ldiskfs/kernel_patches/patches/iopen-2.6.22-vanilla.patch

   1 Index: linux-2.6.16.27-0.9/fs/ext3/iopen.c
   2 ===================================================================
   3 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
   4 +++ linux-2.6.16.27-0.9/fs/ext3/iopen.c 2007-06-29 08:33:12.000000000 +0200
   5 @@ -0,0 +1,256 @@
   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_ACL_IDX_INO &&
  79 +            //ino != EXT3_ACL_DATA_INO &&
  80 +            ino < EXT3_FIRST_INO(dir->i_sb)) ||
  81 +           ino > le32_to_cpu(EXT3_SB(dir->i_sb)->s_es->s_inodes_count))
  82 +               return ERR_PTR(-ENOENT);
  83 +
  84 +       inode = iget(dir->i_sb, ino);
  85 +       if (!inode)
  86 +               return ERR_PTR(-EACCES);
  87 +       if (is_bad_inode(inode)) {
  88 +               iput(inode);
  89 +               return ERR_PTR(-ENOENT);
  90 +       }
  91 +
  92 +       assert(list_empty(&dentry->d_alias));           /* d_instantiate */
  93 +       assert(d_unhashed(dentry));                     /* d_rehash */
  94 +
  95 +       /* preferrably return a connected dentry */
  96 +       spin_lock(&dcache_lock);
  97 +       list_for_each(lp, &inode->i_dentry) {
  98 +               alternate = list_entry(lp, struct dentry, d_alias);
  99 +               assert(!(alternate->d_flags & DCACHE_DISCONNECTED));
 100 +       }
 101 +
 102 +       if (!list_empty(&inode->i_dentry)) {
 103 +               alternate = list_entry(inode->i_dentry.next,
 104 +                                      struct dentry, d_alias);
 105 +               dget_locked(alternate);
 106 +               spin_lock(&alternate->d_lock);
 107 +               alternate->d_flags |= DCACHE_REFERENCED;
 108 +               spin_unlock(&alternate->d_lock);
 109 +               iput(inode);
 110 +               spin_unlock(&dcache_lock);
 111 +               return alternate;
 112 +       }
 113 +       dentry->d_flags |= DCACHE_DISCONNECTED;
 114 +
 115 +       /* d_add(), but don't drop dcache_lock before adding dentry to inode */
 116 +       list_add(&dentry->d_alias, &inode->i_dentry);   /* d_instantiate */
 117 +       dentry->d_inode = inode;
 118 +
 119 +       d_rehash_cond(dentry, 0);       /* d_rehash */
 120 +       spin_unlock(&dcache_lock);
 121 +
 122 +       return NULL;
 123 +}
 124 +
 125 +/* This function is spliced into ext3_lookup and does the move of a
 126 + * disconnected dentry (if it exists) to a connected dentry.
 127 + */
 128 +struct dentry *iopen_connect_dentry(struct dentry *dentry, struct inode *inode,
 129 +                                   int rehash)
 130 +{
 131 +       struct dentry *tmp, *goal = NULL;
 132 +       struct list_head *lp;
 133 +
 134 +       /* verify this dentry is really new */
 135 +       assert(dentry->d_inode == NULL);
 136 +       assert(list_empty(&dentry->d_alias));           /* d_instantiate */
 137 +       if (rehash)
 138 +               assert(d_unhashed(dentry));             /* d_rehash */
 139 +       assert(list_empty(&dentry->d_subdirs));
 140 +
 141 +       spin_lock(&dcache_lock);
 142 +       if (!inode)
 143 +               goto do_rehash;
 144 +
 145 +       if (!test_opt(inode->i_sb, IOPEN))
 146 +               goto do_instantiate;
 147 +
 148 +       /* preferrably return a connected dentry */
 149 +       list_for_each(lp, &inode->i_dentry) {
 150 +               tmp = list_entry(lp, struct dentry, d_alias);
 151 +               if (tmp->d_flags & DCACHE_DISCONNECTED) {
 152 +                       assert(tmp->d_alias.next == &inode->i_dentry);
 153 +                       assert(tmp->d_alias.prev == &inode->i_dentry);
 154 +                       goal = tmp;
 155 +                       dget_locked(goal);
 156 +                       break;
 157 +               }
 158 +       }
 159 +
 160 +       if (!goal)
 161 +               goto do_instantiate;
 162 +
 163 +       /* Move the goal to the de hash queue */
 164 +       goal->d_flags &= ~DCACHE_DISCONNECTED;
 165 +       security_d_instantiate(goal, inode);
 166 +       __d_drop(dentry);
 167 +       d_rehash_cond(dentry, 0);
 168 +       __d_move(goal, dentry);
 169 +       spin_unlock(&dcache_lock);
 170 +       iput(inode);
 171 +
 172 +       return goal;
 173 +
 174 +       /* d_add(), but don't drop dcache_lock before adding dentry to inode */
 175 +do_instantiate:
 176 +       list_add(&dentry->d_alias, &inode->i_dentry);   /* d_instantiate */
 177 +       dentry->d_inode = inode;
 178 +do_rehash:
 179 +       if (rehash)
 180 +               d_rehash_cond(dentry, 0);       /* d_rehash */
 181 +       spin_unlock(&dcache_lock);
 182 +
 183 +       return NULL;
 184 +}
 185 +
 186 +/*
 187 + * These are the special structures for the iopen pseudo directory.
 188 + */
 189 +
 190 +static struct inode_operations iopen_inode_operations = {
 191 +       lookup:         iopen_lookup,           /* BKL held */
 192 +};
 193 +
 194 +static struct file_operations iopen_file_operations = {
 195 +       read:           generic_read_dir,
 196 +};
 197 +
 198 +static int match_dentry(struct dentry *dentry, const char *name)
 199 +{
 200 +       int     len;
 201 +
 202 +       len = strlen(name);
 203 +       if (dentry->d_name.len != len)
 204 +               return 0;
 205 +       if (strncmp(dentry->d_name.name, name, len))
 206 +               return 0;
 207 +       return 1;
 208 +}
 209 +
 210 +/*
 211 + * This function is spliced into ext3_lookup and returns 1 the file
 212 + * name is __iopen__ and dentry has been filled in appropriately.
 213 + */
 214 +int ext3_check_for_iopen(struct inode *dir, struct dentry *dentry)
 215 +{
 216 +       struct inode *inode;
 217 +
 218 +       if (dir->i_ino != EXT3_ROOT_INO ||
 219 +           !test_opt(dir->i_sb, IOPEN) ||
 220 +           !match_dentry(dentry, "__iopen__"))
 221 +               return 0;
 222 +
 223 +       inode = iget(dir->i_sb, EXT3_BAD_INO);
 224 +
 225 +       if (!inode)
 226 +               return 0;
 227 +       d_add(dentry, inode);
 228 +       return 1;
 229 +}
 230 +
 231 +/*
 232 + * This function is spliced into read_inode; it returns 1 if inode
 233 + * number is the one for /__iopen__, in which case the inode is filled
 234 + * in appropriately.  Otherwise, this fuction returns 0.
 235 + */
 236 +int ext3_iopen_get_inode(struct inode *inode)
 237 +{
 238 +       if (inode->i_ino != EXT3_BAD_INO)
 239 +               return 0;
 240 +
 241 +       inode->i_mode = S_IFDIR | S_IRUSR | S_IXUSR;
 242 +       if (test_opt(inode->i_sb, IOPEN_NOPRIV))
 243 +               inode->i_mode |= 0777;
 244 +       inode->i_uid = 0;
 245 +       inode->i_gid = 0;
 246 +       inode->i_nlink = 1;
 247 +       inode->i_size = 4096;
 248 +       inode->i_atime = CURRENT_TIME;
 249 +       inode->i_ctime = CURRENT_TIME;
 250 +       inode->i_mtime = CURRENT_TIME;
 251 +       EXT3_I(inode)->i_dtime = 0;
 252 +       inode->i_blocks = 0;
 253 +       inode->i_version = 1;
 254 +       inode->i_generation = 0;
 255 +
 256 +       inode->i_op = &iopen_inode_operations;
 257 +       inode->i_fop = &iopen_file_operations;
 258 +       inode->i_mapping->a_ops = 0;
 259 +
 260 +       return 1;
 261 +}
 262 Index: linux-2.6.16.27-0.9/fs/ext3/iopen.h
 263 ===================================================================
 264 --- /dev/null   1970-01-01 00:00:00.000000000 +0000
 265 +++ linux-2.6.16.27-0.9/fs/ext3/iopen.h 2007-06-29 08:24:49.000000000 +0200
 266 @@ -0,0 +1,15 @@
 267 +/*
 268 + * iopen.h
 269 + *
 270 + * Special support for opening files by inode number.
 271 + *
 272 + * Copyright (C) 2001 by Theodore Ts'o (tytso@alum.mit.edu).
 273 + *
 274 + * This file may be redistributed under the terms of the GNU General
 275 + * Public License.
 276 + */
 277 +
 278 +extern int ext3_check_for_iopen(struct inode *dir, struct dentry *dentry);
 279 +extern int ext3_iopen_get_inode(struct inode *inode);
 280 +extern struct dentry *iopen_connect_dentry(struct dentry *dentry,
 281 +                                          struct inode *inode, int rehash);
 282 Index: linux-2.6.16.27-0.9/fs/ext3/inode.c
 283 ===================================================================
 284 --- linux-2.6.16.27-0.9.orig/fs/ext3/inode.c    2007-06-29 08:24:48.000000000 +0200
 285 +++ linux-2.6.16.27-0.9/fs/ext3/inode.c 2007-06-29 08:24:52.000000000 +0200
 286 @@ -37,6 +37,7 @@
 287  #include <linux/uio.h>
 288  #include <linux/bio.h>
 289  #include "xattr.h"
 290 +#include "iopen.h"
 291  #include "acl.h"
 292
 293  static int ext3_writepage_trans_blocks(struct inode *inode);
 294 @@ -2448,6 +2449,8 @@ void ext3_read_inode(struct inode * inod
 295         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 296  #endif
 297         ei->i_block_alloc_info = NULL;
 298 +       if (ext3_iopen_get_inode(inode))
 299 +               return;
 300
 301         if (__ext3_get_inode_loc(inode, &iloc, 0))
 302                 goto bad_inode;
 303 Index: linux-2.6.16.27-0.9/fs/ext3/super.c
 304 ===================================================================
 305 --- linux-2.6.16.27-0.9.orig/fs/ext3/super.c    2007-06-29 08:24:48.000000000 +0200
 306 +++ linux-2.6.16.27-0.9/fs/ext3/super.c 2007-06-29 08:24:52.000000000 +0200
 307 @@ -678,6 +678,7 @@ enum {
 308         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 309         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 310         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 311 +       Opt_iopen, Opt_noiopen, Opt_iopen_nopriv,
 312         Opt_grpquota
 313  };
 314
 315 @@ -726,6 +727,9 @@ static match_table_t tokens = {
 316         {Opt_noquota, "noquota"},
 317         {Opt_quota, "quota"},
 318         {Opt_usrquota, "usrquota"},
 319 +       {Opt_iopen, "iopen"},
 320 +       {Opt_noiopen, "noiopen"},
 321 +       {Opt_iopen_nopriv, "iopen_nopriv"},
 322         {Opt_barrier, "barrier=%u"},
 323         {Opt_err, NULL},
 324         {Opt_resize, "resize"},
 325 @@ -1040,6 +1044,18 @@ clear_qf_name:
 326                         else
 327                                 clear_opt(sbi->s_mount_opt, BARRIER);
 328                         break;
 329 +               case Opt_iopen:
 330 +                       set_opt (sbi->s_mount_opt, IOPEN);
 331 +                       clear_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 332 +                       break;
 333 +               case Opt_noiopen:
 334 +                       clear_opt (sbi->s_mount_opt, IOPEN);
 335 +                       clear_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 336 +                       break;
 337 +               case Opt_iopen_nopriv:
 338 +                       set_opt (sbi->s_mount_opt, IOPEN);
 339 +                       set_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 340 +                       break;
 341                 case Opt_ignore:
 342                         break;
 343                 case Opt_resize:
 344 Index: linux-2.6.16.27-0.9/fs/ext3/namei.c
 345 ===================================================================
 346 --- linux-2.6.16.27-0.9.orig/fs/ext3/namei.c    2007-06-29 08:24:47.000000000 +0200
 347 +++ linux-2.6.16.27-0.9/fs/ext3/namei.c 2007-06-29 08:24:49.000000000 +0200
 348 @@ -39,6 +39,7 @@
 349
 350  #include "namei.h"
 351  #include "xattr.h"
 352 +#include "iopen.h"
 353  #include "acl.h"
 354
 355  /*
 356 @@ -1004,6 +1005,9 @@ static struct dentry *ext3_lookup(struct
 357         if (dentry->d_name.len > EXT3_NAME_LEN)
 358                 return ERR_PTR(-ENAMETOOLONG);
 359
 360 +       if (ext3_check_for_iopen(dir, dentry))
 361 +               return NULL;
 362 +
 363         bh = ext3_find_entry(dentry, &de);
 364         inode = NULL;
 365         if (bh) {
 366 @@ -1014,7 +1018,7 @@ static struct dentry *ext3_lookup(struct
 367                 if (!inode)
 368                         return ERR_PTR(-EACCES);
 369         }
 370 -       return d_splice_alias(inode, dentry);
 371 +       return iopen_connect_dentry(dentry, inode, 1);
 372  }
 373
 374
 375 @@ -2058,10 +2062,6 @@ static int ext3_rmdir (struct inode * di
 376                               inode->i_nlink);
 377         inode->i_version++;
 378         clear_nlink(inode);
 379 -       /* There's no need to set i_disksize: the fact that i_nlink is
 380 -        * zero will ensure that the right thing happens during any
 381 -        * recovery. */
 382 -       inode->i_size = 0;
 383         ext3_orphan_add(handle, inode);
 384         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
 385         ext3_mark_inode_dirty(handle, inode);
 386 @@ -2185,6 +2185,23 @@ out_stop:
 387         return err;
 388  }
 389
 390 +/* Like ext3_add_nondir() except for call to iopen_connect_dentry */
 391 +static int ext3_add_link(handle_t *handle, struct dentry *dentry,
 392 +                        struct inode *inode)
 393 +{
 394 +       int err = ext3_add_entry(handle, dentry, inode);
 395 +       if (!err) {
 396 +               err = ext3_mark_inode_dirty(handle, inode);
 397 +               if (err == 0) {
 398 +                       dput(iopen_connect_dentry(dentry, inode, 0));
 399 +                       return 0;
 400 +               }
 401 +       }
 402 +       ext3_dec_count(handle, inode);
 403 +       iput(inode);
 404 +       return err;
 405 +}
 406 +
 407  static int ext3_link (struct dentry * old_dentry,
 408                 struct inode * dir, struct dentry *dentry)
 409  {
 410 @@ -2208,7 +2225,8 @@ retry:
 411         ext3_inc_count(handle, inode);
 412         atomic_inc(&inode->i_count);
 413
 414 -       err = ext3_add_nondir(handle, dentry, inode);
 415 +       err = ext3_add_link(handle, dentry, inode);
 416 +       ext3_orphan_del(handle, inode);
 417         ext3_journal_stop(handle);
 418         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
 419                 goto retry;
 420 Index: linux-2.6.16.27-0.9/fs/ext3/Makefile
 421 ===================================================================
 422 --- linux-2.6.16.27-0.9.orig/fs/ext3/Makefile   2007-03-13 00:56:52.000000000 +0100
 423 +++ linux-2.6.16.27-0.9/fs/ext3/Makefile        2007-06-29 08:24:49.000000000 +0200
 424 @@ -4,7 +4,7 @@
 425
 426  obj-$(CONFIG_EXT3_FS) += ext3.o
 427
 428 -ext3-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o \
 429 +ext3-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o iopen.o \
 430            ioctl.o namei.o super.o symlink.o hash.o resize.o ext3_jbd.o
 431
 432  ext3-$(CONFIG_EXT3_FS_XATTR)    += xattr.o xattr_user.o xattr_trusted.o
 433 Index: linux-2.6.16.27-0.9/include/linux/ext3_fs.h
 434 ===================================================================
 435 --- linux-2.6.16.27-0.9.orig/include/linux/ext3_fs.h    2007-06-29 08:24:47.000000000 +0200
 436 +++ linux-2.6.16.27-0.9/include/linux/ext3_fs.h 2007-06-29 08:24:49.000000000 +0200
 437 @@ -375,6 +375,8 @@ struct ext3_inode {
 438  #define EXT3_MOUNT_QUOTA               0x80000 /* Some quota option set */
 439  #define EXT3_MOUNT_USRQUOTA            0x100000 /* "old" user quota */
 440  #define EXT3_MOUNT_GRPQUOTA            0x200000 /* "old" group quota */
 441 +#define EXT3_MOUNT_IOPEN               0x400000        /* Allow access via iopen */
 442 +#define EXT3_MOUNT_IOPEN_NOPRIV                0x800000/* Make iopen world-readable */
 443
 444  /* Compatibility, for having both ext2_fs.h and ext3_fs.h included at once */
 445  #ifndef _LINUX_EXT2_FS_H