lustre/obdfilter/filter.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  linux/fs/filter/filter.c
   5  *
   6  * Copyright (C) 2001  Cluster File Systems, Inc.
   7  *
   8  * This code is issued under the GNU General Public License.
   9  * See the file COPYING in this distribution
  10  *
  11  * by Peter Braam <braam@clusterfs.com>
  12  */
  13
  14 #define EXPORT_SYMTAB
  15 #define DEBUG_SUBSYSTEM S_FILTER
  16
  17 #include <linux/module.h>
  18 #include <linux/obd_filter.h>
  19 #include <linux/ext3_jbd.h>
  20 #include <linux/quotaops.h>
  21
  22 extern struct obd_device obd_dev[MAX_OBD_DEVICES];
  23 long filter_memory;
  24
  25 #define FILTER_ROOTINO 2
  26
  27 #define S_SHIFT 12
  28 static char *obd_type_by_mode[S_IFMT >> S_SHIFT] = {
  29         [0]                     NULL,
  30         [S_IFREG >> S_SHIFT]    "R",
  31         [S_IFDIR >> S_SHIFT]    "D",
  32         [S_IFCHR >> S_SHIFT]    "C",
  33         [S_IFBLK >> S_SHIFT]    "B",
  34         [S_IFIFO >> S_SHIFT]    "F",
  35         [S_IFSOCK >> S_SHIFT]   "S",
  36         [S_IFLNK >> S_SHIFT]    "L"
  37 };
  38
  39 static inline const char *obd_mode_to_type(int mode)
  40 {
  41         return obd_type_by_mode[(mode & S_IFMT) >> S_SHIFT];
  42 }
  43
  44 /* write the pathname into the string */
  45 static int filter_id(char *buf, obd_id id, obd_mode mode)
  46 {
  47         return sprintf(buf, "O/%s/%Ld", obd_mode_to_type(mode),
  48                        (unsigned long long)id);
  49 }
  50
  51 /* setup the object store with correct subdirectories */
  52 static int filter_prep(struct obd_device *obddev)
  53 {
  54         struct obd_run_ctxt saved;
  55         struct filter_obd *filter = &obddev->u.filter;
  56         struct dentry *dentry;
  57         struct file *file;
  58         struct inode *inode;
  59         loff_t off;
  60         int rc = 0;
  61         char rootid[128];
  62         __u64 lastino = 2;
  63         int mode = 0;
  64
  65         push_ctxt(&saved, &filter->fo_ctxt);
  66         dentry = simple_mkdir(current->fs->pwd, "O", 0700);
  67         CDEBUG(D_INODE, "got/created O: %p\n", dentry);
  68         if (IS_ERR(dentry)) {
  69                 rc = PTR_ERR(dentry);
  70                 CERROR("cannot open/create O: rc = %d\n", rc);
  71                 GOTO(out, rc);
  72         }
  73         filter->fo_dentry_O = dentry;
  74         dentry = simple_mkdir(current->fs->pwd, "P", 0700);
  75         CDEBUG(D_INODE, "got/created P: %p\n", dentry);
  76         if (IS_ERR(dentry)) {
  77                 rc = PTR_ERR(dentry);
  78                 CERROR("cannot open/create P: rc = %d\n", rc);
  79                 GOTO(out_O, rc);
  80         }
  81         CDEBUG(D_INODE, "putting P: %p, count = %d\n", dentry,
  82                atomic_read(&dentry->d_count) - 1);
  83         dput(dentry);
  84         dentry = simple_mkdir(current->fs->pwd, "D", 0700);
  85         CDEBUG(D_INODE, "got/created D: %p\n", dentry);
  86         if (IS_ERR(dentry)) {
  87                 rc = PTR_ERR(dentry);
  88                 CERROR("cannot open/create D: rc = %d\n", rc);
  89                 GOTO(out_O, rc);
  90         }
  91         CDEBUG(D_INODE, "putting D: %p, count = %d\n", dentry,
  92                atomic_read(&dentry->d_count) - 1);
  93         dput(dentry);
  94
  95         /*
  96          * Create directories and/or get dentries for each object type.
  97          * This saves us from having to do multiple lookups for each one.
  98          */
  99         for (mode = 0; mode < (S_IFMT >> S_SHIFT); mode++) {
 100                 char *type = obd_type_by_mode[mode];
 101
 102                 if (!type) {
 103                         filter->fo_dentry_O_mode[mode] = NULL;
 104                         continue;
 105                 }
 106                 dentry = simple_mkdir(filter->fo_dentry_O, type, 0700);
 107                 CDEBUG(D_INODE, "got/created O/%s: %p\n", type, dentry);
 108                 if (IS_ERR(dentry)) {
 109                         rc = PTR_ERR(dentry);
 110                         CERROR("cannot create O/%s: rc = %d\n", type, rc);
 111                         GOTO(out_O_mode, rc);
 112                 }
 113                 filter->fo_dentry_O_mode[mode] = dentry;
 114         }
 115
 116         filter_id(rootid, FILTER_ROOTINO, S_IFDIR);
 117         file = filp_open(rootid, O_RDWR | O_CREAT, 0755);
 118         if (IS_ERR(file)) {
 119                 rc = PTR_ERR(file);
 120                 CERROR("OBD filter: cannot open/create root %s: rc = %d\n",
 121                        rootid, rc);
 122                 GOTO(out_O_mode, rc);
 123         }
 124         filp_close(file, 0);
 125
 126         file = filp_open("D/status", O_RDWR | O_CREAT, 0700);
 127         if ( !file || IS_ERR(file) ) {
 128                 rc = PTR_ERR(file);
 129                 CERROR("OBD filter: cannot open/create status %s: rc = %d\n",
 130                        "D/status", rc);
 131                 GOTO(out_O_mode, rc);
 132         }
 133
 134         /* steal operations */
 135         inode = file->f_dentry->d_inode;
 136         filter->fo_fop = file->f_op;
 137         filter->fo_iop = inode->i_op;
 138         filter->fo_aops = inode->i_mapping->a_ops;
 139
 140         off = 0;
 141         if (inode->i_size == 0) {
 142                 ssize_t retval = file->f_op->write(file, (char *)&lastino,
 143                                                    sizeof(lastino), &off);
 144                 if (retval != sizeof(lastino)) {
 145                         CDEBUG(D_INODE, "OBD filter: error writing lastino\n");
 146                         filp_close(file, 0);
 147                         GOTO(out_O_mode, rc = -EIO);
 148                 }
 149         } else {
 150                 ssize_t retval = file->f_op->read(file, (char *)&lastino,
 151                                                   sizeof(lastino), &off);
 152                 if (retval != sizeof(lastino)) {
 153                         CDEBUG(D_INODE, "OBD filter: error reading lastino\n");
 154                         filp_close(file, 0);
 155                         GOTO(out_O_mode, rc = -EIO);
 156                 }
 157         }
 158         filter->fo_lastino = lastino;
 159         filp_close(file, 0);
 160
 161         rc = 0;
 162  out:
 163         pop_ctxt(&saved);
 164
 165         return(rc);
 166
 167 out_O_mode:
 168         while (--mode >= 0) {
 169                 struct dentry *dentry = filter->fo_dentry_O_mode[mode];
 170                 if (dentry) {
 171                         CDEBUG(D_INODE, "putting O/%s: %p, count = %d\n",
 172                                obd_type_by_mode[mode], dentry,
 173                                atomic_read(&dentry->d_count) - 1);
 174                         dput(dentry);
 175                         filter->fo_dentry_O_mode[mode] = NULL;
 176                 }
 177         }
 178 out_O:
 179         CDEBUG(D_INODE, "putting O: %p, count = %d\n", filter->fo_dentry_O,
 180                atomic_read(&filter->fo_dentry_O->d_count) - 1);
 181         dput(filter->fo_dentry_O);
 182         filter->fo_dentry_O = NULL;
 183         goto out;
 184 }
 185
 186 /* cleanup the filter: write last used object id to status file */
 187 static void filter_post(struct obd_device *obddev)
 188 {
 189         struct obd_run_ctxt saved;
 190         struct filter_obd *filter = &obddev->u.filter;
 191         long rc;
 192         struct file *file;
 193         loff_t off = 0;
 194         int mode;
 195
 196         push_ctxt(&saved, &filter->fo_ctxt);
 197         file = filp_open("D/status", O_RDWR | O_CREAT, 0700);
 198         if (IS_ERR(file)) {
 199                 CERROR("OBD filter: cannot create status file\n");
 200                 goto out;
 201         }
 202         rc = file->f_op->write(file, (char *)&filter->fo_lastino,
 203                        sizeof(filter->fo_lastino), &off);
 204         if (rc != sizeof(filter->fo_lastino))
 205                 CERROR("OBD filter: error writing lastino: rc = %ld\n", rc);
 206
 207         rc = filp_close(file, NULL);
 208         if (rc)
 209                 CERROR("OBD filter: cannot close status file: rc = %ld\n", rc);
 210
 211         for (mode = 0; mode < (S_IFMT >> S_SHIFT); mode++) {
 212                 struct dentry *dentry = filter->fo_dentry_O_mode[mode];
 213                 if (dentry) {
 214                         CDEBUG(D_INODE, "putting O/%s: %p, count = %d\n",
 215                                obd_type_by_mode[mode], dentry,
 216                                atomic_read(&dentry->d_count) - 1);
 217                         dput(dentry);
 218                         filter->fo_dentry_O_mode[mode] = NULL;
 219                 }
 220         }
 221         CDEBUG(D_INODE, "putting O: %p, count = %d\n", filter->fo_dentry_O,
 222                atomic_read(&filter->fo_dentry_O->d_count) - 1);
 223         dput(filter->fo_dentry_O);
 224 out:
 225         pop_ctxt(&saved);
 226 }
 227
 228
 229 static __u64 filter_next_id(struct obd_device *obddev)
 230 {
 231         __u64 id;
 232         spin_lock(&obddev->u.filter.fo_lock);
 233         obddev->u.filter.fo_lastino++;
 234         id =    obddev->u.filter.fo_lastino;
 235         spin_unlock(&obddev->u.filter.fo_lock);
 236         return id;
 237 }
 238
 239 /* how to get files, dentries, inodes from object id's */
 240 /* parent i_sem is already held if needed for exclusivity */
 241 static struct dentry *filter_fid2dentry(struct obd_device *obddev,
 242                                         struct dentry *dparent,
 243                                         __u64 id, __u32 type)
 244 {
 245         struct super_block *sb = obddev->u.filter.fo_sb;
 246         struct dentry *dchild;
 247         char name[32];
 248         int len;
 249         ENTRY;
 250
 251         if (!sb || !sb->s_dev) {
 252                 CERROR("fatal: device not initialized.\n");
 253                 RETURN(ERR_PTR(-ENXIO));
 254         }
 255
 256         if (id == 0) {
 257                 CERROR("fatal: invalid object #0\n");
 258                 LBUG();
 259                 RETURN(ERR_PTR(-ESTALE));
 260         }
 261
 262         if (!(type & S_IFMT)) {
 263                 CERROR("OBD %s, object %Lu has bad type: %o\n", __FUNCTION__,
 264                        (unsigned long long)id, type);
 265                 RETURN(ERR_PTR(-EINVAL));
 266         }
 267
 268         len = sprintf(name, "%Ld", id);
 269         CDEBUG(D_INODE, "opening object O/%s/%s\n", obd_mode_to_type(type),
 270                name);
 271         dchild = lookup_one_len(name, dparent, len);
 272         CDEBUG(D_INODE, "got child obj O/%s/%s: %p, count = %d\n",
 273                obd_mode_to_type(type), name, dchild,
 274                atomic_read(&dchild->d_count));
 275
 276         if (IS_ERR(dchild)) {
 277                 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
 278                 RETURN(dchild);
 279         }
 280
 281         RETURN(dchild);
 282 }
 283
 284 static struct file *filter_obj_open(struct obd_device *obddev,
 285                                     __u64 id, __u32 type)
 286 {
 287         struct super_block *sb = obddev->u.filter.fo_sb;
 288         struct obd_run_ctxt saved;
 289         char name[24];
 290         struct file *file;
 291         ENTRY;
 292
 293         if (!sb || !sb->s_dev) {
 294                 CERROR("fatal: device not initialized.\n");
 295                 RETURN(ERR_PTR(-ENXIO));
 296         }
 297
 298         if (!id) {
 299                 CERROR("fatal: invalid obdo %Lu\n", (unsigned long long)id);
 300                 RETURN(ERR_PTR(-ESTALE));
 301         }
 302
 303         if (!(type & S_IFMT)) {
 304                 CERROR("OBD %s, no type (%Ld), mode %o!\n", __FUNCTION__,
 305                        (unsigned long long)id, type);
 306                 RETURN(ERR_PTR(-EINVAL));
 307         }
 308
 309         filter_id(name, id, type);
 310         push_ctxt(&saved, &obddev->u.filter.fo_ctxt);
 311         file = filp_open(name, O_RDONLY | O_LARGEFILE, 0 /* type? */);
 312         pop_ctxt(&saved);
 313
 314         CDEBUG(D_INODE, "opening obdo %s: rc = %p\n", name, file);
 315
 316         if (IS_ERR(file))
 317                 file = NULL;
 318         RETURN(file);
 319 }
 320
 321 static struct dentry *filter_parent(struct obd_device *obddev, obd_mode mode)
 322 {
 323         struct filter_obd *filter = &obddev->u.filter;
 324
 325         return filter->fo_dentry_O_mode[(mode & S_IFMT) >> S_SHIFT];
 326 }
 327
 328
 329 static struct inode *filter_inode_from_obj(struct obd_device *obddev,
 330                                            __u64 id, __u32 type)
 331 {
 332         struct dentry *dentry;
 333         struct inode *inode;
 334
 335         dentry = filter_fid2dentry(obddev, filter_parent(obddev, type),
 336                                    id, type);
 337         if (IS_ERR(dentry)) {
 338                 CERROR("%s: lookup failed: rc = %ld\n", __FUNCTION__,
 339                        PTR_ERR(dentry));
 340                 RETURN(NULL);
 341         }
 342
 343         lock_kernel();
 344         inode = iget(dentry->d_inode->i_sb, dentry->d_inode->i_ino);
 345         unlock_kernel();
 346         CDEBUG(D_INODE, "put child %p, count = %d\n", dentry,
 347                atomic_read(&dentry->d_count) - 1);
 348         dput(dentry);
 349         CDEBUG(D_INODE, "got inode %p (%ld), count = %d\n", inode, inode->i_ino,
 350                atomic_read(&inode->i_count));
 351         return inode;
 352 }
 353
 354 /* obd methods */
 355 static int filter_connect(struct obd_conn *conn)
 356 {
 357         int rc;
 358
 359         MOD_INC_USE_COUNT;
 360         rc = gen_connect(conn);
 361
 362         if (rc)
 363                 MOD_DEC_USE_COUNT;
 364
 365         return rc;
 366 }
 367
 368 static int filter_disconnect(struct obd_conn *conn)
 369 {
 370         int rc;
 371
 372         rc = gen_disconnect(conn);
 373         if (!rc)
 374                 MOD_DEC_USE_COUNT;
 375
 376         /* XXX cleanup preallocated inodes */
 377         return rc;
 378 }
 379
 380 /* mount the file system (secretly) */
 381 static int filter_setup(struct obd_device *obddev, obd_count len, void *buf)
 382 {
 383         struct obd_ioctl_data* data = buf;
 384         struct filter_obd *filter;
 385         struct vfsmount *mnt;
 386         int err = 0;
 387         ENTRY;
 388
 389         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2)
 390                 RETURN(-EINVAL);
 391
 392         MOD_INC_USE_COUNT;
 393         mnt = do_kern_mount(data->ioc_inlbuf2, 0, data->ioc_inlbuf1, NULL);
 394         err = PTR_ERR(mnt);
 395         if (IS_ERR(mnt))
 396                 GOTO(err_dec, err);
 397
 398         filter = &obddev->u.filter;;
 399         filter->fo_sb = mnt->mnt_root->d_inode->i_sb;
 400         /* XXX is this even possible if do_kern_mount succeeded? */
 401         if (!filter->fo_sb)
 402                 GOTO(err_put, err = -ENODEV);
 403
 404         filter->fo_vfsmnt = mnt;
 405         filter->fo_fstype = strdup(data->ioc_inlbuf2);
 406
 407         filter->fo_ctxt.pwdmnt = mnt;
 408         filter->fo_ctxt.pwd = mnt->mnt_root;
 409         filter->fo_ctxt.fs = KERNEL_DS;
 410
 411         err = filter_prep(obddev);
 412         if (err)
 413                 GOTO(err_kfree, err);
 414         spin_lock_init(&filter->fo_lock);
 415
 416         RETURN(0);
 417
 418 err_kfree:
 419         kfree(filter->fo_fstype);
 420 err_put:
 421         unlock_kernel();
 422         mntput(filter->fo_vfsmnt);
 423         filter->fo_sb = 0;
 424         lock_kernel();
 425
 426 err_dec:
 427         MOD_DEC_USE_COUNT;
 428         return err;
 429 }
 430
 431
 432 static int filter_cleanup(struct obd_device * obddev)
 433 {
 434         struct super_block *sb;
 435         ENTRY;
 436
 437         if (!(obddev->obd_flags & OBD_SET_UP))
 438                 RETURN(0);
 439
 440         if (!list_empty(&obddev->obd_gen_clients)) {
 441                 CERROR("still has clients!\n");
 442                 RETURN(-EBUSY);
 443         }
 444
 445         sb = obddev->u.filter.fo_sb;
 446         if (!obddev->u.filter.fo_sb)
 447                 RETURN(0);
 448
 449         filter_post(obddev);
 450
 451         shrink_dcache_parent(sb->s_root);
 452         unlock_kernel();
 453         mntput(obddev->u.filter.fo_vfsmnt);
 454         obddev->u.filter.fo_sb = 0;
 455         kfree(obddev->u.filter.fo_fstype);
 456
 457         lock_kernel();
 458
 459         MOD_DEC_USE_COUNT;
 460         RETURN(0);
 461 }
 462
 463
 464 static inline void filter_from_inode(struct obdo *oa, struct inode *inode)
 465 {
 466         int type = oa->o_mode & S_IFMT;
 467         ENTRY;
 468
 469         CDEBUG(D_INFO, "src inode %ld, dst obdo %ld valid 0x%08x\n",
 470                inode->i_ino, (long)oa->o_id, oa->o_valid);
 471         obdo_from_inode(oa, inode);
 472         oa->o_mode &= ~S_IFMT;
 473         oa->o_mode |= type;
 474
 475         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
 476                 obd_rdev rdev = kdev_t_to_nr(inode->i_rdev);
 477                 CDEBUG(D_INODE, "copying device %x from inode to obdo\n",
 478                        rdev);
 479                 *((obd_rdev *)oa->o_inline) = rdev;
 480                 oa->o_obdflags |= OBD_FL_INLINEDATA;
 481                 oa->o_valid |= OBD_MD_FLINLINE;
 482         }
 483
 484 #if 0
 485         else if (filter_has_inline(inode)) {
 486                 CDEBUG(D_INFO, "copying inline from inode to obdo\n");
 487                 memcpy(oa->o_inline, inode->u.ext2_i.i_data,
 488                        MIN(sizeof(inode->u.ext2_i.i_data),OBD_INLINESZ));
 489                 oa->o_obdflags |= OBD_FL_INLINEDATA;
 490                 oa->o_valid |= OBD_MD_FLINLINE;
 491         }
 492
 493         if (filter_has_obdmd(inode)) {
 494                 /* XXX this will change when we don't store the obdmd in data */
 495                 CDEBUG(D_INFO, "copying obdmd from inode to obdo\n");
 496                 memcpy(oa->o_obdmd, inode->u.ext2_i.i_data,
 497                        MIN(sizeof(inode->u.ext2_i.i_data),OBD_INLINESZ));
 498                 oa->o_obdflags |= OBD_FL_OBDMDEXISTS;
 499                 oa->o_valid |= OBD_MD_FLOBDMD;
 500         }
 501 #endif
 502         EXIT;
 503 }
 504
 505 static int filter_getattr(struct obd_conn *conn, struct obdo *oa)
 506 {
 507         struct obd_device *obddev;
 508         struct dentry *dentry;
 509         ENTRY;
 510
 511         if (!gen_client(conn)) {
 512                 CDEBUG(D_IOCTL, "fatal: invalid client %u\n", conn->oc_id);
 513                 RETURN(-EINVAL);
 514         }
 515
 516         obddev = conn->oc_dev;
 517         dentry = filter_fid2dentry(obddev, filter_parent(obddev, oa->o_mode),
 518                                    oa->o_id, oa->o_mode);
 519         if (IS_ERR(dentry))
 520                 RETURN(PTR_ERR(dentry));
 521
 522         oa->o_valid &= ~OBD_MD_FLID;
 523         filter_from_inode(oa, dentry->d_inode);
 524
 525         dput(dentry);
 526         RETURN(0);
 527 }
 528
 529 static int filter_setattr(struct obd_conn *conn, struct obdo *oa)
 530 {
 531         struct obd_run_ctxt saved;
 532         struct obd_device *obddev;
 533         struct dentry *dentry;
 534         struct iattr iattr;
 535         struct inode *inode;
 536         int rc;
 537         ENTRY;
 538
 539         if (!gen_client(conn)) {
 540                 CDEBUG(D_IOCTL, "invalid client %u\n", conn->oc_id);
 541                 RETURN(-EINVAL);
 542         }
 543
 544         obddev = conn->oc_dev;
 545         dentry = filter_fid2dentry(obddev, filter_parent(obddev, oa->o_mode),
 546                                    oa->o_id, oa->o_mode);
 547         if (IS_ERR(dentry))
 548                 RETURN(PTR_ERR(dentry));
 549
 550         inode = dentry->d_inode;
 551         iattr_from_obdo(&iattr, oa);
 552         iattr.ia_mode &= ~S_IFMT;
 553         iattr.ia_mode |= S_IFREG;
 554         lock_kernel();
 555         if (iattr.ia_mode & ATTR_SIZE)
 556                 down(&inode->i_sem);
 557         push_ctxt(&saved, &conn->oc_dev->u.filter.fo_ctxt);
 558         if (inode->i_op->setattr)
 559                 rc = inode->i_op->setattr(dentry, &iattr);
 560         else
 561                 rc = inode_setattr(inode, &iattr);
 562         pop_ctxt(&saved);
 563         if (iattr.ia_mode & ATTR_SIZE) {
 564                 up(&inode->i_sem);
 565                 oa->o_valid = OBD_MD_FLBLOCKS | OBD_MD_FLCTIME | OBD_MD_FLMTIME;
 566                 obdo_from_inode(oa, inode);
 567         }
 568         unlock_kernel();
 569
 570         CDEBUG(D_INODE, "put dentry %p, count = %d\n", inode,
 571                atomic_read(&dentry->d_count) - 1);
 572         dput(dentry);
 573         RETURN(rc);
 574 }
 575
 576 static int filter_open(struct obd_conn *conn, struct obdo *oa)
 577 {
 578         struct obd_device *obddev;
 579         struct dentry *dentry;
 580         /* ENTRY; */
 581
 582         if (!gen_client(conn)) {
 583                 CDEBUG(D_IOCTL, "fatal: invalid client %u\n", conn->oc_id);
 584                 RETURN(-EINVAL);
 585         }
 586
 587         obddev = conn->oc_dev;
 588         dentry = filter_fid2dentry(obddev, filter_parent(obddev, oa->o_mode),
 589                                    oa->o_id, oa->o_mode);
 590         if (IS_ERR(dentry))
 591                 RETURN(PTR_ERR(dentry));
 592
 593         return 0;
 594 } /* filter_open */
 595
 596 static int filter_close(struct obd_conn *conn, struct obdo *oa)
 597 {
 598         struct obd_device *obddev;
 599         struct dentry *dentry;
 600         /* ENTRY; */
 601
 602         if (!gen_client(conn)) {
 603                 CDEBUG(D_IOCTL, "fatal: invalid client %u\n", conn->oc_id);
 604                 RETURN(-EINVAL);
 605         }
 606
 607         obddev = conn->oc_dev;
 608         dentry = filter_fid2dentry(obddev, filter_parent(obddev, oa->o_mode),
 609                                    oa->o_id, oa->o_mode);
 610         if (IS_ERR(dentry))
 611                 RETURN(PTR_ERR(dentry));
 612
 613         CDEBUG(D_INODE, "put dentry %p, count = %d\n", dentry,
 614                atomic_read(&dentry->d_count) - 1);
 615         dput(dentry);  /* for the close */
 616         CDEBUG(D_INODE, "put dentry %p, count = %d\n", dentry,
 617                atomic_read(&dentry->d_count) - 1);
 618         dput(dentry);  /* for this call */
 619         return 0;
 620 } /* filter_close */
 621
 622 static int filter_create(struct obd_conn* conn, struct obdo *oa)
 623 {
 624         char name[64];
 625         struct obd_run_ctxt saved;
 626         struct file *file;
 627         int mode;
 628         struct obd_device *obddev = conn->oc_dev;
 629         struct iattr;
 630         ENTRY;
 631
 632         if (!gen_client(conn)) {
 633                 CERROR("invalid client %u\n", conn->oc_id);
 634                 return -EINVAL;
 635         }
 636
 637         oa->o_id = filter_next_id(conn->oc_dev);
 638         if (!(oa->o_mode && S_IFMT)) {
 639                 CERROR("filter obd: no type!\n");
 640                 return -ENOENT;
 641         }
 642
 643         filter_id(name, oa->o_id, oa->o_mode);
 644         mode = (oa->o_mode & ~S_IFMT) | S_IFREG;
 645         push_ctxt(&saved, &obddev->u.filter.fo_ctxt);
 646         file = filp_open(name, O_RDONLY | O_CREAT, mode);
 647         pop_ctxt(&saved);
 648         if (IS_ERR(file)) {
 649                 CERROR("Error mknod obj %s, err %ld\n", name, PTR_ERR(file));
 650                 return -ENOENT;
 651         }
 652         filp_close(file, 0);
 653
 654         /* Set flags for fields we have set in the inode struct */
 655         oa->o_valid |= OBD_MD_FLID | OBD_MD_FLBLKSZ | OBD_MD_FLBLOCKS |
 656                  OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME |
 657                  OBD_MD_FLUID | OBD_MD_FLGID;
 658
 659         /* XXX Hmm, shouldn't we copy the fields into the obdo here? */
 660         return 0;
 661 }
 662
 663 static int filter_destroy(struct obd_conn *conn, struct obdo *oa)
 664 {
 665         struct obd_run_ctxt saved;
 666         struct obd_device *obddev;
 667         struct obd_client *cli;
 668         struct inode *inode;
 669         struct dentry *dir_dentry, *object_dentry;
 670         int rc;
 671         ENTRY;
 672
 673         if (!(cli = gen_client(conn))) {
 674                 CERROR("invalid client %u\n", conn->oc_id);
 675                 RETURN(-EINVAL);
 676         }
 677
 678         CDEBUG(D_INODE, "destroying object %Ld\n",oa->o_id);
 679         obddev = conn->oc_dev;
 680
 681         dir_dentry = filter_parent(obddev, oa->o_mode);
 682         down(&dir_dentry->d_inode->i_sem);
 683
 684         object_dentry = filter_fid2dentry(obddev, dir_dentry, oa->o_id,
 685                                           oa->o_mode);
 686         if (IS_ERR(object_dentry))
 687                 GOTO(out, rc = -ENOENT);
 688
 689         inode = object_dentry->d_inode;
 690         if (inode->i_nlink != 1) {
 691                 CERROR("destroying inode with nlink = %d\n", inode->i_nlink);
 692                 inode->i_nlink = 1;
 693         }
 694         inode->i_mode = S_IFREG;
 695
 696         push_ctxt(&saved, &obddev->u.filter.fo_ctxt);
 697
 698         rc = vfs_unlink(dir_dentry->d_inode, object_dentry);
 699         pop_ctxt(&saved);
 700         CDEBUG(D_INODE, "put child %p, count = %d\n", object_dentry,
 701                atomic_read(&object_dentry->d_count) - 1);
 702         dput(object_dentry);
 703
 704         EXIT;
 705 out:
 706         up(&dir_dentry->d_inode->i_sem);
 707         return rc;
 708 }
 709
 710 /* NB count and offset are used for punch, but not truncate */
 711 static int filter_truncate(struct obd_conn *conn, struct obdo *oa,
 712                            obd_size count, obd_off offset)
 713 {
 714         int error;
 715         ENTRY;
 716
 717         CDEBUG(D_INODE, "calling truncate for object #%Ld, valid = %x, "
 718                "o_size = %Ld\n", oa->o_id, oa->o_valid, oa->o_size);
 719         error = filter_setattr(conn, oa);
 720
 721         RETURN(error);
 722 }
 723
 724 /* buffer must lie in user memory here */
 725 static int filter_read(struct obd_conn *conn, struct obdo *oa, char *buf,
 726                         obd_size *count, obd_off offset)
 727 {
 728         struct file * file;
 729         unsigned long retval;
 730         int err;
 731         ENTRY;
 732
 733         if (!gen_client(conn)) {
 734                 CDEBUG(D_IOCTL, "invalid client %u\n", conn->oc_id);
 735                 RETURN(-EINVAL);
 736         }
 737
 738         file = filter_obj_open(conn->oc_dev, oa->o_id, oa->o_mode);
 739         if (IS_ERR(file))
 740                 RETURN(PTR_ERR(file));
 741
 742         /* count doubles as retval */
 743         retval = file->f_op->read(file, buf, *count, (loff_t *)&offset);
 744         filp_close(file, 0);
 745
 746         if ( retval >= 0 ) {
 747                 err = 0;
 748                 *count = retval;
 749         } else {
 750                 err = retval;
 751                 *count = 0;
 752         }
 753
 754         return err;
 755 }
 756
 757
 758 /* buffer must lie in user memory here */
 759 static int filter_write(struct obd_conn *conn, struct obdo *oa, char *buf,
 760                          obd_size *count, obd_off offset)
 761 {
 762         struct obd_run_ctxt saved;
 763         int err;
 764         struct file * file;
 765         unsigned long retval;
 766         ENTRY;
 767
 768         if (!gen_client(conn)) {
 769                 CDEBUG(D_IOCTL, "invalid client %u\n", conn->oc_id);
 770                 RETURN(-EINVAL);
 771         }
 772
 773         file = filter_obj_open(conn->oc_dev, oa->o_id, oa->o_mode);
 774         if (IS_ERR(file))
 775                 RETURN(PTR_ERR(file));
 776
 777         /* count doubles as retval */
 778         push_ctxt(&saved, &conn->oc_dev->u.filter.fo_ctxt);
 779         retval = file->f_op->write(file, buf, *count, (loff_t *)&offset);
 780         pop_ctxt(&saved);
 781         filp_close(file, 0);
 782
 783         if ( retval >= 0 ) {
 784                 err = 0;
 785                 *count = retval;
 786                 EXIT;
 787         } else {
 788                 err = retval;
 789                 *count = 0;
 790                 EXIT;
 791         }
 792
 793         return err;
 794 } /* filter_write */
 795
 796 static int filter_pgcache_brw(int rw, struct obd_conn *conn, obd_count num_oa,
 797                                struct obdo **oa, obd_count *oa_bufs,
 798                                struct page **pages, obd_size *count,
 799                                obd_off *offset, obd_flag *flags, void *callback)
 800 {
 801         struct obd_run_ctxt      saved;
 802         struct super_block      *sb;
 803         int                      onum;          /* index to oas */
 804         int                      pnum;          /* index to pages (bufs) */
 805         unsigned long            retval;
 806         int                      error;
 807         struct file             *file;
 808         ENTRY;
 809
 810         if (!gen_client(conn)) {
 811                 CDEBUG(D_IOCTL, "invalid client %u\n", conn->oc_id);
 812                 RETURN(-EINVAL);
 813         }
 814
 815         sb = conn->oc_dev->u.filter.fo_sb;
 816         // if (rw == WRITE)
 817         push_ctxt(&saved, &conn->oc_dev->u.filter.fo_ctxt);
 818         pnum = 0; /* pnum indexes buf 0..num_pages */
 819         for (onum = 0; onum < num_oa; onum++) {
 820                 int pg;
 821
 822                 file = filter_obj_open(conn->oc_dev, oa[onum]->o_id,
 823                                        oa[onum]->o_mode);
 824                 if (IS_ERR(file))
 825                         GOTO(out, retval = PTR_ERR(file));
 826
 827                 /* count doubles as retval */
 828                 for (pg = 0; pg < oa_bufs[onum]; pg++) {
 829                         CDEBUG(D_INODE, "OP %d obdo no/pno: (%d,%d) (%ld,%ld) "
 830                                "off count (%Ld,%Ld)\n",
 831                                rw, onum, pnum, file->f_dentry->d_inode->i_ino,
 832                                (unsigned long)offset[pnum] >> PAGE_CACHE_SHIFT,
 833                                (unsigned long long)offset[pnum],
 834                                (unsigned long long)count[pnum]);
 835                         if (rw == WRITE) {
 836                                 loff_t off;
 837                                 char *buffer;
 838                                 off = offset[pnum];
 839                                 buffer = kmap(pages[pnum]);
 840                                 retval = file->f_op->write(file, buffer, count[pnum], &off);
 841                                 kunmap(pages[pnum]);
 842                                 CDEBUG(D_INODE, "retval %ld\n", retval);
 843                         } else {
 844                                 loff_t off = offset[pnum];
 845                                 char *buffer = kmap(pages[pnum]);
 846
 847                                 if (off >= file->f_dentry->d_inode->i_size) {
 848                                         memset(buffer, 0, count[pnum]);
 849                                         retval = count[pnum];
 850                                 } else {
 851                                         retval = file->f_op->read(file, buffer, count[pnum], &off);
 852                                 }
 853                                 kunmap(pages[pnum]);
 854
 855                                 if (retval != count[pnum]) {
 856                                         filp_close(file, 0);
 857                                         GOTO(out, retval = -EIO);
 858                                 }
 859                                 CDEBUG(D_INODE, "retval %ld\n", retval);
 860                         }
 861                         pnum++;
 862                 }
 863                 /* sizes and blocks are set by generic_file_write */
 864                 /* ctimes/mtimes will follow with a setattr call */
 865                 filp_close(file, 0);
 866         }
 867
 868         EXIT;
 869 out:
 870         // if (rw == WRITE)
 871         pop_ctxt(&saved);
 872         error = (retval >= 0) ? 0 : retval;
 873         return error;
 874 }
 875
 876
 877 struct inode *ioobj_to_inode(struct obd_conn *conn, struct obd_ioobj *o)
 878 {
 879         struct super_block *sb = conn->oc_dev->u.filter.fo_sb;
 880         struct inode *inode = NULL;
 881         ENTRY;
 882
 883         if (!sb || !sb->s_dev) {
 884                 CDEBUG(D_SUPER, "fatal: device not initialized.\n");
 885                 RETURN(NULL);
 886         }
 887
 888         if ( !o->ioo_id ) {
 889                 CDEBUG(D_INODE, "fatal: invalid obdo %lu\n", (long)o->ioo_id);
 890                 RETURN(NULL);
 891         }
 892
 893         inode = filter_inode_from_obj(conn->oc_dev, o->ioo_id, S_IFREG);
 894         if (!inode || inode->i_nlink == 0 || is_bad_inode(inode)) {
 895                 CERROR("from obdo - fatal: invalid inode %ld (%s).\n",
 896                        (long)o->ioo_id, inode ? inode->i_nlink ? "bad inode" :
 897                        "no links" : "NULL");
 898                 iput(inode);
 899                 RETURN(NULL);
 900         }
 901
 902         RETURN(inode);
 903 }
 904
 905 /*
 906  * Calculate the number of buffer credits needed to write multiple pages in
 907  * a single ext3/extN transaction.  No, this shouldn't be here, but as yet
 908  * ext3 doesn't have a nice API for calculating this sort of thing in advance.
 909  *
 910  * See comment above ext3_writepage_trans_blocks for details.  We assume
 911  * no data journaling is being done, but it does allow for all of the pages
 912  * being non-contiguous.  If we are guaranteed contiguous pages we could
 913  * reduce the number of (d)indirect blocks a lot.
 914  *
 915  * With N blocks per page and P pages, for each inode we have at most:
 916  * N*P indirect
 917  * min(N*P, blocksize/4 + 1) dindirect blocks
 918  * 1 tindirect
 919  *
 920  * For the entire filesystem, we have at most:
 921  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
 922  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
 923  * 1 inode block
 924  * 1 superblock
 925  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
 926  */
 927 static int ext3_credits_needed(struct super_block *sb, int objcount,
 928                                struct obd_ioobj *obj)
 929 {
 930         struct obd_ioobj *o = obj;
 931         int blockpp = 1 << (PAGE_CACHE_SHIFT - sb->s_blocksize_bits);
 932         int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
 933         int nbitmaps = 0;
 934         int ngdblocks = 0;
 935         int needed = objcount + 1;
 936         int i;
 937
 938         for (i = 0; i < objcount; i++, o++) {
 939                 int nblocks = o->ioo_bufcnt * blockpp;
 940                 int ndindirect = min(nblocks, addrpp + 1);
 941                 int nindir = nblocks + ndindirect + 1;
 942
 943                 nbitmaps += nindir + nblocks;
 944                 ngdblocks += nindir + nblocks;
 945
 946                 needed += nindir;
 947         }
 948
 949         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
 950                 nbitmaps = EXT3_SB(sb)->s_groups_count;
 951         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
 952                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
 953
 954         needed += nbitmaps + ngdblocks;
 955
 956 #ifdef CONFIG_QUOTA
 957         /* We assume that there will be 1 bit set in s_dquot.flags for each
 958          * quota file that is active.  This is at least true for now.
 959          */
 960         needed += hweight32(sb_any_quota_enabled(sb)) *
 961                 EXT3_SINGLEDATA_TRANS_BLOCKS;
 962 #endif
 963
 964         return needed;
 965 }
 966
 967 /* We have to start a huge journal transaction here to hold all of the
 968  * metadata for the pages being written here.  This is necessitated by
 969  * the fact that we do lots of prepare_write operations before we do
 970  * any of the matching commit_write operations, so even if we split
 971  * up to use "smaller" transactions none of them could complete until
 972  * all of them were opened.  By having a single journal transaction,
 973  * we eliminate duplicate reservations for common blocks like the
 974  * superblock and group descriptors or bitmaps.
 975  *
 976  * We will start the transaction here, but each prepare_write will
 977  * add a refcount to the transaction, and each commit_write will
 978  * remove a refcount.  The transaction will be closed when all of
 979  * the pages have been written.
 980  */
 981 static void *ext3_filter_journal_start(struct filter_obd *filter,
 982                                        int objcount, struct obd_ioobj *obj,
 983                                        int niocount, struct niobuf_remote *nb)
 984 {
 985         journal_t *journal = NULL;
 986         handle_t *handle = NULL;
 987         int needed;
 988
 989         /* Assumes ext3 and extN have same sb_info layout, but avoids issues
 990          * with having extN built properly before filterobd for now.
 991          */
 992         journal = EXT3_SB(filter->fo_sb)->s_journal;
 993         needed = ext3_credits_needed(filter->fo_sb, objcount, obj);
 994
 995         /* The number of blocks we could _possibly_ dirty can very large.
 996          * We reduce our request if it is absurd (and we couldn't get that
 997          * many credits for a single handle anyways).
 998          *
 999          * At some point we have to limit the size of I/Os sent at one time,
1000          * increase the size of the journal, or we have to calculate the
1001          * actual journal requirements more carefully by checking all of
1002          * the blocks instead of being maximally pessimistic.  It remains to
1003          * be seen if this is a real problem or not.
1004          */
1005         if (needed > journal->j_max_transaction_buffers) {
1006                 CERROR("want too many journal credits (%d) using %d instead\n",
1007                        needed, journal->j_max_transaction_buffers);
1008                 needed = journal->j_max_transaction_buffers;
1009         }
1010
1011         handle = journal_start(journal, needed);
1012         if (IS_ERR(handle))
1013                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
1014                        PTR_ERR(handle));
1015
1016         return(handle);
1017 }
1018
1019 static void *filter_journal_start(void **journal_save,
1020                                   struct filter_obd *filter,
1021                                   int objcount, struct obd_ioobj *obj,
1022                                   int niocount, struct niobuf_remote *nb)
1023 {
1024         void *handle = NULL;
1025
1026         /* This may not be necessary - we probably never have a
1027          * transaction started when we enter here, so we can
1028          * remove the saving of the journal state entirely.
1029          * For now leave it in just to see if it ever happens.
1030          */
1031         *journal_save = current->journal_info;
1032         if (*journal_save) {
1033                 CERROR("Already have handle %p???\n", *journal_save);
1034                 LBUG();
1035                 current->journal_info = NULL;
1036         }
1037
1038         if (!strcmp(filter->fo_fstype, "ext3") ||
1039             !strcmp(filter->fo_fstype, "extN"))
1040                 handle = ext3_filter_journal_start(filter, objcount, obj,
1041                                                    niocount, nb);
1042         return handle;
1043 }
1044
1045 static int ext3_filter_journal_stop(void *handle)
1046 {
1047         int rc;
1048
1049         /* We got a refcount on the handle for each call to prepare_write,
1050          * so we can drop the "parent" handle here to avoid the need for
1051          * osc to call back into filterobd to close the handle.  The
1052          * remaining references will be dropped in commit_write.
1053          */
1054         rc = journal_stop((handle_t *)handle);
1055
1056         return rc;
1057 }
1058
1059 static int filter_journal_stop(void *journal_save, struct filter_obd *filter,
1060                                void *handle)
1061 {
1062         int rc = 0;
1063
1064         if (!strcmp(filter->fo_fstype, "ext3") ||
1065             !strcmp(filter->fo_fstype, "extN"))
1066                 rc = ext3_filter_journal_stop(handle);
1067
1068         current->journal_info = journal_save;
1069
1070         return rc;
1071 }
1072
1073 struct page *filter_get_page_write(struct inode *inode, unsigned long index,
1074                                    struct niobuf_local *lnb)
1075 {
1076         struct address_space *mapping = inode->i_mapping;
1077         struct page *page;
1078         int rc;
1079
1080         //ASSERT_PAGE_INDEX(index, GOTO(err, rc = -EINVAL));
1081         page = grab_cache_page_nowait(mapping, index); /* locked page */
1082
1083         /* This page is currently locked, so we grab a new one temporarily */
1084         if (!page) {
1085                 unsigned long addr;
1086                 addr = __get_free_pages(GFP_KERNEL, 0);
1087                 if (!addr) {
1088                         CERROR("no memory for a temp page\n");
1089                         LBUG();
1090                         GOTO(err, rc = -ENOMEM);
1091                 }
1092                 page = virt_to_page(addr);
1093                 kmap(page);
1094                 page->index = index;
1095                 lnb->flags |= N_LOCAL_TEMP_PAGE;
1096         } else if (!IS_ERR(page)) {
1097                 /* Note: Called with "O" and "PAGE_SIZE" this is essentially
1098                  * a no-op for most filesystems, because we write the whole
1099                  * page.  For partial-page I/O this will read in the page.
1100                  */
1101                 kmap(page);
1102                 rc = mapping->a_ops->prepare_write(NULL, page, 0, PAGE_SIZE);
1103                 if (rc) {
1104                         CERROR("page index %lu, rc = %d\n", index, rc);
1105                         if (rc != -ENOSPC)
1106                                 LBUG();
1107                         GOTO(err_unlock, rc);
1108                 }
1109                 /* XXX not sure if we need this if we are overwriting page */
1110                 if (PageError(page)) {
1111                         CERROR("error on page index %lu, rc = %d\n", index, rc);
1112                         LBUG();
1113                         GOTO(err_unlock, rc = -EIO);
1114                 }
1115         }
1116         return page;
1117
1118 err_unlock:
1119         unlock_page(page);
1120         lustre_put_page(page);
1121 err:
1122         return ERR_PTR(rc);
1123 }
1124
1125 static int filter_preprw(int cmd, struct obd_conn *conn,
1126                          int objcount, struct obd_ioobj *obj,
1127                          int niocount, struct niobuf_remote *nb,
1128                          struct niobuf_local *res, void **desc_private)
1129 {
1130         struct obd_run_ctxt saved;
1131         struct obd_device *obddev;
1132         struct obd_ioobj *o = obj;
1133         struct niobuf_remote *b = nb;
1134         struct niobuf_local *r = res;
1135         void *journal_save = NULL;
1136         int rc = 0;
1137         int i;
1138         ENTRY;
1139
1140         memset(res, 0, sizeof(*res) * niocount);
1141         obddev = conn->oc_dev;
1142
1143         push_ctxt(&saved, &obddev->u.filter.fo_ctxt);
1144
1145         if (cmd == OBD_BRW_WRITE) {
1146                 *desc_private = filter_journal_start(&journal_save,
1147                                                      &obddev->u.filter,
1148                                                      objcount, obj, niocount,
1149                                                      nb);
1150                 if (IS_ERR(*desc_private))
1151                         GOTO(out_ctxt, rc = PTR_ERR(*desc_private));
1152         }
1153
1154         for (i = 0; i < objcount; i++, o++) {
1155                 struct dentry *dentry;
1156                 struct inode *inode;
1157                 int j;
1158
1159                 dentry = filter_fid2dentry(obddev,
1160                                            filter_parent(obddev, S_IFREG),
1161                                            o->ioo_id, S_IFREG);
1162                 if (IS_ERR(dentry))
1163                         GOTO(out_ctxt, rc = PTR_ERR(dentry));
1164                 inode = dentry->d_inode;
1165
1166                 for (j = 0; j < o->ioo_bufcnt; j++, b++, r++) {
1167                         unsigned long index = b->offset >> PAGE_SHIFT;
1168                         struct page *page;
1169
1170                         /* XXX  We _might_ change this to a dcount if we
1171                          * wanted to pass a dentry pointer in the niobuf
1172                          * to avoid doing so many igets on an inode we
1173                          * already have.  It appears to be solely for the
1174                          * purpose of having a refcount that we can drop
1175                          * in commitrw where we get one call per page.
1176                          */
1177                         if (j > 0)
1178                                 r->dentry = dget(dentry);
1179                         else
1180                                 r->dentry = dentry;
1181
1182                         /* FIXME: we need to iput all inodes on error */
1183                         if (!inode)
1184                                 GOTO(out_ctxt, rc = -EINVAL);
1185
1186                         if (cmd == OBD_BRW_WRITE) {
1187                                 page = filter_get_page_write(inode, index, r);
1188
1189                                 /* We unlock the page to avoid deadlocks with
1190                                  * the page I/O because we are preparing
1191                                  * multiple pages at one time and we have lock
1192                                  * ordering problems.  Lustre I/O and disk I/O
1193                                  * on this page can happen concurrently.
1194                                  */
1195                         } else
1196                                 page = lustre_get_page_read(inode, index);
1197
1198                         /* FIXME: we need to clean up here... */
1199                         if (IS_ERR(page))
1200                                 GOTO(out_ctxt, rc = PTR_ERR(page));
1201
1202                         r->addr = page_address(page);
1203                         r->offset = b->offset;
1204                         r->page = page;
1205                         r->len = PAGE_SIZE;
1206                 }
1207         }
1208
1209         if (cmd == OBD_BRW_WRITE) {
1210                 /* FIXME: need to clean up here */
1211                 rc = filter_journal_stop(journal_save, &obddev->u.filter,
1212                                          *desc_private);
1213         }
1214 out_ctxt:
1215         pop_ctxt(&saved);
1216         RETURN(rc);
1217 }
1218
1219 static int filter_write_locked_page(struct niobuf_local *lnb)
1220 {
1221         struct page *lpage;
1222         int rc;
1223
1224         lpage = lustre_get_page_write(lnb->dentry->d_inode, lnb->page->index);
1225         /* XXX */
1226
1227         memcpy(page_address(lpage), kmap(lnb->page), PAGE_SIZE);
1228         kunmap(lnb->page);
1229         __free_pages(lnb->page, 0);
1230
1231         rc = lustre_commit_page(lpage, 0, PAGE_SIZE);
1232         dput(lnb->dentry);
1233
1234         return rc;
1235 }
1236
1237 static int filter_commitrw(int cmd, struct obd_conn *conn,
1238                            int objcount, struct obd_ioobj *obj,
1239                            int niocount, struct niobuf_local *res,
1240                            void *private)
1241 {
1242         struct obd_run_ctxt saved;
1243         struct obd_ioobj *o = obj;
1244         struct niobuf_local *r = res;
1245         void *journal_save;
1246         int found_locked = 0;
1247         int i;
1248         ENTRY;
1249
1250         // if (cmd == OBD_BRW_WRITE)
1251         push_ctxt(&saved, &conn->oc_dev->u.filter.fo_ctxt);
1252         journal_save = current->journal_info;
1253         if (journal_save)
1254                 CERROR("Existing handle %p???\n", journal_save);
1255         current->journal_info = private;
1256         for (i = 0; i < objcount; i++, obj++) {
1257                 int j;
1258                 for (j = 0 ; j < o->ioo_bufcnt ; j++, r++) {
1259                         struct page *page = r->page;
1260
1261                         /* If there was an error setting up a particular page
1262                          * for I/O we still need to continue with the rest of
1263                          * the pages in order to balance prepate/commit_write
1264                          * calls, and to complete as much I/O as possible.
1265                          */
1266                         if (!page)
1267                                 LBUG();
1268
1269                         if (r->flags & N_LOCAL_TEMP_PAGE) {
1270                                 found_locked = 1;
1271                                 continue;
1272                         }
1273
1274                         if (cmd == OBD_BRW_WRITE) {
1275                                 int rc;
1276                                 rc = lustre_commit_page(page, 0, PAGE_SIZE);
1277
1278                                 /* FIXME: still need to iput the other inodes */
1279                                 if (rc)
1280                                         RETURN(rc);
1281                         } else
1282                                 lustre_put_page(page);
1283
1284                         CDEBUG(D_INODE,
1285                                "put inode %p (%ld), count = %d, nlink = %d\n",
1286                                page->mapping->host,
1287                                page->mapping->host->i_ino,
1288                                atomic_read(&page->mapping->host->i_count) - 1,
1289                                page->mapping->host->i_nlink);
1290                         dput(r->dentry);
1291                 }
1292         }
1293         if (!found_locked)
1294                 goto out;
1295
1296         for (i = 0; i < objcount; i++, obj++) {
1297                 int j;
1298                 for (j = 0 ; j < o->ioo_bufcnt ; j++, r++) {
1299                         int rc;
1300                         if (!(r->flags & N_LOCAL_TEMP_PAGE))
1301                                 continue;
1302
1303                         rc = filter_write_locked_page(r);
1304                         /* XXX */
1305                 }
1306         }
1307
1308 out:
1309         current->journal_info = journal_save;
1310         pop_ctxt(&saved);
1311         RETURN(0);
1312 }
1313
1314 static int filter_statfs(struct obd_conn *conn, struct statfs * statfs)
1315 {
1316         struct super_block *sb;
1317         int err;
1318         ENTRY;
1319
1320         if (!gen_client(conn)) {
1321                 CDEBUG(D_IOCTL, "invalid client %u\n", conn->oc_id);
1322                 RETURN(-EINVAL);
1323         }
1324
1325         sb = conn->oc_dev->u.filter.fo_sb;
1326
1327         err = sb->s_op->statfs(sb, statfs);
1328         RETURN(err);
1329 } /* filter_statfs */
1330
1331
1332 static int filter_get_info(struct obd_conn *conn, obd_count keylen,
1333                            void *key, obd_count *vallen, void **val)
1334 {
1335         struct obd_device *obddev;
1336         struct obd_client * cli;
1337         ENTRY;
1338
1339         if (!(cli = gen_client(conn))) {
1340                 CDEBUG(D_IOCTL, "invalid client %u\n", conn->oc_id);
1341                 RETURN(-EINVAL);
1342         }
1343
1344         obddev = conn->oc_dev;
1345
1346         if ( keylen == strlen("blocksize") &&
1347              memcmp(key, "blocksize", keylen) == 0 ) {
1348                 *vallen = sizeof(long);
1349                 *val = (void *)(long)obddev->u.filter.fo_sb->s_blocksize;
1350                 RETURN(0);
1351         }
1352
1353         if ( keylen == strlen("blocksize_bits") &&
1354              memcmp(key, "blocksize_bits", keylen) == 0 ){
1355                 *vallen = sizeof(long);
1356                 *val = (void *)(long)obddev->u.filter.fo_sb->s_blocksize_bits;
1357                 RETURN(0);
1358         }
1359
1360         if ( keylen == strlen("root_ino") &&
1361              memcmp(key, "root_ino", keylen) == 0 ){
1362                 *vallen = sizeof(long);
1363                 *val = (void *)(long)FILTER_ROOTINO;
1364                 RETURN(0);
1365         }
1366
1367         CDEBUG(D_IOCTL, "invalid key\n");
1368         RETURN(-EINVAL);
1369 }
1370
1371
1372 static struct obd_ops filter_obd_ops = {
1373         o_get_info:    filter_get_info,
1374         o_setup:       filter_setup,
1375         o_cleanup:     filter_cleanup,
1376         o_connect:     filter_connect,
1377         o_disconnect:  filter_disconnect,
1378         o_statfs:      filter_statfs,
1379         o_getattr:     filter_getattr,
1380         o_create:      filter_create,
1381         o_setattr:     filter_setattr,
1382         o_destroy:     filter_destroy,
1383         o_open:        filter_open,
1384         o_close:       filter_close,
1385         o_read:        filter_read,
1386         o_write:       filter_write,
1387         o_brw:         filter_pgcache_brw,
1388         o_punch:       filter_truncate,
1389         o_preprw:      filter_preprw,
1390         o_commitrw:    filter_commitrw
1391 #if 0
1392         o_preallocate: filter_preallocate_inodes,
1393         o_migrate:     filter_migrate,
1394         o_copy:        gen_copy_data,
1395         o_iterate:     filter_iterate
1396 #endif
1397 };
1398
1399
1400 static int __init obdfilter_init(void)
1401 {
1402         printk(KERN_INFO "Filtering OBD driver  v0.001, braam@clusterfs.com\n");
1403         return obd_register_type(&filter_obd_ops, OBD_FILTER_DEVICENAME);
1404 }
1405
1406 static void __exit obdfilter_exit(void)
1407 {
1408         obd_unregister_type(OBD_FILTER_DEVICENAME);
1409 }
1410
1411 MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
1412 MODULE_DESCRIPTION("Lustre Filtering OBD driver v1.0");
1413 MODULE_LICENSE("GPL");
1414
1415 module_init(obdfilter_init);
1416 module_exit(obdfilter_exit);