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 <lustre_fsfilt.h>
  41 #include <obd.h>
  42 #include <linux/module.h>
  43 #include <linux/init.h>
  44 #include <linux/lustre_compat25.h>
  45 #include <lvfs.h>
  46 #include "lvfs_internal.h"
  47
  48 #include <obd.h>
  49 #include <lustre_lib.h>
  50 #include <lustre_quota.h>
  51
  52 atomic_t obd_memory;
  53 int obd_memmax;
  54 unsigned int obd_fail_val;
  55 unsigned int obd_fail_loc;
  56 unsigned int obd_alloc_fail_rate = 0;
  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) LASSERTF(!segment_eq(get_fs(), get_ds()),\
  62                                               msg)
  63 # define ASSERT_KERNEL_CTXT(msg) LASSERTF(segment_eq(get_fs(), get_ds()), msg)
  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 upcall_cache_entry *uce)
  72 {
  73         struct group_info *ginfo = uce ? uce->ue_group_info : NULL;
  74
  75         if (!ginfo) {
  76                 save->ngroups = current_ngroups;
  77                 current_ngroups = 0;
  78         } else {
  79 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,4)
  80                 task_lock(current);
  81                 save->group_info = current->group_info;
  82                 current->group_info = ginfo;
  83                 task_unlock(current);
  84 #else
  85                 LASSERT(ginfo->ngroups <= NGROUPS);
  86                 LASSERT(current->ngroups <= NGROUPS_SMALL);
  87                 /* save old */
  88                 save->group_info.ngroups = current->ngroups;
  89                 if (current->ngroups)
  90                         memcpy(save->group_info.small_block, current->groups,
  91                                current->ngroups * sizeof(gid_t));
  92                 /* push new */
  93                 current->ngroups = ginfo->ngroups;
  94                 if (ginfo->ngroups)
  95                         memcpy(current->groups, ginfo->small_block,
  96                                current->ngroups * sizeof(gid_t));
  97 #endif
  98         }
  99 }
 100
 101 static void pop_group_info(struct lvfs_run_ctxt *save,
 102                            struct upcall_cache_entry *uce)
 103 {
 104         struct group_info *ginfo = uce ? uce->ue_group_info : NULL;
 105
 106         if (!ginfo) {
 107                 current_ngroups = save->ngroups;
 108         } else {
 109 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,4)
 110                 task_lock(current);
 111                 current->group_info = save->group_info;
 112                 task_unlock(current);
 113 #else
 114                 current->ngroups = save->group_info.ngroups;
 115                 if (current->ngroups)
 116                         memcpy(current->groups, save->group_info.small_block,
 117                                current->ngroups * sizeof(gid_t));
 118 #endif
 119         }
 120 }
 121
 122 /* push / pop to root of obd store */
 123 void push_ctxt(struct lvfs_run_ctxt *save, struct lvfs_run_ctxt *new_ctx,
 124                struct lvfs_ucred *uc)
 125 {
 126         //ASSERT_NOT_KERNEL_CTXT("already in kernel context!\n");
 127         ASSERT_CTXT_MAGIC(new_ctx->magic);
 128         OBD_SET_CTXT_MAGIC(save);
 129
 130         /*
 131         CDEBUG(D_INFO,
 132                "= push %p->%p = cur fs %p pwd %p:d%d:i%d (%.*s), pwdmnt %p:%d\n",
 133                save, current, current->fs, current->fs->pwd,
 134                atomic_read(&current->fs->pwd->d_count),
 135                atomic_read(&current->fs->pwd->d_inode->i_count),
 136                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 137                current->fs->pwdmnt,
 138                atomic_read(&current->fs->pwdmnt->mnt_count));
 139         */
 140
 141         save->fs = get_fs();
 142         LASSERT(atomic_read(&current->fs->pwd->d_count));
 143         LASSERT(atomic_read(&new_ctx->pwd->d_count));
 144         save->pwd = dget(current->fs->pwd);
 145         save->pwdmnt = mntget(current->fs->pwdmnt);
 146         save->luc.luc_umask = current->fs->umask;
 147
 148         LASSERT(save->pwd);
 149         LASSERT(save->pwdmnt);
 150         LASSERT(new_ctx->pwd);
 151         LASSERT(new_ctx->pwdmnt);
 152
 153         if (uc) {
 154                 save->luc.luc_fsuid = current->fsuid;
 155                 save->luc.luc_fsgid = current->fsgid;
 156                 save->luc.luc_cap = current->cap_effective;
 157
 158                 current->fsuid = uc->luc_fsuid;
 159                 current->fsgid = uc->luc_fsgid;
 160                 current->cap_effective = uc->luc_cap;
 161                 push_group_info(save, uc->luc_uce);
 162         }
 163         current->fs->umask = 0; /* umask already applied on client */
 164         set_fs(new_ctx->fs);
 165         ll_set_fs_pwd(current->fs, new_ctx->pwdmnt, new_ctx->pwd);
 166
 167         /*
 168         CDEBUG(D_INFO,
 169                "= push %p->%p = cur fs %p pwd %p:d%d:i%d (%.*s), pwdmnt %p:%d\n",
 170                new_ctx, current, current->fs, current->fs->pwd,
 171                atomic_read(&current->fs->pwd->d_count),
 172                atomic_read(&current->fs->pwd->d_inode->i_count),
 173                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 174                current->fs->pwdmnt,
 175                atomic_read(&current->fs->pwdmnt->mnt_count));
 176         */
 177 }
 178 EXPORT_SYMBOL(push_ctxt);
 179
 180 void pop_ctxt(struct lvfs_run_ctxt *saved, struct lvfs_run_ctxt *new_ctx,
 181               struct lvfs_ucred *uc)
 182 {
 183         //printk("pc0");
 184         ASSERT_CTXT_MAGIC(saved->magic);
 185         //printk("pc1");
 186         ASSERT_KERNEL_CTXT("popping non-kernel context!\n");
 187
 188         /*
 189         CDEBUG(D_INFO,
 190                " = pop  %p==%p = cur %p pwd %p:d%d:i%d (%.*s), pwdmnt %p:%d\n",
 191                new_ctx, current, current->fs, current->fs->pwd,
 192                atomic_read(&current->fs->pwd->d_count),
 193                atomic_read(&current->fs->pwd->d_inode->i_count),
 194                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 195                current->fs->pwdmnt,
 196                atomic_read(&current->fs->pwdmnt->mnt_count));
 197         */
 198
 199         LASSERTF(current->fs->pwd == new_ctx->pwd, "%p != %p\n",
 200                  current->fs->pwd, new_ctx->pwd);
 201         LASSERTF(current->fs->pwdmnt == new_ctx->pwdmnt, "%p != %p\n",
 202                  current->fs->pwdmnt, new_ctx->pwdmnt);
 203
 204         set_fs(saved->fs);
 205         ll_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                 pop_group_info(saved, uc->luc_uce);
 215         }
 216
 217         /*
 218         CDEBUG(D_INFO,
 219                "= pop  %p->%p = cur fs %p pwd %p:d%d:i%d (%.*s), pwdmnt %p:%d\n",
 220                saved, current, current->fs, current->fs->pwd,
 221                atomic_read(&current->fs->pwd->d_count),
 222                atomic_read(&current->fs->pwd->d_inode->i_count),
 223                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 224                current->fs->pwdmnt,
 225                atomic_read(&current->fs->pwdmnt->mnt_count));
 226         */
 227 }
 228 EXPORT_SYMBOL(pop_ctxt);
 229
 230 /* utility to make a file */
 231 struct dentry *simple_mknod(struct dentry *dir, char *name, int mode, int fix)
 232 {
 233         struct dentry *dchild;
 234         int err = 0;
 235         ENTRY;
 236
 237         ASSERT_KERNEL_CTXT("kernel doing mknod outside kernel context\n");
 238         CDEBUG(D_INODE, "creating file %.*s\n", (int)strlen(name), name);
 239
 240         dchild = ll_lookup_one_len(name, dir, strlen(name));
 241         if (IS_ERR(dchild))
 242                 GOTO(out_up, dchild);
 243
 244         if (dchild->d_inode) {
 245                 int old_mode = dchild->d_inode->i_mode;
 246                 if (!S_ISREG(old_mode))
 247                         GOTO(out_err, err = -EEXIST);
 248
 249                 /* Fixup file permissions if necessary */
 250                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 251                         CWARN("fixing permissions on %s from %o to %o\n",
 252                               name, old_mode, mode);
 253                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 254                                                   (old_mode & ~S_IALLUGO);
 255                         mark_inode_dirty(dchild->d_inode);
 256                 }
 257                 GOTO(out_up, dchild);
 258         }
 259
 260         err = ll_vfs_create(dir->d_inode, dchild, (mode & ~S_IFMT) | S_IFREG,
 261                             NULL);
 262         if (err)
 263                 GOTO(out_err, err);
 264
 265         RETURN(dchild);
 266
 267 out_err:
 268         dput(dchild);
 269         dchild = ERR_PTR(err);
 270 out_up:
 271         return dchild;
 272 }
 273 EXPORT_SYMBOL(simple_mknod);
 274
 275 /* utility to make a directory */
 276 struct dentry *simple_mkdir(struct dentry *dir, char *name, int mode, int fix)
 277 {
 278         struct dentry *dchild;
 279         int err = 0;
 280         ENTRY;
 281
 282         ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
 283         CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
 284         dchild = ll_lookup_one_len(name, dir, strlen(name));
 285         if (IS_ERR(dchild))
 286                 GOTO(out_up, dchild);
 287
 288         if (dchild->d_inode) {
 289                 int old_mode = dchild->d_inode->i_mode;
 290                 if (!S_ISDIR(old_mode)) {
 291                         CERROR("found %s (%lu/%u) is mode %o\n", name,
 292                                dchild->d_inode->i_ino,
 293                                dchild->d_inode->i_generation, old_mode);
 294                         GOTO(out_err, err = -ENOTDIR);
 295                 }
 296
 297                 /* Fixup directory permissions if necessary */
 298                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 299                         CDEBUG(D_CONFIG,
 300                                "fixing permissions on %s from %o to %o\n",
 301                                name, old_mode, mode);
 302                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 303                                                   (old_mode & ~S_IALLUGO);
 304                         mark_inode_dirty(dchild->d_inode);
 305                 }
 306                 GOTO(out_up, dchild);
 307         }
 308
 309         err = vfs_mkdir(dir->d_inode, dchild, mode);
 310         if (err)
 311                 GOTO(out_err, err);
 312
 313         RETURN(dchild);
 314
 315 out_err:
 316         dput(dchild);
 317         dchild = ERR_PTR(err);
 318 out_up:
 319         return dchild;
 320 }
 321 EXPORT_SYMBOL(simple_mkdir);
 322
 323 /* utility to rename a file */
 324 int lustre_rename(struct dentry *dir, char *oldname, char *newname)
 325 {
 326         struct dentry *dchild_old, *dchild_new;
 327         int err = 0;
 328         ENTRY;
 329
 330         ASSERT_KERNEL_CTXT("kernel doing rename outside kernel context\n");
 331         CDEBUG(D_INODE, "renaming file %.*s to %.*s\n",
 332                (int)strlen(oldname), oldname, (int)strlen(newname), newname);
 333
 334         dchild_old = ll_lookup_one_len(oldname, dir, strlen(oldname));
 335         if (IS_ERR(dchild_old))
 336                 RETURN(PTR_ERR(dchild_old));
 337
 338         if (!dchild_old->d_inode)
 339                 GOTO(put_old, err = -ENOENT);
 340
 341         dchild_new = ll_lookup_one_len(newname, dir, strlen(newname));
 342         if (IS_ERR(dchild_new))
 343                 GOTO(put_old, err = PTR_ERR(dchild_new));
 344
 345         err = vfs_rename(dir->d_inode, dchild_old, dir->d_inode, dchild_new);
 346
 347         dput(dchild_new);
 348 put_old:
 349         dput(dchild_old);
 350         RETURN(err);
 351 }
 352 EXPORT_SYMBOL(lustre_rename);
 353
 354 /*
 355  * Read a file from within kernel context.  Prior to calling this
 356  * function we should already have done a push_ctxt().
 357  */
 358 int lustre_fread(struct file *file, void *buf, int len, loff_t *off)
 359 {
 360         ASSERT_KERNEL_CTXT("kernel doing read outside kernel context\n");
 361         if (!file || !file->f_op || !file->f_op->read || !off)
 362                 RETURN(-ENOSYS);
 363
 364         return file->f_op->read(file, buf, len, off);
 365 }
 366 EXPORT_SYMBOL(lustre_fread);
 367
 368 /*
 369  * Write a file from within kernel context.  Prior to calling this
 370  * function we should already have done a push_ctxt().
 371  */
 372 int lustre_fwrite(struct file *file, const void *buf, int len, loff_t *off)
 373 {
 374         ENTRY;
 375         ASSERT_KERNEL_CTXT("kernel doing write outside kernel context\n");
 376         if (!file)
 377                 RETURN(-ENOENT);
 378         if (!file->f_op)
 379                 RETURN(-ENOSYS);
 380         if (!off)
 381                 RETURN(-EINVAL);
 382
 383         if (!file->f_op->write)
 384                 RETURN(-EROFS);
 385
 386         RETURN(file->f_op->write(file, buf, len, off));
 387 }
 388 EXPORT_SYMBOL(lustre_fwrite);
 389
 390 /*
 391  * Sync a file from within kernel context.  Prior to calling this
 392  * function we should already have done a push_ctxt().
 393  */
 394 int lustre_fsync(struct file *file)
 395 {
 396         ENTRY;
 397         ASSERT_KERNEL_CTXT("kernel doing sync outside kernel context\n");
 398         if (!file || !file->f_op || !file->f_op->fsync)
 399                 RETURN(-ENOSYS);
 400
 401         RETURN(file->f_op->fsync(file, file->f_dentry, 0));
 402 }
 403 EXPORT_SYMBOL(lustre_fsync);
 404
 405 struct l_file *l_dentry_open(struct lvfs_run_ctxt *ctxt, struct l_dentry *de,
 406                              int flags)
 407 {
 408         mntget(ctxt->pwdmnt);
 409         return dentry_open(de, ctxt->pwdmnt, flags);
 410 }
 411 EXPORT_SYMBOL(l_dentry_open);
 412
 413 #ifdef HAVE_VFS_READDIR_U64_INO
 414 static int l_filldir(void *__buf, const char *name, int namlen, loff_t offset,
 415                      u64 ino, unsigned int d_type)
 416 #else
 417 static int l_filldir(void *__buf, const char *name, int namlen, loff_t offset,
 418                      ino_t ino, unsigned int d_type)
 419 #endif
 420 {
 421         struct l_linux_dirent *dirent;
 422         struct l_readdir_callback *buf = (struct l_readdir_callback *)__buf;
 423
 424         dirent = buf->lrc_dirent;
 425         if (dirent)
 426                dirent->lld_off = offset;
 427
 428         OBD_ALLOC(dirent, sizeof(*dirent));
 429
 430         if (!dirent)
 431                 return -ENOMEM;
 432
 433         list_add_tail(&dirent->lld_list, buf->lrc_list);
 434
 435         buf->lrc_dirent = dirent;
 436         dirent->lld_ino = ino;
 437         LASSERT(sizeof(dirent->lld_name) >= namlen + 1);
 438         memcpy(dirent->lld_name, name, namlen);
 439
 440         return 0;
 441 }
 442
 443 long l_readdir(struct file *file, struct list_head *dentry_list)
 444 {
 445         struct l_linux_dirent *lastdirent;
 446         struct l_readdir_callback buf;
 447         int error;
 448
 449         buf.lrc_dirent = NULL;
 450         buf.lrc_list = dentry_list;
 451
 452         error = vfs_readdir(file, l_filldir, &buf);
 453         if (error < 0)
 454                 return error;
 455
 456         lastdirent = buf.lrc_dirent;
 457         if (lastdirent)
 458                 lastdirent->lld_off = file->f_pos;
 459
 460         return 0;
 461 }
 462 EXPORT_SYMBOL(l_readdir);
 463 EXPORT_SYMBOL(obd_memory);
 464 EXPORT_SYMBOL(obd_memmax);
 465
 466 #ifdef LUSTRE_KERNEL_VERSION
 467 #ifndef HAVE_CLEAR_RDONLY_ON_PUT
 468 #error rdonly patchset must be updated [cfs bz11248]
 469 #endif
 470
 471 void dev_set_rdonly(lvfs_sbdev_type dev);
 472 int dev_check_rdonly(lvfs_sbdev_type dev);
 473
 474 void __lvfs_set_rdonly(lvfs_sbdev_type dev, lvfs_sbdev_type jdev)
 475 {
 476         lvfs_sbdev_sync(dev);
 477         if (jdev && (jdev != dev)) {
 478                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
 479                        (long)jdev);
 480                 dev_set_rdonly(jdev);
 481         }
 482         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
 483         dev_set_rdonly(dev);
 484 }
 485
 486 int lvfs_check_rdonly(lvfs_sbdev_type dev)
 487 {
 488         return dev_check_rdonly(dev);
 489 }
 490
 491 EXPORT_SYMBOL(__lvfs_set_rdonly);
 492 EXPORT_SYMBOL(lvfs_check_rdonly);
 493 #endif /* LUSTRE_KERNEL_VERSION */
 494
 495 int lvfs_check_io_health(struct obd_device *obd, struct file *file)
 496 {
 497         char *write_page = NULL;
 498         loff_t offset = 0;
 499         int rc = 0;
 500         ENTRY;
 501
 502         OBD_ALLOC(write_page, CFS_PAGE_SIZE);
 503         if (!write_page)
 504                 RETURN(-ENOMEM);
 505
 506         rc = fsfilt_write_record(obd, file, write_page, CFS_PAGE_SIZE, &offset, 1);
 507
 508         OBD_FREE(write_page, CFS_PAGE_SIZE);
 509
 510         CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
 511         RETURN(rc);
 512 }
 513 EXPORT_SYMBOL(lvfs_check_io_health);
 514
 515 static int __init lvfs_linux_init(void)
 516 {
 517         RETURN(0);
 518 }
 519
 520 static void __exit lvfs_linux_exit(void)
 521 {
 522         int leaked;
 523         ENTRY;
 524
 525         leaked = atomic_read(&obd_memory);
 526         CDEBUG(leaked ? D_ERROR : D_INFO,
 527                "obd mem max: %d leaked: %d\n", obd_memmax, leaked);
 528
 529         EXIT;
 530         return;
 531 }
 532
 533 EXPORT_SYMBOL(obd_fail_loc);
 534 EXPORT_SYMBOL(obd_alloc_fail_rate);
 535 EXPORT_SYMBOL(obd_fail_val);
 536
 537 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 538 MODULE_DESCRIPTION("Lustre VFS Filesystem Helper v0.1");
 539 MODULE_LICENSE("GPL");
 540
 541 module_init(lvfs_linux_init);
 542 module_exit(lvfs_linux_exit);