fs/fuse/file.c

   1 /*
   2   FUSE: Filesystem in Userspace
   3   Copyright (C) 2001-2006  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/kernel.h>
  14
  15 static const struct file_operations fuse_direct_io_file_operations;
  16
  17 static int fuse_send_open(struct inode *inode, struct file *file, int isdir,
  18                           struct fuse_open_out *outargp)
  19 {
  20         struct fuse_conn *fc = get_fuse_conn(inode);
  21         struct fuse_open_in inarg;
  22         struct fuse_req *req;
  23         int err;
  24
  25         req = fuse_get_req(fc);
  26         if (IS_ERR(req))
  27                 return PTR_ERR(req);
  28
  29         memset(&inarg, 0, sizeof(inarg));
  30         inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC);
  31         req->in.h.opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
  32         req->in.h.nodeid = get_node_id(inode);
  33         req->in.numargs = 1;
  34         req->in.args[0].size = sizeof(inarg);
  35         req->in.args[0].value = &inarg;
  36         req->out.numargs = 1;
  37         req->out.args[0].size = sizeof(*outargp);
  38         req->out.args[0].value = outargp;
  39         request_send(fc, req);
  40         err = req->out.h.error;
  41         fuse_put_request(fc, req);
  42
  43         return err;
  44 }
  45
  46 struct fuse_file *fuse_file_alloc(void)
  47 {
  48         struct fuse_file *ff;
  49         ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL);
  50         if (ff) {
  51                 ff->release_req = fuse_request_alloc();
  52                 if (!ff->release_req) {
  53                         kfree(ff);
  54                         ff = NULL;
  55                 }
  56         }
  57         return ff;
  58 }
  59
  60 void fuse_file_free(struct fuse_file *ff)
  61 {
  62         fuse_request_free(ff->release_req);
  63         kfree(ff);
  64 }
  65
  66 void fuse_finish_open(struct inode *inode, struct file *file,
  67                       struct fuse_file *ff, struct fuse_open_out *outarg)
  68 {
  69         if (outarg->open_flags & FOPEN_DIRECT_IO)
  70                 file->f_op = &fuse_direct_io_file_operations;
  71         if (!(outarg->open_flags & FOPEN_KEEP_CACHE))
  72                 invalidate_inode_pages(inode->i_mapping);
  73         ff->fh = outarg->fh;
  74         file->private_data = ff;
  75 }
  76
  77 int fuse_open_common(struct inode *inode, struct file *file, int isdir)
  78 {
  79         struct fuse_open_out outarg;
  80         struct fuse_file *ff;
  81         int err;
  82
  83         /* VFS checks this, but only _after_ ->open() */
  84         if (file->f_flags & O_DIRECT)
  85                 return -EINVAL;
  86
  87         err = generic_file_open(inode, file);
  88         if (err)
  89                 return err;
  90
  91         /* If opening the root node, no lookup has been performed on
  92            it, so the attributes must be refreshed */
  93         if (get_node_id(inode) == FUSE_ROOT_ID) {
  94                 err = fuse_do_getattr(inode);
  95                 if (err)
  96                         return err;
  97         }
  98
  99         ff = fuse_file_alloc();
 100         if (!ff)
 101                 return -ENOMEM;
 102
 103         err = fuse_send_open(inode, file, isdir, &outarg);
 104         if (err)
 105                 fuse_file_free(ff);
 106         else {
 107                 if (isdir)
 108                         outarg.open_flags &= ~FOPEN_DIRECT_IO;
 109                 fuse_finish_open(inode, file, ff, &outarg);
 110         }
 111
 112         return err;
 113 }
 114
 115 struct fuse_req *fuse_release_fill(struct fuse_file *ff, u64 nodeid, int flags,
 116                                    int opcode)
 117 {
 118         struct fuse_req *req = ff->release_req;
 119         struct fuse_release_in *inarg = &req->misc.release_in;
 120
 121         inarg->fh = ff->fh;
 122         inarg->flags = flags;
 123         req->in.h.opcode = opcode;
 124         req->in.h.nodeid = nodeid;
 125         req->in.numargs = 1;
 126         req->in.args[0].size = sizeof(struct fuse_release_in);
 127         req->in.args[0].value = inarg;
 128         kfree(ff);
 129
 130         return req;
 131 }
 132
 133 int fuse_release_common(struct inode *inode, struct file *file, int isdir)
 134 {
 135         struct fuse_file *ff = file->private_data;
 136         if (ff) {
 137                 struct fuse_conn *fc = get_fuse_conn(inode);
 138                 struct fuse_req *req;
 139
 140                 req = fuse_release_fill(ff, get_node_id(inode), file->f_flags,
 141                                         isdir ? FUSE_RELEASEDIR : FUSE_RELEASE);
 142
 143                 /* Hold vfsmount and dentry until release is finished */
 144                 req->vfsmount = mntget(file->f_vfsmnt);
 145                 req->dentry = dget(file->f_dentry);
 146                 request_send_background(fc, req);
 147         }
 148
 149         /* Return value is ignored by VFS */
 150         return 0;
 151 }
 152
 153 static int fuse_open(struct inode *inode, struct file *file)
 154 {
 155         return fuse_open_common(inode, file, 0);
 156 }
 157
 158 static int fuse_release(struct inode *inode, struct file *file)
 159 {
 160         return fuse_release_common(inode, file, 0);
 161 }
 162
 163 static int fuse_flush(struct file *file, fl_owner_t id)
 164 {
 165         struct inode *inode = file->f_dentry->d_inode;
 166         struct fuse_conn *fc = get_fuse_conn(inode);
 167         struct fuse_file *ff = file->private_data;
 168         struct fuse_req *req;
 169         struct fuse_flush_in inarg;
 170         int err;
 171
 172         if (is_bad_inode(inode))
 173                 return -EIO;
 174
 175         if (fc->no_flush)
 176                 return 0;
 177
 178         req = fuse_get_req(fc);
 179         if (IS_ERR(req))
 180                 return PTR_ERR(req);
 181
 182         memset(&inarg, 0, sizeof(inarg));
 183         inarg.fh = ff->fh;
 184         req->in.h.opcode = FUSE_FLUSH;
 185         req->in.h.nodeid = get_node_id(inode);
 186         req->in.numargs = 1;
 187         req->in.args[0].size = sizeof(inarg);
 188         req->in.args[0].value = &inarg;
 189         request_send(fc, req);
 190         err = req->out.h.error;
 191         fuse_put_request(fc, req);
 192         if (err == -ENOSYS) {
 193                 fc->no_flush = 1;
 194                 err = 0;
 195         }
 196         return err;
 197 }
 198
 199 int fuse_fsync_common(struct file *file, struct dentry *de, int datasync,
 200                       int isdir)
 201 {
 202         struct inode *inode = de->d_inode;
 203         struct fuse_conn *fc = get_fuse_conn(inode);
 204         struct fuse_file *ff = file->private_data;
 205         struct fuse_req *req;
 206         struct fuse_fsync_in inarg;
 207         int err;
 208
 209         if (is_bad_inode(inode))
 210                 return -EIO;
 211
 212         if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
 213                 return 0;
 214
 215         req = fuse_get_req(fc);
 216         if (IS_ERR(req))
 217                 return PTR_ERR(req);
 218
 219         memset(&inarg, 0, sizeof(inarg));
 220         inarg.fh = ff->fh;
 221         inarg.fsync_flags = datasync ? 1 : 0;
 222         req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
 223         req->in.h.nodeid = get_node_id(inode);
 224         req->in.numargs = 1;
 225         req->in.args[0].size = sizeof(inarg);
 226         req->in.args[0].value = &inarg;
 227         request_send(fc, req);
 228         err = req->out.h.error;
 229         fuse_put_request(fc, req);
 230         if (err == -ENOSYS) {
 231                 if (isdir)
 232                         fc->no_fsyncdir = 1;
 233                 else
 234                         fc->no_fsync = 1;
 235                 err = 0;
 236         }
 237         return err;
 238 }
 239
 240 static int fuse_fsync(struct file *file, struct dentry *de, int datasync)
 241 {
 242         return fuse_fsync_common(file, de, datasync, 0);
 243 }
 244
 245 void fuse_read_fill(struct fuse_req *req, struct file *file,
 246                     struct inode *inode, loff_t pos, size_t count, int opcode)
 247 {
 248         struct fuse_file *ff = file->private_data;
 249         struct fuse_read_in *inarg = &req->misc.read_in;
 250
 251         inarg->fh = ff->fh;
 252         inarg->offset = pos;
 253         inarg->size = count;
 254         req->in.h.opcode = opcode;
 255         req->in.h.nodeid = get_node_id(inode);
 256         req->in.numargs = 1;
 257         req->in.args[0].size = sizeof(struct fuse_read_in);
 258         req->in.args[0].value = inarg;
 259         req->out.argpages = 1;
 260         req->out.argvar = 1;
 261         req->out.numargs = 1;
 262         req->out.args[0].size = count;
 263 }
 264
 265 static size_t fuse_send_read(struct fuse_req *req, struct file *file,
 266                              struct inode *inode, loff_t pos, size_t count)
 267 {
 268         struct fuse_conn *fc = get_fuse_conn(inode);
 269         fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
 270         request_send(fc, req);
 271         return req->out.args[0].size;
 272 }
 273
 274 static int fuse_readpage(struct file *file, struct page *page)
 275 {
 276         struct inode *inode = page->mapping->host;
 277         struct fuse_conn *fc = get_fuse_conn(inode);
 278         struct fuse_req *req;
 279         int err;
 280
 281         err = -EIO;
 282         if (is_bad_inode(inode))
 283                 goto out;
 284
 285         req = fuse_get_req(fc);
 286         err = PTR_ERR(req);
 287         if (IS_ERR(req))
 288                 goto out;
 289
 290         req->out.page_zeroing = 1;
 291         req->num_pages = 1;
 292         req->pages[0] = page;
 293         fuse_send_read(req, file, inode, page_offset(page), PAGE_CACHE_SIZE);
 294         err = req->out.h.error;
 295         fuse_put_request(fc, req);
 296         if (!err)
 297                 SetPageUptodate(page);
 298         fuse_invalidate_attr(inode); /* atime changed */
 299  out:
 300         unlock_page(page);
 301         return err;
 302 }
 303
 304 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
 305 {
 306         int i;
 307
 308         fuse_invalidate_attr(req->pages[0]->mapping->host); /* atime changed */
 309
 310         for (i = 0; i < req->num_pages; i++) {
 311                 struct page *page = req->pages[i];
 312                 if (!req->out.h.error)
 313                         SetPageUptodate(page);
 314                 else
 315                         SetPageError(page);
 316                 unlock_page(page);
 317         }
 318         fuse_put_request(fc, req);
 319 }
 320
 321 static void fuse_send_readpages(struct fuse_req *req, struct file *file,
 322                                 struct inode *inode)
 323 {
 324         struct fuse_conn *fc = get_fuse_conn(inode);
 325         loff_t pos = page_offset(req->pages[0]);
 326         size_t count = req->num_pages << PAGE_CACHE_SHIFT;
 327         req->out.page_zeroing = 1;
 328         fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
 329         if (fc->async_read) {
 330                 get_file(file);
 331                 req->file = file;
 332                 req->end = fuse_readpages_end;
 333                 request_send_background(fc, req);
 334         } else {
 335                 request_send(fc, req);
 336                 fuse_readpages_end(fc, req);
 337         }
 338 }
 339
 340 struct fuse_readpages_data {
 341         struct fuse_req *req;
 342         struct file *file;
 343         struct inode *inode;
 344 };
 345
 346 static int fuse_readpages_fill(void *_data, struct page *page)
 347 {
 348         struct fuse_readpages_data *data = _data;
 349         struct fuse_req *req = data->req;
 350         struct inode *inode = data->inode;
 351         struct fuse_conn *fc = get_fuse_conn(inode);
 352
 353         if (req->num_pages &&
 354             (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
 355              (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
 356              req->pages[req->num_pages - 1]->index + 1 != page->index)) {
 357                 fuse_send_readpages(req, data->file, inode);
 358                 data->req = req = fuse_get_req(fc);
 359                 if (IS_ERR(req)) {
 360                         unlock_page(page);
 361                         return PTR_ERR(req);
 362                 }
 363         }
 364         req->pages[req->num_pages] = page;
 365         req->num_pages ++;
 366         return 0;
 367 }
 368
 369 static int fuse_readpages(struct file *file, struct address_space *mapping,
 370                           struct list_head *pages, unsigned nr_pages)
 371 {
 372         struct inode *inode = mapping->host;
 373         struct fuse_conn *fc = get_fuse_conn(inode);
 374         struct fuse_readpages_data data;
 375         int err;
 376
 377         if (is_bad_inode(inode))
 378                 return -EIO;
 379
 380         data.file = file;
 381         data.inode = inode;
 382         data.req = fuse_get_req(fc);
 383         if (IS_ERR(data.req))
 384                 return PTR_ERR(data.req);
 385
 386         err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
 387         if (!err) {
 388                 if (data.req->num_pages)
 389                         fuse_send_readpages(data.req, file, inode);
 390                 else
 391                         fuse_put_request(fc, data.req);
 392         }
 393         return err;
 394 }
 395
 396 static size_t fuse_send_write(struct fuse_req *req, struct file *file,
 397                               struct inode *inode, loff_t pos, size_t count)
 398 {
 399         struct fuse_conn *fc = get_fuse_conn(inode);
 400         struct fuse_file *ff = file->private_data;
 401         struct fuse_write_in inarg;
 402         struct fuse_write_out outarg;
 403
 404         memset(&inarg, 0, sizeof(struct fuse_write_in));
 405         inarg.fh = ff->fh;
 406         inarg.offset = pos;
 407         inarg.size = count;
 408         req->in.h.opcode = FUSE_WRITE;
 409         req->in.h.nodeid = get_node_id(inode);
 410         req->in.argpages = 1;
 411         req->in.numargs = 2;
 412         req->in.args[0].size = sizeof(struct fuse_write_in);
 413         req->in.args[0].value = &inarg;
 414         req->in.args[1].size = count;
 415         req->out.numargs = 1;
 416         req->out.args[0].size = sizeof(struct fuse_write_out);
 417         req->out.args[0].value = &outarg;
 418         request_send(fc, req);
 419         return outarg.size;
 420 }
 421
 422 static int fuse_prepare_write(struct file *file, struct page *page,
 423                               unsigned offset, unsigned to)
 424 {
 425         /* No op */
 426         return 0;
 427 }
 428
 429 static int fuse_commit_write(struct file *file, struct page *page,
 430                              unsigned offset, unsigned to)
 431 {
 432         int err;
 433         size_t nres;
 434         unsigned count = to - offset;
 435         struct inode *inode = page->mapping->host;
 436         struct fuse_conn *fc = get_fuse_conn(inode);
 437         loff_t pos = page_offset(page) + offset;
 438         struct fuse_req *req;
 439
 440         if (is_bad_inode(inode))
 441                 return -EIO;
 442
 443         req = fuse_get_req(fc);
 444         if (IS_ERR(req))
 445                 return PTR_ERR(req);
 446
 447         req->num_pages = 1;
 448         req->pages[0] = page;
 449         req->page_offset = offset;
 450         nres = fuse_send_write(req, file, inode, pos, count);
 451         err = req->out.h.error;
 452         fuse_put_request(fc, req);
 453         if (!err && nres != count)
 454                 err = -EIO;
 455         if (!err) {
 456                 pos += count;
 457                 if (pos > i_size_read(inode))
 458                         i_size_write(inode, pos);
 459
 460                 if (offset == 0 && to == PAGE_CACHE_SIZE) {
 461                         clear_page_dirty(page);
 462                         SetPageUptodate(page);
 463                 }
 464         }
 465         fuse_invalidate_attr(inode);
 466         return err;
 467 }
 468
 469 static void fuse_release_user_pages(struct fuse_req *req, int write)
 470 {
 471         unsigned i;
 472
 473         for (i = 0; i < req->num_pages; i++) {
 474                 struct page *page = req->pages[i];
 475                 if (write)
 476                         set_page_dirty_lock(page);
 477                 put_page(page);
 478         }
 479 }
 480
 481 static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
 482                                unsigned nbytes, int write)
 483 {
 484         unsigned long user_addr = (unsigned long) buf;
 485         unsigned offset = user_addr & ~PAGE_MASK;
 486         int npages;
 487
 488         /* This doesn't work with nfsd */
 489         if (!current->mm)
 490                 return -EPERM;
 491
 492         nbytes = min(nbytes, (unsigned) FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
 493         npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
 494         npages = min(max(npages, 1), FUSE_MAX_PAGES_PER_REQ);
 495         down_read(&current->mm->mmap_sem);
 496         npages = get_user_pages(current, current->mm, user_addr, npages, write,
 497                                 0, req->pages, NULL);
 498         up_read(&current->mm->mmap_sem);
 499         if (npages < 0)
 500                 return npages;
 501
 502         req->num_pages = npages;
 503         req->page_offset = offset;
 504         return 0;
 505 }
 506
 507 static ssize_t fuse_direct_io(struct file *file, const char __user *buf,
 508                               size_t count, loff_t *ppos, int write)
 509 {
 510         struct inode *inode = file->f_dentry->d_inode;
 511         struct fuse_conn *fc = get_fuse_conn(inode);
 512         size_t nmax = write ? fc->max_write : fc->max_read;
 513         loff_t pos = *ppos;
 514         ssize_t res = 0;
 515         struct fuse_req *req;
 516
 517         if (is_bad_inode(inode))
 518                 return -EIO;
 519
 520         req = fuse_get_req(fc);
 521         if (IS_ERR(req))
 522                 return PTR_ERR(req);
 523
 524         while (count) {
 525                 size_t nres;
 526                 size_t nbytes = min(count, nmax);
 527                 int err = fuse_get_user_pages(req, buf, nbytes, !write);
 528                 if (err) {
 529                         res = err;
 530                         break;
 531                 }
 532                 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset;
 533                 nbytes = min(count, nbytes);
 534                 if (write)
 535                         nres = fuse_send_write(req, file, inode, pos, nbytes);
 536                 else
 537                         nres = fuse_send_read(req, file, inode, pos, nbytes);
 538                 fuse_release_user_pages(req, !write);
 539                 if (req->out.h.error) {
 540                         if (!res)
 541                                 res = req->out.h.error;
 542                         break;
 543                 } else if (nres > nbytes) {
 544                         res = -EIO;
 545                         break;
 546                 }
 547                 count -= nres;
 548                 res += nres;
 549                 pos += nres;
 550                 buf += nres;
 551                 if (nres != nbytes)
 552                         break;
 553                 if (count) {
 554                         fuse_put_request(fc, req);
 555                         req = fuse_get_req(fc);
 556                         if (IS_ERR(req))
 557                                 break;
 558                 }
 559         }
 560         fuse_put_request(fc, req);
 561         if (res > 0) {
 562                 if (write && pos > i_size_read(inode))
 563                         i_size_write(inode, pos);
 564                 *ppos = pos;
 565         }
 566         fuse_invalidate_attr(inode);
 567
 568         return res;
 569 }
 570
 571 static ssize_t fuse_direct_read(struct file *file, char __user *buf,
 572                                      size_t count, loff_t *ppos)
 573 {
 574         return fuse_direct_io(file, buf, count, ppos, 0);
 575 }
 576
 577 static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
 578                                  size_t count, loff_t *ppos)
 579 {
 580         struct inode *inode = file->f_dentry->d_inode;
 581         ssize_t res;
 582         /* Don't allow parallel writes to the same file */
 583         mutex_lock(&inode->i_mutex);
 584         res = fuse_direct_io(file, buf, count, ppos, 1);
 585         mutex_unlock(&inode->i_mutex);
 586         return res;
 587 }
 588
 589 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
 590 {
 591         if ((vma->vm_flags & VM_SHARED)) {
 592                 if ((vma->vm_flags & VM_WRITE))
 593                         return -ENODEV;
 594                 else
 595                         vma->vm_flags &= ~VM_MAYWRITE;
 596         }
 597         return generic_file_mmap(file, vma);
 598 }
 599
 600 static int fuse_set_page_dirty(struct page *page)
 601 {
 602         printk("fuse_set_page_dirty: should not happen\n");
 603         dump_stack();
 604         return 0;
 605 }
 606
 607 static const struct file_operations fuse_file_operations = {
 608         .llseek         = generic_file_llseek,
 609         .read           = generic_file_read,
 610         .write          = generic_file_write,
 611         .mmap           = fuse_file_mmap,
 612         .open           = fuse_open,
 613         .flush          = fuse_flush,
 614         .release        = fuse_release,
 615         .fsync          = fuse_fsync,
 616         .sendfile       = generic_file_sendfile,
 617 };
 618
 619 static const struct file_operations fuse_direct_io_file_operations = {
 620         .llseek         = generic_file_llseek,
 621         .read           = fuse_direct_read,
 622         .write          = fuse_direct_write,
 623         .open           = fuse_open,
 624         .flush          = fuse_flush,
 625         .release        = fuse_release,
 626         .fsync          = fuse_fsync,
 627         /* no mmap and sendfile */
 628 };
 629
 630 static struct address_space_operations fuse_file_aops  = {
 631         .readpage       = fuse_readpage,
 632         .prepare_write  = fuse_prepare_write,
 633         .commit_write   = fuse_commit_write,
 634         .readpages      = fuse_readpages,
 635         .set_page_dirty = fuse_set_page_dirty,
 636 };
 637
 638 void fuse_init_file_inode(struct inode *inode)
 639 {
 640         inode->i_fop = &fuse_file_operations;
 641         inode->i_data.a_ops = &fuse_file_aops;
 642 }