ldiskfs/kernel_patches/patches/iopen-2.6.18-rhel5.patch

   1 Index: linux-2.6.18-53.1.21/fs/ext3/iopen.c
   2 ===================================================================
   3 --- /dev/null
   4 +++ linux-2.6.18-53.1.21/fs/ext3/iopen.c
   5 @@ -0,0 +1,291 @@
   6 +/*
   7 + * linux/fs/ext3/iopen.c
   8 + *
   9 + * Special support for open by inode number
  10 + *
  11 + * Copyright (C) 2001 by Theodore Ts'o (tytso@alum.mit.edu).
  12 + *
  13 + * This file may be redistributed under the terms of the GNU General
  14 + * Public License.
  15 + *
  16 + *
  17 + * Invariants:
  18 + *   - there is only ever a single DCACHE_NFSD_DISCONNECTED dentry alias
  19 + *     for an inode at one time.
  20 + *   - there are never both connected and DCACHE_NFSD_DISCONNECTED dentry
  21 + *     aliases on an inode at the same time.
  22 + *
  23 + * If we have any connected dentry aliases for an inode, use one of those
  24 + * in iopen_lookup().  Otherwise, we instantiate a single NFSD_DISCONNECTED
  25 + * dentry for this inode, which thereafter will be found by the dcache
  26 + * when looking up this inode number in __iopen__, so we don't return here
  27 + * until it is gone.
  28 + *
  29 + * If we get an inode via a regular name lookup, then we "rename" the
  30 + * NFSD_DISCONNECTED dentry to the proper name and parent.  This ensures
  31 + * existing users of the disconnected dentry will continue to use the same
  32 + * dentry as the connected users, and there will never be both kinds of
  33 + * dentry aliases at one time.
  34 + */
  35 +
  36 +#include <linux/sched.h>
  37 +#include <linux/fs.h>
  38 +#include <linux/ext3_jbd.h>
  39 +#include <linux/jbd.h>
  40 +#include <linux/ext3_fs.h>
  41 +#include <linux/smp_lock.h>
  42 +#include <linux/dcache.h>
  43 +#include <linux/security.h>
  44 +#include "iopen.h"
  45 +
  46 +#ifndef assert
  47 +#define assert(test) J_ASSERT(test)
  48 +#endif
  49 +
  50 +#define IOPEN_NAME_LEN 32
  51 +
  52 +/*
  53 + * This implements looking up an inode by number.
  54 + */
  55 +static struct dentry *iopen_lookup(struct inode * dir, struct dentry *dentry,
  56 +                                  struct nameidata *nd)
  57 +{
  58 +       struct inode *inode;
  59 +       unsigned long ino;
  60 +       struct list_head *lp;
  61 +       struct dentry *alternate;
  62 +       char buf[IOPEN_NAME_LEN];
  63 +
  64 +       if (dentry->d_name.len >= IOPEN_NAME_LEN)
  65 +               return ERR_PTR(-ENAMETOOLONG);
  66 +
  67 +       memcpy(buf, dentry->d_name.name, dentry->d_name.len);
  68 +       buf[dentry->d_name.len] = 0;
  69 +
  70 +       if (strcmp(buf, ".") == 0)
  71 +               ino = dir->i_ino;
  72 +       else if (strcmp(buf, "..") == 0)
  73 +               ino = EXT3_ROOT_INO;
  74 +       else
  75 +               ino = simple_strtoul(buf, 0, 0);
  76 +
  77 +       if ((ino != EXT3_ROOT_INO &&
  78 +            ino < EXT3_FIRST_INO(dir->i_sb)) ||
  79 +           ino > le32_to_cpu(EXT3_SB(dir->i_sb)->s_es->s_inodes_count))
  80 +               return ERR_PTR(-ENOENT);
  81 +
  82 +       inode = iget(dir->i_sb, ino);
  83 +       if (!inode)
  84 +               return ERR_PTR(-EACCES);
  85 +       if (is_bad_inode(inode)) {
  86 +               iput(inode);
  87 +               return ERR_PTR(-ENOENT);
  88 +       }
  89 +
  90 +       assert(list_empty(&dentry->d_alias));           /* d_instantiate */
  91 +       assert(d_unhashed(dentry));                     /* d_rehash */
  92 +
  93 +       /* preferrably return a connected dentry */
  94 +       spin_lock(&dcache_lock);
  95 +       list_for_each(lp, &inode->i_dentry) {
  96 +               alternate = list_entry(lp, struct dentry, d_alias);
  97 +               assert(!(alternate->d_flags & DCACHE_DISCONNECTED));
  98 +       }
  99 +
 100 +       if (!list_empty(&inode->i_dentry)) {
 101 +               alternate = list_entry(inode->i_dentry.next,
 102 +                                      struct dentry, d_alias);
 103 +               dget_locked(alternate);
 104 +               spin_lock(&alternate->d_lock);
 105 +               alternate->d_flags |= DCACHE_REFERENCED;
 106 +               spin_unlock(&alternate->d_lock);
 107 +               iput(inode);
 108 +               spin_unlock(&dcache_lock);
 109 +               return alternate;
 110 +       }
 111 +       dentry->d_flags |= DCACHE_DISCONNECTED;
 112 +
 113 +       /* d_add(), but don't drop dcache_lock before adding dentry to inode */
 114 +       list_add(&dentry->d_alias, &inode->i_dentry);   /* d_instantiate */
 115 +       dentry->d_inode = inode;
 116 +
 117 +       d_rehash_cond(dentry, 0);
 118 +       spin_unlock(&dcache_lock);
 119 +
 120 +       return NULL;
 121 +}
 122 +
 123 +/* This function is spliced into ext3_lookup and does the move of a
 124 + * disconnected dentry (if it exists) to a connected dentry.
 125 + */
 126 +struct dentry *iopen_connect_dentry(struct dentry *dentry, struct inode *inode,
 127 +                                   int rehash)
 128 +{
 129 +       struct dentry *tmp, *goal = NULL;
 130 +       struct list_head *lp;
 131 +
 132 +       /* verify this dentry is really new */
 133 +       assert(dentry->d_inode == NULL);
 134 +       assert(list_empty(&dentry->d_alias));           /* d_instantiate */
 135 +       if (rehash)
 136 +               assert(d_unhashed(dentry));             /* d_rehash */
 137 +       assert(list_empty(&dentry->d_subdirs));
 138 +
 139 +       spin_lock(&dcache_lock);
 140 +       if (!inode)
 141 +               goto do_rehash;
 142 +
 143 +       if (!test_opt(inode->i_sb, IOPEN))
 144 +               goto do_instantiate;
 145 +
 146 +       /* preferrably return a connected dentry */
 147 +       list_for_each(lp, &inode->i_dentry) {
 148 +               tmp = list_entry(lp, struct dentry, d_alias);
 149 +               if (tmp->d_flags & DCACHE_DISCONNECTED) {
 150 +                       assert(tmp->d_alias.next == &inode->i_dentry);
 151 +                       assert(tmp->d_alias.prev == &inode->i_dentry);
 152 +                       goal = tmp;
 153 +                       dget_locked(goal);
 154 +                       break;
 155 +               }
 156 +       }
 157 +
 158 +       if (!goal)
 159 +               goto do_instantiate;
 160 +
 161 +       /* Move the goal to the de hash queue */
 162 +       goal->d_flags &= ~DCACHE_DISCONNECTED;
 163 +       security_d_instantiate(goal, inode);
 164 +       __d_drop(dentry);
 165 +       d_rehash_cond(dentry, 0);
 166 +       d_move_locked(goal, dentry);
 167 +       spin_unlock(&dcache_lock);
 168 +       iput(inode);
 169 +
 170 +       return goal;
 171 +
 172 +       /* d_add(), but don't drop dcache_lock before adding dentry to inode */
 173 +do_instantiate:
 174 +       list_add(&dentry->d_alias, &inode->i_dentry);   /* d_instantiate */
 175 +       dentry->d_inode = inode;
 176 +do_rehash:
 177 +       if (rehash)
 178 +               d_rehash_cond(dentry, 0);
 179 +       spin_unlock(&dcache_lock);
 180 +
 181 +       return NULL;
 182 +}
 183 +
 184 +/*
 185 + * Similar as d_instantiate() except that it drops the disconnected
 186 + * dentry if any.
 187 + */
 188 +void iopen_d_instantiate(struct dentry *dentry, struct inode * inode)
 189 +{
 190 +       struct dentry *dis_dentry;
 191 +
 192 +       /* verify this dentry is really new */
 193 +       assert(dentry->d_inode == NULL);
 194 +       assert(list_empty(&dentry->d_alias));
 195 +
 196 +       spin_lock(&dcache_lock);
 197 +       if (!inode || !test_opt(inode->i_sb, IOPEN) ||
 198 +           list_empty(&inode->i_dentry))
 199 +               goto do_instantiate;
 200 +
 201 +       /* a disconnected dentry has been added in our back,
 202 +        * we have to drop this dentry, see bug 16362/15713*/
 203 +       dis_dentry = list_entry(inode->i_dentry.next, struct dentry, d_alias);
 204 +       spin_lock(&dis_dentry->d_lock);
 205 +       assert(dis_dentry->d_alias.next == &inode->i_dentry);
 206 +       assert(dis_dentry->d_alias.prev == &inode->i_dentry);
 207 +       assert(dis_dentry->d_flags & DCACHE_DISCONNECTED);
 208 +       __d_drop(dis_dentry);
 209 +       list_del_init(&dis_dentry->d_alias);
 210 +       spin_unlock(&dis_dentry->d_lock);
 211 +
 212 +do_instantiate:
 213 +       if (inode)
 214 +               list_add(&dentry->d_alias, &inode->i_dentry);
 215 +       dentry->d_inode = inode;
 216 +       spin_unlock(&dcache_lock);
 217 +       security_d_instantiate(dentry, inode);
 218 +}
 219 +
 220 +/*
 221 + * These are the special structures for the iopen pseudo directory.
 222 + */
 223 +
 224 +static struct inode_operations iopen_inode_operations = {
 225 +       lookup:         iopen_lookup,           /* BKL held */
 226 +};
 227 +
 228 +static struct file_operations iopen_file_operations = {
 229 +       read:           generic_read_dir,
 230 +};
 231 +
 232 +static int match_dentry(struct dentry *dentry, const char *name)
 233 +{
 234 +       int     len;
 235 +
 236 +       len = strlen(name);
 237 +       if (dentry->d_name.len != len)
 238 +               return 0;
 239 +       if (strncmp(dentry->d_name.name, name, len))
 240 +               return 0;
 241 +       return 1;
 242 +}
 243 +
 244 +/*
 245 + * This function is spliced into ext3_lookup and returns 1 the file
 246 + * name is __iopen__ and dentry has been filled in appropriately.
 247 + */
 248 +int ext3_check_for_iopen(struct inode *dir, struct dentry *dentry)
 249 +{
 250 +       struct inode *inode;
 251 +
 252 +       if (dir->i_ino != EXT3_ROOT_INO ||
 253 +           !test_opt(dir->i_sb, IOPEN) ||
 254 +           !match_dentry(dentry, "__iopen__"))
 255 +               return 0;
 256 +
 257 +       inode = iget(dir->i_sb, EXT3_BAD_INO);
 258 +
 259 +       if (!inode)
 260 +               return 0;
 261 +       d_add(dentry, inode);
 262 +       return 1;
 263 +}
 264 +
 265 +/*
 266 + * This function is spliced into read_inode; it returns 1 if inode
 267 + * number is the one for /__iopen__, in which case the inode is filled
 268 + * in appropriately.  Otherwise, this fuction returns 0.
 269 + */
 270 +int ext3_iopen_get_inode(struct inode *inode)
 271 +{
 272 +       if (inode->i_ino != EXT3_BAD_INO)
 273 +               return 0;
 274 +
 275 +       inode->i_mode = S_IFDIR | S_IRUSR | S_IXUSR;
 276 +       if (test_opt(inode->i_sb, IOPEN_NOPRIV))
 277 +               inode->i_mode |= 0777;
 278 +       inode->i_uid = 0;
 279 +       inode->i_gid = 0;
 280 +       inode->i_nlink = 1;
 281 +       inode->i_size = 4096;
 282 +       inode->i_atime = CURRENT_TIME;
 283 +       inode->i_ctime = CURRENT_TIME;
 284 +       inode->i_mtime = CURRENT_TIME;
 285 +       EXT3_I(inode)->i_dtime = 0;
 286 +       EXT3_I(inode)->i_file_acl = 0;
 287 +       inode->i_blocks = 0;
 288 +       inode->i_version = 1;
 289 +       inode->i_generation = 0;
 290 +
 291 +       inode->i_op = &iopen_inode_operations;
 292 +       inode->i_fop = &iopen_file_operations;
 293 +       inode->i_mapping->a_ops = 0;
 294 +
 295 +       return 1;
 296 +}
 297 Index: linux-2.6.18-53.1.21/fs/ext3/iopen.h
 298 ===================================================================
 299 --- /dev/null
 300 +++ linux-2.6.18-53.1.21/fs/ext3/iopen.h
 301 @@ -0,0 +1,16 @@
 302 +/*
 303 + * iopen.h
 304 + *
 305 + * Special support for opening files by inode number.
 306 + *
 307 + * Copyright (C) 2001 by Theodore Ts'o (tytso@alum.mit.edu).
 308 + *
 309 + * This file may be redistributed under the terms of the GNU General
 310 + * Public License.
 311 + */
 312 +
 313 +extern int ext3_check_for_iopen(struct inode *dir, struct dentry *dentry);
 314 +extern int ext3_iopen_get_inode(struct inode *inode);
 315 +extern struct dentry *iopen_connect_dentry(struct dentry *dentry,
 316 +                                          struct inode *inode, int rehash);
 317 +extern void iopen_d_instantiate(struct dentry *dentry, struct inode * inode);
 318 Index: linux-2.6.18-53.1.21/fs/ext3/inode.c
 319 ===================================================================
 320 --- linux-2.6.18-53.1.21.orig/fs/ext3/inode.c
 321 +++ linux-2.6.18-53.1.21/fs/ext3/inode.c
 322 @@ -37,6 +37,7 @@
 323  #include <linux/mpage.h>
 324  #include <linux/uio.h>
 325  #include "xattr.h"
 326 +#include "iopen.h"
 327  #include "acl.h"
 328
 329  static int ext3_writepage_trans_blocks(struct inode *inode);
 330 @@ -2593,6 +2594,8 @@ void ext3_read_inode(struct inode * inod
 331         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 332  #endif
 333         ei->i_block_alloc_info = NULL;
 334 +       if (ext3_iopen_get_inode(inode))
 335 +               return;
 336
 337         if (__ext3_get_inode_loc(inode, &iloc, 0))
 338                 goto bad_inode;
 339 Index: linux-2.6.18-53.1.21/fs/ext3/super.c
 340 ===================================================================
 341 --- linux-2.6.18-53.1.21.orig/fs/ext3/super.c
 342 +++ linux-2.6.18-53.1.21/fs/ext3/super.c
 343 @@ -677,6 +677,7 @@ enum {
 344         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 345         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 346         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 347 +       Opt_iopen, Opt_noiopen, Opt_iopen_nopriv,
 348         Opt_grpquota
 349  };
 350
 351 @@ -726,6 +727,9 @@ static match_table_t tokens = {
 352         {Opt_noquota, "noquota"},
 353         {Opt_quota, "quota"},
 354         {Opt_usrquota, "usrquota"},
 355 +       {Opt_iopen, "iopen"},
 356 +       {Opt_noiopen, "noiopen"},
 357 +       {Opt_iopen_nopriv, "iopen_nopriv"},
 358         {Opt_barrier, "barrier=%u"},
 359         {Opt_err, NULL},
 360         {Opt_resize, "resize"},
 361 @@ -1041,6 +1045,18 @@ clear_qf_name:
 362                         else
 363                                 clear_opt(sbi->s_mount_opt, BARRIER);
 364                         break;
 365 +               case Opt_iopen:
 366 +                       set_opt (sbi->s_mount_opt, IOPEN);
 367 +                       clear_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 368 +                       break;
 369 +               case Opt_noiopen:
 370 +                       clear_opt (sbi->s_mount_opt, IOPEN);
 371 +                       clear_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 372 +                       break;
 373 +               case Opt_iopen_nopriv:
 374 +                       set_opt (sbi->s_mount_opt, IOPEN);
 375 +                       set_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 376 +                       break;
 377                 case Opt_ignore:
 378                         break;
 379                 case Opt_resize:
 380 Index: linux-2.6.18-53.1.21/fs/ext3/namei.c
 381 ===================================================================
 382 --- linux-2.6.18-53.1.21.orig/fs/ext3/namei.c
 383 +++ linux-2.6.18-53.1.21/fs/ext3/namei.c
 384 @@ -39,6 +39,7 @@
 385
 386  #include "namei.h"
 387  #include "xattr.h"
 388 +#include "iopen.h"
 389  #include "acl.h"
 390
 391  /*
 392 @@ -1020,6 +1021,9 @@ static struct dentry *ext3_lookup(struct
 393         if (dentry->d_name.len > EXT3_NAME_LEN)
 394                 return ERR_PTR(-ENAMETOOLONG);
 395
 396 +       if (ext3_check_for_iopen(dir, dentry))
 397 +               return NULL;
 398 +
 399         bh = ext3_find_entry(dentry, &de);
 400         inode = NULL;
 401         if (bh) {
 402 @@ -1035,7 +1039,8 @@ static struct dentry *ext3_lookup(struct
 403                 if (!inode)
 404                         return ERR_PTR(-EACCES);
 405         }
 406 -       return d_splice_alias(inode, dentry);
 407 +
 408 +       return iopen_connect_dentry(dentry, inode, 1);
 409  }
 410
 411
 412 @@ -1678,7 +1683,7 @@ static int ext3_add_nondir(handle_t *han
 413         int err = ext3_add_entry(handle, dentry, inode);
 414         if (!err) {
 415                 ext3_mark_inode_dirty(handle, inode);
 416 -               d_instantiate(dentry, inode);
 417 +               iopen_d_instantiate(dentry, inode);
 418                 return 0;
 419         }
 420         ext3_dec_count(handle, inode);
 421 @@ -1840,7 +1845,7 @@ retry:
 422         dir->i_nlink++;
 423         ext3_update_dx_flag(dir);
 424         ext3_mark_inode_dirty(handle, dir);
 425 -       d_instantiate(dentry, inode);
 426 +       iopen_d_instantiate(dentry, inode);
 427  out_stop:
 428         ext3_journal_stop(handle);
 429         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
 430 @@ -2108,10 +2113,6 @@ static int ext3_rmdir (struct inode * di
 431                               inode->i_nlink);
 432         inode->i_version++;
 433         inode->i_nlink = 0;
 434 -       /* There's no need to set i_disksize: the fact that i_nlink is
 435 -        * zero will ensure that the right thing happens during any
 436 -        * recovery. */
 437 -       inode->i_size = 0;
 438         ext3_orphan_add(handle, inode);
 439         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
 440         ext3_mark_inode_dirty(handle, inode);
 441 @@ -2235,6 +2236,23 @@ out_stop:
 442         return err;
 443  }
 444
 445 +/* Like ext3_add_nondir() except for call to iopen_connect_dentry */
 446 +static int ext3_add_link(handle_t *handle, struct dentry *dentry,
 447 +                        struct inode *inode)
 448 +{
 449 +       int err = ext3_add_entry(handle, dentry, inode);
 450 +       if (!err) {
 451 +               err = ext3_mark_inode_dirty(handle, inode);
 452 +               if (err == 0) {
 453 +                       dput(iopen_connect_dentry(dentry, inode, 0));
 454 +                       return 0;
 455 +               }
 456 +       }
 457 +       ext3_dec_count(handle, inode);
 458 +       iput(inode);
 459 +       return err;
 460 +}
 461 +
 462  static int ext3_link (struct dentry * old_dentry,
 463                 struct inode * dir, struct dentry *dentry)
 464  {
 465 @@ -2264,7 +2282,8 @@ retry:
 466         ext3_inc_count(handle, inode);
 467         atomic_inc(&inode->i_count);
 468
 469 -       err = ext3_add_nondir(handle, dentry, inode);
 470 +       err = ext3_add_link(handle, dentry, inode);
 471 +       ext3_orphan_del(handle, inode);
 472         ext3_journal_stop(handle);
 473         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
 474                 goto retry;
 475 Index: linux-2.6.18-53.1.21/fs/ext3/Makefile
 476 ===================================================================
 477 --- linux-2.6.18-53.1.21.orig/fs/ext3/Makefile
 478 +++ linux-2.6.18-53.1.21/fs/ext3/Makefile
 479 @@ -4,7 +4,7 @@
 480
 481  obj-$(CONFIG_EXT3_FS) += ext3.o
 482
 483 -ext3-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o \
 484 +ext3-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o iopen.o \
 485            ioctl.o namei.o super.o symlink.o hash.o resize.o
 486
 487  ext3-$(CONFIG_EXT3_FS_XATTR)    += xattr.o xattr_user.o xattr_trusted.o
 488 Index: linux-2.6.18-53.1.21/include/linux/ext3_fs.h
 489 ===================================================================
 490 --- linux-2.6.18-53.1.21.orig/include/linux/ext3_fs.h
 491 +++ linux-2.6.18-53.1.21/include/linux/ext3_fs.h
 492 @@ -371,6 +371,8 @@ struct ext3_inode {
 493  #define EXT3_MOUNT_QUOTA               0x80000 /* Some quota option set */
 494  #define EXT3_MOUNT_USRQUOTA            0x100000 /* "old" user quota */
 495  #define EXT3_MOUNT_GRPQUOTA            0x200000 /* "old" group quota */
 496 +#define EXT3_MOUNT_IOPEN               0x400000        /* Allow access via iopen */
 497 +#define EXT3_MOUNT_IOPEN_NOPRIV                0x800000/* Make iopen world-readable */
 498
 499  /* Compatibility, for having both ext2_fs.h and ext3_fs.h included at once */
 500  #ifndef _LINUX_EXT2_FS_H