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 #include <linux/lustre_mds.h>   /* for mds_grp_hash_entry */
  53
  54 atomic_t obd_memory;
  55 int obd_memmax;
  56
  57
  58 /* Debugging check only needed during development */
  59 #ifdef OBD_CTXT_DEBUG
  60 # define ASSERT_CTXT_MAGIC(magic) LASSERT((magic) == OBD_RUN_CTXT_MAGIC)
  61 # define ASSERT_NOT_KERNEL_CTXT(msg) LASSERT(!segment_eq(get_fs(), get_ds()))
  62 # define ASSERT_KERNEL_CTXT(msg) LASSERT(segment_eq(get_fs(), get_ds()))
  63 #else
  64 # define ASSERT_CTXT_MAGIC(magic) do {} while(0)
  65 # define ASSERT_NOT_KERNEL_CTXT(msg) do {} while(0)
  66 # define ASSERT_KERNEL_CTXT(msg) do {} while(0)
  67 #endif
  68
  69 static void push_group_info(struct lvfs_run_ctxt *save,
  70                             struct group_info *ginfo)
  71 {
  72         if (!ginfo) {
  73                 save->ngroups = current_ngroups;
  74                 current_ngroups = 0;
  75         } else {
  76 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,4)
  77                 task_lock(current);
  78                 save->group_info = current->group_info;
  79                 current->group_info = group_info;
  80                 task_unlock(current);
  81 #else
  82                 LASSERT(ginfo->ngroups <= NGROUPS);
  83                 /* save old */
  84                 save->group_info.ngroups = current->ngroups;
  85                 if (current->ngroups)
  86                         memcpy(save->group_info.small_block, current->groups,
  87                                current->ngroups);
  88                 /* push new */
  89                 current->ngroups = ginfo->ngroups;
  90                 if (ginfo->ngroups)
  91                         memcpy(current->groups, ginfo->small_block,
  92                                current->ngroups);
  93 #endif
  94         }
  95 }
  96
  97 static void pop_group_info(struct lvfs_run_ctxt *save,
  98                            struct group_info *ginfo)
  99 {
 100         if (!ginfo) {
 101                 current_ngroups = save->ngroups;
 102         } else {
 103 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,4)
 104                 task_lock(current);
 105                 current->group_info = save->group_info;
 106                 task_unlock(current);
 107 #else
 108                 current->ngroups = ginfo->ngroups;
 109                 if (current->ngroups)
 110                         memcpy(current->groups, save->group_info.small_block,
 111                                current->ngroups);
 112 #endif
 113         }
 114 }
 115
 116 /* push / pop to root of obd store */
 117 void push_ctxt(struct lvfs_run_ctxt *save, struct lvfs_run_ctxt *new_ctx,
 118                struct lvfs_ucred *uc)
 119 {
 120         //ASSERT_NOT_KERNEL_CTXT("already in kernel context!\n");
 121         ASSERT_CTXT_MAGIC(new_ctx->magic);
 122         LASSERT(save->magic != OBD_RUN_CTXT_MAGIC || save->pid != current->pid);
 123         OBD_SET_CTXT_MAGIC(save);
 124         save->pid = current->pid;
 125
 126         /*
 127         CDEBUG(D_INFO,
 128                "= push %p->%p = cur fs %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 129                save, current, current->fs, current->fs->pwd,
 130                atomic_read(&current->fs->pwd->d_count),
 131                atomic_read(&current->fs->pwd->d_inode->i_count),
 132                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 133                current->fs->pwdmnt,
 134                atomic_read(&current->fs->pwdmnt->mnt_count));
 135         */
 136
 137         save->fs = get_fs();
 138         LASSERT(atomic_read(&current->fs->pwd->d_count));
 139         LASSERT(atomic_read(&new_ctx->pwd->d_count));
 140         save->pwd = dget(current->fs->pwd);
 141         save->pwdmnt = mntget(current->fs->pwdmnt);
 142         save->ngroups = current_ngroups;
 143
 144         LASSERT(save->pwd);
 145         LASSERT(save->pwdmnt);
 146         LASSERT(new_ctx->pwd);
 147         LASSERT(new_ctx->pwdmnt);
 148
 149         if (uc) {
 150                 save->luc.luc_fsuid = current->fsuid;
 151                 save->luc.luc_fsgid = current->fsgid;
 152                 save->luc.luc_cap = current->cap_effective;
 153
 154                 current->fsuid = uc->luc_fsuid;
 155                 current->fsgid = uc->luc_fsgid;
 156                 current->cap_effective = uc->luc_cap;
 157
 158                 push_group_info(save, uc->luc_ginfo);
 159         }
 160         set_fs(new_ctx->fs);
 161         set_fs_pwd(current->fs, new_ctx->pwdmnt, new_ctx->pwd);
 162
 163         /*
 164         CDEBUG(D_INFO,
 165                "= push %p->%p = cur fs %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 166                new_ctx, current, current->fs, current->fs->pwd,
 167                atomic_read(&current->fs->pwd->d_count),
 168                atomic_read(&current->fs->pwd->d_inode->i_count),
 169                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 170                current->fs->pwdmnt,
 171                atomic_read(&current->fs->pwdmnt->mnt_count));
 172         */
 173 }
 174 EXPORT_SYMBOL(push_ctxt);
 175
 176 void pop_ctxt(struct lvfs_run_ctxt *saved, struct lvfs_run_ctxt *new_ctx,
 177               struct lvfs_ucred *uc)
 178 {
 179         //printk("pc0");
 180         ASSERT_CTXT_MAGIC(saved->magic);
 181         LASSERT(saved->pid == current->pid);
 182         saved->magic = 0;
 183         saved->pid = 0;
 184         //printk("pc1");
 185         ASSERT_KERNEL_CTXT("popping non-kernel context!\n");
 186
 187         /*
 188         CDEBUG(D_INFO,
 189                " = pop  %p==%p = cur %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 190                new_ctx, current, current->fs, current->fs->pwd,
 191                atomic_read(&current->fs->pwd->d_count),
 192                atomic_read(&current->fs->pwd->d_inode->i_count),
 193                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 194                current->fs->pwdmnt,
 195                atomic_read(&current->fs->pwdmnt->mnt_count));
 196         */
 197
 198         LASSERT(current->fs->pwd == new_ctx->pwd);
 199         LASSERT(current->fs->pwdmnt == new_ctx->pwdmnt);
 200
 201         set_fs(saved->fs);
 202         set_fs_pwd(current->fs, saved->pwdmnt, saved->pwd);
 203
 204         dput(saved->pwd);
 205         mntput(saved->pwdmnt);
 206         if (uc) {
 207                 current->fsuid = saved->luc.luc_fsuid;
 208                 current->fsgid = saved->luc.luc_fsgid;
 209                 current->cap_effective = saved->luc.luc_cap;
 210
 211                 pop_group_info(saved, uc->luc_ginfo);
 212         }
 213
 214         /*
 215         CDEBUG(D_INFO,
 216                "= pop  %p->%p = cur fs %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 217                saved, current, current->fs, current->fs->pwd,
 218                atomic_read(&current->fs->pwd->d_count),
 219                atomic_read(&current->fs->pwd->d_inode->i_count),
 220                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 221                current->fs->pwdmnt,
 222                atomic_read(&current->fs->pwdmnt->mnt_count));
 223         */
 224 }
 225 EXPORT_SYMBOL(pop_ctxt);
 226
 227 /* utility to make a file */
 228 struct dentry *simple_mknod(struct dentry *dir, char *name, int mode, int fix)
 229 {
 230         struct dentry *dchild;
 231         int err = 0;
 232         ENTRY;
 233
 234         ASSERT_KERNEL_CTXT("kernel doing mknod outside kernel context\n");
 235         CDEBUG(D_INODE, "creating file %*s\n", (int)strlen(name), name);
 236
 237         dchild = ll_lookup_one_len(name, dir, strlen(name));
 238         if (IS_ERR(dchild))
 239                 GOTO(out_up, dchild);
 240
 241         if (dchild->d_inode) {
 242                 int old_mode = dchild->d_inode->i_mode;
 243                 if (!S_ISREG(old_mode))
 244                         GOTO(out_err, err = -EEXIST);
 245
 246                 /* Fixup file permissions if necessary */
 247                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 248                         CWARN("fixing permissions on %s from %o to %o\n",
 249                               name, old_mode, mode);
 250                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 251                                                   (old_mode & ~S_IALLUGO);
 252                         mark_inode_dirty(dchild->d_inode);
 253                 }
 254                 GOTO(out_up, dchild);
 255         }
 256
 257         err = ll_vfs_create(dir->d_inode, dchild, (mode & ~S_IFMT) | S_IFREG,
 258                             NULL);
 259         if (err)
 260                 GOTO(out_err, err);
 261
 262         RETURN(dchild);
 263
 264 out_err:
 265         dput(dchild);
 266         dchild = ERR_PTR(err);
 267 out_up:
 268         return dchild;
 269 }
 270 EXPORT_SYMBOL(simple_mknod);
 271
 272 /* utility to make a directory */
 273 struct dentry *simple_mkdir(struct dentry *dir, char *name, int mode, int fix)
 274 {
 275         struct dentry *dchild;
 276         int err = 0;
 277         ENTRY;
 278
 279         ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
 280         CDEBUG(D_INODE, "creating directory %*s\n", (int)strlen(name), name);
 281         dchild = ll_lookup_one_len(name, dir, strlen(name));
 282         if (IS_ERR(dchild))
 283                 GOTO(out_up, dchild);
 284
 285         if (dchild->d_inode) {
 286                 int old_mode = dchild->d_inode->i_mode;
 287                 if (!S_ISDIR(old_mode))
 288                         GOTO(out_err, err = -ENOTDIR);
 289
 290                 /* Fixup directory permissions if necessary */
 291                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 292                         CWARN("fixing permissions on %s from %o to %o\n",
 293                               name, old_mode, mode);
 294                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 295                                                   (old_mode & ~S_IALLUGO);
 296                         mark_inode_dirty(dchild->d_inode);
 297                 }
 298                 GOTO(out_up, dchild);
 299         }
 300
 301         err = vfs_mkdir(dir->d_inode, dchild, mode);
 302         if (err)
 303                 GOTO(out_err, err);
 304
 305         RETURN(dchild);
 306
 307 out_err:
 308         dput(dchild);
 309         dchild = ERR_PTR(err);
 310 out_up:
 311         return dchild;
 312 }
 313 EXPORT_SYMBOL(simple_mkdir);
 314
 315 /*
 316  * Read a file from within kernel context.  Prior to calling this
 317  * function we should already have done a push_ctxt().
 318  */
 319 int lustre_fread(struct file *file, void *buf, int len, loff_t *off)
 320 {
 321         ASSERT_KERNEL_CTXT("kernel doing read outside kernel context\n");
 322         if (!file || !file->f_op || !file->f_op->read || !off)
 323                 RETURN(-ENOSYS);
 324
 325         return file->f_op->read(file, buf, len, off);
 326 }
 327 EXPORT_SYMBOL(lustre_fread);
 328
 329 /*
 330  * Write a file from within kernel context.  Prior to calling this
 331  * function we should already have done a push_ctxt().
 332  */
 333 int lustre_fwrite(struct file *file, const void *buf, int len, loff_t *off)
 334 {
 335         ENTRY;
 336         ASSERT_KERNEL_CTXT("kernel doing write outside kernel context\n");
 337         if (!file)
 338                 RETURN(-ENOENT);
 339         if (!file->f_op)
 340                 RETURN(-ENOSYS);
 341         if (!off)
 342                 RETURN(-EINVAL);
 343
 344         if (!file->f_op->write)
 345                 RETURN(-EROFS);
 346
 347         RETURN(file->f_op->write(file, buf, len, off));
 348 }
 349 EXPORT_SYMBOL(lustre_fwrite);
 350
 351 /*
 352  * Sync a file from within kernel context.  Prior to calling this
 353  * function we should already have done a push_ctxt().
 354  */
 355 int lustre_fsync(struct file *file)
 356 {
 357         ENTRY;
 358         ASSERT_KERNEL_CTXT("kernel doing sync outside kernel context\n");
 359         if (!file || !file->f_op || !file->f_op->fsync)
 360                 RETURN(-ENOSYS);
 361
 362         RETURN(file->f_op->fsync(file, file->f_dentry, 0));
 363 }
 364 EXPORT_SYMBOL(lustre_fsync);
 365
 366 struct l_file *l_dentry_open(struct lvfs_run_ctxt *ctxt, struct l_dentry *de,
 367                              int flags)
 368 {
 369         mntget(ctxt->pwdmnt);
 370         return dentry_open(de, ctxt->pwdmnt, flags);
 371 }
 372 EXPORT_SYMBOL(l_dentry_open);
 373
 374 static int l_filldir(void *__buf, const char *name, int namlen, loff_t offset,
 375                      ino_t ino, unsigned int d_type)
 376 {
 377         struct l_linux_dirent *dirent;
 378         struct l_readdir_callback *buf = (struct l_readdir_callback *)__buf;
 379
 380         dirent = buf->lrc_dirent;
 381         if (dirent)
 382                dirent->lld_off = offset;
 383
 384         OBD_ALLOC(dirent, sizeof(*dirent));
 385
 386         list_add_tail(&dirent->lld_list, buf->lrc_list);
 387
 388         buf->lrc_dirent = dirent;
 389         dirent->lld_ino = ino;
 390         LASSERT(sizeof(dirent->lld_name) >= namlen + 1);
 391         memcpy(dirent->lld_name, name, namlen);
 392
 393         return 0;
 394 }
 395
 396 long l_readdir(struct file *file, struct list_head *dentry_list)
 397 {
 398         struct l_linux_dirent *lastdirent;
 399         struct l_readdir_callback buf;
 400         int error;
 401
 402         buf.lrc_dirent = NULL;
 403         buf.lrc_list = dentry_list;
 404
 405         error = vfs_readdir(file, l_filldir, &buf);
 406         if (error < 0)
 407                 return error;
 408
 409         lastdirent = buf.lrc_dirent;
 410         if (lastdirent)
 411                 lastdirent->lld_off = file->f_pos;
 412
 413         return 0;
 414 }
 415 EXPORT_SYMBOL(l_readdir);
 416 EXPORT_SYMBOL(obd_memory);
 417 EXPORT_SYMBOL(obd_memmax);
 418
 419 static int __init lvfs_linux_init(void)
 420 {
 421         RETURN(0);
 422 }
 423
 424 static void __exit lvfs_linux_exit(void)
 425 {
 426         int leaked;
 427         ENTRY;
 428
 429         leaked = atomic_read(&obd_memory);
 430         CDEBUG(leaked ? D_ERROR : D_INFO,
 431                "obd mem max: %d leaked: %d\n", obd_memmax, leaked);
 432
 433         return;
 434 }
 435
 436 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 437 MODULE_DESCRIPTION("Lustre VFS Filesystem Helper v0.1");
 438 MODULE_LICENSE("GPL");
 439
 440 module_init(lvfs_linux_init);
 441 module_exit(lvfs_linux_exit);