code.grnet.gr Git - archipelago/blob - xseg/sys/xsegdev.c

   1 /*
   2  *
   3  */
   4
   5 #include <linux/kernel.h>
   6 #include <linux/module.h>
   7 #include <linux/moduleparam.h>
   8 #include <linux/init.h>
   9 #include <linux/mm.h>
  10 #include <linux/fs.h>
  11 #include <linux/init.h>
  12 #include <linux/list.h>
  13 #include <linux/cdev.h>
  14 #include <linux/poll.h>
  15 #include <linux/slab.h>
  16 #include <linux/sched.h>
  17 #include <linux/ioctl.h>
  18 #include <linux/types.h>
  19 #include <linux/module.h>
  20 #include <linux/mmzone.h>
  21 #include <linux/vmalloc.h>
  22 #include <linux/spinlock.h>
  23 #include <linux/wait.h>
  24
  25 #include "xsegdev.h"
  26
  27 static struct xsegdev xsegdev;
  28
  29 int xsegdev_create_segment(struct xsegdev *dev, u64 segsize, char reserved)
  30 {
  31         void *segment;
  32         int ret = mutex_lock_interruptible(&dev->mutex);
  33         if (ret)
  34                 goto out;
  35
  36         ret = -EBUSY;
  37         if (dev->segment)
  38                 goto out_unlock;
  39
  40         /* vmalloc can handle large sizes */
  41         ret = -ENOMEM;
  42         segment = vmalloc(segsize);
  43         if (!segment)
  44                 goto out_unlock;
  45
  46         dev->segsize = segsize;
  47         dev->segment = segment;
  48         memset(dev->segment, 0, segsize);
  49         set_bit(XSEGDEV_READY, &dev->flags);
  50         if (reserved)
  51                 set_bit(XSEGDEV_RESERVED, &dev->flags);
  52         ret = 0;
  53
  54 out_unlock:
  55         mutex_unlock(&dev->mutex);
  56 out:
  57         return ret;
  58 }
  59
  60 EXPORT_SYMBOL(xsegdev_create_segment);
  61
  62 int xsegdev_destroy_segment(struct xsegdev *dev)
  63 {
  64         int ret = mutex_lock_interruptible(&dev->mutex);
  65         if (ret)
  66                 goto out;
  67
  68         /* VERIFY:
  69          * The segment trully dies when everyone in userspace has unmapped it.
  70          * However, the kernel mapping is immediately destroyed.
  71          * Kernel users are notified to abort via switching of XSEGDEV_READY.
  72          * The mapping deallocation is performed when all kernel users
  73          * have stopped using the segment as reported by usercount.
  74         */
  75
  76         ret = -EINVAL;
  77         if (!dev->segment)
  78                 goto out_unlock;
  79
  80         ret = -EBUSY;
  81         if (test_bit(XSEGDEV_RESERVED, &dev->flags))
  82                 goto out_unlock;
  83
  84         clear_bit(XSEGDEV_READY, &dev->flags);
  85         ret = wait_event_interruptible(dev->wq, atomic_read(&dev->usercount) <= 1);
  86         if (ret)
  87                 goto out_unlock;
  88
  89         vfree(dev->segment);
  90         dev->segment = NULL;
  91         dev->segsize = 0;
  92         ret = 0;
  93
  94 out_unlock:
  95         mutex_unlock(&dev->mutex);
  96         set_bit(XSEGDEV_READY, &dev->flags);
  97 out:
  98         return ret;
  99 }
 100
 101 EXPORT_SYMBOL(xsegdev_destroy_segment);
 102
 103 struct xsegdev *xsegdev_get(int minor)
 104 {
 105         struct xsegdev *dev = ERR_PTR(-ENODEV);
 106         if (minor)
 107                 goto out;
 108
 109         dev = &xsegdev;
 110         atomic_inc(&dev->usercount);
 111         if (!test_bit(XSEGDEV_READY, &dev->flags))
 112                 goto fail_busy;
 113 out:
 114         return dev;
 115
 116 fail_busy:
 117         atomic_dec(&dev->usercount);
 118         dev = ERR_PTR(-EBUSY);
 119         goto out;
 120 }
 121
 122 EXPORT_SYMBOL(xsegdev_get);
 123
 124 void xsegdev_put(struct xsegdev *dev)
 125 {
 126         atomic_dec(&dev->usercount);
 127         wake_up(&dev->wq);
 128         /* ain't all this too heavy ? */
 129 }
 130
 131 EXPORT_SYMBOL(xsegdev_put);
 132
 133 /* ********************* */
 134 /* ** File Operations ** */
 135 /* ********************* */
 136
 137 struct xsegdev_file {
 138         int minor;
 139 };
 140
 141 static int xsegdev_open(struct inode *inode, struct file *file)
 142 {
 143         struct xsegdev_file *vf = kmalloc(sizeof(struct xsegdev_file), GFP_KERNEL);
 144         if (!vf)
 145                 return -ENOMEM;
 146         vf->minor = 0;
 147         file->private_data = vf;
 148         return 0;
 149 }
 150
 151 static int xsegdev_release(struct inode *inode, struct file *file)
 152 {
 153         struct xsegdev_file *vf = file->private_data;
 154         kfree(vf);
 155         return 0;
 156 }
 157
 158 static long xsegdev_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 159 {
 160         struct xsegdev *dev;
 161         char *seg;
 162         long size;
 163         int ret = -EINVAL;
 164
 165         switch (cmd) {
 166
 167         case XSEGDEV_IOC_CREATESEG:
 168                 dev = xsegdev_get(0);
 169                 ret = IS_ERR(dev) ? PTR_ERR(dev) : 0;
 170                 if (ret)
 171                         goto out;
 172
 173                 ret = xsegdev_create_segment(dev, (u64)arg, 0);
 174                 xsegdev_put(dev);
 175                 goto out;
 176
 177         case XSEGDEV_IOC_DESTROYSEG:
 178                 dev = xsegdev_get(0);
 179                 ret = xsegdev_destroy_segment(&xsegdev);
 180                 xsegdev_put(dev);
 181                 goto out;
 182
 183         case XSEGDEV_IOC_SEGSIZE:
 184                 dev = xsegdev_get(0);
 185
 186                 ret = IS_ERR(dev) ? PTR_ERR(dev) : 0;
 187                 if (ret)
 188                         goto out;
 189
 190                 size = dev->segsize;
 191                 seg = dev->segment;
 192                 xsegdev_put(dev);
 193
 194                 ret = -ENODEV;
 195                 if (!seg)
 196                         goto out;
 197
 198                 return size;
 199         }
 200
 201 out:
 202         return ret;
 203 }
 204
 205 static ssize_t xsegdev_read(struct file *file, char __user *buf,
 206                            size_t count, loff_t *f_pos)
 207 {
 208         return 0;
 209 }
 210
 211 static ssize_t xsegdev_write(struct file *file, const char __user *buf,
 212                             size_t count, loff_t *f_pos)
 213 {
 214         struct xsegdev_file *vf = file->private_data;
 215         struct xsegdev *dev = xsegdev_get(vf->minor);
 216         int ret = -ENODEV;
 217         if (!dev)
 218                 goto out;
 219
 220         ret = -ENOSYS;
 221         if (dev->callback)
 222                 ret = dev->callback(dev->callarg);
 223
 224         xsegdev_put(dev);
 225 out:
 226         return ret;
 227 }
 228
 229 static int xsegdev_mmap(struct file *file, struct vm_area_struct *vma)
 230 {
 231         struct xsegdev_file *vf = file->private_data;
 232         struct xsegdev *dev;
 233         size_t size = vma->vm_end - vma->vm_start;
 234         unsigned long start = vma->vm_start, end = start + size;
 235         char *ptr;
 236         int ret = -ENODEV;
 237
 238         dev = xsegdev_get(vf->minor);
 239         if (IS_ERR(dev))
 240                 goto out;
 241
 242         ptr = dev->segment;
 243         if (!ptr)
 244                 goto out_put;
 245
 246         ret = -EINVAL;
 247
 248         /* do not allow offset mappings, for now */
 249         if (vma->vm_pgoff || size > dev->segsize)
 250                 goto out_put;
 251
 252         /* allow only shared, read-write mappings */
 253         if (!(vma->vm_flags & VM_SHARED))
 254                 goto out_put;
 255
 256         /* the segment is vmalloc() so we have to iterate through
 257          * all pages and laboriously map them one by one. */
 258         for (; start < end; start += PAGE_SIZE, ptr += PAGE_SIZE) {
 259                 ret = remap_pfn_range(vma, start, vmalloc_to_pfn(ptr),
 260                                       PAGE_SIZE, vma->vm_page_prot);
 261                 if (ret)
 262                         goto out_put; /* mmap syscall should clean up, right? */
 263         }
 264
 265         ret = 0;
 266
 267 out_put:
 268         xsegdev_put(dev);
 269 out:
 270         return ret;
 271 }
 272
 273 static struct file_operations xsegdev_ops =
 274 {
 275         .owner          = THIS_MODULE,
 276         .open           = xsegdev_open,
 277         .release        = xsegdev_release,
 278         .read           = xsegdev_read,
 279         .write          = xsegdev_write,
 280         .mmap           = xsegdev_mmap,
 281         .unlocked_ioctl = xsegdev_ioctl,
 282 };
 283
 284
 285 /* *************************** */
 286 /* ** Module Initialization ** */
 287 /* *************************** */
 288
 289 static void xsegdev_init(struct xsegdev *dev, int minor)
 290 {
 291         dev->minor = 0;
 292         dev->segment = NULL;
 293         dev->segsize = 0;
 294         dev->flags = 0;
 295         atomic_set(&dev->usercount, 0);
 296         init_waitqueue_head(&dev->wq);
 297         cdev_init(&dev->cdev, &xsegdev_ops);
 298         mutex_init(&dev->mutex);
 299         spin_lock_init(&dev->lock);
 300         dev->cdev.owner = THIS_MODULE;
 301         set_bit(XSEGDEV_READY, &dev->flags);
 302 }
 303
 304 int __init xsegdev_mod_init(void)
 305 {
 306         int ret;
 307         dev_t dev_no = MKDEV(XSEGDEV_MAJOR, 0);
 308         ret = register_chrdev_region(dev_no, 1, "xsegdev");
 309         if (ret < 0)
 310                 goto out;
 311
 312         xsegdev_init(&xsegdev, 0);
 313         ret = cdev_add(&xsegdev.cdev, dev_no, 1);
 314         if (ret < 0)
 315                 goto out_unregister;
 316
 317         return ret;
 318
 319 out_unregister:
 320         unregister_chrdev_region(dev_no, 1);
 321 out:
 322         return ret;
 323 }
 324
 325 void __exit xsegdev_mod_exit(void)
 326 {
 327         dev_t dev_no = MKDEV(XSEGDEV_MAJOR, 0);
 328         xsegdev_destroy_segment(&xsegdev);
 329         cdev_del(&xsegdev.cdev);
 330         unregister_chrdev_region(dev_no, 1);
 331 }
 332
 333 module_init(xsegdev_mod_init);
 334 module_exit(xsegdev_mod_exit);
 335
 336 MODULE_DESCRIPTION("xsegdev");
 337 MODULE_AUTHOR("XSEG");
 338 MODULE_LICENSE("GPL");
 339