
 161
 162         if (default_acl) {
 163                 error = -xfs_inherit_acl(inode, default_acl);
 164                 default_acl = NULL;
 165                 if (unlikely(error))
 166                         goto out_cleanup_inode;
 167         }
 168
 169
 170         d_instantiate(dentry, inode);
 171         return -error;
 172
 173  out_cleanup_inode:
 174         xfs_cleanup_inode(dir, inode, dentry);
 175  out_free_acl:
 176         posix_acl_release(default_acl);
 177         return -error;
 178 }
 179
 180 STATIC int
 181 xfs_vn_create(
 182         struct inode    *dir,
 183         struct dentry   *dentry,
 184         umode_t         mode,
 185         struct nameidata *nd)
 186 {
 187         return xfs_vn_mknod(dir, dentry, mode, 0);
 188 }
 189
 190 STATIC int
 191 xfs_vn_mkdir(
 192         struct inode    *dir,
 193         struct dentry   *dentry,
 194         umode_t         mode)
 195 {
 196         return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0);
 197 }
 198
 199 STATIC struct dentry *
 200 xfs_vn_lookup(
 201         struct inode    *dir,
 202         struct dentry   *dentry,
 203         struct nameidata *nd)
 204 {
 205         struct xfs_inode *cip;
 206         struct xfs_name name;
 207         int             error;
 208
 209         if (dentry->d_name.len >= MAXNAMELEN)
 210                 return ERR_PTR(-ENAMETOOLONG);
 211
 212         xfs_dentry_to_name(&name, dentry);
 213         error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
 214         if (unlikely(error)) {
 215                 if (unlikely(error != ENOENT))
 216                         return ERR_PTR(-error);
 217                 d_add(dentry, NULL);
 218                 return NULL;
 219         }
 220
 221         return d_splice_alias(VFS_I(cip), dentry);
 222 }
 223
 224 STATIC struct dentry *
 225 xfs_vn_ci_lookup(
 226         struct inode    *dir,
 227         struct dentry   *dentry,
 228         struct nameidata *nd)
 229 {
 230         struct xfs_inode *ip;
 231         struct xfs_name xname;
 232         struct xfs_name ci_name;
 233         struct qstr     dname;
 234         int             error;
 235
 236         if (dentry->d_name.len >= MAXNAMELEN)
 237                 return ERR_PTR(-ENAMETOOLONG);
 238
 239         xfs_dentry_to_name(&xname, dentry);
 240         error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
 241         if (unlikely(error)) {
 242                 if (unlikely(error != ENOENT))
 243                         return ERR_PTR(-error);
 244                 /*
 245                  * call d_add(dentry, NULL) here when d_drop_negative_children
 246                  * is called in xfs_vn_mknod (ie. allow negative dentries
 247                  * with CI filesystems).
 248                  */
 249                 return NULL;
 250         }
 251
 252         /* if exact match, just splice and exit */
 253         if (!ci_name.name)
 254                 return d_splice_alias(VFS_I(ip), dentry);
 255
 256         /* else case-insensitive match... */
 257         dname.name = ci_name.name;
 258         dname.len = ci_name.len;
 259         dentry = d_add_ci(dentry, VFS_I(ip), &dname);
 260         kmem_free(ci_name.name);
 261         return dentry;
 262 }
 263
 264 STATIC int
 265 xfs_vn_link(
 266         struct dentry   *old_dentry,
 267         struct inode    *dir,
 268         struct dentry   *dentry)
 269 {
 270         struct inode    *inode = old_dentry->d_inode;
 271         struct xfs_name name;
 272         int             error;
 273
 274         xfs_dentry_to_name(&name, dentry);
 275
 276         error = xfs_link(XFS_I(dir), XFS_I(inode), &name);
 277         if (unlikely(error))
 278                 return -error;
 279
 280         ihold(inode);
 281         d_instantiate(dentry, inode);
 282         return 0;
 283 }
 284
 285 STATIC int
 286 xfs_vn_unlink(
 287         struct inode    *dir,
 288         struct dentry   *dentry)
 289 {
 290         struct xfs_name name;
 291         int             error;
 292
 293         xfs_dentry_to_name(&name, dentry);
 294
 295         error = -xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode));
 296         if (error)
 297                 return error;
 298
 299         /*
 300          * With unlink, the VFS makes the dentry "negative": no inode,
 301          * but still hashed. This is incompatible with case-insensitive
 302          * mode, so invalidate (unhash) the dentry in CI-mode.
 303          */
 304         if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb))
 305                 d_invalidate(dentry);
 306         return 0;
 307 }
 308
 309 STATIC int
 310 xfs_vn_symlink(
 311         struct inode    *dir,
 312         struct dentry   *dentry,
 313         const char      *symname)
 314 {
 315         struct inode    *inode;
 316         struct xfs_inode *cip = NULL;
 317         struct xfs_name name;
 318         int             error;
 319         umode_t         mode;
 320
 321         mode = S_IFLNK |
 322                 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO);
 323         xfs_dentry_to_name(&name, dentry);
 324
 325         error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip);
 326         if (unlikely(error))
 327                 goto out;
 328
 329         inode = VFS_I(cip);
 330
 331         error = xfs_init_security(inode, dir, &dentry->d_name);
 332         if (unlikely(error))
 333                 goto out_cleanup_inode;
 334
 335         d_instantiate(dentry, inode);
 336         return 0;
 337
 338  out_cleanup_inode:
 339         xfs_cleanup_inode(dir, inode, dentry);
 340  out:
 341         return -error;
 342 }
 343
 344 STATIC int
 345 xfs_vn_rename(
 346         struct inode    *odir,
 347         struct dentry   *odentry,
 348         struct inode    *ndir,
 349         struct dentry   *ndentry)
 350 {
 351         struct inode    *new_inode = ndentry->d_inode;
 352         struct xfs_name oname;
 353         struct xfs_name nname;
 354
 355         xfs_dentry_to_name(&oname, odentry);
 356         xfs_dentry_to_name(&nname, ndentry);
 357
 358         return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode),
 359                            XFS_I(ndir), &nname, new_inode ?
 360                                                 XFS_I(new_inode) : NULL);
 361 }
 362
 363 /*
 364  * careful here - this function can get called recursively, so
 365  * we need to be very careful about how much stack we use.
 366  * uio is kmalloced for this reason...
 367  */
 368 STATIC void *
 369 xfs_vn_follow_link(
 370         struct dentry           *dentry,
 371         struct nameidata        *nd)
 372 {
 373         char                    *link;
 374         int                     error = -ENOMEM;
 375
 376         link = kmalloc(MAXPATHLEN+1, GFP_KERNEL);
 377         if (!link)
 378                 goto out_err;
 379
 380         error = -xfs_readlink(XFS_I(dentry->d_inode), link);
 381         if (unlikely(error))
 382                 goto out_kfree;
 383
 384         nd_set_link(nd, link);
 385         return NULL;
 386
 387  out_kfree:
 388         kfree(link);
 389  out_err:
 390         nd_set_link(nd, ERR_PTR(error));
 391         return NULL;
 392 }
 393
 394 STATIC void
 395 xfs_vn_put_link(
 396         struct dentry   *dentry,
 397         struct nameidata *nd,
 398         void            *p)
 399 {
 400         char            *s = nd_get_link(nd);
 401
 402         if (!IS_ERR(s))
 403                 kfree(s);
 404 }
 405
 406 STATIC int
 407 xfs_vn_getattr(
 408         struct vfsmount         *mnt,
 409         struct dentry           *dentry,
 410         struct kstat            *stat)
 411 {
 412         struct inode            *inode = dentry->d_inode;
 413         struct xfs_inode        *ip = XFS_I(inode);
 414         struct xfs_mount        *mp = ip->i_mount;
 415
 416         trace_xfs_getattr(ip);
 417
 418         if (XFS_FORCED_SHUTDOWN(mp))
 419                 return -XFS_ERROR(EIO);
 420
 421         stat->size = XFS_ISIZE(ip);
 422         stat->dev = inode->i_sb->s_dev;
 423         stat->mode = ip->i_d.di_mode;
 424         stat->nlink = ip->i_d.di_nlink;
 425         stat->uid = ip->i_d.di_uid;
 426         stat->gid = ip->i_d.di_gid;
 427         stat->ino = ip->i_ino;
 428         stat->atime = inode->i_atime;
 429         stat->mtime = inode->i_mtime;
 430         stat->ctime = inode->i_ctime;
 431         stat->blocks =
 432                 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks);
 433
 434
 435         switch (inode->i_mode & S_IFMT) {
 436         case S_IFBLK:
 437         case S_IFCHR:
 438                 stat->blksize = BLKDEV_IOSIZE;
 439                 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
 440                                    sysv_minor(ip->i_df.if_u2.if_rdev));
 441                 break;
 442         default:
 443                 if (XFS_IS_REALTIME_INODE(ip)) {
 444                         /*
 445                          * If the file blocks are being allocated from a
 446                          * realtime volume, then return the inode's realtime
 447                          * extent size or the realtime volume's extent size.
 448                          */
 449                         stat->blksize =
 450                                 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog;
 451                 } else
 452                         stat->blksize = xfs_preferred_iosize(mp);
 453                 stat->rdev = 0;
 454                 break;
 455         }
 456
 457         return 0;
 458 }
 459
 460 static void
 461 xfs_setattr_mode(
 462         struct xfs_trans        *tp,
 463         struct xfs_inode        *ip,
 464         struct iattr            *iattr)
 465 {
 466         struct inode    *inode = VFS_I(ip);
 467         umode_t         mode = iattr->ia_mode;
 468
 469         ASSERT(tp);
 470         ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
 471
 472         if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 473                 mode &= ~S_ISGID;
 474
 475         ip->i_d.di_mode &= S_IFMT;
 476         ip->i_d.di_mode |= mode & ~S_IFMT;
 477
 478         inode->i_mode &= S_IFMT;
 479         inode->i_mode |= mode & ~S_IFMT;
 480 }
 481
 482 int
 483 xfs_setattr_nonsize(
 484         struct xfs_inode        *ip,
 485         struct iattr            *iattr,
 486         int                     flags)
 487 {
 488         xfs_mount_t             *mp = ip->i_mount;
 489         struct inode            *inode = VFS_I(ip);
 490         int                     mask = iattr->ia_valid;
 491         xfs_trans_t             *tp;
 492         int                     error;
 493         uid_t                   uid = 0, iuid = 0;
 494         gid_t                   gid = 0, igid = 0;
 495         struct xfs_dquot        *udqp = NULL, *gdqp = NULL;
 496         struct xfs_dquot        *olddquot1 = NULL, *olddquot2 = NULL;
 497
 498         trace_xfs_setattr(ip);
 499
 500         if (mp->m_flags & XFS_MOUNT_RDONLY)
 501                 return XFS_ERROR(EROFS);
 502
 503         if (XFS_FORCED_SHUTDOWN(mp))
 504                 return XFS_ERROR(EIO);
 505
 506         error = -inode_change_ok(inode, iattr);
 507         if (error)
 508                 return XFS_ERROR(error);
 509
 510         ASSERT((mask & ATTR_SIZE) == 0);
 511
 512         /*
 513          * If disk quotas is on, we make sure that the dquots do exist on disk,
 514          * before we start any other transactions. Trying to do this later
 515          * is messy. We don't care to take a readlock to look at the ids
 516          * in inode here, because we can't hold it across the trans_reserve.
 517          * If the IDs do change before we take the ilock, we're covered
 518          * because the i_*dquot fields will get updated anyway.
 519          */
 520         if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) {
 521                 uint    qflags = 0;
 522
 523                 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) {
 524                         uid = iattr->ia_uid;
 525                         qflags |= XFS_QMOPT_UQUOTA;
 526                 } else {
 527                         uid = ip->i_d.di_uid;
 528                 }
 529                 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) {
 530                         gid = iattr->ia_gid;
 531                         qflags |= XFS_QMOPT_GQUOTA;
 532                 }  else {
 533                         gid = ip->i_d.di_gid;
 534                 }
 535
 536                 /*
 537                  * We take a reference when we initialize udqp and gdqp,
 538                  * so it is important that we never blindly double trip on
 539                  * the same variable. See xfs_create() for an example.
 540                  */
 541                 ASSERT(udqp == NULL);
 542                 ASSERT(gdqp == NULL);
 543                 error = xfs_qm_vop_dqalloc(ip, uid, gid, xfs_get_projid(ip),
 544                                          qflags, &udqp, &gdqp);
 545                 if (error)
 546                         return error;
 547         }
 548
 549         tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE);
 550         error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0);
 551         if (error)
 552                 goto out_dqrele;
 553
 554         xfs_ilock(ip, XFS_ILOCK_EXCL);
 555
 556         /*
 557          * Change file ownership.  Must be the owner or privileged.
 558          */
 559         if (mask & (ATTR_UID|ATTR_GID)) {
 560                 /*
 561                  * These IDs could have changed since we last looked at them.
 562                  * But, we're assured that if the ownership did change
 563                  * while we didn't have the inode locked, inode's dquot(s)
 564                  * would have changed also.
 565                  */
 566                 iuid = ip->i_d.di_uid;
 567                 igid = ip->i_d.di_gid;
 568                 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid;
 569                 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid;
 570
 571                 /*
 572                  * Do a quota reservation only if uid/gid is actually
 573                  * going to change.
 574                  */
 575                 if (XFS_IS_QUOTA_RUNNING(mp) &&
 576                     ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) ||
 577                      (XFS_IS_GQUOTA_ON(mp) && igid != gid))) {
 578                         ASSERT(tp);
 579                         error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp,
 580                                                 capable(CAP_FOWNER) ?
 581                                                 XFS_QMOPT_FORCE_RES : 0);
 582                         if (error)      /* out of quota */
 583                                 goto out_trans_cancel;
 584                 }
 585         }
 586
 587         xfs_trans_ijoin(tp, ip, 0);
 588
 589         /*
 590          * Change file ownership.  Must be the owner or privileged.
 591          */
 592         if (mask & (ATTR_UID|ATTR_GID)) {
 593                 /*
 594                  * CAP_FSETID overrides the following restrictions:
 595                  *
 596                  * The set-user-ID and set-group-ID bits of a file will be
 597                  * cleared upon successful return from chown()
 598                  */
 599                 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) &&
 600                     !capable(CAP_FSETID))
 601                         ip->i_d.di_mode &= ~(S_ISUID|S_ISGID);
 602
 603                 /*
 604                  * Change the ownerships and register quota modifications
 605                  * in the transaction.
 606                  */
 607                 if (iuid != uid) {
 608                         if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) {
 609                                 ASSERT(mask & ATTR_UID);
 610                                 ASSERT(udqp);
 611                                 olddquot1 = xfs_qm_vop_chown(tp, ip,
 612                                                         &ip->i_udquot, udqp);
 613                         }
 614                         ip->i_d.di_uid = uid;
 615                         inode->i_uid = uid;
 616                 }
 617                 if (igid != gid) {
 618                         if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) {
 619                                 ASSERT(!XFS_IS_PQUOTA_ON(mp));
 620                                 ASSERT(mask & ATTR_GID);
 621                                 ASSERT(gdqp);
 622                                 olddquot2 = xfs_qm_vop_chown(tp, ip,
 623                                                         &ip->i_gdquot, gdqp);
 624                         }
 625                         ip->i_d.di_gid = gid;
 626                         inode->i_gid = gid;
 627                 }
 628         }
 629
 630         /*
 631          * Change file access modes.
 632          */
 633         if (mask & ATTR_MODE)
 634                 xfs_setattr_mode(tp, ip, iattr);
 635
 636         /*
 637          * Change file access or modified times.
 638          */
 639         if (mask & ATTR_ATIME) {
 640                 inode->i_atime = iattr->ia_atime;
 641                 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec;
 642                 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec;
 643         }
 644         if (mask & ATTR_CTIME) {
 645                 inode->i_ctime = iattr->ia_ctime;
 646                 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
 647                 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
 648         }
 649         if (mask & ATTR_MTIME) {
 650                 inode->i_mtime = iattr->ia_mtime;
 651                 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
 652                 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
 653         }
 654
 655         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 656
 657         XFS_STATS_INC(xs_ig_attrchg);
 658
 659         if (mp->m_flags & XFS_MOUNT_WSYNC)
 660                 xfs_trans_set_sync(tp);
 661         error = xfs_trans_commit(tp, 0);
 662
 663         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 664
 665         /*
 666          * Release any dquot(s) the inode had kept before chown.
 667          */
 668         xfs_qm_dqrele(olddquot1);
 669         xfs_qm_dqrele(olddquot2);
 670         xfs_qm_dqrele(udqp);
 671         xfs_qm_dqrele(gdqp);
 672
 673         if (error)
 674                 return XFS_ERROR(error);
 675
 676         /*
 677          * XXX(hch): Updating the ACL entries is not atomic vs the i_mode
 678          *           update.  We could avoid this with linked transactions
 679          *           and passing down the transaction pointer all the way
 680          *           to attr_set.  No previous user of the generic
 681          *           Posix ACL code seems to care about this issue either.
 682          */
 683         if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) {
 684                 error = -xfs_acl_chmod(inode);
 685                 if (error)
 686                         return XFS_ERROR(error);
 687         }
 688
 689         return 0;
 690
 691 out_trans_cancel:
 692         xfs_trans_cancel(tp, 0);
 693         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 694 out_dqrele:
 695         xfs_qm_dqrele(udqp);
 696         xfs_qm_dqrele(gdqp);
 697         return error;
 698 }
 699
 700 /*
 701  * Truncate file.  Must have write permission and not be a directory.
 702  */
 703 int
 704 xfs_setattr_size(
 705         struct xfs_inode        *ip,
 706         struct iattr            *iattr,
 707         int                     flags)
 708 {
 709         struct xfs_mount        *mp = ip->i_mount;
 710         struct inode            *inode = VFS_I(ip);
 711         int                     mask = iattr->ia_valid;
 712         xfs_off_t               oldsize, newsize;
 713         struct xfs_trans        *tp;
 714         int                     error;
 715         uint                    lock_flags;
 716         uint                    commit_flags = 0;
 717
 718         trace_xfs_setattr(ip);
 719
 720         if (mp->m_flags & XFS_MOUNT_RDONLY)
 721                 return XFS_ERROR(EROFS);
 722
 723         if (XFS_FORCED_SHUTDOWN(mp))
 724                 return XFS_ERROR(EIO);
 725
 726         error = -inode_change_ok(inode, iattr);
 727         if (error)
 728                 return XFS_ERROR(error);
 729
 730         ASSERT(S_ISREG(ip->i_d.di_mode));
 731         ASSERT((mask & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET|
 732                         ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0);
 733
 734         lock_flags = XFS_ILOCK_EXCL;
 735         if (!(flags & XFS_ATTR_NOLOCK))
 736                 lock_flags |= XFS_IOLOCK_EXCL;
 737         xfs_ilock(ip, lock_flags);
 738
 739         oldsize = inode->i_size;
 740         newsize = iattr->ia_size;
 741
 742         /*
 743          * Short circuit the truncate case for zero length files.
 744          */
 745         if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) {
 746                 if (!(mask & (ATTR_CTIME|ATTR_MTIME)))
 747                         goto out_unlock;
 748
 749                 /*
 750                  * Use the regular setattr path to update the timestamps.
 751                  */
 752                 xfs_iunlock(ip, lock_flags);
 753                 iattr->ia_valid &= ~ATTR_SIZE;
 754                 return xfs_setattr_nonsize(ip, iattr, 0);
 755         }
 756
 757         /*
 758          * Make sure that the dquots are attached to the inode.
 759          */
 760         error = xfs_qm_dqattach_locked(ip, 0);
 761         if (error)
 762                 goto out_unlock;
 763
 764         /*
 765          * Now we can make the changes.  Before we join the inode to the
 766          * transaction, take care of the part of the truncation that must be
 767          * done without the inode lock.  This needs to be done before joining
 768          * the inode to the transaction, because the inode cannot be unlocked
 769          * once it is a part of the transaction.
 770          */
 771         if (newsize > oldsize) {
 772                 /*
 773                  * Do the first part of growing a file: zero any data in the
 774                  * last block that is beyond the old EOF.  We need to do this
 775                  * before the inode is joined to the transaction to modify
 776                  * i_size.
 777                  */
 778                 error = xfs_zero_eof(ip, newsize, oldsize);
 779                 if (error)
 780                         goto out_unlock;
 781         }
 782         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 783         lock_flags &= ~XFS_ILOCK_EXCL;
 784
 785         /*
 786          * We are going to log the inode size change in this transaction so
 787          * any previous writes that are beyond the on disk EOF and the new
 788          * EOF that have not been written out need to be written here.  If we
 789          * do not write the data out, we expose ourselves to the null files
 790          * problem.
 791          *
 792          * Only flush from the on disk size to the smaller of the in memory
 793          * file size or the new size as that's the range we really care about
 794          * here and prevents waiting for other data not within the range we
 795          * care about here.
 796          */
 797         if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) {
 798                 error = xfs_flush_pages(ip, ip->i_d.di_size, newsize, 0,
 799                                         FI_NONE);
 800                 if (error)
 801                         goto out_unlock;
 802         }
 803
 804         /*
 805          * Wait for all direct I/O to complete.
 806          */
 807         inode_dio_wait(inode);
 808
 809         error = -block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks);
 810         if (error)
 811                 goto out_unlock;
 812
 813         tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE);
 814         error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0,
 815                                  XFS_TRANS_PERM_LOG_RES,
 816                                  XFS_ITRUNCATE_LOG_COUNT);
 817         if (error)
 818                 goto out_trans_cancel;
 819
 820         truncate_setsize(inode, newsize);
 821
 822         commit_flags = XFS_TRANS_RELEASE_LOG_RES;
 823         lock_flags |= XFS_ILOCK_EXCL;
 824
 825         xfs_ilock(ip, XFS_ILOCK_EXCL);
 826
 827         xfs_trans_ijoin(tp, ip, 0);
 828
 829         /*
 830          * Only change the c/mtime if we are changing the size or we are
 831          * explicitly asked to change it.  This handles the semantic difference
 832          * between truncate() and ftruncate() as implemented in the VFS.
 833          *
 834          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
 835          * special case where we need to update the times despite not having
 836          * these flags set.  For all other operations the VFS set these flags
 837          * explicitly if it wants a timestamp update.
 838          */
 839         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME)))) {
 840                 iattr->ia_ctime = iattr->ia_mtime =
 841                         current_fs_time(inode->i_sb);
 842                 mask |= ATTR_CTIME | ATTR_MTIME;
 843         }
 844
 845         /*
 846          * The first thing we do is set the size to new_size permanently on
 847          * disk.  This way we don't have to worry about anyone ever being able
 848          * to look at the data being freed even in the face of a crash.
 849          * What we're getting around here is the case where we free a block, it
 850          * is allocated to another file, it is written to, and then we crash.
 851          * If the new data gets written to the file but the log buffers
 852          * containing the free and reallocation don't, then we'd end up with
 853          * garbage in the blocks being freed.  As long as we make the new size
 854          * permanent before actually freeing any blocks it doesn't matter if
 855          * they get written to.
 856          */
 857         ip->i_d.di_size = newsize;
 858         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 859
 860         if (newsize <= oldsize) {
 861                 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize);
 862                 if (error)
 863                         goto out_trans_abort;
 864
 865                 /*
 866                  * Truncated "down", so we're removing references to old data
 867                  * here - if we delay flushing for a long time, we expose
 868                  * ourselves unduly to the notorious NULL files problem.  So,
 869                  * we mark this inode and flush it when the file is closed,
 870                  * and do not wait the usual (long) time for writeout.
 871                  */
 872                 xfs_iflags_set(ip, XFS_ITRUNCATED);
 873         }
 874
 875         /*
 876          * Change file access modes.
 877          */
 878         if (mask & ATTR_MODE)
 879                 xfs_setattr_mode(tp, ip, iattr);
 880
 881         if (mask & ATTR_CTIME) {
 882                 inode->i_ctime = iattr->ia_ctime;
 883                 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec;
 884                 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec;
 885         }
 886         if (mask & ATTR_MTIME) {
 887                 inode->i_mtime = iattr->ia_mtime;
 888                 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec;
 889                 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec;
 890         }
 891
 892         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 893
 894         XFS_STATS_INC(xs_ig_attrchg);
 895
 896         if (mp->m_flags & XFS_MOUNT_WSYNC)
 897                 xfs_trans_set_sync(tp);
 898
 899         error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
 900 out_unlock:
 901         if (lock_flags)
 902                 xfs_iunlock(ip, lock_flags);
 903         return error;
 904
 905 out_trans_abort:
 906         commit_flags |= XFS_TRANS_ABORT;
 907 out_trans_cancel:
 908         xfs_trans_cancel(tp, commit_flags);
 909         goto out_unlock;
 910 }
 911
 912 STATIC int
 913 xfs_vn_setattr(
 914         struct dentry   *dentry,
 915         struct iattr    *iattr)
 916 {
 917         if (iattr->ia_valid & ATTR_SIZE)
 918                 return -xfs_setattr_size(XFS_I(dentry->d_inode), iattr, 0);
 919         return -xfs_setattr_nonsize(XFS_I(dentry->d_inode), iattr, 0);
 920 }
 921
 922 #define XFS_FIEMAP_FLAGS        (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR)
 923
 924 /*
 925  * Call fiemap helper to fill in user data.
 926  * Returns positive errors to xfs_getbmap.
 927  */
 928 STATIC int
 929 xfs_fiemap_format(
 930         void                    **arg,
 931         struct getbmapx         *bmv,
 932         int                     *full)
 933 {
 934         int                     error;
 935         struct fiemap_extent_info *fieinfo = *arg;
 936         u32                     fiemap_flags = 0;
 937         u64                     logical, physical, length;
 938
 939         /* Do nothing for a hole */
 940         if (bmv->bmv_block == -1LL)
 941                 return 0;
 942
 943         logical = BBTOB(bmv->bmv_offset);
 944         physical = BBTOB(bmv->bmv_block);
 945         length = BBTOB(bmv->bmv_length);
 946
 947         if (bmv->bmv_oflags & BMV_OF_PREALLOC)
 948                 fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN;
 949         else if (bmv->bmv_oflags & BMV_OF_DELALLOC) {
 950                 fiemap_flags |= FIEMAP_EXTENT_DELALLOC;
 951                 physical = 0;   /* no block yet */
 952         }
 953         if (bmv->bmv_oflags & BMV_OF_LAST)
 954                 fiemap_flags |= FIEMAP_EXTENT_LAST;
 955
 956         error = fiemap_fill_next_extent(fieinfo, logical, physical,
 957                                         length, fiemap_flags);
 958         if (error > 0) {
 959                 error = 0;
 960                 *full = 1;      /* user array now full */
 961         }
 962
 963         return -error;
 964 }
 965
 966 STATIC int
 967 xfs_vn_fiemap(
 968         struct inode            *inode,
 969         struct fiemap_extent_info *fieinfo,
 970         u64                     start,
 971         u64                     length)
 972 {
 973         xfs_inode_t             *ip = XFS_I(inode);
 974         struct getbmapx         bm;
 975         int                     error;
 976
 977         error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS);
 978         if (error)
 979                 return error;
 980
 981         /* Set up bmap header for xfs internal routine */
 982         bm.bmv_offset = BTOBB(start);
 983         /* Special case for whole file */
 984         if (length == FIEMAP_MAX_OFFSET)
 985                 bm.bmv_length = -1LL;
 986         else
 987                 bm.bmv_length = BTOBB(length);
 988
 989         /* We add one because in getbmap world count includes the header */
 990         bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM :
 991                                         fieinfo->fi_extents_max + 1;
 992         bm.bmv_count = min_t(__s32, bm.bmv_count,
 993                              (PAGE_SIZE * 16 / sizeof(struct getbmapx)));
 994         bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES;
 995         if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR)
 996                 bm.bmv_iflags |= BMV_IF_ATTRFORK;
 997         if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC))
 998                 bm.bmv_iflags |= BMV_IF_DELALLOC;
 999
1000         error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo);
1001         if (error)
1002                 return -error;
1003
1004         return 0;
1005 }
1006
1007 static const struct inode_operations xfs_inode_operations = {
1008         .get_acl                = xfs_get_acl,
1009         .getattr                = xfs_vn_getattr,
1010         .setattr                = xfs_vn_setattr,
1011         .setxattr               = generic_setxattr,
1012         .getxattr               = generic_getxattr,
1013         .removexattr            = generic_removexattr,
1014         .listxattr              = xfs_vn_listxattr,
1015         .fiemap                 = xfs_vn_fiemap,
1016 };
1017
1018 static const struct inode_operations xfs_dir_inode_operations = {
1019         .create                 = xfs_vn_create,
1020         .lookup                 = xfs_vn_lookup,
1021         .link                   = xfs_vn_link,
1022         .unlink                 = xfs_vn_unlink,
1023         .symlink                = xfs_vn_symlink,
1024         .mkdir                  = xfs_vn_mkdir,
1025         /*
1026          * Yes, XFS uses the same method for rmdir and unlink.
1027          *
1028          * There are some subtile differences deeper in the code,
1029          * but we use S_ISDIR to check for those.
1030          */
1031         .rmdir                  = xfs_vn_unlink,
1032         .mknod                  = xfs_vn_mknod,
1033         .rename                 = xfs_vn_rename,
1034         .get_acl                = xfs_get_acl,
1035         .getattr                = xfs_vn_getattr,
1036         .setattr                = xfs_vn_setattr,
1037         .setxattr               = generic_setxattr,
1038         .getxattr               = generic_getxattr,
1039         .removexattr            = generic_removexattr,
1040         .listxattr              = xfs_vn_listxattr,
1041 };
1042
1043 static const struct inode_operations xfs_dir_ci_inode_operations = {
1044         .create                 = xfs_vn_create,
1045         .lookup                 = xfs_vn_ci_lookup,
1046         .link                   = xfs_vn_link,
1047         .unlink                 = xfs_vn_unlink,
1048         .symlink                = xfs_vn_symlink,
1049         .mkdir                  = xfs_vn_mkdir,
1050         /*
1051          * Yes, XFS uses the same method for rmdir and unlink.
1052          *
1053          * There are some subtile differences deeper in the code,
1054          * but we use S_ISDIR to check for those.
1055          */
1056         .rmdir                  = xfs_vn_unlink,
1057         .mknod                  = xfs_vn_mknod,
1058         .rename                 = xfs_vn_rename,
1059         .get_acl                = xfs_get_acl,
1060         .getattr                = xfs_vn_getattr,
1061         .setattr                = xfs_vn_setattr,
1062         .setxattr               = generic_setxattr,
1063         .getxattr               = generic_getxattr,
1064         .removexattr            = generic_removexattr,
1065         .listxattr              = xfs_vn_listxattr,
1066 };
1067
1068 static const struct inode_operations xfs_symlink_inode_operations = {
1069         .readlink               = generic_readlink,
1070         .follow_link            = xfs_vn_follow_link,
1071         .put_link               = xfs_vn_put_link,
1072         .get_acl                = xfs_get_acl,
1073         .getattr                = xfs_vn_getattr,
1074         .setattr                = xfs_vn_setattr,
1075         .setxattr               = generic_setxattr,
1076         .getxattr               = generic_getxattr,
1077         .removexattr            = generic_removexattr,
1078         .listxattr              = xfs_vn_listxattr,
1079 };
1080
1081 STATIC void
1082 xfs_diflags_to_iflags(
1083         struct inode            *inode,
1084         struct xfs_inode        *ip)
1085 {
1086         if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE)
1087                 inode->i_flags |= S_IMMUTABLE;
1088         else
1089                 inode->i_flags &= ~S_IMMUTABLE;
1090         if (ip->i_d.di_flags & XFS_DIFLAG_APPEND)
1091                 inode->i_flags |= S_APPEND;
1092         else
1093                 inode->i_flags &= ~S_APPEND;
1094         if (ip->i_d.di_flags & XFS_DIFLAG_SYNC)
1095                 inode->i_flags |= S_SYNC;
1096         else
1097                 inode->i_flags &= ~S_SYNC;
1098         if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME)
1099                 inode->i_flags |= S_NOATIME;
1100         else
1101                 inode->i_flags &= ~S_NOATIME;
1102 }
1103
1104 /*
1105  * Initialize the Linux inode, set up the operation vectors and
1106  * unlock the inode.
1107  *
1108  * When reading existing inodes from disk this is called directly
1109  * from xfs_iget, when creating a new inode it is called from
1110  * xfs_ialloc after setting up the inode.
1111  *
1112  * We are always called with an uninitialised linux inode here.
1113  * We need to initialise the necessary fields and take a reference
1114  * on it.
1115  */
1116 void
1117 xfs_setup_inode(
1118         struct xfs_inode        *ip)
1119 {
1120         struct inode            *inode = &ip->i_vnode;
1121
1122         inode->i_ino = ip->i_ino;
1123         inode->i_state = I_NEW;
1124
1125         inode_sb_list_add(inode);
1126         /* make the inode look hashed for the writeback code */
1127         hlist_add_fake(&inode->i_hash);
1128
1129         inode->i_mode   = ip->i_d.di_mode;
1130         set_nlink(inode, ip->i_d.di_nlink);
1131         inode->i_uid    = ip->i_d.di_uid;
1132         inode->i_gid    = ip->i_d.di_gid;
1133
1134         switch (inode->i_mode & S_IFMT) {
1135         case S_IFBLK:
1136         case S_IFCHR:
1137                 inode->i_rdev =
1138                         MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff,
1139                               sysv_minor(ip->i_df.if_u2.if_rdev));
1140                 break;
1141         default:
1142                 inode->i_rdev = 0;
1143                 break;
1144         }
1145
1146         inode->i_generation = ip->i_d.di_gen;
1147         i_size_write(inode, ip->i_d.di_size);
1148         inode->i_atime.tv_sec   = ip->i_d.di_atime.t_sec;
1149         inode->i_atime.tv_nsec  = ip->i_d.di_atime.t_nsec;
1150         inode->i_mtime.tv_sec   = ip->i_d.di_mtime.t_sec;
1151         inode->i_mtime.tv_nsec  = ip->i_d.di_mtime.t_nsec;
1152         inode->i_ctime.tv_sec   = ip->i_d.di_ctime.t_sec;
1153         inode->i_ctime.tv_nsec  = ip->i_d.di_ctime.t_nsec;
1154         xfs_diflags_to_iflags(inode, ip);
1155
1156         switch (inode->i_mode & S_IFMT) {
1157         case S_IFREG:
1158                 inode->i_op = &xfs_inode_operations;
1159                 inode->i_fop = &xfs_file_operations;
1160                 inode->i_mapping->a_ops = &xfs_address_space_operations;
1161                 break;
1162         case S_IFDIR:
1163                 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb))
1164                         inode->i_op = &xfs_dir_ci_inode_operations;
1165                 else
1166                         inode->i_op = &xfs_dir_inode_operations;
1167                 inode->i_fop = &xfs_dir_file_operations;
1168                 break;
1169         case S_IFLNK:
1170                 inode->i_op = &xfs_symlink_inode_operations;
1171                 if (!(ip->i_df.if_flags & XFS_IFINLINE))
1172                         inode->i_mapping->a_ops = &xfs_address_space_operations;
1173                 break;
1174         default:
1175                 inode->i_op = &xfs_inode_operations;
1176                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1177                 break;
1178         }
1179
1180         /*
1181          * If there is no attribute fork no ACL can exist on this inode,
1182          * and it can't have any file capabilities attached to it either.
1183          */
1184         if (!XFS_IFORK_Q(ip)) {
1185                 inode_has_no_xattr(inode);
1186                 cache_no_acl(inode);
1187         }
1188
1189         xfs_iflags_clear(ip, XFS_INEW);
1190         barrier();
1191
1192         unlock_new_inode(inode);
1193 }