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         LASSERT(save->magic != OBD_RUN_CTXT_MAGIC || save->pid != current->pid);
  83         OBD_SET_CTXT_MAGIC(save);
  84         save->pid = current->pid;
  85
  86         /*
  87         CDEBUG(D_INFO,
  88                "= push %p->%p = cur fs %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
  89                save, current, current->fs, current->fs->pwd,
  90                atomic_read(&current->fs->pwd->d_count),
  91                atomic_read(&current->fs->pwd->d_inode->i_count),
  92                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
  93                current->fs->pwdmnt,
  94                atomic_read(&current->fs->pwdmnt->mnt_count));
  95         */
  96
  97         save->fs = get_fs();
  98         LASSERT(atomic_read(&current->fs->pwd->d_count));
  99         LASSERT(atomic_read(&new_ctx->pwd->d_count));
 100         save->pwd = dget(current->fs->pwd);
 101         save->pwdmnt = mntget(current->fs->pwdmnt);
 102         save->ngroups = current_ngroups;
 103
 104         LASSERT(save->pwd);
 105         LASSERT(save->pwdmnt);
 106         LASSERT(new_ctx->pwd);
 107         LASSERT(new_ctx->pwdmnt);
 108
 109         if (uc) {
 110                 save->luc.luc_fsuid = current->fsuid;
 111                 save->luc.luc_fsgid = current->fsgid;
 112                 save->luc.luc_cap = current->cap_effective;
 113                 save->luc.luc_suppgid1 = current_groups[0];
 114                 save->luc.luc_suppgid2 = current_groups[1];
 115
 116                 current->fsuid = uc->luc_fsuid;
 117                 current->fsgid = uc->luc_fsgid;
 118                 current->cap_effective = uc->luc_cap;
 119                 current_ngroups = 0;
 120
 121                 if (uc->luc_suppgid1 != -1)
 122                         current_groups[current_ngroups++] = uc->luc_suppgid1;
 123                 if (uc->luc_suppgid2 != -1)
 124                         current_groups[current_ngroups++] = uc->luc_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 lvfs_run_ctxt *saved, struct lvfs_run_ctxt *new_ctx,
 143               struct lvfs_ucred *uc)
 144 {
 145         //printk("pc0");
 146         ASSERT_CTXT_MAGIC(saved->magic);
 147         LASSERT(saved->pid == current->pid);
 148         saved->magic = 0;
 149         saved->pid = 0;
 150         //printk("pc1");
 151         ASSERT_KERNEL_CTXT("popping non-kernel context!\n");
 152
 153         /*
 154         CDEBUG(D_INFO,
 155                " = pop  %p==%p = cur %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 156                new_ctx, current, current->fs, current->fs->pwd,
 157                atomic_read(&current->fs->pwd->d_count),
 158                atomic_read(&current->fs->pwd->d_inode->i_count),
 159                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 160                current->fs->pwdmnt,
 161                atomic_read(&current->fs->pwdmnt->mnt_count));
 162         */
 163
 164         LASSERT(current->fs->pwd == new_ctx->pwd);
 165         LASSERT(current->fs->pwdmnt == new_ctx->pwdmnt);
 166
 167         set_fs(saved->fs);
 168         set_fs_pwd(current->fs, saved->pwdmnt, saved->pwd);
 169
 170         dput(saved->pwd);
 171         mntput(saved->pwdmnt);
 172         if (uc) {
 173                 current->fsuid = saved->luc.luc_fsuid;
 174                 current->fsgid = saved->luc.luc_fsgid;
 175                 current->cap_effective = saved->luc.luc_cap;
 176                 current_ngroups = saved->ngroups;
 177                 current_groups[0] = saved->luc.luc_suppgid1;
 178                 current_groups[1] = saved->luc.luc_suppgid2;
 179         }
 180
 181         /*
 182         CDEBUG(D_INFO,
 183                "= pop  %p->%p = cur fs %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 184                saved, current, current->fs, current->fs->pwd,
 185                atomic_read(&current->fs->pwd->d_count),
 186                atomic_read(&current->fs->pwd->d_inode->i_count),
 187                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 188                current->fs->pwdmnt,
 189                atomic_read(&current->fs->pwdmnt->mnt_count));
 190         */
 191 }
 192 EXPORT_SYMBOL(pop_ctxt);
 193
 194 /* utility to make a file */
 195 struct dentry *simple_mknod(struct dentry *dir, char *name, int mode, int fix)
 196 {
 197         struct dentry *dchild;
 198         int err = 0;
 199         ENTRY;
 200
 201         ASSERT_KERNEL_CTXT("kernel doing mknod outside kernel context\n");
 202         CDEBUG(D_INODE, "creating file %*s\n", (int)strlen(name), name);
 203
 204         dchild = ll_lookup_one_len(name, dir, strlen(name));
 205         if (IS_ERR(dchild))
 206                 GOTO(out_up, dchild);
 207
 208         if (dchild->d_inode) {
 209                 int old_mode = dchild->d_inode->i_mode;
 210                 if (!S_ISREG(old_mode))
 211                         GOTO(out_err, err = -EEXIST);
 212
 213                 /* Fixup file permissions if necessary */
 214                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 215                         CWARN("fixing permissions on %s from %o to %o\n",
 216                               name, old_mode, mode);
 217                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 218                                                   (old_mode & ~S_IALLUGO);
 219                         mark_inode_dirty(dchild->d_inode);
 220                 }
 221                 GOTO(out_up, dchild);
 222         }
 223
 224         err = ll_vfs_create(dir->d_inode, dchild, (mode & ~S_IFMT) | S_IFREG,
 225                             NULL);
 226         if (err)
 227                 GOTO(out_err, err);
 228
 229         RETURN(dchild);
 230
 231 out_err:
 232         dput(dchild);
 233         dchild = ERR_PTR(err);
 234 out_up:
 235         return dchild;
 236 }
 237 EXPORT_SYMBOL(simple_mknod);
 238
 239 /* utility to make a directory */
 240 struct dentry *simple_mkdir(struct dentry *dir, char *name, int mode, int fix)
 241 {
 242         struct dentry *dchild;
 243         int err = 0;
 244         ENTRY;
 245
 246         ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
 247         CDEBUG(D_INODE, "creating directory %*s\n", (int)strlen(name), name);
 248         dchild = ll_lookup_one_len(name, dir, strlen(name));
 249         if (IS_ERR(dchild))
 250                 GOTO(out_up, dchild);
 251
 252         if (dchild->d_inode) {
 253                 int old_mode = dchild->d_inode->i_mode;
 254                 if (!S_ISDIR(old_mode))
 255                         GOTO(out_err, err = -ENOTDIR);
 256
 257                 /* Fixup directory permissions if necessary */
 258                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 259                         CWARN("fixing permissions on %s from %o to %o\n",
 260                               name, old_mode, mode);
 261                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 262                                                   (old_mode & ~S_IALLUGO);
 263                         mark_inode_dirty(dchild->d_inode);
 264                 }
 265                 GOTO(out_up, dchild);
 266         }
 267
 268         err = vfs_mkdir(dir->d_inode, dchild, mode);
 269         if (err)
 270                 GOTO(out_err, err);
 271
 272         RETURN(dchild);
 273
 274 out_err:
 275         dput(dchild);
 276         dchild = ERR_PTR(err);
 277 out_up:
 278         return dchild;
 279 }
 280 EXPORT_SYMBOL(simple_mkdir);
 281
 282 /*
 283  * Read a file from within kernel context.  Prior to calling this
 284  * function we should already have done a push_ctxt().
 285  */
 286 int lustre_fread(struct file *file, void *buf, int len, loff_t *off)
 287 {
 288         ASSERT_KERNEL_CTXT("kernel doing read outside kernel context\n");
 289         if (!file || !file->f_op || !file->f_op->read || !off)
 290                 RETURN(-ENOSYS);
 291
 292         return file->f_op->read(file, buf, len, off);
 293 }
 294 EXPORT_SYMBOL(lustre_fread);
 295
 296 /*
 297  * Write a file from within kernel context.  Prior to calling this
 298  * function we should already have done a push_ctxt().
 299  */
 300 int lustre_fwrite(struct file *file, const void *buf, int len, loff_t *off)
 301 {
 302         ENTRY;
 303         ASSERT_KERNEL_CTXT("kernel doing write outside kernel context\n");
 304         if (!file)
 305                 RETURN(-ENOENT);
 306         if (!file->f_op)
 307                 RETURN(-ENOSYS);
 308         if (!off)
 309                 RETURN(-EINVAL);
 310
 311         if (!file->f_op->write)
 312                 RETURN(-EROFS);
 313
 314         RETURN(file->f_op->write(file, buf, len, off));
 315 }
 316 EXPORT_SYMBOL(lustre_fwrite);
 317
 318 /*
 319  * Sync a file from within kernel context.  Prior to calling this
 320  * function we should already have done a push_ctxt().
 321  */
 322 int lustre_fsync(struct file *file)
 323 {
 324         ENTRY;
 325         ASSERT_KERNEL_CTXT("kernel doing sync outside kernel context\n");
 326         if (!file || !file->f_op || !file->f_op->fsync)
 327                 RETURN(-ENOSYS);
 328
 329         RETURN(file->f_op->fsync(file, file->f_dentry, 0));
 330 }
 331 EXPORT_SYMBOL(lustre_fsync);
 332
 333 struct l_file *l_dentry_open(struct lvfs_run_ctxt *ctxt, struct l_dentry *de,
 334                              int flags)
 335 {
 336         mntget(ctxt->pwdmnt);
 337         return dentry_open(de, ctxt->pwdmnt, flags);
 338 }
 339 EXPORT_SYMBOL(l_dentry_open);
 340
 341 static int l_filldir(void *__buf, const char *name, int namlen, loff_t offset,
 342                      ino_t ino, unsigned int d_type)
 343 {
 344         struct l_linux_dirent *dirent;
 345         struct l_readdir_callback *buf = (struct l_readdir_callback *)__buf;
 346
 347         dirent = buf->lrc_dirent;
 348         if (dirent)
 349                dirent->lld_off = offset;
 350
 351         OBD_ALLOC(dirent, sizeof(*dirent));
 352
 353         list_add_tail(&dirent->lld_list, buf->lrc_list);
 354
 355         buf->lrc_dirent = dirent;
 356         dirent->lld_ino = ino;
 357         LASSERT(sizeof(dirent->lld_name) >= namlen + 1);
 358         memcpy(dirent->lld_name, name, namlen);
 359
 360         return 0;
 361 }
 362
 363 long l_readdir(struct file *file, struct list_head *dentry_list)
 364 {
 365         struct l_linux_dirent *lastdirent;
 366         struct l_readdir_callback buf;
 367         int error;
 368
 369         buf.lrc_dirent = NULL;
 370         buf.lrc_list = dentry_list;
 371
 372         error = vfs_readdir(file, l_filldir, &buf);
 373         if (error < 0)
 374                 return error;
 375
 376         lastdirent = buf.lrc_dirent;
 377         if (lastdirent)
 378                 lastdirent->lld_off = file->f_pos;
 379
 380         return 0;
 381 }
 382 EXPORT_SYMBOL(l_readdir);
 383 EXPORT_SYMBOL(obd_memory);
 384 EXPORT_SYMBOL(obd_memmax);
 385
 386 static int __init lvfs_linux_init(void)
 387 {
 388         RETURN(0);
 389 }
 390
 391 static void __exit lvfs_linux_exit(void)
 392 {
 393         int leaked;
 394         ENTRY;
 395
 396         leaked = atomic_read(&obd_memory);
 397         CDEBUG(leaked ? D_ERROR : D_INFO,
 398                "obd mem max: %d leaked: %d\n", obd_memmax, leaked);
 399
 400         return;
 401 }
 402
 403 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 404 MODULE_DESCRIPTION("Lustre VFS Filesystem Helper v0.1");
 405 MODULE_LICENSE("GPL");
 406
 407 module_init(lvfs_linux_init);
 408 module_exit(lvfs_linux_exit);