fs/file_table.c

   1 /*
   2  *  linux/fs/file_table.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *  Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
   6  */
   7
   8 #include <linux/string.h>
   9 #include <linux/slab.h>
  10 #include <linux/file.h>
  11 #include <linux/fdtable.h>
  12 #include <linux/init.h>
  13 #include <linux/module.h>
  14 #include <linux/fs.h>
  15 #include <linux/security.h>
  16 #include <linux/eventpoll.h>
  17 #include <linux/rcupdate.h>
  18 #include <linux/mount.h>
  19 #include <linux/capability.h>
  20 #include <linux/cdev.h>
  21 #include <linux/fsnotify.h>
  22 #include <linux/sysctl.h>
  23 #include <linux/lglock.h>
  24 #include <linux/percpu_counter.h>
  25 #include <linux/percpu.h>
  26 #include <linux/hardirq.h>
  27 #include <linux/task_work.h>
  28 #include <linux/ima.h>
  29
  30 #include <linux/atomic.h>
  31
  32 #include "internal.h"
  33
  34 /* sysctl tunables... */
  35 struct files_stat_struct files_stat = {
  36         .max_files = NR_FILE
  37 };
  38
  39 /* SLAB cache for file structures */
  40 static struct kmem_cache *filp_cachep __read_mostly;
  41
  42 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
  43
  44 static void file_free_rcu(struct rcu_head *head)
  45 {
  46         struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
  47
  48         put_cred(f->f_cred);
  49         kmem_cache_free(filp_cachep, f);
  50 }
  51
  52 static inline void file_free(struct file *f)
  53 {
  54         percpu_counter_dec(&nr_files);
  55         call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
  56 }
  57
  58 /*
  59  * Return the total number of open files in the system
  60  */
  61 static long get_nr_files(void)
  62 {
  63         return percpu_counter_read_positive(&nr_files);
  64 }
  65
  66 /*
  67  * Return the maximum number of open files in the system
  68  */
  69 unsigned long get_max_files(void)
  70 {
  71         return files_stat.max_files;
  72 }
  73 EXPORT_SYMBOL_GPL(get_max_files);
  74
  75 /*
  76  * Handle nr_files sysctl
  77  */
  78 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
  79 int proc_nr_files(struct ctl_table *table, int write,
  80                      void __user *buffer, size_t *lenp, loff_t *ppos)
  81 {
  82         files_stat.nr_files = get_nr_files();
  83         return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
  84 }
  85 #else
  86 int proc_nr_files(struct ctl_table *table, int write,
  87                      void __user *buffer, size_t *lenp, loff_t *ppos)
  88 {
  89         return -ENOSYS;
  90 }
  91 #endif
  92
  93 /* Find an unused file structure and return a pointer to it.
  94  * Returns an error pointer if some error happend e.g. we over file
  95  * structures limit, run out of memory or operation is not permitted.
  96  *
  97  * Be very careful using this.  You are responsible for
  98  * getting write access to any mount that you might assign
  99  * to this filp, if it is opened for write.  If this is not
 100  * done, you will imbalance int the mount's writer count
 101  * and a warning at __fput() time.
 102  */
 103 struct file *get_empty_filp(void)
 104 {
 105         const struct cred *cred = current_cred();
 106         static long old_max;
 107         struct file *f;
 108         int error;
 109
 110         /*
 111          * Privileged users can go above max_files
 112          */
 113         if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
 114                 /*
 115                  * percpu_counters are inaccurate.  Do an expensive check before
 116                  * we go and fail.
 117                  */
 118                 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
 119                         goto over;
 120         }
 121
 122         f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
 123         if (unlikely(!f))
 124                 return ERR_PTR(-ENOMEM);
 125
 126         percpu_counter_inc(&nr_files);
 127         f->f_cred = get_cred(cred);
 128         error = security_file_alloc(f);
 129         if (unlikely(error)) {
 130                 file_free(f);
 131                 return ERR_PTR(error);
 132         }
 133
 134         atomic_long_set(&f->f_count, 1);
 135         rwlock_init(&f->f_owner.lock);
 136         spin_lock_init(&f->f_lock);
 137         mutex_init(&f->f_pos_lock);
 138         eventpoll_init_file(f);
 139         /* f->f_version: 0 */
 140         return f;
 141
 142 over:
 143         /* Ran out of filps - report that */
 144         if (get_nr_files() > old_max) {
 145                 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
 146                 old_max = get_nr_files();
 147         }
 148         return ERR_PTR(-ENFILE);
 149 }
 150
 151 /**
 152  * alloc_file - allocate and initialize a 'struct file'
 153  * @mnt: the vfsmount on which the file will reside
 154  * @dentry: the dentry representing the new file
 155  * @mode: the mode with which the new file will be opened
 156  * @fop: the 'struct file_operations' for the new file
 157  *
 158  * Use this instead of get_empty_filp() to get a new
 159  * 'struct file'.  Do so because of the same initialization
 160  * pitfalls reasons listed for init_file().  This is a
 161  * preferred interface to using init_file().
 162  *
 163  * If all the callers of init_file() are eliminated, its
 164  * code should be moved into this function.
 165  */
 166 struct file *alloc_file(struct path *path, fmode_t mode,
 167                 const struct file_operations *fop)
 168 {
 169         struct file *file;
 170
 171         file = get_empty_filp();
 172         if (IS_ERR(file))
 173                 return file;
 174
 175         file->f_path = *path;
 176         file->f_inode = path->dentry->d_inode;
 177         file->f_mapping = path->dentry->d_inode->i_mapping;
 178         if ((mode & FMODE_READ) &&
 179              likely(fop->read || fop->aio_read || fop->read_iter))
 180                 mode |= FMODE_CAN_READ;
 181         if ((mode & FMODE_WRITE) &&
 182              likely(fop->write || fop->aio_write || fop->write_iter))
 183                 mode |= FMODE_CAN_WRITE;
 184         file->f_mode = mode;
 185         file->f_op = fop;
 186         if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 187                 i_readcount_inc(path->dentry->d_inode);
 188         return file;
 189 }
 190 EXPORT_SYMBOL(alloc_file);
 191
 192 /* the real guts of fput() - releasing the last reference to file
 193  */
 194 static void __fput(struct file *file)
 195 {
 196         struct dentry *dentry = file->f_path.dentry;
 197         struct vfsmount *mnt = file->f_path.mnt;
 198         struct inode *inode = file->f_inode;
 199
 200         might_sleep();
 201
 202         fsnotify_close(file);
 203         /*
 204          * The function eventpoll_release() should be the first called
 205          * in the file cleanup chain.
 206          */
 207         eventpoll_release(file);
 208         locks_remove_file(file);
 209
 210         if (unlikely(file->f_flags & FASYNC)) {
 211                 if (file->f_op->fasync)
 212                         file->f_op->fasync(-1, file, 0);
 213         }
 214         ima_file_free(file);
 215         if (file->f_op->release)
 216                 file->f_op->release(inode, file);
 217         security_file_free(file);
 218         if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
 219                      !(file->f_mode & FMODE_PATH))) {
 220                 cdev_put(inode->i_cdev);
 221         }
 222         fops_put(file->f_op);
 223         put_pid(file->f_owner.pid);
 224         if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 225                 i_readcount_dec(inode);
 226         if (file->f_mode & FMODE_WRITER) {
 227                 put_write_access(inode);
 228                 __mnt_drop_write(mnt);
 229         }
 230         file->f_path.dentry = NULL;
 231         file->f_path.mnt = NULL;
 232         file->f_inode = NULL;
 233         file_free(file);
 234         dput(dentry);
 235         mntput(mnt);
 236 }
 237
 238 static LLIST_HEAD(delayed_fput_list);
 239 static void delayed_fput(struct work_struct *unused)
 240 {
 241         struct llist_node *node = llist_del_all(&delayed_fput_list);
 242         struct llist_node *next;
 243
 244         for (; node; node = next) {
 245                 next = llist_next(node);
 246                 __fput(llist_entry(node, struct file, f_u.fu_llist));
 247         }
 248 }
 249
 250 static void ____fput(struct callback_head *work)
 251 {
 252         __fput(container_of(work, struct file, f_u.fu_rcuhead));
 253 }
 254
 255 /*
 256  * If kernel thread really needs to have the final fput() it has done
 257  * to complete, call this.  The only user right now is the boot - we
 258  * *do* need to make sure our writes to binaries on initramfs has
 259  * not left us with opened struct file waiting for __fput() - execve()
 260  * won't work without that.  Please, don't add more callers without
 261  * very good reasons; in particular, never call that with locks
 262  * held and never call that from a thread that might need to do
 263  * some work on any kind of umount.
 264  */
 265 void flush_delayed_fput(void)
 266 {
 267         delayed_fput(NULL);
 268 }
 269
 270 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
 271
 272 void fput(struct file *file)
 273 {
 274         if (atomic_long_dec_and_test(&file->f_count)) {
 275                 struct task_struct *task = current;
 276
 277                 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
 278                         init_task_work(&file->f_u.fu_rcuhead, ____fput);
 279                         if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
 280                                 return;
 281                         /*
 282                          * After this task has run exit_task_work(),
 283                          * task_work_add() will fail.  Fall through to delayed
 284                          * fput to avoid leaking *file.
 285                          */
 286                 }
 287
 288                 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
 289                         schedule_delayed_work(&delayed_fput_work, 1);
 290         }
 291 }
 292
 293 /*
 294  * synchronous analog of fput(); for kernel threads that might be needed
 295  * in some umount() (and thus can't use flush_delayed_fput() without
 296  * risking deadlocks), need to wait for completion of __fput() and know
 297  * for this specific struct file it won't involve anything that would
 298  * need them.  Use only if you really need it - at the very least,
 299  * don't blindly convert fput() by kernel thread to that.
 300  */
 301 void __fput_sync(struct file *file)
 302 {
 303         if (atomic_long_dec_and_test(&file->f_count)) {
 304                 struct task_struct *task = current;
 305                 BUG_ON(!(task->flags & PF_KTHREAD));
 306                 __fput(file);
 307         }
 308 }
 309
 310 EXPORT_SYMBOL(fput);
 311
 312 void put_filp(struct file *file)
 313 {
 314         if (atomic_long_dec_and_test(&file->f_count)) {
 315                 security_file_free(file);
 316                 file_free(file);
 317         }
 318 }
 319
 320 void __init files_init(unsigned long mempages)
 321 {
 322         unsigned long n;
 323
 324         filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
 325                         SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
 326
 327         /*
 328          * One file with associated inode and dcache is very roughly 1K.
 329          * Per default don't use more than 10% of our memory for files.
 330          */
 331
 332         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 333         files_stat.max_files = max_t(unsigned long, n, NR_FILE);
 334         percpu_counter_init(&nr_files, 0);
 335 }