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

   1 Index: linux-stage/fs/ext3/Makefile
   2 ===================================================================
   3 --- linux-stage.orig/fs/ext3/Makefile   2005-02-25 14:31:53.151076368 +0200
   4 +++ linux-stage/fs/ext3/Makefile        2005-02-25 14:41:51.259150120 +0200
   5 @@ -4,7 +4,7 @@
   6
   7  obj-$(CONFIG_EXT3_FS) += ext3.o
   8
   9 -ext3-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o \
  10 +ext3-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o iopen.o \
  11            ioctl.o namei.o super.o symlink.o hash.o
  12
  13  ext3-$(CONFIG_EXT3_FS_XATTR)    += xattr.o xattr_user.o xattr_trusted.o
  14 Index: linux-stage/fs/ext3/inode.c
  15 ===================================================================
  16 --- linux-stage.orig/fs/ext3/inode.c    2005-02-25 14:37:30.983718000 +0200
  17 +++ linux-stage/fs/ext3/inode.c 2005-02-25 14:47:42.069818792 +0200
  18 @@ -37,6 +37,7 @@
  19  #include <linux/mpage.h>
  20  #include <linux/uio.h>
  21  #include "xattr.h"
  22 +#include "iopen.h"
  23  #include "acl.h"
  24
  25  /*
  26 @@ -2408,6 +2409,9 @@
  27  #endif
  28         ei->i_rsv_window.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
  29
  30 +       if (ext3_iopen_get_inode(inode))
  31 +               return;
  32 +
  33         if (ext3_get_inode_loc(inode, &iloc, 0))
  34                 goto bad_inode;
  35         bh = iloc.bh;
  36 Index: linux-stage/fs/ext3/iopen.c
  37 ===================================================================
  38 --- linux-2.6.5-sles9.orig/fs/ext3/iopen.c      2003-01-30 13:24:37.000000000 +0300
  39 +++ linux-2.6.5-sles9/fs/ext3/iopen.c   2004-11-09 02:18:27.611913312 +0300
  40 @@ -0,0 +1,278 @@
  41 +/*
  42 + * linux/fs/ext3/iopen.c
  43 + *
  44 + * Special support for open by inode number
  45 + *
  46 + * Copyright (C) 2001 by Theodore Ts'o (tytso@alum.mit.edu).
  47 + *
  48 + * This file may be redistributed under the terms of the GNU General
  49 + * Public License.
  50 + *
  51 + *
  52 + * Invariants:
  53 + *   - there is only ever a single DCACHE_NFSD_DISCONNECTED dentry alias
  54 + *     for an inode at one time.
  55 + *   - there are never both connected and DCACHE_NFSD_DISCONNECTED dentry
  56 + *     aliases on an inode at the same time.
  57 + *
  58 + * If we have any connected dentry aliases for an inode, use one of those
  59 + * in iopen_lookup().  Otherwise, we instantiate a single NFSD_DISCONNECTED
  60 + * dentry for this inode, which thereafter will be found by the dcache
  61 + * when looking up this inode number in __iopen__, so we don't return here
  62 + * until it is gone.
  63 + *
  64 + * If we get an inode via a regular name lookup, then we "rename" the
  65 + * NFSD_DISCONNECTED dentry to the proper name and parent.  This ensures
  66 + * existing users of the disconnected dentry will continue to use the same
  67 + * dentry as the connected users, and there will never be both kinds of
  68 + * dentry aliases at one time.
  69 + */
  70 +
  71 +#include <linux/sched.h>
  72 +#include <linux/fs.h>
  73 +#include <linux/ext3_jbd.h>
  74 +#include <linux/jbd.h>
  75 +#include <linux/ext3_fs.h>
  76 +#include <linux/smp_lock.h>
  77 +#include <linux/dcache.h>
  78 +#include <linux/security.h>
  79 +#include "iopen.h"
  80 +
  81 +#ifndef assert
  82 +#define assert(test) J_ASSERT(test)
  83 +#endif
  84 +
  85 +#define IOPEN_NAME_LEN 32
  86 +
  87 +/*
  88 + * This implements looking up an inode by number.
  89 + */
  90 +static struct dentry *iopen_lookup(struct inode * dir, struct dentry *dentry,
  91 +                                  struct nameidata *nd)
  92 +{
  93 +       struct inode *inode;
  94 +       unsigned long ino;
  95 +       struct list_head *lp;
  96 +       struct dentry *alternate;
  97 +       char buf[IOPEN_NAME_LEN];
  98 +
  99 +       if (dentry->d_name.len >= IOPEN_NAME_LEN)
 100 +               return ERR_PTR(-ENAMETOOLONG);
 101 +
 102 +       memcpy(buf, dentry->d_name.name, dentry->d_name.len);
 103 +       buf[dentry->d_name.len] = 0;
 104 +
 105 +       if (strcmp(buf, ".") == 0)
 106 +               ino = dir->i_ino;
 107 +       else if (strcmp(buf, "..") == 0)
 108 +               ino = EXT3_ROOT_INO;
 109 +       else
 110 +               ino = simple_strtoul(buf, 0, 0);
 111 +
 112 +       if ((ino != EXT3_ROOT_INO &&
 113 +            //ino != EXT3_ACL_IDX_INO &&
 114 +            //ino != EXT3_ACL_DATA_INO &&
 115 +            ino < EXT3_FIRST_INO(dir->i_sb)) ||
 116 +           ino > le32_to_cpu(EXT3_SB(dir->i_sb)->s_es->s_inodes_count))
 117 +               return ERR_PTR(-ENOENT);
 118 +
 119 +       inode = iget(dir->i_sb, ino);
 120 +       if (!inode)
 121 +               return ERR_PTR(-EACCES);
 122 +       if (is_bad_inode(inode)) {
 123 +               iput(inode);
 124 +               return ERR_PTR(-ENOENT);
 125 +       }
 126 +
 127 +       assert(list_empty(&dentry->d_alias));           /* d_instantiate */
 128 +       assert(d_unhashed(dentry));                     /* d_rehash */
 129 +
 130 +       /* preferrably return a connected dentry */
 131 +       spin_lock(&dcache_lock);
 132 +       list_for_each(lp, &inode->i_dentry) {
 133 +               alternate = list_entry(lp, struct dentry, d_alias);
 134 +               assert(!(alternate->d_flags & DCACHE_DISCONNECTED));
 135 +       }
 136 +
 137 +       if (!list_empty(&inode->i_dentry)) {
 138 +               alternate = list_entry(inode->i_dentry.next,
 139 +                                      struct dentry, d_alias);
 140 +               dget_locked(alternate);
 141 +               spin_lock(&alternate->d_lock);
 142 +               alternate->d_vfs_flags |= DCACHE_REFERENCED;
 143 +               spin_unlock(&alternate->d_lock);
 144 +               iput(inode);
 145 +               spin_unlock(&dcache_lock);
 146 +               return alternate;
 147 +       }
 148 +       dentry->d_flags |= DCACHE_DISCONNECTED;
 149 +
 150 +       /* d_add(), but don't drop dcache_lock before adding dentry to inode */
 151 +       list_add(&dentry->d_alias, &inode->i_dentry);   /* d_instantiate */
 152 +       dentry->d_inode = inode;
 153 +
 154 +       __d_rehash(dentry, 0);                          /* d_rehash */
 155 +       spin_unlock(&dcache_lock);
 156 +
 157 +       return NULL;
 158 +}
 159 +
 160 +#define do_switch(x,y) do { \
 161 +       __typeof__ (x) __tmp = x; \
 162 +       x = y; y = __tmp; } while (0)
 163 +
 164 +static inline void switch_names(struct dentry *dentry, struct dentry *target)
 165 +{
 166 +       const unsigned char *old_name, *new_name;
 167 +
 168 +       memcpy(dentry->d_iname, target->d_iname, DNAME_INLINE_LEN);
 169 +       old_name = target->d_name.name;
 170 +       new_name = dentry->d_name.name;
 171 +       if (old_name == target->d_iname)
 172 +               old_name = dentry->d_iname;
 173 +       if (new_name == dentry->d_iname)
 174 +               new_name = target->d_iname;
 175 +       target->d_name.name = new_name;
 176 +       dentry->d_name.name = old_name;
 177 +}
 178 +
 179 +/* This function is spliced into ext3_lookup and does the move of a
 180 + * disconnected dentry (if it exists) to a connected dentry.
 181 + */
 182 +struct dentry *iopen_connect_dentry(struct dentry *dentry, struct inode *inode,
 183 +                                   int rehash)
 184 +{
 185 +       struct dentry *tmp, *goal = NULL;
 186 +       struct list_head *lp;
 187 +
 188 +       /* verify this dentry is really new */
 189 +       assert(dentry->d_inode == NULL);
 190 +       assert(list_empty(&dentry->d_alias));           /* d_instantiate */
 191 +       if (rehash)
 192 +               assert(d_unhashed(dentry));     /* d_rehash */
 193 +       assert(list_empty(&dentry->d_subdirs));
 194 +
 195 +       spin_lock(&dcache_lock);
 196 +       if (!inode)
 197 +               goto do_rehash;
 198 +
 199 +       if (!test_opt(inode->i_sb, IOPEN))
 200 +               goto do_instantiate;
 201 +
 202 +       /* preferrably return a connected dentry */
 203 +       list_for_each(lp, &inode->i_dentry) {
 204 +               tmp = list_entry(lp, struct dentry, d_alias);
 205 +               if (tmp->d_flags & DCACHE_DISCONNECTED) {
 206 +                       assert(tmp->d_alias.next == &inode->i_dentry);
 207 +                       assert(tmp->d_alias.prev == &inode->i_dentry);
 208 +                       goal = tmp;
 209 +                       dget_locked(goal);
 210 +                       break;
 211 +               }
 212 +       }
 213 +
 214 +       if (!goal)
 215 +               goto do_instantiate;
 216 +
 217 +       /* Move the goal to the de hash queue */
 218 +       goal->d_flags &= ~DCACHE_DISCONNECTED;
 219 +       security_d_instantiate(goal, inode);
 220 +       __d_drop(dentry);
 221 +       __d_rehash(dentry, 0);
 222 +       __d_move(goal, dentry);
 223 +       spin_unlock(&dcache_lock);
 224 +       iput(inode);
 225 +
 226 +       return goal;
 227 +
 228 +       /* d_add(), but don't drop dcache_lock before adding dentry to inode */
 229 +do_instantiate:
 230 +       list_add(&dentry->d_alias, &inode->i_dentry);   /* d_instantiate */
 231 +       dentry->d_inode = inode;
 232 +do_rehash:
 233 +       if (rehash)
 234 +               __d_rehash(dentry, 0);                  /* d_rehash */
 235 +       spin_unlock(&dcache_lock);
 236 +
 237 +       return NULL;
 238 +}
 239 +
 240 +/*
 241 + * These are the special structures for the iopen pseudo directory.
 242 + */
 243 +
 244 +static struct inode_operations iopen_inode_operations = {
 245 +       lookup:         iopen_lookup,           /* BKL held */
 246 +};
 247 +
 248 +static struct file_operations iopen_file_operations = {
 249 +       read:           generic_read_dir,
 250 +};
 251 +
 252 +static int match_dentry(struct dentry *dentry, const char *name)
 253 +{
 254 +       int     len;
 255 +
 256 +       len = strlen(name);
 257 +       if (dentry->d_name.len != len)
 258 +               return 0;
 259 +       if (strncmp(dentry->d_name.name, name, len))
 260 +               return 0;
 261 +       return 1;
 262 +}
 263 +
 264 +/*
 265 + * This function is spliced into ext3_lookup and returns 1 the file
 266 + * name is __iopen__ and dentry has been filled in appropriately.
 267 + */
 268 +int ext3_check_for_iopen(struct inode *dir, struct dentry *dentry)
 269 +{
 270 +       struct inode *inode;
 271 +
 272 +       if (dir->i_ino != EXT3_ROOT_INO ||
 273 +           !test_opt(dir->i_sb, IOPEN) ||
 274 +           !match_dentry(dentry, "__iopen__"))
 275 +               return 0;
 276 +
 277 +       inode = iget(dir->i_sb, EXT3_BAD_INO);
 278 +
 279 +       if (!inode)
 280 +               return 0;
 281 +       d_add(dentry, inode);
 282 +       return 1;
 283 +}
 284 +
 285 +/*
 286 + * This function is spliced into read_inode; it returns 1 if inode
 287 + * number is the one for /__iopen__, in which case the inode is filled
 288 + * in appropriately.  Otherwise, this fuction returns 0.
 289 + */
 290 +int ext3_iopen_get_inode(struct inode *inode)
 291 +{
 292 +       if (inode->i_ino != EXT3_BAD_INO)
 293 +               return 0;
 294 +
 295 +       inode->i_mode = S_IFDIR | S_IRUSR | S_IXUSR;
 296 +       if (test_opt(inode->i_sb, IOPEN_NOPRIV))
 297 +               inode->i_mode |= 0777;
 298 +       inode->i_uid = 0;
 299 +       inode->i_gid = 0;
 300 +       inode->i_nlink = 1;
 301 +       inode->i_size = 4096;
 302 +       inode->i_atime = CURRENT_TIME;
 303 +       inode->i_ctime = CURRENT_TIME;
 304 +       inode->i_mtime = CURRENT_TIME;
 305 +       EXT3_I(inode)->i_dtime = 0;
 306 +       inode->i_blksize = PAGE_SIZE;   /* This is the optimal IO size
 307 +                                        * (for stat), not the fs block
 308 +                                        * size */
 309 +       inode->i_blocks = 0;
 310 +       inode->i_version = 1;
 311 +       inode->i_generation = 0;
 312 +
 313 +       inode->i_op = &iopen_inode_operations;
 314 +       inode->i_fop = &iopen_file_operations;
 315 +       inode->i_mapping->a_ops = 0;
 316 +
 317 +       return 1;
 318 +}
 319 Index: linux-stage/fs/ext3/iopen.h
 320 ===================================================================
 321 --- linux-stage.orig/fs/ext3/iopen.h    2005-02-25 14:41:01.017787968 +0200
 322 +++ linux-stage/fs/ext3/iopen.h 2005-02-25 14:41:01.045783712 +0200
 323 @@ -0,0 +1,15 @@
 324 +/*
 325 + * iopen.h
 326 + *
 327 + * Special support for opening files by inode number.
 328 + *
 329 + * Copyright (C) 2001 by Theodore Ts'o (tytso@alum.mit.edu).
 330 + *
 331 + * This file may be redistributed under the terms of the GNU General
 332 + * Public License.
 333 + */
 334 +
 335 +extern int ext3_check_for_iopen(struct inode *dir, struct dentry *dentry);
 336 +extern int ext3_iopen_get_inode(struct inode *inode);
 337 +extern struct dentry *iopen_connect_dentry(struct dentry *dentry,
 338 +                                          struct inode *inode, int rehash);
 339 Index: linux-stage/fs/ext3/namei.c
 340 ===================================================================
 341 --- linux-stage.orig/fs/ext3/namei.c    2005-02-25 14:37:28.975023368 +0200
 342 +++ linux-stage/fs/ext3/namei.c 2005-02-25 14:46:43.090784968 +0200
 343 @@ -37,6 +37,7 @@
 344  #include <linux/buffer_head.h>
 345  #include <linux/smp_lock.h>
 346  #include "xattr.h"
 347 +#include "iopen.h"
 348  #include "acl.h"
 349
 350  /*
 351 @@ -980,6 +981,9 @@
 352         if (dentry->d_name.len > EXT3_NAME_LEN)
 353                 return ERR_PTR(-ENAMETOOLONG);
 354
 355 +       if (ext3_check_for_iopen(dir, dentry))
 356 +               return NULL;
 357 +
 358         bh = ext3_find_entry(dentry, &de);
 359         inode = NULL;
 360         if (bh) {
 361 @@ -990,10 +994,8 @@
 362                 if (!inode)
 363                         return ERR_PTR(-EACCES);
 364         }
 365 -       if (inode)
 366 -               return d_splice_alias(inode, dentry);
 367 -       d_add(dentry, inode);
 368 -       return NULL;
 369 +
 370 +       return iopen_connect_dentry(dentry, inode, 1);
 371  }
 372
 373
 374 @@ -2037,10 +2039,6 @@
 375                               inode->i_nlink);
 376         inode->i_version++;
 377         inode->i_nlink = 0;
 378 -       /* There's no need to set i_disksize: the fact that i_nlink is
 379 -        * zero will ensure that the right thing happens during any
 380 -        * recovery. */
 381 -       inode->i_size = 0;
 382         ext3_orphan_add(handle, inode);
 383         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 384         ext3_mark_inode_dirty(handle, inode);
 385 @@ -2163,6 +2161,23 @@
 386         return err;
 387  }
 388
 389 +/* Like ext3_add_nondir() except for call to iopen_connect_dentry */
 390 +static int ext3_add_link(handle_t *handle, struct dentry *dentry,
 391 +                        struct inode *inode)
 392 +{
 393 +       int err = ext3_add_entry(handle, dentry, inode);
 394 +       if (!err) {
 395 +               err = ext3_mark_inode_dirty(handle, inode);
 396 +               if (err == 0) {
 397 +                       dput(iopen_connect_dentry(dentry, inode, 0));
 398 +                       return 0;
 399 +               }
 400 +       }
 401 +       ext3_dec_count(handle, inode);
 402 +       iput(inode);
 403 +       return err;
 404 +}
 405 +
 406  static int ext3_link (struct dentry * old_dentry,
 407                 struct inode * dir, struct dentry *dentry)
 408  {
 409 @@ -2186,7 +2201,8 @@
 410         ext3_inc_count(handle, inode);
 411         atomic_inc(&inode->i_count);
 412
 413 -       err = ext3_add_nondir(handle, dentry, inode);
 414 +       err = ext3_add_link(handle, dentry, inode);
 415 +       ext3_orphan_del(handle, inode);
 416         ext3_journal_stop(handle);
 417         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
 418                 goto retry;
 419 Index: linux-stage/fs/ext3/super.c
 420 ===================================================================
 421 --- linux-stage.orig/fs/ext3/super.c    2005-02-25 14:37:30.987717392 +0200
 422 +++ linux-stage/fs/ext3/super.c 2005-02-25 14:44:50.495901992 +0200
 423 @@ -586,6 +586,7 @@
 424         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 425         Opt_ignore, Opt_barrier,
 426         Opt_err,
 427 +       Opt_iopen, Opt_noiopen, Opt_iopen_nopriv,
 428  };
 429
 430  static match_table_t tokens = {
 431 @@ -633,6 +634,9 @@
 432         {Opt_ignore, "noquota"},
 433         {Opt_ignore, "quota"},
 434         {Opt_ignore, "usrquota"},
 435 +       {Opt_iopen, "iopen"},
 436 +       {Opt_noiopen, "noiopen"},
 437 +       {Opt_iopen_nopriv, "iopen_nopriv"},
 438         {Opt_barrier, "barrier=%u"},
 439         {Opt_err, NULL}
 440  };
 441 @@ -914,6 +918,18 @@
 442                         else
 443                                 clear_opt(sbi->s_mount_opt, BARRIER);
 444                         break;
 445 +               case Opt_iopen:
 446 +                       set_opt (sbi->s_mount_opt, IOPEN);
 447 +                       clear_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 448 +                       break;
 449 +               case Opt_noiopen:
 450 +                       clear_opt (sbi->s_mount_opt, IOPEN);
 451 +                       clear_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 452 +                       break;
 453 +               case Opt_iopen_nopriv:
 454 +                       set_opt (sbi->s_mount_opt, IOPEN);
 455 +                       set_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 456 +                       break;
 457                 case Opt_ignore:
 458                         break;
 459                 default:
 460 Index: linux-stage/include/linux/ext3_fs.h
 461 ===================================================================
 462 --- linux-stage.orig/include/linux/ext3_fs.h    2005-02-25 14:37:28.977023064 +0200
 463 +++ linux-stage/include/linux/ext3_fs.h 2005-02-25 14:49:00.569884968 +0200
 464 @@ -355,6 +355,8 @@
 465  #define EXT3_MOUNT_POSIX_ACL           0x08000 /* POSIX Access Control Lists */
 466  #define EXT3_MOUNT_RESERVATION         0x10000 /* Preallocation */
 467  #define EXT3_MOUNT_BARRIER             0x20000 /* Use block barriers */
 468 +#define EXT3_MOUNT_IOPEN               0x80000 /* Allow access via iopen */
 469 +#define EXT3_MOUNT_IOPEN_NOPRIV                0x100000/* Make iopen world-readable */
 470
 471  /* Compatibility, for having both ext2_fs.h and ext3_fs.h included at once */
 472  #ifndef _LINUX_EXT2_FS_H