tools/virtiofsd/passthrough_ll.c

   1 /*
   2  * FUSE: Filesystem in Userspace
   3  * Copyright (C) 2001-2007  Miklos Szeredi <miklos@szeredi.hu>
   4  *
   5  * This program can be distributed under the terms of the GNU GPLv2.
   6  * See the file COPYING.
   7  */
   8
   9 /*
  10  *
  11  * This file system mirrors the existing file system hierarchy of the
  12  * system, starting at the root file system. This is implemented by
  13  * just "passing through" all requests to the corresponding user-space
  14  * libc functions. In contrast to passthrough.c and passthrough_fh.c,
  15  * this implementation uses the low-level API. Its performance should
  16  * be the least bad among the three, but many operations are not
  17  * implemented. In particular, it is not possible to remove files (or
  18  * directories) because the code necessary to defer actual removal
  19  * until the file is not opened anymore would make the example much
  20  * more complicated.
  21  *
  22  * When writeback caching is enabled (-o writeback mount option), it
  23  * is only possible to write to files for which the mounting user has
  24  * read permissions. This is because the writeback cache requires the
  25  * kernel to be able to issue read requests for all files (which the
  26  * passthrough filesystem cannot satisfy if it can't read the file in
  27  * the underlying filesystem).
  28  *
  29  * Compile with:
  30  *
  31  *     gcc -Wall passthrough_ll.c `pkg-config fuse3 --cflags --libs` -o
  32  * passthrough_ll
  33  *
  34  * ## Source code ##
  35  * \include passthrough_ll.c
  36  */
  37
  38 #define _GNU_SOURCE
  39 #define FUSE_USE_VERSION 31
  40
  41 #include "config.h"
  42
  43 #include <assert.h>
  44 #include <dirent.h>
  45 #include <errno.h>
  46 #include <fuse_lowlevel.h>
  47 #include <inttypes.h>
  48 #include <limits.h>
  49 #include <pthread.h>
  50 #include <stdbool.h>
  51 #include <stddef.h>
  52 #include <stdio.h>
  53 #include <stdlib.h>
  54 #include <string.h>
  55 #include <sys/file.h>
  56 #include <sys/xattr.h>
  57 #include <unistd.h>
  58
  59 #include "passthrough_helpers.h"
  60
  61 /*
  62  * We are re-using pointers to our `struct lo_inode` and `struct
  63  * lo_dirp` elements as inodes. This means that we must be able to
  64  * store uintptr_t values in a fuse_ino_t variable. The following
  65  * incantation checks this condition at compile time.
  66  */
  67 #if defined(__GNUC__) &&                                      \
  68     (__GNUC__ > 4 || __GNUC__ == 4 && __GNUC_MINOR__ >= 6) && \
  69     !defined __cplusplus
  70 _Static_assert(sizeof(fuse_ino_t) >= sizeof(uintptr_t),
  71                "fuse_ino_t too small to hold uintptr_t values!");
  72 #else
  73 struct _uintptr_to_must_hold_fuse_ino_t_dummy_struct {
  74     unsigned _uintptr_to_must_hold_fuse_ino_t
  75         : ((sizeof(fuse_ino_t) >= sizeof(uintptr_t)) ? 1 : -1);
  76 };
  77 #endif
  78
  79 struct lo_inode {
  80     struct lo_inode *next; /* protected by lo->mutex */
  81     struct lo_inode *prev; /* protected by lo->mutex */
  82     int fd;
  83     bool is_symlink;
  84     ino_t ino;
  85     dev_t dev;
  86     uint64_t refcount; /* protected by lo->mutex */
  87 };
  88
  89 enum {
  90     CACHE_NEVER,
  91     CACHE_NORMAL,
  92     CACHE_ALWAYS,
  93 };
  94
  95 struct lo_data {
  96     pthread_mutex_t mutex;
  97     int debug;
  98     int writeback;
  99     int flock;
 100     int xattr;
 101     const char *source;
 102     double timeout;
 103     int cache;
 104     int timeout_set;
 105     struct lo_inode root; /* protected by lo->mutex */
 106 };
 107
 108 static const struct fuse_opt lo_opts[] = {
 109     { "writeback", offsetof(struct lo_data, writeback), 1 },
 110     { "no_writeback", offsetof(struct lo_data, writeback), 0 },
 111     { "source=%s", offsetof(struct lo_data, source), 0 },
 112     { "flock", offsetof(struct lo_data, flock), 1 },
 113     { "no_flock", offsetof(struct lo_data, flock), 0 },
 114     { "xattr", offsetof(struct lo_data, xattr), 1 },
 115     { "no_xattr", offsetof(struct lo_data, xattr), 0 },
 116     { "timeout=%lf", offsetof(struct lo_data, timeout), 0 },
 117     { "timeout=", offsetof(struct lo_data, timeout_set), 1 },
 118     { "cache=never", offsetof(struct lo_data, cache), CACHE_NEVER },
 119     { "cache=auto", offsetof(struct lo_data, cache), CACHE_NORMAL },
 120     { "cache=always", offsetof(struct lo_data, cache), CACHE_ALWAYS },
 121
 122     FUSE_OPT_END
 123 };
 124
 125 static struct lo_data *lo_data(fuse_req_t req)
 126 {
 127     return (struct lo_data *)fuse_req_userdata(req);
 128 }
 129
 130 static struct lo_inode *lo_inode(fuse_req_t req, fuse_ino_t ino)
 131 {
 132     if (ino == FUSE_ROOT_ID) {
 133         return &lo_data(req)->root;
 134     } else {
 135         return (struct lo_inode *)(uintptr_t)ino;
 136     }
 137 }
 138
 139 static int lo_fd(fuse_req_t req, fuse_ino_t ino)
 140 {
 141     return lo_inode(req, ino)->fd;
 142 }
 143
 144 static bool lo_debug(fuse_req_t req)
 145 {
 146     return lo_data(req)->debug != 0;
 147 }
 148
 149 static void lo_init(void *userdata, struct fuse_conn_info *conn)
 150 {
 151     struct lo_data *lo = (struct lo_data *)userdata;
 152
 153     if (conn->capable & FUSE_CAP_EXPORT_SUPPORT) {
 154         conn->want |= FUSE_CAP_EXPORT_SUPPORT;
 155     }
 156
 157     if (lo->writeback && conn->capable & FUSE_CAP_WRITEBACK_CACHE) {
 158         if (lo->debug) {
 159             fuse_log(FUSE_LOG_DEBUG, "lo_init: activating writeback\n");
 160         }
 161         conn->want |= FUSE_CAP_WRITEBACK_CACHE;
 162     }
 163     if (lo->flock && conn->capable & FUSE_CAP_FLOCK_LOCKS) {
 164         if (lo->debug) {
 165             fuse_log(FUSE_LOG_DEBUG, "lo_init: activating flock locks\n");
 166         }
 167         conn->want |= FUSE_CAP_FLOCK_LOCKS;
 168     }
 169 }
 170
 171 static void lo_getattr(fuse_req_t req, fuse_ino_t ino,
 172                        struct fuse_file_info *fi)
 173 {
 174     int res;
 175     struct stat buf;
 176     struct lo_data *lo = lo_data(req);
 177
 178     (void)fi;
 179
 180     res =
 181         fstatat(lo_fd(req, ino), "", &buf, AT_EMPTY_PATH | AT_SYMLINK_NOFOLLOW);
 182     if (res == -1) {
 183         return (void)fuse_reply_err(req, errno);
 184     }
 185
 186     fuse_reply_attr(req, &buf, lo->timeout);
 187 }
 188
 189 static int utimensat_empty_nofollow(struct lo_inode *inode,
 190                                     const struct timespec *tv)
 191 {
 192     int res;
 193     char procname[64];
 194
 195     if (inode->is_symlink) {
 196         res = utimensat(inode->fd, "", tv, AT_EMPTY_PATH | AT_SYMLINK_NOFOLLOW);
 197         if (res == -1 && errno == EINVAL) {
 198             /* Sorry, no race free way to set times on symlink. */
 199             errno = EPERM;
 200         }
 201         return res;
 202     }
 203     sprintf(procname, "/proc/self/fd/%i", inode->fd);
 204
 205     return utimensat(AT_FDCWD, procname, tv, 0);
 206 }
 207
 208 static void lo_setattr(fuse_req_t req, fuse_ino_t ino, struct stat *attr,
 209                        int valid, struct fuse_file_info *fi)
 210 {
 211     int saverr;
 212     char procname[64];
 213     struct lo_inode *inode = lo_inode(req, ino);
 214     int ifd = inode->fd;
 215     int res;
 216
 217     if (valid & FUSE_SET_ATTR_MODE) {
 218         if (fi) {
 219             res = fchmod(fi->fh, attr->st_mode);
 220         } else {
 221             sprintf(procname, "/proc/self/fd/%i", ifd);
 222             res = chmod(procname, attr->st_mode);
 223         }
 224         if (res == -1) {
 225             goto out_err;
 226         }
 227     }
 228     if (valid & (FUSE_SET_ATTR_UID | FUSE_SET_ATTR_GID)) {
 229         uid_t uid = (valid & FUSE_SET_ATTR_UID) ? attr->st_uid : (uid_t)-1;
 230         gid_t gid = (valid & FUSE_SET_ATTR_GID) ? attr->st_gid : (gid_t)-1;
 231
 232         res = fchownat(ifd, "", uid, gid, AT_EMPTY_PATH | AT_SYMLINK_NOFOLLOW);
 233         if (res == -1) {
 234             goto out_err;
 235         }
 236     }
 237     if (valid & FUSE_SET_ATTR_SIZE) {
 238         if (fi) {
 239             res = ftruncate(fi->fh, attr->st_size);
 240         } else {
 241             sprintf(procname, "/proc/self/fd/%i", ifd);
 242             res = truncate(procname, attr->st_size);
 243         }
 244         if (res == -1) {
 245             goto out_err;
 246         }
 247     }
 248     if (valid & (FUSE_SET_ATTR_ATIME | FUSE_SET_ATTR_MTIME)) {
 249         struct timespec tv[2];
 250
 251         tv[0].tv_sec = 0;
 252         tv[1].tv_sec = 0;
 253         tv[0].tv_nsec = UTIME_OMIT;
 254         tv[1].tv_nsec = UTIME_OMIT;
 255
 256         if (valid & FUSE_SET_ATTR_ATIME_NOW) {
 257             tv[0].tv_nsec = UTIME_NOW;
 258         } else if (valid & FUSE_SET_ATTR_ATIME) {
 259             tv[0] = attr->st_atim;
 260         }
 261
 262         if (valid & FUSE_SET_ATTR_MTIME_NOW) {
 263             tv[1].tv_nsec = UTIME_NOW;
 264         } else if (valid & FUSE_SET_ATTR_MTIME) {
 265             tv[1] = attr->st_mtim;
 266         }
 267
 268         if (fi) {
 269             res = futimens(fi->fh, tv);
 270         } else {
 271             res = utimensat_empty_nofollow(inode, tv);
 272         }
 273         if (res == -1) {
 274             goto out_err;
 275         }
 276     }
 277
 278     return lo_getattr(req, ino, fi);
 279
 280 out_err:
 281     saverr = errno;
 282     fuse_reply_err(req, saverr);
 283 }
 284
 285 static struct lo_inode *lo_find(struct lo_data *lo, struct stat *st)
 286 {
 287     struct lo_inode *p;
 288     struct lo_inode *ret = NULL;
 289
 290     pthread_mutex_lock(&lo->mutex);
 291     for (p = lo->root.next; p != &lo->root; p = p->next) {
 292         if (p->ino == st->st_ino && p->dev == st->st_dev) {
 293             assert(p->refcount > 0);
 294             ret = p;
 295             ret->refcount++;
 296             break;
 297         }
 298     }
 299     pthread_mutex_unlock(&lo->mutex);
 300     return ret;
 301 }
 302
 303 static int lo_do_lookup(fuse_req_t req, fuse_ino_t parent, const char *name,
 304                         struct fuse_entry_param *e)
 305 {
 306     int newfd;
 307     int res;
 308     int saverr;
 309     struct lo_data *lo = lo_data(req);
 310     struct lo_inode *inode;
 311
 312     memset(e, 0, sizeof(*e));
 313     e->attr_timeout = lo->timeout;
 314     e->entry_timeout = lo->timeout;
 315
 316     newfd = openat(lo_fd(req, parent), name, O_PATH | O_NOFOLLOW);
 317     if (newfd == -1) {
 318         goto out_err;
 319     }
 320
 321     res = fstatat(newfd, "", &e->attr, AT_EMPTY_PATH | AT_SYMLINK_NOFOLLOW);
 322     if (res == -1) {
 323         goto out_err;
 324     }
 325
 326     inode = lo_find(lo_data(req), &e->attr);
 327     if (inode) {
 328         close(newfd);
 329         newfd = -1;
 330     } else {
 331         struct lo_inode *prev, *next;
 332
 333         saverr = ENOMEM;
 334         inode = calloc(1, sizeof(struct lo_inode));
 335         if (!inode) {
 336             goto out_err;
 337         }
 338
 339         inode->is_symlink = S_ISLNK(e->attr.st_mode);
 340         inode->refcount = 1;
 341         inode->fd = newfd;
 342         inode->ino = e->attr.st_ino;
 343         inode->dev = e->attr.st_dev;
 344
 345         pthread_mutex_lock(&lo->mutex);
 346         prev = &lo->root;
 347         next = prev->next;
 348         next->prev = inode;
 349         inode->next = next;
 350         inode->prev = prev;
 351         prev->next = inode;
 352         pthread_mutex_unlock(&lo->mutex);
 353     }
 354     e->ino = (uintptr_t)inode;
 355
 356     if (lo_debug(req)) {
 357         fuse_log(FUSE_LOG_DEBUG, "  %lli/%s -> %lli\n",
 358                  (unsigned long long)parent, name, (unsigned long long)e->ino);
 359     }
 360
 361     return 0;
 362
 363 out_err:
 364     saverr = errno;
 365     if (newfd != -1) {
 366         close(newfd);
 367     }
 368     return saverr;
 369 }
 370
 371 static void lo_lookup(fuse_req_t req, fuse_ino_t parent, const char *name)
 372 {
 373     struct fuse_entry_param e;
 374     int err;
 375
 376     if (lo_debug(req)) {
 377         fuse_log(FUSE_LOG_DEBUG, "lo_lookup(parent=%" PRIu64 ", name=%s)\n",
 378                  parent, name);
 379     }
 380
 381     err = lo_do_lookup(req, parent, name, &e);
 382     if (err) {
 383         fuse_reply_err(req, err);
 384     } else {
 385         fuse_reply_entry(req, &e);
 386     }
 387 }
 388
 389 static void lo_mknod_symlink(fuse_req_t req, fuse_ino_t parent,
 390                              const char *name, mode_t mode, dev_t rdev,
 391                              const char *link)
 392 {
 393     int res;
 394     int saverr;
 395     struct lo_inode *dir = lo_inode(req, parent);
 396     struct fuse_entry_param e;
 397
 398     saverr = ENOMEM;
 399
 400     res = mknod_wrapper(dir->fd, name, link, mode, rdev);
 401
 402     saverr = errno;
 403     if (res == -1) {
 404         goto out;
 405     }
 406
 407     saverr = lo_do_lookup(req, parent, name, &e);
 408     if (saverr) {
 409         goto out;
 410     }
 411
 412     if (lo_debug(req)) {
 413         fuse_log(FUSE_LOG_DEBUG, "  %lli/%s -> %lli\n",
 414                  (unsigned long long)parent, name, (unsigned long long)e.ino);
 415     }
 416
 417     fuse_reply_entry(req, &e);
 418     return;
 419
 420 out:
 421     fuse_reply_err(req, saverr);
 422 }
 423
 424 static void lo_mknod(fuse_req_t req, fuse_ino_t parent, const char *name,
 425                      mode_t mode, dev_t rdev)
 426 {
 427     lo_mknod_symlink(req, parent, name, mode, rdev, NULL);
 428 }
 429
 430 static void lo_mkdir(fuse_req_t req, fuse_ino_t parent, const char *name,
 431                      mode_t mode)
 432 {
 433     lo_mknod_symlink(req, parent, name, S_IFDIR | mode, 0, NULL);
 434 }
 435
 436 static void lo_symlink(fuse_req_t req, const char *link, fuse_ino_t parent,
 437                        const char *name)
 438 {
 439     lo_mknod_symlink(req, parent, name, S_IFLNK, 0, link);
 440 }
 441
 442 static int linkat_empty_nofollow(struct lo_inode *inode, int dfd,
 443                                  const char *name)
 444 {
 445     int res;
 446     char procname[64];
 447
 448     if (inode->is_symlink) {
 449         res = linkat(inode->fd, "", dfd, name, AT_EMPTY_PATH);
 450         if (res == -1 && (errno == ENOENT || errno == EINVAL)) {
 451             /* Sorry, no race free way to hard-link a symlink. */
 452             errno = EPERM;
 453         }
 454         return res;
 455     }
 456
 457     sprintf(procname, "/proc/self/fd/%i", inode->fd);
 458
 459     return linkat(AT_FDCWD, procname, dfd, name, AT_SYMLINK_FOLLOW);
 460 }
 461
 462 static void lo_link(fuse_req_t req, fuse_ino_t ino, fuse_ino_t parent,
 463                     const char *name)
 464 {
 465     int res;
 466     struct lo_data *lo = lo_data(req);
 467     struct lo_inode *inode = lo_inode(req, ino);
 468     struct fuse_entry_param e;
 469     int saverr;
 470
 471     memset(&e, 0, sizeof(struct fuse_entry_param));
 472     e.attr_timeout = lo->timeout;
 473     e.entry_timeout = lo->timeout;
 474
 475     res = linkat_empty_nofollow(inode, lo_fd(req, parent), name);
 476     if (res == -1) {
 477         goto out_err;
 478     }
 479
 480     res = fstatat(inode->fd, "", &e.attr, AT_EMPTY_PATH | AT_SYMLINK_NOFOLLOW);
 481     if (res == -1) {
 482         goto out_err;
 483     }
 484
 485     pthread_mutex_lock(&lo->mutex);
 486     inode->refcount++;
 487     pthread_mutex_unlock(&lo->mutex);
 488     e.ino = (uintptr_t)inode;
 489
 490     if (lo_debug(req)) {
 491         fuse_log(FUSE_LOG_DEBUG, "  %lli/%s -> %lli\n",
 492                  (unsigned long long)parent, name, (unsigned long long)e.ino);
 493     }
 494
 495     fuse_reply_entry(req, &e);
 496     return;
 497
 498 out_err:
 499     saverr = errno;
 500     fuse_reply_err(req, saverr);
 501 }
 502
 503 static void lo_rmdir(fuse_req_t req, fuse_ino_t parent, const char *name)
 504 {
 505     int res;
 506
 507     res = unlinkat(lo_fd(req, parent), name, AT_REMOVEDIR);
 508
 509     fuse_reply_err(req, res == -1 ? errno : 0);
 510 }
 511
 512 static void lo_rename(fuse_req_t req, fuse_ino_t parent, const char *name,
 513                       fuse_ino_t newparent, const char *newname,
 514                       unsigned int flags)
 515 {
 516     int res;
 517
 518     if (flags) {
 519         fuse_reply_err(req, EINVAL);
 520         return;
 521     }
 522
 523     res = renameat(lo_fd(req, parent), name, lo_fd(req, newparent), newname);
 524
 525     fuse_reply_err(req, res == -1 ? errno : 0);
 526 }
 527
 528 static void lo_unlink(fuse_req_t req, fuse_ino_t parent, const char *name)
 529 {
 530     int res;
 531
 532     res = unlinkat(lo_fd(req, parent), name, 0);
 533
 534     fuse_reply_err(req, res == -1 ? errno : 0);
 535 }
 536
 537 static void unref_inode(struct lo_data *lo, struct lo_inode *inode, uint64_t n)
 538 {
 539     if (!inode) {
 540         return;
 541     }
 542
 543     pthread_mutex_lock(&lo->mutex);
 544     assert(inode->refcount >= n);
 545     inode->refcount -= n;
 546     if (!inode->refcount) {
 547         struct lo_inode *prev, *next;
 548
 549         prev = inode->prev;
 550         next = inode->next;
 551         next->prev = prev;
 552         prev->next = next;
 553
 554         pthread_mutex_unlock(&lo->mutex);
 555         close(inode->fd);
 556         free(inode);
 557
 558     } else {
 559         pthread_mutex_unlock(&lo->mutex);
 560     }
 561 }
 562
 563 static void lo_forget_one(fuse_req_t req, fuse_ino_t ino, uint64_t nlookup)
 564 {
 565     struct lo_data *lo = lo_data(req);
 566     struct lo_inode *inode = lo_inode(req, ino);
 567
 568     if (lo_debug(req)) {
 569         fuse_log(FUSE_LOG_DEBUG, "  forget %lli %lli -%lli\n",
 570                  (unsigned long long)ino, (unsigned long long)inode->refcount,
 571                  (unsigned long long)nlookup);
 572     }
 573
 574     unref_inode(lo, inode, nlookup);
 575 }
 576
 577 static void lo_forget(fuse_req_t req, fuse_ino_t ino, uint64_t nlookup)
 578 {
 579     lo_forget_one(req, ino, nlookup);
 580     fuse_reply_none(req);
 581 }
 582
 583 static void lo_forget_multi(fuse_req_t req, size_t count,
 584                             struct fuse_forget_data *forgets)
 585 {
 586     int i;
 587
 588     for (i = 0; i < count; i++) {
 589         lo_forget_one(req, forgets[i].ino, forgets[i].nlookup);
 590     }
 591     fuse_reply_none(req);
 592 }
 593
 594 static void lo_readlink(fuse_req_t req, fuse_ino_t ino)
 595 {
 596     char buf[PATH_MAX + 1];
 597     int res;
 598
 599     res = readlinkat(lo_fd(req, ino), "", buf, sizeof(buf));
 600     if (res == -1) {
 601         return (void)fuse_reply_err(req, errno);
 602     }
 603
 604     if (res == sizeof(buf)) {
 605         return (void)fuse_reply_err(req, ENAMETOOLONG);
 606     }
 607
 608     buf[res] = '\0';
 609
 610     fuse_reply_readlink(req, buf);
 611 }
 612
 613 struct lo_dirp {
 614     DIR *dp;
 615     struct dirent *entry;
 616     off_t offset;
 617 };
 618
 619 static struct lo_dirp *lo_dirp(struct fuse_file_info *fi)
 620 {
 621     return (struct lo_dirp *)(uintptr_t)fi->fh;
 622 }
 623
 624 static void lo_opendir(fuse_req_t req, fuse_ino_t ino,
 625                        struct fuse_file_info *fi)
 626 {
 627     int error = ENOMEM;
 628     struct lo_data *lo = lo_data(req);
 629     struct lo_dirp *d;
 630     int fd;
 631
 632     d = calloc(1, sizeof(struct lo_dirp));
 633     if (d == NULL) {
 634         goto out_err;
 635     }
 636
 637     fd = openat(lo_fd(req, ino), ".", O_RDONLY);
 638     if (fd == -1) {
 639         goto out_errno;
 640     }
 641
 642     d->dp = fdopendir(fd);
 643     if (d->dp == NULL) {
 644         goto out_errno;
 645     }
 646
 647     d->offset = 0;
 648     d->entry = NULL;
 649
 650     fi->fh = (uintptr_t)d;
 651     if (lo->cache == CACHE_ALWAYS) {
 652         fi->keep_cache = 1;
 653     }
 654     fuse_reply_open(req, fi);
 655     return;
 656
 657 out_errno:
 658     error = errno;
 659 out_err:
 660     if (d) {
 661         if (fd != -1) {
 662             close(fd);
 663         }
 664         free(d);
 665     }
 666     fuse_reply_err(req, error);
 667 }
 668
 669 static int is_dot_or_dotdot(const char *name)
 670 {
 671     return name[0] == '.' &&
 672            (name[1] == '\0' || (name[1] == '.' && name[2] == '\0'));
 673 }
 674
 675 static void lo_do_readdir(fuse_req_t req, fuse_ino_t ino, size_t size,
 676                           off_t offset, struct fuse_file_info *fi, int plus)
 677 {
 678     struct lo_dirp *d = lo_dirp(fi);
 679     char *buf;
 680     char *p;
 681     size_t rem = size;
 682     int err;
 683
 684     (void)ino;
 685
 686     buf = calloc(1, size);
 687     if (!buf) {
 688         err = ENOMEM;
 689         goto error;
 690     }
 691     p = buf;
 692
 693     if (offset != d->offset) {
 694         seekdir(d->dp, offset);
 695         d->entry = NULL;
 696         d->offset = offset;
 697     }
 698     while (1) {
 699         size_t entsize;
 700         off_t nextoff;
 701         const char *name;
 702
 703         if (!d->entry) {
 704             errno = 0;
 705             d->entry = readdir(d->dp);
 706             if (!d->entry) {
 707                 if (errno) { /* Error */
 708                     err = errno;
 709                     goto error;
 710                 } else { /* End of stream */
 711                     break;
 712                 }
 713             }
 714         }
 715         nextoff = d->entry->d_off;
 716         name = d->entry->d_name;
 717         fuse_ino_t entry_ino = 0;
 718         if (plus) {
 719             struct fuse_entry_param e;
 720             if (is_dot_or_dotdot(name)) {
 721                 e = (struct fuse_entry_param){
 722                     .attr.st_ino = d->entry->d_ino,
 723                     .attr.st_mode = d->entry->d_type << 12,
 724                 };
 725             } else {
 726                 err = lo_do_lookup(req, ino, name, &e);
 727                 if (err) {
 728                     goto error;
 729                 }
 730                 entry_ino = e.ino;
 731             }
 732
 733             entsize = fuse_add_direntry_plus(req, p, rem, name, &e, nextoff);
 734         } else {
 735             struct stat st = {
 736                 .st_ino = d->entry->d_ino,
 737                 .st_mode = d->entry->d_type << 12,
 738             };
 739             entsize = fuse_add_direntry(req, p, rem, name, &st, nextoff);
 740         }
 741         if (entsize > rem) {
 742             if (entry_ino != 0) {
 743                 lo_forget_one(req, entry_ino, 1);
 744             }
 745             break;
 746         }
 747
 748         p += entsize;
 749         rem -= entsize;
 750
 751         d->entry = NULL;
 752         d->offset = nextoff;
 753     }
 754
 755     err = 0;
 756 error:
 757     /*
 758      * If there's an error, we can only signal it if we haven't stored
 759      * any entries yet - otherwise we'd end up with wrong lookup
 760      * counts for the entries that are already in the buffer. So we
 761      * return what we've collected until that point.
 762      */
 763     if (err && rem == size) {
 764         fuse_reply_err(req, err);
 765     } else {
 766         fuse_reply_buf(req, buf, size - rem);
 767     }
 768     free(buf);
 769 }
 770
 771 static void lo_readdir(fuse_req_t req, fuse_ino_t ino, size_t size,
 772                        off_t offset, struct fuse_file_info *fi)
 773 {
 774     lo_do_readdir(req, ino, size, offset, fi, 0);
 775 }
 776
 777 static void lo_readdirplus(fuse_req_t req, fuse_ino_t ino, size_t size,
 778                            off_t offset, struct fuse_file_info *fi)
 779 {
 780     lo_do_readdir(req, ino, size, offset, fi, 1);
 781 }
 782
 783 static void lo_releasedir(fuse_req_t req, fuse_ino_t ino,
 784                           struct fuse_file_info *fi)
 785 {
 786     struct lo_dirp *d = lo_dirp(fi);
 787     (void)ino;
 788     closedir(d->dp);
 789     free(d);
 790     fuse_reply_err(req, 0);
 791 }
 792
 793 static void lo_create(fuse_req_t req, fuse_ino_t parent, const char *name,
 794                       mode_t mode, struct fuse_file_info *fi)
 795 {
 796     int fd;
 797     struct lo_data *lo = lo_data(req);
 798     struct fuse_entry_param e;
 799     int err;
 800
 801     if (lo_debug(req)) {
 802         fuse_log(FUSE_LOG_DEBUG, "lo_create(parent=%" PRIu64 ", name=%s)\n",
 803                  parent, name);
 804     }
 805
 806     fd = openat(lo_fd(req, parent), name, (fi->flags | O_CREAT) & ~O_NOFOLLOW,
 807                 mode);
 808     if (fd == -1) {
 809         return (void)fuse_reply_err(req, errno);
 810     }
 811
 812     fi->fh = fd;
 813     if (lo->cache == CACHE_NEVER) {
 814         fi->direct_io = 1;
 815     } else if (lo->cache == CACHE_ALWAYS) {
 816         fi->keep_cache = 1;
 817     }
 818
 819     err = lo_do_lookup(req, parent, name, &e);
 820     if (err) {
 821         fuse_reply_err(req, err);
 822     } else {
 823         fuse_reply_create(req, &e, fi);
 824     }
 825 }
 826
 827 static void lo_fsyncdir(fuse_req_t req, fuse_ino_t ino, int datasync,
 828                         struct fuse_file_info *fi)
 829 {
 830     int res;
 831     int fd = dirfd(lo_dirp(fi)->dp);
 832     (void)ino;
 833     if (datasync) {
 834         res = fdatasync(fd);
 835     } else {
 836         res = fsync(fd);
 837     }
 838     fuse_reply_err(req, res == -1 ? errno : 0);
 839 }
 840
 841 static void lo_open(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi)
 842 {
 843     int fd;
 844     char buf[64];
 845     struct lo_data *lo = lo_data(req);
 846
 847     if (lo_debug(req)) {
 848         fuse_log(FUSE_LOG_DEBUG, "lo_open(ino=%" PRIu64 ", flags=%d)\n", ino,
 849                  fi->flags);
 850     }
 851
 852     /*
 853      * With writeback cache, kernel may send read requests even
 854      * when userspace opened write-only
 855      */
 856     if (lo->writeback && (fi->flags & O_ACCMODE) == O_WRONLY) {
 857         fi->flags &= ~O_ACCMODE;
 858         fi->flags |= O_RDWR;
 859     }
 860
 861     /*
 862      * With writeback cache, O_APPEND is handled by the kernel.
 863      * This breaks atomicity (since the file may change in the
 864      * underlying filesystem, so that the kernel's idea of the
 865      * end of the file isn't accurate anymore). In this example,
 866      * we just accept that. A more rigorous filesystem may want
 867      * to return an error here
 868      */
 869     if (lo->writeback && (fi->flags & O_APPEND)) {
 870         fi->flags &= ~O_APPEND;
 871     }
 872
 873     sprintf(buf, "/proc/self/fd/%i", lo_fd(req, ino));
 874     fd = open(buf, fi->flags & ~O_NOFOLLOW);
 875     if (fd == -1) {
 876         return (void)fuse_reply_err(req, errno);
 877     }
 878
 879     fi->fh = fd;
 880     if (lo->cache == CACHE_NEVER) {
 881         fi->direct_io = 1;
 882     } else if (lo->cache == CACHE_ALWAYS) {
 883         fi->keep_cache = 1;
 884     }
 885     fuse_reply_open(req, fi);
 886 }
 887
 888 static void lo_release(fuse_req_t req, fuse_ino_t ino,
 889                        struct fuse_file_info *fi)
 890 {
 891     (void)ino;
 892
 893     close(fi->fh);
 894     fuse_reply_err(req, 0);
 895 }
 896
 897 static void lo_flush(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi)
 898 {
 899     int res;
 900     (void)ino;
 901     res = close(dup(fi->fh));
 902     fuse_reply_err(req, res == -1 ? errno : 0);
 903 }
 904
 905 static void lo_fsync(fuse_req_t req, fuse_ino_t ino, int datasync,
 906                      struct fuse_file_info *fi)
 907 {
 908     int res;
 909     (void)ino;
 910     if (datasync) {
 911         res = fdatasync(fi->fh);
 912     } else {
 913         res = fsync(fi->fh);
 914     }
 915     fuse_reply_err(req, res == -1 ? errno : 0);
 916 }
 917
 918 static void lo_read(fuse_req_t req, fuse_ino_t ino, size_t size, off_t offset,
 919                     struct fuse_file_info *fi)
 920 {
 921     struct fuse_bufvec buf = FUSE_BUFVEC_INIT(size);
 922
 923     if (lo_debug(req)) {
 924         fuse_log(FUSE_LOG_DEBUG,
 925                  "lo_read(ino=%" PRIu64 ", size=%zd, "
 926                  "off=%lu)\n",
 927                  ino, size, (unsigned long)offset);
 928     }
 929
 930     buf.buf[0].flags = FUSE_BUF_IS_FD | FUSE_BUF_FD_SEEK;
 931     buf.buf[0].fd = fi->fh;
 932     buf.buf[0].pos = offset;
 933
 934     fuse_reply_data(req, &buf);
 935 }
 936
 937 static void lo_write_buf(fuse_req_t req, fuse_ino_t ino,
 938                          struct fuse_bufvec *in_buf, off_t off,
 939                          struct fuse_file_info *fi)
 940 {
 941     (void)ino;
 942     ssize_t res;
 943     struct fuse_bufvec out_buf = FUSE_BUFVEC_INIT(fuse_buf_size(in_buf));
 944
 945     out_buf.buf[0].flags = FUSE_BUF_IS_FD | FUSE_BUF_FD_SEEK;
 946     out_buf.buf[0].fd = fi->fh;
 947     out_buf.buf[0].pos = off;
 948
 949     if (lo_debug(req)) {
 950         fuse_log(FUSE_LOG_DEBUG,
 951                  "lo_write(ino=%" PRIu64 ", size=%zd, off=%lu)\n", ino,
 952                  out_buf.buf[0].size, (unsigned long)off);
 953     }
 954
 955     res = fuse_buf_copy(&out_buf, in_buf);
 956     if (res < 0) {
 957         fuse_reply_err(req, -res);
 958     } else {
 959         fuse_reply_write(req, (size_t)res);
 960     }
 961 }
 962
 963 static void lo_statfs(fuse_req_t req, fuse_ino_t ino)
 964 {
 965     int res;
 966     struct statvfs stbuf;
 967
 968     res = fstatvfs(lo_fd(req, ino), &stbuf);
 969     if (res == -1) {
 970         fuse_reply_err(req, errno);
 971     } else {
 972         fuse_reply_statfs(req, &stbuf);
 973     }
 974 }
 975
 976 static void lo_fallocate(fuse_req_t req, fuse_ino_t ino, int mode, off_t offset,
 977                          off_t length, struct fuse_file_info *fi)
 978 {
 979     int err = EOPNOTSUPP;
 980     (void)ino;
 981
 982 #ifdef HAVE_FALLOCATE
 983     err = fallocate(fi->fh, mode, offset, length);
 984     if (err < 0) {
 985         err = errno;
 986     }
 987
 988 #elif defined(HAVE_POSIX_FALLOCATE)
 989     if (mode) {
 990         fuse_reply_err(req, EOPNOTSUPP);
 991         return;
 992     }
 993
 994     err = posix_fallocate(fi->fh, offset, length);
 995 #endif
 996
 997     fuse_reply_err(req, err);
 998 }
 999
1000 static void lo_flock(fuse_req_t req, fuse_ino_t ino, struct fuse_file_info *fi,
1001                      int op)
1002 {
1003     int res;
1004     (void)ino;
1005
1006     res = flock(fi->fh, op);
1007
1008     fuse_reply_err(req, res == -1 ? errno : 0);
1009 }
1010
1011 static void lo_getxattr(fuse_req_t req, fuse_ino_t ino, const char *name,
1012                         size_t size)
1013 {
1014     char *value = NULL;
1015     char procname[64];
1016     struct lo_inode *inode = lo_inode(req, ino);
1017     ssize_t ret;
1018     int saverr;
1019
1020     saverr = ENOSYS;
1021     if (!lo_data(req)->xattr) {
1022         goto out;
1023     }
1024
1025     if (lo_debug(req)) {
1026         fuse_log(FUSE_LOG_DEBUG,
1027                  "lo_getxattr(ino=%" PRIu64 ", name=%s size=%zd)\n", ino, name,
1028                  size);
1029     }
1030
1031     if (inode->is_symlink) {
1032         /* Sorry, no race free way to getxattr on symlink. */
1033         saverr = EPERM;
1034         goto out;
1035     }
1036
1037     sprintf(procname, "/proc/self/fd/%i", inode->fd);
1038
1039     if (size) {
1040         value = malloc(size);
1041         if (!value) {
1042             goto out_err;
1043         }
1044
1045         ret = getxattr(procname, name, value, size);
1046         if (ret == -1) {
1047             goto out_err;
1048         }
1049         saverr = 0;
1050         if (ret == 0) {
1051             goto out;
1052         }
1053
1054         fuse_reply_buf(req, value, ret);
1055     } else {
1056         ret = getxattr(procname, name, NULL, 0);
1057         if (ret == -1) {
1058             goto out_err;
1059         }
1060
1061         fuse_reply_xattr(req, ret);
1062     }
1063 out_free:
1064     free(value);
1065     return;
1066
1067 out_err:
1068     saverr = errno;
1069 out:
1070     fuse_reply_err(req, saverr);
1071     goto out_free;
1072 }
1073
1074 static void lo_listxattr(fuse_req_t req, fuse_ino_t ino, size_t size)
1075 {
1076     char *value = NULL;
1077     char procname[64];
1078     struct lo_inode *inode = lo_inode(req, ino);
1079     ssize_t ret;
1080     int saverr;
1081
1082     saverr = ENOSYS;
1083     if (!lo_data(req)->xattr) {
1084         goto out;
1085     }
1086
1087     if (lo_debug(req)) {
1088         fuse_log(FUSE_LOG_DEBUG, "lo_listxattr(ino=%" PRIu64 ", size=%zd)\n",
1089                  ino, size);
1090     }
1091
1092     if (inode->is_symlink) {
1093         /* Sorry, no race free way to listxattr on symlink. */
1094         saverr = EPERM;
1095         goto out;
1096     }
1097
1098     sprintf(procname, "/proc/self/fd/%i", inode->fd);
1099
1100     if (size) {
1101         value = malloc(size);
1102         if (!value) {
1103             goto out_err;
1104         }
1105
1106         ret = listxattr(procname, value, size);
1107         if (ret == -1) {
1108             goto out_err;
1109         }
1110         saverr = 0;
1111         if (ret == 0) {
1112             goto out;
1113         }
1114
1115         fuse_reply_buf(req, value, ret);
1116     } else {
1117         ret = listxattr(procname, NULL, 0);
1118         if (ret == -1) {
1119             goto out_err;
1120         }
1121
1122         fuse_reply_xattr(req, ret);
1123     }
1124 out_free:
1125     free(value);
1126     return;
1127
1128 out_err:
1129     saverr = errno;
1130 out:
1131     fuse_reply_err(req, saverr);
1132     goto out_free;
1133 }
1134
1135 static void lo_setxattr(fuse_req_t req, fuse_ino_t ino, const char *name,
1136                         const char *value, size_t size, int flags)
1137 {
1138     char procname[64];
1139     struct lo_inode *inode = lo_inode(req, ino);
1140     ssize_t ret;
1141     int saverr;
1142
1143     saverr = ENOSYS;
1144     if (!lo_data(req)->xattr) {
1145         goto out;
1146     }
1147
1148     if (lo_debug(req)) {
1149         fuse_log(FUSE_LOG_DEBUG,
1150                  "lo_setxattr(ino=%" PRIu64 ", name=%s value=%s size=%zd)\n",
1151                  ino, name, value, size);
1152     }
1153
1154     if (inode->is_symlink) {
1155         /* Sorry, no race free way to setxattr on symlink. */
1156         saverr = EPERM;
1157         goto out;
1158     }
1159
1160     sprintf(procname, "/proc/self/fd/%i", inode->fd);
1161
1162     ret = setxattr(procname, name, value, size, flags);
1163     saverr = ret == -1 ? errno : 0;
1164
1165 out:
1166     fuse_reply_err(req, saverr);
1167 }
1168
1169 static void lo_removexattr(fuse_req_t req, fuse_ino_t ino, const char *name)
1170 {
1171     char procname[64];
1172     struct lo_inode *inode = lo_inode(req, ino);
1173     ssize_t ret;
1174     int saverr;
1175
1176     saverr = ENOSYS;
1177     if (!lo_data(req)->xattr) {
1178         goto out;
1179     }
1180
1181     if (lo_debug(req)) {
1182         fuse_log(FUSE_LOG_DEBUG, "lo_removexattr(ino=%" PRIu64 ", name=%s)\n",
1183                  ino, name);
1184     }
1185
1186     if (inode->is_symlink) {
1187         /* Sorry, no race free way to setxattr on symlink. */
1188         saverr = EPERM;
1189         goto out;
1190     }
1191
1192     sprintf(procname, "/proc/self/fd/%i", inode->fd);
1193
1194     ret = removexattr(procname, name);
1195     saverr = ret == -1 ? errno : 0;
1196
1197 out:
1198     fuse_reply_err(req, saverr);
1199 }
1200
1201 #ifdef HAVE_COPY_FILE_RANGE
1202 static void lo_copy_file_range(fuse_req_t req, fuse_ino_t ino_in, off_t off_in,
1203                                struct fuse_file_info *fi_in, fuse_ino_t ino_out,
1204                                off_t off_out, struct fuse_file_info *fi_out,
1205                                size_t len, int flags)
1206 {
1207     ssize_t res;
1208
1209     if (lo_debug(req))
1210         fuse_log(FUSE_LOG_DEBUG,
1211                  "lo_copy_file_range(ino=%" PRIu64 "/fd=%lu, "
1212                  "off=%lu, ino=%" PRIu64 "/fd=%lu, "
1213                  "off=%lu, size=%zd, flags=0x%x)\n",
1214                  ino_in, fi_in->fh, off_in, ino_out, fi_out->fh, off_out, len,
1215                  flags);
1216
1217     res = copy_file_range(fi_in->fh, &off_in, fi_out->fh, &off_out, len, flags);
1218     if (res < 0) {
1219         fuse_reply_err(req, -errno);
1220     } else {
1221         fuse_reply_write(req, res);
1222     }
1223 }
1224 #endif
1225
1226 static void lo_lseek(fuse_req_t req, fuse_ino_t ino, off_t off, int whence,
1227                      struct fuse_file_info *fi)
1228 {
1229     off_t res;
1230
1231     (void)ino;
1232     res = lseek(fi->fh, off, whence);
1233     if (res != -1) {
1234         fuse_reply_lseek(req, res);
1235     } else {
1236         fuse_reply_err(req, errno);
1237     }
1238 }
1239
1240 static struct fuse_lowlevel_ops lo_oper = {
1241     .init = lo_init,
1242     .lookup = lo_lookup,
1243     .mkdir = lo_mkdir,
1244     .mknod = lo_mknod,
1245     .symlink = lo_symlink,
1246     .link = lo_link,
1247     .unlink = lo_unlink,
1248     .rmdir = lo_rmdir,
1249     .rename = lo_rename,
1250     .forget = lo_forget,
1251     .forget_multi = lo_forget_multi,
1252     .getattr = lo_getattr,
1253     .setattr = lo_setattr,
1254     .readlink = lo_readlink,
1255     .opendir = lo_opendir,
1256     .readdir = lo_readdir,
1257     .readdirplus = lo_readdirplus,
1258     .releasedir = lo_releasedir,
1259     .fsyncdir = lo_fsyncdir,
1260     .create = lo_create,
1261     .open = lo_open,
1262     .release = lo_release,
1263     .flush = lo_flush,
1264     .fsync = lo_fsync,
1265     .read = lo_read,
1266     .write_buf = lo_write_buf,
1267     .statfs = lo_statfs,
1268     .fallocate = lo_fallocate,
1269     .flock = lo_flock,
1270     .getxattr = lo_getxattr,
1271     .listxattr = lo_listxattr,
1272     .setxattr = lo_setxattr,
1273     .removexattr = lo_removexattr,
1274 #ifdef HAVE_COPY_FILE_RANGE
1275     .copy_file_range = lo_copy_file_range,
1276 #endif
1277     .lseek = lo_lseek,
1278 };
1279
1280 int main(int argc, char *argv[])
1281 {
1282     struct fuse_args args = FUSE_ARGS_INIT(argc, argv);
1283     struct fuse_session *se;
1284     struct fuse_cmdline_opts opts;
1285     struct lo_data lo = { .debug = 0, .writeback = 0 };
1286     int ret = -1;
1287
1288     /* Don't mask creation mode, kernel already did that */
1289     umask(0);
1290
1291     pthread_mutex_init(&lo.mutex, NULL);
1292     lo.root.next = lo.root.prev = &lo.root;
1293     lo.root.fd = -1;
1294     lo.cache = CACHE_NORMAL;
1295
1296     if (fuse_parse_cmdline(&args, &opts) != 0) {
1297         return 1;
1298     }
1299     if (opts.show_help) {
1300         printf("usage: %s [options]\n\n", argv[0]);
1301         fuse_cmdline_help();
1302         fuse_lowlevel_help();
1303         ret = 0;
1304         goto err_out1;
1305     } else if (opts.show_version) {
1306         fuse_lowlevel_version();
1307         ret = 0;
1308         goto err_out1;
1309     }
1310
1311     if (fuse_opt_parse(&args, &lo, lo_opts, NULL) == -1) {
1312         return 1;
1313     }
1314
1315     lo.debug = opts.debug;
1316     lo.root.refcount = 2;
1317     if (lo.source) {
1318         struct stat stat;
1319         int res;
1320
1321         res = lstat(lo.source, &stat);
1322         if (res == -1) {
1323             fuse_log(FUSE_LOG_ERR, "failed to stat source (\"%s\"): %m\n",
1324                      lo.source);
1325             exit(1);
1326         }
1327         if (!S_ISDIR(stat.st_mode)) {
1328             fuse_log(FUSE_LOG_ERR, "source is not a directory\n");
1329             exit(1);
1330         }
1331
1332     } else {
1333         lo.source = "/";
1334     }
1335     lo.root.is_symlink = false;
1336     if (!lo.timeout_set) {
1337         switch (lo.cache) {
1338         case CACHE_NEVER:
1339             lo.timeout = 0.0;
1340             break;
1341
1342         case CACHE_NORMAL:
1343             lo.timeout = 1.0;
1344             break;
1345
1346         case CACHE_ALWAYS:
1347             lo.timeout = 86400.0;
1348             break;
1349         }
1350     } else if (lo.timeout < 0) {
1351         fuse_log(FUSE_LOG_ERR, "timeout is negative (%lf)\n", lo.timeout);
1352         exit(1);
1353     }
1354
1355     lo.root.fd = open(lo.source, O_PATH);
1356     if (lo.root.fd == -1) {
1357         fuse_log(FUSE_LOG_ERR, "open(\"%s\", O_PATH): %m\n", lo.source);
1358         exit(1);
1359     }
1360
1361     se = fuse_session_new(&args, &lo_oper, sizeof(lo_oper), &lo);
1362     if (se == NULL) {
1363         goto err_out1;
1364     }
1365
1366     if (fuse_set_signal_handlers(se) != 0) {
1367         goto err_out2;
1368     }
1369
1370     if (fuse_session_mount(se) != 0) {
1371         goto err_out3;
1372     }
1373
1374     fuse_daemonize(opts.foreground);
1375
1376     /* Block until ctrl+c or fusermount -u */
1377     if (opts.singlethread) {
1378         ret = fuse_session_loop(se);
1379     } else {
1380         ret = fuse_session_loop_mt(se, opts.clone_fd);
1381     }
1382
1383     fuse_session_unmount(se);
1384 err_out3:
1385     fuse_remove_signal_handlers(se);
1386 err_out2:
1387     fuse_session_destroy(se);
1388 err_out1:
1389     fuse_opt_free_args(&args);
1390
1391     if (lo.root.fd >= 0) {
1392         close(lo.root.fd);
1393     }
1394
1395     return ret ? 1 : 0;
1396 }