fuse.c

/*
  FUSE: Filesystem in Userspace
  Copyright (C) 2001-2007  Miklos Szeredi <miklos@szeredi.hu>

  Implementation of the high-level FUSE API on top of the low-level
  API.

  This program can be distributed under the terms of the GNU LGPLv2.
  See the file COPYING.LIB
*/


/* For pthread_rwlock_t */
#define _GNU_SOURCE

#include "config.h"
#include "fuse_i.h"
#include "fuse_lowlevel.h"
#include "fuse_opt.h"
#include "fuse_misc.h"
#include "fuse_kernel.h"

#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <stddef.h>
#include <stdbool.h>
#include <unistd.h>
#include <time.h>
#include <fcntl.h>
#include <limits.h>
#include <errno.h>
#include <signal.h>
#include <dlfcn.h>
#include <assert.h>
#include <poll.h>
#include <sys/param.h>
#include <sys/uio.h>
#include <sys/time.h>
#include <sys/mman.h>
#include <sys/file.h>

#define FUSE_NODE_SLAB 1

#ifndef MAP_ANONYMOUS
#undef FUSE_NODE_SLAB
#endif

#ifndef RENAME_EXCHANGE
#define RENAME_EXCHANGE         (1 << 1)        /* Exchange source and dest */
#endif

#define FUSE_DEFAULT_INTR_SIGNAL SIGUSR1

#define FUSE_UNKNOWN_INO 0xffffffff
#define OFFSET_MAX 0x7fffffffffffffffLL

#define NODE_TABLE_MIN_SIZE 8192

struct fuse_fs {
        struct fuse_operations op;
        struct fuse_module *m;
        void *user_data;
        int debug;
};

struct fusemod_so {
        void *handle;
        int ctr;
};

struct lock_queue_element {
        struct lock_queue_element *next;
        pthread_cond_t cond;
        fuse_ino_t nodeid1;
        const char *name1;
        char **path1;
        struct node **wnode1;
        fuse_ino_t nodeid2;
        const char *name2;
        char **path2;
        struct node **wnode2;
        int err;
        bool first_locked : 1;
        bool second_locked : 1;
        bool done : 1;
};

struct node_table {
        struct node **array;
        size_t use;
        size_t size;
        size_t split;
};

#define container_of(ptr, type, member) ({                              \
                        const typeof( ((type *)0)->member ) *__mptr = (ptr); \
                        (type *)( (char *)__mptr - offsetof(type,member) );})

#define list_entry(ptr, type, member)           \
        container_of(ptr, type, member)

struct list_head {
        struct list_head *next;
        struct list_head *prev;
};

struct node_slab {
        struct list_head list;  /* must be the first member */
        struct list_head freelist;
        int used;
};

struct fuse {
        struct fuse_session *se;
        struct node_table name_table;
        struct node_table id_table;
        struct list_head lru_table;
        fuse_ino_t ctr;
        unsigned int generation;
        unsigned int hidectr;
        pthread_mutex_t lock;
        struct fuse_config conf;
        int intr_installed;
        struct fuse_fs *fs;
        struct lock_queue_element *lockq;
        int pagesize;
        struct list_head partial_slabs;
        struct list_head full_slabs;
        pthread_t prune_thread;
};

struct lock {
        int type;
        off_t start;
        off_t end;
        pid_t pid;
        uint64_t owner;
        struct lock *next;
};

struct node {
        struct node *name_next;
        struct node *id_next;
        fuse_ino_t nodeid;
        unsigned int generation;
        int refctr;
        struct node *parent;
        char *name;
        uint64_t nlookup;
        int open_count;
        struct timespec stat_updated;
        struct timespec mtime;
        off_t size;
        struct lock *locks;
        unsigned int is_hidden : 1;
        unsigned int cache_valid : 1;
        int treelock;
        char inline_name[32];
};

#define TREELOCK_WRITE -1
#define TREELOCK_WAIT_OFFSET INT_MIN

struct node_lru {
        struct node node;
        struct list_head lru;
        struct timespec forget_time;
};

struct fuse_direntry {
        struct stat stat;
        char *name;
        struct fuse_direntry *next;
};

struct fuse_dh {
        pthread_mutex_t lock;
        struct fuse *fuse;
        fuse_req_t req;
        char *contents;
        struct fuse_direntry *first;
        struct fuse_direntry **last;
        unsigned len;
        unsigned size;
        unsigned needlen;
        int filled;
        uint64_t fh;
        int error;
        fuse_ino_t nodeid;
};

struct fuse_context_i {
        struct fuse_context ctx;
        fuse_req_t req;
};

/* Defined by FUSE_REGISTER_MODULE() in lib/modules/subdir.c and iconv.c.  */
extern fuse_module_factory_t fuse_module_subdir_factory;
#ifdef HAVE_ICONV
extern fuse_module_factory_t fuse_module_iconv_factory;
#endif

static pthread_key_t fuse_context_key;
static pthread_mutex_t fuse_context_lock = PTHREAD_MUTEX_INITIALIZER;
static int fuse_context_ref;
static struct fuse_module *fuse_modules = NULL;

static int fuse_register_module(const char *name,
                                fuse_module_factory_t factory,
                                struct fusemod_so *so)
{
        struct fuse_module *mod;

        mod = calloc(1, sizeof(struct fuse_module));
        if (!mod) {
                fprintf(stderr, "fuse: failed to allocate module\n");
                return -1;
        }
        mod->name = strdup(name);
        if (!mod->name) {
                fprintf(stderr, "fuse: failed to allocate module name\n");
                free(mod);
                return -1;
        }
        mod->factory = factory;
        mod->ctr = 0;
        mod->so = so;
        if (mod->so)
                mod->so->ctr++;
        mod->next = fuse_modules;
        fuse_modules = mod;

        return 0;
}

static void fuse_unregister_module(struct fuse_module *m)
{
        struct fuse_module **mp;
        for (mp = &fuse_modules; *mp; mp = &(*mp)->next) {
                if (*mp == m) {
                        *mp = (*mp)->next;
                        break;
                }
        }
        free(m->name);
        free(m);
}

static int fuse_load_so_module(const char *module)
{
        int ret = -1;
        char *tmp;
        struct fusemod_so *so;
        fuse_module_factory_t factory;

        tmp = malloc(strlen(module) + 64);
        if (!tmp) {
                fprintf(stderr, "fuse: memory allocation failed\n");
                return -1;
        }
        sprintf(tmp, "libfusemod_%s.so", module);
        so = calloc(1, sizeof(struct fusemod_so));
        if (!so) {
                fprintf(stderr, "fuse: failed to allocate module so\n");
                goto out;
        }

        so->handle = dlopen(tmp, RTLD_NOW);
        if (so->handle == NULL) {
                fprintf(stderr, "fuse: dlopen(%s) failed: %s\n",
                        tmp, dlerror());
                goto out_free_so;
        }

        sprintf(tmp, "fuse_module_%s_factory", module);
        *(void**)(&factory) = dlsym(so->handle, tmp);
        if (factory == NULL) {
                fprintf(stderr, "fuse: symbol <%s> not found in module: %s\n",
                        tmp, dlerror());
                goto out_dlclose;
        }
        ret = fuse_register_module(module, factory, so);
        if (ret)
                goto out_dlclose;

out:
        free(tmp);
        return ret;

out_dlclose:
        dlclose(so->handle);
out_free_so:
        free(so);
        goto out;
}

static struct fuse_module *fuse_find_module(const char *module)
{
        struct fuse_module *m;
        for (m = fuse_modules; m; m = m->next) {
                if (strcmp(module, m->name) == 0) {
                        m->ctr++;
                        break;
                }
        }
        return m;
}

static struct fuse_module *fuse_get_module(const char *module)
{
        struct fuse_module *m;

        pthread_mutex_lock(&fuse_context_lock);
        m = fuse_find_module(module);
        if (!m) {
                int err = fuse_load_so_module(module);
                if (!err)
                        m = fuse_find_module(module);
        }
        pthread_mutex_unlock(&fuse_context_lock);
        return m;
}

static void fuse_put_module(struct fuse_module *m)
{
        pthread_mutex_lock(&fuse_context_lock);
        if (m->so)
                assert(m->ctr > 0);
        /* Builtin modules may already have m->ctr == 0 */
        if (m->ctr > 0)
                m->ctr--;
        if (!m->ctr && m->so) {
                struct fusemod_so *so = m->so;
                assert(so->ctr > 0);
                so->ctr--;
                if (!so->ctr) {
                        struct fuse_module **mp;
                        for (mp = &fuse_modules; *mp;) {
                                if ((*mp)->so == so)
                                        fuse_unregister_module(*mp);
                                else
                                        mp = &(*mp)->next;
                        }
                        dlclose(so->handle);
                        free(so);
                }
        } else if (!m->ctr) {
                fuse_unregister_module(m);
        }
        pthread_mutex_unlock(&fuse_context_lock);
}

static void init_list_head(struct list_head *list)
{
        list->next = list;
        list->prev = list;
}

static int list_empty(const struct list_head *head)
{
        return head->next == head;
}

static void list_add(struct list_head *new, struct list_head *prev,
                     struct list_head *next)
{
        next->prev = new;
        new->next = next;
        new->prev = prev;
        prev->next = new;
}

static inline void list_add_head(struct list_head *new, struct list_head *head)
{
        list_add(new, head, head->next);
}

static inline void list_add_tail(struct list_head *new, struct list_head *head)
{
        list_add(new, head->prev, head);
}

static inline void list_del(struct list_head *entry)
{
        struct list_head *prev = entry->prev;
        struct list_head *next = entry->next;

        next->prev = prev;
        prev->next = next;
}

static inline int lru_enabled(struct fuse *f)
{
        return f->conf.remember > 0;
}

static struct node_lru *node_lru(struct node *node)
{
        return (struct node_lru *) node;
}

static size_t get_node_size(struct fuse *f)
{
        if (lru_enabled(f))
                return sizeof(struct node_lru);
        else
                return sizeof(struct node);
}

#ifdef FUSE_NODE_SLAB
static struct node_slab *list_to_slab(struct list_head *head)
{
        return (struct node_slab *) head;
}

static struct node_slab *node_to_slab(struct fuse *f, struct node *node)
{
        return (struct node_slab *) (((uintptr_t) node) & ~((uintptr_t) f->pagesize - 1));
}

static int alloc_slab(struct fuse *f)
{
        void *mem;
        struct node_slab *slab;
        char *start;
        size_t num;
        size_t i;
        size_t node_size = get_node_size(f);

        mem = mmap(NULL, f->pagesize, PROT_READ | PROT_WRITE,
                   MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);

        if (mem == MAP_FAILED)
                return -1;

        slab = mem;
        init_list_head(&slab->freelist);
        slab->used = 0;
        num = (f->pagesize - sizeof(struct node_slab)) / node_size;

        start = (char *) mem + f->pagesize - num * node_size;
        for (i = 0; i < num; i++) {
                struct list_head *n;

                n = (struct list_head *) (start + i * node_size);
                list_add_tail(n, &slab->freelist);
        }
        list_add_tail(&slab->list, &f->partial_slabs);

        return 0;
}

static struct node *alloc_node(struct fuse *f)
{
        struct node_slab *slab;
        struct list_head *node;

        if (list_empty(&f->partial_slabs)) {
                int res = alloc_slab(f);
                if (res != 0)
                        return NULL;
        }
        slab = list_to_slab(f->partial_slabs.next);
        slab->used++;
        node = slab->freelist.next;
        list_del(node);
        if (list_empty(&slab->freelist)) {
                list_del(&slab->list);
                list_add_tail(&slab->list, &f->full_slabs);
        }
        memset(node, 0, sizeof(struct node));

        return (struct node *) node;
}

static void free_slab(struct fuse *f, struct node_slab *slab)
{
        int res;

        list_del(&slab->list);
        res = munmap(slab, f->pagesize);
        if (res == -1)
                fprintf(stderr, "fuse warning: munmap(%p) failed\n", slab);
}

static void free_node_mem(struct fuse *f, struct node *node)
{
        struct node_slab *slab = node_to_slab(f, node);
        struct list_head *n = (struct list_head *) node;

        slab->used--;
        if (slab->used) {
                if (list_empty(&slab->freelist)) {
                        list_del(&slab->list);
                        list_add_tail(&slab->list, &f->partial_slabs);
                }
                list_add_head(n, &slab->freelist);
        } else {
                free_slab(f, slab);
        }
}
#else
static struct node *alloc_node(struct fuse *f)
{
        return (struct node *) calloc(1, get_node_size(f));
}

static void free_node_mem(struct fuse *f, struct node *node)
{
        (void) f;
        free(node);
}
#endif

static size_t id_hash(struct fuse *f, fuse_ino_t ino)
{
        uint64_t hash = ((uint32_t) ino * 2654435761U) % f->id_table.size;
        uint64_t oldhash = hash % (f->id_table.size / 2);

        if (oldhash >= f->id_table.split)
                return oldhash;
        else
                return hash;
}

static struct node *get_node_nocheck(struct fuse *f, fuse_ino_t nodeid)
{
        size_t hash = id_hash(f, nodeid);
        struct node *node;

        for (node = f->id_table.array[hash]; node != NULL; node = node->id_next)
                if (node->nodeid == nodeid)
                        return node;

        return NULL;
}

static struct node *get_node(struct fuse *f, fuse_ino_t nodeid)
{
        struct node *node = get_node_nocheck(f, nodeid);
        if (!node) {
                fprintf(stderr, "fuse internal error: node %llu not found\n",
                        (unsigned long long) nodeid);
                abort();
        }
        return node;
}

static void curr_time(struct timespec *now);
static double diff_timespec(const struct timespec *t1,
                           const struct timespec *t2);

static void remove_node_lru(struct node *node)
{
        struct node_lru *lnode = node_lru(node);
        list_del(&lnode->lru);
        init_list_head(&lnode->lru);
}

static void set_forget_time(struct fuse *f, struct node *node)
{
        struct node_lru *lnode = node_lru(node);

        list_del(&lnode->lru);
        list_add_tail(&lnode->lru, &f->lru_table);
        curr_time(&lnode->forget_time);
}

static void free_node(struct fuse *f, struct node *node)
{
        if (node->name != node->inline_name)
                free(node->name);
        free_node_mem(f, node);
}

static void node_table_reduce(struct node_table *t)
{
        size_t newsize = t->size / 2;
        void *newarray;

        if (newsize < NODE_TABLE_MIN_SIZE)
                return;

        newarray = realloc(t->array, sizeof(struct node *) * newsize);
        if (newarray != NULL)
                t->array = newarray;

        t->size = newsize;
        t->split = t->size / 2;
}

static void remerge_id(struct fuse *f)
{
        struct node_table *t = &f->id_table;
        int iter;

        if (t->split == 0)
                node_table_reduce(t);

        for (iter = 8; t->split > 0 && iter; iter--) {
                struct node **upper;

                t->split--;
                upper = &t->array[t->split + t->size / 2];
                if (*upper) {
                        struct node **nodep;

                        for (nodep = &t->array[t->split]; *nodep;
                             nodep = &(*nodep)->id_next);

                        *nodep = *upper;
                        *upper = NULL;
                        break;
                }
        }
}

static void unhash_id(struct fuse *f, struct node *node)
{
        struct node **nodep = &f->id_table.array[id_hash(f, node->nodeid)];

        for (; *nodep != NULL; nodep = &(*nodep)->id_next)
                if (*nodep == node) {
                        *nodep = node->id_next;
                        f->id_table.use--;

                        if(f->id_table.use < f->id_table.size / 4)
                                remerge_id(f);
                        return;
                }
}

static int node_table_resize(struct node_table *t)
{
        size_t newsize = t->size * 2;
        void *newarray;

        newarray = realloc(t->array, sizeof(struct node *) * newsize);
        if (newarray == NULL)
                return -1;

        t->array = newarray;
        memset(t->array + t->size, 0, t->size * sizeof(struct node *));
        t->size = newsize;
        t->split = 0;

        return 0;
}

static void rehash_id(struct fuse *f)
{
        struct node_table *t = &f->id_table;
        struct node **nodep;
        struct node **next;
        size_t hash;

        if (t->split == t->size / 2)
                return;

        hash = t->split;
        t->split++;
        for (nodep = &t->array[hash]; *nodep != NULL; nodep = next) {
                struct node *node = *nodep;
                size_t newhash = id_hash(f, node->nodeid);

                if (newhash != hash) {
                        next = nodep;
                        *nodep = node->id_next;
                        node->id_next = t->array[newhash];
                        t->array[newhash] = node;
                } else {
                        next = &node->id_next;
                }
        }
        if (t->split == t->size / 2)
                node_table_resize(t);
}

static void hash_id(struct fuse *f, struct node *node)
{
        size_t hash = id_hash(f, node->nodeid);
        node->id_next = f->id_table.array[hash];
        f->id_table.array[hash] = node;
        f->id_table.use++;

        if (f->id_table.use >= f->id_table.size / 2)
                rehash_id(f);
}

static size_t name_hash(struct fuse *f, fuse_ino_t parent,
                        const char *name)
{
        uint64_t hash = parent;
        uint64_t oldhash;

        for (; *name; name++)
                hash = hash * 31 + (unsigned char) *name;

        hash %= f->name_table.size;
        oldhash = hash % (f->name_table.size / 2);
        if (oldhash >= f->name_table.split)
                return oldhash;
        else
                return hash;
}

static void unref_node(struct fuse *f, struct node *node);

static void remerge_name(struct fuse *f)
{
        struct node_table *t = &f->name_table;
        int iter;

        if (t->split == 0)
                node_table_reduce(t);

        for (iter = 8; t->split > 0 && iter; iter--) {
                struct node **upper;

                t->split--;
                upper = &t->array[t->split + t->size / 2];
                if (*upper) {
                        struct node **nodep;

                        for (nodep = &t->array[t->split]; *nodep;
                             nodep = &(*nodep)->name_next);

                        *nodep = *upper;
                        *upper = NULL;
                        break;
                }
        }
}

static void unhash_name(struct fuse *f, struct node *node)
{
        if (node->name) {
                size_t hash = name_hash(f, node->parent->nodeid, node->name);
                struct node **nodep = &f->name_table.array[hash];

                for (; *nodep != NULL; nodep = &(*nodep)->name_next)
                        if (*nodep == node) {
                                *nodep = node->name_next;
                                node->name_next = NULL;
                                unref_node(f, node->parent);
                                if (node->name != node->inline_name)
                                        free(node->name);
                                node->name = NULL;
                                node->parent = NULL;
                                f->name_table.use--;

                                if (f->name_table.use < f->name_table.size / 4)
                                        remerge_name(f);
                                return;
                        }
                fprintf(stderr,
                        "fuse internal error: unable to unhash node: %llu\n",
                        (unsigned long long) node->nodeid);
                abort();
        }
}

static void rehash_name(struct fuse *f)
{
        struct node_table *t = &f->name_table;
        struct node **nodep;
        struct node **next;
        size_t hash;

        if (t->split == t->size / 2)
                return;

        hash = t->split;
        t->split++;
        for (nodep = &t->array[hash]; *nodep != NULL; nodep = next) {
                struct node *node = *nodep;
                size_t newhash = name_hash(f, node->parent->nodeid, node->name);

                if (newhash != hash) {
                        next = nodep;
                            if (node->name != node-&gnoenline_name)