lustre/kernel_patches/patches/iopen-2.4.21-chaos.patch

   1  Documentation/filesystems/ext2.txt |   16 ++
   2  fs/ext3/Makefile                   |    2
   3  fs/ext3/inode.c                    |    4
   4  fs/ext3/iopen.c                    |  259 +++++++++++++++++++++++++++++++++++++
   5  fs/ext3/iopen.h                    |   13 +
   6  fs/ext3/namei.c                    |   13 +
   7  fs/ext3/super.c                    |   11 +
   8  include/linux/ext3_fs.h            |    2
   9  8 files changed, 318 insertions(+), 2 deletions(-)
  10
  11 Index: linux-ia64/Documentation/filesystems/ext2.txt
  12 ===================================================================
  13 --- linux-ia64.orig/Documentation/filesystems/ext2.txt  2004-03-17 15:47:15.000000000 -0800
  14 +++ linux-ia64/Documentation/filesystems/ext2.txt       2004-03-17 18:03:15.000000000 -0800
  15 @@ -35,6 +35,22 @@ resgid=n                     The group ID which may use th
  16
  17  sb=n                           Use alternate superblock at this location.
  18
  19 +iopen                          Makes an invisible pseudo-directory called
  20 +                               __iopen__ available in the root directory
  21 +                               of the filesystem.  Allows open-by-inode-
  22 +                               number.  i.e., inode 3145 can be accessed
  23 +                               via /mntpt/__iopen__/3145
  24 +
  25 +iopen_nopriv                   This option makes the iopen directory be
  26 +                               world-readable.  This may be safer since it
  27 +                               allows daemons to run as an unprivileged user,
  28 +                               however it significantly changes the security
  29 +                               model of a Unix filesystem, since previously
  30 +                               all files under a mode 700 directory were not
  31 +                               generally avilable even if the
  32 +                               permissions on the file itself is
  33 +                               world-readable.
  34 +
  35  grpquota,noquota,quota,usrquota        Quota options are silently ignored by ext2.
  36
  37
  38 Index: linux-ia64/fs/ext3/Makefile
  39 ===================================================================
  40 --- linux-ia64.orig/fs/ext3/Makefile    2004-03-17 18:03:14.000000000 -0800
  41 +++ linux-ia64/fs/ext3/Makefile 2004-03-17 18:03:15.000000000 -0800
  42 @@ -11,7 +11,7 @@ O_TARGET := ext3.o
  43
  44  export-objs := ext3-exports.o
  45
  46 -obj-y    := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o \
  47 +obj-y    := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o iopen.o \
  48                 ioctl.o namei.o super.o symlink.o hash.o ext3-exports.o
  49  obj-m    := $(O_TARGET)
  50
  51 Index: linux-ia64/fs/ext3/inode.c
  52 ===================================================================
  53 --- linux-ia64.orig/fs/ext3/inode.c     2004-03-17 18:03:15.000000000 -0800
  54 +++ linux-ia64/fs/ext3/inode.c  2004-03-17 18:10:36.000000000 -0800
  55 @@ -34,6 +34,7 @@
  56  #include <linux/highuid.h>
  57  #include <linux/quotaops.h>
  58  #include <linux/module.h>
  59 +#include "iopen.h"
  60
  61  /*
  62   * SEARCH_FROM_ZERO forces each block allocation to search from the start
  63 @@ -2430,6 +2431,9 @@ void ext3_read_inode(struct inode * inod
  64         struct buffer_head *bh;
  65         int block;
  66
  67 +       if (ext3_iopen_get_inode(inode))
  68 +               return;
  69 +
  70         if(ext3_get_inode_loc(inode, &iloc))
  71                 goto bad_inode;
  72         bh = iloc.bh;
  73 Index: linux-ia64/fs/ext3/iopen.c
  74 ===================================================================
  75 --- linux-ia64.orig/fs/ext3/iopen.c     2004-03-17 18:02:08.000000000 -0800
  76 +++ linux-ia64/fs/ext3/iopen.c  2004-03-17 18:10:58.000000000 -0800
  77 @@ -8,3 +8,275 @@
  78   * This file may be redistributed under the terms of the GNU General
  79   * Public License.
  80   *
  81 + *
  82 + * Invariants:
  83 + *   - there is only ever a single DCACHE_NFSD_DISCONNECTED dentry alias
  84 + *     for an inode at one time.
  85 + *   - there are never both connected and DCACHE_NFSD_DISCONNECTED dentry
  86 + *     aliases on an inode at the same time.
  87 + *
  88 + * If we have any connected dentry aliases for an inode, use one of those
  89 + * in iopen_lookup().  Otherwise, we instantiate a single NFSD_DISCONNECTED
  90 + * dentry for this inode, which thereafter will be found by the dcache
  91 + * when looking up this inode number in __iopen__, so we don't return here
  92 + * until it is gone.
  93 + *
  94 + * If we get an inode via a regular name lookup, then we "rename" the
  95 + * NFSD_DISCONNECTED dentry to the proper name and parent.  This ensures
  96 + * existing users of the disconnected dentry will continue to use the same
  97 + * dentry as the connected users, and there will never be both kinds of
  98 + * dentry aliases at one time.
  99 + */
 100 +
 101 +#include <linux/sched.h>
 102 +#include <linux/fs.h>
 103 +#include <linux/locks.h>
 104 +#include <linux/ext3_jbd.h>
 105 +#include <linux/jbd.h>
 106 +#include <linux/ext3_fs.h>
 107 +#include <linux/smp_lock.h>
 108 +#include "iopen.h"
 109 +
 110 +#ifndef assert
 111 +#define assert(test) J_ASSERT(test)
 112 +#endif
 113 +
 114 +#define IOPEN_NAME_LEN 32
 115 +
 116 +/*
 117 + * This implements looking up an inode by number.
 118 + */
 119 +static struct dentry *iopen_lookup(struct inode *dir, struct dentry *dentry)
 120 +{
 121 +       struct inode *inode;
 122 +       unsigned long ino;
 123 +       struct list_head *lp;
 124 +       struct dentry *alternate;
 125 +       char buf[IOPEN_NAME_LEN];
 126 +
 127 +       if (dentry->d_name.len >= IOPEN_NAME_LEN)
 128 +               return ERR_PTR(-ENAMETOOLONG);
 129 +
 130 +       memcpy(buf, dentry->d_name.name, dentry->d_name.len);
 131 +       buf[dentry->d_name.len] = 0;
 132 +
 133 +       if (strcmp(buf, ".") == 0)
 134 +               ino = dir->i_ino;
 135 +       else if (strcmp(buf, "..") == 0)
 136 +               ino = EXT3_ROOT_INO;
 137 +       else
 138 +               ino = simple_strtoul(buf, 0, 0);
 139 +
 140 +       if ((ino != EXT3_ROOT_INO &&
 141 +            //ino != EXT3_ACL_IDX_INO &&
 142 +            //ino != EXT3_ACL_DATA_INO &&
 143 +            ino < EXT3_FIRST_INO(dir->i_sb)) ||
 144 +           ino > le32_to_cpu(dir->i_sb->u.ext3_sb.s_es->s_inodes_count))
 145 +               return ERR_PTR(-ENOENT);
 146 +
 147 +       inode = iget(dir->i_sb, ino);
 148 +       if (!inode)
 149 +               return ERR_PTR(-EACCES);
 150 +       if (is_bad_inode(inode)) {
 151 +               iput(inode);
 152 +               return ERR_PTR(-ENOENT);
 153 +       }
 154 +
 155 +       assert(list_empty(&dentry->d_alias));           /* d_instantiate */
 156 +       assert(list_empty(&dentry->d_hash));            /* d_rehash */
 157 +
 158 +       /* preferrably return a connected dentry */
 159 +       spin_lock(&dcache_lock);
 160 +       list_for_each(lp, &inode->i_dentry) {
 161 +               alternate = list_entry(lp, struct dentry, d_alias);
 162 +               assert(!(alternate->d_flags & DCACHE_NFSD_DISCONNECTED));
 163 +       }
 164 +
 165 +       if (!list_empty(&inode->i_dentry)) {
 166 +               alternate = list_entry(inode->i_dentry.next,
 167 +                                      struct dentry, d_alias);
 168 +               dget_locked(alternate);
 169 +               alternate->d_vfs_flags |= DCACHE_REFERENCED;
 170 +               iput(inode);
 171 +               spin_unlock(&dcache_lock);
 172 +               return alternate;
 173 +       }
 174 +       dentry->d_flags |= DCACHE_NFSD_DISCONNECTED;
 175 +
 176 +       /* d_add(), but don't drop dcache_lock before adding dentry to inode */
 177 +       list_add(&dentry->d_alias, &inode->i_dentry);   /* d_instantiate */
 178 +       dentry->d_inode = inode;
 179 +
 180 +       __d_rehash(dentry, 0);                          /* d_rehash */
 181 +       spin_unlock(&dcache_lock);
 182 +
 183 +       return NULL;
 184 +}
 185 +
 186 +#define do_switch(x,y) do { \
 187 +       __typeof__ (x) __tmp = x; \
 188 +       x = y; y = __tmp; } while (0)
 189 +
 190 +static inline void switch_names(struct dentry *dentry, struct dentry *target)
 191 +{
 192 +       const unsigned char *old_name, *new_name;
 193 +
 194 +       memcpy(dentry->d_iname, target->d_iname, DNAME_INLINE_LEN);
 195 +       old_name = target->d_name.name;
 196 +       new_name = dentry->d_name.name;
 197 +       if (old_name == target->d_iname)
 198 +               old_name = dentry->d_iname;
 199 +       if (new_name == dentry->d_iname)
 200 +               new_name = target->d_iname;
 201 +       target->d_name.name = new_name;
 202 +       dentry->d_name.name = old_name;
 203 +}
 204 +
 205 +/* This function is spliced into ext3_lookup and does the move of a
 206 + * disconnected dentry (if it exists) to a connected dentry.
 207 + */
 208 +struct dentry *iopen_connect_dentry(struct dentry *dentry, struct inode *inode,
 209 +                                   int rehash)
 210 +{
 211 +       struct dentry *tmp, *goal = NULL;
 212 +       struct list_head *lp;
 213 +
 214 +       /* verify this dentry is really new */
 215 +       assert(dentry->d_inode == NULL);
 216 +       assert(list_empty(&dentry->d_alias));           /* d_instantiate */
 217 +       if (rehash)
 218 +               assert(list_empty(&dentry->d_hash));    /* d_rehash */
 219 +       assert(list_empty(&dentry->d_subdirs));
 220 +
 221 +       spin_lock(&dcache_lock);
 222 +       if (!inode)
 223 +               goto do_rehash;
 224 +
 225 +       /* preferrably return a connected dentry */
 226 +       list_for_each(lp, &inode->i_dentry) {
 227 +               tmp = list_entry(lp, struct dentry, d_alias);
 228 +               if (tmp->d_flags & DCACHE_NFSD_DISCONNECTED) {
 229 +                       assert(tmp->d_alias.next == &inode->i_dentry);
 230 +                       assert(tmp->d_alias.prev == &inode->i_dentry);
 231 +                       goal = tmp;
 232 +                       dget_locked(goal);
 233 +                       break;
 234 +               }
 235 +       }
 236 +
 237 +       if (!goal)
 238 +               goto do_instantiate;
 239 +
 240 +       /* Move the goal to the de hash queue - like d_move() */
 241 +       goal->d_flags &= ~DCACHE_NFSD_DISCONNECTED;
 242 +       list_del_init(&goal->d_hash);
 243 +
 244 +       list_del(&goal->d_child);
 245 +       list_del(&dentry->d_child);
 246 +
 247 +       /* Switch the parents and the names.. */
 248 +       switch_names(goal, dentry);
 249 +       do_switch(goal->d_parent, dentry->d_parent);
 250 +       do_switch(goal->d_name.len, dentry->d_name.len);
 251 +       do_switch(goal->d_name.hash, dentry->d_name.hash);
 252 +
 253 +       /* And add them back to the (new) parent lists */
 254 +       list_add(&goal->d_child, &goal->d_parent->d_subdirs);
 255 +       list_add(&dentry->d_child, &dentry->d_parent->d_subdirs);
 256 +       __d_rehash(goal, 0);
 257 +       spin_unlock(&dcache_lock);
 258 +       iput(inode);
 259 +
 260 +       return goal;
 261 +
 262 +       /* d_add(), but don't drop dcache_lock before adding dentry to inode */
 263 +do_instantiate:
 264 +       list_add(&dentry->d_alias, &inode->i_dentry);   /* d_instantiate */
 265 +       dentry->d_inode = inode;
 266 +do_rehash:
 267 +       if (rehash)
 268 +               __d_rehash(dentry, 0);                  /* d_rehash */
 269 +       spin_unlock(&dcache_lock);
 270 +
 271 +       return NULL;
 272 +}
 273 +
 274 +/*
 275 + * These are the special structures for the iopen pseudo directory.
 276 + */
 277 +
 278 +static struct inode_operations iopen_inode_operations = {
 279 +       lookup:         iopen_lookup,           /* BKL held */
 280 +};
 281 +
 282 +static struct file_operations iopen_file_operations = {
 283 +       read:           generic_read_dir,
 284 +};
 285 +
 286 +static int match_dentry(struct dentry *dentry, const char *name)
 287 +{
 288 +       int     len;
 289 +
 290 +       len = strlen(name);
 291 +       if (dentry->d_name.len != len)
 292 +               return 0;
 293 +       if (strncmp(dentry->d_name.name, name, len))
 294 +               return 0;
 295 +       return 1;
 296 +}
 297 +
 298 +/*
 299 + * This function is spliced into ext3_lookup and returns 1 the file
 300 + * name is __iopen__ and dentry has been filled in appropriately.
 301 + */
 302 +int ext3_check_for_iopen(struct inode *dir, struct dentry *dentry)
 303 +{
 304 +       struct inode *inode;
 305 +
 306 +       if (dir->i_ino != EXT3_ROOT_INO ||
 307 +           !test_opt(dir->i_sb, IOPEN) ||
 308 +           !match_dentry(dentry, "__iopen__"))
 309 +               return 0;
 310 +
 311 +       inode = iget(dir->i_sb, EXT3_BAD_INO);
 312 +
 313 +       if (!inode)
 314 +               return 0;
 315 +       d_add(dentry, inode);
 316 +       return 1;
 317 +}
 318 +
 319 +/*
 320 + * This function is spliced into read_inode; it returns 1 if inode
 321 + * number is the one for /__iopen__, in which case the inode is filled
 322 + * in appropriately.  Otherwise, this fuction returns 0.
 323 + */
 324 +int ext3_iopen_get_inode(struct inode *inode)
 325 +{
 326 +       if (inode->i_ino != EXT3_BAD_INO)
 327 +               return 0;
 328 +
 329 +       inode->i_mode = S_IFDIR | S_IRUSR | S_IXUSR;
 330 +       if (test_opt(inode->i_sb, IOPEN_NOPRIV))
 331 +               inode->i_mode |= 0777;
 332 +       inode->i_uid = 0;
 333 +       inode->i_gid = 0;
 334 +       inode->i_nlink = 1;
 335 +       inode->i_size = 4096;
 336 +       inode->i_atime = CURRENT_TIME;
 337 +       inode->i_ctime = CURRENT_TIME;
 338 +       inode->i_mtime = CURRENT_TIME;
 339 +       inode->u.ext3_i.i_dtime = 0;
 340 +       inode->i_blksize = PAGE_SIZE;   /* This is the optimal IO size
 341 +                                        * (for stat), not the fs block
 342 +                                        * size */
 343 +       inode->i_blocks = 0;
 344 +       inode->i_version = 1;
 345 +       inode->i_generation = 0;
 346 +
 347 +       inode->i_op = &iopen_inode_operations;
 348 +       inode->i_fop = &iopen_file_operations;
 349 +       inode->i_mapping->a_ops = 0;
 350 +
 351 +       return 1;
 352 +}
 353 Index: linux-ia64/fs/ext3/iopen.h
 354 ===================================================================
 355 --- linux-ia64.orig/fs/ext3/iopen.h     2004-03-17 15:47:15.000000000 -0800
 356 +++ linux-ia64/fs/ext3/iopen.h  2004-03-17 18:03:15.000000000 -0800
 357 @@ -0,0 +1,15 @@
 358 +/*
 359 + * iopen.h
 360 + *
 361 + * Special support for opening files by inode number.
 362 + *
 363 + * Copyright (C) 2001 by Theodore Ts'o (tytso@alum.mit.edu).
 364 + *
 365 + * This file may be redistributed under the terms of the GNU General
 366 + * Public License.
 367 + */
 368 +
 369 +extern int ext3_check_for_iopen(struct inode *dir, struct dentry *dentry);
 370 +extern int ext3_iopen_get_inode(struct inode *inode);
 371 +extern struct dentry *iopen_connect_dentry(struct dentry *dentry,
 372 +                                          struct inode *inode, int rehash);
 373 Index: linux-ia64/fs/ext3/namei.c
 374 ===================================================================
 375 --- linux-ia64.orig/fs/ext3/namei.c     2004-03-17 18:03:15.000000000 -0800
 376 +++ linux-ia64/fs/ext3/namei.c  2004-03-17 18:10:35.000000000 -0800
 377 @@ -36,7 +36,7 @@
 378  #include <linux/string.h>
 379  #include <linux/locks.h>
 380  #include <linux/quotaops.h>
 381 -
 382 +#include "iopen.h"
 383
 384  /*
 385   * define how far ahead to read directories while searching them.
 386 @@ -932,6 +932,9 @@ static struct dentry *ext3_lookup(struct
 387         if (dentry->d_name.len > EXT3_NAME_LEN)
 388                 return ERR_PTR(-ENAMETOOLONG);
 389
 390 +       if (ext3_check_for_iopen(dir, dentry))
 391 +               return NULL;
 392 +
 393         bh = ext3_find_entry(dentry, &de);
 394         inode = NULL;
 395         if (bh) {
 396 @@ -943,8 +946,8 @@ static struct dentry *ext3_lookup(struct
 397                         return ERR_PTR(-EACCES);
 398                 }
 399         }
 400 -       d_add(dentry, inode);
 401 -       return NULL;
 402 +
 403 +       return iopen_connect_dentry(dentry, inode, 1);
 404  }
 405
 406  #define S_SHIFT 12
 407 @@ -1935,10 +1938,6 @@ static int ext3_rmdir (struct inode * di
 408                               inode->i_nlink);
 409         inode->i_version = ++event;
 410         inode->i_nlink = 0;
 411 -       /* There's no need to set i_disksize: the fact that i_nlink is
 412 -        * zero will ensure that the right thing happens during any
 413 -        * recovery. */
 414 -       inode->i_size = 0;
 415         ext3_orphan_add(handle, inode);
 416         dir->i_nlink--;
 417         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 418 @@ -2057,6 +2056,23 @@ out_stop:
 419         return err;
 420  }
 421
 422 +/* Like ext3_add_nondir() except for call to iopen_connect_dentry */
 423 +static int ext3_add_link(handle_t *handle, struct dentry *dentry,
 424 +                        struct inode *inode)
 425 +{
 426 +       int err = ext3_add_entry(handle, dentry, inode);
 427 +       if (!err) {
 428 +               err = ext3_mark_inode_dirty(handle, inode);
 429 +               if (err == 0) {
 430 +                       (void)iopen_connect_dentry(dentry, inode, 0);
 431 +                       return 0;
 432 +               }
 433 +       }
 434 +       ext3_dec_count(handle, inode);
 435 +       iput(inode);
 436 +       return err;
 437 +}
 438 +
 439  static int ext3_link (struct dentry * old_dentry,
 440                 struct inode * dir, struct dentry *dentry)
 441  {
 442 @@ -2084,7 +2100,8 @@ static int ext3_link (struct dentry * ol
 443         ext3_inc_count(handle, inode);
 444         atomic_inc(&inode->i_count);
 445
 446 -       err = ext3_add_nondir(handle, dentry, inode);
 447 +       err = ext3_add_link(handle, dentry, inode);
 448 +       ext3_orphan_del(handle, inode);
 449         ext3_journal_stop(handle, dir);
 450         return err;
 451  }
 452 Index: linux-ia64/fs/ext3/super.c
 453 ===================================================================
 454 --- linux-ia64.orig/fs/ext3/super.c     2004-03-17 18:03:14.000000000 -0800
 455 +++ linux-ia64/fs/ext3/super.c  2004-03-17 18:10:35.000000000 -0800
 456 @@ -891,6 +891,18 @@ static int parse_options (char * options
 457                          || !strcmp (this_char, "quota")
 458                          || !strcmp (this_char, "usrquota"))
 459                         /* Don't do anything ;-) */ ;
 460 +               else if (!strcmp (this_char, "iopen")) {
 461 +                       set_opt (sbi->s_mount_opt, IOPEN);
 462 +                       clear_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 463 +               }
 464 +               else if (!strcmp (this_char, "noiopen")) {
 465 +                       clear_opt (sbi->s_mount_opt, IOPEN);
 466 +                       clear_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 467 +               }
 468 +               else if (!strcmp (this_char, "iopen_nopriv")) {
 469 +                       set_opt (sbi->s_mount_opt, IOPEN);
 470 +                       set_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 471 +               }
 472                 else if (!strcmp (this_char, "journal")) {
 473                         /* @@@ FIXME */
 474                         /* Eventually we will want to be able to create
 475 Index: linux-ia64/include/linux/ext3_fs.h
 476 ===================================================================
 477 --- linux-ia64.orig/include/linux/ext3_fs.h     2004-03-17 18:03:15.000000000 -0800
 478 +++ linux-ia64/include/linux/ext3_fs.h  2004-03-17 18:03:15.000000000 -0800
 479 @@ -328,6 +328,8 @@ struct ext3_inode {
 480  #define EXT3_MOUNT_XATTR_USER          0x4000  /* Extended user attributes */
 481  #define EXT3_MOUNT_POSIX_ACL           0x8000  /* POSIX Access Control Lists */
 482  #define EXT3_MOUNT_ASYNCDEL            0x20000 /* Delayed deletion */
 483 +#define EXT3_MOUNT_IOPEN               0x40000 /* Allow access via iopen */
 484 +#define EXT3_MOUNT_IOPEN_NOPRIV                0x80000 /* Make iopen world-readable */
 485
 486  /* Compatibility, for having both ext2_fs.h and ext3_fs.h included at once */
 487  #ifndef _LINUX_EXT2_FS_H