lustre/kernel_patches/patches/iopen-2.6-fc5.patch

   1 Index: linux-2.6.16.i686/fs/ext3/iopen.c
   2 ===================================================================
   3 --- linux-2.6.16.i686.orig/fs/ext3/iopen.c      2006-05-31 04:14:15.752410384 +0800
   4 +++ linux-2.6.16.i686/fs/ext3/iopen.c   2006-05-30 22:52:38.000000000 +0800
   5 @@ -0,0 +1,259 @@
   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 +       spin_unlock(&dcache_lock);
 119 +
 120 +       d_rehash(dentry);
 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 +       spin_unlock(&dcache_lock);
 168 +       d_rehash(dentry);
 169 +       d_move(goal, dentry);
 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 +       spin_unlock(&dcache_lock);
 180 +       if (rehash)
 181 +               d_rehash(dentry);
 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_blksize = PAGE_SIZE;   /* This is the optimal IO size
 253 +                                        * (for stat), not the fs block
 254 +                                        * size */
 255 +       inode->i_blocks = 0;
 256 +       inode->i_version = 1;
 257 +       inode->i_generation = 0;
 258 +
 259 +       inode->i_op = &iopen_inode_operations;
 260 +       inode->i_fop = &iopen_file_operations;
 261 +       inode->i_mapping->a_ops = 0;
 262 +
 263 +       return 1;
 264 +}
 265 Index: linux-2.6.16.i686/fs/ext3/iopen.h
 266 ===================================================================
 267 --- linux-2.6.16.i686.orig/fs/ext3/iopen.h      2006-05-31 04:14:15.752410384 +0800
 268 +++ linux-2.6.16.i686/fs/ext3/iopen.h   2006-05-30 22:52:38.000000000 +0800
 269 @@ -0,0 +1,15 @@
 270 +/*
 271 + * iopen.h
 272 + *
 273 + * Special support for opening files by inode number.
 274 + *
 275 + * Copyright (C) 2001 by Theodore Ts'o (tytso@alum.mit.edu).
 276 + *
 277 + * This file may be redistributed under the terms of the GNU General
 278 + * Public License.
 279 + */
 280 +
 281 +extern int ext3_check_for_iopen(struct inode *dir, struct dentry *dentry);
 282 +extern int ext3_iopen_get_inode(struct inode *inode);
 283 +extern struct dentry *iopen_connect_dentry(struct dentry *dentry,
 284 +                                          struct inode *inode, int rehash);
 285 Index: linux-2.6.16.i686/fs/ext3/inode.c
 286 ===================================================================
 287 --- linux-2.6.16.i686.orig/fs/ext3/inode.c      2006-05-30 22:52:03.000000000 +0800
 288 +++ linux-2.6.16.i686/fs/ext3/inode.c   2006-05-30 22:52:38.000000000 +0800
 289 @@ -37,6 +37,7 @@
 290  #include <linux/mpage.h>
 291  #include <linux/uio.h>
 292  #include "xattr.h"
 293 +#include "iopen.h"
 294  #include "acl.h"
 295
 296  static int ext3_writepage_trans_blocks(struct inode *inode);
 297 @@ -2448,6 +2449,8 @@
 298         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 299  #endif
 300         ei->i_block_alloc_info = NULL;
 301 +       if (ext3_iopen_get_inode(inode))
 302 +               return;
 303
 304         if (__ext3_get_inode_loc(inode, &iloc, 0))
 305                 goto bad_inode;
 306 Index: linux-2.6.16.i686/fs/ext3/super.c
 307 ===================================================================
 308 --- linux-2.6.16.i686.orig/fs/ext3/super.c      2006-05-30 22:52:03.000000000 +0800
 309 +++ linux-2.6.16.i686/fs/ext3/super.c   2006-05-30 22:52:38.000000000 +0800
 310 @@ -634,6 +634,7 @@
 311         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 312         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_quota, Opt_noquota,
 313         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
 314 +       Opt_iopen, Opt_noiopen, Opt_iopen_nopriv,
 315         Opt_grpquota
 316  };
 317
 318 @@ -682,6 +683,9 @@
 319         {Opt_noquota, "noquota"},
 320         {Opt_quota, "quota"},
 321         {Opt_usrquota, "usrquota"},
 322 +       {Opt_iopen, "iopen"},
 323 +       {Opt_noiopen, "noiopen"},
 324 +       {Opt_iopen_nopriv, "iopen_nopriv"},
 325         {Opt_barrier, "barrier=%u"},
 326         {Opt_err, NULL},
 327         {Opt_resize, "resize"},
 328 @@ -996,6 +1000,18 @@
 329                         else
 330                                 clear_opt(sbi->s_mount_opt, BARRIER);
 331                         break;
 332 +               case Opt_iopen:
 333 +                       set_opt (sbi->s_mount_opt, IOPEN);
 334 +                       clear_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 335 +                       break;
 336 +               case Opt_noiopen:
 337 +                       clear_opt (sbi->s_mount_opt, IOPEN);
 338 +                       clear_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 339 +                       break;
 340 +               case Opt_iopen_nopriv:
 341 +                       set_opt (sbi->s_mount_opt, IOPEN);
 342 +                       set_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 343 +                       break;
 344                 case Opt_ignore:
 345                         break;
 346                 case Opt_resize:
 347 Index: linux-2.6.16.i686/fs/ext3/namei.c
 348 ===================================================================
 349 --- linux-2.6.16.i686.orig/fs/ext3/namei.c      2006-05-30 22:52:00.000000000 +0800
 350 +++ linux-2.6.16.i686/fs/ext3/namei.c   2006-05-30 22:55:19.000000000 +0800
 351 @@ -39,6 +39,7 @@
 352
 353  #include "namei.h"
 354  #include "xattr.h"
 355 +#include "iopen.h"
 356  #include "acl.h"
 357
 358  /*
 359 @@ -995,6 +996,9 @@
 360         if (dentry->d_name.len > EXT3_NAME_LEN)
 361                 return ERR_PTR(-ENAMETOOLONG);
 362
 363 +       if (ext3_check_for_iopen(dir, dentry))
 364 +               return NULL;
 365 +
 366         bh = ext3_find_entry(dentry, &de);
 367         inode = NULL;
 368         if (bh) {
 369 @@ -1005,7 +1009,7 @@
 370                 if (!inode)
 371                         return ERR_PTR(-EACCES);
 372         }
 373 -       return d_splice_alias(inode, dentry);
 374 +       return iopen_connect_dentry(dentry, inode, 1);
 375  }
 376
 377
 378 @@ -2046,10 +2050,6 @@
 379                               inode->i_nlink);
 380         inode->i_version++;
 381         inode->i_nlink = 0;
 382 -       /* There's no need to set i_disksize: the fact that i_nlink is
 383 -        * zero will ensure that the right thing happens during any
 384 -        * recovery. */
 385 -       inode->i_size = 0;
 386         ext3_orphan_add(handle, inode);
 387         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
 388         ext3_mark_inode_dirty(handle, inode);
 389 @@ -2173,6 +2173,23 @@
 390         return err;
 391  }
 392
 393 +/* Like ext3_add_nondir() except for call to iopen_connect_dentry */
 394 +static int ext3_add_link(handle_t *handle, struct dentry *dentry,
 395 +                        struct inode *inode)
 396 +{
 397 +       int err = ext3_add_entry(handle, dentry, inode);
 398 +       if (!err) {
 399 +               err = ext3_mark_inode_dirty(handle, inode);
 400 +               if (err == 0) {
 401 +                       dput(iopen_connect_dentry(dentry, inode, 0));
 402 +                       return 0;
 403 +               }
 404 +       }
 405 +       ext3_dec_count(handle, inode);
 406 +       iput(inode);
 407 +       return err;
 408 +}
 409 +
 410  static int ext3_link (struct dentry * old_dentry,
 411                 struct inode * dir, struct dentry *dentry)
 412  {
 413 @@ -2196,7 +2213,8 @@
 414         ext3_inc_count(handle, inode);
 415         atomic_inc(&inode->i_count);
 416
 417 -       err = ext3_add_nondir(handle, dentry, inode);
 418 +       err = ext3_add_link(handle, dentry, inode);
 419 +       ext3_orphan_del(handle, inode);
 420         ext3_journal_stop(handle);
 421         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
 422                 goto retry;
 423 Index: linux-2.6.16.i686/fs/ext3/Makefile
 424 ===================================================================
 425 --- linux-2.6.16.i686.orig/fs/ext3/Makefile     2006-03-20 13:53:29.000000000 +0800
 426 +++ linux-2.6.16.i686/fs/ext3/Makefile  2006-05-30 22:52:38.000000000 +0800
 427 @@ -4,7 +4,7 @@
 428
 429  obj-$(CONFIG_EXT3_FS) += ext3.o
 430
 431 -ext3-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o \
 432 +ext3-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o iopen.o \
 433            ioctl.o namei.o super.o symlink.o hash.o resize.o
 434
 435  ext3-$(CONFIG_EXT3_FS_XATTR)    += xattr.o xattr_user.o xattr_trusted.o
 436 Index: linux-2.6.16.i686/include/linux/ext3_fs.h
 437 ===================================================================
 438 --- linux-2.6.16.i686.orig/include/linux/ext3_fs.h      2006-05-30 22:52:00.000000000 +0800
 439 +++ linux-2.6.16.i686/include/linux/ext3_fs.h   2006-05-30 22:52:38.000000000 +0800
 440 @@ -375,6 +375,8 @@
 441  #define EXT3_MOUNT_QUOTA               0x80000 /* Some quota option set */
 442  #define EXT3_MOUNT_USRQUOTA            0x100000 /* "old" user quota */
 443  #define EXT3_MOUNT_GRPQUOTA            0x200000 /* "old" group quota */
 444 +#define EXT3_MOUNT_IOPEN               0x400000        /* Allow access via iopen */
 445 +#define EXT3_MOUNT_IOPEN_NOPRIV                0x800000/* Make iopen world-readable */
 446
 447  /* Compatibility, for having both ext2_fs.h and ext3_fs.h included at once */
 448  #ifndef _LINUX_EXT2_FS_H