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 #include <linux/sched.h>
  15
  16 static const struct file_operations fuse_direct_io_file_operations;
  17
  18 static int fuse_send_open(struct inode *inode, struct file *file, int isdir,
  19                           struct fuse_open_out *outargp)
  20 {
  21         struct fuse_conn *fc = get_fuse_conn(inode);
  22         struct fuse_open_in inarg;
  23         struct fuse_req *req;
  24         int err;
  25
  26         req = fuse_get_req(fc);
  27         if (IS_ERR(req))
  28                 return PTR_ERR(req);
  29
  30         memset(&inarg, 0, sizeof(inarg));
  31         inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY);
  32         if (!fc->atomic_o_trunc)
  33                 inarg.flags &= ~O_TRUNC;
  34         req->in.h.opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
  35         req->in.h.nodeid = get_node_id(inode);
  36         req->in.numargs = 1;
  37         req->in.args[0].size = sizeof(inarg);
  38         req->in.args[0].value = &inarg;
  39         req->out.numargs = 1;
  40         req->out.args[0].size = sizeof(*outargp);
  41         req->out.args[0].value = outargp;
  42         request_send(fc, req);
  43         err = req->out.h.error;
  44         fuse_put_request(fc, req);
  45
  46         return err;
  47 }
  48
  49 struct fuse_file *fuse_file_alloc(void)
  50 {
  51         struct fuse_file *ff;
  52         ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL);
  53         if (ff) {
  54                 ff->reserved_req = fuse_request_alloc();
  55                 if (!ff->reserved_req) {
  56                         kfree(ff);
  57                         ff = NULL;
  58                 } else {
  59                         INIT_LIST_HEAD(&ff->write_entry);
  60                         atomic_set(&ff->count, 0);
  61                 }
  62         }
  63         return ff;
  64 }
  65
  66 void fuse_file_free(struct fuse_file *ff)
  67 {
  68         fuse_request_free(ff->reserved_req);
  69         kfree(ff);
  70 }
  71
  72 static struct fuse_file *fuse_file_get(struct fuse_file *ff)
  73 {
  74         atomic_inc(&ff->count);
  75         return ff;
  76 }
  77
  78 static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
  79 {
  80         dput(req->misc.release.dentry);
  81         mntput(req->misc.release.vfsmount);
  82         fuse_put_request(fc, req);
  83 }
  84
  85 static void fuse_file_put(struct fuse_file *ff)
  86 {
  87         if (atomic_dec_and_test(&ff->count)) {
  88                 struct fuse_req *req = ff->reserved_req;
  89                 struct inode *inode = req->misc.release.dentry->d_inode;
  90                 struct fuse_conn *fc = get_fuse_conn(inode);
  91                 req->end = fuse_release_end;
  92                 request_send_background(fc, req);
  93                 kfree(ff);
  94         }
  95 }
  96
  97 void fuse_finish_open(struct inode *inode, struct file *file,
  98                       struct fuse_file *ff, struct fuse_open_out *outarg)
  99 {
 100         if (outarg->open_flags & FOPEN_DIRECT_IO)
 101                 file->f_op = &fuse_direct_io_file_operations;
 102         if (!(outarg->open_flags & FOPEN_KEEP_CACHE))
 103                 invalidate_inode_pages2(inode->i_mapping);
 104         ff->fh = outarg->fh;
 105         file->private_data = fuse_file_get(ff);
 106 }
 107
 108 int fuse_open_common(struct inode *inode, struct file *file, int isdir)
 109 {
 110         struct fuse_open_out outarg;
 111         struct fuse_file *ff;
 112         int err;
 113
 114         /* VFS checks this, but only _after_ ->open() */
 115         if (file->f_flags & O_DIRECT)
 116                 return -EINVAL;
 117
 118         err = generic_file_open(inode, file);
 119         if (err)
 120                 return err;
 121
 122         ff = fuse_file_alloc();
 123         if (!ff)
 124                 return -ENOMEM;
 125
 126         err = fuse_send_open(inode, file, isdir, &outarg);
 127         if (err)
 128                 fuse_file_free(ff);
 129         else {
 130                 if (isdir)
 131                         outarg.open_flags &= ~FOPEN_DIRECT_IO;
 132                 fuse_finish_open(inode, file, ff, &outarg);
 133         }
 134
 135         return err;
 136 }
 137
 138 void fuse_release_fill(struct fuse_file *ff, u64 nodeid, int flags, int opcode)
 139 {
 140         struct fuse_req *req = ff->reserved_req;
 141         struct fuse_release_in *inarg = &req->misc.release.in;
 142
 143         inarg->fh = ff->fh;
 144         inarg->flags = flags;
 145         req->in.h.opcode = opcode;
 146         req->in.h.nodeid = nodeid;
 147         req->in.numargs = 1;
 148         req->in.args[0].size = sizeof(struct fuse_release_in);
 149         req->in.args[0].value = inarg;
 150 }
 151
 152 int fuse_release_common(struct inode *inode, struct file *file, int isdir)
 153 {
 154         struct fuse_file *ff = file->private_data;
 155         if (ff) {
 156                 struct fuse_conn *fc = get_fuse_conn(inode);
 157                 struct fuse_req *req = ff->reserved_req;
 158
 159                 fuse_release_fill(ff, get_node_id(inode), file->f_flags,
 160                                   isdir ? FUSE_RELEASEDIR : FUSE_RELEASE);
 161
 162                 /* Hold vfsmount and dentry until release is finished */
 163                 req->misc.release.vfsmount = mntget(file->f_path.mnt);
 164                 req->misc.release.dentry = dget(file->f_path.dentry);
 165
 166                 spin_lock(&fc->lock);
 167                 list_del(&ff->write_entry);
 168                 spin_unlock(&fc->lock);
 169                 /*
 170                  * Normally this will send the RELEASE request,
 171                  * however if some asynchronous READ or WRITE requests
 172                  * are outstanding, the sending will be delayed
 173                  */
 174                 fuse_file_put(ff);
 175         }
 176
 177         /* Return value is ignored by VFS */
 178         return 0;
 179 }
 180
 181 static int fuse_open(struct inode *inode, struct file *file)
 182 {
 183         return fuse_open_common(inode, file, 0);
 184 }
 185
 186 static int fuse_release(struct inode *inode, struct file *file)
 187 {
 188         return fuse_release_common(inode, file, 0);
 189 }
 190
 191 /*
 192  * Scramble the ID space with XTEA, so that the value of the files_struct
 193  * pointer is not exposed to userspace.
 194  */
 195 u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
 196 {
 197         u32 *k = fc->scramble_key;
 198         u64 v = (unsigned long) id;
 199         u32 v0 = v;
 200         u32 v1 = v >> 32;
 201         u32 sum = 0;
 202         int i;
 203
 204         for (i = 0; i < 32; i++) {
 205                 v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
 206                 sum += 0x9E3779B9;
 207                 v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
 208         }
 209
 210         return (u64) v0 + ((u64) v1 << 32);
 211 }
 212
 213 /*
 214  * Check if page is under writeback
 215  *
 216  * This is currently done by walking the list of writepage requests
 217  * for the inode, which can be pretty inefficient.
 218  */
 219 static bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
 220 {
 221         struct fuse_conn *fc = get_fuse_conn(inode);
 222         struct fuse_inode *fi = get_fuse_inode(inode);
 223         struct fuse_req *req;
 224         bool found = false;
 225
 226         spin_lock(&fc->lock);
 227         list_for_each_entry(req, &fi->writepages, writepages_entry) {
 228                 pgoff_t curr_index;
 229
 230                 BUG_ON(req->inode != inode);
 231                 curr_index = req->misc.write.in.offset >> PAGE_CACHE_SHIFT;
 232                 if (curr_index == index) {
 233                         found = true;
 234                         break;
 235                 }
 236         }
 237         spin_unlock(&fc->lock);
 238
 239         return found;
 240 }
 241
 242 /*
 243  * Wait for page writeback to be completed.
 244  *
 245  * Since fuse doesn't rely on the VM writeback tracking, this has to
 246  * use some other means.
 247  */
 248 static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
 249 {
 250         struct fuse_inode *fi = get_fuse_inode(inode);
 251
 252         wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index));
 253         return 0;
 254 }
 255
 256 static int fuse_flush(struct file *file, fl_owner_t id)
 257 {
 258         struct inode *inode = file->f_path.dentry->d_inode;
 259         struct fuse_conn *fc = get_fuse_conn(inode);
 260         struct fuse_file *ff = file->private_data;
 261         struct fuse_req *req;
 262         struct fuse_flush_in inarg;
 263         int err;
 264
 265         if (is_bad_inode(inode))
 266                 return -EIO;
 267
 268         if (fc->no_flush)
 269                 return 0;
 270
 271         req = fuse_get_req_nofail(fc, file);
 272         memset(&inarg, 0, sizeof(inarg));
 273         inarg.fh = ff->fh;
 274         inarg.lock_owner = fuse_lock_owner_id(fc, id);
 275         req->in.h.opcode = FUSE_FLUSH;
 276         req->in.h.nodeid = get_node_id(inode);
 277         req->in.numargs = 1;
 278         req->in.args[0].size = sizeof(inarg);
 279         req->in.args[0].value = &inarg;
 280         req->force = 1;
 281         request_send(fc, req);
 282         err = req->out.h.error;
 283         fuse_put_request(fc, req);
 284         if (err == -ENOSYS) {
 285                 fc->no_flush = 1;
 286                 err = 0;
 287         }
 288         return err;
 289 }
 290
 291 /*
 292  * Wait for all pending writepages on the inode to finish.
 293  *
 294  * This is currently done by blocking further writes with FUSE_NOWRITE
 295  * and waiting for all sent writes to complete.
 296  *
 297  * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
 298  * could conflict with truncation.
 299  */
 300 static void fuse_sync_writes(struct inode *inode)
 301 {
 302         fuse_set_nowrite(inode);
 303         fuse_release_nowrite(inode);
 304 }
 305
 306 int fuse_fsync_common(struct file *file, struct dentry *de, int datasync,
 307                       int isdir)
 308 {
 309         struct inode *inode = de->d_inode;
 310         struct fuse_conn *fc = get_fuse_conn(inode);
 311         struct fuse_file *ff = file->private_data;
 312         struct fuse_req *req;
 313         struct fuse_fsync_in inarg;
 314         int err;
 315
 316         if (is_bad_inode(inode))
 317                 return -EIO;
 318
 319         if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
 320                 return 0;
 321
 322         /*
 323          * Start writeback against all dirty pages of the inode, then
 324          * wait for all outstanding writes, before sending the FSYNC
 325          * request.
 326          */
 327         err = write_inode_now(inode, 0);
 328         if (err)
 329                 return err;
 330
 331         fuse_sync_writes(inode);
 332
 333         req = fuse_get_req(fc);
 334         if (IS_ERR(req))
 335                 return PTR_ERR(req);
 336
 337         memset(&inarg, 0, sizeof(inarg));
 338         inarg.fh = ff->fh;
 339         inarg.fsync_flags = datasync ? 1 : 0;
 340         req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
 341         req->in.h.nodeid = get_node_id(inode);
 342         req->in.numargs = 1;
 343         req->in.args[0].size = sizeof(inarg);
 344         req->in.args[0].value = &inarg;
 345         request_send(fc, req);
 346         err = req->out.h.error;
 347         fuse_put_request(fc, req);
 348         if (err == -ENOSYS) {
 349                 if (isdir)
 350                         fc->no_fsyncdir = 1;
 351                 else
 352                         fc->no_fsync = 1;
 353                 err = 0;
 354         }
 355         return err;
 356 }
 357
 358 static int fuse_fsync(struct file *file, struct dentry *de, int datasync)
 359 {
 360         return fuse_fsync_common(file, de, datasync, 0);
 361 }
 362
 363 void fuse_read_fill(struct fuse_req *req, struct file *file,
 364                     struct inode *inode, loff_t pos, size_t count, int opcode)
 365 {
 366         struct fuse_read_in *inarg = &req->misc.read.in;
 367         struct fuse_file *ff = file->private_data;
 368
 369         inarg->fh = ff->fh;
 370         inarg->offset = pos;
 371         inarg->size = count;
 372         inarg->flags = file->f_flags;
 373         req->in.h.opcode = opcode;
 374         req->in.h.nodeid = get_node_id(inode);
 375         req->in.numargs = 1;
 376         req->in.args[0].size = sizeof(struct fuse_read_in);
 377         req->in.args[0].value = inarg;
 378         req->out.argpages = 1;
 379         req->out.argvar = 1;
 380         req->out.numargs = 1;
 381         req->out.args[0].size = count;
 382 }
 383
 384 static size_t fuse_send_read(struct fuse_req *req, struct file *file,
 385                              struct inode *inode, loff_t pos, size_t count,
 386                              fl_owner_t owner)
 387 {
 388         struct fuse_conn *fc = get_fuse_conn(inode);
 389
 390         fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
 391         if (owner != NULL) {
 392                 struct fuse_read_in *inarg = &req->misc.read.in;
 393
 394                 inarg->read_flags |= FUSE_READ_LOCKOWNER;
 395                 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
 396         }
 397         request_send(fc, req);
 398         return req->out.args[0].size;
 399 }
 400
 401 static void fuse_read_update_size(struct inode *inode, loff_t size,
 402                                   u64 attr_ver)
 403 {
 404         struct fuse_conn *fc = get_fuse_conn(inode);
 405         struct fuse_inode *fi = get_fuse_inode(inode);
 406
 407         spin_lock(&fc->lock);
 408         if (attr_ver == fi->attr_version && size < inode->i_size) {
 409                 fi->attr_version = ++fc->attr_version;
 410                 i_size_write(inode, size);
 411         }
 412         spin_unlock(&fc->lock);
 413 }
 414
 415 static int fuse_readpage(struct file *file, struct page *page)
 416 {
 417         struct inode *inode = page->mapping->host;
 418         struct fuse_conn *fc = get_fuse_conn(inode);
 419         struct fuse_req *req;
 420         size_t num_read;
 421         loff_t pos = page_offset(page);
 422         size_t count = PAGE_CACHE_SIZE;
 423         u64 attr_ver;
 424         int err;
 425
 426         err = -EIO;
 427         if (is_bad_inode(inode))
 428                 goto out;
 429
 430         /*
 431          * Page writeback can extend beyond the liftime of the
 432          * page-cache page, so make sure we read a properly synced
 433          * page.
 434          */
 435         fuse_wait_on_page_writeback(inode, page->index);
 436
 437         req = fuse_get_req(fc);
 438         err = PTR_ERR(req);
 439         if (IS_ERR(req))
 440                 goto out;
 441
 442         attr_ver = fuse_get_attr_version(fc);
 443
 444         req->out.page_zeroing = 1;
 445         req->num_pages = 1;
 446         req->pages[0] = page;
 447         num_read = fuse_send_read(req, file, inode, pos, count, NULL);
 448         err = req->out.h.error;
 449         fuse_put_request(fc, req);
 450
 451         if (!err) {
 452                 /*
 453                  * Short read means EOF.  If file size is larger, truncate it
 454                  */
 455                 if (num_read < count)
 456                         fuse_read_update_size(inode, pos + num_read, attr_ver);
 457
 458                 SetPageUptodate(page);
 459         }
 460
 461         fuse_invalidate_attr(inode); /* atime changed */
 462  out:
 463         unlock_page(page);
 464         return err;
 465 }
 466
 467 static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
 468 {
 469         int i;
 470         size_t count = req->misc.read.in.size;
 471         size_t num_read = req->out.args[0].size;
 472         struct inode *inode = req->pages[0]->mapping->host;
 473
 474         /*
 475          * Short read means EOF.  If file size is larger, truncate it
 476          */
 477         if (!req->out.h.error && num_read < count) {
 478                 loff_t pos = page_offset(req->pages[0]) + num_read;
 479                 fuse_read_update_size(inode, pos, req->misc.read.attr_ver);
 480         }
 481
 482         fuse_invalidate_attr(inode); /* atime changed */
 483
 484         for (i = 0; i < req->num_pages; i++) {
 485                 struct page *page = req->pages[i];
 486                 if (!req->out.h.error)
 487                         SetPageUptodate(page);
 488                 else
 489                         SetPageError(page);
 490                 unlock_page(page);
 491         }
 492         if (req->ff)
 493                 fuse_file_put(req->ff);
 494         fuse_put_request(fc, req);
 495 }
 496
 497 static void fuse_send_readpages(struct fuse_req *req, struct file *file,
 498                                 struct inode *inode)
 499 {
 500         struct fuse_conn *fc = get_fuse_conn(inode);
 501         loff_t pos = page_offset(req->pages[0]);
 502         size_t count = req->num_pages << PAGE_CACHE_SHIFT;
 503         req->out.page_zeroing = 1;
 504         fuse_read_fill(req, file, inode, pos, count, FUSE_READ);
 505         req->misc.read.attr_ver = fuse_get_attr_version(fc);
 506         if (fc->async_read) {
 507                 struct fuse_file *ff = file->private_data;
 508                 req->ff = fuse_file_get(ff);
 509                 req->end = fuse_readpages_end;
 510                 request_send_background(fc, req);
 511         } else {
 512                 request_send(fc, req);
 513                 fuse_readpages_end(fc, req);
 514         }
 515 }
 516
 517 struct fuse_fill_data {
 518         struct fuse_req *req;
 519         struct file *file;
 520         struct inode *inode;
 521 };
 522
 523 static int fuse_readpages_fill(void *_data, struct page *page)
 524 {
 525         struct fuse_fill_data *data = _data;
 526         struct fuse_req *req = data->req;
 527         struct inode *inode = data->inode;
 528         struct fuse_conn *fc = get_fuse_conn(inode);
 529
 530         fuse_wait_on_page_writeback(inode, page->index);
 531
 532         if (req->num_pages &&
 533             (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
 534              (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
 535              req->pages[req->num_pages - 1]->index + 1 != page->index)) {
 536                 fuse_send_readpages(req, data->file, inode);
 537                 data->req = req = fuse_get_req(fc);
 538                 if (IS_ERR(req)) {
 539                         unlock_page(page);
 540                         return PTR_ERR(req);
 541                 }
 542         }
 543         req->pages[req->num_pages] = page;
 544         req->num_pages ++;
 545         return 0;
 546 }
 547
 548 static int fuse_readpages(struct file *file, struct address_space *mapping,
 549                           struct list_head *pages, unsigned nr_pages)
 550 {
 551         struct inode *inode = mapping->host;
 552         struct fuse_conn *fc = get_fuse_conn(inode);
 553         struct fuse_fill_data data;
 554         int err;
 555
 556         err = -EIO;
 557         if (is_bad_inode(inode))
 558                 goto out;
 559
 560         data.file = file;
 561         data.inode = inode;
 562         data.req = fuse_get_req(fc);
 563         err = PTR_ERR(data.req);
 564         if (IS_ERR(data.req))
 565                 goto out;
 566
 567         err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
 568         if (!err) {
 569                 if (data.req->num_pages)
 570                         fuse_send_readpages(data.req, file, inode);
 571                 else
 572                         fuse_put_request(fc, data.req);
 573         }
 574 out:
 575         return err;
 576 }
 577
 578 static ssize_t fuse_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
 579                                   unsigned long nr_segs, loff_t pos)
 580 {
 581         struct inode *inode = iocb->ki_filp->f_mapping->host;
 582
 583         if (pos + iov_length(iov, nr_segs) > i_size_read(inode)) {
 584                 int err;
 585                 /*
 586                  * If trying to read past EOF, make sure the i_size
 587                  * attribute is up-to-date.
 588                  */
 589                 err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL);
 590                 if (err)
 591                         return err;
 592         }
 593
 594         return generic_file_aio_read(iocb, iov, nr_segs, pos);
 595 }
 596
 597 static void fuse_write_fill(struct fuse_req *req, struct file *file,
 598                             struct fuse_file *ff, struct inode *inode,
 599                             loff_t pos, size_t count, int writepage)
 600 {
 601         struct fuse_conn *fc = get_fuse_conn(inode);
 602         struct fuse_write_in *inarg = &req->misc.write.in;
 603         struct fuse_write_out *outarg = &req->misc.write.out;
 604
 605         memset(inarg, 0, sizeof(struct fuse_write_in));
 606         inarg->fh = ff->fh;
 607         inarg->offset = pos;
 608         inarg->size = count;
 609         inarg->write_flags = writepage ? FUSE_WRITE_CACHE : 0;
 610         inarg->flags = file ? file->f_flags : 0;
 611         req->in.h.opcode = FUSE_WRITE;
 612         req->in.h.nodeid = get_node_id(inode);
 613         req->in.argpages = 1;
 614         req->in.numargs = 2;
 615         if (fc->minor < 9)
 616                 req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
 617         else
 618                 req->in.args[0].size = sizeof(struct fuse_write_in);
 619         req->in.args[0].value = inarg;
 620         req->in.args[1].size = count;
 621         req->out.numargs = 1;
 622         req->out.args[0].size = sizeof(struct fuse_write_out);
 623         req->out.args[0].value = outarg;
 624 }
 625
 626 static size_t fuse_send_write(struct fuse_req *req, struct file *file,
 627                               struct inode *inode, loff_t pos, size_t count,
 628                               fl_owner_t owner)
 629 {
 630         struct fuse_conn *fc = get_fuse_conn(inode);
 631         fuse_write_fill(req, file, file->private_data, inode, pos, count, 0);
 632         if (owner != NULL) {
 633                 struct fuse_write_in *inarg = &req->misc.write.in;
 634                 inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
 635                 inarg->lock_owner = fuse_lock_owner_id(fc, owner);
 636         }
 637         request_send(fc, req);
 638         return req->misc.write.out.size;
 639 }
 640
 641 static int fuse_write_begin(struct file *file, struct address_space *mapping,
 642                         loff_t pos, unsigned len, unsigned flags,
 643                         struct page **pagep, void **fsdata)
 644 {
 645         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
 646
 647         *pagep = __grab_cache_page(mapping, index);
 648         if (!*pagep)
 649                 return -ENOMEM;
 650         return 0;
 651 }
 652
 653 static void fuse_write_update_size(struct inode *inode, loff_t pos)
 654 {
 655         struct fuse_conn *fc = get_fuse_conn(inode);
 656         struct fuse_inode *fi = get_fuse_inode(inode);
 657
 658         spin_lock(&fc->lock);
 659         fi->attr_version = ++fc->attr_version;
 660         if (pos > inode->i_size)
 661                 i_size_write(inode, pos);
 662         spin_unlock(&fc->lock);
 663 }
 664
 665 static int fuse_buffered_write(struct file *file, struct inode *inode,
 666                                loff_t pos, unsigned count, struct page *page)
 667 {
 668         int err;
 669         size_t nres;
 670         struct fuse_conn *fc = get_fuse_conn(inode);
 671         unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
 672         struct fuse_req *req;
 673
 674         if (is_bad_inode(inode))
 675                 return -EIO;
 676
 677         /*
 678          * Make sure writepages on the same page are not mixed up with
 679          * plain writes.
 680          */
 681         fuse_wait_on_page_writeback(inode, page->index);
 682
 683         req = fuse_get_req(fc);
 684         if (IS_ERR(req))
 685                 return PTR_ERR(req);
 686
 687         req->num_pages = 1;
 688         req->pages[0] = page;
 689         req->page_offset = offset;
 690         nres = fuse_send_write(req, file, inode, pos, count, NULL);
 691         err = req->out.h.error;
 692         fuse_put_request(fc, req);
 693         if (!err && !nres)
 694                 err = -EIO;
 695         if (!err) {
 696                 pos += nres;
 697                 fuse_write_update_size(inode, pos);
 698                 if (count == PAGE_CACHE_SIZE)
 699                         SetPageUptodate(page);
 700         }
 701         fuse_invalidate_attr(inode);
 702         return err ? err : nres;
 703 }
 704
 705 static int fuse_write_end(struct file *file, struct address_space *mapping,
 706                         loff_t pos, unsigned len, unsigned copied,
 707                         struct page *page, void *fsdata)
 708 {
 709         struct inode *inode = mapping->host;
 710         int res = 0;
 711
 712         if (copied)
 713                 res = fuse_buffered_write(file, inode, pos, copied, page);
 714
 715         unlock_page(page);
 716         page_cache_release(page);
 717         return res;
 718 }
 719
 720 static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file,
 721                                     struct inode *inode, loff_t pos,
 722                                     size_t count)
 723 {
 724         size_t res;
 725         unsigned offset;
 726         unsigned i;
 727
 728         for (i = 0; i < req->num_pages; i++)
 729                 fuse_wait_on_page_writeback(inode, req->pages[i]->index);
 730
 731         res = fuse_send_write(req, file, inode, pos, count, NULL);
 732
 733         offset = req->page_offset;
 734         count = res;
 735         for (i = 0; i < req->num_pages; i++) {
 736                 struct page *page = req->pages[i];
 737
 738                 if (!req->out.h.error && !offset && count >= PAGE_CACHE_SIZE)
 739                         SetPageUptodate(page);
 740
 741                 if (count > PAGE_CACHE_SIZE - offset)
 742                         count -= PAGE_CACHE_SIZE - offset;
 743                 else
 744                         count = 0;
 745                 offset = 0;
 746
 747                 unlock_page(page);
 748                 page_cache_release(page);
 749         }
 750
 751         return res;
 752 }
 753
 754 static ssize_t fuse_fill_write_pages(struct fuse_req *req,
 755                                struct address_space *mapping,
 756                                struct iov_iter *ii, loff_t pos)
 757 {
 758         struct fuse_conn *fc = get_fuse_conn(mapping->host);
 759         unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
 760         size_t count = 0;
 761         int err;
 762
 763         req->page_offset = offset;
 764
 765         do {
 766                 size_t tmp;
 767                 struct page *page;
 768                 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
 769                 size_t bytes = min_t(size_t, PAGE_CACHE_SIZE - offset,
 770                                      iov_iter_count(ii));
 771
 772                 bytes = min_t(size_t, bytes, fc->max_write - count);
 773
 774  again:
 775                 err = -EFAULT;
 776                 if (iov_iter_fault_in_readable(ii, bytes))
 777                         break;
 778
 779                 err = -ENOMEM;
 780                 page = __grab_cache_page(mapping, index);
 781                 if (!page)
 782                         break;
 783
 784                 pagefault_disable();
 785                 tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes);
 786                 pagefault_enable();
 787                 flush_dcache_page(page);
 788
 789                 if (!tmp) {
 790                         unlock_page(page);
 791                         page_cache_release(page);
 792                         bytes = min(bytes, iov_iter_single_seg_count(ii));
 793                         goto again;
 794                 }
 795
 796                 err = 0;
 797                 req->pages[req->num_pages] = page;
 798                 req->num_pages++;
 799
 800                 iov_iter_advance(ii, tmp);
 801                 count += tmp;
 802                 pos += tmp;
 803                 offset += tmp;
 804                 if (offset == PAGE_CACHE_SIZE)
 805                         offset = 0;
 806
 807                 if (!fc->big_writes)
 808                         break;
 809         } while (iov_iter_count(ii) && count < fc->max_write &&
 810                  req->num_pages < FUSE_MAX_PAGES_PER_REQ && offset == 0);
 811
 812         return count > 0 ? count : err;
 813 }
 814
 815 static ssize_t fuse_perform_write(struct file *file,
 816                                   struct address_space *mapping,
 817                                   struct iov_iter *ii, loff_t pos)
 818 {
 819         struct inode *inode = mapping->host;
 820         struct fuse_conn *fc = get_fuse_conn(inode);
 821         int err = 0;
 822         ssize_t res = 0;
 823
 824         if (is_bad_inode(inode))
 825                 return -EIO;
 826
 827         do {
 828                 struct fuse_req *req;
 829                 ssize_t count;
 830
 831                 req = fuse_get_req(fc);
 832                 if (IS_ERR(req)) {
 833                         err = PTR_ERR(req);
 834                         break;
 835                 }
 836
 837                 count = fuse_fill_write_pages(req, mapping, ii, pos);
 838                 if (count <= 0) {
 839                         err = count;
 840                 } else {
 841                         size_t num_written;
 842
 843                         num_written = fuse_send_write_pages(req, file, inode,
 844                                                             pos, count);
 845                         err = req->out.h.error;
 846                         if (!err) {
 847                                 res += num_written;
 848                                 pos += num_written;
 849
 850                                 /* break out of the loop on short write */
 851                                 if (num_written != count)
 852                                         err = -EIO;
 853                         }
 854                 }
 855                 fuse_put_request(fc, req);
 856         } while (!err && iov_iter_count(ii));
 857
 858         if (res > 0)
 859                 fuse_write_update_size(inode, pos);
 860
 861         fuse_invalidate_attr(inode);
 862
 863         return res > 0 ? res : err;
 864 }
 865
 866 static ssize_t fuse_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
 867                                    unsigned long nr_segs, loff_t pos)
 868 {
 869         struct file *file = iocb->ki_filp;
 870         struct address_space *mapping = file->f_mapping;
 871         size_t count = 0;
 872         ssize_t written = 0;
 873         struct inode *inode = mapping->host;
 874         ssize_t err;
 875         struct iov_iter i;
 876
 877         WARN_ON(iocb->ki_pos != pos);
 878
 879         err = generic_segment_checks(iov, &nr_segs, &count, VERIFY_READ);
 880         if (err)
 881                 return err;
 882
 883         mutex_lock(&inode->i_mutex);
 884         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 885
 886         /* We can write back this queue in page reclaim */
 887         current->backing_dev_info = mapping->backing_dev_info;
 888
 889         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 890         if (err)
 891                 goto out;
 892
 893         if (count == 0)
 894                 goto out;
 895
 896         err = file_remove_suid(file);
 897         if (err)
 898                 goto out;
 899
 900         file_update_time(file);
 901
 902         iov_iter_init(&i, iov, nr_segs, count, 0);
 903         written = fuse_perform_write(file, mapping, &i, pos);
 904         if (written >= 0)
 905                 iocb->ki_pos = pos + written;
 906
 907 out:
 908         current->backing_dev_info = NULL;
 909         mutex_unlock(&inode->i_mutex);
 910
 911         return written ? written : err;
 912 }
 913
 914 static void fuse_release_user_pages(struct fuse_req *req, int write)
 915 {
 916         unsigned i;
 917
 918         for (i = 0; i < req->num_pages; i++) {
 919                 struct page *page = req->pages[i];
 920                 if (write)
 921                         set_page_dirty_lock(page);
 922                 put_page(page);
 923         }
 924 }
 925
 926 static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf,
 927                                unsigned nbytes, int write)
 928 {
 929         unsigned long user_addr = (unsigned long) buf;
 930         unsigned offset = user_addr & ~PAGE_MASK;
 931         int npages;
 932
 933         /* This doesn't work with nfsd */
 934         if (!current->mm)
 935                 return -EPERM;
 936
 937         nbytes = min(nbytes, (unsigned) FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
 938         npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT;
 939         npages = clamp(npages, 1, FUSE_MAX_PAGES_PER_REQ);
 940         down_read(&current->mm->mmap_sem);
 941         npages = get_user_pages(current, current->mm, user_addr, npages, write,
 942                                 0, req->pages, NULL);
 943         up_read(&current->mm->mmap_sem);
 944         if (npages < 0)
 945                 return npages;
 946
 947         req->num_pages = npages;
 948         req->page_offset = offset;
 949         return 0;
 950 }
 951
 952 static ssize_t fuse_direct_io(struct file *file, const char __user *buf,
 953                               size_t count, loff_t *ppos, int write)
 954 {
 955         struct inode *inode = file->f_path.dentry->d_inode;
 956         struct fuse_conn *fc = get_fuse_conn(inode);
 957         size_t nmax = write ? fc->max_write : fc->max_read;
 958         loff_t pos = *ppos;
 959         ssize_t res = 0;
 960         struct fuse_req *req;
 961
 962         if (is_bad_inode(inode))
 963                 return -EIO;
 964
 965         req = fuse_get_req(fc);
 966         if (IS_ERR(req))
 967                 return PTR_ERR(req);
 968
 969         while (count) {
 970                 size_t nres;
 971                 size_t nbytes_limit = min(count, nmax);
 972                 size_t nbytes;
 973                 int err = fuse_get_user_pages(req, buf, nbytes_limit, !write);
 974                 if (err) {
 975                         res = err;
 976                         break;
 977                 }
 978                 nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset;
 979                 nbytes = min(nbytes_limit, nbytes);
 980                 if (write)
 981                         nres = fuse_send_write(req, file, inode, pos, nbytes,
 982                                                current->files);
 983                 else
 984                         nres = fuse_send_read(req, file, inode, pos, nbytes,
 985                                               current->files);
 986                 fuse_release_user_pages(req, !write);
 987                 if (req->out.h.error) {
 988                         if (!res)
 989                                 res = req->out.h.error;
 990                         break;
 991                 } else if (nres > nbytes) {
 992                         res = -EIO;
 993                         break;
 994                 }
 995                 count -= nres;
 996                 res += nres;
 997                 pos += nres;
 998                 buf += nres;
 999                 if (nres != nbytes)
1000                         break;
1001                 if (count) {
1002                         fuse_put_request(fc, req);
1003                         req = fuse_get_req(fc);
1004                         if (IS_ERR(req))
1005                                 break;
1006                 }
1007         }
1008         fuse_put_request(fc, req);
1009         if (res > 0) {
1010                 if (write)
1011                         fuse_write_update_size(inode, pos);
1012                 *ppos = pos;
1013         }
1014         fuse_invalidate_attr(inode);
1015
1016         return res;
1017 }
1018
1019 static ssize_t fuse_direct_read(struct file *file, char __user *buf,
1020                                      size_t count, loff_t *ppos)
1021 {
1022         return fuse_direct_io(file, buf, count, ppos, 0);
1023 }
1024
1025 static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
1026                                  size_t count, loff_t *ppos)
1027 {
1028         struct inode *inode = file->f_path.dentry->d_inode;
1029         ssize_t res;
1030         /* Don't allow parallel writes to the same file */
1031         mutex_lock(&inode->i_mutex);
1032         res = generic_write_checks(file, ppos, &count, 0);
1033         if (!res)
1034                 res = fuse_direct_io(file, buf, count, ppos, 1);
1035         mutex_unlock(&inode->i_mutex);
1036         return res;
1037 }
1038
1039 static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req)
1040 {
1041         __free_page(req->pages[0]);
1042         fuse_file_put(req->ff);
1043         fuse_put_request(fc, req);
1044 }
1045
1046 static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
1047 {
1048         struct inode *inode = req->inode;
1049         struct fuse_inode *fi = get_fuse_inode(inode);
1050         struct backing_dev_info *bdi = inode->i_mapping->backing_dev_info;
1051
1052         list_del(&req->writepages_entry);
1053         dec_bdi_stat(bdi, BDI_WRITEBACK);
1054         dec_zone_page_state(req->pages[0], NR_WRITEBACK_TEMP);
1055         bdi_writeout_inc(bdi);
1056         wake_up(&fi->page_waitq);
1057 }
1058
1059 /* Called under fc->lock, may release and reacquire it */
1060 static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req)
1061 {
1062         struct fuse_inode *fi = get_fuse_inode(req->inode);
1063         loff_t size = i_size_read(req->inode);
1064         struct fuse_write_in *inarg = &req->misc.write.in;
1065
1066         if (!fc->connected)
1067                 goto out_free;
1068
1069         if (inarg->offset + PAGE_CACHE_SIZE <= size) {
1070                 inarg->size = PAGE_CACHE_SIZE;
1071         } else if (inarg->offset < size) {
1072                 inarg->size = size & (PAGE_CACHE_SIZE - 1);
1073         } else {
1074                 /* Got truncated off completely */
1075                 goto out_free;
1076         }
1077
1078         req->in.args[1].size = inarg->size;
1079         fi->writectr++;
1080         request_send_background_locked(fc, req);
1081         return;
1082
1083  out_free:
1084         fuse_writepage_finish(fc, req);
1085         spin_unlock(&fc->lock);
1086         fuse_writepage_free(fc, req);
1087         spin_lock(&fc->lock);
1088 }
1089
1090 /*
1091  * If fi->writectr is positive (no truncate or fsync going on) send
1092  * all queued writepage requests.
1093  *
1094  * Called with fc->lock
1095  */
1096 void fuse_flush_writepages(struct inode *inode)
1097 {
1098         struct fuse_conn *fc = get_fuse_conn(inode);
1099         struct fuse_inode *fi = get_fuse_inode(inode);
1100         struct fuse_req *req;
1101
1102         while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) {
1103                 req = list_entry(fi->queued_writes.next, struct fuse_req, list);
1104                 list_del_init(&req->list);
1105                 fuse_send_writepage(fc, req);
1106         }
1107 }
1108
1109 static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
1110 {
1111         struct inode *inode = req->inode;
1112         struct fuse_inode *fi = get_fuse_inode(inode);
1113
1114         mapping_set_error(inode->i_mapping, req->out.h.error);
1115         spin_lock(&fc->lock);
1116         fi->writectr--;
1117         fuse_writepage_finish(fc, req);
1118         spin_unlock(&fc->lock);
1119         fuse_writepage_free(fc, req);
1120 }
1121
1122 static int fuse_writepage_locked(struct page *page)
1123 {
1124         struct address_space *mapping = page->mapping;
1125         struct inode *inode = mapping->host;
1126         struct fuse_conn *fc = get_fuse_conn(inode);
1127         struct fuse_inode *fi = get_fuse_inode(inode);
1128         struct fuse_req *req;
1129         struct fuse_file *ff;
1130         struct page *tmp_page;
1131
1132         set_page_writeback(page);
1133
1134         req = fuse_request_alloc_nofs();
1135         if (!req)
1136                 goto err;
1137
1138         tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1139         if (!tmp_page)
1140                 goto err_free;
1141
1142         spin_lock(&fc->lock);
1143         BUG_ON(list_empty(&fi->write_files));
1144         ff = list_entry(fi->write_files.next, struct fuse_file, write_entry);
1145         req->ff = fuse_file_get(ff);
1146         spin_unlock(&fc->lock);
1147
1148         fuse_write_fill(req, NULL, ff, inode, page_offset(page), 0, 1);
1149
1150         copy_highpage(tmp_page, page);
1151         req->num_pages = 1;
1152         req->pages[0] = tmp_page;
1153         req->page_offset = 0;
1154         req->end = fuse_writepage_end;
1155         req->inode = inode;
1156
1157         inc_bdi_stat(mapping->backing_dev_info, BDI_WRITEBACK);
1158         inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP);
1159         end_page_writeback(page);
1160
1161         spin_lock(&fc->lock);
1162         list_add(&req->writepages_entry, &fi->writepages);
1163         list_add_tail(&req->list, &fi->queued_writes);
1164         fuse_flush_writepages(inode);
1165         spin_unlock(&fc->lock);
1166
1167         return 0;
1168
1169 err_free:
1170         fuse_request_free(req);
1171 err:
1172         end_page_writeback(page);
1173         return -ENOMEM;
1174 }
1175
1176 static int fuse_writepage(struct page *page, struct writeback_control *wbc)
1177 {
1178         int err;
1179
1180         err = fuse_writepage_locked(page);
1181         unlock_page(page);
1182
1183         return err;
1184 }
1185
1186 static int fuse_launder_page(struct page *page)
1187 {
1188         int err = 0;
1189         if (clear_page_dirty_for_io(page)) {
1190                 struct inode *inode = page->mapping->host;
1191                 err = fuse_writepage_locked(page);
1192                 if (!err)
1193                         fuse_wait_on_page_writeback(inode, page->index);
1194         }
1195         return err;
1196 }
1197
1198 /*
1199  * Write back dirty pages now, because there may not be any suitable
1200  * open files later
1201  */
1202 static void fuse_vma_close(struct vm_area_struct *vma)
1203 {
1204         filemap_write_and_wait(vma->vm_file->f_mapping);
1205 }
1206
1207 /*
1208  * Wait for writeback against this page to complete before allowing it
1209  * to be marked dirty again, and hence written back again, possibly
1210  * before the previous writepage completed.
1211  *
1212  * Block here, instead of in ->writepage(), so that the userspace fs
1213  * can only block processes actually operating on the filesystem.
1214  *
1215  * Otherwise unprivileged userspace fs would be able to block
1216  * unrelated:
1217  *
1218  * - page migration
1219  * - sync(2)
1220  * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
1221  */
1222 static int fuse_page_mkwrite(struct vm_area_struct *vma, struct page *page)
1223 {
1224         /*
1225          * Don't use page->mapping as it may become NULL from a
1226          * concurrent truncate.
1227          */
1228         struct inode *inode = vma->vm_file->f_mapping->host;
1229
1230         fuse_wait_on_page_writeback(inode, page->index);
1231         return 0;
1232 }
1233
1234 static struct vm_operations_struct fuse_file_vm_ops = {
1235         .close          = fuse_vma_close,
1236         .fault          = filemap_fault,
1237         .page_mkwrite   = fuse_page_mkwrite,
1238 };
1239
1240 static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
1241 {
1242         if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) {
1243                 struct inode *inode = file->f_dentry->d_inode;
1244                 struct fuse_conn *fc = get_fuse_conn(inode);
1245                 struct fuse_inode *fi = get_fuse_inode(inode);
1246                 struct fuse_file *ff = file->private_data;
1247                 /*
1248                  * file may be written through mmap, so chain it onto the
1249                  * inodes's write_file list
1250                  */
1251                 spin_lock(&fc->lock);
1252                 if (list_empty(&ff->write_entry))
1253                         list_add(&ff->write_entry, &fi->write_files);
1254                 spin_unlock(&fc->lock);
1255         }
1256         file_accessed(file);
1257         vma->vm_ops = &fuse_file_vm_ops;
1258         return 0;
1259 }
1260
1261 static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
1262                                   struct file_lock *fl)
1263 {
1264         switch (ffl->type) {
1265         case F_UNLCK:
1266                 break;
1267
1268         case F_RDLCK:
1269         case F_WRLCK:
1270                 if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
1271                     ffl->end < ffl->start)
1272                         return -EIO;
1273
1274                 fl->fl_start = ffl->start;
1275                 fl->fl_end = ffl->end;
1276                 fl->fl_pid = ffl->pid;
1277                 break;
1278
1279         default:
1280                 return -EIO;
1281         }
1282         fl->fl_type = ffl->type;
1283         return 0;
1284 }
1285
1286 static void fuse_lk_fill(struct fuse_req *req, struct file *file,
1287                          const struct file_lock *fl, int opcode, pid_t pid,
1288                          int flock)
1289 {
1290         struct inode *inode = file->f_path.dentry->d_inode;
1291         struct fuse_conn *fc = get_fuse_conn(inode);
1292         struct fuse_file *ff = file->private_data;
1293         struct fuse_lk_in *arg = &req->misc.lk_in;
1294
1295         arg->fh = ff->fh;
1296         arg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
1297         arg->lk.start = fl->fl_start;
1298         arg->lk.end = fl->fl_end;
1299         arg->lk.type = fl->fl_type;
1300         arg->lk.pid = pid;
1301         if (flock)
1302                 arg->lk_flags |= FUSE_LK_FLOCK;
1303         req->in.h.opcode = opcode;
1304         req->in.h.nodeid = get_node_id(inode);
1305         req->in.numargs = 1;
1306         req->in.args[0].size = sizeof(*arg);
1307         req->in.args[0].value = arg;
1308 }
1309
1310 static int fuse_getlk(struct file *file, struct file_lock *fl)
1311 {
1312         struct inode *inode = file->f_path.dentry->d_inode;
1313         struct fuse_conn *fc = get_fuse_conn(inode);
1314         struct fuse_req *req;
1315         struct fuse_lk_out outarg;
1316         int err;
1317
1318         req = fuse_get_req(fc);
1319         if (IS_ERR(req))
1320                 return PTR_ERR(req);
1321
1322         fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0);
1323         req->out.numargs = 1;
1324         req->out.args[0].size = sizeof(outarg);
1325         req->out.args[0].value = &outarg;
1326         request_send(fc, req);
1327         err = req->out.h.error;
1328         fuse_put_request(fc, req);
1329         if (!err)
1330                 err = convert_fuse_file_lock(&outarg.lk, fl);
1331
1332         return err;
1333 }
1334
1335 static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
1336 {
1337         struct inode *inode = file->f_path.dentry->d_inode;
1338         struct fuse_conn *fc = get_fuse_conn(inode);
1339         struct fuse_req *req;
1340         int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
1341         pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
1342         int err;
1343
1344         if (fl->fl_lmops && fl->fl_lmops->fl_grant) {
1345                 /* NLM needs asynchronous locks, which we don't support yet */
1346                 return -ENOLCK;
1347         }
1348
1349         /* Unlock on close is handled by the flush method */
1350         if (fl->fl_flags & FL_CLOSE)
1351                 return 0;
1352
1353         req = fuse_get_req(fc);
1354         if (IS_ERR(req))
1355                 return PTR_ERR(req);
1356
1357         fuse_lk_fill(req, file, fl, opcode, pid, flock);
1358         request_send(fc, req);
1359         err = req->out.h.error;
1360         /* locking is restartable */
1361         if (err == -EINTR)
1362                 err = -ERESTARTSYS;
1363         fuse_put_request(fc, req);
1364         return err;
1365 }
1366
1367 static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
1368 {
1369         struct inode *inode = file->f_path.dentry->d_inode;
1370         struct fuse_conn *fc = get_fuse_conn(inode);
1371         int err;
1372
1373         if (cmd == F_CANCELLK) {
1374                 err = 0;
1375         } else if (cmd == F_GETLK) {
1376                 if (fc->no_lock) {
1377                         posix_test_lock(file, fl);
1378                         err = 0;
1379                 } else
1380                         err = fuse_getlk(file, fl);
1381         } else {
1382                 if (fc->no_lock)
1383                         err = posix_lock_file(file, fl, NULL);
1384                 else
1385                         err = fuse_setlk(file, fl, 0);
1386         }
1387         return err;
1388 }
1389
1390 static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
1391 {
1392         struct inode *inode = file->f_path.dentry->d_inode;
1393         struct fuse_conn *fc = get_fuse_conn(inode);
1394         int err;
1395
1396         if (fc->no_lock) {
1397                 err = flock_lock_file_wait(file, fl);
1398         } else {
1399                 /* emulate flock with POSIX locks */
1400                 fl->fl_owner = (fl_owner_t) file;
1401                 err = fuse_setlk(file, fl, 1);
1402         }
1403
1404         return err;
1405 }
1406
1407 static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
1408 {
1409         struct inode *inode = mapping->host;
1410         struct fuse_conn *fc = get_fuse_conn(inode);
1411         struct fuse_req *req;
1412         struct fuse_bmap_in inarg;
1413         struct fuse_bmap_out outarg;
1414         int err;
1415
1416         if (!inode->i_sb->s_bdev || fc->no_bmap)
1417                 return 0;
1418
1419         req = fuse_get_req(fc);
1420         if (IS_ERR(req))
1421                 return 0;
1422
1423         memset(&inarg, 0, sizeof(inarg));
1424         inarg.block = block;
1425         inarg.blocksize = inode->i_sb->s_blocksize;
1426         req->in.h.opcode = FUSE_BMAP;
1427         req->in.h.nodeid = get_node_id(inode);
1428         req->in.numargs = 1;
1429         req->in.args[0].size = sizeof(inarg);
1430         req->in.args[0].value = &inarg;
1431         req->out.numargs = 1;
1432         req->out.args[0].size = sizeof(outarg);
1433         req->out.args[0].value = &outarg;
1434         request_send(fc, req);
1435         err = req->out.h.error;
1436         fuse_put_request(fc, req);
1437         if (err == -ENOSYS)
1438                 fc->no_bmap = 1;
1439
1440         return err ? 0 : outarg.block;
1441 }
1442
1443 static loff_t fuse_file_llseek(struct file *file, loff_t offset, int origin)
1444 {
1445         loff_t retval;
1446         struct inode *inode = file->f_path.dentry->d_inode;
1447
1448         mutex_lock(&inode->i_mutex);
1449         switch (origin) {
1450         case SEEK_END:
1451                 offset += i_size_read(inode);
1452                 break;
1453         case SEEK_CUR:
1454                 offset += file->f_pos;
1455         }
1456         retval = -EINVAL;
1457         if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
1458                 if (offset != file->f_pos) {
1459                         file->f_pos = offset;
1460                         file->f_version = 0;
1461                 }
1462                 retval = offset;
1463         }
1464         mutex_unlock(&inode->i_mutex);
1465         return retval;
1466 }
1467
1468 static const struct file_operations fuse_file_operations = {
1469         .llseek         = fuse_file_llseek,
1470         .read           = do_sync_read,
1471         .aio_read       = fuse_file_aio_read,
1472         .write          = do_sync_write,
1473         .aio_write      = fuse_file_aio_write,
1474         .mmap           = fuse_file_mmap,
1475         .open           = fuse_open,
1476         .flush          = fuse_flush,
1477         .release        = fuse_release,
1478         .fsync          = fuse_fsync,
1479         .lock           = fuse_file_lock,
1480         .flock          = fuse_file_flock,
1481         .splice_read    = generic_file_splice_read,
1482 };
1483
1484 static const struct file_operations fuse_direct_io_file_operations = {
1485         .llseek         = fuse_file_llseek,
1486         .read           = fuse_direct_read,
1487         .write          = fuse_direct_write,
1488         .open           = fuse_open,
1489         .flush          = fuse_flush,
1490         .release        = fuse_release,
1491         .fsync          = fuse_fsync,
1492         .lock           = fuse_file_lock,
1493         .flock          = fuse_file_flock,
1494         /* no mmap and splice_read */
1495 };
1496
1497 static const struct address_space_operations fuse_file_aops  = {
1498         .readpage       = fuse_readpage,
1499         .writepage      = fuse_writepage,
1500         .launder_page   = fuse_launder_page,
1501         .write_begin    = fuse_write_begin,
1502         .write_end      = fuse_write_end,
1503         .readpages      = fuse_readpages,
1504         .set_page_dirty = __set_page_dirty_nobuffers,
1505         .bmap           = fuse_bmap,
1506 };
1507
1508 void fuse_init_file_inode(struct inode *inode)
1509 {
1510         inode->i_fop = &fuse_file_operations;
1511         inode->i_data.a_ops = &fuse_file_aops;
1512 }