linux/fs/fuse/file.c
<<
>>
Prefs
   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/kernel.h>
  14#include <linux/sched.h>
  15#include <linux/module.h>
  16#include <linux/compat.h>
  17#include <linux/swap.h>
  18#include <linux/aio.h>
  19#include <linux/falloc.h>
  20
  21static const struct file_operations fuse_direct_io_file_operations;
  22
  23static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
  24                          int opcode, struct fuse_open_out *outargp)
  25{
  26        struct fuse_open_in inarg;
  27        FUSE_ARGS(args);
  28
  29        memset(&inarg, 0, sizeof(inarg));
  30        inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY);
  31        if (!fc->atomic_o_trunc)
  32                inarg.flags &= ~O_TRUNC;
  33        args.in.h.opcode = opcode;
  34        args.in.h.nodeid = nodeid;
  35        args.in.numargs = 1;
  36        args.in.args[0].size = sizeof(inarg);
  37        args.in.args[0].value = &inarg;
  38        args.out.numargs = 1;
  39        args.out.args[0].size = sizeof(*outargp);
  40        args.out.args[0].value = outargp;
  41
  42        return fuse_simple_request(fc, &args);
  43}
  44
  45struct fuse_file *fuse_file_alloc(struct fuse_conn *fc)
  46{
  47        struct fuse_file *ff;
  48
  49        ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL);
  50        if (unlikely(!ff))
  51                return NULL;
  52
  53        ff->fc = fc;
  54        ff->reserved_req = fuse_request_alloc(0);
  55        if (unlikely(!ff->reserved_req)) {
  56                kfree(ff);
  57                return NULL;
  58        }
  59
  60        INIT_LIST_HEAD(&ff->write_entry);
  61        atomic_set(&ff->count, 0);
  62        RB_CLEAR_NODE(&ff->polled_node);
  63        init_waitqueue_head(&ff->poll_wait);
  64
  65        spin_lock(&fc->lock);
  66        ff->kh = ++fc->khctr;
  67        spin_unlock(&fc->lock);
  68
  69        return ff;
  70}
  71
  72void fuse_file_free(struct fuse_file *ff)
  73{
  74        fuse_request_free(ff->reserved_req);
  75        kfree(ff);
  76}
  77
  78struct fuse_file *fuse_file_get(struct fuse_file *ff)
  79{
  80        atomic_inc(&ff->count);
  81        return ff;
  82}
  83
  84static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req)
  85{
  86        iput(req->misc.release.inode);
  87}
  88
  89static void fuse_file_put(struct fuse_file *ff, bool sync)
  90{
  91        if (atomic_dec_and_test(&ff->count)) {
  92                struct fuse_req *req = ff->reserved_req;
  93
  94                if (ff->fc->no_open) {
  95                        /*
  96                         * Drop the release request when client does not
  97                         * implement 'open'
  98                         */
  99                        req->background = 0;
 100                        iput(req->misc.release.inode);
 101                        fuse_put_request(ff->fc, req);
 102                } else if (sync) {
 103                        req->background = 0;
 104                        fuse_request_send(ff->fc, req);
 105                        iput(req->misc.release.inode);
 106                        fuse_put_request(ff->fc, req);
 107                } else {
 108                        req->end = fuse_release_end;
 109                        req->background = 1;
 110                        fuse_request_send_background(ff->fc, req);
 111                }
 112                kfree(ff);
 113        }
 114}
 115
 116int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file,
 117                 bool isdir)
 118{
 119        struct fuse_file *ff;
 120        int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN;
 121
 122        ff = fuse_file_alloc(fc);
 123        if (!ff)
 124                return -ENOMEM;
 125
 126        ff->fh = 0;
 127        ff->open_flags = FOPEN_KEEP_CACHE; /* Default for no-open */
 128        if (!fc->no_open || isdir) {
 129                struct fuse_open_out outarg;
 130                int err;
 131
 132                err = fuse_send_open(fc, nodeid, file, opcode, &outarg);
 133                if (!err) {
 134                        ff->fh = outarg.fh;
 135                        ff->open_flags = outarg.open_flags;
 136
 137                } else if (err != -ENOSYS || isdir) {
 138                        fuse_file_free(ff);
 139                        return err;
 140                } else {
 141                        fc->no_open = 1;
 142                }
 143        }
 144
 145        if (isdir)
 146                ff->open_flags &= ~FOPEN_DIRECT_IO;
 147
 148        ff->nodeid = nodeid;
 149        file->private_data = fuse_file_get(ff);
 150
 151        return 0;
 152}
 153EXPORT_SYMBOL_GPL(fuse_do_open);
 154
 155static void fuse_link_write_file(struct file *file)
 156{
 157        struct inode *inode = file_inode(file);
 158        struct fuse_conn *fc = get_fuse_conn(inode);
 159        struct fuse_inode *fi = get_fuse_inode(inode);
 160        struct fuse_file *ff = file->private_data;
 161        /*
 162         * file may be written through mmap, so chain it onto the
 163         * inodes's write_file list
 164         */
 165        spin_lock(&fc->lock);
 166        if (list_empty(&ff->write_entry))
 167                list_add(&ff->write_entry, &fi->write_files);
 168        spin_unlock(&fc->lock);
 169}
 170
 171void fuse_finish_open(struct inode *inode, struct file *file)
 172{
 173        struct fuse_file *ff = file->private_data;
 174        struct fuse_conn *fc = get_fuse_conn(inode);
 175
 176        if (ff->open_flags & FOPEN_DIRECT_IO)
 177                file->f_op = &fuse_direct_io_file_operations;
 178        if (!(ff->open_flags & FOPEN_KEEP_CACHE))
 179                invalidate_inode_pages2(inode->i_mapping);
 180        if (ff->open_flags & FOPEN_NONSEEKABLE)
 181                nonseekable_open(inode, file);
 182        if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) {
 183                struct fuse_inode *fi = get_fuse_inode(inode);
 184
 185                spin_lock(&fc->lock);
 186                fi->attr_version = ++fc->attr_version;
 187                i_size_write(inode, 0);
 188                spin_unlock(&fc->lock);
 189                fuse_invalidate_attr(inode);
 190                if (fc->writeback_cache)
 191                        file_update_time(file);
 192        }
 193        if ((file->f_mode & FMODE_WRITE) && fc->writeback_cache)
 194                fuse_link_write_file(file);
 195}
 196
 197int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
 198{
 199        struct fuse_conn *fc = get_fuse_conn(inode);
 200        int err;
 201        bool lock_inode = (file->f_flags & O_TRUNC) &&
 202                          fc->atomic_o_trunc &&
 203                          fc->writeback_cache;
 204
 205        err = generic_file_open(inode, file);
 206        if (err)
 207                return err;
 208
 209        if (lock_inode)
 210                mutex_lock(&inode->i_mutex);
 211
 212        err = fuse_do_open(fc, get_node_id(inode), file, isdir);
 213
 214        if (!err)
 215                fuse_finish_open(inode, file);
 216
 217        if (lock_inode)
 218                mutex_unlock(&inode->i_mutex);
 219
 220        return err;
 221}
 222
 223static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode)
 224{
 225        struct fuse_conn *fc = ff->fc;
 226        struct fuse_req *req = ff->reserved_req;
 227        struct fuse_release_in *inarg = &req->misc.release.in;
 228
 229        spin_lock(&fc->lock);
 230        list_del(&ff->write_entry);
 231        if (!RB_EMPTY_NODE(&ff->polled_node))
 232                rb_erase(&ff->polled_node, &fc->polled_files);
 233        spin_unlock(&fc->lock);
 234
 235        wake_up_interruptible_all(&ff->poll_wait);
 236
 237        inarg->fh = ff->fh;
 238        inarg->flags = flags;
 239        req->in.h.opcode = opcode;
 240        req->in.h.nodeid = ff->nodeid;
 241        req->in.numargs = 1;
 242        req->in.args[0].size = sizeof(struct fuse_release_in);
 243        req->in.args[0].value = inarg;
 244}
 245
 246void fuse_release_common(struct file *file, int opcode)
 247{
 248        struct fuse_file *ff;
 249        struct fuse_req *req;
 250
 251        ff = file->private_data;
 252        if (unlikely(!ff))
 253                return;
 254
 255        req = ff->reserved_req;
 256        fuse_prepare_release(ff, file->f_flags, opcode);
 257
 258        if (ff->flock) {
 259                struct fuse_release_in *inarg = &req->misc.release.in;
 260                inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK;
 261                inarg->lock_owner = fuse_lock_owner_id(ff->fc,
 262                                                       (fl_owner_t) file);
 263        }
 264        /* Hold inode until release is finished */
 265        req->misc.release.inode = igrab(file_inode(file));
 266
 267        /*
 268         * Normally this will send the RELEASE request, however if
 269         * some asynchronous READ or WRITE requests are outstanding,
 270         * the sending will be delayed.
 271         *
 272         * Make the release synchronous if this is a fuseblk mount,
 273         * synchronous RELEASE is allowed (and desirable) in this case
 274         * because the server can be trusted not to screw up.
 275         */
 276        fuse_file_put(ff, ff->fc->destroy_req != NULL);
 277}
 278
 279static int fuse_open(struct inode *inode, struct file *file)
 280{
 281        return fuse_open_common(inode, file, false);
 282}
 283
 284static int fuse_release(struct inode *inode, struct file *file)
 285{
 286        struct fuse_conn *fc = get_fuse_conn(inode);
 287
 288        /* see fuse_vma_close() for !writeback_cache case */
 289        if (fc->writeback_cache)
 290                write_inode_now(inode, 1);
 291
 292        fuse_release_common(file, FUSE_RELEASE);
 293
 294        /* return value is ignored by VFS */
 295        return 0;
 296}
 297
 298void fuse_sync_release(struct fuse_file *ff, int flags)
 299{
 300        WARN_ON(atomic_read(&ff->count) > 1);
 301        fuse_prepare_release(ff, flags, FUSE_RELEASE);
 302        ff->reserved_req->force = 1;
 303        ff->reserved_req->background = 0;
 304        fuse_request_send(ff->fc, ff->reserved_req);
 305        fuse_put_request(ff->fc, ff->reserved_req);
 306        kfree(ff);
 307}
 308EXPORT_SYMBOL_GPL(fuse_sync_release);
 309
 310/*
 311 * Scramble the ID space with XTEA, so that the value of the files_struct
 312 * pointer is not exposed to userspace.
 313 */
 314u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id)
 315{
 316        u32 *k = fc->scramble_key;
 317        u64 v = (unsigned long) id;
 318        u32 v0 = v;
 319        u32 v1 = v >> 32;
 320        u32 sum = 0;
 321        int i;
 322
 323        for (i = 0; i < 32; i++) {
 324                v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]);
 325                sum += 0x9E3779B9;
 326                v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]);
 327        }
 328
 329        return (u64) v0 + ((u64) v1 << 32);
 330}
 331
 332/*
 333 * Check if any page in a range is under writeback
 334 *
 335 * This is currently done by walking the list of writepage requests
 336 * for the inode, which can be pretty inefficient.
 337 */
 338static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from,
 339                                   pgoff_t idx_to)
 340{
 341        struct fuse_conn *fc = get_fuse_conn(inode);
 342        struct fuse_inode *fi = get_fuse_inode(inode);
 343        struct fuse_req *req;
 344        bool found = false;
 345
 346        spin_lock(&fc->lock);
 347        list_for_each_entry(req, &fi->writepages, writepages_entry) {
 348                pgoff_t curr_index;
 349
 350                BUG_ON(req->inode != inode);
 351                curr_index = req->misc.write.in.offset >> PAGE_CACHE_SHIFT;
 352                if (idx_from < curr_index + req->num_pages &&
 353                    curr_index <= idx_to) {
 354                        found = true;
 355                        break;
 356                }
 357        }
 358        spin_unlock(&fc->lock);
 359
 360        return found;
 361}
 362
 363static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index)
 364{
 365        return fuse_range_is_writeback(inode, index, index);
 366}
 367
 368/*
 369 * Wait for page writeback to be completed.
 370 *
 371 * Since fuse doesn't rely on the VM writeback tracking, this has to
 372 * use some other means.
 373 */
 374static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index)
 375{
 376        struct fuse_inode *fi = get_fuse_inode(inode);
 377
 378        wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index));
 379        return 0;
 380}
 381
 382/*
 383 * Wait for all pending writepages on the inode to finish.
 384 *
 385 * This is currently done by blocking further writes with FUSE_NOWRITE
 386 * and waiting for all sent writes to complete.
 387 *
 388 * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage
 389 * could conflict with truncation.
 390 */
 391static void fuse_sync_writes(struct inode *inode)
 392{
 393        fuse_set_nowrite(inode);
 394        fuse_release_nowrite(inode);
 395}
 396
 397static int fuse_flush(struct file *file, fl_owner_t id)
 398{
 399        struct inode *inode = file_inode(file);
 400        struct fuse_conn *fc = get_fuse_conn(inode);
 401        struct fuse_file *ff = file->private_data;
 402        struct fuse_req *req;
 403        struct fuse_flush_in inarg;
 404        int err;
 405
 406        if (is_bad_inode(inode))
 407                return -EIO;
 408
 409        if (fc->no_flush)
 410                return 0;
 411
 412        err = write_inode_now(inode, 1);
 413        if (err)
 414                return err;
 415
 416        mutex_lock(&inode->i_mutex);
 417        fuse_sync_writes(inode);
 418        mutex_unlock(&inode->i_mutex);
 419
 420        req = fuse_get_req_nofail_nopages(fc, file);
 421        memset(&inarg, 0, sizeof(inarg));
 422        inarg.fh = ff->fh;
 423        inarg.lock_owner = fuse_lock_owner_id(fc, id);
 424        req->in.h.opcode = FUSE_FLUSH;
 425        req->in.h.nodeid = get_node_id(inode);
 426        req->in.numargs = 1;
 427        req->in.args[0].size = sizeof(inarg);
 428        req->in.args[0].value = &inarg;
 429        req->force = 1;
 430        fuse_request_send(fc, req);
 431        err = req->out.h.error;
 432        fuse_put_request(fc, req);
 433        if (err == -ENOSYS) {
 434                fc->no_flush = 1;
 435                err = 0;
 436        }
 437        return err;
 438}
 439
 440int fuse_fsync_common(struct file *file, loff_t start, loff_t end,
 441                      int datasync, int isdir)
 442{
 443        struct inode *inode = file->f_mapping->host;
 444        struct fuse_conn *fc = get_fuse_conn(inode);
 445        struct fuse_file *ff = file->private_data;
 446        FUSE_ARGS(args);
 447        struct fuse_fsync_in inarg;
 448        int err;
 449
 450        if (is_bad_inode(inode))
 451                return -EIO;
 452
 453        mutex_lock(&inode->i_mutex);
 454
 455        /*
 456         * Start writeback against all dirty pages of the inode, then
 457         * wait for all outstanding writes, before sending the FSYNC
 458         * request.
 459         */
 460        err = filemap_write_and_wait_range(inode->i_mapping, start, end);
 461        if (err)
 462                goto out;
 463
 464        fuse_sync_writes(inode);
 465        err = sync_inode_metadata(inode, 1);
 466        if (err)
 467                goto out;
 468
 469        if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir))
 470                goto out;
 471
 472        memset(&inarg, 0, sizeof(inarg));
 473        inarg.fh = ff->fh;
 474        inarg.fsync_flags = datasync ? 1 : 0;
 475        args.in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC;
 476        args.in.h.nodeid = get_node_id(inode);
 477        args.in.numargs = 1;
 478        args.in.args[0].size = sizeof(inarg);
 479        args.in.args[0].value = &inarg;
 480        err = fuse_simple_request(fc, &args);
 481        if (err == -ENOSYS) {
 482                if (isdir)
 483                        fc->no_fsyncdir = 1;
 484                else
 485                        fc->no_fsync = 1;
 486                err = 0;
 487        }
 488out:
 489        mutex_unlock(&inode->i_mutex);
 490        return err;
 491}
 492
 493static int fuse_fsync(struct file *file, loff_t start, loff_t end,
 494                      int datasync)
 495{
 496        return fuse_fsync_common(file, start, end, datasync, 0);
 497}
 498
 499void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos,
 500                    size_t count, int opcode)
 501{
 502        struct fuse_read_in *inarg = &req->misc.read.in;
 503        struct fuse_file *ff = file->private_data;
 504
 505        inarg->fh = ff->fh;
 506        inarg->offset = pos;
 507        inarg->size = count;
 508        inarg->flags = file->f_flags;
 509        req->in.h.opcode = opcode;
 510        req->in.h.nodeid = ff->nodeid;
 511        req->in.numargs = 1;
 512        req->in.args[0].size = sizeof(struct fuse_read_in);
 513        req->in.args[0].value = inarg;
 514        req->out.argvar = 1;
 515        req->out.numargs = 1;
 516        req->out.args[0].size = count;
 517}
 518
 519static void fuse_release_user_pages(struct fuse_req *req, int write)
 520{
 521        unsigned i;
 522
 523        for (i = 0; i < req->num_pages; i++) {
 524                struct page *page = req->pages[i];
 525                if (write)
 526                        set_page_dirty_lock(page);
 527                put_page(page);
 528        }
 529}
 530
 531/**
 532 * In case of short read, the caller sets 'pos' to the position of
 533 * actual end of fuse request in IO request. Otherwise, if bytes_requested
 534 * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1.
 535 *
 536 * An example:
 537 * User requested DIO read of 64K. It was splitted into two 32K fuse requests,
 538 * both submitted asynchronously. The first of them was ACKed by userspace as
 539 * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The
 540 * second request was ACKed as short, e.g. only 1K was read, resulting in
 541 * pos == 33K.
 542 *
 543 * Thus, when all fuse requests are completed, the minimal non-negative 'pos'
 544 * will be equal to the length of the longest contiguous fragment of
 545 * transferred data starting from the beginning of IO request.
 546 */
 547static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos)
 548{
 549        int left;
 550
 551        spin_lock(&io->lock);
 552        if (err)
 553                io->err = io->err ? : err;
 554        else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes))
 555                io->bytes = pos;
 556
 557        left = --io->reqs;
 558        spin_unlock(&io->lock);
 559
 560        if (!left) {
 561                long res;
 562
 563                if (io->err)
 564                        res = io->err;
 565                else if (io->bytes >= 0 && io->write)
 566                        res = -EIO;
 567                else {
 568                        res = io->bytes < 0 ? io->size : io->bytes;
 569
 570                        if (!is_sync_kiocb(io->iocb)) {
 571                                struct inode *inode = file_inode(io->iocb->ki_filp);
 572                                struct fuse_conn *fc = get_fuse_conn(inode);
 573                                struct fuse_inode *fi = get_fuse_inode(inode);
 574
 575                                spin_lock(&fc->lock);
 576                                fi->attr_version = ++fc->attr_version;
 577                                spin_unlock(&fc->lock);
 578                        }
 579                }
 580
 581                aio_complete(io->iocb, res, 0);
 582                kfree(io);
 583        }
 584}
 585
 586static void fuse_aio_complete_req(struct fuse_conn *fc, struct fuse_req *req)
 587{
 588        struct fuse_io_priv *io = req->io;
 589        ssize_t pos = -1;
 590
 591        fuse_release_user_pages(req, !io->write);
 592
 593        if (io->write) {
 594                if (req->misc.write.in.size != req->misc.write.out.size)
 595                        pos = req->misc.write.in.offset - io->offset +
 596                                req->misc.write.out.size;
 597        } else {
 598                if (req->misc.read.in.size != req->out.args[0].size)
 599                        pos = req->misc.read.in.offset - io->offset +
 600                                req->out.args[0].size;
 601        }
 602
 603        fuse_aio_complete(io, req->out.h.error, pos);
 604}
 605
 606static size_t fuse_async_req_send(struct fuse_conn *fc, struct fuse_req *req,
 607                size_t num_bytes, struct fuse_io_priv *io)
 608{
 609        spin_lock(&io->lock);
 610        io->size += num_bytes;
 611        io->reqs++;
 612        spin_unlock(&io->lock);
 613
 614        req->io = io;
 615        req->end = fuse_aio_complete_req;
 616
 617        __fuse_get_request(req);
 618        fuse_request_send_background(fc, req);
 619
 620        return num_bytes;
 621}
 622
 623static size_t fuse_send_read(struct fuse_req *req, struct fuse_io_priv *io,
 624                             loff_t pos, size_t count, fl_owner_t owner)
 625{
 626        struct file *file = io->file;
 627        struct fuse_file *ff = file->private_data;
 628        struct fuse_conn *fc = ff->fc;
 629
 630        fuse_read_fill(req, file, pos, count, FUSE_READ);
 631        if (owner != NULL) {
 632                struct fuse_read_in *inarg = &req->misc.read.in;
 633
 634                inarg->read_flags |= FUSE_READ_LOCKOWNER;
 635                inarg->lock_owner = fuse_lock_owner_id(fc, owner);
 636        }
 637
 638        if (io->async)
 639                return fuse_async_req_send(fc, req, count, io);
 640
 641        fuse_request_send(fc, req);
 642        return req->out.args[0].size;
 643}
 644
 645static void fuse_read_update_size(struct inode *inode, loff_t size,
 646                                  u64 attr_ver)
 647{
 648        struct fuse_conn *fc = get_fuse_conn(inode);
 649        struct fuse_inode *fi = get_fuse_inode(inode);
 650
 651        spin_lock(&fc->lock);
 652        if (attr_ver == fi->attr_version && size < inode->i_size &&
 653            !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 654                fi->attr_version = ++fc->attr_version;
 655                i_size_write(inode, size);
 656        }
 657        spin_unlock(&fc->lock);
 658}
 659
 660static void fuse_short_read(struct fuse_req *req, struct inode *inode,
 661                            u64 attr_ver)
 662{
 663        size_t num_read = req->out.args[0].size;
 664        struct fuse_conn *fc = get_fuse_conn(inode);
 665
 666        if (fc->writeback_cache) {
 667                /*
 668                 * A hole in a file. Some data after the hole are in page cache,
 669                 * but have not reached the client fs yet. So, the hole is not
 670                 * present there.
 671                 */
 672                int i;
 673                int start_idx = num_read >> PAGE_CACHE_SHIFT;
 674                size_t off = num_read & (PAGE_CACHE_SIZE - 1);
 675
 676                for (i = start_idx; i < req->num_pages; i++) {
 677                        zero_user_segment(req->pages[i], off, PAGE_CACHE_SIZE);
 678                        off = 0;
 679                }
 680        } else {
 681                loff_t pos = page_offset(req->pages[0]) + num_read;
 682                fuse_read_update_size(inode, pos, attr_ver);
 683        }
 684}
 685
 686static int fuse_do_readpage(struct file *file, struct page *page)
 687{
 688        struct fuse_io_priv io = { .async = 0, .file = file };
 689        struct inode *inode = page->mapping->host;
 690        struct fuse_conn *fc = get_fuse_conn(inode);
 691        struct fuse_req *req;
 692        size_t num_read;
 693        loff_t pos = page_offset(page);
 694        size_t count = PAGE_CACHE_SIZE;
 695        u64 attr_ver;
 696        int err;
 697
 698        /*
 699         * Page writeback can extend beyond the lifetime of the
 700         * page-cache page, so make sure we read a properly synced
 701         * page.
 702         */
 703        fuse_wait_on_page_writeback(inode, page->index);
 704
 705        req = fuse_get_req(fc, 1);
 706        if (IS_ERR(req))
 707                return PTR_ERR(req);
 708
 709        attr_ver = fuse_get_attr_version(fc);
 710
 711        req->out.page_zeroing = 1;
 712        req->out.argpages = 1;
 713        req->num_pages = 1;
 714        req->pages[0] = page;
 715        req->page_descs[0].length = count;
 716        num_read = fuse_send_read(req, &io, pos, count, NULL);
 717        err = req->out.h.error;
 718
 719        if (!err) {
 720                /*
 721                 * Short read means EOF.  If file size is larger, truncate it
 722                 */
 723                if (num_read < count)
 724                        fuse_short_read(req, inode, attr_ver);
 725
 726                SetPageUptodate(page);
 727        }
 728
 729        fuse_put_request(fc, req);
 730
 731        return err;
 732}
 733
 734static int fuse_readpage(struct file *file, struct page *page)
 735{
 736        struct inode *inode = page->mapping->host;
 737        int err;
 738
 739        err = -EIO;
 740        if (is_bad_inode(inode))
 741                goto out;
 742
 743        err = fuse_do_readpage(file, page);
 744        fuse_invalidate_atime(inode);
 745 out:
 746        unlock_page(page);
 747        return err;
 748}
 749
 750static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req)
 751{
 752        int i;
 753        size_t count = req->misc.read.in.size;
 754        size_t num_read = req->out.args[0].size;
 755        struct address_space *mapping = NULL;
 756
 757        for (i = 0; mapping == NULL && i < req->num_pages; i++)
 758                mapping = req->pages[i]->mapping;
 759
 760        if (mapping) {
 761                struct inode *inode = mapping->host;
 762
 763                /*
 764                 * Short read means EOF. If file size is larger, truncate it
 765                 */
 766                if (!req->out.h.error && num_read < count)
 767                        fuse_short_read(req, inode, req->misc.read.attr_ver);
 768
 769                fuse_invalidate_atime(inode);
 770        }
 771
 772        for (i = 0; i < req->num_pages; i++) {
 773                struct page *page = req->pages[i];
 774                if (!req->out.h.error)
 775                        SetPageUptodate(page);
 776                else
 777                        SetPageError(page);
 778                unlock_page(page);
 779                page_cache_release(page);
 780        }
 781        if (req->ff)
 782                fuse_file_put(req->ff, false);
 783}
 784
 785static void fuse_send_readpages(struct fuse_req *req, struct file *file)
 786{
 787        struct fuse_file *ff = file->private_data;
 788        struct fuse_conn *fc = ff->fc;
 789        loff_t pos = page_offset(req->pages[0]);
 790        size_t count = req->num_pages << PAGE_CACHE_SHIFT;
 791
 792        req->out.argpages = 1;
 793        req->out.page_zeroing = 1;
 794        req->out.page_replace = 1;
 795        fuse_read_fill(req, file, pos, count, FUSE_READ);
 796        req->misc.read.attr_ver = fuse_get_attr_version(fc);
 797        if (fc->async_read) {
 798                req->ff = fuse_file_get(ff);
 799                req->end = fuse_readpages_end;
 800                fuse_request_send_background(fc, req);
 801        } else {
 802                fuse_request_send(fc, req);
 803                fuse_readpages_end(fc, req);
 804                fuse_put_request(fc, req);
 805        }
 806}
 807
 808struct fuse_fill_data {
 809        struct fuse_req *req;
 810        struct file *file;
 811        struct inode *inode;
 812        unsigned nr_pages;
 813};
 814
 815static int fuse_readpages_fill(void *_data, struct page *page)
 816{
 817        struct fuse_fill_data *data = _data;
 818        struct fuse_req *req = data->req;
 819        struct inode *inode = data->inode;
 820        struct fuse_conn *fc = get_fuse_conn(inode);
 821
 822        fuse_wait_on_page_writeback(inode, page->index);
 823
 824        if (req->num_pages &&
 825            (req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
 826             (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read ||
 827             req->pages[req->num_pages - 1]->index + 1 != page->index)) {
 828                int nr_alloc = min_t(unsigned, data->nr_pages,
 829                                     FUSE_MAX_PAGES_PER_REQ);
 830                fuse_send_readpages(req, data->file);
 831                if (fc->async_read)
 832                        req = fuse_get_req_for_background(fc, nr_alloc);
 833                else
 834                        req = fuse_get_req(fc, nr_alloc);
 835
 836                data->req = req;
 837                if (IS_ERR(req)) {
 838                        unlock_page(page);
 839                        return PTR_ERR(req);
 840                }
 841        }
 842
 843        if (WARN_ON(req->num_pages >= req->max_pages)) {
 844                fuse_put_request(fc, req);
 845                return -EIO;
 846        }
 847
 848        page_cache_get(page);
 849        req->pages[req->num_pages] = page;
 850        req->page_descs[req->num_pages].length = PAGE_SIZE;
 851        req->num_pages++;
 852        data->nr_pages--;
 853        return 0;
 854}
 855
 856static int fuse_readpages(struct file *file, struct address_space *mapping,
 857                          struct list_head *pages, unsigned nr_pages)
 858{
 859        struct inode *inode = mapping->host;
 860        struct fuse_conn *fc = get_fuse_conn(inode);
 861        struct fuse_fill_data data;
 862        int err;
 863        int nr_alloc = min_t(unsigned, nr_pages, FUSE_MAX_PAGES_PER_REQ);
 864
 865        err = -EIO;
 866        if (is_bad_inode(inode))
 867                goto out;
 868
 869        data.file = file;
 870        data.inode = inode;
 871        if (fc->async_read)
 872                data.req = fuse_get_req_for_background(fc, nr_alloc);
 873        else
 874                data.req = fuse_get_req(fc, nr_alloc);
 875        data.nr_pages = nr_pages;
 876        err = PTR_ERR(data.req);
 877        if (IS_ERR(data.req))
 878                goto out;
 879
 880        err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data);
 881        if (!err) {
 882                if (data.req->num_pages)
 883                        fuse_send_readpages(data.req, file);
 884                else
 885                        fuse_put_request(fc, data.req);
 886        }
 887out:
 888        return err;
 889}
 890
 891static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 892{
 893        struct inode *inode = iocb->ki_filp->f_mapping->host;
 894        struct fuse_conn *fc = get_fuse_conn(inode);
 895
 896        /*
 897         * In auto invalidate mode, always update attributes on read.
 898         * Otherwise, only update if we attempt to read past EOF (to ensure
 899         * i_size is up to date).
 900         */
 901        if (fc->auto_inval_data ||
 902            (iocb->ki_pos + iov_iter_count(to) > i_size_read(inode))) {
 903                int err;
 904                err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL);
 905                if (err)
 906                        return err;
 907        }
 908
 909        return generic_file_read_iter(iocb, to);
 910}
 911
 912static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff,
 913                            loff_t pos, size_t count)
 914{
 915        struct fuse_write_in *inarg = &req->misc.write.in;
 916        struct fuse_write_out *outarg = &req->misc.write.out;
 917
 918        inarg->fh = ff->fh;
 919        inarg->offset = pos;
 920        inarg->size = count;
 921        req->in.h.opcode = FUSE_WRITE;
 922        req->in.h.nodeid = ff->nodeid;
 923        req->in.numargs = 2;
 924        if (ff->fc->minor < 9)
 925                req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE;
 926        else
 927                req->in.args[0].size = sizeof(struct fuse_write_in);
 928        req->in.args[0].value = inarg;
 929        req->in.args[1].size = count;
 930        req->out.numargs = 1;
 931        req->out.args[0].size = sizeof(struct fuse_write_out);
 932        req->out.args[0].value = outarg;
 933}
 934
 935static size_t fuse_send_write(struct fuse_req *req, struct fuse_io_priv *io,
 936                              loff_t pos, size_t count, fl_owner_t owner)
 937{
 938        struct file *file = io->file;
 939        struct fuse_file *ff = file->private_data;
 940        struct fuse_conn *fc = ff->fc;
 941        struct fuse_write_in *inarg = &req->misc.write.in;
 942
 943        fuse_write_fill(req, ff, pos, count);
 944        inarg->flags = file->f_flags;
 945        if (owner != NULL) {
 946                inarg->write_flags |= FUSE_WRITE_LOCKOWNER;
 947                inarg->lock_owner = fuse_lock_owner_id(fc, owner);
 948        }
 949
 950        if (io->async)
 951                return fuse_async_req_send(fc, req, count, io);
 952
 953        fuse_request_send(fc, req);
 954        return req->misc.write.out.size;
 955}
 956
 957bool fuse_write_update_size(struct inode *inode, loff_t pos)
 958{
 959        struct fuse_conn *fc = get_fuse_conn(inode);
 960        struct fuse_inode *fi = get_fuse_inode(inode);
 961        bool ret = false;
 962
 963        spin_lock(&fc->lock);
 964        fi->attr_version = ++fc->attr_version;
 965        if (pos > inode->i_size) {
 966                i_size_write(inode, pos);
 967                ret = true;
 968        }
 969        spin_unlock(&fc->lock);
 970
 971        return ret;
 972}
 973
 974static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file,
 975                                    struct inode *inode, loff_t pos,
 976                                    size_t count)
 977{
 978        size_t res;
 979        unsigned offset;
 980        unsigned i;
 981        struct fuse_io_priv io = { .async = 0, .file = file };
 982
 983        for (i = 0; i < req->num_pages; i++)
 984                fuse_wait_on_page_writeback(inode, req->pages[i]->index);
 985
 986        res = fuse_send_write(req, &io, pos, count, NULL);
 987
 988        offset = req->page_descs[0].offset;
 989        count = res;
 990        for (i = 0; i < req->num_pages; i++) {
 991                struct page *page = req->pages[i];
 992
 993                if (!req->out.h.error && !offset && count >= PAGE_CACHE_SIZE)
 994                        SetPageUptodate(page);
 995
 996                if (count > PAGE_CACHE_SIZE - offset)
 997                        count -= PAGE_CACHE_SIZE - offset;
 998                else
 999                        count = 0;
1000                offset = 0;
1001
1002                unlock_page(page);
1003                page_cache_release(page);
1004        }
1005
1006        return res;
1007}
1008
1009static ssize_t fuse_fill_write_pages(struct fuse_req *req,
1010                               struct address_space *mapping,
1011                               struct iov_iter *ii, loff_t pos)
1012{
1013        struct fuse_conn *fc = get_fuse_conn(mapping->host);
1014        unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1015        size_t count = 0;
1016        int err;
1017
1018        req->in.argpages = 1;
1019        req->page_descs[0].offset = offset;
1020
1021        do {
1022                size_t tmp;
1023                struct page *page;
1024                pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1025                size_t bytes = min_t(size_t, PAGE_CACHE_SIZE - offset,
1026                                     iov_iter_count(ii));
1027
1028                bytes = min_t(size_t, bytes, fc->max_write - count);
1029
1030 again:
1031                err = -EFAULT;
1032                if (iov_iter_fault_in_readable(ii, bytes))
1033                        break;
1034
1035                err = -ENOMEM;
1036                page = grab_cache_page_write_begin(mapping, index, 0);
1037                if (!page)
1038                        break;
1039
1040                if (mapping_writably_mapped(mapping))
1041                        flush_dcache_page(page);
1042
1043                tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes);
1044                flush_dcache_page(page);
1045
1046                if (!tmp) {
1047                        unlock_page(page);
1048                        page_cache_release(page);
1049                        bytes = min(bytes, iov_iter_single_seg_count(ii));
1050                        goto again;
1051                }
1052
1053                err = 0;
1054                req->pages[req->num_pages] = page;
1055                req->page_descs[req->num_pages].length = tmp;
1056                req->num_pages++;
1057
1058                iov_iter_advance(ii, tmp);
1059                count += tmp;
1060                pos += tmp;
1061                offset += tmp;
1062                if (offset == PAGE_CACHE_SIZE)
1063                        offset = 0;
1064
1065                if (!fc->big_writes)
1066                        break;
1067        } while (iov_iter_count(ii) && count < fc->max_write &&
1068                 req->num_pages < req->max_pages && offset == 0);
1069
1070        return count > 0 ? count : err;
1071}
1072
1073static inline unsigned fuse_wr_pages(loff_t pos, size_t len)
1074{
1075        return min_t(unsigned,
1076                     ((pos + len - 1) >> PAGE_CACHE_SHIFT) -
1077                     (pos >> PAGE_CACHE_SHIFT) + 1,
1078                     FUSE_MAX_PAGES_PER_REQ);
1079}
1080
1081static ssize_t fuse_perform_write(struct file *file,
1082                                  struct address_space *mapping,
1083                                  struct iov_iter *ii, loff_t pos)
1084{
1085        struct inode *inode = mapping->host;
1086        struct fuse_conn *fc = get_fuse_conn(inode);
1087        struct fuse_inode *fi = get_fuse_inode(inode);
1088        int err = 0;
1089        ssize_t res = 0;
1090
1091        if (is_bad_inode(inode))
1092                return -EIO;
1093
1094        if (inode->i_size < pos + iov_iter_count(ii))
1095                set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1096
1097        do {
1098                struct fuse_req *req;
1099                ssize_t count;
1100                unsigned nr_pages = fuse_wr_pages(pos, iov_iter_count(ii));
1101
1102                req = fuse_get_req(fc, nr_pages);
1103                if (IS_ERR(req)) {
1104                        err = PTR_ERR(req);
1105                        break;
1106                }
1107
1108                count = fuse_fill_write_pages(req, mapping, ii, pos);
1109                if (count <= 0) {
1110                        err = count;
1111                } else {
1112                        size_t num_written;
1113
1114                        num_written = fuse_send_write_pages(req, file, inode,
1115                                                            pos, count);
1116                        err = req->out.h.error;
1117                        if (!err) {
1118                                res += num_written;
1119                                pos += num_written;
1120
1121                                /* break out of the loop on short write */
1122                                if (num_written != count)
1123                                        err = -EIO;
1124                        }
1125                }
1126                fuse_put_request(fc, req);
1127        } while (!err && iov_iter_count(ii));
1128
1129        if (res > 0)
1130                fuse_write_update_size(inode, pos);
1131
1132        clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
1133        fuse_invalidate_attr(inode);
1134
1135        return res > 0 ? res : err;
1136}
1137
1138static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1139{
1140        struct file *file = iocb->ki_filp;
1141        struct address_space *mapping = file->f_mapping;
1142        size_t count = iov_iter_count(from);
1143        ssize_t written = 0;
1144        ssize_t written_buffered = 0;
1145        struct inode *inode = mapping->host;
1146        ssize_t err;
1147        loff_t endbyte = 0;
1148        loff_t pos = iocb->ki_pos;
1149
1150        if (get_fuse_conn(inode)->writeback_cache) {
1151                /* Update size (EOF optimization) and mode (SUID clearing) */
1152                err = fuse_update_attributes(mapping->host, NULL, file, NULL);
1153                if (err)
1154                        return err;
1155
1156                return generic_file_write_iter(iocb, from);
1157        }
1158
1159        mutex_lock(&inode->i_mutex);
1160
1161        /* We can write back this queue in page reclaim */
1162        current->backing_dev_info = mapping->backing_dev_info;
1163
1164        err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1165        if (err)
1166                goto out;
1167
1168        if (count == 0)
1169                goto out;
1170
1171        iov_iter_truncate(from, count);
1172        err = file_remove_suid(file);
1173        if (err)
1174                goto out;
1175
1176        err = file_update_time(file);
1177        if (err)
1178                goto out;
1179
1180        if (file->f_flags & O_DIRECT) {
1181                written = generic_file_direct_write(iocb, from, pos);
1182                if (written < 0 || !iov_iter_count(from))
1183                        goto out;
1184
1185                pos += written;
1186
1187                written_buffered = fuse_perform_write(file, mapping, from, pos);
1188                if (written_buffered < 0) {
1189                        err = written_buffered;
1190                        goto out;
1191                }
1192                endbyte = pos + written_buffered - 1;
1193
1194                err = filemap_write_and_wait_range(file->f_mapping, pos,
1195                                                   endbyte);
1196                if (err)
1197                        goto out;
1198
1199                invalidate_mapping_pages(file->f_mapping,
1200                                         pos >> PAGE_CACHE_SHIFT,
1201                                         endbyte >> PAGE_CACHE_SHIFT);
1202
1203                written += written_buffered;
1204                iocb->ki_pos = pos + written_buffered;
1205        } else {
1206                written = fuse_perform_write(file, mapping, from, pos);
1207                if (written >= 0)
1208                        iocb->ki_pos = pos + written;
1209        }
1210out:
1211        current->backing_dev_info = NULL;
1212        mutex_unlock(&inode->i_mutex);
1213
1214        return written ? written : err;
1215}
1216
1217static inline void fuse_page_descs_length_init(struct fuse_req *req,
1218                unsigned index, unsigned nr_pages)
1219{
1220        int i;
1221
1222        for (i = index; i < index + nr_pages; i++)
1223                req->page_descs[i].length = PAGE_SIZE -
1224                        req->page_descs[i].offset;
1225}
1226
1227static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii)
1228{
1229        return (unsigned long)ii->iov->iov_base + ii->iov_offset;
1230}
1231
1232static inline size_t fuse_get_frag_size(const struct iov_iter *ii,
1233                                        size_t max_size)
1234{
1235        return min(iov_iter_single_seg_count(ii), max_size);
1236}
1237
1238static int fuse_get_user_pages(struct fuse_req *req, struct iov_iter *ii,
1239                               size_t *nbytesp, int write)
1240{
1241        size_t nbytes = 0;  /* # bytes already packed in req */
1242
1243        /* Special case for kernel I/O: can copy directly into the buffer */
1244        if (ii->type & ITER_KVEC) {
1245                unsigned long user_addr = fuse_get_user_addr(ii);
1246                size_t frag_size = fuse_get_frag_size(ii, *nbytesp);
1247
1248                if (write)
1249                        req->in.args[1].value = (void *) user_addr;
1250                else
1251                        req->out.args[0].value = (void *) user_addr;
1252
1253                iov_iter_advance(ii, frag_size);
1254                *nbytesp = frag_size;
1255                return 0;
1256        }
1257
1258        while (nbytes < *nbytesp && req->num_pages < req->max_pages) {
1259                unsigned npages;
1260                size_t start;
1261                ssize_t ret = iov_iter_get_pages(ii,
1262                                        &req->pages[req->num_pages],
1263                                        *nbytesp - nbytes,
1264                                        req->max_pages - req->num_pages,
1265                                        &start);
1266                if (ret < 0)
1267                        return ret;
1268
1269                iov_iter_advance(ii, ret);
1270                nbytes += ret;
1271
1272                ret += start;
1273                npages = (ret + PAGE_SIZE - 1) / PAGE_SIZE;
1274
1275                req->page_descs[req->num_pages].offset = start;
1276                fuse_page_descs_length_init(req, req->num_pages, npages);
1277
1278                req->num_pages += npages;
1279                req->page_descs[req->num_pages - 1].length -=
1280                        (PAGE_SIZE - ret) & (PAGE_SIZE - 1);
1281        }
1282
1283        if (write)
1284                req->in.argpages = 1;
1285        else
1286                req->out.argpages = 1;
1287
1288        *nbytesp = nbytes;
1289
1290        return 0;
1291}
1292
1293static inline int fuse_iter_npages(const struct iov_iter *ii_p)
1294{
1295        return iov_iter_npages(ii_p, FUSE_MAX_PAGES_PER_REQ);
1296}
1297
1298ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter,
1299                       loff_t *ppos, int flags)
1300{
1301        int write = flags & FUSE_DIO_WRITE;
1302        int cuse = flags & FUSE_DIO_CUSE;
1303        struct file *file = io->file;
1304        struct inode *inode = file->f_mapping->host;
1305        struct fuse_file *ff = file->private_data;
1306        struct fuse_conn *fc = ff->fc;
1307        size_t nmax = write ? fc->max_write : fc->max_read;
1308        loff_t pos = *ppos;
1309        size_t count = iov_iter_count(iter);
1310        pgoff_t idx_from = pos >> PAGE_CACHE_SHIFT;
1311        pgoff_t idx_to = (pos + count - 1) >> PAGE_CACHE_SHIFT;
1312        ssize_t res = 0;
1313        struct fuse_req *req;
1314
1315        if (io->async)
1316                req = fuse_get_req_for_background(fc, fuse_iter_npages(iter));
1317        else
1318                req = fuse_get_req(fc, fuse_iter_npages(iter));
1319        if (IS_ERR(req))
1320                return PTR_ERR(req);
1321
1322        if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) {
1323                if (!write)
1324                        mutex_lock(&inode->i_mutex);
1325                fuse_sync_writes(inode);
1326                if (!write)
1327                        mutex_unlock(&inode->i_mutex);
1328        }
1329
1330        while (count) {
1331                size_t nres;
1332                fl_owner_t owner = current->files;
1333                size_t nbytes = min(count, nmax);
1334                int err = fuse_get_user_pages(req, iter, &nbytes, write);
1335                if (err) {
1336                        res = err;
1337                        break;
1338                }
1339
1340                if (write)
1341                        nres = fuse_send_write(req, io, pos, nbytes, owner);
1342                else
1343                        nres = fuse_send_read(req, io, pos, nbytes, owner);
1344
1345                if (!io->async)
1346                        fuse_release_user_pages(req, !write);
1347                if (req->out.h.error) {
1348                        if (!res)
1349                                res = req->out.h.error;
1350                        break;
1351                } else if (nres > nbytes) {
1352                        res = -EIO;
1353                        break;
1354                }
1355                count -= nres;
1356                res += nres;
1357                pos += nres;
1358                if (nres != nbytes)
1359                        break;
1360                if (count) {
1361                        fuse_put_request(fc, req);
1362                        if (io->async)
1363                                req = fuse_get_req_for_background(fc,
1364                                        fuse_iter_npages(iter));
1365                        else
1366                                req = fuse_get_req(fc, fuse_iter_npages(iter));
1367                        if (IS_ERR(req))
1368                                break;
1369                }
1370        }
1371        if (!IS_ERR(req))
1372                fuse_put_request(fc, req);
1373        if (res > 0)
1374                *ppos = pos;
1375
1376        return res;
1377}
1378EXPORT_SYMBOL_GPL(fuse_direct_io);
1379
1380static ssize_t __fuse_direct_read(struct fuse_io_priv *io,
1381                                  struct iov_iter *iter,
1382                                  loff_t *ppos)
1383{
1384        ssize_t res;
1385        struct file *file = io->file;
1386        struct inode *inode = file_inode(file);
1387
1388        if (is_bad_inode(inode))
1389                return -EIO;
1390
1391        res = fuse_direct_io(io, iter, ppos, 0);
1392
1393        fuse_invalidate_attr(inode);
1394
1395        return res;
1396}
1397
1398static ssize_t fuse_direct_read(struct file *file, char __user *buf,
1399                                     size_t count, loff_t *ppos)
1400{
1401        struct fuse_io_priv io = { .async = 0, .file = file };
1402        struct iovec iov = { .iov_base = buf, .iov_len = count };
1403        struct iov_iter ii;
1404        iov_iter_init(&ii, READ, &iov, 1, count);
1405        return __fuse_direct_read(&io, &ii, ppos);
1406}
1407
1408static ssize_t __fuse_direct_write(struct fuse_io_priv *io,
1409                                   struct iov_iter *iter,
1410                                   loff_t *ppos)
1411{
1412        struct file *file = io->file;
1413        struct inode *inode = file_inode(file);
1414        size_t count = iov_iter_count(iter);
1415        ssize_t res;
1416
1417
1418        res = generic_write_checks(file, ppos, &count, 0);
1419        if (!res) {
1420                iov_iter_truncate(iter, count);
1421                res = fuse_direct_io(io, iter, ppos, FUSE_DIO_WRITE);
1422        }
1423
1424        fuse_invalidate_attr(inode);
1425
1426        return res;
1427}
1428
1429static ssize_t fuse_direct_write(struct file *file, const char __user *buf,
1430                                 size_t count, loff_t *ppos)
1431{
1432        struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = count };
1433        struct inode *inode = file_inode(file);
1434        ssize_t res;
1435        struct fuse_io_priv io = { .async = 0, .file = file };
1436        struct iov_iter ii;
1437        iov_iter_init(&ii, WRITE, &iov, 1, count);
1438
1439        if (is_bad_inode(inode))
1440                return -EIO;
1441
1442        /* Don't allow parallel writes to the same file */
1443        mutex_lock(&inode->i_mutex);
1444        res = __fuse_direct_write(&io, &ii, ppos);
1445        if (res > 0)
1446                fuse_write_update_size(inode, *ppos);
1447        mutex_unlock(&inode->i_mutex);
1448
1449        return res;
1450}
1451
1452static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req)
1453{
1454        int i;
1455
1456        for (i = 0; i < req->num_pages; i++)
1457                __free_page(req->pages[i]);
1458
1459        if (req->ff)
1460                fuse_file_put(req->ff, false);
1461}
1462
1463static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req)
1464{
1465        struct inode *inode = req->inode;
1466        struct fuse_inode *fi = get_fuse_inode(inode);
1467        struct backing_dev_info *bdi = inode->i_mapping->backing_dev_info;
1468        int i;
1469
1470        list_del(&req->writepages_entry);
1471        for (i = 0; i < req->num_pages; i++) {
1472                dec_bdi_stat(bdi, BDI_WRITEBACK);
1473                dec_zone_page_state(req->pages[i], NR_WRITEBACK_TEMP);
1474                bdi_writeout_inc(bdi);
1475        }
1476        wake_up(&fi->page_waitq);
1477}
1478
1479/* Called under fc->lock, may release and reacquire it */
1480static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req,
1481                                loff_t size)
1482__releases(fc->lock)
1483__acquires(fc->lock)
1484{
1485        struct fuse_inode *fi = get_fuse_inode(req->inode);
1486        struct fuse_write_in *inarg = &req->misc.write.in;
1487        __u64 data_size = req->num_pages * PAGE_CACHE_SIZE;
1488
1489        if (!fc->connected)
1490                goto out_free;
1491
1492        if (inarg->offset + data_size <= size) {
1493                inarg->size = data_size;
1494        } else if (inarg->offset < size) {
1495                inarg->size = size - inarg->offset;
1496        } else {
1497                /* Got truncated off completely */
1498                goto out_free;
1499        }
1500
1501        req->in.args[1].size = inarg->size;
1502        fi->writectr++;
1503        fuse_request_send_background_locked(fc, req);
1504        return;
1505
1506 out_free:
1507        fuse_writepage_finish(fc, req);
1508        spin_unlock(&fc->lock);
1509        fuse_writepage_free(fc, req);
1510        fuse_put_request(fc, req);
1511        spin_lock(&fc->lock);
1512}
1513
1514/*
1515 * If fi->writectr is positive (no truncate or fsync going on) send
1516 * all queued writepage requests.
1517 *
1518 * Called with fc->lock
1519 */
1520void fuse_flush_writepages(struct inode *inode)
1521__releases(fc->lock)
1522__acquires(fc->lock)
1523{
1524        struct fuse_conn *fc = get_fuse_conn(inode);
1525        struct fuse_inode *fi = get_fuse_inode(inode);
1526        size_t crop = i_size_read(inode);
1527        struct fuse_req *req;
1528
1529        while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) {
1530                req = list_entry(fi->queued_writes.next, struct fuse_req, list);
1531                list_del_init(&req->list);
1532                fuse_send_writepage(fc, req, crop);
1533        }
1534}
1535
1536static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req)
1537{
1538        struct inode *inode = req->inode;
1539        struct fuse_inode *fi = get_fuse_inode(inode);
1540
1541        mapping_set_error(inode->i_mapping, req->out.h.error);
1542        spin_lock(&fc->lock);
1543        while (req->misc.write.next) {
1544                struct fuse_conn *fc = get_fuse_conn(inode);
1545                struct fuse_write_in *inarg = &req->misc.write.in;
1546                struct fuse_req *next = req->misc.write.next;
1547                req->misc.write.next = next->misc.write.next;
1548                next->misc.write.next = NULL;
1549                next->ff = fuse_file_get(req->ff);
1550                list_add(&next->writepages_entry, &fi->writepages);
1551
1552                /*
1553                 * Skip fuse_flush_writepages() to make it easy to crop requests
1554                 * based on primary request size.
1555                 *
1556                 * 1st case (trivial): there are no concurrent activities using
1557                 * fuse_set/release_nowrite.  Then we're on safe side because
1558                 * fuse_flush_writepages() would call fuse_send_writepage()
1559                 * anyway.
1560                 *
1561                 * 2nd case: someone called fuse_set_nowrite and it is waiting
1562                 * now for completion of all in-flight requests.  This happens
1563                 * rarely and no more than once per page, so this should be
1564                 * okay.
1565                 *
1566                 * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle
1567                 * of fuse_set_nowrite..fuse_release_nowrite section.  The fact
1568                 * that fuse_set_nowrite returned implies that all in-flight
1569                 * requests were completed along with all of their secondary
1570                 * requests.  Further primary requests are blocked by negative
1571                 * writectr.  Hence there cannot be any in-flight requests and
1572                 * no invocations of fuse_writepage_end() while we're in
1573                 * fuse_set_nowrite..fuse_release_nowrite section.
1574                 */
1575                fuse_send_writepage(fc, next, inarg->offset + inarg->size);
1576        }
1577        fi->writectr--;
1578        fuse_writepage_finish(fc, req);
1579        spin_unlock(&fc->lock);
1580        fuse_writepage_free(fc, req);
1581}
1582
1583static struct fuse_file *__fuse_write_file_get(struct fuse_conn *fc,
1584                                               struct fuse_inode *fi)
1585{
1586        struct fuse_file *ff = NULL;
1587
1588        spin_lock(&fc->lock);
1589        if (!list_empty(&fi->write_files)) {
1590                ff = list_entry(fi->write_files.next, struct fuse_file,
1591                                write_entry);
1592                fuse_file_get(ff);
1593        }
1594        spin_unlock(&fc->lock);
1595
1596        return ff;
1597}
1598
1599static struct fuse_file *fuse_write_file_get(struct fuse_conn *fc,
1600                                             struct fuse_inode *fi)
1601{
1602        struct fuse_file *ff = __fuse_write_file_get(fc, fi);
1603        WARN_ON(!ff);
1604        return ff;
1605}
1606
1607int fuse_write_inode(struct inode *inode, struct writeback_control *wbc)
1608{
1609        struct fuse_conn *fc = get_fuse_conn(inode);
1610        struct fuse_inode *fi = get_fuse_inode(inode);
1611        struct fuse_file *ff;
1612        int err;
1613
1614        ff = __fuse_write_file_get(fc, fi);
1615        err = fuse_flush_times(inode, ff);
1616        if (ff)
1617                fuse_file_put(ff, 0);
1618
1619        return err;
1620}
1621
1622static int fuse_writepage_locked(struct page *page)
1623{
1624        struct address_space *mapping = page->mapping;
1625        struct inode *inode = mapping->host;
1626        struct fuse_conn *fc = get_fuse_conn(inode);
1627        struct fuse_inode *fi = get_fuse_inode(inode);
1628        struct fuse_req *req;
1629        struct page *tmp_page;
1630        int error = -ENOMEM;
1631
1632        set_page_writeback(page);
1633
1634        req = fuse_request_alloc_nofs(1);
1635        if (!req)
1636                goto err;
1637
1638        req->background = 1; /* writeback always goes to bg_queue */
1639        tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1640        if (!tmp_page)
1641                goto err_free;
1642
1643        error = -EIO;
1644        req->ff = fuse_write_file_get(fc, fi);
1645        if (!req->ff)
1646                goto err_nofile;
1647
1648        fuse_write_fill(req, req->ff, page_offset(page), 0);
1649
1650        copy_highpage(tmp_page, page);
1651        req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
1652        req->misc.write.next = NULL;
1653        req->in.argpages = 1;
1654        req->num_pages = 1;
1655        req->pages[0] = tmp_page;
1656        req->page_descs[0].offset = 0;
1657        req->page_descs[0].length = PAGE_SIZE;
1658        req->end = fuse_writepage_end;
1659        req->inode = inode;
1660
1661        inc_bdi_stat(mapping->backing_dev_info, BDI_WRITEBACK);
1662        inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP);
1663
1664        spin_lock(&fc->lock);
1665        list_add(&req->writepages_entry, &fi->writepages);
1666        list_add_tail(&req->list, &fi->queued_writes);
1667        fuse_flush_writepages(inode);
1668        spin_unlock(&fc->lock);
1669
1670        end_page_writeback(page);
1671
1672        return 0;
1673
1674err_nofile:
1675        __free_page(tmp_page);
1676err_free:
1677        fuse_request_free(req);
1678err:
1679        end_page_writeback(page);
1680        return error;
1681}
1682
1683static int fuse_writepage(struct page *page, struct writeback_control *wbc)
1684{
1685        int err;
1686
1687        if (fuse_page_is_writeback(page->mapping->host, page->index)) {
1688                /*
1689                 * ->writepages() should be called for sync() and friends.  We
1690                 * should only get here on direct reclaim and then we are
1691                 * allowed to skip a page which is already in flight
1692                 */
1693                WARN_ON(wbc->sync_mode == WB_SYNC_ALL);
1694
1695                redirty_page_for_writepage(wbc, page);
1696                return 0;
1697        }
1698
1699        err = fuse_writepage_locked(page);
1700        unlock_page(page);
1701
1702        return err;
1703}
1704
1705struct fuse_fill_wb_data {
1706        struct fuse_req *req;
1707        struct fuse_file *ff;
1708        struct inode *inode;
1709        struct page **orig_pages;
1710};
1711
1712static void fuse_writepages_send(struct fuse_fill_wb_data *data)
1713{
1714        struct fuse_req *req = data->req;
1715        struct inode *inode = data->inode;
1716        struct fuse_conn *fc = get_fuse_conn(inode);
1717        struct fuse_inode *fi = get_fuse_inode(inode);
1718        int num_pages = req->num_pages;
1719        int i;
1720
1721        req->ff = fuse_file_get(data->ff);
1722        spin_lock(&fc->lock);
1723        list_add_tail(&req->list, &fi->queued_writes);
1724        fuse_flush_writepages(inode);
1725        spin_unlock(&fc->lock);
1726
1727        for (i = 0; i < num_pages; i++)
1728                end_page_writeback(data->orig_pages[i]);
1729}
1730
1731static bool fuse_writepage_in_flight(struct fuse_req *new_req,
1732                                     struct page *page)
1733{
1734        struct fuse_conn *fc = get_fuse_conn(new_req->inode);
1735        struct fuse_inode *fi = get_fuse_inode(new_req->inode);
1736        struct fuse_req *tmp;
1737        struct fuse_req *old_req;
1738        bool found = false;
1739        pgoff_t curr_index;
1740
1741        BUG_ON(new_req->num_pages != 0);
1742
1743        spin_lock(&fc->lock);
1744        list_del(&new_req->writepages_entry);
1745        list_for_each_entry(old_req, &fi->writepages, writepages_entry) {
1746                BUG_ON(old_req->inode != new_req->inode);
1747                curr_index = old_req->misc.write.in.offset >> PAGE_CACHE_SHIFT;
1748                if (curr_index <= page->index &&
1749                    page->index < curr_index + old_req->num_pages) {
1750                        found = true;
1751                        break;
1752                }
1753        }
1754        if (!found) {
1755                list_add(&new_req->writepages_entry, &fi->writepages);
1756                goto out_unlock;
1757        }
1758
1759        new_req->num_pages = 1;
1760        for (tmp = old_req; tmp != NULL; tmp = tmp->misc.write.next) {
1761                BUG_ON(tmp->inode != new_req->inode);
1762                curr_index = tmp->misc.write.in.offset >> PAGE_CACHE_SHIFT;
1763                if (tmp->num_pages == 1 &&
1764                    curr_index == page->index) {
1765                        old_req = tmp;
1766                }
1767        }
1768
1769        if (old_req->num_pages == 1 && (old_req->state == FUSE_REQ_INIT ||
1770                                        old_req->state == FUSE_REQ_PENDING)) {
1771                struct backing_dev_info *bdi = page->mapping->backing_dev_info;
1772
1773                copy_highpage(old_req->pages[0], page);
1774                spin_unlock(&fc->lock);
1775
1776                dec_bdi_stat(bdi, BDI_WRITEBACK);
1777                dec_zone_page_state(page, NR_WRITEBACK_TEMP);
1778                bdi_writeout_inc(bdi);
1779                fuse_writepage_free(fc, new_req);
1780                fuse_request_free(new_req);
1781                goto out;
1782        } else {
1783                new_req->misc.write.next = old_req->misc.write.next;
1784                old_req->misc.write.next = new_req;
1785        }
1786out_unlock:
1787        spin_unlock(&fc->lock);
1788out:
1789        return found;
1790}
1791
1792static int fuse_writepages_fill(struct page *page,
1793                struct writeback_control *wbc, void *_data)
1794{
1795        struct fuse_fill_wb_data *data = _data;
1796        struct fuse_req *req = data->req;
1797        struct inode *inode = data->inode;
1798        struct fuse_conn *fc = get_fuse_conn(inode);
1799        struct page *tmp_page;
1800        bool is_writeback;
1801        int err;
1802
1803        if (!data->ff) {
1804                err = -EIO;
1805                data->ff = fuse_write_file_get(fc, get_fuse_inode(inode));
1806                if (!data->ff)
1807                        goto out_unlock;
1808        }
1809
1810        /*
1811         * Being under writeback is unlikely but possible.  For example direct
1812         * read to an mmaped fuse file will set the page dirty twice; once when
1813         * the pages are faulted with get_user_pages(), and then after the read
1814         * completed.
1815         */
1816        is_writeback = fuse_page_is_writeback(inode, page->index);
1817
1818        if (req && req->num_pages &&
1819            (is_writeback || req->num_pages == FUSE_MAX_PAGES_PER_REQ ||
1820             (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_write ||
1821             data->orig_pages[req->num_pages - 1]->index + 1 != page->index)) {
1822                fuse_writepages_send(data);
1823                data->req = NULL;
1824        }
1825        err = -ENOMEM;
1826        tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
1827        if (!tmp_page)
1828                goto out_unlock;
1829
1830        /*
1831         * The page must not be redirtied until the writeout is completed
1832         * (i.e. userspace has sent a reply to the write request).  Otherwise
1833         * there could be more than one temporary page instance for each real
1834         * page.
1835         *
1836         * This is ensured by holding the page lock in page_mkwrite() while
1837         * checking fuse_page_is_writeback().  We already hold the page lock
1838         * since clear_page_dirty_for_io() and keep it held until we add the
1839         * request to the fi->writepages list and increment req->num_pages.
1840         * After this fuse_page_is_writeback() will indicate that the page is
1841         * under writeback, so we can release the page lock.
1842         */
1843        if (data->req == NULL) {
1844                struct fuse_inode *fi = get_fuse_inode(inode);
1845
1846                err = -ENOMEM;
1847                req = fuse_request_alloc_nofs(FUSE_MAX_PAGES_PER_REQ);
1848                if (!req) {
1849                        __free_page(tmp_page);
1850                        goto out_unlock;
1851                }
1852
1853                fuse_write_fill(req, data->ff, page_offset(page), 0);
1854                req->misc.write.in.write_flags |= FUSE_WRITE_CACHE;
1855                req->misc.write.next = NULL;
1856                req->in.argpages = 1;
1857                req->background = 1;
1858                req->num_pages = 0;
1859                req->end = fuse_writepage_end;
1860                req->inode = inode;
1861
1862                spin_lock(&fc->lock);
1863                list_add(&req->writepages_entry, &fi->writepages);
1864                spin_unlock(&fc->lock);
1865
1866                data->req = req;
1867        }
1868        set_page_writeback(page);
1869
1870        copy_highpage(tmp_page, page);
1871        req->pages[req->num_pages] = tmp_page;
1872        req->page_descs[req->num_pages].offset = 0;
1873        req->page_descs[req->num_pages].length = PAGE_SIZE;
1874
1875        inc_bdi_stat(page->mapping->backing_dev_info, BDI_WRITEBACK);
1876        inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP);
1877
1878        err = 0;
1879        if (is_writeback && fuse_writepage_in_flight(req, page)) {
1880                end_page_writeback(page);
1881                data->req = NULL;
1882                goto out_unlock;
1883        }
1884        data->orig_pages[req->num_pages] = page;
1885
1886        /*
1887         * Protected by fc->lock against concurrent access by
1888         * fuse_page_is_writeback().
1889         */
1890        spin_lock(&fc->lock);
1891        req->num_pages++;
1892        spin_unlock(&fc->lock);
1893
1894out_unlock:
1895        unlock_page(page);
1896
1897        return err;
1898}
1899
1900static int fuse_writepages(struct address_space *mapping,
1901                           struct writeback_control *wbc)
1902{
1903        struct inode *inode = mapping->host;
1904        struct fuse_fill_wb_data data;
1905        int err;
1906
1907        err = -EIO;
1908        if (is_bad_inode(inode))
1909                goto out;
1910
1911        data.inode = inode;
1912        data.req = NULL;
1913        data.ff = NULL;
1914
1915        err = -ENOMEM;
1916        data.orig_pages = kcalloc(FUSE_MAX_PAGES_PER_REQ,
1917                                  sizeof(struct page *),
1918                                  GFP_NOFS);
1919        if (!data.orig_pages)
1920                goto out;
1921
1922        err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data);
1923        if (data.req) {
1924                /* Ignore errors if we can write at least one page */
1925                BUG_ON(!data.req->num_pages);
1926                fuse_writepages_send(&data);
1927                err = 0;
1928        }
1929        if (data.ff)
1930                fuse_file_put(data.ff, false);
1931
1932        kfree(data.orig_pages);
1933out:
1934        return err;
1935}
1936
1937/*
1938 * It's worthy to make sure that space is reserved on disk for the write,
1939 * but how to implement it without killing performance need more thinking.
1940 */
1941static int fuse_write_begin(struct file *file, struct address_space *mapping,
1942                loff_t pos, unsigned len, unsigned flags,
1943                struct page **pagep, void **fsdata)
1944{
1945        pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1946        struct fuse_conn *fc = get_fuse_conn(file_inode(file));
1947        struct page *page;
1948        loff_t fsize;
1949        int err = -ENOMEM;
1950
1951        WARN_ON(!fc->writeback_cache);
1952
1953        page = grab_cache_page_write_begin(mapping, index, flags);
1954        if (!page)
1955                goto error;
1956
1957        fuse_wait_on_page_writeback(mapping->host, page->index);
1958
1959        if (PageUptodate(page) || len == PAGE_CACHE_SIZE)
1960                goto success;
1961        /*
1962         * Check if the start this page comes after the end of file, in which
1963         * case the readpage can be optimized away.
1964         */
1965        fsize = i_size_read(mapping->host);
1966        if (fsize <= (pos & PAGE_CACHE_MASK)) {
1967                size_t off = pos & ~PAGE_CACHE_MASK;
1968                if (off)
1969                        zero_user_segment(page, 0, off);
1970                goto success;
1971        }
1972        err = fuse_do_readpage(file, page);
1973        if (err)
1974                goto cleanup;
1975success:
1976        *pagep = page;
1977        return 0;
1978
1979cleanup:
1980        unlock_page(page);
1981        page_cache_release(page);
1982error:
1983        return err;
1984}
1985
1986static int fuse_write_end(struct file *file, struct address_space *mapping,
1987                loff_t pos, unsigned len, unsigned copied,
1988                struct page *page, void *fsdata)
1989{
1990        struct inode *inode = page->mapping->host;
1991
1992        if (!PageUptodate(page)) {
1993                /* Zero any unwritten bytes at the end of the page */
1994                size_t endoff = (pos + copied) & ~PAGE_CACHE_MASK;
1995                if (endoff)
1996                        zero_user_segment(page, endoff, PAGE_CACHE_SIZE);
1997                SetPageUptodate(page);
1998        }
1999
2000        fuse_write_update_size(inode, pos + copied);
2001        set_page_dirty(page);
2002        unlock_page(page);
2003        page_cache_release(page);
2004
2005        return copied;
2006}
2007
2008static int fuse_launder_page(struct page *page)
2009{
2010        int err = 0;
2011        if (clear_page_dirty_for_io(page)) {
2012                struct inode *inode = page->mapping->host;
2013                err = fuse_writepage_locked(page);
2014                if (!err)
2015                        fuse_wait_on_page_writeback(inode, page->index);
2016        }
2017        return err;
2018}
2019
2020/*
2021 * Write back dirty pages now, because there may not be any suitable
2022 * open files later
2023 */
2024static void fuse_vma_close(struct vm_area_struct *vma)
2025{
2026        filemap_write_and_wait(vma->vm_file->f_mapping);
2027}
2028
2029/*
2030 * Wait for writeback against this page to complete before allowing it
2031 * to be marked dirty again, and hence written back again, possibly
2032 * before the previous writepage completed.
2033 *
2034 * Block here, instead of in ->writepage(), so that the userspace fs
2035 * can only block processes actually operating on the filesystem.
2036 *
2037 * Otherwise unprivileged userspace fs would be able to block
2038 * unrelated:
2039 *
2040 * - page migration
2041 * - sync(2)
2042 * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER
2043 */
2044static int fuse_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2045{
2046        struct page *page = vmf->page;
2047        struct inode *inode = file_inode(vma->vm_file);
2048
2049        file_update_time(vma->vm_file);
2050        lock_page(page);
2051        if (page->mapping != inode->i_mapping) {
2052                unlock_page(page);
2053                return VM_FAULT_NOPAGE;
2054        }
2055
2056        fuse_wait_on_page_writeback(inode, page->index);
2057        return VM_FAULT_LOCKED;
2058}
2059
2060static const struct vm_operations_struct fuse_file_vm_ops = {
2061        .close          = fuse_vma_close,
2062        .fault          = filemap_fault,
2063        .map_pages      = filemap_map_pages,
2064        .page_mkwrite   = fuse_page_mkwrite,
2065        .remap_pages    = generic_file_remap_pages,
2066};
2067
2068static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma)
2069{
2070        if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2071                fuse_link_write_file(file);
2072
2073        file_accessed(file);
2074        vma->vm_ops = &fuse_file_vm_ops;
2075        return 0;
2076}
2077
2078static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma)
2079{
2080        /* Can't provide the coherency needed for MAP_SHARED */
2081        if (vma->vm_flags & VM_MAYSHARE)
2082                return -ENODEV;
2083
2084        invalidate_inode_pages2(file->f_mapping);
2085
2086        return generic_file_mmap(file, vma);
2087}
2088
2089static int convert_fuse_file_lock(const struct fuse_file_lock *ffl,
2090                                  struct file_lock *fl)
2091{
2092        switch (ffl->type) {
2093        case F_UNLCK:
2094                break;
2095
2096        case F_RDLCK:
2097        case F_WRLCK:
2098                if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX ||
2099                    ffl->end < ffl->start)
2100                        return -EIO;
2101
2102                fl->fl_start = ffl->start;
2103                fl->fl_end = ffl->end;
2104                fl->fl_pid = ffl->pid;
2105                break;
2106
2107        default:
2108                return -EIO;
2109        }
2110        fl->fl_type = ffl->type;
2111        return 0;
2112}
2113
2114static void fuse_lk_fill(struct fuse_args *args, struct file *file,
2115                         const struct file_lock *fl, int opcode, pid_t pid,
2116                         int flock, struct fuse_lk_in *inarg)
2117{
2118        struct inode *inode = file_inode(file);
2119        struct fuse_conn *fc = get_fuse_conn(inode);
2120        struct fuse_file *ff = file->private_data;
2121
2122        memset(inarg, 0, sizeof(*inarg));
2123        inarg->fh = ff->fh;
2124        inarg->owner = fuse_lock_owner_id(fc, fl->fl_owner);
2125        inarg->lk.start = fl->fl_start;
2126        inarg->lk.end = fl->fl_end;
2127        inarg->lk.type = fl->fl_type;
2128        inarg->lk.pid = pid;
2129        if (flock)
2130                inarg->lk_flags |= FUSE_LK_FLOCK;
2131        args->in.h.opcode = opcode;
2132        args->in.h.nodeid = get_node_id(inode);
2133        args->in.numargs = 1;
2134        args->in.args[0].size = sizeof(*inarg);
2135        args->in.args[0].value = inarg;
2136}
2137
2138static int fuse_getlk(struct file *file, struct file_lock *fl)
2139{
2140        struct inode *inode = file_inode(file);
2141        struct fuse_conn *fc = get_fuse_conn(inode);
2142        FUSE_ARGS(args);
2143        struct fuse_lk_in inarg;
2144        struct fuse_lk_out outarg;
2145        int err;
2146
2147        fuse_lk_fill(&args, file, fl, FUSE_GETLK, 0, 0, &inarg);
2148        args.out.numargs = 1;
2149        args.out.args[0].size = sizeof(outarg);
2150        args.out.args[0].value = &outarg;
2151        err = fuse_simple_request(fc, &args);
2152        if (!err)
2153                err = convert_fuse_file_lock(&outarg.lk, fl);
2154
2155        return err;
2156}
2157
2158static int fuse_setlk(struct file *file, struct file_lock *fl, int flock)
2159{
2160        struct inode *inode = file_inode(file);
2161        struct fuse_conn *fc = get_fuse_conn(inode);
2162        FUSE_ARGS(args);
2163        struct fuse_lk_in inarg;
2164        int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK;
2165        pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0;
2166        int err;
2167
2168        if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
2169                /* NLM needs asynchronous locks, which we don't support yet */
2170                return -ENOLCK;
2171        }
2172
2173        /* Unlock on close is handled by the flush method */
2174        if (fl->fl_flags & FL_CLOSE)
2175                return 0;
2176
2177        fuse_lk_fill(&args, file, fl, opcode, pid, flock, &inarg);
2178        err = fuse_simple_request(fc, &args);
2179
2180        /* locking is restartable */
2181        if (err == -EINTR)
2182                err = -ERESTARTSYS;
2183
2184        return err;
2185}
2186
2187static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl)
2188{
2189        struct inode *inode = file_inode(file);
2190        struct fuse_conn *fc = get_fuse_conn(inode);
2191        int err;
2192
2193        if (cmd == F_CANCELLK) {
2194                err = 0;
2195        } else if (cmd == F_GETLK) {
2196                if (fc->no_lock) {
2197                        posix_test_lock(file, fl);
2198                        err = 0;
2199                } else
2200                        err = fuse_getlk(file, fl);
2201        } else {
2202                if (fc->no_lock)
2203                        err = posix_lock_file(file, fl, NULL);
2204                else
2205                        err = fuse_setlk(file, fl, 0);
2206        }
2207        return err;
2208}
2209
2210static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl)
2211{
2212        struct inode *inode = file_inode(file);
2213        struct fuse_conn *fc = get_fuse_conn(inode);
2214        int err;
2215
2216        if (fc->no_flock) {
2217                err = flock_lock_file_wait(file, fl);
2218        } else {
2219                struct fuse_file *ff = file->private_data;
2220
2221                /* emulate flock with POSIX locks */
2222                ff->flock = true;
2223                err = fuse_setlk(file, fl, 1);
2224        }
2225
2226        return err;
2227}
2228
2229static sector_t fuse_bmap(struct address_space *mapping, sector_t block)
2230{
2231        struct inode *inode = mapping->host;
2232        struct fuse_conn *fc = get_fuse_conn(inode);
2233        FUSE_ARGS(args);
2234        struct fuse_bmap_in inarg;
2235        struct fuse_bmap_out outarg;
2236        int err;
2237
2238        if (!inode->i_sb->s_bdev || fc->no_bmap)
2239                return 0;
2240
2241        memset(&inarg, 0, sizeof(inarg));
2242        inarg.block = block;
2243        inarg.blocksize = inode->i_sb->s_blocksize;
2244        args.in.h.opcode = FUSE_BMAP;
2245        args.in.h.nodeid = get_node_id(inode);
2246        args.in.numargs = 1;
2247        args.in.args[0].size = sizeof(inarg);
2248        args.in.args[0].value = &inarg;
2249        args.out.numargs = 1;
2250        args.out.args[0].size = sizeof(outarg);
2251        args.out.args[0].value = &outarg;
2252        err = fuse_simple_request(fc, &args);
2253        if (err == -ENOSYS)
2254                fc->no_bmap = 1;
2255
2256        return err ? 0 : outarg.block;
2257}
2258
2259static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence)
2260{
2261        loff_t retval;
2262        struct inode *inode = file_inode(file);
2263
2264        /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */
2265        if (whence == SEEK_CUR || whence == SEEK_SET)
2266                return generic_file_llseek(file, offset, whence);
2267
2268        mutex_lock(&inode->i_mutex);
2269        retval = fuse_update_attributes(inode, NULL, file, NULL);
2270        if (!retval)
2271                retval = generic_file_llseek(file, offset, whence);
2272        mutex_unlock(&inode->i_mutex);
2273
2274        return retval;
2275}
2276
2277static int fuse_ioctl_copy_user(struct page **pages, struct iovec *iov,
2278                        unsigned int nr_segs, size_t bytes, bool to_user)
2279{
2280        struct iov_iter ii;
2281        int page_idx = 0;
2282
2283        if (!bytes)
2284                return 0;
2285
2286        iov_iter_init(&ii, to_user ? READ : WRITE, iov, nr_segs, bytes);
2287
2288        while (iov_iter_count(&ii)) {
2289                struct page *page = pages[page_idx++];
2290                size_t todo = min_t(size_t, PAGE_SIZE, iov_iter_count(&ii));
2291                void *kaddr;
2292
2293                kaddr = kmap(page);
2294
2295                while (todo) {
2296                        char __user *uaddr = ii.iov->iov_base + ii.iov_offset;
2297                        size_t iov_len = ii.iov->iov_len - ii.iov_offset;
2298                        size_t copy = min(todo, iov_len);
2299                        size_t left;
2300
2301                        if (!to_user)
2302                                left = copy_from_user(kaddr, uaddr, copy);
2303                        else
2304                                left = copy_to_user(uaddr, kaddr, copy);
2305
2306                        if (unlikely(left))
2307                                return -EFAULT;
2308
2309                        iov_iter_advance(&ii, copy);
2310                        todo -= copy;
2311                        kaddr += copy;
2312                }
2313
2314                kunmap(page);
2315        }
2316
2317        return 0;
2318}
2319
2320/*
2321 * CUSE servers compiled on 32bit broke on 64bit kernels because the
2322 * ABI was defined to be 'struct iovec' which is different on 32bit
2323 * and 64bit.  Fortunately we can determine which structure the server
2324 * used from the size of the reply.
2325 */
2326static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src,
2327                                     size_t transferred, unsigned count,
2328                                     bool is_compat)
2329{
2330#ifdef CONFIG_COMPAT
2331        if (count * sizeof(struct compat_iovec) == transferred) {
2332                struct compat_iovec *ciov = src;
2333                unsigned i;
2334
2335                /*
2336                 * With this interface a 32bit server cannot support
2337                 * non-compat (i.e. ones coming from 64bit apps) ioctl
2338                 * requests
2339                 */
2340                if (!is_compat)
2341                        return -EINVAL;
2342
2343                for (i = 0; i < count; i++) {
2344                        dst[i].iov_base = compat_ptr(ciov[i].iov_base);
2345                        dst[i].iov_len = ciov[i].iov_len;
2346                }
2347                return 0;
2348        }
2349#endif
2350
2351        if (count * sizeof(struct iovec) != transferred)
2352                return -EIO;
2353
2354        memcpy(dst, src, transferred);
2355        return 0;
2356}
2357
2358/* Make sure iov_length() won't overflow */
2359static int fuse_verify_ioctl_iov(struct iovec *iov, size_t count)
2360{
2361        size_t n;
2362        u32 max = FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT;
2363
2364        for (n = 0; n < count; n++, iov++) {
2365                if (iov->iov_len > (size_t) max)
2366                        return -ENOMEM;
2367                max -= iov->iov_len;
2368        }
2369        return 0;
2370}
2371
2372static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst,
2373                                 void *src, size_t transferred, unsigned count,
2374                                 bool is_compat)
2375{
2376        unsigned i;
2377        struct fuse_ioctl_iovec *fiov = src;
2378
2379        if (fc->minor < 16) {
2380                return fuse_copy_ioctl_iovec_old(dst, src, transferred,
2381                                                 count, is_compat);
2382        }
2383
2384        if (count * sizeof(struct fuse_ioctl_iovec) != transferred)
2385                return -EIO;
2386
2387        for (i = 0; i < count; i++) {
2388                /* Did the server supply an inappropriate value? */
2389                if (fiov[i].base != (unsigned long) fiov[i].base ||
2390                    fiov[i].len != (unsigned long) fiov[i].len)
2391                        return -EIO;
2392
2393                dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base;
2394                dst[i].iov_len = (size_t) fiov[i].len;
2395
2396#ifdef CONFIG_COMPAT
2397                if (is_compat &&
2398                    (ptr_to_compat(dst[i].iov_base) != fiov[i].base ||
2399                     (compat_size_t) dst[i].iov_len != fiov[i].len))
2400                        return -EIO;
2401#endif
2402        }
2403
2404        return 0;
2405}
2406
2407
2408/*
2409 * For ioctls, there is no generic way to determine how much memory
2410 * needs to be read and/or written.  Furthermore, ioctls are allowed
2411 * to dereference the passed pointer, so the parameter requires deep
2412 * copying but FUSE has no idea whatsoever about what to copy in or
2413 * out.
2414 *
2415 * This is solved by allowing FUSE server to retry ioctl with
2416 * necessary in/out iovecs.  Let's assume the ioctl implementation
2417 * needs to read in the following structure.
2418 *
2419 * struct a {
2420 *      char    *buf;
2421 *      size_t  buflen;
2422 * }
2423 *
2424 * On the first callout to FUSE server, inarg->in_size and
2425 * inarg->out_size will be NULL; then, the server completes the ioctl
2426 * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and
2427 * the actual iov array to
2428 *
2429 * { { .iov_base = inarg.arg,   .iov_len = sizeof(struct a) } }
2430 *
2431 * which tells FUSE to copy in the requested area and retry the ioctl.
2432 * On the second round, the server has access to the structure and
2433 * from that it can tell what to look for next, so on the invocation,
2434 * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to
2435 *
2436 * { { .iov_base = inarg.arg,   .iov_len = sizeof(struct a)     },
2437 *   { .iov_base = a.buf,       .iov_len = a.buflen             } }
2438 *
2439 * FUSE will copy both struct a and the pointed buffer from the
2440 * process doing the ioctl and retry ioctl with both struct a and the
2441 * buffer.
2442 *
2443 * This time, FUSE server has everything it needs and completes ioctl
2444 * without FUSE_IOCTL_RETRY which finishes the ioctl call.
2445 *
2446 * Copying data out works the same way.
2447 *
2448 * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel
2449 * automatically initializes in and out iovs by decoding @cmd with
2450 * _IOC_* macros and the server is not allowed to request RETRY.  This
2451 * limits ioctl data transfers to well-formed ioctls and is the forced
2452 * behavior for all FUSE servers.
2453 */
2454long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
2455                   unsigned int flags)
2456{
2457        struct fuse_file *ff = file->private_data;
2458        struct fuse_conn *fc = ff->fc;
2459        struct fuse_ioctl_in inarg = {
2460                .fh = ff->fh,
2461                .cmd = cmd,
2462                .arg = arg,
2463                .flags = flags
2464        };
2465        struct fuse_ioctl_out outarg;
2466        struct fuse_req *req = NULL;
2467        struct page **pages = NULL;
2468        struct iovec *iov_page = NULL;
2469        struct iovec *in_iov = NULL, *out_iov = NULL;
2470        unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages;
2471        size_t in_size, out_size, transferred;
2472        int err;
2473
2474#if BITS_PER_LONG == 32
2475        inarg.flags |= FUSE_IOCTL_32BIT;
2476#else
2477        if (flags & FUSE_IOCTL_COMPAT)
2478                inarg.flags |= FUSE_IOCTL_32BIT;
2479#endif
2480
2481        /* assume all the iovs returned by client always fits in a page */
2482        BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE);
2483
2484        err = -ENOMEM;
2485        pages = kcalloc(FUSE_MAX_PAGES_PER_REQ, sizeof(pages[0]), GFP_KERNEL);
2486        iov_page = (struct iovec *) __get_free_page(GFP_KERNEL);
2487        if (!pages || !iov_page)
2488                goto out;
2489
2490        /*
2491         * If restricted, initialize IO parameters as encoded in @cmd.
2492         * RETRY from server is not allowed.
2493         */
2494        if (!(flags & FUSE_IOCTL_UNRESTRICTED)) {
2495                struct iovec *iov = iov_page;
2496
2497                iov->iov_base = (void __user *)arg;
2498                iov->iov_len = _IOC_SIZE(cmd);
2499
2500                if (_IOC_DIR(cmd) & _IOC_WRITE) {
2501                        in_iov = iov;
2502                        in_iovs = 1;
2503                }
2504
2505                if (_IOC_DIR(cmd) & _IOC_READ) {
2506                        out_iov = iov;
2507                        out_iovs = 1;
2508                }
2509        }
2510
2511 retry:
2512        inarg.in_size = in_size = iov_length(in_iov, in_iovs);
2513        inarg.out_size = out_size = iov_length(out_iov, out_iovs);
2514
2515        /*
2516         * Out data can be used either for actual out data or iovs,
2517         * make sure there always is at least one page.
2518         */
2519        out_size = max_t(size_t, out_size, PAGE_SIZE);
2520        max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE);
2521
2522        /* make sure there are enough buffer pages and init request with them */
2523        err = -ENOMEM;
2524        if (max_pages > FUSE_MAX_PAGES_PER_REQ)
2525                goto out;
2526        while (num_pages < max_pages) {
2527                pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2528                if (!pages[num_pages])
2529                        goto out;
2530                num_pages++;
2531        }
2532
2533        req = fuse_get_req(fc, num_pages);
2534        if (IS_ERR(req)) {
2535                err = PTR_ERR(req);
2536                req = NULL;
2537                goto out;
2538        }
2539        memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages);
2540        req->num_pages = num_pages;
2541        fuse_page_descs_length_init(req, 0, req->num_pages);
2542
2543        /* okay, let's send it to the client */
2544        req->in.h.opcode = FUSE_IOCTL;
2545        req->in.h.nodeid = ff->nodeid;
2546        req->in.numargs = 1;
2547        req->in.args[0].size = sizeof(inarg);
2548        req->in.args[0].value = &inarg;
2549        if (in_size) {
2550                req->in.numargs++;
2551                req->in.args[1].size = in_size;
2552                req->in.argpages = 1;
2553
2554                err = fuse_ioctl_copy_user(pages, in_iov, in_iovs, in_size,
2555                                           false);
2556                if (err)
2557                        goto out;
2558        }
2559
2560        req->out.numargs = 2;
2561        req->out.args[0].size = sizeof(outarg);
2562        req->out.args[0].value = &outarg;
2563        req->out.args[1].size = out_size;
2564        req->out.argpages = 1;
2565        req->out.argvar = 1;
2566
2567        fuse_request_send(fc, req);
2568        err = req->out.h.error;
2569        transferred = req->out.args[1].size;
2570        fuse_put_request(fc, req);
2571        req = NULL;
2572        if (err)
2573                goto out;
2574
2575        /* did it ask for retry? */
2576        if (outarg.flags & FUSE_IOCTL_RETRY) {
2577                void *vaddr;
2578
2579                /* no retry if in restricted mode */
2580                err = -EIO;
2581                if (!(flags & FUSE_IOCTL_UNRESTRICTED))
2582                        goto out;
2583
2584                in_iovs = outarg.in_iovs;
2585                out_iovs = outarg.out_iovs;
2586
2587                /*
2588                 * Make sure things are in boundary, separate checks
2589                 * are to protect against overflow.
2590                 */
2591                err = -ENOMEM;
2592                if (in_iovs > FUSE_IOCTL_MAX_IOV ||
2593                    out_iovs > FUSE_IOCTL_MAX_IOV ||
2594                    in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV)
2595                        goto out;
2596
2597                vaddr = kmap_atomic(pages[0]);
2598                err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr,
2599                                            transferred, in_iovs + out_iovs,
2600                                            (flags & FUSE_IOCTL_COMPAT) != 0);
2601                kunmap_atomic(vaddr);
2602                if (err)
2603                        goto out;
2604
2605                in_iov = iov_page;
2606                out_iov = in_iov + in_iovs;
2607
2608                err = fuse_verify_ioctl_iov(in_iov, in_iovs);
2609                if (err)
2610                        goto out;
2611
2612                err = fuse_verify_ioctl_iov(out_iov, out_iovs);
2613                if (err)
2614                        goto out;
2615
2616                goto retry;
2617        }
2618
2619        err = -EIO;
2620        if (transferred > inarg.out_size)
2621                goto out;
2622
2623        err = fuse_ioctl_copy_user(pages, out_iov, out_iovs, transferred, true);
2624 out:
2625        if (req)
2626                fuse_put_request(fc, req);
2627        free_page((unsigned long) iov_page);
2628        while (num_pages)
2629                __free_page(pages[--num_pages]);
2630        kfree(pages);
2631
2632        return err ? err : outarg.result;
2633}
2634EXPORT_SYMBOL_GPL(fuse_do_ioctl);
2635
2636long fuse_ioctl_common(struct file *file, unsigned int cmd,
2637                       unsigned long arg, unsigned int flags)
2638{
2639        struct inode *inode = file_inode(file);
2640        struct fuse_conn *fc = get_fuse_conn(inode);
2641
2642        if (!fuse_allow_current_process(fc))
2643                return -EACCES;
2644
2645        if (is_bad_inode(inode))
2646                return -EIO;
2647
2648        return fuse_do_ioctl(file, cmd, arg, flags);
2649}
2650
2651static long fuse_file_ioctl(struct file *file, unsigned int cmd,
2652                            unsigned long arg)
2653{
2654        return fuse_ioctl_common(file, cmd, arg, 0);
2655}
2656
2657static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd,
2658                                   unsigned long arg)
2659{
2660        return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT);
2661}
2662
2663/*
2664 * All files which have been polled are linked to RB tree
2665 * fuse_conn->polled_files which is indexed by kh.  Walk the tree and
2666 * find the matching one.
2667 */
2668static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh,
2669                                              struct rb_node **parent_out)
2670{
2671        struct rb_node **link = &fc->polled_files.rb_node;
2672        struct rb_node *last = NULL;
2673
2674        while (*link) {
2675                struct fuse_file *ff;
2676
2677                last = *link;
2678                ff = rb_entry(last, struct fuse_file, polled_node);
2679
2680                if (kh < ff->kh)
2681                        link = &last->rb_left;
2682                else if (kh > ff->kh)
2683                        link = &last->rb_right;
2684                else
2685                        return link;
2686        }
2687
2688        if (parent_out)
2689                *parent_out = last;
2690        return link;
2691}
2692
2693/*
2694 * The file is about to be polled.  Make sure it's on the polled_files
2695 * RB tree.  Note that files once added to the polled_files tree are
2696 * not removed before the file is released.  This is because a file
2697 * polled once is likely to be polled again.
2698 */
2699static void fuse_register_polled_file(struct fuse_conn *fc,
2700                                      struct fuse_file *ff)
2701{
2702        spin_lock(&fc->lock);
2703        if (RB_EMPTY_NODE(&ff->polled_node)) {
2704                struct rb_node **link, *uninitialized_var(parent);
2705
2706                link = fuse_find_polled_node(fc, ff->kh, &parent);
2707                BUG_ON(*link);
2708                rb_link_node(&ff->polled_node, parent, link);
2709                rb_insert_color(&ff->polled_node, &fc->polled_files);
2710        }
2711        spin_unlock(&fc->lock);
2712}
2713
2714unsigned fuse_file_poll(struct file *file, poll_table *wait)
2715{
2716        struct fuse_file *ff = file->private_data;
2717        struct fuse_conn *fc = ff->fc;
2718        struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh };
2719        struct fuse_poll_out outarg;
2720        FUSE_ARGS(args);
2721        int err;
2722
2723        if (fc->no_poll)
2724                return DEFAULT_POLLMASK;
2725
2726        poll_wait(file, &ff->poll_wait, wait);
2727        inarg.events = (__u32)poll_requested_events(wait);
2728
2729        /*
2730         * Ask for notification iff there's someone waiting for it.
2731         * The client may ignore the flag and always notify.
2732         */
2733        if (waitqueue_active(&ff->poll_wait)) {
2734                inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY;
2735                fuse_register_polled_file(fc, ff);
2736        }
2737
2738        args.in.h.opcode = FUSE_POLL;
2739        args.in.h.nodeid = ff->nodeid;
2740        args.in.numargs = 1;
2741        args.in.args[0].size = sizeof(inarg);
2742        args.in.args[0].value = &inarg;
2743        args.out.numargs = 1;
2744        args.out.args[0].size = sizeof(outarg);
2745        args.out.args[0].value = &outarg;
2746        err = fuse_simple_request(fc, &args);
2747
2748        if (!err)
2749                return outarg.revents;
2750        if (err == -ENOSYS) {
2751                fc->no_poll = 1;
2752                return DEFAULT_POLLMASK;
2753        }
2754        return POLLERR;
2755}
2756EXPORT_SYMBOL_GPL(fuse_file_poll);
2757
2758/*
2759 * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and
2760 * wakes up the poll waiters.
2761 */
2762int fuse_notify_poll_wakeup(struct fuse_conn *fc,
2763                            struct fuse_notify_poll_wakeup_out *outarg)
2764{
2765        u64 kh = outarg->kh;
2766        struct rb_node **link;
2767
2768        spin_lock(&fc->lock);
2769
2770        link = fuse_find_polled_node(fc, kh, NULL);
2771        if (*link) {
2772                struct fuse_file *ff;
2773
2774                ff = rb_entry(*link, struct fuse_file, polled_node);
2775                wake_up_interruptible_sync(&ff->poll_wait);
2776        }
2777
2778        spin_unlock(&fc->lock);
2779        return 0;
2780}
2781
2782static void fuse_do_truncate(struct file *file)
2783{
2784        struct inode *inode = file->f_mapping->host;
2785        struct iattr attr;
2786
2787        attr.ia_valid = ATTR_SIZE;
2788        attr.ia_size = i_size_read(inode);
2789
2790        attr.ia_file = file;
2791        attr.ia_valid |= ATTR_FILE;
2792
2793        fuse_do_setattr(inode, &attr, file);
2794}
2795
2796static inline loff_t fuse_round_up(loff_t off)
2797{
2798        return round_up(off, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT);
2799}
2800
2801static ssize_t
2802fuse_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter,
2803                        loff_t offset)
2804{
2805        ssize_t ret = 0;
2806        struct file *file = iocb->ki_filp;
2807        struct fuse_file *ff = file->private_data;
2808        bool async_dio = ff->fc->async_dio;
2809        loff_t pos = 0;
2810        struct inode *inode;
2811        loff_t i_size;
2812        size_t count = iov_iter_count(iter);
2813        struct fuse_io_priv *io;
2814
2815        pos = offset;
2816        inode = file->f_mapping->host;
2817        i_size = i_size_read(inode);
2818
2819        if ((rw == READ) && (offset > i_size))
2820                return 0;
2821
2822        /* optimization for short read */
2823        if (async_dio && rw != WRITE && offset + count > i_size) {
2824                if (offset >= i_size)
2825                        return 0;
2826                count = min_t(loff_t, count, fuse_round_up(i_size - offset));
2827                iov_iter_truncate(iter, count);
2828        }
2829
2830        io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL);
2831        if (!io)
2832                return -ENOMEM;
2833        spin_lock_init(&io->lock);
2834        io->reqs = 1;
2835        io->bytes = -1;
2836        io->size = 0;
2837        io->offset = offset;
2838        io->write = (rw == WRITE);
2839        io->err = 0;
2840        io->file = file;
2841        /*
2842         * By default, we want to optimize all I/Os with async request
2843         * submission to the client filesystem if supported.
2844         */
2845        io->async = async_dio;
2846        io->iocb = iocb;
2847
2848        /*
2849         * We cannot asynchronously extend the size of a file. We have no method
2850         * to wait on real async I/O requests, so we must submit this request
2851         * synchronously.
2852         */
2853        if (!is_sync_kiocb(iocb) && (offset + count > i_size) && rw == WRITE)
2854                io->async = false;
2855
2856        if (rw == WRITE)
2857                ret = __fuse_direct_write(io, iter, &pos);
2858        else
2859                ret = __fuse_direct_read(io, iter, &pos);
2860
2861        if (io->async) {
2862                fuse_aio_complete(io, ret < 0 ? ret : 0, -1);
2863
2864                /* we have a non-extending, async request, so return */
2865                if (!is_sync_kiocb(iocb))
2866                        return -EIOCBQUEUED;
2867
2868                ret = wait_on_sync_kiocb(iocb);
2869        } else {
2870                kfree(io);
2871        }
2872
2873        if (rw == WRITE) {
2874                if (ret > 0)
2875                        fuse_write_update_size(inode, pos);
2876                else if (ret < 0 && offset + count > i_size)
2877                        fuse_do_truncate(file);
2878        }
2879
2880        return ret;
2881}
2882
2883static long fuse_file_fallocate(struct file *file, int mode, loff_t offset,
2884                                loff_t length)
2885{
2886        struct fuse_file *ff = file->private_data;
2887        struct inode *inode = file_inode(file);
2888        struct fuse_inode *fi = get_fuse_inode(inode);
2889        struct fuse_conn *fc = ff->fc;
2890        FUSE_ARGS(args);
2891        struct fuse_fallocate_in inarg = {
2892                .fh = ff->fh,
2893                .offset = offset,
2894                .length = length,
2895                .mode = mode
2896        };
2897        int err;
2898        bool lock_inode = !(mode & FALLOC_FL_KEEP_SIZE) ||
2899                           (mode & FALLOC_FL_PUNCH_HOLE);
2900
2901        if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2902                return -EOPNOTSUPP;
2903
2904        if (fc->no_fallocate)
2905                return -EOPNOTSUPP;
2906
2907        if (lock_inode) {
2908                mutex_lock(&inode->i_mutex);
2909                if (mode & FALLOC_FL_PUNCH_HOLE) {
2910                        loff_t endbyte = offset + length - 1;
2911                        err = filemap_write_and_wait_range(inode->i_mapping,
2912                                                           offset, endbyte);
2913                        if (err)
2914                                goto out;
2915
2916                        fuse_sync_writes(inode);
2917                }
2918        }
2919
2920        if (!(mode & FALLOC_FL_KEEP_SIZE))
2921                set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
2922
2923        args.in.h.opcode = FUSE_FALLOCATE;
2924        args.in.h.nodeid = ff->nodeid;
2925        args.in.numargs = 1;
2926        args.in.args[0].size = sizeof(inarg);
2927        args.in.args[0].value = &inarg;
2928        err = fuse_simple_request(fc, &args);
2929        if (err == -ENOSYS) {
2930                fc->no_fallocate = 1;
2931                err = -EOPNOTSUPP;
2932        }
2933        if (err)
2934                goto out;
2935
2936        /* we could have extended the file */
2937        if (!(mode & FALLOC_FL_KEEP_SIZE)) {
2938                bool changed = fuse_write_update_size(inode, offset + length);
2939
2940                if (changed && fc->writeback_cache)
2941                        file_update_time(file);
2942        }
2943
2944        if (mode & FALLOC_FL_PUNCH_HOLE)
2945                truncate_pagecache_range(inode, offset, offset + length - 1);
2946
2947        fuse_invalidate_attr(inode);
2948
2949out:
2950        if (!(mode & FALLOC_FL_KEEP_SIZE))
2951                clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state);
2952
2953        if (lock_inode)
2954                mutex_unlock(&inode->i_mutex);
2955
2956        return err;
2957}
2958
2959static const struct file_operations fuse_file_operations = {
2960        .llseek         = fuse_file_llseek,
2961        .read           = new_sync_read,
2962        .read_iter      = fuse_file_read_iter,
2963        .write          = new_sync_write,
2964        .write_iter     = fuse_file_write_iter,
2965        .mmap           = fuse_file_mmap,
2966        .open           = fuse_open,
2967        .flush          = fuse_flush,
2968        .release        = fuse_release,
2969        .fsync          = fuse_fsync,
2970        .lock           = fuse_file_lock,
2971        .flock          = fuse_file_flock,
2972        .splice_read    = generic_file_splice_read,
2973        .unlocked_ioctl = fuse_file_ioctl,
2974        .compat_ioctl   = fuse_file_compat_ioctl,
2975        .poll           = fuse_file_poll,
2976        .fallocate      = fuse_file_fallocate,
2977};
2978
2979static const struct file_operations fuse_direct_io_file_operations = {
2980        .llseek         = fuse_file_llseek,
2981        .read           = fuse_direct_read,
2982        .write          = fuse_direct_write,
2983        .mmap           = fuse_direct_mmap,
2984        .open           = fuse_open,
2985        .flush          = fuse_flush,
2986        .release        = fuse_release,
2987        .fsync          = fuse_fsync,
2988        .lock           = fuse_file_lock,
2989        .flock          = fuse_file_flock,
2990        .unlocked_ioctl = fuse_file_ioctl,
2991        .compat_ioctl   = fuse_file_compat_ioctl,
2992        .poll           = fuse_file_poll,
2993        .fallocate      = fuse_file_fallocate,
2994        /* no splice_read */
2995};
2996
2997static const struct address_space_operations fuse_file_aops  = {
2998        .readpage       = fuse_readpage,
2999        .writepage      = fuse_writepage,
3000        .writepages     = fuse_writepages,
3001        .launder_page   = fuse_launder_page,
3002        .readpages      = fuse_readpages,
3003        .set_page_dirty = __set_page_dirty_nobuffers,
3004        .bmap           = fuse_bmap,
3005        .direct_IO      = fuse_direct_IO,
3006        .write_begin    = fuse_write_begin,
3007        .write_end      = fuse_write_end,
3008};
3009
3010void fuse_init_file_inode(struct inode *inode)
3011{
3012        inode->i_fop = &fuse_file_operations;
3013        inode->i_data.a_ops = &fuse_file_aops;
3014}
3015
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.