fs/fuse/inode.c

   1 /*
   2   FUSE: Filesystem in Userspace
   3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
   4
   5   This program can be distributed under the terms of the GNU GPL.
   6   See the file COPYING.
   7 */
   8
   9 #include "fuse_i.h"
  10
  11 #include <linux/pagemap.h>
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/seq_file.h>
  15 #include <linux/init.h>
  16 #include <linux/module.h>
  17 #include <linux/moduleparam.h>
  18 #include <linux/parser.h>
  19 #include <linux/statfs.h>
  20 #include <linux/random.h>
  21 #include <linux/sched.h>
  22 #include <linux/exportfs.h>
  23
  24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  25 MODULE_DESCRIPTION("Filesystem in Userspace");
  26 MODULE_LICENSE("GPL");
  27
  28 static struct kmem_cache *fuse_inode_cachep;
  29 struct list_head fuse_conn_list;
  30 DEFINE_MUTEX(fuse_mutex);
  31
  32 static int set_global_limit(const char *val, struct kernel_param *kp);
  33
  34 unsigned max_user_bgreq;
  35 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
  36                   &max_user_bgreq, 0644);
  37 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
  38 MODULE_PARM_DESC(max_user_bgreq,
  39  "Global limit for the maximum number of backgrounded requests an "
  40  "unprivileged user can set");
  41
  42 unsigned max_user_congthresh;
  43 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
  44                   &max_user_congthresh, 0644);
  45 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
  46 MODULE_PARM_DESC(max_user_congthresh,
  47  "Global limit for the maximum congestion threshold an "
  48  "unprivileged user can set");
  49
  50 #define FUSE_SUPER_MAGIC 0x65735546
  51
  52 #define FUSE_DEFAULT_BLKSIZE 512
  53
  54 /** Maximum number of outstanding background requests */
  55 #define FUSE_DEFAULT_MAX_BACKGROUND 12
  56
  57 /** Congestion starts at 75% of maximum */
  58 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
  59
  60 struct fuse_mount_data {
  61         int fd;
  62         unsigned rootmode;
  63         unsigned user_id;
  64         unsigned group_id;
  65         unsigned fd_present:1;
  66         unsigned rootmode_present:1;
  67         unsigned user_id_present:1;
  68         unsigned group_id_present:1;
  69         unsigned flags;
  70         unsigned max_read;
  71         unsigned blksize;
  72 };
  73
  74 static struct inode *fuse_alloc_inode(struct super_block *sb)
  75 {
  76         struct inode *inode;
  77         struct fuse_inode *fi;
  78
  79         inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
  80         if (!inode)
  81                 return NULL;
  82
  83         fi = get_fuse_inode(inode);
  84         fi->i_time = 0;
  85         fi->nodeid = 0;
  86         fi->nlookup = 0;
  87         fi->attr_version = 0;
  88         fi->writectr = 0;
  89         INIT_LIST_HEAD(&fi->write_files);
  90         INIT_LIST_HEAD(&fi->queued_writes);
  91         INIT_LIST_HEAD(&fi->writepages);
  92         init_waitqueue_head(&fi->page_waitq);
  93         fi->forget_req = fuse_request_alloc();
  94         if (!fi->forget_req) {
  95                 kmem_cache_free(fuse_inode_cachep, inode);
  96                 return NULL;
  97         }
  98
  99         return inode;
 100 }
 101
 102 static void fuse_destroy_inode(struct inode *inode)
 103 {
 104         struct fuse_inode *fi = get_fuse_inode(inode);
 105         BUG_ON(!list_empty(&fi->write_files));
 106         BUG_ON(!list_empty(&fi->queued_writes));
 107         if (fi->forget_req)
 108                 fuse_request_free(fi->forget_req);
 109         kmem_cache_free(fuse_inode_cachep, inode);
 110 }
 111
 112 void fuse_send_forget(struct fuse_conn *fc, struct fuse_req *req,
 113                       u64 nodeid, u64 nlookup)
 114 {
 115         struct fuse_forget_in *inarg = &req->misc.forget_in;
 116         inarg->nlookup = nlookup;
 117         req->in.h.opcode = FUSE_FORGET;
 118         req->in.h.nodeid = nodeid;
 119         req->in.numargs = 1;
 120         req->in.args[0].size = sizeof(struct fuse_forget_in);
 121         req->in.args[0].value = inarg;
 122         fuse_request_send_noreply(fc, req);
 123 }
 124
 125 static void fuse_clear_inode(struct inode *inode)
 126 {
 127         if (inode->i_sb->s_flags & MS_ACTIVE) {
 128                 struct fuse_conn *fc = get_fuse_conn(inode);
 129                 struct fuse_inode *fi = get_fuse_inode(inode);
 130                 fuse_send_forget(fc, fi->forget_req, fi->nodeid, fi->nlookup);
 131                 fi->forget_req = NULL;
 132         }
 133 }
 134
 135 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
 136 {
 137         if (*flags & MS_MANDLOCK)
 138                 return -EINVAL;
 139
 140         return 0;
 141 }
 142
 143 void fuse_truncate(struct address_space *mapping, loff_t offset)
 144 {
 145         /* See vmtruncate() */
 146         unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
 147         truncate_inode_pages(mapping, offset);
 148         unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
 149 }
 150
 151 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 152                                    u64 attr_valid)
 153 {
 154         struct fuse_conn *fc = get_fuse_conn(inode);
 155         struct fuse_inode *fi = get_fuse_inode(inode);
 156
 157         fi->attr_version = ++fc->attr_version;
 158         fi->i_time = attr_valid;
 159
 160         inode->i_ino     = attr->ino;
 161         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 162         inode->i_nlink   = attr->nlink;
 163         inode->i_uid     = attr->uid;
 164         inode->i_gid     = attr->gid;
 165         inode->i_blocks  = attr->blocks;
 166         inode->i_atime.tv_sec   = attr->atime;
 167         inode->i_atime.tv_nsec  = attr->atimensec;
 168         inode->i_mtime.tv_sec   = attr->mtime;
 169         inode->i_mtime.tv_nsec  = attr->mtimensec;
 170         inode->i_ctime.tv_sec   = attr->ctime;
 171         inode->i_ctime.tv_nsec  = attr->ctimensec;
 172
 173         if (attr->blksize != 0)
 174                 inode->i_blkbits = ilog2(attr->blksize);
 175         else
 176                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 177
 178         /*
 179          * Don't set the sticky bit in i_mode, unless we want the VFS
 180          * to check permissions.  This prevents failures due to the
 181          * check in may_delete().
 182          */
 183         fi->orig_i_mode = inode->i_mode;
 184         if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
 185                 inode->i_mode &= ~S_ISVTX;
 186 }
 187
 188 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 189                             u64 attr_valid, u64 attr_version)
 190 {
 191         struct fuse_conn *fc = get_fuse_conn(inode);
 192         struct fuse_inode *fi = get_fuse_inode(inode);
 193         loff_t oldsize;
 194
 195         spin_lock(&fc->lock);
 196         if (attr_version != 0 && fi->attr_version > attr_version) {
 197                 spin_unlock(&fc->lock);
 198                 return;
 199         }
 200
 201         fuse_change_attributes_common(inode, attr, attr_valid);
 202
 203         oldsize = inode->i_size;
 204         i_size_write(inode, attr->size);
 205         spin_unlock(&fc->lock);
 206
 207         if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
 208                 if (attr->size < oldsize)
 209                         fuse_truncate(inode->i_mapping, attr->size);
 210                 invalidate_inode_pages2(inode->i_mapping);
 211         }
 212 }
 213
 214 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 215 {
 216         inode->i_mode = attr->mode & S_IFMT;
 217         inode->i_size = attr->size;
 218         if (S_ISREG(inode->i_mode)) {
 219                 fuse_init_common(inode);
 220                 fuse_init_file_inode(inode);
 221         } else if (S_ISDIR(inode->i_mode))
 222                 fuse_init_dir(inode);
 223         else if (S_ISLNK(inode->i_mode))
 224                 fuse_init_symlink(inode);
 225         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 226                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 227                 fuse_init_common(inode);
 228                 init_special_inode(inode, inode->i_mode,
 229                                    new_decode_dev(attr->rdev));
 230         } else
 231                 BUG();
 232 }
 233
 234 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 235 {
 236         u64 nodeid = *(u64 *) _nodeidp;
 237         if (get_node_id(inode) == nodeid)
 238                 return 1;
 239         else
 240                 return 0;
 241 }
 242
 243 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 244 {
 245         u64 nodeid = *(u64 *) _nodeidp;
 246         get_fuse_inode(inode)->nodeid = nodeid;
 247         return 0;
 248 }
 249
 250 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 251                         int generation, struct fuse_attr *attr,
 252                         u64 attr_valid, u64 attr_version)
 253 {
 254         struct inode *inode;
 255         struct fuse_inode *fi;
 256         struct fuse_conn *fc = get_fuse_conn_super(sb);
 257
 258  retry:
 259         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 260         if (!inode)
 261                 return NULL;
 262
 263         if ((inode->i_state & I_NEW)) {
 264                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
 265                 inode->i_generation = generation;
 266                 inode->i_data.backing_dev_info = &fc->bdi;
 267                 fuse_init_inode(inode, attr);
 268                 unlock_new_inode(inode);
 269         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 270                 /* Inode has changed type, any I/O on the old should fail */
 271                 make_bad_inode(inode);
 272                 iput(inode);
 273                 goto retry;
 274         }
 275
 276         fi = get_fuse_inode(inode);
 277         spin_lock(&fc->lock);
 278         fi->nlookup++;
 279         spin_unlock(&fc->lock);
 280         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 281
 282         return inode;
 283 }
 284
 285 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 286                              loff_t offset, loff_t len)
 287 {
 288         struct inode *inode;
 289         pgoff_t pg_start;
 290         pgoff_t pg_end;
 291
 292         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 293         if (!inode)
 294                 return -ENOENT;
 295
 296         fuse_invalidate_attr(inode);
 297         if (offset >= 0) {
 298                 pg_start = offset >> PAGE_CACHE_SHIFT;
 299                 if (len <= 0)
 300                         pg_end = -1;
 301                 else
 302                         pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
 303                 invalidate_inode_pages2_range(inode->i_mapping,
 304                                               pg_start, pg_end);
 305         }
 306         iput(inode);
 307         return 0;
 308 }
 309
 310 static void fuse_umount_begin(struct super_block *sb)
 311 {
 312         fuse_abort_conn(get_fuse_conn_super(sb));
 313 }
 314
 315 static void fuse_send_destroy(struct fuse_conn *fc)
 316 {
 317         struct fuse_req *req = fc->destroy_req;
 318         if (req && fc->conn_init) {
 319                 fc->destroy_req = NULL;
 320                 req->in.h.opcode = FUSE_DESTROY;
 321                 req->force = 1;
 322                 fuse_request_send(fc, req);
 323                 fuse_put_request(fc, req);
 324         }
 325 }
 326
 327 static void fuse_bdi_destroy(struct fuse_conn *fc)
 328 {
 329         if (fc->bdi_initialized)
 330                 bdi_destroy(&fc->bdi);
 331 }
 332
 333 void fuse_conn_kill(struct fuse_conn *fc)
 334 {
 335         spin_lock(&fc->lock);
 336         fc->connected = 0;
 337         fc->blocked = 0;
 338         spin_unlock(&fc->lock);
 339         /* Flush all readers on this fs */
 340         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
 341         wake_up_all(&fc->waitq);
 342         wake_up_all(&fc->blocked_waitq);
 343         wake_up_all(&fc->reserved_req_waitq);
 344         mutex_lock(&fuse_mutex);
 345         list_del(&fc->entry);
 346         fuse_ctl_remove_conn(fc);
 347         mutex_unlock(&fuse_mutex);
 348         fuse_bdi_destroy(fc);
 349 }
 350 EXPORT_SYMBOL_GPL(fuse_conn_kill);
 351
 352 static void fuse_put_super(struct super_block *sb)
 353 {
 354         struct fuse_conn *fc = get_fuse_conn_super(sb);
 355
 356         fuse_send_destroy(fc);
 357         fuse_conn_kill(fc);
 358         fuse_conn_put(fc);
 359 }
 360
 361 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 362 {
 363         stbuf->f_type    = FUSE_SUPER_MAGIC;
 364         stbuf->f_bsize   = attr->bsize;
 365         stbuf->f_frsize  = attr->frsize;
 366         stbuf->f_blocks  = attr->blocks;
 367         stbuf->f_bfree   = attr->bfree;
 368         stbuf->f_bavail  = attr->bavail;
 369         stbuf->f_files   = attr->files;
 370         stbuf->f_ffree   = attr->ffree;
 371         stbuf->f_namelen = attr->namelen;
 372         /* fsid is left zero */
 373 }
 374
 375 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 376 {
 377         struct super_block *sb = dentry->d_sb;
 378         struct fuse_conn *fc = get_fuse_conn_super(sb);
 379         struct fuse_req *req;
 380         struct fuse_statfs_out outarg;
 381         int err;
 382
 383         if (!fuse_allow_task(fc, current)) {
 384                 buf->f_type = FUSE_SUPER_MAGIC;
 385                 return 0;
 386         }
 387
 388         req = fuse_get_req(fc);
 389         if (IS_ERR(req))
 390                 return PTR_ERR(req);
 391
 392         memset(&outarg, 0, sizeof(outarg));
 393         req->in.numargs = 0;
 394         req->in.h.opcode = FUSE_STATFS;
 395         req->in.h.nodeid = get_node_id(dentry->d_inode);
 396         req->out.numargs = 1;
 397         req->out.args[0].size =
 398                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
 399         req->out.args[0].value = &outarg;
 400         fuse_request_send(fc, req);
 401         err = req->out.h.error;
 402         if (!err)
 403                 convert_fuse_statfs(buf, &outarg.st);
 404         fuse_put_request(fc, req);
 405         return err;
 406 }
 407
 408 enum {
 409         OPT_FD,
 410         OPT_ROOTMODE,
 411         OPT_USER_ID,
 412         OPT_GROUP_ID,
 413         OPT_DEFAULT_PERMISSIONS,
 414         OPT_ALLOW_OTHER,
 415         OPT_MAX_READ,
 416         OPT_BLKSIZE,
 417         OPT_ERR
 418 };
 419
 420 static const match_table_t tokens = {
 421         {OPT_FD,                        "fd=%u"},
 422         {OPT_ROOTMODE,                  "rootmode=%o"},
 423         {OPT_USER_ID,                   "user_id=%u"},
 424         {OPT_GROUP_ID,                  "group_id=%u"},
 425         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 426         {OPT_ALLOW_OTHER,               "allow_other"},
 427         {OPT_MAX_READ,                  "max_read=%u"},
 428         {OPT_BLKSIZE,                   "blksize=%u"},
 429         {OPT_ERR,                       NULL}
 430 };
 431
 432 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
 433 {
 434         char *p;
 435         memset(d, 0, sizeof(struct fuse_mount_data));
 436         d->max_read = ~0;
 437         d->blksize = FUSE_DEFAULT_BLKSIZE;
 438
 439         while ((p = strsep(&opt, ",")) != NULL) {
 440                 int token;
 441                 int value;
 442                 substring_t args[MAX_OPT_ARGS];
 443                 if (!*p)
 444                         continue;
 445
 446                 token = match_token(p, tokens, args);
 447                 switch (token) {
 448                 case OPT_FD:
 449                         if (match_int(&args[0], &value))
 450                                 return 0;
 451                         d->fd = value;
 452                         d->fd_present = 1;
 453                         break;
 454
 455                 case OPT_ROOTMODE:
 456                         if (match_octal(&args[0], &value))
 457                                 return 0;
 458                         if (!fuse_valid_type(value))
 459                                 return 0;
 460                         d->rootmode = value;
 461                         d->rootmode_present = 1;
 462                         break;
 463
 464                 case OPT_USER_ID:
 465                         if (match_int(&args[0], &value))
 466                                 return 0;
 467                         d->user_id = value;
 468                         d->user_id_present = 1;
 469                         break;
 470
 471                 case OPT_GROUP_ID:
 472                         if (match_int(&args[0], &value))
 473                                 return 0;
 474                         d->group_id = value;
 475                         d->group_id_present = 1;
 476                         break;
 477
 478                 case OPT_DEFAULT_PERMISSIONS:
 479                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
 480                         break;
 481
 482                 case OPT_ALLOW_OTHER:
 483                         d->flags |= FUSE_ALLOW_OTHER;
 484                         break;
 485
 486                 case OPT_MAX_READ:
 487                         if (match_int(&args[0], &value))
 488                                 return 0;
 489                         d->max_read = value;
 490                         break;
 491
 492                 case OPT_BLKSIZE:
 493                         if (!is_bdev || match_int(&args[0], &value))
 494                                 return 0;
 495                         d->blksize = value;
 496                         break;
 497
 498                 default:
 499                         return 0;
 500                 }
 501         }
 502
 503         if (!d->fd_present || !d->rootmode_present ||
 504             !d->user_id_present || !d->group_id_present)
 505                 return 0;
 506
 507         return 1;
 508 }
 509
 510 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
 511 {
 512         struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
 513
 514         seq_printf(m, ",user_id=%u", fc->user_id);
 515         seq_printf(m, ",group_id=%u", fc->group_id);
 516         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
 517                 seq_puts(m, ",default_permissions");
 518         if (fc->flags & FUSE_ALLOW_OTHER)
 519                 seq_puts(m, ",allow_other");
 520         if (fc->max_read != ~0)
 521                 seq_printf(m, ",max_read=%u", fc->max_read);
 522         if (mnt->mnt_sb->s_bdev &&
 523             mnt->mnt_sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 524                 seq_printf(m, ",blksize=%lu", mnt->mnt_sb->s_blocksize);
 525         return 0;
 526 }
 527
 528 void fuse_conn_init(struct fuse_conn *fc)
 529 {
 530         memset(fc, 0, sizeof(*fc));
 531         spin_lock_init(&fc->lock);
 532         mutex_init(&fc->inst_mutex);
 533         init_rwsem(&fc->killsb);
 534         atomic_set(&fc->count, 1);
 535         init_waitqueue_head(&fc->waitq);
 536         init_waitqueue_head(&fc->blocked_waitq);
 537         init_waitqueue_head(&fc->reserved_req_waitq);
 538         INIT_LIST_HEAD(&fc->pending);
 539         INIT_LIST_HEAD(&fc->processing);
 540         INIT_LIST_HEAD(&fc->io);
 541         INIT_LIST_HEAD(&fc->interrupts);
 542         INIT_LIST_HEAD(&fc->bg_queue);
 543         INIT_LIST_HEAD(&fc->entry);
 544         atomic_set(&fc->num_waiting, 0);
 545         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 546         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 547         fc->khctr = 0;
 548         fc->polled_files = RB_ROOT;
 549         fc->reqctr = 0;
 550         fc->blocked = 1;
 551         fc->attr_version = 1;
 552         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 553 }
 554 EXPORT_SYMBOL_GPL(fuse_conn_init);
 555
 556 void fuse_conn_put(struct fuse_conn *fc)
 557 {
 558         if (atomic_dec_and_test(&fc->count)) {
 559                 if (fc->destroy_req)
 560                         fuse_request_free(fc->destroy_req);
 561                 mutex_destroy(&fc->inst_mutex);
 562                 fc->release(fc);
 563         }
 564 }
 565 EXPORT_SYMBOL_GPL(fuse_conn_put);
 566
 567 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 568 {
 569         atomic_inc(&fc->count);
 570         return fc;
 571 }
 572 EXPORT_SYMBOL_GPL(fuse_conn_get);
 573
 574 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 575 {
 576         struct fuse_attr attr;
 577         memset(&attr, 0, sizeof(attr));
 578
 579         attr.mode = mode;
 580         attr.ino = FUSE_ROOT_ID;
 581         attr.nlink = 1;
 582         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 583 }
 584
 585 struct fuse_inode_handle {
 586         u64 nodeid;
 587         u32 generation;
 588 };
 589
 590 static struct dentry *fuse_get_dentry(struct super_block *sb,
 591                                       struct fuse_inode_handle *handle)
 592 {
 593         struct fuse_conn *fc = get_fuse_conn_super(sb);
 594         struct inode *inode;
 595         struct dentry *entry;
 596         int err = -ESTALE;
 597
 598         if (handle->nodeid == 0)
 599                 goto out_err;
 600
 601         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 602         if (!inode) {
 603                 struct fuse_entry_out outarg;
 604                 struct qstr name;
 605
 606                 if (!fc->export_support)
 607                         goto out_err;
 608
 609                 name.len = 1;
 610                 name.name = ".";
 611                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 612                                        &inode);
 613                 if (err && err != -ENOENT)
 614                         goto out_err;
 615                 if (err || !inode) {
 616                         err = -ESTALE;
 617                         goto out_err;
 618                 }
 619                 err = -EIO;
 620                 if (get_node_id(inode) != handle->nodeid)
 621                         goto out_iput;
 622         }
 623         err = -ESTALE;
 624         if (inode->i_generation != handle->generation)
 625                 goto out_iput;
 626
 627         entry = d_obtain_alias(inode);
 628         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID) {
 629                 entry->d_op = &fuse_dentry_operations;
 630                 fuse_invalidate_entry_cache(entry);
 631         }
 632
 633         return entry;
 634
 635  out_iput:
 636         iput(inode);
 637  out_err:
 638         return ERR_PTR(err);
 639 }
 640
 641 static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
 642                            int connectable)
 643 {
 644         struct inode *inode = dentry->d_inode;
 645         bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
 646         int len = encode_parent ? 6 : 3;
 647         u64 nodeid;
 648         u32 generation;
 649
 650         if (*max_len < len)
 651                 return  255;
 652
 653         nodeid = get_fuse_inode(inode)->nodeid;
 654         generation = inode->i_generation;
 655
 656         fh[0] = (u32)(nodeid >> 32);
 657         fh[1] = (u32)(nodeid & 0xffffffff);
 658         fh[2] = generation;
 659
 660         if (encode_parent) {
 661                 struct inode *parent;
 662
 663                 spin_lock(&dentry->d_lock);
 664                 parent = dentry->d_parent->d_inode;
 665                 nodeid = get_fuse_inode(parent)->nodeid;
 666                 generation = parent->i_generation;
 667                 spin_unlock(&dentry->d_lock);
 668
 669                 fh[3] = (u32)(nodeid >> 32);
 670                 fh[4] = (u32)(nodeid & 0xffffffff);
 671                 fh[5] = generation;
 672         }
 673
 674         *max_len = len;
 675         return encode_parent ? 0x82 : 0x81;
 676 }
 677
 678 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 679                 struct fid *fid, int fh_len, int fh_type)
 680 {
 681         struct fuse_inode_handle handle;
 682
 683         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 684                 return NULL;
 685
 686         handle.nodeid = (u64) fid->raw[0] << 32;
 687         handle.nodeid |= (u64) fid->raw[1];
 688         handle.generation = fid->raw[2];
 689         return fuse_get_dentry(sb, &handle);
 690 }
 691
 692 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 693                 struct fid *fid, int fh_len, int fh_type)
 694 {
 695         struct fuse_inode_handle parent;
 696
 697         if (fh_type != 0x82 || fh_len < 6)
 698                 return NULL;
 699
 700         parent.nodeid = (u64) fid->raw[3] << 32;
 701         parent.nodeid |= (u64) fid->raw[4];
 702         parent.generation = fid->raw[5];
 703         return fuse_get_dentry(sb, &parent);
 704 }
 705
 706 static struct dentry *fuse_get_parent(struct dentry *child)
 707 {
 708         struct inode *child_inode = child->d_inode;
 709         struct fuse_conn *fc = get_fuse_conn(child_inode);
 710         struct inode *inode;
 711         struct dentry *parent;
 712         struct fuse_entry_out outarg;
 713         struct qstr name;
 714         int err;
 715
 716         if (!fc->export_support)
 717                 return ERR_PTR(-ESTALE);
 718
 719         name.len = 2;
 720         name.name = "..";
 721         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 722                                &name, &outarg, &inode);
 723         if (err) {
 724                 if (err == -ENOENT)
 725                         return ERR_PTR(-ESTALE);
 726                 return ERR_PTR(err);
 727         }
 728
 729         parent = d_obtain_alias(inode);
 730         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID) {
 731                 parent->d_op = &fuse_dentry_operations;
 732                 fuse_invalidate_entry_cache(parent);
 733         }
 734
 735         return parent;
 736 }
 737
 738 static const struct export_operations fuse_export_operations = {
 739         .fh_to_dentry   = fuse_fh_to_dentry,
 740         .fh_to_parent   = fuse_fh_to_parent,
 741         .encode_fh      = fuse_encode_fh,
 742         .get_parent     = fuse_get_parent,
 743 };
 744
 745 static const struct super_operations fuse_super_operations = {
 746         .alloc_inode    = fuse_alloc_inode,
 747         .destroy_inode  = fuse_destroy_inode,
 748         .clear_inode    = fuse_clear_inode,
 749         .drop_inode     = generic_delete_inode,
 750         .remount_fs     = fuse_remount_fs,
 751         .put_super      = fuse_put_super,
 752         .umount_begin   = fuse_umount_begin,
 753         .statfs         = fuse_statfs,
 754         .show_options   = fuse_show_options,
 755 };
 756
 757 static void sanitize_global_limit(unsigned *limit)
 758 {
 759         if (*limit == 0)
 760                 *limit = ((num_physpages << PAGE_SHIFT) >> 13) /
 761                          sizeof(struct fuse_req);
 762
 763         if (*limit >= 1 << 16)
 764                 *limit = (1 << 16) - 1;
 765 }
 766
 767 static int set_global_limit(const char *val, struct kernel_param *kp)
 768 {
 769         int rv;
 770
 771         rv = param_set_uint(val, kp);
 772         if (rv)
 773                 return rv;
 774
 775         sanitize_global_limit((unsigned *)kp->arg);
 776
 777         return 0;
 778 }
 779
 780 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 781 {
 782         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 783
 784         if (arg->minor < 13)
 785                 return;
 786
 787         sanitize_global_limit(&max_user_bgreq);
 788         sanitize_global_limit(&max_user_congthresh);
 789
 790         if (arg->max_background) {
 791                 fc->max_background = arg->max_background;
 792
 793                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 794                         fc->max_background = max_user_bgreq;
 795         }
 796         if (arg->congestion_threshold) {
 797                 fc->congestion_threshold = arg->congestion_threshold;
 798
 799                 if (!cap_sys_admin &&
 800                     fc->congestion_threshold > max_user_congthresh)
 801                         fc->congestion_threshold = max_user_congthresh;
 802         }
 803 }
 804
 805 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 806 {
 807         struct fuse_init_out *arg = &req->misc.init_out;
 808
 809         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 810                 fc->conn_error = 1;
 811         else {
 812                 unsigned long ra_pages;
 813
 814                 process_init_limits(fc, arg);
 815
 816                 if (arg->minor >= 6) {
 817                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
 818                         if (arg->flags & FUSE_ASYNC_READ)
 819                                 fc->async_read = 1;
 820                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 821                                 fc->no_lock = 1;
 822                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 823                                 fc->atomic_o_trunc = 1;
 824                         if (arg->minor >= 9) {
 825                                 /* LOOKUP has dependency on proto version */
 826                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 827                                         fc->export_support = 1;
 828                         }
 829                         if (arg->flags & FUSE_BIG_WRITES)
 830                                 fc->big_writes = 1;
 831                         if (arg->flags & FUSE_DONT_MASK)
 832                                 fc->dont_mask = 1;
 833                 } else {
 834                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
 835                         fc->no_lock = 1;
 836                 }
 837
 838                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
 839                 fc->minor = arg->minor;
 840                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 841                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 842                 fc->conn_init = 1;
 843         }
 844         fc->blocked = 0;
 845         wake_up_all(&fc->blocked_waitq);
 846 }
 847
 848 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 849 {
 850         struct fuse_init_in *arg = &req->misc.init_in;
 851
 852         arg->major = FUSE_KERNEL_VERSION;
 853         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 854         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
 855         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 856                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK;
 857         req->in.h.opcode = FUSE_INIT;
 858         req->in.numargs = 1;
 859         req->in.args[0].size = sizeof(*arg);
 860         req->in.args[0].value = arg;
 861         req->out.numargs = 1;
 862         /* Variable length arguement used for backward compatibility
 863            with interface version < 7.5.  Rest of init_out is zeroed
 864            by do_get_request(), so a short reply is not a problem */
 865         req->out.argvar = 1;
 866         req->out.args[0].size = sizeof(struct fuse_init_out);
 867         req->out.args[0].value = &req->misc.init_out;
 868         req->end = process_init_reply;
 869         fuse_request_send_background(fc, req);
 870 }
 871
 872 static void fuse_free_conn(struct fuse_conn *fc)
 873 {
 874         kfree(fc);
 875 }
 876
 877 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 878 {
 879         int err;
 880
 881         fc->bdi.name = "fuse";
 882         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 883         fc->bdi.unplug_io_fn = default_unplug_io_fn;
 884         /* fuse does it's own writeback accounting */
 885         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
 886
 887         err = bdi_init(&fc->bdi);
 888         if (err)
 889                 return err;
 890
 891         fc->bdi_initialized = 1;
 892
 893         if (sb->s_bdev) {
 894                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
 895                                     MAJOR(fc->dev), MINOR(fc->dev));
 896         } else {
 897                 err = bdi_register_dev(&fc->bdi, fc->dev);
 898         }
 899
 900         if (err)
 901                 return err;
 902
 903         /*
 904          * For a single fuse filesystem use max 1% of dirty +
 905          * writeback threshold.
 906          *
 907          * This gives about 1M of write buffer for memory maps on a
 908          * machine with 1G and 10% dirty_ratio, which should be more
 909          * than enough.
 910          *
 911          * Privileged users can raise it by writing to
 912          *
 913          *    /sys/class/bdi/<bdi>/max_ratio
 914          */
 915         bdi_set_max_ratio(&fc->bdi, 1);
 916
 917         return 0;
 918 }
 919
 920 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
 921 {
 922         struct fuse_conn *fc;
 923         struct inode *root;
 924         struct fuse_mount_data d;
 925         struct file *file;
 926         struct dentry *root_dentry;
 927         struct fuse_req *init_req;
 928         int err;
 929         int is_bdev = sb->s_bdev != NULL;
 930
 931         err = -EINVAL;
 932         if (sb->s_flags & MS_MANDLOCK)
 933                 goto err;
 934
 935         if (!parse_fuse_opt((char *) data, &d, is_bdev))
 936                 goto err;
 937
 938         if (is_bdev) {
 939 #ifdef CONFIG_BLOCK
 940                 err = -EINVAL;
 941                 if (!sb_set_blocksize(sb, d.blksize))
 942                         goto err;
 943 #endif
 944         } else {
 945                 sb->s_blocksize = PAGE_CACHE_SIZE;
 946                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 947         }
 948         sb->s_magic = FUSE_SUPER_MAGIC;
 949         sb->s_op = &fuse_super_operations;
 950         sb->s_maxbytes = MAX_LFS_FILESIZE;
 951         sb->s_export_op = &fuse_export_operations;
 952
 953         file = fget(d.fd);
 954         err = -EINVAL;
 955         if (!file)
 956                 goto err;
 957
 958         if (file->f_op != &fuse_dev_operations)
 959                 goto err_fput;
 960
 961         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
 962         err = -ENOMEM;
 963         if (!fc)
 964                 goto err_fput;
 965
 966         fuse_conn_init(fc);
 967
 968         fc->dev = sb->s_dev;
 969         fc->sb = sb;
 970         err = fuse_bdi_init(fc, sb);
 971         if (err)
 972                 goto err_put_conn;
 973
 974         sb->s_bdi = &fc->bdi;
 975
 976         /* Handle umasking inside the fuse code */
 977         if (sb->s_flags & MS_POSIXACL)
 978                 fc->dont_mask = 1;
 979         sb->s_flags |= MS_POSIXACL;
 980
 981         fc->release = fuse_free_conn;
 982         fc->flags = d.flags;
 983         fc->user_id = d.user_id;
 984         fc->group_id = d.group_id;
 985         fc->max_read = max_t(unsigned, 4096, d.max_read);
 986
 987         /* Used by get_root_inode() */
 988         sb->s_fs_info = fc;
 989
 990         err = -ENOMEM;
 991         root = fuse_get_root_inode(sb, d.rootmode);
 992         if (!root)
 993                 goto err_put_conn;
 994
 995         root_dentry = d_alloc_root(root);
 996         if (!root_dentry) {
 997                 iput(root);
 998                 goto err_put_conn;
 999         }
1000
1001         init_req = fuse_request_alloc();
1002         if (!init_req)
1003                 goto err_put_root;
1004
1005         if (is_bdev) {
1006                 fc->destroy_req = fuse_request_alloc();
1007                 if (!fc->destroy_req)
1008                         goto err_free_init_req;
1009         }
1010
1011         mutex_lock(&fuse_mutex);
1012         err = -EINVAL;
1013         if (file->private_data)
1014                 goto err_unlock;
1015
1016         err = fuse_ctl_add_conn(fc);
1017         if (err)
1018                 goto err_unlock;
1019
1020         list_add_tail(&fc->entry, &fuse_conn_list);
1021         sb->s_root = root_dentry;
1022         fc->connected = 1;
1023         file->private_data = fuse_conn_get(fc);
1024         mutex_unlock(&fuse_mutex);
1025         /*
1026          * atomic_dec_and_test() in fput() provides the necessary
1027          * memory barrier for file->private_data to be visible on all
1028          * CPUs after this
1029          */
1030         fput(file);
1031
1032         fuse_send_init(fc, init_req);
1033
1034         return 0;
1035
1036  err_unlock:
1037         mutex_unlock(&fuse_mutex);
1038  err_free_init_req:
1039         fuse_request_free(init_req);
1040  err_put_root:
1041         dput(root_dentry);
1042  err_put_conn:
1043         fuse_bdi_destroy(fc);
1044         fuse_conn_put(fc);
1045  err_fput:
1046         fput(file);
1047  err:
1048         return err;
1049 }
1050
1051 static int fuse_get_sb(struct file_system_type *fs_type,
1052                        int flags, const char *dev_name,
1053                        void *raw_data, struct vfsmount *mnt)
1054 {
1055         return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
1056 }
1057
1058 static void fuse_kill_sb_anon(struct super_block *sb)
1059 {
1060         struct fuse_conn *fc = get_fuse_conn_super(sb);
1061
1062         if (fc) {
1063                 down_write(&fc->killsb);
1064                 fc->sb = NULL;
1065                 up_write(&fc->killsb);
1066         }
1067
1068         kill_anon_super(sb);
1069 }
1070
1071 static struct file_system_type fuse_fs_type = {
1072         .owner          = THIS_MODULE,
1073         .name           = "fuse",
1074         .fs_flags       = FS_HAS_SUBTYPE,
1075         .get_sb         = fuse_get_sb,
1076         .kill_sb        = fuse_kill_sb_anon,
1077 };
1078
1079 #ifdef CONFIG_BLOCK
1080 static int fuse_get_sb_blk(struct file_system_type *fs_type,
1081                            int flags, const char *dev_name,
1082                            void *raw_data, struct vfsmount *mnt)
1083 {
1084         return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
1085                            mnt);
1086 }
1087
1088 static void fuse_kill_sb_blk(struct super_block *sb)
1089 {
1090         struct fuse_conn *fc = get_fuse_conn_super(sb);
1091
1092         if (fc) {
1093                 down_write(&fc->killsb);
1094                 fc->sb = NULL;
1095                 up_write(&fc->killsb);
1096         }
1097
1098         kill_block_super(sb);
1099 }
1100
1101 static struct file_system_type fuseblk_fs_type = {
1102         .owner          = THIS_MODULE,
1103         .name           = "fuseblk",
1104         .get_sb         = fuse_get_sb_blk,
1105         .kill_sb        = fuse_kill_sb_blk,
1106         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1107 };
1108
1109 static inline int register_fuseblk(void)
1110 {
1111         return register_filesystem(&fuseblk_fs_type);
1112 }
1113
1114 static inline void unregister_fuseblk(void)
1115 {
1116         unregister_filesystem(&fuseblk_fs_type);
1117 }
1118 #else
1119 static inline int register_fuseblk(void)
1120 {
1121         return 0;
1122 }
1123
1124 static inline void unregister_fuseblk(void)
1125 {
1126 }
1127 #endif
1128
1129 static void fuse_inode_init_once(void *foo)
1130 {
1131         struct inode *inode = foo;
1132
1133         inode_init_once(inode);
1134 }
1135
1136 static int __init fuse_fs_init(void)
1137 {
1138         int err;
1139
1140         err = register_filesystem(&fuse_fs_type);
1141         if (err)
1142                 goto out;
1143
1144         err = register_fuseblk();
1145         if (err)
1146                 goto out_unreg;
1147
1148         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1149                                               sizeof(struct fuse_inode),
1150                                               0, SLAB_HWCACHE_ALIGN,
1151                                               fuse_inode_init_once);
1152         err = -ENOMEM;
1153         if (!fuse_inode_cachep)
1154                 goto out_unreg2;
1155
1156         return 0;
1157
1158  out_unreg2:
1159         unregister_fuseblk();
1160  out_unreg:
1161         unregister_filesystem(&fuse_fs_type);
1162  out:
1163         return err;
1164 }
1165
1166 static void fuse_fs_cleanup(void)
1167 {
1168         unregister_filesystem(&fuse_fs_type);
1169         unregister_fuseblk();
1170         kmem_cache_destroy(fuse_inode_cachep);
1171 }
1172
1173 static struct kobject *fuse_kobj;
1174 static struct kobject *connections_kobj;
1175
1176 static int fuse_sysfs_init(void)
1177 {
1178         int err;
1179
1180         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1181         if (!fuse_kobj) {
1182                 err = -ENOMEM;
1183                 goto out_err;
1184         }
1185
1186         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1187         if (!connections_kobj) {
1188                 err = -ENOMEM;
1189                 goto out_fuse_unregister;
1190         }
1191
1192         return 0;
1193
1194  out_fuse_unregister:
1195         kobject_put(fuse_kobj);
1196  out_err:
1197         return err;
1198 }
1199
1200 static void fuse_sysfs_cleanup(void)
1201 {
1202         kobject_put(connections_kobj);
1203         kobject_put(fuse_kobj);
1204 }
1205
1206 static int __init fuse_init(void)
1207 {
1208         int res;
1209
1210         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1211                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1212
1213         INIT_LIST_HEAD(&fuse_conn_list);
1214         res = fuse_fs_init();
1215         if (res)
1216                 goto err;
1217
1218         res = fuse_dev_init();
1219         if (res)
1220                 goto err_fs_cleanup;
1221
1222         res = fuse_sysfs_init();
1223         if (res)
1224                 goto err_dev_cleanup;
1225
1226         res = fuse_ctl_init();
1227         if (res)
1228                 goto err_sysfs_cleanup;
1229
1230         sanitize_global_limit(&max_user_bgreq);
1231         sanitize_global_limit(&max_user_congthresh);
1232
1233         return 0;
1234
1235  err_sysfs_cleanup:
1236         fuse_sysfs_cleanup();
1237  err_dev_cleanup:
1238         fuse_dev_cleanup();
1239  err_fs_cleanup:
1240         fuse_fs_cleanup();
1241  err:
1242         return res;
1243 }
1244
1245 static void __exit fuse_exit(void)
1246 {
1247         printk(KERN_DEBUG "fuse exit\n");
1248
1249         fuse_ctl_cleanup();
1250         fuse_sysfs_cleanup();
1251         fuse_fs_cleanup();
1252         fuse_dev_cleanup();
1253 }
1254
1255 module_init(fuse_init);
1256 module_exit(fuse_exit);