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 obd_run_ctxt *save, struct obd_run_ctxt *new_ctx,
  78                struct obd_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->ouc.ouc_fsuid = current->fsuid;
 109                 save->ouc.ouc_fsgid = current->fsgid;
 110                 save->ouc.ouc_cap = current->cap_effective;
 111                 save->ouc.ouc_suppgid1 = current_groups[0];
 112                 save->ouc.ouc_suppgid2 = current_groups[1];
 113                 save->ouc.ouc_umask = current->fs->umask;
 114
 115                 current->fsuid = uc->ouc_fsuid;
 116                 current->fsgid = uc->ouc_fsgid;
 117                 current->cap_effective = uc->ouc_cap;
 118                 current_ngroups = 0;
 119                 current->fs->umask = 0; /* umask already applied on client */
 120
 121                 if (uc->ouc_suppgid1 != -1)
 122                         current_groups[current_ngroups++] = uc->ouc_suppgid1;
 123                 if (uc->ouc_suppgid2 != -1)
 124                         current_groups[current_ngroups++] = uc->ouc_suppgid2;
 125         }
 126         set_fs(new_ctx->fs);
 127         set_fs_pwd(current->fs, new_ctx->pwdmnt, new_ctx->pwd);
 128
 129         /*
 130         CDEBUG(D_INFO,
 131                "= push %p->%p = cur fs %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 132                new_ctx, current, current->fs, current->fs->pwd,
 133                atomic_read(&current->fs->pwd->d_count),
 134                atomic_read(&current->fs->pwd->d_inode->i_count),
 135                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 136                current->fs->pwdmnt,
 137                atomic_read(&current->fs->pwdmnt->mnt_count));
 138         */
 139 }
 140 EXPORT_SYMBOL(push_ctxt);
 141
 142 void pop_ctxt(struct obd_run_ctxt *saved, struct obd_run_ctxt *new_ctx,
 143               struct obd_ucred *uc)
 144 {
 145         //printk("pc0");
 146         ASSERT_CTXT_MAGIC(saved->magic);
 147         //printk("pc1");
 148         ASSERT_KERNEL_CTXT("popping non-kernel context!\n");
 149
 150         /*
 151         CDEBUG(D_INFO,
 152                " = pop  %p==%p = cur %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 153                new_ctx, current, current->fs, current->fs->pwd,
 154                atomic_read(&current->fs->pwd->d_count),
 155                atomic_read(&current->fs->pwd->d_inode->i_count),
 156                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 157                current->fs->pwdmnt,
 158                atomic_read(&current->fs->pwdmnt->mnt_count));
 159         */
 160
 161         LASSERT(current->fs->pwd == new_ctx->pwd);
 162         LASSERT(current->fs->pwdmnt == new_ctx->pwdmnt);
 163
 164         set_fs(saved->fs);
 165         set_fs_pwd(current->fs, saved->pwdmnt, saved->pwd);
 166
 167         dput(saved->pwd);
 168         mntput(saved->pwdmnt);
 169         if (uc) {
 170                 current->fsuid = saved->ouc.ouc_fsuid;
 171                 current->fsgid = saved->ouc.ouc_fsgid;
 172                 current->cap_effective = saved->ouc.ouc_cap;
 173                 current_ngroups = saved->ngroups;
 174                 current_groups[0] = saved->ouc.ouc_suppgid1;
 175                 current_groups[1] = saved->ouc.ouc_suppgid2;
 176                 current->fs->umask = saved->ouc.ouc_umask;
 177         }
 178
 179         /*
 180         CDEBUG(D_INFO,
 181                "= pop  %p->%p = cur fs %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 182                saved, current, current->fs, current->fs->pwd,
 183                atomic_read(&current->fs->pwd->d_count),
 184                atomic_read(&current->fs->pwd->d_inode->i_count),
 185                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 186                current->fs->pwdmnt,
 187                atomic_read(&current->fs->pwdmnt->mnt_count));
 188         */
 189 }
 190 EXPORT_SYMBOL(pop_ctxt);
 191
 192 /* utility to make a file */
 193 struct dentry *simple_mknod(struct dentry *dir, char *name, int mode, int fix)
 194 {
 195         struct dentry *dchild;
 196         int err = 0;
 197         ENTRY;
 198
 199         ASSERT_KERNEL_CTXT("kernel doing mknod outside kernel context\n");
 200         CDEBUG(D_INODE, "creating file %*s\n", (int)strlen(name), name);
 201
 202         dchild = ll_lookup_one_len(name, dir, strlen(name));
 203         if (IS_ERR(dchild))
 204                 GOTO(out_up, dchild);
 205
 206         if (dchild->d_inode) {
 207                 int old_mode = dchild->d_inode->i_mode;
 208                 if (!S_ISREG(old_mode))
 209                         GOTO(out_err, err = -EEXIST);
 210
 211                 /* Fixup file permissions if necessary */
 212                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 213                         CWARN("fixing permissions on %s from %o to %o\n",
 214                               name, old_mode, mode);
 215                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 216                                                   (old_mode & ~S_IALLUGO);
 217                         mark_inode_dirty(dchild->d_inode);
 218                 }
 219                 GOTO(out_up, dchild);
 220         }
 221
 222         err = ll_vfs_create(dir->d_inode, dchild, (mode & ~S_IFMT) | S_IFREG,
 223                             NULL);
 224         if (err)
 225                 GOTO(out_err, err);
 226
 227         RETURN(dchild);
 228
 229 out_err:
 230         dput(dchild);
 231         dchild = ERR_PTR(err);
 232 out_up:
 233         return dchild;
 234 }
 235 EXPORT_SYMBOL(simple_mknod);
 236
 237 /* utility to make a directory */
 238 struct dentry *simple_mkdir(struct dentry *dir, char *name, int mode, int fix)
 239 {
 240         struct dentry *dchild;
 241         int err = 0;
 242         ENTRY;
 243
 244         ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
 245         CDEBUG(D_INODE, "creating directory %*s\n", (int)strlen(name), name);
 246         dchild = ll_lookup_one_len(name, dir, strlen(name));
 247         if (IS_ERR(dchild))
 248                 GOTO(out_up, dchild);
 249
 250         if (dchild->d_inode) {
 251                 int old_mode = dchild->d_inode->i_mode;
 252                 if (!S_ISDIR(old_mode))
 253                         GOTO(out_err, err = -ENOTDIR);
 254
 255                 /* Fixup directory permissions if necessary */
 256                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 257                         CWARN("fixing permissions on %s from %o to %o\n",
 258                               name, old_mode, mode);
 259                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 260                                                   (old_mode & ~S_IALLUGO);
 261                         mark_inode_dirty(dchild->d_inode);
 262                 }
 263                 GOTO(out_up, dchild);
 264         }
 265
 266         err = vfs_mkdir(dir->d_inode, dchild, mode);
 267         if (err)
 268                 GOTO(out_err, err);
 269
 270         RETURN(dchild);
 271
 272 out_err:
 273         dput(dchild);
 274         dchild = ERR_PTR(err);
 275 out_up:
 276         return dchild;
 277 }
 278 EXPORT_SYMBOL(simple_mkdir);
 279
 280 /*
 281  * Read a file from within kernel context.  Prior to calling this
 282  * function we should already have done a push_ctxt().
 283  */
 284 int lustre_fread(struct file *file, void *buf, int len, loff_t *off)
 285 {
 286         ASSERT_KERNEL_CTXT("kernel doing read outside kernel context\n");
 287         if (!file || !file->f_op || !file->f_op->read || !off)
 288                 RETURN(-ENOSYS);
 289
 290         return file->f_op->read(file, buf, len, off);
 291 }
 292 EXPORT_SYMBOL(lustre_fread);
 293
 294 /*
 295  * Write a file from within kernel context.  Prior to calling this
 296  * function we should already have done a push_ctxt().
 297  */
 298 int lustre_fwrite(struct file *file, const void *buf, int len, loff_t *off)
 299 {
 300         ENTRY;
 301         ASSERT_KERNEL_CTXT("kernel doing write outside kernel context\n");
 302         if (!file)
 303                 RETURN(-ENOENT);
 304         if (!file->f_op)
 305                 RETURN(-ENOSYS);
 306         if (!off)
 307                 RETURN(-EINVAL);
 308
 309         if (!file->f_op->write)
 310                 RETURN(-EROFS);
 311
 312         RETURN(file->f_op->write(file, buf, len, off));
 313 }
 314 EXPORT_SYMBOL(lustre_fwrite);
 315
 316 /*
 317  * Sync a file from within kernel context.  Prior to calling this
 318  * function we should already have done a push_ctxt().
 319  */
 320 int lustre_fsync(struct file *file)
 321 {
 322         ENTRY;
 323         ASSERT_KERNEL_CTXT("kernel doing sync outside kernel context\n");
 324         if (!file || !file->f_op || !file->f_op->fsync)
 325                 RETURN(-ENOSYS);
 326
 327         RETURN(file->f_op->fsync(file, file->f_dentry, 0));
 328 }
 329 EXPORT_SYMBOL(lustre_fsync);
 330
 331 struct l_file *l_dentry_open(struct obd_run_ctxt *ctxt, struct l_dentry *de,
 332                              int flags)
 333 {
 334         mntget(ctxt->pwdmnt);
 335         return dentry_open(de, ctxt->pwdmnt, flags);
 336 }
 337 EXPORT_SYMBOL(l_dentry_open);
 338
 339 static int l_filldir(void *__buf, const char *name, int namlen, loff_t offset,
 340                      ino_t ino, unsigned int d_type)
 341 {
 342         struct l_linux_dirent *dirent;
 343         struct l_readdir_callback *buf = (struct l_readdir_callback *)__buf;
 344
 345         dirent = buf->lrc_dirent;
 346         if (dirent)
 347                dirent->lld_off = offset;
 348
 349         OBD_ALLOC(dirent, sizeof(*dirent));
 350
 351         list_add_tail(&dirent->lld_list, buf->lrc_list);
 352
 353         buf->lrc_dirent = dirent;
 354         dirent->lld_ino = ino;
 355         LASSERT(sizeof(dirent->lld_name) >= namlen + 1);
 356         memcpy(dirent->lld_name, name, namlen);
 357
 358         return 0;
 359 }
 360
 361 long l_readdir(struct file *file, struct list_head *dentry_list)
 362 {
 363         struct l_linux_dirent *lastdirent;
 364         struct l_readdir_callback buf;
 365         int error;
 366
 367         buf.lrc_dirent = NULL;
 368         buf.lrc_list = dentry_list;
 369
 370         error = vfs_readdir(file, l_filldir, &buf);
 371         if (error < 0)
 372                 return error;
 373
 374         lastdirent = buf.lrc_dirent;
 375         if (lastdirent)
 376                 lastdirent->lld_off = file->f_pos;
 377
 378         return 0;
 379 }
 380 EXPORT_SYMBOL(l_readdir);
 381 EXPORT_SYMBOL(obd_memory);
 382 EXPORT_SYMBOL(obd_memmax);
 383
 384 static int __init lvfs_linux_init(void)
 385 {
 386         RETURN(0);
 387 }
 388
 389 static void __exit lvfs_linux_exit(void)
 390 {
 391         int leaked;
 392         ENTRY;
 393
 394         leaked = atomic_read(&obd_memory);
 395         CDEBUG(leaked ? D_ERROR : D_INFO,
 396                "obd mem max: %d leaked: %d\n", obd_memmax, leaked);
 397
 398         return;
 399 }
 400
 401 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 402 MODULE_DESCRIPTION("Lustre VFS Filesystem Helper v0.1");
 403 MODULE_LICENSE("GPL");
 404
 405 module_init(lvfs_linux_init);
 406 module_exit(lvfs_linux_exit);