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/cred.h>
  17 #include <linux/eventpoll.h>
  18 #include <linux/rcupdate.h>
  19 #include <linux/mount.h>
  20 #include <linux/capability.h>
  21 #include <linux/cdev.h>
  22 #include <linux/fsnotify.h>
  23 #include <linux/sysctl.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 #include <linux/swap.h>
  30
  31 #include <linux/atomic.h>
  32
  33 #include "internal.h"
  34
  35 /* sysctl tunables... */
  36 struct files_stat_struct files_stat = {
  37         .max_files = NR_FILE
  38 };
  39
  40 /* SLAB cache for file structures */
  41 static struct kmem_cache *filp_cachep __read_mostly;
  42
  43 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
  44
  45 static void file_free_rcu(struct rcu_head *head)
  46 {
  47         struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
  48
  49         put_cred(f->f_cred);
  50         kmem_cache_free(filp_cachep, f);
  51 }
  52
  53 static inline void file_free(struct file *f)
  54 {
  55         percpu_counter_dec(&nr_files);
  56         call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
  57 }
  58
  59 /*
  60  * Return the total number of open files in the system
  61  */
  62 static long get_nr_files(void)
  63 {
  64         return percpu_counter_read_positive(&nr_files);
  65 }
  66
  67 /*
  68  * Return the maximum number of open files in the system
  69  */
  70 unsigned long get_max_files(void)
  71 {
  72         return files_stat.max_files;
  73 }
  74 EXPORT_SYMBOL_GPL(get_max_files);
  75
  76 /*
  77  * Handle nr_files sysctl
  78  */
  79 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
  80 int proc_nr_files(struct ctl_table *table, int write,
  81                      void __user *buffer, size_t *lenp, loff_t *ppos)
  82 {
  83         files_stat.nr_files = get_nr_files();
  84         return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
  85 }
  86 #else
  87 int proc_nr_files(struct ctl_table *table, int write,
  88                      void __user *buffer, size_t *lenp, loff_t *ppos)
  89 {
  90         return -ENOSYS;
  91 }
  92 #endif
  93
  94 /* Find an unused file structure and return a pointer to it.
  95  * Returns an error pointer if some error happend e.g. we over file
  96  * structures limit, run out of memory or operation is not permitted.
  97  *
  98  * Be very careful using this.  You are responsible for
  99  * getting write access to any mount that you might assign
 100  * to this filp, if it is opened for write.  If this is not
 101  * done, you will imbalance int the mount's writer count
 102  * and a warning at __fput() time.
 103  */
 104 struct file *get_empty_filp(void)
 105 {
 106         const struct cred *cred = current_cred();
 107         static long old_max;
 108         struct file *f;
 109         int error;
 110
 111         /*
 112          * Privileged users can go above max_files
 113          */
 114         if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
 115                 /*
 116                  * percpu_counters are inaccurate.  Do an expensive check before
 117                  * we go and fail.
 118                  */
 119                 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
 120                         goto over;
 121         }
 122
 123         f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
 124         if (unlikely(!f))
 125                 return ERR_PTR(-ENOMEM);
 126
 127         percpu_counter_inc(&nr_files);
 128         f->f_cred = get_cred(cred);
 129         error = security_file_alloc(f);
 130         if (unlikely(error)) {
 131                 file_free(f);
 132                 return ERR_PTR(error);
 133         }
 134
 135         atomic_long_set(&f->f_count, 1);
 136         rwlock_init(&f->f_owner.lock);
 137         spin_lock_init(&f->f_lock);
 138         mutex_init(&f->f_pos_lock);
 139         eventpoll_init_file(f);
 140         /* f->f_version: 0 */
 141         return f;
 142
 143 over:
 144         /* Ran out of filps - report that */
 145         if (get_nr_files() > old_max) {
 146                 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
 147                 old_max = get_nr_files();
 148         }
 149         return ERR_PTR(-ENFILE);
 150 }
 151 EXPORT_SYMBOL_GPL(get_empty_filp);
 152
 153 /**
 154  * alloc_file - allocate and initialize a 'struct file'
 155  *
 156  * @path: the (dentry, vfsmount) pair for the new file
 157  * @mode: the mode with which the new file will be opened
 158  * @fop: the 'struct file_operations' for the new file
 159  */
 160 struct file *alloc_file(const struct path *path, fmode_t mode,
 161                 const struct file_operations *fop)
 162 {
 163         struct file *file;
 164
 165         file = get_empty_filp();
 166         if (IS_ERR(file))
 167                 return file;
 168
 169         file->f_path = *path;
 170         file->f_inode = path->dentry->d_inode;
 171         file->f_mapping = path->dentry->d_inode->i_mapping;
 172         file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
 173         if ((mode & FMODE_READ) &&
 174              likely(fop->read || fop->read_iter))
 175                 mode |= FMODE_CAN_READ;
 176         if ((mode & FMODE_WRITE) &&
 177              likely(fop->write || fop->write_iter))
 178                 mode |= FMODE_CAN_WRITE;
 179         file->f_mode = mode;
 180         file->f_op = fop;
 181         if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 182                 i_readcount_inc(path->dentry->d_inode);
 183         return file;
 184 }
 185 EXPORT_SYMBOL(alloc_file);
 186
 187 /* the real guts of fput() - releasing the last reference to file
 188  */
 189 static void __fput(struct file *file)
 190 {
 191         struct dentry *dentry = file->f_path.dentry;
 192         struct vfsmount *mnt = file->f_path.mnt;
 193         struct inode *inode = file->f_inode;
 194
 195         might_sleep();
 196
 197         fsnotify_close(file);
 198         /*
 199          * The function eventpoll_release() should be the first called
 200          * in the file cleanup chain.
 201          */
 202         eventpoll_release(file);
 203         locks_remove_file(file);
 204
 205         if (unlikely(file->f_flags & FASYNC)) {
 206                 if (file->f_op->fasync)
 207                         file->f_op->fasync(-1, file, 0);
 208         }
 209         ima_file_free(file);
 210         if (file->f_op->release)
 211                 file->f_op->release(inode, file);
 212         security_file_free(file);
 213         if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
 214                      !(file->f_mode & FMODE_PATH))) {
 215                 cdev_put(inode->i_cdev);
 216         }
 217         fops_put(file->f_op);
 218         put_pid(file->f_owner.pid);
 219         if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 220                 i_readcount_dec(inode);
 221         if (file->f_mode & FMODE_WRITER) {
 222                 put_write_access(inode);
 223                 __mnt_drop_write(mnt);
 224         }
 225         file->f_path.dentry = NULL;
 226         file->f_path.mnt = NULL;
 227         file->f_inode = NULL;
 228         file_free(file);
 229         dput(dentry);
 230         mntput(mnt);
 231 }
 232
 233 static LLIST_HEAD(delayed_fput_list);
 234 static void delayed_fput(struct work_struct *unused)
 235 {
 236         struct llist_node *node = llist_del_all(&delayed_fput_list);
 237         struct llist_node *next;
 238
 239         for (; node; node = next) {
 240                 next = llist_next(node);
 241                 __fput(llist_entry(node, struct file, f_u.fu_llist));
 242         }
 243 }
 244
 245 static void ____fput(struct callback_head *work)
 246 {
 247         __fput(container_of(work, struct file, f_u.fu_rcuhead));
 248 }
 249
 250 /*
 251  * If kernel thread really needs to have the final fput() it has done
 252  * to complete, call this.  The only user right now is the boot - we
 253  * *do* need to make sure our writes to binaries on initramfs has
 254  * not left us with opened struct file waiting for __fput() - execve()
 255  * won't work without that.  Please, don't add more callers without
 256  * very good reasons; in particular, never call that with locks
 257  * held and never call that from a thread that might need to do
 258  * some work on any kind of umount.
 259  */
 260 void flush_delayed_fput(void)
 261 {
 262         delayed_fput(NULL);
 263 }
 264 EXPORT_SYMBOL_GPL(flush_delayed_fput);
 265
 266 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
 267
 268 void fput(struct file *file)
 269 {
 270         if (atomic_long_dec_and_test(&file->f_count)) {
 271                 struct task_struct *task = current;
 272
 273                 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
 274                         init_task_work(&file->f_u.fu_rcuhead, ____fput);
 275                         if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
 276                                 return;
 277                         /*
 278                          * After this task has run exit_task_work(),
 279                          * task_work_add() will fail.  Fall through to delayed
 280                          * fput to avoid leaking *file.
 281                          */
 282                 }
 283
 284                 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
 285                         schedule_delayed_work(&delayed_fput_work, 1);
 286         }
 287 }
 288
 289 /*
 290  * synchronous analog of fput(); for kernel threads that might be needed
 291  * in some umount() (and thus can't use flush_delayed_fput() without
 292  * risking deadlocks), need to wait for completion of __fput() and know
 293  * for this specific struct file it won't involve anything that would
 294  * need them.  Use only if you really need it - at the very least,
 295  * don't blindly convert fput() by kernel thread to that.
 296  */
 297 void __fput_sync(struct file *file)
 298 {
 299         if (atomic_long_dec_and_test(&file->f_count)) {
 300                 struct task_struct *task = current;
 301                 BUG_ON(!(task->flags & PF_KTHREAD));
 302                 __fput(file);
 303         }
 304 }
 305
 306 EXPORT_SYMBOL(fput);
 307 EXPORT_SYMBOL_GPL(__fput_sync);
 308
 309 void put_filp(struct file *file)
 310 {
 311         if (atomic_long_dec_and_test(&file->f_count)) {
 312                 security_file_free(file);
 313                 file_free(file);
 314         }
 315 }
 316 EXPORT_SYMBOL_GPL(put_filp);
 317
 318 void __init files_init(void)
 319 {
 320         filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
 321                         SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
 322         percpu_counter_init(&nr_files, 0, GFP_KERNEL);
 323 }
 324
 325 /*
 326  * One file with associated inode and dcache is very roughly 1K. Per default
 327  * do not use more than 10% of our memory for files.
 328  */
 329 void __init files_maxfiles_init(void)
 330 {
 331         unsigned long n;
 332         unsigned long memreserve = (totalram_pages - nr_free_pages()) * 3/2;
 333
 334         memreserve = min(memreserve, totalram_pages - 1);
 335         n = ((totalram_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
 336
 337         files_stat.max_files = max_t(unsigned long, n, NR_FILE);
 338 }