lustre/lvfs/lvfs_linux.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  lustre/lib/fsfilt_ext3.c
   5  *  Lustre filesystem abstraction routines
   6  *
   7  *  Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   8  *   Author: Andreas Dilger <adilger@clusterfs.com>
   9  *
  10  *   This file is part of Lustre, http://www.lustre.org.
  11  *
  12  *   Lustre is free software; you can redistribute it and/or
  13  *   modify it under the terms of version 2 of the GNU General Public
  14  *   License as published by the Free Software Foundation.
  15  *
  16  *   Lustre is distributed in the hope that it will be useful,
  17  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  19  *   GNU General Public License for more details.
  20  *
  21  *   You should have received a copy of the GNU General Public License
  22  *   along with Lustre; if not, write to the Free Software
  23  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  24  */
  25
  26 #ifndef EXPORT_SYMTAB
  27 # define EXPORT_SYMTAB
  28 #endif
  29
  30 #define DEBUG_SUBSYSTEM S_FILTER
  31
  32 #include <linux/version.h>
  33 #include <linux/fs.h>
  34 #include <asm/unistd.h>
  35 #include <linux/jbd.h>
  36 #include <linux/slab.h>
  37 #include <linux/pagemap.h>
  38 #include <linux/quotaops.h>
  39 #include <linux/version.h>
  40 #include <linux/kp30.h>
  41 #include <linux/lustre_fsfilt.h>
  42 #include <linux/obd.h>
  43 #include <linux/obd_class.h>
  44 #include <linux/module.h>
  45 #include <linux/init.h>
  46 #include <linux/lustre_compat25.h>
  47 #include <linux/lvfs.h>
  48 #include "lvfs_internal.h"
  49
  50 #include <linux/obd.h>
  51 #include <linux/lustre_lib.h>
  52
  53 atomic_t obd_memory;
  54 int obd_memmax;
  55
  56
  57 /* Debugging check only needed during development */
  58 #ifdef OBD_CTXT_DEBUG
  59 # define ASSERT_CTXT_MAGIC(magic) LASSERT((magic) == OBD_RUN_CTXT_MAGIC)
  60 # define ASSERT_NOT_KERNEL_CTXT(msg) LASSERT(!segment_eq(get_fs(), get_ds()))
  61 # define ASSERT_KERNEL_CTXT(msg) LASSERT(segment_eq(get_fs(), get_ds()))
  62 #else
  63 # define ASSERT_CTXT_MAGIC(magic) do {} while(0)
  64 # define ASSERT_NOT_KERNEL_CTXT(msg) do {} while(0)
  65 # define ASSERT_KERNEL_CTXT(msg) do {} while(0)
  66 #endif
  67
  68 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0))
  69 #define current_ngroups current->group_info->ngroups
  70 #define current_groups current->group_info->small_block
  71 #else
  72 #define current_ngroups current->ngroups
  73 #define current_groups current->groups
  74 #endif
  75
  76 /* push / pop to root of obd store */
  77 void push_ctxt(struct lvfs_run_ctxt *save, struct lvfs_run_ctxt *new_ctx,
  78                struct lvfs_ucred *uc)
  79 {
  80         //ASSERT_NOT_KERNEL_CTXT("already in kernel context!\n");
  81         ASSERT_CTXT_MAGIC(new_ctx->magic);
  82         OBD_SET_CTXT_MAGIC(save);
  83
  84         /*
  85         CDEBUG(D_INFO,
  86                "= push %p->%p = cur fs %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
  87                save, current, current->fs, current->fs->pwd,
  88                atomic_read(&current->fs->pwd->d_count),
  89                atomic_read(&current->fs->pwd->d_inode->i_count),
  90                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
  91                current->fs->pwdmnt,
  92                atomic_read(&current->fs->pwdmnt->mnt_count));
  93         */
  94
  95         save->fs = get_fs();
  96         LASSERT(atomic_read(&current->fs->pwd->d_count));
  97         LASSERT(atomic_read(&new_ctx->pwd->d_count));
  98         save->pwd = dget(current->fs->pwd);
  99         save->pwdmnt = mntget(current->fs->pwdmnt);
 100         save->ngroups = current_ngroups;
 101
 102         LASSERT(save->pwd);
 103         LASSERT(save->pwdmnt);
 104         LASSERT(new_ctx->pwd);
 105         LASSERT(new_ctx->pwdmnt);
 106
 107         if (uc) {
 108                 save->luc.luc_fsuid = current->fsuid;
 109                 save->luc.luc_fsgid = current->fsgid;
 110                 save->luc.luc_cap = current->cap_effective;
 111                 save->luc.luc_suppgid1 = current_groups[0];
 112                 save->luc.luc_suppgid2 = current_groups[1];
 113
 114                 current->fsuid = uc->luc_fsuid;
 115                 current->fsgid = uc->luc_fsgid;
 116                 current->cap_effective = uc->luc_cap;
 117                 current_ngroups = 0;
 118
 119                 if (uc->luc_suppgid1 != -1)
 120                         current_groups[current_ngroups++] = uc->luc_suppgid1;
 121                 if (uc->luc_suppgid2 != -1)
 122                         current_groups[current_ngroups++] = uc->luc_suppgid2;
 123         }
 124         set_fs(new_ctx->fs);
 125         set_fs_pwd(current->fs, new_ctx->pwdmnt, new_ctx->pwd);
 126
 127         /*
 128         CDEBUG(D_INFO,
 129                "= push %p->%p = cur fs %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 130                new_ctx, current, current->fs, current->fs->pwd,
 131                atomic_read(&current->fs->pwd->d_count),
 132                atomic_read(&current->fs->pwd->d_inode->i_count),
 133                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 134                current->fs->pwdmnt,
 135                atomic_read(&current->fs->pwdmnt->mnt_count));
 136         */
 137 }
 138 EXPORT_SYMBOL(push_ctxt);
 139
 140 void pop_ctxt(struct lvfs_run_ctxt *saved, struct lvfs_run_ctxt *new_ctx,
 141               struct lvfs_ucred *uc)
 142 {
 143         //printk("pc0");
 144         ASSERT_CTXT_MAGIC(saved->magic);
 145         //printk("pc1");
 146         ASSERT_KERNEL_CTXT("popping non-kernel context!\n");
 147
 148         /*
 149         CDEBUG(D_INFO,
 150                " = pop  %p==%p = cur %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 151                new_ctx, current, current->fs, current->fs->pwd,
 152                atomic_read(&current->fs->pwd->d_count),
 153                atomic_read(&current->fs->pwd->d_inode->i_count),
 154                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 155                current->fs->pwdmnt,
 156                atomic_read(&current->fs->pwdmnt->mnt_count));
 157         */
 158
 159         LASSERT(current->fs->pwd == new_ctx->pwd);
 160         LASSERT(current->fs->pwdmnt == new_ctx->pwdmnt);
 161
 162         set_fs(saved->fs);
 163         set_fs_pwd(current->fs, saved->pwdmnt, saved->pwd);
 164
 165         dput(saved->pwd);
 166         mntput(saved->pwdmnt);
 167         if (uc) {
 168                 current->fsuid = saved->luc.luc_fsuid;
 169                 current->fsgid = saved->luc.luc_fsgid;
 170                 current->cap_effective = saved->luc.luc_cap;
 171                 current_ngroups = saved->ngroups;
 172                 current_groups[0] = saved->luc.luc_suppgid1;
 173                 current_groups[1] = saved->luc.luc_suppgid2;
 174         }
 175
 176         /*
 177         CDEBUG(D_INFO,
 178                "= pop  %p->%p = cur fs %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 179                saved, current, current->fs, current->fs->pwd,
 180                atomic_read(&current->fs->pwd->d_count),
 181                atomic_read(&current->fs->pwd->d_inode->i_count),
 182                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 183                current->fs->pwdmnt,
 184                atomic_read(&current->fs->pwdmnt->mnt_count));
 185         */
 186 }
 187 EXPORT_SYMBOL(pop_ctxt);
 188
 189 /* utility to make a file */
 190 struct dentry *simple_mknod(struct dentry *dir, char *name, int mode, int fix)
 191 {
 192         struct dentry *dchild;
 193         int err = 0;
 194         ENTRY;
 195
 196         ASSERT_KERNEL_CTXT("kernel doing mknod outside kernel context\n");
 197         CDEBUG(D_INODE, "creating file %*s\n", (int)strlen(name), name);
 198
 199         dchild = ll_lookup_one_len(name, dir, strlen(name));
 200         if (IS_ERR(dchild))
 201                 GOTO(out_up, dchild);
 202
 203         if (dchild->d_inode) {
 204                 int old_mode = dchild->d_inode->i_mode;
 205                 if (!S_ISREG(old_mode))
 206                         GOTO(out_err, err = -EEXIST);
 207
 208                 /* Fixup file permissions if necessary */
 209                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 210                         CWARN("fixing permissions on %s from %o to %o\n",
 211                               name, old_mode, mode);
 212                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 213                                                   (old_mode & ~S_IALLUGO);
 214                         mark_inode_dirty(dchild->d_inode);
 215                 }
 216                 GOTO(out_up, dchild);
 217         }
 218
 219         err = ll_vfs_create(dir->d_inode, dchild, (mode & ~S_IFMT) | S_IFREG,
 220                             NULL);
 221         if (err)
 222                 GOTO(out_err, err);
 223
 224         RETURN(dchild);
 225
 226 out_err:
 227         dput(dchild);
 228         dchild = ERR_PTR(err);
 229 out_up:
 230         return dchild;
 231 }
 232 EXPORT_SYMBOL(simple_mknod);
 233
 234 /* utility to make a directory */
 235 struct dentry *simple_mkdir(struct dentry *dir, char *name, int mode, int fix)
 236 {
 237         struct dentry *dchild;
 238         int err = 0;
 239         ENTRY;
 240
 241         ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
 242         CDEBUG(D_INODE, "creating directory %*s\n", (int)strlen(name), name);
 243         dchild = ll_lookup_one_len(name, dir, strlen(name));
 244         if (IS_ERR(dchild))
 245                 GOTO(out_up, dchild);
 246
 247         if (dchild->d_inode) {
 248                 int old_mode = dchild->d_inode->i_mode;
 249                 if (!S_ISDIR(old_mode))
 250                         GOTO(out_err, err = -ENOTDIR);
 251
 252                 /* Fixup directory permissions if necessary */
 253                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 254                         CWARN("fixing permissions on %s from %o to %o\n",
 255                               name, old_mode, mode);
 256                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 257                                                   (old_mode & ~S_IALLUGO);
 258                         mark_inode_dirty(dchild->d_inode);
 259                 }
 260                 GOTO(out_up, dchild);
 261         }
 262
 263         err = vfs_mkdir(dir->d_inode, dchild, mode);
 264         if (err)
 265                 GOTO(out_err, err);
 266
 267         RETURN(dchild);
 268
 269 out_err:
 270         dput(dchild);
 271         dchild = ERR_PTR(err);
 272 out_up:
 273         return dchild;
 274 }
 275 EXPORT_SYMBOL(simple_mkdir);
 276
 277 /*
 278  * Read a file from within kernel context.  Prior to calling this
 279  * function we should already have done a push_ctxt().
 280  */
 281 int lustre_fread(struct file *file, void *buf, int len, loff_t *off)
 282 {
 283         ASSERT_KERNEL_CTXT("kernel doing read outside kernel context\n");
 284         if (!file || !file->f_op || !file->f_op->read || !off)
 285                 RETURN(-ENOSYS);
 286
 287         return file->f_op->read(file, buf, len, off);
 288 }
 289 EXPORT_SYMBOL(lustre_fread);
 290
 291 /*
 292  * Write a file from within kernel context.  Prior to calling this
 293  * function we should already have done a push_ctxt().
 294  */
 295 int lustre_fwrite(struct file *file, const void *buf, int len, loff_t *off)
 296 {
 297         ENTRY;
 298         ASSERT_KERNEL_CTXT("kernel doing write outside kernel context\n");
 299         if (!file)
 300                 RETURN(-ENOENT);
 301         if (!file->f_op)
 302                 RETURN(-ENOSYS);
 303         if (!off)
 304                 RETURN(-EINVAL);
 305
 306         if (!file->f_op->write)
 307                 RETURN(-EROFS);
 308
 309         RETURN(file->f_op->write(file, buf, len, off));
 310 }
 311 EXPORT_SYMBOL(lustre_fwrite);
 312
 313 /*
 314  * Sync a file from within kernel context.  Prior to calling this
 315  * function we should already have done a push_ctxt().
 316  */
 317 int lustre_fsync(struct file *file)
 318 {
 319         ENTRY;
 320         ASSERT_KERNEL_CTXT("kernel doing sync outside kernel context\n");
 321         if (!file || !file->f_op || !file->f_op->fsync)
 322                 RETURN(-ENOSYS);
 323
 324         RETURN(file->f_op->fsync(file, file->f_dentry, 0));
 325 }
 326 EXPORT_SYMBOL(lustre_fsync);
 327
 328 struct l_file *l_dentry_open(struct lvfs_run_ctxt *ctxt, struct l_dentry *de,
 329                              int flags)
 330 {
 331         mntget(ctxt->pwdmnt);
 332         return dentry_open(de, ctxt->pwdmnt, flags);
 333 }
 334 EXPORT_SYMBOL(l_dentry_open);
 335
 336 static int l_filldir(void *__buf, const char *name, int namlen, loff_t offset,
 337                      ino_t ino, unsigned int d_type)
 338 {
 339         struct l_linux_dirent *dirent;
 340         struct l_readdir_callback *buf = (struct l_readdir_callback *)__buf;
 341
 342         dirent = buf->lrc_dirent;
 343         if (dirent)
 344                dirent->lld_off = offset;
 345
 346         OBD_ALLOC(dirent, sizeof(*dirent));
 347
 348         list_add_tail(&dirent->lld_list, buf->lrc_list);
 349
 350         buf->lrc_dirent = dirent;
 351         dirent->lld_ino = ino;
 352         LASSERT(sizeof(dirent->lld_name) >= namlen + 1);
 353         memcpy(dirent->lld_name, name, namlen);
 354
 355         return 0;
 356 }
 357
 358 long l_readdir(struct file *file, struct list_head *dentry_list)
 359 {
 360         struct l_linux_dirent *lastdirent;
 361         struct l_readdir_callback buf;
 362         int error;
 363
 364         buf.lrc_dirent = NULL;
 365         buf.lrc_list = dentry_list;
 366
 367         error = vfs_readdir(file, l_filldir, &buf);
 368         if (error < 0)
 369                 return error;
 370
 371         lastdirent = buf.lrc_dirent;
 372         if (lastdirent)
 373                 lastdirent->lld_off = file->f_pos;
 374
 375         return 0;
 376 }
 377 EXPORT_SYMBOL(l_readdir);
 378 EXPORT_SYMBOL(obd_memory);
 379 EXPORT_SYMBOL(obd_memmax);
 380
 381 static int __init lvfs_linux_init(void)
 382 {
 383         RETURN(0);
 384 }
 385
 386 static void __exit lvfs_linux_exit(void)
 387 {
 388         int leaked;
 389         ENTRY;
 390
 391         leaked = atomic_read(&obd_memory);
 392         CDEBUG(leaked ? D_ERROR : D_INFO,
 393                "obd mem max: %d leaked: %d\n", obd_memmax, leaked);
 394
 395         return;
 396 }
 397
 398 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 399 MODULE_DESCRIPTION("Lustre VFS Filesystem Helper v0.1");
 400 MODULE_LICENSE("GPL");
 401
 402 module_init(lvfs_linux_init);
 403 module_exit(lvfs_linux_exit);