fs/backing-file.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Common helpers for stackable filesystems and backing files.
   4  *
   5  * Forked from fs/overlayfs/file.c.
   6  *
   7  * Copyright (C) 2017 Red Hat, Inc.
   8  * Copyright (C) 2023 CTERA Networks.
   9  */
  10
  11 #include <linux/fs.h>
  12 #include <linux/backing-file.h>
  13 #include <linux/splice.h>
  14
  15 #include "internal.h"
  16
  17 /**
  18  * backing_file_open - open a backing file for kernel internal use
  19  * @user_path:  path that the user reuqested to open
  20  * @flags:      open flags
  21  * @real_path:  path of the backing file
  22  * @cred:       credentials for open
  23  *
  24  * Open a backing file for a stackable filesystem (e.g., overlayfs).
  25  * @user_path may be on the stackable filesystem and @real_path on the
  26  * underlying filesystem.  In this case, we want to be able to return the
  27  * @user_path of the stackable filesystem. This is done by embedding the
  28  * returned file into a container structure that also stores the stacked
  29  * file's path, which can be retrieved using backing_file_user_path().
  30  */
  31 struct file *backing_file_open(const struct path *user_path, int flags,
  32                                const struct path *real_path,
  33                                const struct cred *cred)
  34 {
  35         struct file *f;
  36         int error;
  37
  38         f = alloc_empty_backing_file(flags, cred);
  39         if (IS_ERR(f))
  40                 return f;
  41
  42         path_get(user_path);
  43         *backing_file_user_path(f) = *user_path;
  44         error = vfs_open(real_path, f);
  45         if (error) {
  46                 fput(f);
  47                 f = ERR_PTR(error);
  48         }
  49
  50         return f;
  51 }
  52 EXPORT_SYMBOL_GPL(backing_file_open);
  53
  54 struct backing_aio {
  55         struct kiocb iocb;
  56         refcount_t ref;
  57         struct kiocb *orig_iocb;
  58         /* used for aio completion */
  59         void (*end_write)(struct file *);
  60         struct work_struct work;
  61         long res;
  62 };
  63
  64 static struct kmem_cache *backing_aio_cachep;
  65
  66 #define BACKING_IOCB_MASK \
  67         (IOCB_NOWAIT | IOCB_HIPRI | IOCB_DSYNC | IOCB_SYNC | IOCB_APPEND)
  68
  69 static rwf_t iocb_to_rw_flags(int flags)
  70 {
  71         return (__force rwf_t)(flags & BACKING_IOCB_MASK);
  72 }
  73
  74 static void backing_aio_put(struct backing_aio *aio)
  75 {
  76         if (refcount_dec_and_test(&aio->ref)) {
  77                 fput(aio->iocb.ki_filp);
  78                 kmem_cache_free(backing_aio_cachep, aio);
  79         }
  80 }
  81
  82 static void backing_aio_cleanup(struct backing_aio *aio, long res)
  83 {
  84         struct kiocb *iocb = &aio->iocb;
  85         struct kiocb *orig_iocb = aio->orig_iocb;
  86
  87         if (aio->end_write)
  88                 aio->end_write(orig_iocb->ki_filp);
  89
  90         orig_iocb->ki_pos = iocb->ki_pos;
  91         backing_aio_put(aio);
  92 }
  93
  94 static void backing_aio_rw_complete(struct kiocb *iocb, long res)
  95 {
  96         struct backing_aio *aio = container_of(iocb, struct backing_aio, iocb);
  97         struct kiocb *orig_iocb = aio->orig_iocb;
  98
  99         if (iocb->ki_flags & IOCB_WRITE)
 100                 kiocb_end_write(iocb);
 101
 102         backing_aio_cleanup(aio, res);
 103         orig_iocb->ki_complete(orig_iocb, res);
 104 }
 105
 106 static void backing_aio_complete_work(struct work_struct *work)
 107 {
 108         struct backing_aio *aio = container_of(work, struct backing_aio, work);
 109
 110         backing_aio_rw_complete(&aio->iocb, aio->res);
 111 }
 112
 113 static void backing_aio_queue_completion(struct kiocb *iocb, long res)
 114 {
 115         struct backing_aio *aio = container_of(iocb, struct backing_aio, iocb);
 116
 117         /*
 118          * Punt to a work queue to serialize updates of mtime/size.
 119          */
 120         aio->res = res;
 121         INIT_WORK(&aio->work, backing_aio_complete_work);
 122         queue_work(file_inode(aio->orig_iocb->ki_filp)->i_sb->s_dio_done_wq,
 123                    &aio->work);
 124 }
 125
 126 static int backing_aio_init_wq(struct kiocb *iocb)
 127 {
 128         struct super_block *sb = file_inode(iocb->ki_filp)->i_sb;
 129
 130         if (sb->s_dio_done_wq)
 131                 return 0;
 132
 133         return sb_init_dio_done_wq(sb);
 134 }
 135
 136
 137 ssize_t backing_file_read_iter(struct file *file, struct iov_iter *iter,
 138                                struct kiocb *iocb, int flags,
 139                                struct backing_file_ctx *ctx)
 140 {
 141         struct backing_aio *aio = NULL;
 142         const struct cred *old_cred;
 143         ssize_t ret;
 144
 145         if (WARN_ON_ONCE(!(file->f_mode & FMODE_BACKING)))
 146                 return -EIO;
 147
 148         if (!iov_iter_count(iter))
 149                 return 0;
 150
 151         if (iocb->ki_flags & IOCB_DIRECT &&
 152             !(file->f_mode & FMODE_CAN_ODIRECT))
 153                 return -EINVAL;
 154
 155         old_cred = override_creds(ctx->cred);
 156         if (is_sync_kiocb(iocb)) {
 157                 rwf_t rwf = iocb_to_rw_flags(flags);
 158
 159                 ret = vfs_iter_read(file, iter, &iocb->ki_pos, rwf);
 160         } else {
 161                 ret = -ENOMEM;
 162                 aio = kmem_cache_zalloc(backing_aio_cachep, GFP_KERNEL);
 163                 if (!aio)
 164                         goto out;
 165
 166                 aio->orig_iocb = iocb;
 167                 kiocb_clone(&aio->iocb, iocb, get_file(file));
 168                 aio->iocb.ki_complete = backing_aio_rw_complete;
 169                 refcount_set(&aio->ref, 2);
 170                 ret = vfs_iocb_iter_read(file, &aio->iocb, iter);
 171                 backing_aio_put(aio);
 172                 if (ret != -EIOCBQUEUED)
 173                         backing_aio_cleanup(aio, ret);
 174         }
 175 out:
 176         revert_creds(old_cred);
 177
 178         if (ctx->accessed)
 179                 ctx->accessed(ctx->user_file);
 180
 181         return ret;
 182 }
 183 EXPORT_SYMBOL_GPL(backing_file_read_iter);
 184
 185 ssize_t backing_file_write_iter(struct file *file, struct iov_iter *iter,
 186                                 struct kiocb *iocb, int flags,
 187                                 struct backing_file_ctx *ctx)
 188 {
 189         const struct cred *old_cred;
 190         ssize_t ret;
 191
 192         if (WARN_ON_ONCE(!(file->f_mode & FMODE_BACKING)))
 193                 return -EIO;
 194
 195         if (!iov_iter_count(iter))
 196                 return 0;
 197
 198         ret = file_remove_privs(ctx->user_file);
 199         if (ret)
 200                 return ret;
 201
 202         if (iocb->ki_flags & IOCB_DIRECT &&
 203             !(file->f_mode & FMODE_CAN_ODIRECT))
 204                 return -EINVAL;
 205
 206         /*
 207          * Stacked filesystems don't support deferred completions, don't copy
 208          * this property in case it is set by the issuer.
 209          */
 210         flags &= ~IOCB_DIO_CALLER_COMP;
 211
 212         old_cred = override_creds(ctx->cred);
 213         if (is_sync_kiocb(iocb)) {
 214                 rwf_t rwf = iocb_to_rw_flags(flags);
 215
 216                 ret = vfs_iter_write(file, iter, &iocb->ki_pos, rwf);
 217                 if (ctx->end_write)
 218                         ctx->end_write(ctx->user_file);
 219         } else {
 220                 struct backing_aio *aio;
 221
 222                 ret = backing_aio_init_wq(iocb);
 223                 if (ret)
 224                         goto out;
 225
 226                 ret = -ENOMEM;
 227                 aio = kmem_cache_zalloc(backing_aio_cachep, GFP_KERNEL);
 228                 if (!aio)
 229                         goto out;
 230
 231                 aio->orig_iocb = iocb;
 232                 aio->end_write = ctx->end_write;
 233                 kiocb_clone(&aio->iocb, iocb, get_file(file));
 234                 aio->iocb.ki_flags = flags;
 235                 aio->iocb.ki_complete = backing_aio_queue_completion;
 236                 refcount_set(&aio->ref, 2);
 237                 ret = vfs_iocb_iter_write(file, &aio->iocb, iter);
 238                 backing_aio_put(aio);
 239                 if (ret != -EIOCBQUEUED)
 240                         backing_aio_cleanup(aio, ret);
 241         }
 242 out:
 243         revert_creds(old_cred);
 244
 245         return ret;
 246 }
 247 EXPORT_SYMBOL_GPL(backing_file_write_iter);
 248
 249 ssize_t backing_file_splice_read(struct file *in, loff_t *ppos,
 250                                  struct pipe_inode_info *pipe, size_t len,
 251                                  unsigned int flags,
 252                                  struct backing_file_ctx *ctx)
 253 {
 254         const struct cred *old_cred;
 255         ssize_t ret;
 256
 257         if (WARN_ON_ONCE(!(in->f_mode & FMODE_BACKING)))
 258                 return -EIO;
 259
 260         old_cred = override_creds(ctx->cred);
 261         ret = vfs_splice_read(in, ppos, pipe, len, flags);
 262         revert_creds(old_cred);
 263
 264         if (ctx->accessed)
 265                 ctx->accessed(ctx->user_file);
 266
 267         return ret;
 268 }
 269 EXPORT_SYMBOL_GPL(backing_file_splice_read);
 270
 271 ssize_t backing_file_splice_write(struct pipe_inode_info *pipe,
 272                                   struct file *out, loff_t *ppos, size_t len,
 273                                   unsigned int flags,
 274                                   struct backing_file_ctx *ctx)
 275 {
 276         const struct cred *old_cred;
 277         ssize_t ret;
 278
 279         if (WARN_ON_ONCE(!(out->f_mode & FMODE_BACKING)))
 280                 return -EIO;
 281
 282         ret = file_remove_privs(ctx->user_file);
 283         if (ret)
 284                 return ret;
 285
 286         old_cred = override_creds(ctx->cred);
 287         file_start_write(out);
 288         ret = iter_file_splice_write(pipe, out, ppos, len, flags);
 289         file_end_write(out);
 290         revert_creds(old_cred);
 291
 292         if (ctx->end_write)
 293                 ctx->end_write(ctx->user_file);
 294
 295         return ret;
 296 }
 297 EXPORT_SYMBOL_GPL(backing_file_splice_write);
 298
 299 static int __init backing_aio_init(void)
 300 {
 301         backing_aio_cachep = kmem_cache_create("backing_aio",
 302                                                sizeof(struct backing_aio),
 303                                                0, SLAB_HWCACHE_ALIGN, NULL);
 304         if (!backing_aio_cachep)
 305                 return -ENOMEM;
 306
 307         return 0;
 308 }
 309 fs_initcall(backing_aio_init);