lustre/kernel_patches/patches/iopen-2.6.10-fc3.patch

   1  fs/ext3/inode.c                    |    3
   2  fs/ext3/iopen.c                    |  239 +++++++++++++++++++++++++++++++++++++
   3  fs/ext3/iopen.h                    |   15 ++
   4  fs/ext3/namei.c                    |   13 ++
   5  fs/ext3/super.c                    |   17 ++
   6  include/linux/ext3_fs.h            |    2
   7  7 files changed, 304 insertions(+), 1 deletion(-)
   8
   9 Index: linux-2.6.10/include/linux/ext3_fs.h
  10 ===================================================================
  11 --- linux-2.6.10.orig/include/linux/ext3_fs.h   2005-04-05 12:25:13.635136112 +0800
  12 +++ linux-2.6.10/include/linux/ext3_fs.h        2005-04-05 12:25:13.801110880 +0800
  13 @@ -357,6 +357,8 @@
  14  #define EXT3_MOUNT_RESERVATION         0x10000 /* Preallocation */
  15  #define EXT3_MOUNT_BARRIER             0x20000 /* Use block barriers */
  16  #define EXT3_MOUNT_PDIROPS             0x800000/* Parallel dir operations */
  17 +#define EXT3_MOUNT_IOPEN               0x40000 /* Allow access via iopen */
  18 +#define EXT3_MOUNT_IOPEN_NOPRIV                0x80000 /* Make iopen world-readable */
  19
  20  /* Compatibility, for having both ext2_fs.h and ext3_fs.h included at once */
  21  #ifndef _LINUX_EXT2_FS_H
  22 Index: linux-2.6.10/fs/ext3/inode.c
  23 ===================================================================
  24 --- linux-2.6.10.orig/fs/ext3/inode.c   2005-04-05 12:25:13.726122280 +0800
  25 +++ linux-2.6.10/fs/ext3/inode.c        2005-04-05 12:25:13.794111944 +0800
  26 @@ -37,6 +37,7 @@
  27  #include <linux/mpage.h>
  28  #include <linux/uio.h>
  29  #include "xattr.h"
  30 +#include "iopen.h"
  31  #include "acl.h"
  32
  33  /*
  34 @@ -2411,6 +2412,9 @@
  35  #endif
  36         ei->i_rsv_window.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
  37
  38 +       if (ext3_iopen_get_inode(inode))
  39 +               return;
  40 +
  41         if (ext3_get_inode_loc(inode, &iloc, 0))
  42                 goto bad_inode;
  43         bh = iloc.bh;
  44 Index: linux-2.6.10/fs/ext3/super.c
  45 ===================================================================
  46 --- linux-2.6.10.orig/fs/ext3/super.c   2005-04-05 12:25:13.728121976 +0800
  47 +++ linux-2.6.10/fs/ext3/super.c        2005-04-05 12:25:13.797111488 +0800
  48 @@ -592,6 +592,7 @@
  49         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
  50         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
  51         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_pdirops,
  52 +       Opt_iopen, Opt_noiopen, Opt_iopen_nopriv,
  53         Opt_ignore, Opt_barrier, Opt_err, Opt_resize,
  54  };
  55
  56 @@ -641,6 +642,9 @@
  57         {Opt_ignore, "usrquota"},
  58         {Opt_barrier, "barrier=%u"},
  59         {Opt_pdirops, "pdirops"},
  60 +       {Opt_iopen,  "iopen"},
  61 +       {Opt_noiopen,  "noiopen"},
  62 +       {Opt_iopen_nopriv,  "iopen_nopriv"},
  63         {Opt_err, NULL},
  64         {Opt_resize, "resize"},
  65  };
  66 @@ -921,6 +925,18 @@
  67                         else
  68                                 clear_opt(sbi->s_mount_opt, BARRIER);
  69                         break;
  70 +               case Opt_iopen:
  71 +                       set_opt (sbi->s_mount_opt, IOPEN);
  72 +                       clear_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
  73 +                       break;
  74 +               case Opt_noiopen:
  75 +                       clear_opt (sbi->s_mount_opt, IOPEN);
  76 +                       clear_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
  77 +                       break;
  78 +               case Opt_iopen_nopriv:
  79 +                       set_opt (sbi->s_mount_opt, IOPEN);
  80 +                       set_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
  81 +                       break;
  82                 case Opt_ignore:
  83                         break;
  84                 case Opt_resize:
  85 Index: linux-2.6.10/fs/ext3/iopen.c
  86 ===================================================================
  87 --- linux-2.6.10.orig/fs/ext3/iopen.c   2005-04-05 19:01:49.158500672 +0800
  88 +++ linux-2.6.10/fs/ext3/iopen.c        2005-04-05 12:25:13.791112400 +0800
  89 @@ -0,0 +1,274 @@
  90 +/*
  91 + * linux/fs/ext3/iopen.c
  92 + *
  93 + * Special support for open by inode number
  94 + *
  95 + * Copyright (C) 2001 by Theodore Ts'o (tytso@alum.mit.edu).
  96 + *
  97 + * This file may be redistributed under the terms of the GNU General
  98 + * Public License.
  99 + *
 100 + *
 101 + * Invariants:
 102 + *   - there is only ever a single DCACHE_NFSD_DISCONNECTED dentry alias
 103 + *     for an inode at one time.
 104 + *   - there are never both connected and DCACHE_NFSD_DISCONNECTED dentry
 105 + *     aliases on an inode at the same time.
 106 + *
 107 + * If we have any connected dentry aliases for an inode, use one of those
 108 + * in iopen_lookup().  Otherwise, we instantiate a single NFSD_DISCONNECTED
 109 + * dentry for this inode, which thereafter will be found by the dcache
 110 + * when looking up this inode number in __iopen__, so we don't return here
 111 + * until it is gone.
 112 + *
 113 + * If we get an inode via a regular name lookup, then we "rename" the
 114 + * NFSD_DISCONNECTED dentry to the proper name and parent.  This ensures
 115 + * existing users of the disconnected dentry will continue to use the same
 116 + * dentry as the connected users, and there will never be both kinds of
 117 + * dentry aliases at one time.
 118 + */
 119 +
 120 +#include <linux/sched.h>
 121 +#include <linux/fs.h>
 122 +#include <linux/ext3_jbd.h>
 123 +#include <linux/jbd.h>
 124 +#include <linux/ext3_fs.h>
 125 +#include <linux/smp_lock.h>
 126 +#include <linux/dcache.h>
 127 +#include <linux/security.h>
 128 +#include "iopen.h"
 129 +
 130 +#ifndef assert
 131 +#define assert(test) J_ASSERT(test)
 132 +#endif
 133 +
 134 +#define IOPEN_NAME_LEN 32
 135 +
 136 +/*
 137 + * This implements looking up an inode by number.
 138 + */
 139 +static struct dentry *iopen_lookup(struct inode * dir, struct dentry *dentry,
 140 +                                  struct nameidata *nd)
 141 +{
 142 +       struct inode *inode;
 143 +       unsigned long ino;
 144 +       struct list_head *lp;
 145 +       struct dentry *alternate;
 146 +       char buf[IOPEN_NAME_LEN];
 147 +
 148 +       if (dentry->d_name.len >= IOPEN_NAME_LEN)
 149 +               return ERR_PTR(-ENAMETOOLONG);
 150 +
 151 +       memcpy(buf, dentry->d_name.name, dentry->d_name.len);
 152 +       buf[dentry->d_name.len] = 0;
 153 +
 154 +       if (strcmp(buf, ".") == 0)
 155 +               ino = dir->i_ino;
 156 +       else if (strcmp(buf, "..") == 0)
 157 +               ino = EXT3_ROOT_INO;
 158 +       else
 159 +               ino = simple_strtoul(buf, 0, 0);
 160 +
 161 +       if ((ino != EXT3_ROOT_INO &&
 162 +            //ino != EXT3_ACL_IDX_INO &&
 163 +            //ino != EXT3_ACL_DATA_INO &&
 164 +            ino < EXT3_FIRST_INO(dir->i_sb)) ||
 165 +           ino > le32_to_cpu(EXT3_SB(dir->i_sb)->s_es->s_inodes_count))
 166 +               return ERR_PTR(-ENOENT);
 167 +
 168 +       inode = iget(dir->i_sb, ino);
 169 +       if (!inode)
 170 +               return ERR_PTR(-EACCES);
 171 +       if (is_bad_inode(inode)) {
 172 +               iput(inode);
 173 +               return ERR_PTR(-ENOENT);
 174 +       }
 175 +
 176 +       assert(list_empty(&dentry->d_alias));           /* d_instantiate */
 177 +       assert(d_unhashed(dentry));             /* d_rehash */
 178 +
 179 +       /* preferrably return a connected dentry */
 180 +       spin_lock(&dcache_lock);
 181 +       list_for_each(lp, &inode->i_dentry) {
 182 +               alternate = list_entry(lp, struct dentry, d_alias);
 183 +               assert(!(alternate->d_flags & DCACHE_DISCONNECTED));
 184 +       }
 185 +
 186 +       if (!list_empty(&inode->i_dentry)) {
 187 +               alternate = list_entry(inode->i_dentry.next,
 188 +                                      struct dentry, d_alias);
 189 +               dget_locked(alternate);
 190 +               spin_lock(&alternate->d_lock);
 191 +               alternate->d_flags |= DCACHE_REFERENCED;
 192 +               spin_unlock(&alternate->d_lock);
 193 +               iput(inode);
 194 +               spin_unlock(&dcache_lock);
 195 +               return alternate;
 196 +       }
 197 +       dentry->d_flags |= DCACHE_DISCONNECTED;
 198 +
 199 +       /* d_add(), but don't drop dcache_lock before adding dentry to inode */
 200 +       list_add(&dentry->d_alias, &inode->i_dentry);   /* d_instantiate */
 201 +       dentry->d_inode = inode;
 202 +
 203 +       __d_rehash(dentry);                             /* d_rehash */
 204 +       spin_unlock(&dcache_lock);
 205 +
 206 +       return NULL;
 207 +}
 208 +
 209 +#define do_switch(x,y) do { \
 210 +       __typeof__ (x) __tmp = x; \
 211 +       x = y; y = __tmp; } while (0)
 212 +
 213 +static inline void switch_names(struct dentry *dentry, struct dentry *target)
 214 +{
 215 +       const unsigned char *old_name, *new_name;
 216 +
 217 +       memcpy(dentry->d_iname, target->d_iname, DNAME_INLINE_LEN_MIN);
 218 +       old_name = target->d_name.name;
 219 +       new_name = dentry->d_name.name;
 220 +       if (old_name == target->d_iname)
 221 +               old_name = dentry->d_iname;
 222 +       if (new_name == dentry->d_iname)
 223 +               new_name = target->d_iname;
 224 +       target->d_name.name = new_name;
 225 +       dentry->d_name.name = old_name;
 226 +}
 227 +
 228 +/* This function is spliced into ext3_lookup and does the move of a
 229 + * disconnected dentry (if it exists) to a connected dentry.
 230 + */
 231 +struct dentry *iopen_connect_dentry(struct dentry *dentry, struct inode *inode,
 232 +                                   int rehash)
 233 +{
 234 +       struct dentry *tmp, *goal = NULL;
 235 +       struct list_head *lp;
 236 +
 237 +       /* verify this dentry is really new */
 238 +       assert(dentry->d_inode == NULL);
 239 +       assert(list_empty(&dentry->d_alias));           /* d_instantiate */
 240 +       if (rehash)
 241 +               assert(d_unhashed(dentry));     /* d_rehash */
 242 +       assert(list_empty(&dentry->d_subdirs));
 243 +
 244 +       spin_lock(&dcache_lock);
 245 +       if (!inode)
 246 +               goto do_rehash;
 247 +
 248 +       /* preferrably return a connected dentry */
 249 +       list_for_each(lp, &inode->i_dentry) {
 250 +               tmp = list_entry(lp, struct dentry, d_alias);
 251 +               if (tmp->d_flags & DCACHE_DISCONNECTED) {
 252 +                       assert(tmp->d_alias.next == &inode->i_dentry);
 253 +                       assert(tmp->d_alias.prev == &inode->i_dentry);
 254 +                       goal = tmp;
 255 +                       dget_locked(goal);
 256 +                       break;
 257 +               }
 258 +       }
 259 +
 260 +       if (!goal)
 261 +               goto do_instantiate;
 262 +
 263 +       /* Move the goal to the de hash queue */
 264 +       goal->d_flags &= ~ DCACHE_DISCONNECTED;
 265 +       security_d_instantiate(goal, inode);
 266 +       __d_rehash(dentry);
 267 +       __d_move(goal, dentry);
 268 +       spin_unlock(&dcache_lock);
 269 +       iput(inode);
 270 +
 271 +       return goal;
 272 +
 273 +       /* d_add(), but don't drop dcache_lock before adding dentry to inode */
 274 +do_instantiate:
 275 +       list_add(&dentry->d_alias, &inode->i_dentry);   /* d_instantiate */
 276 +       dentry->d_inode = inode;
 277 +do_rehash:
 278 +       if (rehash)
 279 +               __d_rehash(dentry);                     /* d_rehash */
 280 +       spin_unlock(&dcache_lock);
 281 +
 282 +       return NULL;
 283 +}
 284 +
 285 +/*
 286 + * These are the special structures for the iopen pseudo directory.
 287 + */
 288 +
 289 +static struct inode_operations iopen_inode_operations = {
 290 +       lookup:         iopen_lookup,           /* BKL held */
 291 +};
 292 +
 293 +static struct file_operations iopen_file_operations = {
 294 +       read:           generic_read_dir,
 295 +};
 296 +
 297 +static int match_dentry(struct dentry *dentry, const char *name)
 298 +{
 299 +       int     len;
 300 +
 301 +       len = strlen(name);
 302 +       if (dentry->d_name.len != len)
 303 +               return 0;
 304 +       if (strncmp(dentry->d_name.name, name, len))
 305 +               return 0;
 306 +       return 1;
 307 +}
 308 +
 309 +/*
 310 + * This function is spliced into ext3_lookup and returns 1 the file
 311 + * name is __iopen__ and dentry has been filled in appropriately.
 312 + */
 313 +int ext3_check_for_iopen(struct inode *dir, struct dentry *dentry)
 314 +{
 315 +       struct inode *inode;
 316 +
 317 +       if (dir->i_ino != EXT3_ROOT_INO ||
 318 +           !test_opt(dir->i_sb, IOPEN) ||
 319 +           !match_dentry(dentry, "__iopen__"))
 320 +               return 0;
 321 +
 322 +       inode = iget(dir->i_sb, EXT3_BAD_INO);
 323 +
 324 +       if (!inode)
 325 +               return 0;
 326 +       d_add(dentry, inode);
 327 +       return 1;
 328 +}
 329 +
 330 +/*
 331 + * This function is spliced into read_inode; it returns 1 if inode
 332 + * number is the one for /__iopen__, in which case the inode is filled
 333 + * in appropriately.  Otherwise, this fuction returns 0.
 334 + */
 335 +int ext3_iopen_get_inode(struct inode *inode)
 336 +{
 337 +       if (inode->i_ino != EXT3_BAD_INO)
 338 +               return 0;
 339 +
 340 +       inode->i_mode = S_IFDIR | S_IRUSR | S_IXUSR;
 341 +       if (test_opt(inode->i_sb, IOPEN_NOPRIV))
 342 +               inode->i_mode |= 0777;
 343 +       inode->i_uid = 0;
 344 +       inode->i_gid = 0;
 345 +       inode->i_nlink = 1;
 346 +       inode->i_size = 4096;
 347 +       inode->i_atime = CURRENT_TIME;
 348 +       inode->i_ctime = CURRENT_TIME;
 349 +       inode->i_mtime = CURRENT_TIME;
 350 +       EXT3_I(inode)->i_dtime = 0;
 351 +       inode->i_blksize = PAGE_SIZE;   /* This is the optimal IO size
 352 +                                        * (for stat), not the fs block
 353 +                                        * size */
 354 +       inode->i_blocks = 0;
 355 +       inode->i_version = 1;
 356 +       inode->i_generation = 0;
 357 +
 358 +       inode->i_op = &iopen_inode_operations;
 359 +       inode->i_fop = &iopen_file_operations;
 360 +       inode->i_mapping->a_ops = 0;
 361 +
 362 +       return 1;
 363 +}
 364 Index: linux-2.6.10/fs/ext3/iopen.h
 365 ===================================================================
 366 --- linux-2.6.10.orig/fs/ext3/iopen.h   2005-04-05 19:01:49.158500672 +0800
 367 +++ linux-2.6.10/fs/ext3/iopen.h        2005-04-05 12:25:13.792112248 +0800
 368 @@ -0,0 +1,15 @@
 369 +/*
 370 + * iopen.h
 371 + *
 372 + * Special support for opening files by inode number.
 373 + *
 374 + * Copyright (C) 2001 by Theodore Ts'o (tytso@alum.mit.edu).
 375 + *
 376 + * This file may be redistributed under the terms of the GNU General
 377 + * Public License.
 378 + */
 379 +
 380 +extern int ext3_check_for_iopen(struct inode *dir, struct dentry *dentry);
 381 +extern int ext3_iopen_get_inode(struct inode *inode);
 382 +extern struct dentry *iopen_connect_dentry(struct dentry *dentry,
 383 +                                          struct inode *inode, int rehash);
 384 Index: linux-2.6.10/fs/ext3/Makefile
 385 ===================================================================
 386 --- linux-2.6.10.orig/fs/ext3/Makefile  2004-12-25 05:33:52.000000000 +0800
 387 +++ linux-2.6.10/fs/ext3/Makefile       2005-04-05 12:26:06.897039072 +0800
 388 @@ -5,7 +5,7 @@
 389  obj-$(CONFIG_EXT3_FS) += ext3.o
 390
 391  ext3-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o \
 392 -          ioctl.o namei.o super.o symlink.o hash.o resize.o
 393 +          ioctl.o namei.o super.o symlink.o hash.o resize.o iopen.o
 394
 395  ext3-$(CONFIG_EXT3_FS_XATTR)    += xattr.o xattr_user.o xattr_trusted.o
 396  ext3-$(CONFIG_EXT3_FS_POSIX_ACL) += acl.o
 397 Index: linux-2.6.10/fs/ext3/namei.c
 398 ===================================================================
 399 --- linux-2.6.10.orig/fs/ext3/namei.c   2005-04-05 12:25:13.633136416 +0800
 400 +++ linux-2.6.10/fs/ext3/namei.c        2005-04-05 12:25:13.799111184 +0800
 401 @@ -37,6 +37,7 @@
 402  #include <linux/buffer_head.h>
 403  #include <linux/smp_lock.h>
 404  #include "xattr.h"
 405 +#include "iopen.h"
 406  #include "acl.h"
 407
 408  /*
 409 @@ -1140,6 +1141,9 @@
 410         if (dentry->d_name.len > EXT3_NAME_LEN)
 411                 return ERR_PTR(-ENAMETOOLONG);
 412
 413 +       if (ext3_check_for_iopen(dir, dentry))
 414 +               return NULL;
 415 +
 416         bh = ext3_find_entry(dentry, &de, 0, &lock);
 417         inode = NULL;
 418         if (bh) {
 419 @@ -1151,10 +1155,8 @@
 420                 if (!inode)
 421                         return ERR_PTR(-EACCES);
 422         }
 423 -       if (inode)
 424 -               return d_splice_alias(inode, dentry);
 425 -       d_add(dentry, inode);
 426 -       return NULL;
 427 +
 428 +       return iopen_connect_dentry(dentry, inode, 1);
 429  }
 430
 431
 432 @@ -2367,10 +2369,6 @@
 433                               inode->i_nlink);
 434         inode->i_version++;
 435         inode->i_nlink = 0;
 436 -       /* There's no need to set i_disksize: the fact that i_nlink is
 437 -        * zero will ensure that the right thing happens during any
 438 -        * recovery. */
 439 -       inode->i_size = 0;
 440         ext3_orphan_add(handle, inode);
 441         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 442         ext3_mark_inode_dirty(handle, inode);
 443 @@ -2497,6 +2495,23 @@
 444         return err;
 445  }
 446
 447 +/* Like ext3_add_nondir() except for call to iopen_connect_dentry */
 448 +static int ext3_add_link(handle_t *handle, struct dentry *dentry,
 449 +                        struct inode *inode)
 450 +{
 451 +       int err = ext3_add_entry(handle, dentry, inode);
 452 +       if (!err) {
 453 +               err = ext3_mark_inode_dirty(handle, inode);
 454 +               if (err == 0) {
 455 +                       dput(iopen_connect_dentry(dentry, inode, 0));
 456 +                       return 0;
 457 +               }
 458 +       }
 459 +       ext3_dec_count(handle, inode);
 460 +       iput(inode);
 461 +       return err;
 462 +}
 463 +
 464  static int ext3_link (struct dentry * old_dentry,
 465                 struct inode * dir, struct dentry *dentry)
 466  {
 467 @@ -2520,7 +2535,8 @@
 468         ext3_inc_count(handle, inode);
 469         atomic_inc(&inode->i_count);
 470
 471 -       err = ext3_add_nondir(handle, dentry, inode);
 472 +       err = ext3_add_link(handle, dentry, inode);
 473 +       ext3_orphan_del(handle,inode);
 474         ext3_journal_stop(handle);
 475         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
 476                 goto retry;