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/libcfs.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 __u64 obd_max_pages = 0;
  68 __u64 obd_max_alloc = 0;
  69 struct lprocfs_stats *obd_memory = NULL;
  70 spinlock_t obd_updatemax_lock = SPIN_LOCK_UNLOCKED;
  71 /* refine later and change to seqlock or simlar from libcfs */
  72
  73 /* Debugging check only needed during development */
  74 #ifdef OBD_CTXT_DEBUG
  75 # define ASSERT_CTXT_MAGIC(magic) LASSERT((magic) == OBD_RUN_CTXT_MAGIC)
  76 # define ASSERT_NOT_KERNEL_CTXT(msg) LASSERTF(!segment_eq(get_fs(), get_ds()),\
  77                                               msg)
  78 # define ASSERT_KERNEL_CTXT(msg) LASSERTF(segment_eq(get_fs(), get_ds()), msg)
  79 #else
  80 # define ASSERT_CTXT_MAGIC(magic) do {} while(0)
  81 # define ASSERT_NOT_KERNEL_CTXT(msg) do {} while(0)
  82 # define ASSERT_KERNEL_CTXT(msg) do {} while(0)
  83 #endif
  84
  85 static void push_group_info(struct lvfs_run_ctxt *save,
  86                             struct group_info *ginfo)
  87 {
  88         if (!ginfo) {
  89                 save->ngroups = current_ngroups;
  90                 current_ngroups = 0;
  91         } else {
  92                 task_lock(current);
  93                 save->group_info = current->group_info;
  94                 current->group_info = ginfo;
  95                 task_unlock(current);
  96         }
  97 }
  98
  99 static void pop_group_info(struct lvfs_run_ctxt *save,
 100                            struct group_info *ginfo)
 101 {
 102         if (!ginfo) {
 103                 current_ngroups = save->ngroups;
 104         } else {
 105                 task_lock(current);
 106                 current->group_info = save->group_info;
 107                 task_unlock(current);
 108         }
 109 }
 110
 111 /* push / pop to root of obd store */
 112 void push_ctxt(struct lvfs_run_ctxt *save, struct lvfs_run_ctxt *new_ctx,
 113                struct lvfs_ucred *uc)
 114 {
 115         //ASSERT_NOT_KERNEL_CTXT("already in kernel context!\n");
 116         ASSERT_CTXT_MAGIC(new_ctx->magic);
 117         OBD_SET_CTXT_MAGIC(save);
 118
 119         save->fs = get_fs();
 120         LASSERT(atomic_read(&current->fs->pwd->d_count));
 121         LASSERT(atomic_read(&new_ctx->pwd->d_count));
 122         save->pwd = dget(current->fs->pwd);
 123         save->pwdmnt = mntget(current->fs->pwdmnt);
 124         save->luc.luc_umask = current->fs->umask;
 125         save->ngroups = current->group_info->ngroups;
 126
 127         LASSERT(save->pwd);
 128         LASSERT(save->pwdmnt);
 129         LASSERT(new_ctx->pwd);
 130         LASSERT(new_ctx->pwdmnt);
 131
 132         if (uc) {
 133                 save->luc.luc_uid = current->uid;
 134                 save->luc.luc_gid = current->gid;
 135                 save->luc.luc_fsuid = current->fsuid;
 136                 save->luc.luc_fsgid = current->fsgid;
 137                 save->luc.luc_cap = current->cap_effective;
 138
 139                 current->uid = uc->luc_uid;
 140                 current->gid = uc->luc_gid;
 141                 current->fsuid = uc->luc_fsuid;
 142                 current->fsgid = uc->luc_fsgid;
 143                 current->cap_effective = uc->luc_cap;
 144
 145                 push_group_info(save,
 146                                 uc->luc_ginfo ?:
 147                                 uc->luc_identity ? uc->luc_identity->mi_ginfo :
 148                                                    NULL);
 149         }
 150         current->fs->umask = 0; /* umask already applied on client */
 151         set_fs(new_ctx->fs);
 152         ll_set_fs_pwd(current->fs, new_ctx->pwdmnt, new_ctx->pwd);
 153 }
 154 EXPORT_SYMBOL(push_ctxt);
 155
 156 void pop_ctxt(struct lvfs_run_ctxt *saved, struct lvfs_run_ctxt *new_ctx,
 157               struct lvfs_ucred *uc)
 158 {
 159         ASSERT_CTXT_MAGIC(saved->magic);
 160         ASSERT_KERNEL_CTXT("popping non-kernel context!\n");
 161
 162         LASSERTF(current->fs->pwd == new_ctx->pwd, "%p != %p\n",
 163                  current->fs->pwd, new_ctx->pwd);
 164         LASSERTF(current->fs->pwdmnt == new_ctx->pwdmnt, "%p != %p\n",
 165                  current->fs->pwdmnt, new_ctx->pwdmnt);
 166
 167         set_fs(saved->fs);
 168         ll_set_fs_pwd(current->fs, saved->pwdmnt, saved->pwd);
 169
 170         dput(saved->pwd);
 171         mntput(saved->pwdmnt);
 172         current->fs->umask = saved->luc.luc_umask;
 173         if (uc) {
 174                 current->uid = saved->luc.luc_uid;
 175                 current->gid = saved->luc.luc_gid;
 176                 current->fsuid = saved->luc.luc_fsuid;
 177                 current->fsgid = saved->luc.luc_fsgid;
 178                 current->cap_effective = saved->luc.luc_cap;
 179                 pop_group_info(saved,
 180                                uc->luc_ginfo ?:
 181                                uc->luc_identity ? uc->luc_identity->mi_ginfo :
 182                                                   NULL);
 183         }
 184 }
 185 EXPORT_SYMBOL(pop_ctxt);
 186
 187 /* utility to make a file */
 188 struct dentry *simple_mknod(struct dentry *dir, char *name, int mode, int fix)
 189 {
 190         struct dentry *dchild;
 191         int err = 0;
 192         ENTRY;
 193
 194         // ASSERT_KERNEL_CTXT("kernel doing mknod outside kernel context\n");
 195         CDEBUG(D_INODE, "creating file %.*s\n", (int)strlen(name), name);
 196
 197         dchild = ll_lookup_one_len(name, dir, strlen(name));
 198         if (IS_ERR(dchild))
 199                 GOTO(out_up, dchild);
 200
 201         if (dchild->d_inode) {
 202                 int old_mode = dchild->d_inode->i_mode;
 203                 if (!S_ISREG(old_mode))
 204                         GOTO(out_err, err = -EEXIST);
 205
 206                 /* Fixup file permissions if necessary */
 207                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 208                         CWARN("fixing permissions on %s from %o to %o\n",
 209                               name, old_mode, mode);
 210                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 211                                                   (old_mode & ~S_IALLUGO);
 212                         mark_inode_dirty(dchild->d_inode);
 213                 }
 214                 GOTO(out_up, dchild);
 215         }
 216
 217         err = ll_vfs_create(dir->d_inode, dchild, (mode & ~S_IFMT) | S_IFREG,
 218                             NULL);
 219         if (err)
 220                 GOTO(out_err, err);
 221
 222         RETURN(dchild);
 223
 224 out_err:
 225         dput(dchild);
 226         dchild = ERR_PTR(err);
 227 out_up:
 228         return dchild;
 229 }
 230 EXPORT_SYMBOL(simple_mknod);
 231
 232 /* utility to make a directory */
 233 struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
 234                             const char *name, int mode, int fix)
 235 {
 236         struct dentry *dchild;
 237         int err = 0;
 238         ENTRY;
 239
 240         // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
 241         CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
 242         dchild = ll_lookup_one_len(name, dir, strlen(name));
 243         if (IS_ERR(dchild))
 244                 GOTO(out_up, dchild);
 245
 246         if (dchild->d_inode) {
 247                 int old_mode = dchild->d_inode->i_mode;
 248                 if (!S_ISDIR(old_mode)) {
 249                         CERROR("found %s (%lu/%u) is mode %o\n", name,
 250                                dchild->d_inode->i_ino,
 251                                dchild->d_inode->i_generation, old_mode);
 252                         GOTO(out_err, err = -ENOTDIR);
 253                 }
 254
 255                 /* Fixup directory permissions if necessary */
 256                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 257                         CDEBUG(D_CONFIG,
 258                                "fixing permissions on %s from %o to %o\n",
 259                                name, old_mode, mode);
 260                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 261                                                   (old_mode & ~S_IALLUGO);
 262                         mark_inode_dirty(dchild->d_inode);
 263                 }
 264                 GOTO(out_up, dchild);
 265         }
 266
 267         err = ll_vfs_mkdir(dir->d_inode, dchild, mnt, mode);
 268         if (err)
 269                 GOTO(out_err, err);
 270
 271         RETURN(dchild);
 272
 273 out_err:
 274         dput(dchild);
 275         dchild = ERR_PTR(err);
 276 out_up:
 277         return dchild;
 278 }
 279 EXPORT_SYMBOL(simple_mkdir);
 280
 281 /* utility to rename a file */
 282 int lustre_rename(struct dentry *dir, struct vfsmount *mnt,
 283                   char *oldname, char *newname)
 284 {
 285         struct dentry *dchild_old, *dchild_new;
 286         int err = 0;
 287         ENTRY;
 288
 289         ASSERT_KERNEL_CTXT("kernel doing rename outside kernel context\n");
 290         CDEBUG(D_INODE, "renaming file %.*s to %.*s\n",
 291                (int)strlen(oldname), oldname, (int)strlen(newname), newname);
 292
 293         dchild_old = ll_lookup_one_len(oldname, dir, strlen(oldname));
 294         if (IS_ERR(dchild_old))
 295                 RETURN(PTR_ERR(dchild_old));
 296
 297         if (!dchild_old->d_inode)
 298                 GOTO(put_old, err = -ENOENT);
 299
 300         dchild_new = ll_lookup_one_len(newname, dir, strlen(newname));
 301         if (IS_ERR(dchild_new))
 302                 GOTO(put_old, err = PTR_ERR(dchild_new));
 303
 304         err = ll_vfs_rename(dir->d_inode, dchild_old, mnt,
 305                             dir->d_inode, dchild_new, mnt);
 306
 307         dput(dchild_new);
 308 put_old:
 309         dput(dchild_old);
 310         RETURN(err);
 311 }
 312 EXPORT_SYMBOL(lustre_rename);
 313
 314 /*
 315  * Read a file from within kernel context.  Prior to calling this
 316  * function we should already have done a push_ctxt().
 317  */
 318 int lustre_fread(struct file *file, void *buf, int len, loff_t *off)
 319 {
 320         ASSERT_KERNEL_CTXT("kernel doing read outside kernel context\n");
 321         if (!file || !file->f_op || !file->f_op->read || !off)
 322                 RETURN(-ENOSYS);
 323
 324         return file->f_op->read(file, buf, len, off);
 325 }
 326 EXPORT_SYMBOL(lustre_fread);
 327
 328 /*
 329  * Write a file from within kernel context.  Prior to calling this
 330  * function we should already have done a push_ctxt().
 331  */
 332 int lustre_fwrite(struct file *file, const void *buf, int len, loff_t *off)
 333 {
 334         ENTRY;
 335         ASSERT_KERNEL_CTXT("kernel doing write outside kernel context\n");
 336         if (!file)
 337                 RETURN(-ENOENT);
 338         if (!file->f_op)
 339                 RETURN(-ENOSYS);
 340         if (!off)
 341                 RETURN(-EINVAL);
 342
 343         if (!file->f_op->write)
 344                 RETURN(-EROFS);
 345
 346         RETURN(file->f_op->write(file, buf, len, off));
 347 }
 348 EXPORT_SYMBOL(lustre_fwrite);
 349
 350 /*
 351  * Sync a file from within kernel context.  Prior to calling this
 352  * function we should already have done a push_ctxt().
 353  */
 354 int lustre_fsync(struct file *file)
 355 {
 356         ENTRY;
 357         ASSERT_KERNEL_CTXT("kernel doing sync outside kernel context\n");
 358         if (!file || !file->f_op || !file->f_op->fsync)
 359                 RETURN(-ENOSYS);
 360
 361         RETURN(file->f_op->fsync(file, file->f_dentry, 0));
 362 }
 363 EXPORT_SYMBOL(lustre_fsync);
 364
 365 struct l_file *l_dentry_open(struct lvfs_run_ctxt *ctxt, struct l_dentry *de,
 366                              int flags)
 367 {
 368         mntget(ctxt->pwdmnt);
 369         return dentry_open(de, ctxt->pwdmnt, flags);
 370 }
 371 EXPORT_SYMBOL(l_dentry_open);
 372
 373 #ifdef HAVE_VFS_READDIR_U64_INO
 374 static int l_filldir(void *__buf, const char *name, int namlen, loff_t offset,
 375                      u64 ino, unsigned int d_type)
 376 #else
 377 static int l_filldir(void *__buf, const char *name, int namlen, loff_t offset,
 378                      ino_t ino, unsigned int d_type)
 379 #endif
 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         if (!dirent)
 391                 return -ENOMEM;
 392
 393         list_add_tail(&dirent->lld_list, buf->lrc_list);
 394
 395         buf->lrc_dirent = dirent;
 396         dirent->lld_ino = ino;
 397         LASSERT(sizeof(dirent->lld_name) >= namlen + 1);
 398         memcpy(dirent->lld_name, name, namlen);
 399
 400         return 0;
 401 }
 402
 403 long l_readdir(struct file *file, struct list_head *dentry_list)
 404 {
 405         struct l_linux_dirent *lastdirent;
 406         struct l_readdir_callback buf;
 407         int error;
 408
 409         buf.lrc_dirent = NULL;
 410         buf.lrc_list = dentry_list;
 411
 412         error = vfs_readdir(file, l_filldir, &buf);
 413         if (error < 0)
 414                 return error;
 415
 416         lastdirent = buf.lrc_dirent;
 417         if (lastdirent)
 418                 lastdirent->lld_off = file->f_pos;
 419
 420         return 0;
 421 }
 422 EXPORT_SYMBOL(l_readdir);
 423
 424 int l_notify_change(struct vfsmount *mnt, struct dentry *dchild,
 425                  struct iattr *newattrs)
 426 {
 427         int rc;
 428
 429         LOCK_INODE_MUTEX(dchild->d_inode);
 430 #ifdef HAVE_SECURITY_PLUG
 431         rc = notify_change(dchild, mnt, newattrs);
 432 #else
 433         rc = notify_change(dchild, newattrs);
 434 #endif
 435         UNLOCK_INODE_MUTEX(dchild->d_inode);
 436         return rc;
 437 }
 438 EXPORT_SYMBOL(l_notify_change);
 439
 440 /* utility to truncate a file */
 441 int simple_truncate(struct dentry *dir, struct vfsmount *mnt,
 442                  char *name, loff_t length)
 443 {
 444         struct dentry *dchild;
 445         struct iattr newattrs;
 446         int err = 0;
 447         ENTRY;
 448
 449         CDEBUG(D_INODE, "truncating file %.*s to %lld\n", (int)strlen(name),
 450                name, (long long)length);
 451         dchild = ll_lookup_one_len(name, dir, strlen(name));
 452         if (IS_ERR(dchild))
 453                 GOTO(out, err = PTR_ERR(dchild));
 454
 455         if (dchild->d_inode) {
 456                 int old_mode = dchild->d_inode->i_mode;
 457                 if (S_ISDIR(old_mode)) {
 458                         CERROR("found %s (%lu/%u) is mode %o\n", name,
 459                                dchild->d_inode->i_ino,
 460                                dchild->d_inode->i_generation, old_mode);
 461                         GOTO(out_dput, err = -EISDIR);
 462                 }
 463
 464                 newattrs.ia_size = length;
 465                 newattrs.ia_valid = ATTR_SIZE;
 466                 err = l_notify_change(mnt, dchild, &newattrs);
 467         }
 468         EXIT;
 469 out_dput:
 470         dput(dchild);
 471 out:
 472         return err;
 473 }
 474 EXPORT_SYMBOL(simple_truncate);
 475
 476 #ifdef LUSTRE_KERNEL_VERSION
 477 #ifndef HAVE_CLEAR_RDONLY_ON_PUT
 478 #error rdonly patchset must be updated [cfs bz11248]
 479 #endif
 480 void dev_set_rdonly(lvfs_sbdev_type dev);
 481 int dev_check_rdonly(lvfs_sbdev_type dev);
 482
 483 void __lvfs_set_rdonly(lvfs_sbdev_type dev, lvfs_sbdev_type jdev)
 484 {
 485         if (jdev && (jdev != dev)) {
 486                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
 487                        (long)jdev);
 488                 dev_set_rdonly(jdev);
 489         }
 490         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
 491         dev_set_rdonly(dev);
 492 }
 493
 494 int lvfs_check_rdonly(lvfs_sbdev_type dev)
 495 {
 496         return dev_check_rdonly(dev);
 497 }
 498
 499 EXPORT_SYMBOL(__lvfs_set_rdonly);
 500 EXPORT_SYMBOL(lvfs_check_rdonly);
 501
 502 int lvfs_check_io_health(struct obd_device *obd, struct file *file)
 503 {
 504         char *write_page = NULL;
 505         loff_t offset = 0;
 506         int rc = 0;
 507         ENTRY;
 508
 509         OBD_ALLOC(write_page, CFS_PAGE_SIZE);
 510         if (!write_page)
 511                 RETURN(-ENOMEM);
 512
 513         rc = fsfilt_write_record(obd, file, write_page, CFS_PAGE_SIZE, &offset, 1);
 514
 515         OBD_FREE(write_page, CFS_PAGE_SIZE);
 516
 517         CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
 518         RETURN(rc);
 519 }
 520 EXPORT_SYMBOL(lvfs_check_io_health);
 521 #endif /* LUSTRE_KERNEL_VERSION */
 522
 523 void obd_update_maxusage()
 524 {
 525         __u64 max1, max2;
 526
 527         max1 = obd_pages_sum();
 528         max2 = obd_memory_sum();
 529
 530         spin_lock(&obd_updatemax_lock);
 531         if (max1 > obd_max_pages)
 532                 obd_max_pages = max1;
 533         if (max2 > obd_max_alloc)
 534                 obd_max_alloc = max2;
 535         spin_unlock(&obd_updatemax_lock);
 536
 537 }
 538
 539 __u64 obd_memory_max(void)
 540 {
 541         __u64 ret;
 542
 543         spin_lock(&obd_updatemax_lock);
 544         ret = obd_max_alloc;
 545         spin_unlock(&obd_updatemax_lock);
 546
 547         return ret;
 548 }
 549
 550 __u64 obd_pages_max(void)
 551 {
 552         __u64 ret;
 553
 554         spin_lock(&obd_updatemax_lock);
 555         ret = obd_max_pages;
 556         spin_unlock(&obd_updatemax_lock);
 557
 558         return ret;
 559 }
 560
 561 EXPORT_SYMBOL(obd_update_maxusage);
 562 EXPORT_SYMBOL(obd_pages_max);
 563 EXPORT_SYMBOL(obd_memory_max);
 564 EXPORT_SYMBOL(obd_memory);
 565
 566 #ifdef LPROCFS
 567 __s64 lprocfs_read_helper(struct lprocfs_counter *lc,
 568                           enum lprocfs_fields_flags field)
 569 {
 570         __s64 ret = 0;
 571         int centry;
 572
 573         if (!lc)
 574                 RETURN(0);
 575         do {
 576                 centry = atomic_read(&lc->lc_cntl.la_entry);
 577
 578                 switch (field) {
 579                         case LPROCFS_FIELDS_FLAGS_CONFIG:
 580                                 ret = lc->lc_config;
 581                                 break;
 582                         case LPROCFS_FIELDS_FLAGS_SUM:
 583                                 ret = lc->lc_sum;
 584                                 break;
 585                         case LPROCFS_FIELDS_FLAGS_MIN:
 586                                 ret = lc->lc_min;
 587                                 break;
 588                         case LPROCFS_FIELDS_FLAGS_MAX:
 589                                 ret = lc->lc_max;
 590                                 break;
 591                         case LPROCFS_FIELDS_FLAGS_AVG:
 592                                 ret = (lc->lc_max - lc->lc_min)/2;
 593                                 break;
 594                         case LPROCFS_FIELDS_FLAGS_SUMSQUARE:
 595                                 ret = lc->lc_sumsquare;
 596                                 break;
 597                         case LPROCFS_FIELDS_FLAGS_COUNT:
 598                                 ret = lc->lc_count;
 599                                 break;
 600                         default:
 601                                 break;
 602                 };
 603         } while (centry != atomic_read(&lc->lc_cntl.la_entry) &&
 604                  centry != atomic_read(&lc->lc_cntl.la_exit));
 605
 606         RETURN(ret);
 607 }
 608 EXPORT_SYMBOL(lprocfs_read_helper);
 609 #endif /* LPROCFS */
 610
 611 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 612 MODULE_DESCRIPTION("Lustre VFS Filesystem Helper v0.1");
 613 MODULE_LICENSE("GPL");