lustre/lvfs/lvfs_linux.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  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 #define DEBUG_SUBSYSTEM S_FILTER
  42
  43 #include <linux/version.h>
  44 #include <linux/fs.h>
  45 #include <asm/unistd.h>
  46 #include <linux/slab.h>
  47 #include <linux/pagemap.h>
  48 #include <linux/quotaops.h>
  49 #include <linux/version.h>
  50 #include <libcfs/libcfs.h>
  51 #include <lustre_fsfilt.h>
  52 #include <obd.h>
  53 #include <linux/module.h>
  54 #include <linux/init.h>
  55 #include <linux/lustre_compat25.h>
  56 #include <lvfs.h>
  57
  58 #include <obd.h>
  59 #include <lustre_lib.h>
  60
  61 __u64 obd_max_pages = 0;
  62 __u64 obd_max_alloc = 0;
  63 struct lprocfs_stats *obd_memory = NULL;
  64 EXPORT_SYMBOL(obd_memory);
  65 DEFINE_SPINLOCK(obd_updatemax_lock);
  66 /* refine later and change to seqlock or simlar from libcfs */
  67
  68 /* Debugging check only needed during development */
  69 #ifdef OBD_CTXT_DEBUG
  70 # define ASSERT_CTXT_MAGIC(magic) LASSERT((magic) == OBD_RUN_CTXT_MAGIC)
  71 # define ASSERT_NOT_KERNEL_CTXT(msg) LASSERTF(!segment_eq(get_fs(), get_ds()),\
  72                                               msg)
  73 # define ASSERT_KERNEL_CTXT(msg) LASSERTF(segment_eq(get_fs(), get_ds()), msg)
  74 #else
  75 # define ASSERT_CTXT_MAGIC(magic) do {} while(0)
  76 # define ASSERT_NOT_KERNEL_CTXT(msg) do {} while(0)
  77 # define ASSERT_KERNEL_CTXT(msg) do {} while(0)
  78 #endif
  79
  80 static void push_group_info(struct lvfs_run_ctxt *save,
  81                             struct group_info *ginfo)
  82 {
  83         if (!ginfo) {
  84                 save->ngroups = current_ngroups;
  85                 current_ngroups = 0;
  86         } else {
  87                 struct cred *cred;
  88                 task_lock(current);
  89                 save->group_info = current_cred()->group_info;
  90                 if ((cred = prepare_creds())) {
  91                         cred->group_info = ginfo;
  92                         commit_creds(cred);
  93                 }
  94                 task_unlock(current);
  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                 struct cred *cred;
 105                 task_lock(current);
 106                 if ((cred = prepare_creds())) {
 107                         cred->group_info = save->group_info;
 108                         commit_creds(cred);
 109                 }
 110                 task_unlock(current);
 111         }
 112 }
 113
 114 /* push / pop to root of obd store */
 115 void push_ctxt(struct lvfs_run_ctxt *save, struct lvfs_run_ctxt *new_ctx,
 116                struct lvfs_ucred *uc)
 117 {
 118         /* if there is underlaying dt_device then push_ctxt is not needed */
 119         if (new_ctx->dt != NULL)
 120                 return;
 121
 122         //ASSERT_NOT_KERNEL_CTXT("already in kernel context!\n");
 123         ASSERT_CTXT_MAGIC(new_ctx->magic);
 124         OBD_SET_CTXT_MAGIC(save);
 125
 126         save->fs = get_fs();
 127         LASSERT(d_refcount(cfs_fs_pwd(current->fs)));
 128         LASSERT(d_refcount(new_ctx->pwd));
 129         save->pwd = dget(cfs_fs_pwd(current->fs));
 130         save->pwdmnt = mntget(cfs_fs_mnt(current->fs));
 131         save->luc.luc_umask = cfs_curproc_umask();
 132         save->ngroups = current_cred()->group_info->ngroups;
 133
 134         LASSERT(save->pwd);
 135         LASSERT(save->pwdmnt);
 136         LASSERT(new_ctx->pwd);
 137         LASSERT(new_ctx->pwdmnt);
 138
 139         if (uc) {
 140                 struct cred *cred;
 141                 save->luc.luc_uid = current_uid();
 142                 save->luc.luc_gid = current_gid();
 143                 save->luc.luc_fsuid = current_fsuid();
 144                 save->luc.luc_fsgid = current_fsgid();
 145                 save->luc.luc_cap = current_cap();
 146
 147                 if ((cred = prepare_creds())) {
 148                         cred->uid = uc->luc_uid;
 149                         cred->gid = uc->luc_gid;
 150                         cred->fsuid = uc->luc_fsuid;
 151                         cred->fsgid = uc->luc_fsgid;
 152                         cred->cap_effective = uc->luc_cap;
 153                         commit_creds(cred);
 154                 }
 155
 156                 push_group_info(save,
 157                                 uc->luc_ginfo ?:
 158                                 uc->luc_identity ? uc->luc_identity->mi_ginfo :
 159                                                    NULL);
 160         }
 161         current->fs->umask = 0; /* umask already applied on client */
 162         set_fs(new_ctx->fs);
 163         ll_set_fs_pwd(current->fs, new_ctx->pwdmnt, new_ctx->pwd);
 164 }
 165 EXPORT_SYMBOL(push_ctxt);
 166
 167 void pop_ctxt(struct lvfs_run_ctxt *saved, struct lvfs_run_ctxt *new_ctx,
 168               struct lvfs_ucred *uc)
 169 {
 170         /* if there is underlaying dt_device then pop_ctxt is not needed */
 171         if (new_ctx->dt != NULL)
 172                 return;
 173
 174         ASSERT_CTXT_MAGIC(saved->magic);
 175         ASSERT_KERNEL_CTXT("popping non-kernel context!\n");
 176
 177         LASSERTF(cfs_fs_pwd(current->fs) == new_ctx->pwd, "%p != %p\n",
 178                  cfs_fs_pwd(current->fs), new_ctx->pwd);
 179         LASSERTF(cfs_fs_mnt(current->fs) == new_ctx->pwdmnt, "%p != %p\n",
 180                  cfs_fs_mnt(current->fs), new_ctx->pwdmnt);
 181
 182         set_fs(saved->fs);
 183         ll_set_fs_pwd(current->fs, saved->pwdmnt, saved->pwd);
 184
 185         dput(saved->pwd);
 186         mntput(saved->pwdmnt);
 187         current->fs->umask = saved->luc.luc_umask;
 188         if (uc) {
 189                 struct cred *cred;
 190                 if ((cred = prepare_creds())) {
 191                         cred->uid = saved->luc.luc_uid;
 192                         cred->gid = saved->luc.luc_gid;
 193                         cred->fsuid = saved->luc.luc_fsuid;
 194                         cred->fsgid = saved->luc.luc_fsgid;
 195                         cred->cap_effective = saved->luc.luc_cap;
 196                         commit_creds(cred);
 197                 }
 198
 199                 pop_group_info(saved,
 200                                uc->luc_ginfo ?:
 201                                uc->luc_identity ? uc->luc_identity->mi_ginfo :
 202                                                   NULL);
 203         }
 204 }
 205 EXPORT_SYMBOL(pop_ctxt);
 206
 207 /* utility to make a directory */
 208 struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
 209                             const char *name, int mode, int fix)
 210 {
 211         struct dentry *dchild;
 212         int err = 0;
 213         ENTRY;
 214
 215         // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
 216         CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
 217         dchild = ll_lookup_one_len(name, dir, strlen(name));
 218         if (IS_ERR(dchild))
 219                 GOTO(out_up, dchild);
 220
 221         if (dchild->d_inode) {
 222                 int old_mode = dchild->d_inode->i_mode;
 223                 if (!S_ISDIR(old_mode)) {
 224                         CERROR("found %s (%lu/%u) is mode %o\n", name,
 225                                dchild->d_inode->i_ino,
 226                                dchild->d_inode->i_generation, old_mode);
 227                         GOTO(out_err, err = -ENOTDIR);
 228                 }
 229
 230                 /* Fixup directory permissions if necessary */
 231                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
 232                         CDEBUG(D_CONFIG,
 233                                "fixing permissions on %s from %o to %o\n",
 234                                name, old_mode, mode);
 235                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
 236                                                   (old_mode & ~S_IALLUGO);
 237                         mark_inode_dirty(dchild->d_inode);
 238                 }
 239                 GOTO(out_up, dchild);
 240         }
 241
 242         err = ll_vfs_mkdir(dir->d_inode, dchild, mnt, mode);
 243         if (err)
 244                 GOTO(out_err, err);
 245
 246         RETURN(dchild);
 247
 248 out_err:
 249         dput(dchild);
 250         dchild = ERR_PTR(err);
 251 out_up:
 252         return dchild;
 253 }
 254 EXPORT_SYMBOL(simple_mkdir);
 255
 256 /* utility to rename a file */
 257 int lustre_rename(struct dentry *dir, struct vfsmount *mnt,
 258                   char *oldname, char *newname)
 259 {
 260         struct dentry *dchild_old, *dchild_new;
 261         int err = 0;
 262         ENTRY;
 263
 264         ASSERT_KERNEL_CTXT("kernel doing rename outside kernel context\n");
 265         CDEBUG(D_INODE, "renaming file %.*s to %.*s\n",
 266                (int)strlen(oldname), oldname, (int)strlen(newname), newname);
 267
 268         dchild_old = ll_lookup_one_len(oldname, dir, strlen(oldname));
 269         if (IS_ERR(dchild_old))
 270                 RETURN(PTR_ERR(dchild_old));
 271
 272         if (!dchild_old->d_inode)
 273                 GOTO(put_old, err = -ENOENT);
 274
 275         dchild_new = ll_lookup_one_len(newname, dir, strlen(newname));
 276         if (IS_ERR(dchild_new))
 277                 GOTO(put_old, err = PTR_ERR(dchild_new));
 278
 279         err = ll_vfs_rename(dir->d_inode, dchild_old, mnt,
 280                             dir->d_inode, dchild_new, mnt);
 281
 282         dput(dchild_new);
 283 put_old:
 284         dput(dchild_old);
 285         RETURN(err);
 286 }
 287 EXPORT_SYMBOL(lustre_rename);
 288
 289 /* Note: dput(dchild) will be called if there is an error */
 290 struct l_file *l_dentry_open(struct lvfs_run_ctxt *ctxt, struct l_dentry *de,
 291                              int flags)
 292 {
 293         mntget(ctxt->pwdmnt);
 294         return ll_dentry_open(de, ctxt->pwdmnt, flags, current_cred());
 295 }
 296 EXPORT_SYMBOL(l_dentry_open);
 297
 298 int __lvfs_set_rdonly(lvfs_sbdev_type dev, lvfs_sbdev_type jdev)
 299 {
 300 #ifdef HAVE_DEV_SET_RDONLY
 301         if (jdev && (jdev != dev)) {
 302                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
 303                        (long)jdev);
 304                 dev_set_rdonly(jdev);
 305         }
 306         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
 307         dev_set_rdonly(dev);
 308
 309         return 0;
 310 #else
 311         CERROR("DEV %lx CANNOT BE SET READONLY\n", (long)dev);
 312
 313         return -EOPNOTSUPP;
 314 #endif
 315 }
 316 EXPORT_SYMBOL(__lvfs_set_rdonly);
 317
 318 int lvfs_check_rdonly(lvfs_sbdev_type dev)
 319 {
 320 #ifdef HAVE_DEV_SET_RDONLY
 321         return dev_check_rdonly(dev);
 322 #else
 323         return 0;
 324 #endif
 325 }
 326 EXPORT_SYMBOL(lvfs_check_rdonly);
 327
 328 void obd_update_maxusage()
 329 {
 330         __u64 max1, max2;
 331
 332         max1 = obd_pages_sum();
 333         max2 = obd_memory_sum();
 334
 335         spin_lock(&obd_updatemax_lock);
 336         if (max1 > obd_max_pages)
 337                 obd_max_pages = max1;
 338         if (max2 > obd_max_alloc)
 339                 obd_max_alloc = max2;
 340         spin_unlock(&obd_updatemax_lock);
 341
 342 }
 343 EXPORT_SYMBOL(obd_update_maxusage);
 344
 345 __u64 obd_memory_max(void)
 346 {
 347         __u64 ret;
 348
 349         spin_lock(&obd_updatemax_lock);
 350         ret = obd_max_alloc;
 351         spin_unlock(&obd_updatemax_lock);
 352
 353         return ret;
 354 }
 355 EXPORT_SYMBOL(obd_memory_max);
 356
 357 __u64 obd_pages_max(void)
 358 {
 359         __u64 ret;
 360
 361         spin_lock(&obd_updatemax_lock);
 362         ret = obd_max_pages;
 363         spin_unlock(&obd_updatemax_lock);
 364
 365         return ret;
 366 }
 367 EXPORT_SYMBOL(obd_pages_max);
 368
 369 #ifdef LPROCFS
 370 __s64 lprocfs_read_helper(struct lprocfs_counter *lc,
 371                           struct lprocfs_counter_header *header,
 372                           enum lprocfs_stats_flags flags,
 373                           enum lprocfs_fields_flags field)
 374 {
 375         __s64 ret = 0;
 376
 377         if (lc == NULL || header == NULL)
 378                 RETURN(0);
 379
 380         switch (field) {
 381                 case LPROCFS_FIELDS_FLAGS_CONFIG:
 382                         ret = header->lc_config;
 383                         break;
 384                 case LPROCFS_FIELDS_FLAGS_SUM:
 385                         ret = lc->lc_sum;
 386                         if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
 387                                 ret += lc->lc_sum_irq;
 388                         break;
 389                 case LPROCFS_FIELDS_FLAGS_MIN:
 390                         ret = lc->lc_min;
 391                         break;
 392                 case LPROCFS_FIELDS_FLAGS_MAX:
 393                         ret = lc->lc_max;
 394                         break;
 395                 case LPROCFS_FIELDS_FLAGS_AVG:
 396                         ret = (lc->lc_max - lc->lc_min) / 2;
 397                         break;
 398                 case LPROCFS_FIELDS_FLAGS_SUMSQUARE:
 399                         ret = lc->lc_sumsquare;
 400                         break;
 401                 case LPROCFS_FIELDS_FLAGS_COUNT:
 402                         ret = lc->lc_count;
 403                         break;
 404                 default:
 405                         break;
 406         };
 407
 408         RETURN(ret);
 409 }
 410 EXPORT_SYMBOL(lprocfs_read_helper);
 411 #endif /* LPROCFS */
 412
 413 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
 414 MODULE_DESCRIPTION("Lustre VFS Filesystem Helper v0.1");
 415 MODULE_LICENSE("GPL");