Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
block: fine-granular CAP_SYS_ADMIN for Persistent Reservation
Allow of unprivileged Persistent Reservation operations on devices
if the write permission check on the device node has passed.
brw-rw---- 1 root disk 259, 0 Jun 13 07:09 /dev/nvme0n1
In the example above, the "disk" group of nvme0n1 is also allowed to
make reservations on the device even without CAP_SYS_ADMIN.
Signed-off-by: Jingbo Xu <jefflexu@linux.alibaba.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20230613084008.93795-3-jefflexu@linux.alibaba.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
authored by
Jingbo Xu
and committed by
Jens Axboe
9a72a024
12629621
+22
-19
+22
-19
block/ioctl.c
+22
-19
block/ioctl.c
···
254
254
EXPORT_SYMBOL(blkdev_compat_ptr_ioctl);
255
255
#endif
256
256
257
-
static bool blkdev_pr_allowed(struct block_device *bdev)
257
+
static bool blkdev_pr_allowed(struct block_device *bdev, blk_mode_t mode)
258
258
{
259
259
/* no sense to make reservations for partitions */
260
260
if (bdev_is_partition(bdev))
···
262
262
263
263
if (capable(CAP_SYS_ADMIN))
264
264
return true;
265
-
266
-
return false;
265
+
/*
266
+
* Only allow unprivileged reservations if the file descriptor is open
267
+
* for writing.
268
+
*/
269
+
return mode & BLK_OPEN_WRITE;
267
270
}
268
271
269
-
static int blkdev_pr_register(struct block_device *bdev,
272
+
static int blkdev_pr_register(struct block_device *bdev, blk_mode_t mode,
270
273
struct pr_registration __user *arg)
271
274
{
272
275
const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
273
276
struct pr_registration reg;
274
277
275
-
if (!blkdev_pr_allowed(bdev))
278
+
if (!blkdev_pr_allowed(bdev, mode))
276
279
return -EPERM;
277
280
if (!ops || !ops->pr_register)
278
281
return -EOPNOTSUPP;
···
287
284
return ops->pr_register(bdev, reg.old_key, reg.new_key, reg.flags);
288
285
}
289
286
290
-
static int blkdev_pr_reserve(struct block_device *bdev,
287
+
static int blkdev_pr_reserve(struct block_device *bdev, blk_mode_t mode,
291
288
struct pr_reservation __user *arg)
292
289
{
293
290
const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
294
291
struct pr_reservation rsv;
295
292
296
-
if (!blkdev_pr_allowed(bdev))
293
+
if (!blkdev_pr_allowed(bdev, mode))
297
294
return -EPERM;
298
295
if (!ops || !ops->pr_reserve)
299
296
return -EOPNOTSUPP;
···
305
302
return ops->pr_reserve(bdev, rsv.key, rsv.type, rsv.flags);
306
303
}
307
304
308
-
static int blkdev_pr_release(struct block_device *bdev,
305
+
static int blkdev_pr_release(struct block_device *bdev, blk_mode_t mode,
309
306
struct pr_reservation __user *arg)
310
307
{
311
308
const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
312
309
struct pr_reservation rsv;
313
310
314
-
if (!blkdev_pr_allowed(bdev))
311
+
if (!blkdev_pr_allowed(bdev, mode))
315
312
return -EPERM;
316
313
if (!ops || !ops->pr_release)
317
314
return -EOPNOTSUPP;
···
323
320
return ops->pr_release(bdev, rsv.key, rsv.type);
324
321
}
325
322
326
-
static int blkdev_pr_preempt(struct block_device *bdev,
323
+
static int blkdev_pr_preempt(struct block_device *bdev, blk_mode_t mode,
327
324
struct pr_preempt __user *arg, bool abort)
328
325
{
329
326
const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
330
327
struct pr_preempt p;
331
328
332
-
if (!blkdev_pr_allowed(bdev))
329
+
if (!blkdev_pr_allowed(bdev, mode))
333
330
return -EPERM;
334
331
if (!ops || !ops->pr_preempt)
335
332
return -EOPNOTSUPP;
···
341
338
return ops->pr_preempt(bdev, p.old_key, p.new_key, p.type, abort);
342
339
}
343
340
344
-
static int blkdev_pr_clear(struct block_device *bdev,
341
+
static int blkdev_pr_clear(struct block_device *bdev, blk_mode_t mode,
345
342
struct pr_clear __user *arg)
346
343
{
347
344
const struct pr_ops *ops = bdev->bd_disk->fops->pr_ops;
348
345
struct pr_clear c;
349
346
350
-
if (!blkdev_pr_allowed(bdev))
347
+
if (!blkdev_pr_allowed(bdev, mode))
351
348
return -EPERM;
352
349
if (!ops || !ops->pr_clear)
353
350
return -EOPNOTSUPP;
···
549
546
case BLKTRACETEARDOWN:
550
547
return blk_trace_ioctl(bdev, cmd, argp);
551
548
case IOC_PR_REGISTER:
552
-
return blkdev_pr_register(bdev, argp);
549
+
return blkdev_pr_register(bdev, mode, argp);
553
550
case IOC_PR_RESERVE:
554
-
return blkdev_pr_reserve(bdev, argp);
551
+
return blkdev_pr_reserve(bdev, mode, argp);
555
552
case IOC_PR_RELEASE:
556
-
return blkdev_pr_release(bdev, argp);
553
+
return blkdev_pr_release(bdev, mode, argp);
557
554
case IOC_PR_PREEMPT:
558
-
return blkdev_pr_preempt(bdev, argp, false);
555
+
return blkdev_pr_preempt(bdev, mode, argp, false);
559
556
case IOC_PR_PREEMPT_ABORT:
560
-
return blkdev_pr_preempt(bdev, argp, true);
557
+
return blkdev_pr_preempt(bdev, mode, argp, true);
561
558
case IOC_PR_CLEAR:
562
-
return blkdev_pr_clear(bdev, argp);
559
+
return blkdev_pr_clear(bdev, mode, argp);
563
560
default:
564
561
return -ENOIOCTLCMD;
565
562
}