lustre/kernel_patches/patches/iopen-2.6-suse.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-stage/fs/ext3/Makefile
  10 ===================================================================
  11 --- linux-stage.orig/fs/ext3/Makefile   2004-11-03 14:41:24.747805262 -0500
  12 +++ linux-stage/fs/ext3/Makefile        2004-11-03 14:41:25.123696274 -0500
  13 @@ -4,7 +4,7 @@
  14
  15  obj-$(CONFIG_EXT3_FS) += ext3.o
  16
  17 -ext3-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o \
  18 +ext3-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o iopen.o \
  19            ioctl.o namei.o super.o symlink.o hash.o
  20
  21  ext3-$(CONFIG_EXT3_FS_XATTR)    += xattr.o xattr_user.o xattr_trusted.o
  22 Index: linux-stage/fs/ext3/inode.c
  23 ===================================================================
  24 --- linux-stage.orig/fs/ext3/inode.c    2004-11-03 14:41:25.040720333 -0500
  25 +++ linux-stage/fs/ext3/inode.c 2004-11-03 14:46:08.458515670 -0500
  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 @@ -2401,6 +2402,9 @@
  35         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
  36  #endif
  37         ei->i_rsv_window.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
  38 +
  39 +    if (ext3_iopen_get_inode(inode))
  40 +        return;
  41
  42         if (ext3_get_inode_loc(inode, &iloc, 0))
  43                 goto bad_inode;
  44 Index: linux-stage/fs/ext3/iopen.c
  45 ===================================================================
  46 --- linux-stage.orig/fs/ext3/iopen.c    1969-12-31 19:00:00.000000000 -0500
  47 +++ linux-stage/fs/ext3/iopen.c 2004-11-03 14:41:25.125695694 -0500
  48 @@ -0,0 +1,272 @@
  49 +/*
  50 + * linux/fs/ext3/iopen.c
  51 + *
  52 + * Special support for open by inode number
  53 + *
  54 + * Copyright (C) 2001 by Theodore Ts'o (tytso@alum.mit.edu).
  55 + *
  56 + * This file may be redistributed under the terms of the GNU General
  57 + * Public License.
  58 + *
  59 + *
  60 + * Invariants:
  61 + *   - there is only ever a single DCACHE_NFSD_DISCONNECTED dentry alias
  62 + *     for an inode at one time.
  63 + *   - there are never both connected and DCACHE_NFSD_DISCONNECTED dentry
  64 + *     aliases on an inode at the same time.
  65 + *
  66 + * If we have any connected dentry aliases for an inode, use one of those
  67 + * in iopen_lookup().  Otherwise, we instantiate a single NFSD_DISCONNECTED
  68 + * dentry for this inode, which thereafter will be found by the dcache
  69 + * when looking up this inode number in __iopen__, so we don't return here
  70 + * until it is gone.
  71 + *
  72 + * If we get an inode via a regular name lookup, then we "rename" the
  73 + * NFSD_DISCONNECTED dentry to the proper name and parent.  This ensures
  74 + * existing users of the disconnected dentry will continue to use the same
  75 + * dentry as the connected users, and there will never be both kinds of
  76 + * dentry aliases at one time.
  77 + */
  78 +
  79 +#include <linux/sched.h>
  80 +#include <linux/fs.h>
  81 +#include <linux/ext3_jbd.h>
  82 +#include <linux/jbd.h>
  83 +#include <linux/ext3_fs.h>
  84 +#include <linux/smp_lock.h>
  85 +#include <linux/dcache.h>
  86 +#include <linux/security.h>
  87 +#include "iopen.h"
  88 +
  89 +#ifndef assert
  90 +#define assert(test) J_ASSERT(test)
  91 +#endif
  92 +
  93 +#define IOPEN_NAME_LEN 32
  94 +
  95 +/*
  96 + * This implements looking up an inode by number.
  97 + */
  98 +static struct dentry *iopen_lookup(struct inode * dir, struct dentry *dentry,
  99 +                                  struct nameidata *nd)
 100 +{
 101 +       struct inode *inode;
 102 +       unsigned long ino;
 103 +       struct list_head *lp;
 104 +       struct dentry *alternate;
 105 +       char buf[IOPEN_NAME_LEN];
 106 +
 107 +       if (dentry->d_name.len >= IOPEN_NAME_LEN)
 108 +               return ERR_PTR(-ENAMETOOLONG);
 109 +
 110 +       memcpy(buf, dentry->d_name.name, dentry->d_name.len);
 111 +       buf[dentry->d_name.len] = 0;
 112 +
 113 +       if (strcmp(buf, ".") == 0)
 114 +               ino = dir->i_ino;
 115 +       else if (strcmp(buf, "..") == 0)
 116 +               ino = EXT3_ROOT_INO;
 117 +       else
 118 +               ino = simple_strtoul(buf, 0, 0);
 119 +
 120 +       if ((ino != EXT3_ROOT_INO &&
 121 +            //ino != EXT3_ACL_IDX_INO &&
 122 +            //ino != EXT3_ACL_DATA_INO &&
 123 +            ino < EXT3_FIRST_INO(dir->i_sb)) ||
 124 +           ino > le32_to_cpu(EXT3_SB(dir->i_sb)->s_es->s_inodes_count))
 125 +               return ERR_PTR(-ENOENT);
 126 +
 127 +       inode = iget(dir->i_sb, ino);
 128 +       if (!inode)
 129 +               return ERR_PTR(-EACCES);
 130 +       if (is_bad_inode(inode)) {
 131 +               iput(inode);
 132 +               return ERR_PTR(-ENOENT);
 133 +       }
 134 +
 135 +       assert(list_empty(&dentry->d_alias));           /* d_instantiate */
 136 +       assert(d_unhashed(dentry));             /* d_rehash */
 137 +
 138 +       /* preferrably return a connected dentry */
 139 +       spin_lock(&dcache_lock);
 140 +       list_for_each(lp, &inode->i_dentry) {
 141 +               alternate = list_entry(lp, struct dentry, d_alias);
 142 +               assert(!(alternate->d_flags & DCACHE_DISCONNECTED));
 143 +       }
 144 +
 145 +       if (!list_empty(&inode->i_dentry)) {
 146 +               alternate = list_entry(inode->i_dentry.next,
 147 +                                      struct dentry, d_alias);
 148 +               dget_locked(alternate);
 149 +               alternate->d_vfs_flags |= DCACHE_REFERENCED;
 150 +               iput(inode);
 151 +               spin_unlock(&dcache_lock);
 152 +               return alternate;
 153 +       }
 154 +       dentry->d_flags |= DCACHE_DISCONNECTED;
 155 +
 156 +       /* d_add(), but don't drop dcache_lock before adding dentry to inode */
 157 +       list_add(&dentry->d_alias, &inode->i_dentry);   /* d_instantiate */
 158 +       dentry->d_inode = inode;
 159 +
 160 +       __d_rehash(dentry, 0);                          /* d_rehash */
 161 +       spin_unlock(&dcache_lock);
 162 +
 163 +       return NULL;
 164 +}
 165 +
 166 +#define do_switch(x,y) do { \
 167 +       __typeof__ (x) __tmp = x; \
 168 +       x = y; y = __tmp; } while (0)
 169 +
 170 +static inline void switch_names(struct dentry *dentry, struct dentry *target)
 171 +{
 172 +       const unsigned char *old_name, *new_name;
 173 +
 174 +       memcpy(dentry->d_iname, target->d_iname, DNAME_INLINE_LEN);
 175 +       old_name = target->d_name.name;
 176 +       new_name = dentry->d_name.name;
 177 +       if (old_name == target->d_iname)
 178 +               old_name = dentry->d_iname;
 179 +       if (new_name == dentry->d_iname)
 180 +               new_name = target->d_iname;
 181 +       target->d_name.name = new_name;
 182 +       dentry->d_name.name = old_name;
 183 +}
 184 +
 185 +/* This function is spliced into ext3_lookup and does the move of a
 186 + * disconnected dentry (if it exists) to a connected dentry.
 187 + */
 188 +struct dentry *iopen_connect_dentry(struct dentry *dentry, struct inode *inode,
 189 +                                   int rehash)
 190 +{
 191 +       struct dentry *tmp, *goal = NULL;
 192 +       struct list_head *lp;
 193 +
 194 +       /* verify this dentry is really new */
 195 +       assert(dentry->d_inode == NULL);
 196 +       assert(list_empty(&dentry->d_alias));           /* d_instantiate */
 197 +       if (rehash)
 198 +               assert(d_unhashed(dentry));     /* d_rehash */
 199 +       assert(list_empty(&dentry->d_subdirs));
 200 +
 201 +       spin_lock(&dcache_lock);
 202 +       if (!inode)
 203 +               goto do_rehash;
 204 +
 205 +       /* preferrably return a connected dentry */
 206 +       list_for_each(lp, &inode->i_dentry) {
 207 +               tmp = list_entry(lp, struct dentry, d_alias);
 208 +               if (tmp->d_flags & DCACHE_DISCONNECTED) {
 209 +                       assert(tmp->d_alias.next == &inode->i_dentry);
 210 +                       assert(tmp->d_alias.prev == &inode->i_dentry);
 211 +                       goal = tmp;
 212 +                       dget_locked(goal);
 213 +                       break;
 214 +               }
 215 +       }
 216 +
 217 +       if (!goal)
 218 +               goto do_instantiate;
 219 +
 220 +       /* Move the goal to the de hash queue */
 221 +       goal->d_flags &= ~ DCACHE_DISCONNECTED;
 222 +       security_d_instantiate(goal, inode);
 223 +       __d_rehash(dentry, 0);
 224 +       __d_move(goal, dentry);
 225 +       spin_unlock(&dcache_lock);
 226 +       iput(inode);
 227 +
 228 +       return goal;
 229 +
 230 +       /* d_add(), but don't drop dcache_lock before adding dentry to inode */
 231 +do_instantiate:
 232 +       list_add(&dentry->d_alias, &inode->i_dentry);   /* d_instantiate */
 233 +       dentry->d_inode = inode;
 234 +do_rehash:
 235 +       if (rehash)
 236 +               __d_rehash(dentry, 0);                  /* d_rehash */
 237 +       spin_unlock(&dcache_lock);
 238 +
 239 +       return NULL;
 240 +}
 241 +
 242 +/*
 243 + * These are the special structures for the iopen pseudo directory.
 244 + */
 245 +
 246 +static struct inode_operations iopen_inode_operations = {
 247 +       lookup:         iopen_lookup,           /* BKL held */
 248 +};
 249 +
 250 +static struct file_operations iopen_file_operations = {
 251 +       read:           generic_read_dir,
 252 +};
 253 +
 254 +static int match_dentry(struct dentry *dentry, const char *name)
 255 +{
 256 +       int     len;
 257 +
 258 +       len = strlen(name);
 259 +       if (dentry->d_name.len != len)
 260 +               return 0;
 261 +       if (strncmp(dentry->d_name.name, name, len))
 262 +               return 0;
 263 +       return 1;
 264 +}
 265 +
 266 +/*
 267 + * This function is spliced into ext3_lookup and returns 1 the file
 268 + * name is __iopen__ and dentry has been filled in appropriately.
 269 + */
 270 +int ext3_check_for_iopen(struct inode *dir, struct dentry *dentry)
 271 +{
 272 +       struct inode *inode;
 273 +
 274 +       if (dir->i_ino != EXT3_ROOT_INO ||
 275 +           !test_opt(dir->i_sb, IOPEN) ||
 276 +           !match_dentry(dentry, "__iopen__"))
 277 +               return 0;
 278 +
 279 +       inode = iget(dir->i_sb, EXT3_BAD_INO);
 280 +
 281 +       if (!inode)
 282 +               return 0;
 283 +       d_add(dentry, inode);
 284 +       return 1;
 285 +}
 286 +
 287 +/*
 288 + * This function is spliced into read_inode; it returns 1 if inode
 289 + * number is the one for /__iopen__, in which case the inode is filled
 290 + * in appropriately.  Otherwise, this fuction returns 0.
 291 + */
 292 +int ext3_iopen_get_inode(struct inode *inode)
 293 +{
 294 +       if (inode->i_ino != EXT3_BAD_INO)
 295 +               return 0;
 296 +
 297 +       inode->i_mode = S_IFDIR | S_IRUSR | S_IXUSR;
 298 +       if (test_opt(inode->i_sb, IOPEN_NOPRIV))
 299 +               inode->i_mode |= 0777;
 300 +       inode->i_uid = 0;
 301 +       inode->i_gid = 0;
 302 +       inode->i_nlink = 1;
 303 +       inode->i_size = 4096;
 304 +       inode->i_atime = CURRENT_TIME;
 305 +       inode->i_ctime = CURRENT_TIME;
 306 +       inode->i_mtime = CURRENT_TIME;
 307 +       EXT3_I(inode)->i_dtime = 0;
 308 +       inode->i_blksize = PAGE_SIZE;   /* This is the optimal IO size
 309 +                                        * (for stat), not the fs block
 310 +                                        * size */
 311 +       inode->i_blocks = 0;
 312 +       inode->i_version = 1;
 313 +       inode->i_generation = 0;
 314 +
 315 +       inode->i_op = &iopen_inode_operations;
 316 +       inode->i_fop = &iopen_file_operations;
 317 +       inode->i_mapping->a_ops = 0;
 318 +
 319 +       return 1;
 320 +}
 321 Index: linux-stage/fs/ext3/iopen.h
 322 ===================================================================
 323 --- linux-stage.orig/fs/ext3/iopen.h    1969-12-31 19:00:00.000000000 -0500
 324 +++ linux-stage/fs/ext3/iopen.h 2004-11-03 14:41:25.126695404 -0500
 325 @@ -0,0 +1,15 @@
 326 +/*
 327 + * iopen.h
 328 + *
 329 + * Special support for opening files by inode number.
 330 + *
 331 + * Copyright (C) 2001 by Theodore Ts'o (tytso@alum.mit.edu).
 332 + *
 333 + * This file may be redistributed under the terms of the GNU General
 334 + * Public License.
 335 + */
 336 +
 337 +extern int ext3_check_for_iopen(struct inode *dir, struct dentry *dentry);
 338 +extern int ext3_iopen_get_inode(struct inode *inode);
 339 +extern struct dentry *iopen_connect_dentry(struct dentry *dentry,
 340 +                                          struct inode *inode, int rehash);
 341 Index: linux-stage/fs/ext3/namei.c
 342 ===================================================================
 343 --- linux-stage.orig/fs/ext3/namei.c    2004-11-03 14:41:24.957744391 -0500
 344 +++ linux-stage/fs/ext3/namei.c 2004-11-03 14:41:25.127695114 -0500
 345 @@ -37,6 +37,7 @@
 346  #include <linux/buffer_head.h>
 347  #include <linux/smp_lock.h>
 348  #include "xattr.h"
 349 +#include "iopen.h"
 350  #include "acl.h"
 351
 352  /*
 353 @@ -979,6 +980,9 @@
 354         if (dentry->d_name.len > EXT3_NAME_LEN)
 355                 return ERR_PTR(-ENAMETOOLONG);
 356
 357 +       if (ext3_check_for_iopen(dir, dentry))
 358 +               return NULL;
 359 +
 360         bh = ext3_find_entry(dentry, &de);
 361         inode = NULL;
 362         if (bh) {
 363 @@ -989,10 +993,8 @@
 364                 if (!inode)
 365                         return ERR_PTR(-EACCES);
 366         }
 367 -       if (inode)
 368 -               return d_splice_alias(inode, dentry);
 369 -       d_add(dentry, inode);
 370 -       return NULL;
 371 +
 372 +       return iopen_connect_dentry(dentry, inode, 1);
 373  }
 374
 375
 376 @@ -2029,10 +2031,6 @@
 377                               inode->i_nlink);
 378         inode->i_version++;
 379         inode->i_nlink = 0;
 380 -       /* There's no need to set i_disksize: the fact that i_nlink is
 381 -        * zero will ensure that the right thing happens during any
 382 -        * recovery. */
 383 -       inode->i_size = 0;
 384         ext3_orphan_add(handle, inode);
 385         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 386         ext3_mark_inode_dirty(handle, inode);
 387 @@ -2152,6 +2150,23 @@
 388         return err;
 389  }
 390
 391 +/* Like ext3_add_nondir() except for call to iopen_connect_dentry */
 392 +static int ext3_add_link(handle_t *handle, struct dentry *dentry,
 393 +                        struct inode *inode)
 394 +{
 395 +       int err = ext3_add_entry(handle, dentry, inode);
 396 +       if (!err) {
 397 +               err = ext3_mark_inode_dirty(handle, inode);
 398 +               if (err == 0) {
 399 +                       (void)iopen_connect_dentry(dentry, inode, 0);
 400 +                       return 0;
 401 +               }
 402 +       }
 403 +       ext3_dec_count(handle, inode);
 404 +       iput(inode);
 405 +       return err;
 406 +}
 407 +
 408  static int ext3_link (struct dentry * old_dentry,
 409                 struct inode * dir, struct dentry *dentry)
 410  {
 411 @@ -2175,7 +2190,8 @@
 412         ext3_inc_count(handle, inode);
 413         atomic_inc(&inode->i_count);
 414
 415 -       err = ext3_add_nondir(handle, dentry, inode);
 416 +       err = ext3_add_link(handle, dentry, inode);
 417 +       ext3_orphan_del(handle,inode);
 418         ext3_journal_stop(handle);
 419         if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
 420                 goto retry;
 421 Index: linux-stage/fs/ext3/super.c
 422 ===================================================================
 423 --- linux-stage.orig/fs/ext3/super.c    2004-11-03 14:41:25.043719463 -0500
 424 +++ linux-stage/fs/ext3/super.c 2004-11-03 14:41:25.129694535 -0500
 425 @@ -534,7 +534,7 @@
 426         Opt_reservation, Opt_noreservation, Opt_noload,
 427         Opt_commit, Opt_journal_update, Opt_journal_inum,
 428         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 429 -       Opt_ignore, Opt_barrier,
 430 +       Opt_ignore, Opt_barrier, Opt_iopen, Opt_noiopen, Opt_iopen_nopriv,
 431         Opt_err,
 432  };
 433
 434 @@ -577,6 +577,9 @@
 435         {Opt_ignore, "quota"},
 436         {Opt_ignore, "usrquota"},
 437         {Opt_barrier, "barrier=%u"},
 438 +       {Opt_iopen, "iopen"},
 439 +       {Opt_noiopen, "noiopen"},
 440 +       {Opt_iopen_nopriv, "iopen_nopriv"},
 441         {Opt_err, NULL}
 442  };
 443
 444 @@ -778,6 +781,18 @@
 445                         else
 446                                 clear_opt(sbi->s_mount_opt, BARRIER);
 447                         break;
 448 +               case Opt_iopen:
 449 +                       set_opt (sbi->s_mount_opt, IOPEN);
 450 +                       clear_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 451 +                       break;
 452 +               case Opt_noiopen:
 453 +                       clear_opt (sbi->s_mount_opt, IOPEN);
 454 +                       clear_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 455 +                       break;
 456 +               case Opt_iopen_nopriv:
 457 +                       set_opt (sbi->s_mount_opt, IOPEN);
 458 +                       set_opt (sbi->s_mount_opt, IOPEN_NOPRIV);
 459 +                       break;
 460                 case Opt_ignore:
 461                         break;
 462                 default:
 463 Index: linux-stage/include/linux/ext3_fs.h
 464 ===================================================================
 465 --- linux-stage.orig/include/linux/ext3_fs.h    2004-05-11 17:21:20.000000000 -0400
 466 +++ linux-stage/include/linux/ext3_fs.h 2004-05-11 17:21:21.000000000 -0400
 467 @@ -326,6 +326,8 @@
 468  #define EXT3_MOUNT_XATTR_USER          0x4000  /* Extended user attributes */
 469  #define EXT3_MOUNT_POSIX_ACL           0x8000  /* POSIX Access Control Lists */
 470  #define EXT3_MOUNT_BARRIER             0x10000 /* Use block barriers */
 471 +#define EXT3_MOUNT_IOPEN                0x20000 /* Allow access via iopen */
 472 +#define EXT3_MOUNT_IOPEN_NOPRIV         0x40000 /* Make iopen world-readable */
 473
 474  /* Compatibility, for having both ext2_fs.h and ext3_fs.h included at once */
 475  #ifndef _LINUX_EXT2_FS_H