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 <linux/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 #include <linux/lustre_mds.h>   /* for mds_grp_hash_entry */
  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) LASSERTF(!segment_eq(get_fs(), get_ds()),\
  61                                               msg)
  62 # define ASSERT_KERNEL_CTXT(msg) LASSERTF(segment_eq(get_fs(), get_ds()), msg)
  63
  64 #else
  65 # define ASSERT_CTXT_MAGIC(magic) do {} while(0)
  66 # define ASSERT_NOT_KERNEL_CTXT(msg) do {} while(0)
  67 # define ASSERT_KERNEL_CTXT(msg) do {} while(0)
  68 #endif
  69
  70 static void push_group_info(struct lvfs_run_ctxt *save,
  71                             struct group_info *ginfo)
  72 {
  73         if (!ginfo) {
  74                 save->ngroups = current_ngroups;
  75                 current_ngroups = 0;
  76         } else {
  77 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,4)
  78                 task_lock(current);
  79                 save->group_info = current->group_info;
  80                 current->group_info = ginfo;
  81                 task_unlock(current);
  82 #else
  83                 LASSERT(ginfo->ngroups <= NGROUPS);
  84                 /* save old */
  85                 save->group_info.ngroups = current->ngroups;
  86                 if (current->ngroups)
  87                         memcpy(save->group_info.small_block, current->groups,
  88                                current->ngroups);
  89                 /* push new */
  90                 current->ngroups = ginfo->ngroups;
  91                 if (ginfo->ngroups)
  92                         memcpy(current->groups, ginfo->small_block,
  93                                current->ngroups);
  94 #endif
  95         }
  96 }
  97
  98 static void pop_group_info(struct lvfs_run_ctxt *save,
  99                            struct group_info *ginfo)
 100 {
 101         if (!ginfo) {
 102                 current_ngroups = save->ngroups;
 103         } else {
 104 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,4)
 105                 task_lock(current);
 106                 current->group_info = save->group_info;
 107                 task_unlock(current);
 108 #else
 109                 current->ngroups = ginfo->ngroups;
 110                 if (current->ngroups)
 111                         memcpy(current->groups, save->group_info.small_block,
 112                                current->ngroups);
 113 #endif
 114         }
 115 }
 116
 117 /* push / pop to root of obd store */
 118 void push_ctxt(struct lvfs_run_ctxt *save, struct lvfs_run_ctxt *new_ctx,
 119                struct lvfs_ucred *uc)
 120 {
 121         //ASSERT_NOT_KERNEL_CTXT("already in kernel context!\n");
 122         ASSERT_CTXT_MAGIC(new_ctx->magic);
 123         LASSERT(save->magic != OBD_RUN_CTXT_MAGIC || save->pid != current->pid);
 124         OBD_SET_CTXT_MAGIC(save);
 125         save->pid = current->pid;
 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                save, 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         save->fs = get_fs();
 139         LASSERT(atomic_read(&current->fs->pwd->d_count));
 140         LASSERT(atomic_read(&new_ctx->pwd->d_count));
 141         save->pwd = dget(current->fs->pwd);
 142         save->pwdmnt = mntget(current->fs->pwdmnt);
 143         save->ngroups = current_ngroups;
 144         save->luc.luc_umask = current->fs->umask;
 145
 146         LASSERT(save->pwd);
 147         LASSERT(save->pwdmnt);
 148         LASSERT(new_ctx->pwd);
 149         LASSERT(new_ctx->pwdmnt);
 150
 151         if (uc) {
 152                 save->luc.luc_fsuid = current->fsuid;
 153                 save->luc.luc_fsgid = current->fsgid;
 154                 save->luc.luc_cap = current->cap_effective;
 155
 156                 current->fsuid = uc->luc_fsuid;
 157                 current->fsgid = uc->luc_fsgid;
 158                 current->cap_effective = uc->luc_cap;
 159
 160                 push_group_info(save, uc->luc_ginfo);
 161         }
 162         current->fs->umask = 0; /* umask already applied on client */
 163         set_fs(new_ctx->fs);
 164         set_fs_pwd(current->fs, new_ctx->pwdmnt, new_ctx->pwd);
 165
 166         /*
 167         CDEBUG(D_INFO,
 168                "= push %p->%p = cur fs %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 169                new_ctx, current, current->fs, current->fs->pwd,
 170                atomic_read(&current->fs->pwd->d_count),
 171                atomic_read(&current->fs->pwd->d_inode->i_count),
 172                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 173                current->fs->pwdmnt,
 174                atomic_read(&current->fs->pwdmnt->mnt_count));
 175         */
 176 }
 177 EXPORT_SYMBOL(push_ctxt);
 178
 179 void pop_ctxt(struct lvfs_run_ctxt *saved, struct lvfs_run_ctxt *new_ctx,
 180               struct lvfs_ucred *uc)
 181 {
 182         //printk("pc0");
 183         ASSERT_CTXT_MAGIC(saved->magic);
 184         LASSERT(saved->pid == current->pid);
 185         saved->magic = 0;
 186         saved->pid = 0;
 187         //printk("pc1");
 188         ASSERT_KERNEL_CTXT("popping non-kernel context!\n");
 189
 190         /*
 191         CDEBUG(D_INFO,
 192                " = pop  %p==%p = cur %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 193                new_ctx, current, current->fs, current->fs->pwd,
 194                atomic_read(&current->fs->pwd->d_count),
 195                atomic_read(&current->fs->pwd->d_inode->i_count),
 196                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 197                current->fs->pwdmnt,
 198                atomic_read(&current->fs->pwdmnt->mnt_count));
 199         */
 200
 201         LASSERT(current->fs->pwd == new_ctx->pwd);
 202         LASSERT(current->fs->pwdmnt == new_ctx->pwdmnt);
 203
 204         set_fs(saved->fs);
 205         set_fs_pwd(current->fs, saved->pwdmnt, saved->pwd);
 206
 207         dput(saved->pwd);
 208         mntput(saved->pwdmnt);
 209         current->fs->umask = saved->luc.luc_umask;
 210         if (uc) {
 211                 current->fsuid = saved->luc.luc_fsuid;
 212                 current->fsgid = saved->luc.luc_fsgid;
 213                 current->cap_effective = saved->luc.luc_cap;
 214
 215                 pop_group_info(saved, uc->luc_ginfo);
 216         }
 217
 218         /*
 219         CDEBUG(D_INFO,
 220                "= pop  %p->%p = cur fs %p pwd %p:d%d:i%d (%*s), pwdmnt %p:%d\n",
 221                saved, current, current->fs, current->fs->pwd,
 222                atomic_read(&current->fs->pwd->d_count),
 223                atomic_read(&current->fs->pwd->d_inode->i_count),
 224                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 225                current->fs->pwdmnt,
 226                atomic_read(&current->fs->pwdmnt->mnt_count));
 227         */
 228 }
 229 EXPORT_SYMBOL(pop_ctxt);
 230
 231 /* utility to make a file */
 232 struct dentry *simple_mknod(struct dentry *dir, char *name, int mode, int fix)
 233 {
 234         struct dentry *dchild;
 235         int err = 0;
 236         ENTRY;
 237
 238         ASSERT_KERNEL_CTXT("kernel doing mknod outside kernel context\n");
 239         CDEBUG(D_INODE, "creating file %*s\n", (int)strlen(name), name);
 240
 241         dchild = ll_lookup_one_len(name, dir, strlen(name));
 242         if (IS_ERR(dchild))
 243                 GOTO(out_up, dchild);
 244
 245         if (dchild->d_inode) {
 246                 int old_mode = dchild->d_inode->i_mode;
 247                 if (!S_ISREG(old_mode))
 248                         GOTO(out_err, err = -EEXIST);
 249
 250                 /* Fixup file permissions if necessary */
 251                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 252                         CWARN("fixing permissions on %s from %o to %o\n",
 253                               name, old_mode, mode);
 254                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 255                                                   (old_mode & ~S_IALLUGO);
 256                         mark_inode_dirty(dchild->d_inode);
 257                 }
 258                 GOTO(out_up, dchild);
 259         }
 260
 261         err = ll_vfs_create(dir->d_inode, dchild, (mode & ~S_IFMT) | S_IFREG,
 262                             NULL);
 263         if (err)
 264                 GOTO(out_err, err);
 265
 266         RETURN(dchild);
 267
 268 out_err:
 269         dput(dchild);
 270         dchild = ERR_PTR(err);
 271 out_up:
 272         return dchild;
 273 }
 274 EXPORT_SYMBOL(simple_mknod);
 275
 276 /* utility to make a directory */
 277 struct dentry *simple_mkdir(struct dentry *dir, char *name, int mode, int fix)
 278 {
 279         struct dentry *dchild;
 280         int err = 0;
 281         ENTRY;
 282
 283         ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
 284         CDEBUG(D_INODE, "creating directory %*s\n", (int)strlen(name), name);
 285         dchild = ll_lookup_one_len(name, dir, strlen(name));
 286         if (IS_ERR(dchild))
 287                 GOTO(out_up, dchild);
 288
 289         if (dchild->d_inode) {
 290                 int old_mode = dchild->d_inode->i_mode;
 291                 if (!S_ISDIR(old_mode))
 292                         GOTO(out_err, err = -ENOTDIR);
 293
 294                 /* Fixup directory permissions if necessary */
 295                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 296                         CWARN("fixing permissions on %s from %o to %o\n",
 297                               name, old_mode, mode);
 298                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 299                                                   (old_mode & ~S_IALLUGO);
 300                         mark_inode_dirty(dchild->d_inode);
 301                 }
 302                 GOTO(out_up, dchild);
 303         }
 304
 305         err = vfs_mkdir(dir->d_inode, dchild, mode);
 306         if (err)
 307                 GOTO(out_err, err);
 308
 309         RETURN(dchild);
 310
 311 out_err:
 312         dput(dchild);
 313         dchild = ERR_PTR(err);
 314 out_up:
 315         return dchild;
 316 }
 317 EXPORT_SYMBOL(simple_mkdir);
 318
 319 /*
 320  * Read a file from within kernel context.  Prior to calling this
 321  * function we should already have done a push_ctxt().
 322  */
 323 int lustre_fread(struct file *file, void *buf, int len, loff_t *off)
 324 {
 325         ASSERT_KERNEL_CTXT("kernel doing read outside kernel context\n");
 326         if (!file || !file->f_op || !file->f_op->read || !off)
 327                 RETURN(-ENOSYS);
 328
 329         return file->f_op->read(file, buf, len, off);
 330 }
 331 EXPORT_SYMBOL(lustre_fread);
 332
 333 /*
 334  * Write a file from within kernel context.  Prior to calling this
 335  * function we should already have done a push_ctxt().
 336  */
 337 int lustre_fwrite(struct file *file, const void *buf, int len, loff_t *off)
 338 {
 339         ENTRY;
 340         ASSERT_KERNEL_CTXT("kernel doing write outside kernel context\n");
 341         if (!file)
 342                 RETURN(-ENOENT);
 343         if (!file->f_op)
 344                 RETURN(-ENOSYS);
 345         if (!off)
 346                 RETURN(-EINVAL);
 347
 348         if (!file->f_op->write)
 349                 RETURN(-EROFS);
 350
 351         RETURN(file->f_op->write(file, buf, len, off));
 352 }
 353 EXPORT_SYMBOL(lustre_fwrite);
 354
 355 /*
 356  * Sync a file from within kernel context.  Prior to calling this
 357  * function we should already have done a push_ctxt().
 358  */
 359 int lustre_fsync(struct file *file)
 360 {
 361         ENTRY;
 362         ASSERT_KERNEL_CTXT("kernel doing sync outside kernel context\n");
 363         if (!file || !file->f_op || !file->f_op->fsync)
 364                 RETURN(-ENOSYS);
 365
 366         RETURN(file->f_op->fsync(file, file->f_dentry, 0));
 367 }
 368 EXPORT_SYMBOL(lustre_fsync);
 369
 370 struct l_file *l_dentry_open(struct lvfs_run_ctxt *ctxt, struct l_dentry *de,
 371                              int flags)
 372 {
 373         mntget(ctxt->pwdmnt);
 374         return dentry_open(de, ctxt->pwdmnt, flags);
 375 }
 376 EXPORT_SYMBOL(l_dentry_open);
 377
 378 static int l_filldir(void *__buf, const char *name, int namlen, loff_t offset,
 379                      ino_t ino, unsigned int d_type)
 380 {
 381         struct l_linux_dirent *dirent;
 382         struct l_readdir_callback *buf = (struct l_readdir_callback *)__buf;
 383
 384         dirent = buf->lrc_dirent;
 385         if (dirent)
 386                dirent->lld_off = offset;
 387
 388         OBD_ALLOC(dirent, sizeof(*dirent));
 389
 390         list_add_tail(&dirent->lld_list, buf->lrc_list);
 391
 392         buf->lrc_dirent = dirent;
 393         dirent->lld_ino = ino;
 394         LASSERT(sizeof(dirent->lld_name) >= namlen + 1);
 395         memcpy(dirent->lld_name, name, namlen);
 396
 397         return 0;
 398 }
 399
 400 long l_readdir(struct file *file, struct list_head *dentry_list)
 401 {
 402         struct l_linux_dirent *lastdirent;
 403         struct l_readdir_callback buf;
 404         int error;
 405
 406         buf.lrc_dirent = NULL;
 407         buf.lrc_list = dentry_list;
 408
 409         error = vfs_readdir(file, l_filldir, &buf);
 410         if (error < 0)
 411                 return error;
 412
 413         lastdirent = buf.lrc_dirent;
 414         if (lastdirent)
 415                 lastdirent->lld_off = file->f_pos;
 416
 417         return 0;
 418 }
 419 EXPORT_SYMBOL(l_readdir);
 420 EXPORT_SYMBOL(obd_memory);
 421 EXPORT_SYMBOL(obd_memmax);
 422
 423 static int __init lvfs_linux_init(void)
 424 {
 425         RETURN(0);
 426 }
 427
 428 static void __exit lvfs_linux_exit(void)
 429 {
 430         int leaked;
 431         ENTRY;
 432
 433         leaked = atomic_read(&obd_memory);
 434         CDEBUG(leaked ? D_ERROR : D_INFO,
 435                "obd mem max: %d leaked: %d\n", obd_memmax, leaked);
 436
 437         return;
 438 }
 439
 440 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 441 MODULE_DESCRIPTION("Lustre VFS Filesystem Helper v0.1");
 442 MODULE_LICENSE("GPL");
 443
 444 module_init(lvfs_linux_init);
 445 module_exit(lvfs_linux_exit);