* Manage the dynamic fd arrays in the process files_struct.
*/
+#include <linux/module.h>
#include <linux/fs.h>
#include <linux/mm.h>
#include <linux/time.h>
#include <linux/slab.h>
#include <linux/vmalloc.h>
#include <linux/file.h>
+#include <linux/fdtable.h>
#include <linux/bitops.h>
+#include <linux/interrupt.h>
+#include <linux/spinlock.h>
+#include <linux/rcupdate.h>
+#include <linux/workqueue.h>
+struct fdtable_defer {
+ spinlock_t lock;
+ struct work_struct wq;
+ struct fdtable *next;
+};
+
+int sysctl_nr_open __read_mostly = 1024*1024;
+int sysctl_nr_open_min = BITS_PER_LONG;
+int sysctl_nr_open_max = 1024 * 1024; /* raised later */
/*
- * Allocate an fd array, using kmalloc or vmalloc.
- * Note: the array isn't cleared at allocation time.
+ * We use this list to defer free fdtables that have vmalloced
+ * sets/arrays. By keeping a per-cpu list, we avoid having to embed
+ * the work_struct in fdtable itself which avoids a 64 byte (i386) increase in
+ * this per-task structure.
*/
-struct file ** alloc_fd_array(int num)
-{
- struct file **new_fds;
- int size = num * sizeof(struct file *);
+static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list);
+static inline void * alloc_fdmem(unsigned int size)
+{
if (size <= PAGE_SIZE)
- new_fds = (struct file **) kmalloc(size, GFP_KERNEL);
- else
- new_fds = (struct file **) vmalloc(size);
- return new_fds;
+ return kmalloc(size, GFP_KERNEL);
+ else
+ return vmalloc(size);
}
-void free_fd_array(struct file **array, int num)
+static inline void free_fdarr(struct fdtable *fdt)
{
- int size = num * sizeof(struct file *);
+ if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *)))
+ kfree(fdt->fd);
+ else
+ vfree(fdt->fd);
+}
- if (!array) {
- printk (KERN_ERR "free_fd_array: array = 0 (num = %d)\n", num);
- return;
+static inline void free_fdset(struct fdtable *fdt)
+{
+ if (fdt->max_fds <= (PAGE_SIZE * BITS_PER_BYTE / 2))
+ kfree(fdt->open_fds);
+ else
+ vfree(fdt->open_fds);
+}
+
+static void free_fdtable_work(struct work_struct *work)
+{
+ struct fdtable_defer *f =
+ container_of(work, struct fdtable_defer, wq);
+ struct fdtable *fdt;
+
+ spin_lock_bh(&f->lock);
+ fdt = f->next;
+ f->next = NULL;
+ spin_unlock_bh(&f->lock);
+ while(fdt) {
+ struct fdtable *next = fdt->next;
+ vfree(fdt->fd);
+ free_fdset(fdt);
+ kfree(fdt);
+ fdt = next;
}
+}
+
+void free_fdtable_rcu(struct rcu_head *rcu)
+{
+ struct fdtable *fdt = container_of(rcu, struct fdtable, rcu);
+ struct fdtable_defer *fddef;
- if (num <= NR_OPEN_DEFAULT) /* Don't free the embedded fd array! */
+ BUG_ON(!fdt);
+
+ if (fdt->max_fds <= NR_OPEN_DEFAULT) {
+ /*
+ * This fdtable is embedded in the files structure and that
+ * structure itself is getting destroyed.
+ */
+ kmem_cache_free(files_cachep,
+ container_of(fdt, struct files_struct, fdtab));
return;
- else if (size <= PAGE_SIZE)
- kfree(array);
- else
- vfree(array);
+ }
+ if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *))) {
+ kfree(fdt->fd);
+ kfree(fdt->open_fds);
+ kfree(fdt);
+ } else {
+ fddef = &get_cpu_var(fdtable_defer_list);
+ spin_lock(&fddef->lock);
+ fdt->next = fddef->next;
+ fddef->next = fdt;
+ /* vmallocs are handled from the workqueue context */
+ schedule_work(&fddef->wq);
+ spin_unlock(&fddef->lock);
+ put_cpu_var(fdtable_defer_list);
+ }
}
/*
- * Expand the fd array in the files_struct. Called with the files
- * spinlock held for write.
+ * Expand the fdset in the files_struct. Called with the files spinlock
+ * held for write.
*/
+static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
+{
+ unsigned int cpy, set;
-static int expand_fd_array(struct files_struct *files, int nr)
- __releases(files->file_lock)
- __acquires(files->file_lock)
+ BUG_ON(nfdt->max_fds < ofdt->max_fds);
+
+ cpy = ofdt->max_fds * sizeof(struct file *);
+ set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
+ memcpy(nfdt->fd, ofdt->fd, cpy);
+ memset((char *)(nfdt->fd) + cpy, 0, set);
+
+ cpy = ofdt->max_fds / BITS_PER_BYTE;
+ set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE;
+ memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
+ memset((char *)(nfdt->open_fds) + cpy, 0, set);
+ memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
+ memset((char *)(nfdt->close_on_exec) + cpy, 0, set);
+}
+
+static struct fdtable * alloc_fdtable(unsigned int nr)
{
- struct file **new_fds;
- int error, nfds;
struct fdtable *fdt;
+ char *data;
-
- error = -EMFILE;
- fdt = files_fdtable(files);
- if (fdt->max_fds >= NR_OPEN || nr >= NR_OPEN)
+ /*
+ * Figure out how many fds we actually want to support in this fdtable.
+ * Allocation steps are keyed to the size of the fdarray, since it
+ * grows far faster than any of the other dynamic data. We try to fit
+ * the fdarray into comfortable page-tuned chunks: starting at 1024B
+ * and growing in powers of two from there on.
+ */
+ nr /= (1024 / sizeof(struct file *));
+ nr = roundup_pow_of_two(nr + 1);
+ nr *= (1024 / sizeof(struct file *));
+ /*
+ * Note that this can drive nr *below* what we had passed if sysctl_nr_open
+ * had been set lower between the check in expand_files() and here. Deal
+ * with that in caller, it's cheaper that way.
+ *
+ * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
+ * bitmaps handling below becomes unpleasant, to put it mildly...
+ */
+ if (unlikely(nr > sysctl_nr_open))
+ nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
+
+ fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
+ if (!fdt)
goto out;
+ fdt->max_fds = nr;
+ data = alloc_fdmem(nr * sizeof(struct file *));
+ if (!data)
+ goto out_fdt;
+ fdt->fd = (struct file **)data;
+ data = alloc_fdmem(max_t(unsigned int,
+ 2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES));
+ if (!data)
+ goto out_arr;
+ fdt->open_fds = (fd_set *)data;
+ data += nr / BITS_PER_BYTE;
+ fdt->close_on_exec = (fd_set *)data;
+ INIT_RCU_HEAD(&fdt->rcu);
+ fdt->next = NULL;
- nfds = fdt->max_fds;
- spin_unlock(&files->file_lock);
+ return fdt;
- /*
- * Expand to the max in easy steps, and keep expanding it until
- * we have enough for the requested fd array size.
- */
+out_arr:
+ free_fdarr(fdt);
+out_fdt:
+ kfree(fdt);
+out:
+ return NULL;
+}
- do {
-#if NR_OPEN_DEFAULT < 256
- if (nfds < 256)
- nfds = 256;
- else
-#endif
- if (nfds < (PAGE_SIZE / sizeof(struct file *)))
- nfds = PAGE_SIZE / sizeof(struct file *);
- else {
- nfds = nfds * 2;
- if (nfds > NR_OPEN)
- nfds = NR_OPEN;
- }
- } while (nfds <= nr);
+/*
+ * Expand the file descriptor table.
+ * This function will allocate a new fdtable and both fd array and fdset, of
+ * the given size.
+ * Return <0 error code on error; 1 on successful completion.
+ * The files->file_lock should be held on entry, and will be held on exit.
+ */
+static int expand_fdtable(struct files_struct *files, int nr)
+ __releases(files->file_lock)
+ __acquires(files->file_lock)
+{
+ struct fdtable *new_fdt, *cur_fdt;
- error = -ENOMEM;
- new_fds = alloc_fd_array(nfds);
+ spin_unlock(&files->file_lock);
+ new_fdt = alloc_fdtable(nr);
spin_lock(&files->file_lock);
- if (!new_fds)
- goto out;
-
- /* Copy the existing array and install the new pointer */
- fdt = files_fdtable(files);
-
- if (nfds > fdt->max_fds) {
- struct file **old_fds;
- int i;
-
- old_fds = xchg(&fdt->fd, new_fds);
- i = xchg(&fdt->max_fds, nfds);
-
- /* Don't copy/clear the array if we are creating a new
- fd array for fork() */
- if (i) {
- memcpy(new_fds, old_fds, i * sizeof(struct file *));
- /* clear the remainder of the array */
- memset(&new_fds[i], 0,
- (nfds-i) * sizeof(struct file *));
-
- spin_unlock(&files->file_lock);
- free_fd_array(old_fds, i);
- spin_lock(&files->file_lock);
- }
+ if (!new_fdt)
+ return -ENOMEM;
+ /*
+ * extremely unlikely race - sysctl_nr_open decreased between the check in
+ * caller and alloc_fdtable(). Cheaper to catch it here...
+ */
+ if (unlikely(new_fdt->max_fds <= nr)) {
+ free_fdarr(new_fdt);
+ free_fdset(new_fdt);
+ kfree(new_fdt);
+ return -EMFILE;
+ }
+ /*
+ * Check again since another task may have expanded the fd table while
+ * we dropped the lock
+ */
+ cur_fdt = files_fdtable(files);
+ if (nr >= cur_fdt->max_fds) {
+ /* Continue as planned */
+ copy_fdtable(new_fdt, cur_fdt);
+ rcu_assign_pointer(files->fdt, new_fdt);
+ if (cur_fdt->max_fds > NR_OPEN_DEFAULT)
+ free_fdtable(cur_fdt);
} else {
- /* Somebody expanded the array while we slept ... */
- spin_unlock(&files->file_lock);
- free_fd_array(new_fds, nfds);
- spin_lock(&files->file_lock);
+ /* Somebody else expanded, so undo our attempt */
+ free_fdarr(new_fdt);
+ free_fdset(new_fdt);
+ kfree(new_fdt);
}
- error = 0;
-out:
- return error;
+ return 1;
}
/*
- * Allocate an fdset array, using kmalloc or vmalloc.
- * Note: the array isn't cleared at allocation time.
+ * Expand files.
+ * This function will expand the file structures, if the requested size exceeds
+ * the current capacity and there is room for expansion.
+ * Return <0 error code on error; 0 when nothing done; 1 when files were
+ * expanded and execution may have blocked.
+ * The files->file_lock should be held on entry, and will be held on exit.
*/
-fd_set * alloc_fdset(int num)
+int expand_files(struct files_struct *files, int nr)
{
- fd_set *new_fdset;
- int size = num / 8;
+ struct fdtable *fdt;
- if (size <= PAGE_SIZE)
- new_fdset = (fd_set *) kmalloc(size, GFP_KERNEL);
- else
- new_fdset = (fd_set *) vmalloc(size);
- return new_fdset;
+ fdt = files_fdtable(files);
+
+ /*
+ * N.B. For clone tasks sharing a files structure, this test
+ * will limit the total number of files that can be opened.
+ */
+ if (nr >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
+ return -EMFILE;
+
+ /* Do we need to expand? */
+ if (nr < fdt->max_fds)
+ return 0;
+
+ /* Can we expand? */
+ if (nr >= sysctl_nr_open)
+ return -EMFILE;
+
+ /* All good, so we try */
+ return expand_fdtable(files, nr);
}
-void free_fdset(fd_set *array, int num)
+static int count_open_files(struct fdtable *fdt)
{
- int size = num / 8;
+ int size = fdt->max_fds;
+ int i;
- if (num <= __FD_SETSIZE) /* Don't free an embedded fdset */
- return;
- else if (size <= PAGE_SIZE)
- kfree(array);
- else
- vfree(array);
+ /* Find the last open fd */
+ for (i = size/(8*sizeof(long)); i > 0; ) {
+ if (fdt->open_fds->fds_bits[--i])
+ break;
+ }
+ i = (i+1) * 8 * sizeof(long);
+ return i;
}
/*
- * Expand the fdset in the files_struct. Called with the files spinlock
- * held for write.
+ * Allocate a new files structure and copy contents from the
+ * passed in files structure.
+ * errorp will be valid only when the returned files_struct is NULL.
*/
-static int expand_fdset(struct files_struct *files, int nr)
- __releases(file->file_lock)
- __acquires(file->file_lock)
+struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
{
- fd_set *new_openset = NULL, *new_execset = NULL;
- int error, nfds = 0;
- struct fdtable *fdt;
+ struct files_struct *newf;
+ struct file **old_fds, **new_fds;
+ int open_files, size, i;
+ struct fdtable *old_fdt, *new_fdt;
- error = -EMFILE;
- fdt = files_fdtable(files);
- if (fdt->max_fdset >= NR_OPEN || nr >= NR_OPEN)
+ *errorp = -ENOMEM;
+ newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
+ if (!newf)
goto out;
- nfds = fdt->max_fdset;
- spin_unlock(&files->file_lock);
+ atomic_set(&newf->count, 1);
+
+ spin_lock_init(&newf->file_lock);
+ newf->next_fd = 0;
+ new_fdt = &newf->fdtab;
+ new_fdt->max_fds = NR_OPEN_DEFAULT;
+ new_fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
+ new_fdt->open_fds = (fd_set *)&newf->open_fds_init;
+ new_fdt->fd = &newf->fd_array[0];
+ INIT_RCU_HEAD(&new_fdt->rcu);
+ new_fdt->next = NULL;
+
+ spin_lock(&oldf->file_lock);
+ old_fdt = files_fdtable(oldf);
+ open_files = count_open_files(old_fdt);
+
+ /*
+ * Check whether we need to allocate a larger fd array and fd set.
+ */
+ while (unlikely(open_files > new_fdt->max_fds)) {
+ spin_unlock(&oldf->file_lock);
- /* Expand to the max in easy steps */
- do {
- if (nfds < (PAGE_SIZE * 8))
- nfds = PAGE_SIZE * 8;
- else {
- nfds = nfds * 2;
- if (nfds > NR_OPEN)
- nfds = NR_OPEN;
+ if (new_fdt != &newf->fdtab) {
+ free_fdarr(new_fdt);
+ free_fdset(new_fdt);
+ kfree(new_fdt);
}
- } while (nfds <= nr);
- error = -ENOMEM;
- new_openset = alloc_fdset(nfds);
- new_execset = alloc_fdset(nfds);
- spin_lock(&files->file_lock);
- if (!new_openset || !new_execset)
- goto out;
+ new_fdt = alloc_fdtable(open_files - 1);
+ if (!new_fdt) {
+ *errorp = -ENOMEM;
+ goto out_release;
+ }
- error = 0;
-
- /* Copy the existing tables and install the new pointers */
- fdt = files_fdtable(files);
- if (nfds > fdt->max_fdset) {
- int i = fdt->max_fdset / (sizeof(unsigned long) * 8);
- int count = (nfds - fdt->max_fdset) / 8;
-
- /*
- * Don't copy the entire array if the current fdset is
- * not yet initialised.
+ /* beyond sysctl_nr_open; nothing to do */
+ if (unlikely(new_fdt->max_fds < open_files)) {
+ free_fdarr(new_fdt);
+ free_fdset(new_fdt);
+ kfree(new_fdt);
+ *errorp = -EMFILE;
+ goto out_release;
+ }
+
+ /*
+ * Reacquire the oldf lock and a pointer to its fd table
+ * who knows it may have a new bigger fd table. We need
+ * the latest pointer.
*/
- if (i) {
- memcpy (new_openset, fdt->open_fds, fdt->max_fdset/8);
- memcpy (new_execset, fdt->close_on_exec, fdt->max_fdset/8);
- memset (&new_openset->fds_bits[i], 0, count);
- memset (&new_execset->fds_bits[i], 0, count);
+ spin_lock(&oldf->file_lock);
+ old_fdt = files_fdtable(oldf);
+ open_files = count_open_files(old_fdt);
+ }
+
+ old_fds = old_fdt->fd;
+ new_fds = new_fdt->fd;
+
+ memcpy(new_fdt->open_fds->fds_bits,
+ old_fdt->open_fds->fds_bits, open_files/8);
+ memcpy(new_fdt->close_on_exec->fds_bits,
+ old_fdt->close_on_exec->fds_bits, open_files/8);
+
+ for (i = open_files; i != 0; i--) {
+ struct file *f = *old_fds++;
+ if (f) {
+ get_file(f);
+ } else {
+ /*
+ * The fd may be claimed in the fd bitmap but not yet
+ * instantiated in the files array if a sibling thread
+ * is partway through open(). So make sure that this
+ * fd is available to the new process.
+ */
+ FD_CLR(open_files - i, new_fdt->open_fds);
}
-
- nfds = xchg(&fdt->max_fdset, nfds);
- new_openset = xchg(&fdt->open_fds, new_openset);
- new_execset = xchg(&fdt->close_on_exec, new_execset);
- spin_unlock(&files->file_lock);
- free_fdset (new_openset, nfds);
- free_fdset (new_execset, nfds);
- spin_lock(&files->file_lock);
- return 0;
- }
- /* Somebody expanded the array while we slept ... */
+ rcu_assign_pointer(*new_fds++, f);
+ }
+ spin_unlock(&oldf->file_lock);
+ /* compute the remainder to be cleared */
+ size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
+
+ /* This is long word aligned thus could use a optimized version */
+ memset(new_fds, 0, size);
+
+ if (new_fdt->max_fds > open_files) {
+ int left = (new_fdt->max_fds-open_files)/8;
+ int start = open_files / (8 * sizeof(unsigned long));
+
+ memset(&new_fdt->open_fds->fds_bits[start], 0, left);
+ memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
+ }
+
+ rcu_assign_pointer(newf->fdt, new_fdt);
+
+ return newf;
+
+out_release:
+ kmem_cache_free(files_cachep, newf);
out:
- spin_unlock(&files->file_lock);
- if (new_openset)
- free_fdset(new_openset, nfds);
- if (new_execset)
- free_fdset(new_execset, nfds);
- spin_lock(&files->file_lock);
- return error;
+ return NULL;
+}
+
+static void __devinit fdtable_defer_list_init(int cpu)
+{
+ struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu);
+ spin_lock_init(&fddef->lock);
+ INIT_WORK(&fddef->wq, free_fdtable_work);
+ fddef->next = NULL;
+}
+
+void __init files_defer_init(void)
+{
+ int i;
+ for_each_possible_cpu(i)
+ fdtable_defer_list_init(i);
+ sysctl_nr_open_max = min((size_t)INT_MAX, ~(size_t)0/sizeof(void *)) &
+ -BITS_PER_LONG;
}
+struct files_struct init_files = {
+ .count = ATOMIC_INIT(1),
+ .fdt = &init_files.fdtab,
+ .fdtab = {
+ .max_fds = NR_OPEN_DEFAULT,
+ .fd = &init_files.fd_array[0],
+ .close_on_exec = (fd_set *)&init_files.close_on_exec_init,
+ .open_fds = (fd_set *)&init_files.open_fds_init,
+ .rcu = RCU_HEAD_INIT,
+ },
+ .file_lock = __SPIN_LOCK_UNLOCKED(init_task.file_lock),
+};
+
/*
- * Expand files.
- * Return <0 on error; 0 nothing done; 1 files expanded, we may have blocked.
- * Should be called with the files->file_lock spinlock held for write.
+ * allocate a file descriptor, mark it busy.
*/
-int expand_files(struct files_struct *files, int nr)
+int alloc_fd(unsigned start, unsigned flags)
{
- int err, expand = 0;
+ struct files_struct *files = current->files;
+ unsigned int fd;
+ int error;
struct fdtable *fdt;
+ spin_lock(&files->file_lock);
+repeat:
fdt = files_fdtable(files);
- if (nr >= fdt->max_fdset) {
- expand = 1;
- if ((err = expand_fdset(files, nr)))
- goto out;
- }
- if (nr >= fdt->max_fds) {
- expand = 1;
- if ((err = expand_fd_array(files, nr)))
- goto out;
+ fd = start;
+ if (fd < files->next_fd)
+ fd = files->next_fd;
+
+ if (fd < fdt->max_fds)
+ fd = find_next_zero_bit(fdt->open_fds->fds_bits,
+ fdt->max_fds, fd);
+
+ error = expand_files(files, fd);
+ if (error < 0)
+ goto out;
+
+ /*
+ * If we needed to expand the fs array we
+ * might have blocked - try again.
+ */
+ if (error)
+ goto repeat;
+
+ if (start <= files->next_fd)
+ files->next_fd = fd + 1;
+
+ FD_SET(fd, fdt->open_fds);
+ if (flags & O_CLOEXEC)
+ FD_SET(fd, fdt->close_on_exec);
+ else
+ FD_CLR(fd, fdt->close_on_exec);
+ error = fd;
+#if 1
+ /* Sanity check */
+ if (rcu_dereference(fdt->fd[fd]) != NULL) {
+ printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
+ rcu_assign_pointer(fdt->fd[fd], NULL);
}
- err = expand;
+#endif
+
out:
- return err;
+ spin_unlock(&files->file_lock);
+ return error;
+}
+
+int get_unused_fd(void)
+{
+ return alloc_fd(0, 0);
}
+EXPORT_SYMBOL(get_unused_fd);