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/lvfs_linux.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/slab.h>
  36 #include <linux/pagemap.h>
  37 #include <linux/quotaops.h>
  38 #include <linux/version.h>
  39 #include <libcfs/kp30.h>
  40 #include <linux/lustre_fsfilt.h>
  41 #include <linux/obd.h>
  42 #include <linux/obd_class.h>
  43 #include <linux/module.h>
  44 #include <linux/init.h>
  45 #include <linux/lustre_compat25.h>
  46 #include <linux/lvfs.h>
  47 #include "lvfs_internal.h"
  48
  49 #include <linux/obd.h>
  50 #include <linux/lustre_lib.h>
  51
  52 atomic_t obd_memory;
  53 int obd_memmax;
  54
  55
  56 /* Debugging check only needed during development */
  57 #ifdef OBD_CTXT_DEBUG
  58 # define ASSERT_CTXT_MAGIC(magic) LASSERT((magic) == OBD_RUN_CTXT_MAGIC)
  59 # define ASSERT_NOT_KERNEL_CTXT(msg) LASSERT(!segment_eq(get_fs(), get_ds()))
  60 # define ASSERT_KERNEL_CTXT(msg) LASSERT(segment_eq(get_fs(), get_ds()))
  61 #else
  62 # define ASSERT_CTXT_MAGIC(magic) do {} while(0)
  63 # define ASSERT_NOT_KERNEL_CTXT(msg) do {} while(0)
  64 # define ASSERT_KERNEL_CTXT(msg) do {} while(0)
  65 #endif
  66
  67 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,4))
  68 #define current_ngroups current->group_info->ngroups
  69 #define current_groups current->group_info->small_block
  70 #else
  71 #define current_ngroups current->ngroups
  72 #define current_groups current->groups
  73 #endif
  74
  75 /* push / pop to root of obd store */
  76 void push_ctxt(struct obd_run_ctxt *save, struct obd_run_ctxt *new_ctx,
  77                struct obd_ucred *uc)
  78 {
  79         //ASSERT_NOT_KERNEL_CTXT("already in kernel context!\n");
  80         ASSERT_CTXT_MAGIC(new_ctx->magic);
  81         OBD_SET_CTXT_MAGIC(save);
  82
  83         /*
  84         CDEBUG(D_INFO,
  85                "= push %p->%p = cur fs %p pwd %p:d%d:i%d (%.*s), pwdmnt %p:%d\n",
  86                save, current, current->fs, current->fs->pwd,
  87                atomic_read(&current->fs->pwd->d_count),
  88                atomic_read(&current->fs->pwd->d_inode->i_count),
  89                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
  90                current->fs->pwdmnt,
  91                atomic_read(&current->fs->pwdmnt->mnt_count));
  92         */
  93
  94         save->fs = get_fs();
  95         LASSERT(atomic_read(&current->fs->pwd->d_count));
  96         LASSERT(atomic_read(&new_ctx->pwd->d_count));
  97         save->pwd = dget(current->fs->pwd);
  98         save->pwdmnt = mntget(current->fs->pwdmnt);
  99         save->ngroups = current_ngroups;
 100         save->ouc.ouc_umask = current->fs->umask;
 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
 114                 current->fsuid = uc->ouc_fsuid;
 115                 current->fsgid = uc->ouc_fsgid;
 116                 current->cap_effective = uc->ouc_cap;
 117                 current_ngroups = 0;
 118
 119                 if (uc->ouc_suppgid1 != -1)
 120                         current_groups[current_ngroups++] = uc->ouc_suppgid1;
 121                 if (uc->ouc_suppgid2 != -1)
 122                         current_groups[current_ngroups++] = uc->ouc_suppgid2;
 123 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,4)
 124                 if (uc->ouc_suppgid1 != -1 && uc->ouc_suppgid2 != -1 &&
 125                     (uc->ouc_suppgid1 > uc->ouc_suppgid2)) {
 126                         current_groups[0] = uc->ouc_suppgid2;
 127                         current_groups[1] = uc->ouc_suppgid1;
 128                 }
 129 #endif
 130         }
 131         current->fs->umask = 0; /* umask already applied on client */
 132         set_fs(new_ctx->fs);
 133         set_fs_pwd(current->fs, new_ctx->pwdmnt, new_ctx->pwd);
 134
 135         /*
 136         CDEBUG(D_INFO,
 137                "= push %p->%p = cur fs %p pwd %p:d%d:i%d (%.*s), pwdmnt %p:%d\n",
 138                new_ctx, current, current->fs, current->fs->pwd,
 139                atomic_read(&current->fs->pwd->d_count),
 140                atomic_read(&current->fs->pwd->d_inode->i_count),
 141                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 142                current->fs->pwdmnt,
 143                atomic_read(&current->fs->pwdmnt->mnt_count));
 144         */
 145 }
 146 EXPORT_SYMBOL(push_ctxt);
 147
 148 void pop_ctxt(struct obd_run_ctxt *saved, struct obd_run_ctxt *new_ctx,
 149               struct obd_ucred *uc)
 150 {
 151         //printk("pc0");
 152         ASSERT_CTXT_MAGIC(saved->magic);
 153         //printk("pc1");
 154         ASSERT_KERNEL_CTXT("popping non-kernel context!\n");
 155
 156         /*
 157         CDEBUG(D_INFO,
 158                " = pop  %p==%p = cur %p pwd %p:d%d:i%d (%.*s), pwdmnt %p:%d\n",
 159                new_ctx, current, current->fs, current->fs->pwd,
 160                atomic_read(&current->fs->pwd->d_count),
 161                atomic_read(&current->fs->pwd->d_inode->i_count),
 162                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 163                current->fs->pwdmnt,
 164                atomic_read(&current->fs->pwdmnt->mnt_count));
 165         */
 166
 167         LASSERT(current->fs->pwd == new_ctx->pwd);
 168         LASSERT(current->fs->pwdmnt == new_ctx->pwdmnt);
 169
 170         set_fs(saved->fs);
 171         set_fs_pwd(current->fs, saved->pwdmnt, saved->pwd);
 172
 173         dput(saved->pwd);
 174         mntput(saved->pwdmnt);
 175         current->fs->umask = saved->ouc.ouc_umask;
 176         if (uc) {
 177                 current->fsuid = saved->ouc.ouc_fsuid;
 178                 current->fsgid = saved->ouc.ouc_fsgid;
 179                 current->cap_effective = saved->ouc.ouc_cap;
 180                 current_ngroups = saved->ngroups;
 181                 current_groups[0] = saved->ouc.ouc_suppgid1;
 182                 current_groups[1] = saved->ouc.ouc_suppgid2;
 183         }
 184
 185         /*
 186         CDEBUG(D_INFO,
 187                "= pop  %p->%p = cur fs %p pwd %p:d%d:i%d (%.*s), pwdmnt %p:%d\n",
 188                saved, current, current->fs, current->fs->pwd,
 189                atomic_read(&current->fs->pwd->d_count),
 190                atomic_read(&current->fs->pwd->d_inode->i_count),
 191                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 192                current->fs->pwdmnt,
 193                atomic_read(&current->fs->pwdmnt->mnt_count));
 194         */
 195 }
 196 EXPORT_SYMBOL(pop_ctxt);
 197
 198 /* utility to make a file */
 199 struct dentry *simple_mknod(struct dentry *dir, char *name, int mode, int fix)
 200 {
 201         struct dentry *dchild;
 202         int err = 0;
 203         ENTRY;
 204
 205         ASSERT_KERNEL_CTXT("kernel doing mknod outside kernel context\n");
 206         CDEBUG(D_INODE, "creating file %.*s\n", (int)strlen(name), name);
 207
 208         dchild = ll_lookup_one_len(name, dir, strlen(name));
 209         if (IS_ERR(dchild))
 210                 GOTO(out_up, dchild);
 211
 212         if (dchild->d_inode) {
 213                 int old_mode = dchild->d_inode->i_mode;
 214                 if (!S_ISREG(old_mode))
 215                         GOTO(out_err, err = -EEXIST);
 216
 217                 /* Fixup file permissions if necessary */
 218                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 219                         CWARN("fixing permissions on %s from %o to %o\n",
 220                               name, old_mode, mode);
 221                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 222                                                   (old_mode & ~S_IALLUGO);
 223                         mark_inode_dirty(dchild->d_inode);
 224                 }
 225                 GOTO(out_up, dchild);
 226         }
 227
 228         err = ll_vfs_create(dir->d_inode, dchild, (mode & ~S_IFMT) | S_IFREG,
 229                             NULL);
 230         if (err)
 231                 GOTO(out_err, err);
 232
 233         RETURN(dchild);
 234
 235 out_err:
 236         dput(dchild);
 237         dchild = ERR_PTR(err);
 238 out_up:
 239         return dchild;
 240 }
 241 EXPORT_SYMBOL(simple_mknod);
 242
 243 /* utility to make a directory */
 244 struct dentry *simple_mkdir(struct dentry *dir, char *name, int mode, int fix)
 245 {
 246         struct dentry *dchild;
 247         int err = 0;
 248         ENTRY;
 249
 250         ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
 251         CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
 252         dchild = ll_lookup_one_len(name, dir, strlen(name));
 253         if (IS_ERR(dchild))
 254                 GOTO(out_up, dchild);
 255
 256         if (dchild->d_inode) {
 257                 int old_mode = dchild->d_inode->i_mode;
 258                 if (!S_ISDIR(old_mode)) {
 259                         CERROR("found %s (%lu/%u) is mode %o\n", name,
 260                                dchild->d_inode->i_ino,
 261                                dchild->d_inode->i_generation, old_mode);
 262                         GOTO(out_err, err = -ENOTDIR);
 263                 }
 264
 265                 /* Fixup directory permissions if necessary */
 266                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 267                         CWARN("fixing permissions on %s from %o to %o\n",
 268                               name, old_mode, mode);
 269                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 270                                                   (old_mode & ~S_IALLUGO);
 271                         mark_inode_dirty(dchild->d_inode);
 272                 }
 273                 GOTO(out_up, dchild);
 274         }
 275
 276         err = vfs_mkdir(dir->d_inode, dchild, mode);
 277         if (err)
 278                 GOTO(out_err, err);
 279
 280         RETURN(dchild);
 281
 282 out_err:
 283         dput(dchild);
 284         dchild = ERR_PTR(err);
 285 out_up:
 286         return dchild;
 287 }
 288 EXPORT_SYMBOL(simple_mkdir);
 289
 290 /*
 291  * Read a file from within kernel context.  Prior to calling this
 292  * function we should already have done a push_ctxt().
 293  */
 294 int lustre_fread(struct file *file, void *buf, int len, loff_t *off)
 295 {
 296         ASSERT_KERNEL_CTXT("kernel doing read outside kernel context\n");
 297         if (!file || !file->f_op || !file->f_op->read || !off)
 298                 RETURN(-ENOSYS);
 299
 300         return file->f_op->read(file, buf, len, off);
 301 }
 302 EXPORT_SYMBOL(lustre_fread);
 303
 304 /*
 305  * Write a file from within kernel context.  Prior to calling this
 306  * function we should already have done a push_ctxt().
 307  */
 308 int lustre_fwrite(struct file *file, const void *buf, int len, loff_t *off)
 309 {
 310         ENTRY;
 311         ASSERT_KERNEL_CTXT("kernel doing write outside kernel context\n");
 312         if (!file)
 313                 RETURN(-ENOENT);
 314         if (!file->f_op)
 315                 RETURN(-ENOSYS);
 316         if (!off)
 317                 RETURN(-EINVAL);
 318
 319         if (!file->f_op->write)
 320                 RETURN(-EROFS);
 321
 322         RETURN(file->f_op->write(file, buf, len, off));
 323 }
 324 EXPORT_SYMBOL(lustre_fwrite);
 325
 326 /*
 327  * Sync a file from within kernel context.  Prior to calling this
 328  * function we should already have done a push_ctxt().
 329  */
 330 int lustre_fsync(struct file *file)
 331 {
 332         ENTRY;
 333         ASSERT_KERNEL_CTXT("kernel doing sync outside kernel context\n");
 334         if (!file || !file->f_op || !file->f_op->fsync)
 335                 RETURN(-ENOSYS);
 336
 337         RETURN(file->f_op->fsync(file, file->f_dentry, 0));
 338 }
 339 EXPORT_SYMBOL(lustre_fsync);
 340
 341 struct l_file *l_dentry_open(struct obd_run_ctxt *ctxt, struct l_dentry *de,
 342                              int flags)
 343 {
 344         mntget(ctxt->pwdmnt);
 345         return dentry_open(de, ctxt->pwdmnt, flags);
 346 }
 347 EXPORT_SYMBOL(l_dentry_open);
 348
 349 static int l_filldir(void *__buf, const char *name, int namlen, loff_t offset,
 350                      ino_t ino, unsigned int d_type)
 351 {
 352         struct l_linux_dirent *dirent;
 353         struct l_readdir_callback *buf = (struct l_readdir_callback *)__buf;
 354
 355         dirent = buf->lrc_dirent;
 356         if (dirent)
 357                dirent->lld_off = offset;
 358
 359         OBD_ALLOC(dirent, sizeof(*dirent));
 360
 361         list_add_tail(&dirent->lld_list, buf->lrc_list);
 362
 363         buf->lrc_dirent = dirent;
 364         dirent->lld_ino = ino;
 365         LASSERT(sizeof(dirent->lld_name) >= namlen + 1);
 366         memcpy(dirent->lld_name, name, namlen);
 367
 368         return 0;
 369 }
 370
 371 long l_readdir(struct file *file, struct list_head *dentry_list)
 372 {
 373         struct l_linux_dirent *lastdirent;
 374         struct l_readdir_callback buf;
 375         int error;
 376
 377         buf.lrc_dirent = NULL;
 378         buf.lrc_list = dentry_list;
 379
 380         error = vfs_readdir(file, l_filldir, &buf);
 381         if (error < 0)
 382                 return error;
 383
 384         lastdirent = buf.lrc_dirent;
 385         if (lastdirent)
 386                 lastdirent->lld_off = file->f_pos;
 387
 388         return 0;
 389 }
 390 EXPORT_SYMBOL(l_readdir);
 391 EXPORT_SYMBOL(obd_memory);
 392 EXPORT_SYMBOL(obd_memmax);
 393
 394 static int __init lvfs_linux_init(void)
 395 {
 396         RETURN(0);
 397 }
 398
 399 static void __exit lvfs_linux_exit(void)
 400 {
 401         int leaked;
 402         ENTRY;
 403
 404         leaked = atomic_read(&obd_memory);
 405         CDEBUG(leaked ? D_ERROR : D_INFO,
 406                "obd mem max: %d leaked: %d\n", obd_memmax, leaked);
 407
 408         return;
 409 }
 410
 411 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 412 MODULE_DESCRIPTION("Lustre VFS Filesystem Helper v0.1");
 413 MODULE_LICENSE("GPL");
 414
 415 module_init(lvfs_linux_init);
 416 module_exit(lvfs_linux_exit);