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  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/lvfs/lvfs_linux.c
  37  *
  38  * Author: Andreas Dilger <adilger@clusterfs.com>
  39  */
  40
  41 #ifndef EXPORT_SYMTAB
  42 # define EXPORT_SYMTAB
  43 #endif
  44
  45 #define DEBUG_SUBSYSTEM S_FILTER
  46
  47 #include <linux/version.h>
  48 #include <linux/fs.h>
  49 #include <asm/unistd.h>
  50 #include <linux/slab.h>
  51 #include <linux/pagemap.h>
  52 #include <linux/quotaops.h>
  53 #include <linux/version.h>
  54 #include <libcfs/kp30.h>
  55 #include <lustre_fsfilt.h>
  56 #include <obd.h>
  57 #include <linux/module.h>
  58 #include <linux/init.h>
  59 #include <linux/lustre_compat25.h>
  60 #include <lvfs.h>
  61 #include "lvfs_internal.h"
  62
  63 #include <obd.h>
  64 #include <lustre_lib.h>
  65 #include <lustre_quota.h>
  66
  67 /* Debugging check only needed during development */
  68 #ifdef OBD_CTXT_DEBUG
  69 # define ASSERT_CTXT_MAGIC(magic) LASSERT((magic) == OBD_RUN_CTXT_MAGIC)
  70 # define ASSERT_NOT_KERNEL_CTXT(msg) LASSERTF(!segment_eq(get_fs(), get_ds()),\
  71                                               msg)
  72 # define ASSERT_KERNEL_CTXT(msg) LASSERTF(segment_eq(get_fs(), get_ds()), msg)
  73 #else
  74 # define ASSERT_CTXT_MAGIC(magic) do {} while(0)
  75 # define ASSERT_NOT_KERNEL_CTXT(msg) do {} while(0)
  76 # define ASSERT_KERNEL_CTXT(msg) do {} while(0)
  77 #endif
  78
  79 static void push_group_info(struct lvfs_run_ctxt *save,
  80                             struct upcall_cache_entry *uce)
  81 {
  82         struct group_info *ginfo = uce ? uce->ue_group_info : NULL;
  83
  84         if (!ginfo) {
  85                 save->ngroups = current_ngroups;
  86                 current_ngroups = 0;
  87         } else {
  88 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,4)
  89                 task_lock(current);
  90                 save->group_info = current->group_info;
  91                 current->group_info = ginfo;
  92                 task_unlock(current);
  93 #else
  94                 LASSERT(ginfo->ngroups <= NGROUPS);
  95                 LASSERT(current->ngroups <= NGROUPS_SMALL);
  96                 /* save old */
  97                 save->group_info.ngroups = current->ngroups;
  98                 if (current->ngroups)
  99                         memcpy(save->group_info.small_block, current->groups,
 100                                current->ngroups * sizeof(gid_t));
 101                 /* push new */
 102                 current->ngroups = ginfo->ngroups;
 103                 if (ginfo->ngroups)
 104                         memcpy(current->groups, ginfo->small_block,
 105                                current->ngroups * sizeof(gid_t));
 106 #endif
 107         }
 108 }
 109
 110 static void pop_group_info(struct lvfs_run_ctxt *save,
 111                            struct upcall_cache_entry *uce)
 112 {
 113         struct group_info *ginfo = uce ? uce->ue_group_info : NULL;
 114
 115         if (!ginfo) {
 116                 current_ngroups = save->ngroups;
 117         } else {
 118 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,4)
 119                 task_lock(current);
 120                 current->group_info = save->group_info;
 121                 task_unlock(current);
 122 #else
 123                 current->ngroups = save->group_info.ngroups;
 124                 if (current->ngroups)
 125                         memcpy(current->groups, save->group_info.small_block,
 126                                current->ngroups * sizeof(gid_t));
 127 #endif
 128         }
 129 }
 130
 131 /* push / pop to root of obd store */
 132 void push_ctxt(struct lvfs_run_ctxt *save, struct lvfs_run_ctxt *new_ctx,
 133                struct lvfs_ucred *uc)
 134 {
 135         //ASSERT_NOT_KERNEL_CTXT("already in kernel context!\n");
 136         ASSERT_CTXT_MAGIC(new_ctx->magic);
 137         OBD_SET_CTXT_MAGIC(save);
 138
 139         /*
 140         CDEBUG(D_INFO,
 141                "= push %p->%p = cur fs %p pwd %p:d%d:i%d (%.*s), pwdmnt %p:%d\n",
 142                save, current, current->fs, current->fs->pwd,
 143                atomic_read(&current->fs->pwd->d_count),
 144                atomic_read(&current->fs->pwd->d_inode->i_count),
 145                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 146                current->fs->pwdmnt,
 147                atomic_read(&current->fs->pwdmnt->mnt_count));
 148         */
 149
 150         save->fs = get_fs();
 151         LASSERT(atomic_read(&current->fs->pwd->d_count));
 152         LASSERT(atomic_read(&new_ctx->pwd->d_count));
 153         save->pwd = dget(current->fs->pwd);
 154         save->pwdmnt = mntget(current->fs->pwdmnt);
 155         save->luc.luc_umask = current->fs->umask;
 156
 157         LASSERT(save->pwd);
 158         LASSERT(save->pwdmnt);
 159         LASSERT(new_ctx->pwd);
 160         LASSERT(new_ctx->pwdmnt);
 161
 162         if (uc) {
 163                 save->luc.luc_fsuid = current->fsuid;
 164                 save->luc.luc_fsgid = current->fsgid;
 165                 save->luc.luc_cap = current->cap_effective;
 166
 167                 current->fsuid = uc->luc_fsuid;
 168                 current->fsgid = uc->luc_fsgid;
 169                 current->cap_effective = uc->luc_cap;
 170                 push_group_info(save, uc->luc_uce);
 171         }
 172         current->fs->umask = 0; /* umask already applied on client */
 173         set_fs(new_ctx->fs);
 174         ll_set_fs_pwd(current->fs, new_ctx->pwdmnt, new_ctx->pwd);
 175
 176         /*
 177         CDEBUG(D_INFO,
 178                "= push %p->%p = cur fs %p pwd %p:d%d:i%d (%.*s), pwdmnt %p:%d\n",
 179                new_ctx, current, current->fs, current->fs->pwd,
 180                atomic_read(&current->fs->pwd->d_count),
 181                atomic_read(&current->fs->pwd->d_inode->i_count),
 182                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 183                current->fs->pwdmnt,
 184                atomic_read(&current->fs->pwdmnt->mnt_count));
 185         */
 186 }
 187 EXPORT_SYMBOL(push_ctxt);
 188
 189 void pop_ctxt(struct lvfs_run_ctxt *saved, struct lvfs_run_ctxt *new_ctx,
 190               struct lvfs_ucred *uc)
 191 {
 192         //printk("pc0");
 193         ASSERT_CTXT_MAGIC(saved->magic);
 194         //printk("pc1");
 195         ASSERT_KERNEL_CTXT("popping non-kernel context!\n");
 196
 197         /*
 198         CDEBUG(D_INFO,
 199                " = pop  %p==%p = cur %p pwd %p:d%d:i%d (%.*s), pwdmnt %p:%d\n",
 200                new_ctx, current, current->fs, current->fs->pwd,
 201                atomic_read(&current->fs->pwd->d_count),
 202                atomic_read(&current->fs->pwd->d_inode->i_count),
 203                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 204                current->fs->pwdmnt,
 205                atomic_read(&current->fs->pwdmnt->mnt_count));
 206         */
 207
 208         LASSERTF(current->fs->pwd == new_ctx->pwd, "%p != %p\n",
 209                  current->fs->pwd, new_ctx->pwd);
 210         LASSERTF(current->fs->pwdmnt == new_ctx->pwdmnt, "%p != %p\n",
 211                  current->fs->pwdmnt, new_ctx->pwdmnt);
 212
 213         set_fs(saved->fs);
 214         ll_set_fs_pwd(current->fs, saved->pwdmnt, saved->pwd);
 215
 216         dput(saved->pwd);
 217         mntput(saved->pwdmnt);
 218         current->fs->umask = saved->luc.luc_umask;
 219         if (uc) {
 220                 current->fsuid = saved->luc.luc_fsuid;
 221                 current->fsgid = saved->luc.luc_fsgid;
 222                 current->cap_effective = saved->luc.luc_cap;
 223                 pop_group_info(saved, uc->luc_uce);
 224         }
 225
 226         /*
 227         CDEBUG(D_INFO,
 228                "= pop  %p->%p = cur fs %p pwd %p:d%d:i%d (%.*s), pwdmnt %p:%d\n",
 229                saved, current, current->fs, current->fs->pwd,
 230                atomic_read(&current->fs->pwd->d_count),
 231                atomic_read(&current->fs->pwd->d_inode->i_count),
 232                current->fs->pwd->d_name.len, current->fs->pwd->d_name.name,
 233                current->fs->pwdmnt,
 234                atomic_read(&current->fs->pwdmnt->mnt_count));
 235         */
 236 }
 237 EXPORT_SYMBOL(pop_ctxt);
 238
 239 /* utility to make a file */
 240 struct dentry *simple_mknod(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 mknod outside kernel context\n");
 247         CDEBUG(D_INODE, "creating file %.*s\n", (int)strlen(name), name);
 248
 249         dchild = ll_lookup_one_len(name, dir, strlen(name));
 250         if (IS_ERR(dchild))
 251                 GOTO(out_up, dchild);
 252
 253         if (dchild->d_inode) {
 254                 int old_mode = dchild->d_inode->i_mode;
 255                 if (!S_ISREG(old_mode))
 256                         GOTO(out_err, err = -EEXIST);
 257
 258                 /* Fixup file permissions if necessary */
 259                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 260                         CWARN("fixing permissions on %s from %o to %o\n",
 261                               name, old_mode, mode);
 262                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 263                                                   (old_mode & ~S_IALLUGO);
 264                         mark_inode_dirty(dchild->d_inode);
 265                 }
 266                 GOTO(out_up, dchild);
 267         }
 268
 269         err = ll_vfs_create(dir->d_inode, dchild, (mode & ~S_IFMT) | S_IFREG,
 270                             NULL);
 271         if (err)
 272                 GOTO(out_err, err);
 273
 274         RETURN(dchild);
 275
 276 out_err:
 277         dput(dchild);
 278         dchild = ERR_PTR(err);
 279 out_up:
 280         return dchild;
 281 }
 282 EXPORT_SYMBOL(simple_mknod);
 283
 284 /* utility to make a directory */
 285 struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
 286                             char *name, int mode, int fix)
 287 {
 288         struct dentry *dchild;
 289         int err = 0;
 290         ENTRY;
 291
 292         ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
 293         CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
 294         dchild = ll_lookup_one_len(name, dir, strlen(name));
 295         if (IS_ERR(dchild))
 296                 GOTO(out_up, dchild);
 297
 298         if (dchild->d_inode) {
 299                 int old_mode = dchild->d_inode->i_mode;
 300                 if (!S_ISDIR(old_mode)) {
 301                         CERROR("found %s (%lu/%u) is mode %o\n", name,
 302                                dchild->d_inode->i_ino,
 303                                dchild->d_inode->i_generation, old_mode);
 304                         GOTO(out_err, err = -ENOTDIR);
 305                 }
 306
 307                 /* Fixup directory permissions if necessary */
 308                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 309                         CDEBUG(D_CONFIG,
 310                                "fixing permissions on %s from %o to %o\n",
 311                                name, old_mode, mode);
 312                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 313                                                   (old_mode & ~S_IALLUGO);
 314                         mark_inode_dirty(dchild->d_inode);
 315                 }
 316                 GOTO(out_up, dchild);
 317         }
 318
 319         err = ll_vfs_mkdir(dir->d_inode, dchild, mnt, mode);
 320         if (err)
 321                 GOTO(out_err, err);
 322
 323         RETURN(dchild);
 324
 325 out_err:
 326         dput(dchild);
 327         dchild = ERR_PTR(err);
 328 out_up:
 329         return dchild;
 330 }
 331 EXPORT_SYMBOL(simple_mkdir);
 332
 333 /* utility to rename a file */
 334 int lustre_rename(struct dentry *dir, struct vfsmount *mnt,
 335                   char *oldname, char *newname)
 336 {
 337         struct dentry *dchild_old, *dchild_new;
 338         int err = 0;
 339         ENTRY;
 340
 341         ASSERT_KERNEL_CTXT("kernel doing rename outside kernel context\n");
 342         CDEBUG(D_INODE, "renaming file %.*s to %.*s\n",
 343                (int)strlen(oldname), oldname, (int)strlen(newname), newname);
 344
 345         dchild_old = ll_lookup_one_len(oldname, dir, strlen(oldname));
 346         if (IS_ERR(dchild_old))
 347                 RETURN(PTR_ERR(dchild_old));
 348
 349         if (!dchild_old->d_inode)
 350                 GOTO(put_old, err = -ENOENT);
 351
 352         dchild_new = ll_lookup_one_len(newname, dir, strlen(newname));
 353         if (IS_ERR(dchild_new))
 354                 GOTO(put_old, err = PTR_ERR(dchild_new));
 355
 356         err = ll_vfs_rename(dir->d_inode, dchild_old, mnt,
 357                             dir->d_inode, dchild_new, mnt);
 358
 359         dput(dchild_new);
 360 put_old:
 361         dput(dchild_old);
 362         RETURN(err);
 363 }
 364 EXPORT_SYMBOL(lustre_rename);
 365
 366 /*
 367  * Read a file from within kernel context.  Prior to calling this
 368  * function we should already have done a push_ctxt().
 369  */
 370 int lustre_fread(struct file *file, void *buf, int len, loff_t *off)
 371 {
 372         ASSERT_KERNEL_CTXT("kernel doing read outside kernel context\n");
 373         if (!file || !file->f_op || !file->f_op->read || !off)
 374                 RETURN(-ENOSYS);
 375
 376         return file->f_op->read(file, buf, len, off);
 377 }
 378 EXPORT_SYMBOL(lustre_fread);
 379
 380 /*
 381  * Write a file from within kernel context.  Prior to calling this
 382  * function we should already have done a push_ctxt().
 383  */
 384 int lustre_fwrite(struct file *file, const void *buf, int len, loff_t *off)
 385 {
 386         ENTRY;
 387         ASSERT_KERNEL_CTXT("kernel doing write outside kernel context\n");
 388         if (!file)
 389                 RETURN(-ENOENT);
 390         if (!file->f_op)
 391                 RETURN(-ENOSYS);
 392         if (!off)
 393                 RETURN(-EINVAL);
 394
 395         if (!file->f_op->write)
 396                 RETURN(-EROFS);
 397
 398         RETURN(file->f_op->write(file, buf, len, off));
 399 }
 400 EXPORT_SYMBOL(lustre_fwrite);
 401
 402 /*
 403  * Sync a file from within kernel context.  Prior to calling this
 404  * function we should already have done a push_ctxt().
 405  */
 406 int lustre_fsync(struct file *file)
 407 {
 408         ENTRY;
 409         ASSERT_KERNEL_CTXT("kernel doing sync outside kernel context\n");
 410         if (!file || !file->f_op || !file->f_op->fsync)
 411                 RETURN(-ENOSYS);
 412
 413         RETURN(file->f_op->fsync(file, file->f_dentry, 0));
 414 }
 415 EXPORT_SYMBOL(lustre_fsync);
 416
 417 struct l_file *l_dentry_open(struct lvfs_run_ctxt *ctxt, struct l_dentry *de,
 418                              int flags)
 419 {
 420         mntget(ctxt->pwdmnt);
 421         return dentry_open(de, ctxt->pwdmnt, flags);
 422 }
 423 EXPORT_SYMBOL(l_dentry_open);
 424
 425 #ifdef HAVE_VFS_READDIR_U64_INO
 426 static int l_filldir(void *__buf, const char *name, int namlen, loff_t offset,
 427                      u64 ino, unsigned int d_type)
 428 #else
 429 static int l_filldir(void *__buf, const char *name, int namlen, loff_t offset,
 430                      ino_t ino, unsigned int d_type)
 431 #endif
 432 {
 433         struct l_linux_dirent *dirent;
 434         struct l_readdir_callback *buf = (struct l_readdir_callback *)__buf;
 435
 436         dirent = buf->lrc_dirent;
 437         if (dirent)
 438                dirent->lld_off = offset;
 439
 440         OBD_ALLOC(dirent, sizeof(*dirent));
 441
 442         if (!dirent)
 443                 return -ENOMEM;
 444
 445         list_add_tail(&dirent->lld_list, buf->lrc_list);
 446
 447         buf->lrc_dirent = dirent;
 448         dirent->lld_ino = ino;
 449         LASSERT(sizeof(dirent->lld_name) >= namlen + 1);
 450         memcpy(dirent->lld_name, name, namlen);
 451
 452         return 0;
 453 }
 454
 455 long l_readdir(struct file *file, struct list_head *dentry_list)
 456 {
 457         struct l_linux_dirent *lastdirent;
 458         struct l_readdir_callback buf;
 459         int error;
 460
 461         buf.lrc_dirent = NULL;
 462         buf.lrc_list = dentry_list;
 463
 464         error = vfs_readdir(file, l_filldir, &buf);
 465         if (error < 0)
 466                 return error;
 467
 468         lastdirent = buf.lrc_dirent;
 469         if (lastdirent)
 470                 lastdirent->lld_off = file->f_pos;
 471
 472         return 0;
 473 }
 474 EXPORT_SYMBOL(l_readdir);
 475
 476 int l_notify_change(struct vfsmount *mnt, struct dentry *dchild,
 477                     struct iattr *newattrs)
 478 {
 479         int rc;
 480
 481         LOCK_INODE_MUTEX(dchild->d_inode);
 482 #ifdef HAVE_SECURITY_PLUG
 483         rc = notify_change(dchild, mnt, newattrs);
 484 #else
 485         rc = notify_change(dchild, newattrs);
 486 #endif
 487         UNLOCK_INODE_MUTEX(dchild->d_inode);
 488         return rc;
 489 }
 490 EXPORT_SYMBOL(l_notify_change);
 491
 492 /* utility to truncate a file */
 493 int simple_truncate(struct dentry *dir, struct vfsmount *mnt,
 494                     char *name, loff_t length)
 495 {
 496         struct dentry *dchild;
 497         struct iattr newattrs;
 498         int err = 0;
 499         ENTRY;
 500
 501         CDEBUG(D_INODE, "truncating file %.*s to %lld\n", (int)strlen(name),
 502                name, (long long)length);
 503         dchild = ll_lookup_one_len(name, dir, strlen(name));
 504         if (IS_ERR(dchild))
 505                 GOTO(out, err = PTR_ERR(dchild));
 506
 507         if (dchild->d_inode) {
 508                 int old_mode = dchild->d_inode->i_mode;
 509                 if (S_ISDIR(old_mode)) {
 510                         CERROR("found %s (%lu/%u) is mode %o\n", name,
 511                                dchild->d_inode->i_ino,
 512                                dchild->d_inode->i_generation, old_mode);
 513                         GOTO(out_dput, err = -EISDIR);
 514                 }
 515
 516                 newattrs.ia_size = length;
 517                 newattrs.ia_valid = ATTR_SIZE;
 518                 err = l_notify_change(mnt, dchild, &newattrs);
 519         }
 520         EXIT;
 521 out_dput:
 522         dput(dchild);
 523 out:
 524         return err;
 525 }
 526 EXPORT_SYMBOL(simple_truncate);
 527
 528 #ifdef LUSTRE_KERNEL_VERSION
 529 #ifndef HAVE_CLEAR_RDONLY_ON_PUT
 530 #error rdonly patchset must be updated [cfs bz11248]
 531 #endif
 532
 533 void dev_set_rdonly(lvfs_sbdev_type dev);
 534 int dev_check_rdonly(lvfs_sbdev_type dev);
 535
 536 void __lvfs_set_rdonly(lvfs_sbdev_type dev, lvfs_sbdev_type jdev)
 537 {
 538         lvfs_sbdev_sync(dev);
 539         if (jdev && (jdev != dev)) {
 540                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
 541                        (long)jdev);
 542                 dev_set_rdonly(jdev);
 543         }
 544         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
 545         dev_set_rdonly(dev);
 546 }
 547
 548 int lvfs_check_rdonly(lvfs_sbdev_type dev)
 549 {
 550         return dev_check_rdonly(dev);
 551 }
 552
 553 EXPORT_SYMBOL(__lvfs_set_rdonly);
 554 EXPORT_SYMBOL(lvfs_check_rdonly);
 555 #endif /* LUSTRE_KERNEL_VERSION */
 556
 557 int lvfs_check_io_health(struct obd_device *obd, struct file *file)
 558 {
 559         char *write_page = NULL;
 560         loff_t offset = 0;
 561         int rc = 0;
 562         ENTRY;
 563
 564         OBD_ALLOC(write_page, CFS_PAGE_SIZE);
 565         if (!write_page)
 566                 RETURN(-ENOMEM);
 567
 568         rc = fsfilt_write_record(obd, file, write_page, CFS_PAGE_SIZE, &offset, 1);
 569
 570         OBD_FREE(write_page, CFS_PAGE_SIZE);
 571
 572         CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
 573         RETURN(rc);
 574 }
 575 EXPORT_SYMBOL(lvfs_check_io_health);
 576
 577 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 578 MODULE_DESCRIPTION("Lustre VFS Filesystem Helper v0.1");
 579 MODULE_LICENSE("GPL");