1/*
   2 * fs/kernfs/file.c - kernfs file implementation
   3 *
   4 * Copyright (c) 2001-3 Patrick Mochel
   5 * Copyright (c) 2007 SUSE Linux Products GmbH
   6 * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org>
   7 *
   8 * This file is released under the GPLv2.
   9 */
  10
  11#include <linux/fs.h>
  12#include <linux/seq_file.h>
  13#include <linux/slab.h>
  14#include <linux/poll.h>
  15#include <linux/pagemap.h>
  16#include <linux/sched/mm.h>
  17#include <linux/fsnotify.h>
  18
  19#include "kernfs-internal.h"
  20
  21/*
  22 * There's one kernfs_open_file for each open file and one kernfs_open_node
  23 * for each kernfs_node with one or more open files.
  24 *
  25 * kernfs_node->attr.open points to kernfs_open_node.  attr.open is
  26 * protected by kernfs_open_node_lock.
  27 *
  28 * filp->private_data points to seq_file whose ->private points to
  29 * kernfs_open_file.  kernfs_open_files are chained at
  30 * kernfs_open_node->files, which is protected by kernfs_open_file_mutex.
  31 */
  32static DEFINE_SPINLOCK(kernfs_open_node_lock);
  33static DEFINE_MUTEX(kernfs_open_file_mutex);
  34
  35struct kernfs_open_node {
  36	atomic_t		refcnt;
  37	atomic_t		event;
  38	wait_queue_head_t	poll;
  39	struct list_head	files; /* goes through kernfs_open_file.list */
  40};
  41
  42/*
  43 * kernfs_notify() may be called from any context and bounces notifications
  44 * through a work item.  To minimize space overhead in kernfs_node, the
  45 * pending queue is implemented as a singly linked list of kernfs_nodes.
  46 * The list is terminated with the self pointer so that whether a
  47 * kernfs_node is on the list or not can be determined by testing the next
  48 * pointer for NULL.
  49 */
  50#define KERNFS_NOTIFY_EOL			((void *)&kernfs_notify_list)
  51
  52static DEFINE_SPINLOCK(kernfs_notify_lock);
  53static struct kernfs_node *kernfs_notify_list = KERNFS_NOTIFY_EOL;
  54
  55static struct kernfs_open_file *kernfs_of(struct file *file)
  56{
  57	return ((struct seq_file *)file->private_data)->private;
  58}
  59
  60/*
  61 * Determine the kernfs_ops for the given kernfs_node.  This function must
  62 * be called while holding an active reference.
  63 */
  64static const struct kernfs_ops *kernfs_ops(struct kernfs_node *kn)
  65{
  66	if (kn->flags & KERNFS_LOCKDEP)
  67		lockdep_assert_held(kn);
  68	return kn->attr.ops;
  69}
  70
  71/*
  72 * As kernfs_seq_stop() is also called after kernfs_seq_start() or
  73 * kernfs_seq_next() failure, it needs to distinguish whether it's stopping
  74 * a seq_file iteration which is fully initialized with an active reference
  75 * or an aborted kernfs_seq_start() due to get_active failure.  The
  76 * position pointer is the only context for each seq_file iteration and
  77 * thus the stop condition should be encoded in it.  As the return value is
  78 * directly visible to userland, ERR_PTR(-ENODEV) is the only acceptable
  79 * choice to indicate get_active failure.
  80 *
  81 * Unfortunately, this is complicated due to the optional custom seq_file
  82 * operations which may return ERR_PTR(-ENODEV) too.  kernfs_seq_stop()
  83 * can't distinguish whether ERR_PTR(-ENODEV) is from get_active failure or
  84 * custom seq_file operations and thus can't decide whether put_active
  85 * should be performed or not only on ERR_PTR(-ENODEV).
  86 *
  87 * This is worked around by factoring out the custom seq_stop() and
  88 * put_active part into kernfs_seq_stop_active(), skipping it from
  89 * kernfs_seq_stop() if ERR_PTR(-ENODEV) while invoking it directly after
  90 * custom seq_file operations fail with ERR_PTR(-ENODEV) - this ensures
  91 * that kernfs_seq_stop_active() is skipped only after get_active failure.
  92 */
  93static void kernfs_seq_stop_active(struct seq_file *sf, void *v)
  94{
  95	struct kernfs_open_file *of = sf->private;
  96	const struct kernfs_ops *ops = kernfs_ops(of->kn);
  97
  98	if (ops->seq_stop)
  99		ops->seq_stop(sf, v);
 100	kernfs_put_active(of->kn);
 101}
 102
 103static void *kernfs_seq_start(struct seq_file *sf, loff_t *ppos)
 104{
 105	struct kernfs_open_file *of = sf->private;
 106	const struct kernfs_ops *ops;
 107
 108	/*
 109	 * @of->mutex nests outside active ref and is primarily to ensure that
 110	 * the ops aren't called concurrently for the same open file.
 111	 */
 112	mutex_lock(&of->mutex);
 113	if (!kernfs_get_active(of->kn))
 114		return ERR_PTR(-ENODEV);
 115
 116	ops = kernfs_ops(of->kn);
 117	if (ops->seq_start) {
 118		void *next = ops->seq_start(sf, ppos);
 119		/* see the comment above kernfs_seq_stop_active() */
 120		if (next == ERR_PTR(-ENODEV))
 121			kernfs_seq_stop_active(sf, next);
 122		return next;
 123	} else {
 124		/*
 125		 * The same behavior and code as single_open().  Returns
 126		 * !NULL if pos is at the beginning; otherwise, NULL.
 127		 */
 128		return NULL + !*ppos;
 129	}
 130}
 131
 132static void *kernfs_seq_next(struct seq_file *sf, void *v, loff_t *ppos)
 133{
 134	struct kernfs_open_file *of = sf->private;
 135	const struct kernfs_ops *ops = kernfs_ops(of->kn);
 136
 137	if (ops->seq_next) {
 138		void *next = ops->seq_next(sf, v, ppos);
 139		/* see the comment above kernfs_seq_stop_active() */
 140		if (next == ERR_PTR(-ENODEV))
 141			kernfs_seq_stop_active(sf, next);
 142		return next;
 143	} else {
 144		/*
 145		 * The same behavior and code as single_open(), always
 146		 * terminate after the initial read.
 147		 */
 148		++*ppos;
 149		return NULL;
 150	}
 151}
 152
 153static void kernfs_seq_stop(struct seq_file *sf, void *v)
 154{
 155	struct kernfs_open_file *of = sf->private;
 156
 157	if (v != ERR_PTR(-ENODEV))
 158		kernfs_seq_stop_active(sf, v);
 159	mutex_unlock(&of->mutex);
 160}
 161
 162static int kernfs_seq_show(struct seq_file *sf, void *v)
 163{
 164	struct kernfs_open_file *of = sf->private;
 165
 166	of->event = atomic_read(&of->kn->attr.open->event);
 167
 168	return of->kn->attr.ops->seq_show(sf, v);
 169}
 170
 171static const struct seq_operations kernfs_seq_ops = {
 172	.start = kernfs_seq_start,
 173	.next = kernfs_seq_next,
 174	.stop = kernfs_seq_stop,
 175	.show = kernfs_seq_show,
 176};
 177
 178/*
 179 * As reading a bin file can have side-effects, the exact offset and bytes
 180 * specified in read(2) call should be passed to the read callback making
 181 * it difficult to use seq_file.  Implement simplistic custom buffering for
 182 * bin files.
 183 */
 184static ssize_t kernfs_file_direct_read(struct kernfs_open_file *of,
 185				       char __user *user_buf, size_t count,
 186				       loff_t *ppos)
 187{
 188	ssize_t len = min_t(size_t, count, PAGE_SIZE);
 189	const struct kernfs_ops *ops;
 190	char *buf;
 191
 192	buf = of->prealloc_buf;
 193	if (buf)
 194		mutex_lock(&of->prealloc_mutex);
 195	else
 196		buf = kmalloc(len, GFP_KERNEL);
 197	if (!buf)
 198		return -ENOMEM;
 199
 200	/*
 201	 * @of->mutex nests outside active ref and is used both to ensure that
 202	 * the ops aren't called concurrently for the same open file.
 203	 */
 204	mutex_lock(&of->mutex);
 205	if (!kernfs_get_active(of->kn)) {
 206		len = -ENODEV;
 207		mutex_unlock(&of->mutex);
 208		goto out_free;
 209	}
 210
 211	of->event = atomic_read(&of->kn->attr.open->event);
 212	ops = kernfs_ops(of->kn);
 213	if (ops->read)
 214		len = ops->read(of, buf, len, *ppos);
 215	else
 216		len = -EINVAL;
 217
 218	kernfs_put_active(of->kn);
 219	mutex_unlock(&of->mutex);
 220
 221	if (len < 0)
 222		goto out_free;
 223
 224	if (copy_to_user(user_buf, buf, len)) {
 225		len = -EFAULT;
 226		goto out_free;
 227	}
 228
 229	*ppos += len;
 230
 231 out_free:
 232	if (buf == of->prealloc_buf)
 233		mutex_unlock(&of->prealloc_mutex);
 234	else
 235		kfree(buf);
 236	return len;
 237}
 238
 239/**
 240 * kernfs_fop_read - kernfs vfs read callback
 241 * @file: file pointer
 242 * @user_buf: data to write
 243 * @count: number of bytes
 244 * @ppos: starting offset
 245 */
 246static ssize_t kernfs_fop_read(struct file *file, char __user *user_buf,
 247			       size_t count, loff_t *ppos)
 248{
 249	struct kernfs_open_file *of = kernfs_of(file);
 250
 251	if (of->kn->flags & KERNFS_HAS_SEQ_SHOW)
 252		return seq_read(file, user_buf, count, ppos);
 253	else
 254		return kernfs_file_direct_read(of, user_buf, count, ppos);
 255}
 256
 257/**
 258 * kernfs_fop_write - kernfs vfs write callback
 259 * @file: file pointer
 260 * @user_buf: data to write
 261 * @count: number of bytes
 262 * @ppos: starting offset
 263 *
 264 * Copy data in from userland and pass it to the matching kernfs write
 265 * operation.
 266 *
 267 * There is no easy way for us to know if userspace is only doing a partial
 268 * write, so we don't support them. We expect the entire buffer to come on
 269 * the first write.  Hint: if you're writing a value, first read the file,
 270 * modify only the the value you're changing, then write entire buffer
 271 * back.
 272 */
 273static ssize_t kernfs_fop_write(struct file *file, const char __user *user_buf,
 274				size_t count, loff_t *ppos)
 275{
 276	struct kernfs_open_file *of = kernfs_of(file);
 277	const struct kernfs_ops *ops;
 278	ssize_t len;
 279	char *buf;
 280
 281	if (of->atomic_write_len) {
 282		len = count;
 283		if (len > of->atomic_write_len)
 284			return -E2BIG;
 285	} else {
 286		len = min_t(size_t, count, PAGE_SIZE);
 287	}
 288
 289	buf = of->prealloc_buf;
 290	if (buf)
 291		mutex_lock(&of->prealloc_mutex);
 292	else
 293		buf = kmalloc(len + 1, GFP_KERNEL);
 294	if (!buf)
 295		return -ENOMEM;
 296
 297	if (copy_from_user(buf, user_buf, len)) {
 298		len = -EFAULT;
 299		goto out_free;
 300	}
 301	buf[len] = '\0';	/* guarantee string termination */
 302
 303	/*
 304	 * @of->mutex nests outside active ref and is used both to ensure that
 305	 * the ops aren't called concurrently for the same open file.
 306	 */
 307	mutex_lock(&of->mutex);
 308	if (!kernfs_get_active(of->kn)) {
 309		mutex_unlock(&of->mutex);
 310		len = -ENODEV;
 311		goto out_free;
 312	}
 313
 314	ops = kernfs_ops(of->kn);
 315	if (ops->write)
 316		len = ops->write(of, buf, len, *ppos);
 317	else
 318		len = -EINVAL;
 319
 320	kernfs_put_active(of->kn);
 321	mutex_unlock(&of->mutex);
 322
 323	if (len > 0)
 324		*ppos += len;
 325
 326out_free:
 327	if (buf == of->prealloc_buf)
 328		mutex_unlock(&of->prealloc_mutex);
 329	else
 330		kfree(buf);
 331	return len;
 332}
 333
 334static void kernfs_vma_open(struct vm_area_struct *vma)
 335{
 336	struct file *file = vma->vm_file;
 337	struct kernfs_open_file *of = kernfs_of(file);
 338
 339	if (!of->vm_ops)
 340		return;
 341
 342	if (!kernfs_get_active(of->kn))
 343		return;
 344
 345	if (of->vm_ops->open)
 346		of->vm_ops->open(vma);
 347
 348	kernfs_put_active(of->kn);
 349}
 350
 351static int kernfs_vma_fault(struct vm_fault *vmf)
 352{
 353	struct file *file = vmf->vma->vm_file;
 354	struct kernfs_open_file *of = kernfs_of(file);
 355	int ret;
 356
 357	if (!of->vm_ops)
 358		return VM_FAULT_SIGBUS;
 359
 360	if (!kernfs_get_active(of->kn))
 361		return VM_FAULT_SIGBUS;
 362
 363	ret = VM_FAULT_SIGBUS;
 364	if (of->vm_ops->fault)
 365		ret = of->vm_ops->fault(vmf);
 366
 367	kernfs_put_active(of->kn);
 368	return ret;
 369}
 370
 371static int kernfs_vma_page_mkwrite(struct vm_fault *vmf)
 
 372{
 373	struct file *file = vmf->vma->vm_file;
 374	struct kernfs_open_file *of = kernfs_of(file);
 375	int ret;
 376
 377	if (!of->vm_ops)
 378		return VM_FAULT_SIGBUS;
 379
 380	if (!kernfs_get_active(of->kn))
 381		return VM_FAULT_SIGBUS;
 382
 383	ret = 0;
 384	if (of->vm_ops->page_mkwrite)
 385		ret = of->vm_ops->page_mkwrite(vmf);
 386	else
 387		file_update_time(file);
 388
 389	kernfs_put_active(of->kn);
 390	return ret;
 391}
 392
 393static int kernfs_vma_access(struct vm_area_struct *vma, unsigned long addr,
 394			     void *buf, int len, int write)
 395{
 396	struct file *file = vma->vm_file;
 397	struct kernfs_open_file *of = kernfs_of(file);
 398	int ret;
 399
 400	if (!of->vm_ops)
 401		return -EINVAL;
 402
 403	if (!kernfs_get_active(of->kn))
 404		return -EINVAL;
 405
 406	ret = -EINVAL;
 407	if (of->vm_ops->access)
 408		ret = of->vm_ops->access(vma, addr, buf, len, write);
 409
 410	kernfs_put_active(of->kn);
 411	return ret;
 412}
 413
 414#ifdef CONFIG_NUMA
 415static int kernfs_vma_set_policy(struct vm_area_struct *vma,
 416				 struct mempolicy *new)
 417{
 418	struct file *file = vma->vm_file;
 419	struct kernfs_open_file *of = kernfs_of(file);
 420	int ret;
 421
 422	if (!of->vm_ops)
 423		return 0;
 424
 425	if (!kernfs_get_active(of->kn))
 426		return -EINVAL;
 427
 428	ret = 0;
 429	if (of->vm_ops->set_policy)
 430		ret = of->vm_ops->set_policy(vma, new);
 431
 432	kernfs_put_active(of->kn);
 433	return ret;
 434}
 435
 436static struct mempolicy *kernfs_vma_get_policy(struct vm_area_struct *vma,
 437					       unsigned long addr)
 438{
 439	struct file *file = vma->vm_file;
 440	struct kernfs_open_file *of = kernfs_of(file);
 441	struct mempolicy *pol;
 442
 443	if (!of->vm_ops)
 444		return vma->vm_policy;
 445
 446	if (!kernfs_get_active(of->kn))
 447		return vma->vm_policy;
 448
 449	pol = vma->vm_policy;
 450	if (of->vm_ops->get_policy)
 451		pol = of->vm_ops->get_policy(vma, addr);
 452
 453	kernfs_put_active(of->kn);
 454	return pol;
 455}
 456
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 457#endif
 458
 459static const struct vm_operations_struct kernfs_vm_ops = {
 460	.open		= kernfs_vma_open,
 461	.fault		= kernfs_vma_fault,
 462	.page_mkwrite	= kernfs_vma_page_mkwrite,
 463	.access		= kernfs_vma_access,
 464#ifdef CONFIG_NUMA
 465	.set_policy	= kernfs_vma_set_policy,
 466	.get_policy	= kernfs_vma_get_policy,
 
 467#endif
 468};
 469
 470static int kernfs_fop_mmap(struct file *file, struct vm_area_struct *vma)
 471{
 472	struct kernfs_open_file *of = kernfs_of(file);
 473	const struct kernfs_ops *ops;
 474	int rc;
 475
 476	/*
 477	 * mmap path and of->mutex are prone to triggering spurious lockdep
 478	 * warnings and we don't want to add spurious locking dependency
 479	 * between the two.  Check whether mmap is actually implemented
 480	 * without grabbing @of->mutex by testing HAS_MMAP flag.  See the
 481	 * comment in kernfs_file_open() for more details.
 482	 */
 483	if (!(of->kn->flags & KERNFS_HAS_MMAP))
 484		return -ENODEV;
 485
 486	mutex_lock(&of->mutex);
 487
 488	rc = -ENODEV;
 489	if (!kernfs_get_active(of->kn))
 490		goto out_unlock;
 491
 492	ops = kernfs_ops(of->kn);
 493	rc = ops->mmap(of, vma);
 494	if (rc)
 495		goto out_put;
 496
 497	/*
 498	 * PowerPC's pci_mmap of legacy_mem uses shmem_zero_setup()
 499	 * to satisfy versions of X which crash if the mmap fails: that
 500	 * substitutes a new vm_file, and we don't then want bin_vm_ops.
 501	 */
 502	if (vma->vm_file != file)
 503		goto out_put;
 504
 505	rc = -EINVAL;
 506	if (of->mmapped && of->vm_ops != vma->vm_ops)
 507		goto out_put;
 508
 509	/*
 510	 * It is not possible to successfully wrap close.
 511	 * So error if someone is trying to use close.
 512	 */
 513	rc = -EINVAL;
 514	if (vma->vm_ops && vma->vm_ops->close)
 515		goto out_put;
 516
 517	rc = 0;
 518	of->mmapped = true;
 519	of->vm_ops = vma->vm_ops;
 520	vma->vm_ops = &kernfs_vm_ops;
 521out_put:
 522	kernfs_put_active(of->kn);
 523out_unlock:
 524	mutex_unlock(&of->mutex);
 525
 526	return rc;
 527}
 528
 529/**
 530 *	kernfs_get_open_node - get or create kernfs_open_node
 531 *	@kn: target kernfs_node
 532 *	@of: kernfs_open_file for this instance of open
 533 *
 534 *	If @kn->attr.open exists, increment its reference count; otherwise,
 535 *	create one.  @of is chained to the files list.
 536 *
 537 *	LOCKING:
 538 *	Kernel thread context (may sleep).
 539 *
 540 *	RETURNS:
 541 *	0 on success, -errno on failure.
 542 */
 543static int kernfs_get_open_node(struct kernfs_node *kn,
 544				struct kernfs_open_file *of)
 545{
 546	struct kernfs_open_node *on, *new_on = NULL;
 547
 548 retry:
 549	mutex_lock(&kernfs_open_file_mutex);
 550	spin_lock_irq(&kernfs_open_node_lock);
 551
 552	if (!kn->attr.open && new_on) {
 553		kn->attr.open = new_on;
 554		new_on = NULL;
 555	}
 556
 557	on = kn->attr.open;
 558	if (on) {
 559		atomic_inc(&on->refcnt);
 560		list_add_tail(&of->list, &on->files);
 561	}
 562
 563	spin_unlock_irq(&kernfs_open_node_lock);
 564	mutex_unlock(&kernfs_open_file_mutex);
 565
 566	if (on) {
 567		kfree(new_on);
 568		return 0;
 569	}
 570
 571	/* not there, initialize a new one and retry */
 572	new_on = kmalloc(sizeof(*new_on), GFP_KERNEL);
 573	if (!new_on)
 574		return -ENOMEM;
 575
 576	atomic_set(&new_on->refcnt, 0);
 577	atomic_set(&new_on->event, 1);
 578	init_waitqueue_head(&new_on->poll);
 579	INIT_LIST_HEAD(&new_on->files);
 580	goto retry;
 581}
 582
 583/**
 584 *	kernfs_put_open_node - put kernfs_open_node
 585 *	@kn: target kernfs_nodet
 586 *	@of: associated kernfs_open_file
 587 *
 588 *	Put @kn->attr.open and unlink @of from the files list.  If
 589 *	reference count reaches zero, disassociate and free it.
 590 *
 591 *	LOCKING:
 592 *	None.
 593 */
 594static void kernfs_put_open_node(struct kernfs_node *kn,
 595				 struct kernfs_open_file *of)
 596{
 597	struct kernfs_open_node *on = kn->attr.open;
 598	unsigned long flags;
 599
 600	mutex_lock(&kernfs_open_file_mutex);
 601	spin_lock_irqsave(&kernfs_open_node_lock, flags);
 602
 603	if (of)
 604		list_del(&of->list);
 605
 606	if (atomic_dec_and_test(&on->refcnt))
 607		kn->attr.open = NULL;
 608	else
 609		on = NULL;
 610
 611	spin_unlock_irqrestore(&kernfs_open_node_lock, flags);
 612	mutex_unlock(&kernfs_open_file_mutex);
 613
 614	kfree(on);
 615}
 616
 617static int kernfs_fop_open(struct inode *inode, struct file *file)
 618{
 619	struct kernfs_node *kn = inode->i_private;
 620	struct kernfs_root *root = kernfs_root(kn);
 621	const struct kernfs_ops *ops;
 622	struct kernfs_open_file *of;
 623	bool has_read, has_write, has_mmap;
 624	int error = -EACCES;
 625
 626	if (!kernfs_get_active(kn))
 627		return -ENODEV;
 628
 629	ops = kernfs_ops(kn);
 630
 631	has_read = ops->seq_show || ops->read || ops->mmap;
 632	has_write = ops->write || ops->mmap;
 633	has_mmap = ops->mmap;
 634
 635	/* see the flag definition for details */
 636	if (root->flags & KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK) {
 637		if ((file->f_mode & FMODE_WRITE) &&
 638		    (!(inode->i_mode & S_IWUGO) || !has_write))
 639			goto err_out;
 640
 641		if ((file->f_mode & FMODE_READ) &&
 642		    (!(inode->i_mode & S_IRUGO) || !has_read))
 643			goto err_out;
 644	}
 645
 646	/* allocate a kernfs_open_file for the file */
 647	error = -ENOMEM;
 648	of = kzalloc(sizeof(struct kernfs_open_file), GFP_KERNEL);
 649	if (!of)
 650		goto err_out;
 651
 652	/*
 653	 * The following is done to give a different lockdep key to
 654	 * @of->mutex for files which implement mmap.  This is a rather
 655	 * crude way to avoid false positive lockdep warning around
 656	 * mm->mmap_sem - mmap nests @of->mutex under mm->mmap_sem and
 657	 * reading /sys/block/sda/trace/act_mask grabs sr_mutex, under
 658	 * which mm->mmap_sem nests, while holding @of->mutex.  As each
 659	 * open file has a separate mutex, it's okay as long as those don't
 660	 * happen on the same file.  At this point, we can't easily give
 661	 * each file a separate locking class.  Let's differentiate on
 662	 * whether the file has mmap or not for now.
 663	 *
 664	 * Both paths of the branch look the same.  They're supposed to
 665	 * look that way and give @of->mutex different static lockdep keys.
 666	 */
 667	if (has_mmap)
 668		mutex_init(&of->mutex);
 669	else
 670		mutex_init(&of->mutex);
 671
 672	of->kn = kn;
 673	of->file = file;
 674
 675	/*
 676	 * Write path needs to atomic_write_len outside active reference.
 677	 * Cache it in open_file.  See kernfs_fop_write() for details.
 678	 */
 679	of->atomic_write_len = ops->atomic_write_len;
 680
 681	error = -EINVAL;
 682	/*
 683	 * ->seq_show is incompatible with ->prealloc,
 684	 * as seq_read does its own allocation.
 685	 * ->read must be used instead.
 686	 */
 687	if (ops->prealloc && ops->seq_show)
 688		goto err_free;
 689	if (ops->prealloc) {
 690		int len = of->atomic_write_len ?: PAGE_SIZE;
 691		of->prealloc_buf = kmalloc(len + 1, GFP_KERNEL);
 692		error = -ENOMEM;
 693		if (!of->prealloc_buf)
 694			goto err_free;
 695		mutex_init(&of->prealloc_mutex);
 696	}
 697
 698	/*
 699	 * Always instantiate seq_file even if read access doesn't use
 700	 * seq_file or is not requested.  This unifies private data access
 701	 * and readable regular files are the vast majority anyway.
 702	 */
 703	if (ops->seq_show)
 704		error = seq_open(file, &kernfs_seq_ops);
 705	else
 706		error = seq_open(file, NULL);
 707	if (error)
 708		goto err_free;
 709
 710	of->seq_file = file->private_data;
 711	of->seq_file->private = of;
 712
 713	/* seq_file clears PWRITE unconditionally, restore it if WRITE */
 714	if (file->f_mode & FMODE_WRITE)
 715		file->f_mode |= FMODE_PWRITE;
 716
 717	/* make sure we have open node struct */
 718	error = kernfs_get_open_node(kn, of);
 719	if (error)
 720		goto err_seq_release;
 721
 722	if (ops->open) {
 723		/* nobody has access to @of yet, skip @of->mutex */
 724		error = ops->open(of);
 725		if (error)
 726			goto err_put_node;
 727	}
 728
 729	/* open succeeded, put active references */
 730	kernfs_put_active(kn);
 731	return 0;
 732
 733err_put_node:
 734	kernfs_put_open_node(kn, of);
 735err_seq_release:
 736	seq_release(inode, file);
 737err_free:
 738	kfree(of->prealloc_buf);
 739	kfree(of);
 740err_out:
 741	kernfs_put_active(kn);
 742	return error;
 743}
 744
 745/* used from release/drain to ensure that ->release() is called exactly once */
 746static void kernfs_release_file(struct kernfs_node *kn,
 747				struct kernfs_open_file *of)
 748{
 749	/*
 750	 * @of is guaranteed to have no other file operations in flight and
 751	 * we just want to synchronize release and drain paths.
 752	 * @kernfs_open_file_mutex is enough.  @of->mutex can't be used
 753	 * here because drain path may be called from places which can
 754	 * cause circular dependency.
 755	 */
 756	lockdep_assert_held(&kernfs_open_file_mutex);
 757
 758	if (!of->released) {
 759		/*
 760		 * A file is never detached without being released and we
 761		 * need to be able to release files which are deactivated
 762		 * and being drained.  Don't use kernfs_ops().
 763		 */
 764		kn->attr.ops->release(of);
 765		of->released = true;
 766	}
 767}
 768
 769static int kernfs_fop_release(struct inode *inode, struct file *filp)
 770{
 771	struct kernfs_node *kn = inode->i_private;
 772	struct kernfs_open_file *of = kernfs_of(filp);
 773
 774	if (kn->flags & KERNFS_HAS_RELEASE) {
 775		mutex_lock(&kernfs_open_file_mutex);
 776		kernfs_release_file(kn, of);
 777		mutex_unlock(&kernfs_open_file_mutex);
 778	}
 779
 780	kernfs_put_open_node(kn, of);
 781	seq_release(inode, filp);
 782	kfree(of->prealloc_buf);
 783	kfree(of);
 784
 785	return 0;
 786}
 787
 788void kernfs_drain_open_files(struct kernfs_node *kn)
 789{
 790	struct kernfs_open_node *on;
 791	struct kernfs_open_file *of;
 792
 793	if (!(kn->flags & (KERNFS_HAS_MMAP | KERNFS_HAS_RELEASE)))
 794		return;
 795
 796	spin_lock_irq(&kernfs_open_node_lock);
 797	on = kn->attr.open;
 798	if (on)
 799		atomic_inc(&on->refcnt);
 800	spin_unlock_irq(&kernfs_open_node_lock);
 801	if (!on)
 802		return;
 803
 804	mutex_lock(&kernfs_open_file_mutex);
 805
 806	list_for_each_entry(of, &on->files, list) {
 807		struct inode *inode = file_inode(of->file);
 808
 809		if (kn->flags & KERNFS_HAS_MMAP)
 810			unmap_mapping_range(inode->i_mapping, 0, 0, 1);
 811
 812		if (kn->flags & KERNFS_HAS_RELEASE)
 813			kernfs_release_file(kn, of);
 814	}
 815
 816	mutex_unlock(&kernfs_open_file_mutex);
 817
 818	kernfs_put_open_node(kn, NULL);
 819}
 820
 821/*
 822 * Kernfs attribute files are pollable.  The idea is that you read
 823 * the content and then you use 'poll' or 'select' to wait for
 824 * the content to change.  When the content changes (assuming the
 825 * manager for the kobject supports notification), poll will
 826 * return EPOLLERR|EPOLLPRI, and select will return the fd whether
 827 * it is waiting for read, write, or exceptions.
 828 * Once poll/select indicates that the value has changed, you
 829 * need to close and re-open the file, or seek to 0 and read again.
 830 * Reminder: this only works for attributes which actively support
 831 * it, and it is not possible to test an attribute from userspace
 832 * to see if it supports poll (Neither 'poll' nor 'select' return
 833 * an appropriate error code).  When in doubt, set a suitable timeout value.
 834 */
 835static __poll_t kernfs_fop_poll(struct file *filp, poll_table *wait)
 836{
 837	struct kernfs_open_file *of = kernfs_of(filp);
 838	struct kernfs_node *kn = kernfs_dentry_node(filp->f_path.dentry);
 839	struct kernfs_open_node *on = kn->attr.open;
 840
 
 841	if (!kernfs_get_active(kn))
 842		goto trigger;
 843
 844	poll_wait(filp, &on->poll, wait);
 845
 846	kernfs_put_active(kn);
 847
 848	if (of->event != atomic_read(&on->event))
 849		goto trigger;
 850
 851	return DEFAULT_POLLMASK;
 852
 853 trigger:
 854	return DEFAULT_POLLMASK|EPOLLERR|EPOLLPRI;
 855}
 856
 857static void kernfs_notify_workfn(struct work_struct *work)
 858{
 859	struct kernfs_node *kn;
 860	struct kernfs_open_node *on;
 861	struct kernfs_super_info *info;
 862repeat:
 863	/* pop one off the notify_list */
 864	spin_lock_irq(&kernfs_notify_lock);
 865	kn = kernfs_notify_list;
 866	if (kn == KERNFS_NOTIFY_EOL) {
 867		spin_unlock_irq(&kernfs_notify_lock);
 868		return;
 869	}
 870	kernfs_notify_list = kn->attr.notify_next;
 871	kn->attr.notify_next = NULL;
 872	spin_unlock_irq(&kernfs_notify_lock);
 873
 874	/* kick poll */
 875	spin_lock_irq(&kernfs_open_node_lock);
 876
 877	on = kn->attr.open;
 878	if (on) {
 879		atomic_inc(&on->event);
 880		wake_up_interruptible(&on->poll);
 881	}
 882
 883	spin_unlock_irq(&kernfs_open_node_lock);
 884
 885	/* kick fsnotify */
 886	mutex_lock(&kernfs_mutex);
 887
 888	list_for_each_entry(info, &kernfs_root(kn)->supers, node) {
 889		struct kernfs_node *parent;
 890		struct inode *inode;
 891
 892		/*
 893		 * We want fsnotify_modify() on @kn but as the
 894		 * modifications aren't originating from userland don't
 895		 * have the matching @file available.  Look up the inodes
 896		 * and generate the events manually.
 897		 */
 898		inode = ilookup(info->sb, kn->id.ino);
 899		if (!inode)
 900			continue;
 901
 902		parent = kernfs_get_parent(kn);
 903		if (parent) {
 904			struct inode *p_inode;
 905
 906			p_inode = ilookup(info->sb, parent->id.ino);
 907			if (p_inode) {
 908				fsnotify(p_inode, FS_MODIFY | FS_EVENT_ON_CHILD,
 909					 inode, FSNOTIFY_EVENT_INODE, kn->name, 0);
 910				iput(p_inode);
 911			}
 912
 913			kernfs_put(parent);
 914		}
 915
 916		fsnotify(inode, FS_MODIFY, inode, FSNOTIFY_EVENT_INODE,
 917			 kn->name, 0);
 918		iput(inode);
 919	}
 920
 921	mutex_unlock(&kernfs_mutex);
 922	kernfs_put(kn);
 923	goto repeat;
 924}
 925
 926/**
 927 * kernfs_notify - notify a kernfs file
 928 * @kn: file to notify
 929 *
 930 * Notify @kn such that poll(2) on @kn wakes up.  Maybe be called from any
 931 * context.
 932 */
 933void kernfs_notify(struct kernfs_node *kn)
 934{
 935	static DECLARE_WORK(kernfs_notify_work, kernfs_notify_workfn);
 936	unsigned long flags;
 937
 938	if (WARN_ON(kernfs_type(kn) != KERNFS_FILE))
 939		return;
 940
 941	spin_lock_irqsave(&kernfs_notify_lock, flags);
 942	if (!kn->attr.notify_next) {
 943		kernfs_get(kn);
 944		kn->attr.notify_next = kernfs_notify_list;
 945		kernfs_notify_list = kn;
 946		schedule_work(&kernfs_notify_work);
 947	}
 948	spin_unlock_irqrestore(&kernfs_notify_lock, flags);
 
 949}
 950EXPORT_SYMBOL_GPL(kernfs_notify);
 951
 952const struct file_operations kernfs_file_fops = {
 953	.read		= kernfs_fop_read,
 954	.write		= kernfs_fop_write,
 955	.llseek		= generic_file_llseek,
 956	.mmap		= kernfs_fop_mmap,
 957	.open		= kernfs_fop_open,
 958	.release	= kernfs_fop_release,
 959	.poll		= kernfs_fop_poll,
 960	.fsync		= noop_fsync,
 961};
 962
 963/**
 964 * __kernfs_create_file - kernfs internal function to create a file
 965 * @parent: directory to create the file in
 966 * @name: name of the file
 967 * @mode: mode of the file
 968 * @size: size of the file
 969 * @ops: kernfs operations for the file
 970 * @priv: private data for the file
 971 * @ns: optional namespace tag of the file
 
 972 * @key: lockdep key for the file's active_ref, %NULL to disable lockdep
 973 *
 974 * Returns the created node on success, ERR_PTR() value on error.
 975 */
 976struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
 977					 const char *name,
 978					 umode_t mode, loff_t size,
 979					 const struct kernfs_ops *ops,
 980					 void *priv, const void *ns,
 
 981					 struct lock_class_key *key)
 982{
 983	struct kernfs_node *kn;
 984	unsigned flags;
 985	int rc;
 986
 987	flags = KERNFS_FILE;
 
 
 988
 989	kn = kernfs_new_node(parent, name, (mode & S_IALLUGO) | S_IFREG, flags);
 990	if (!kn)
 991		return ERR_PTR(-ENOMEM);
 992
 993	kn->attr.ops = ops;
 994	kn->attr.size = size;
 995	kn->ns = ns;
 996	kn->priv = priv;
 997
 998#ifdef CONFIG_DEBUG_LOCK_ALLOC
 999	if (key) {
1000		lockdep_init_map(&kn->dep_map, "kn->count", key, 0);
1001		kn->flags |= KERNFS_LOCKDEP;
1002	}
1003#endif
1004
1005	/*
1006	 * kn->attr.ops is accesible only while holding active ref.  We
1007	 * need to know whether some ops are implemented outside active
1008	 * ref.  Cache their existence in flags.
1009	 */
1010	if (ops->seq_show)
1011		kn->flags |= KERNFS_HAS_SEQ_SHOW;
1012	if (ops->mmap)
1013		kn->flags |= KERNFS_HAS_MMAP;
1014	if (ops->release)
1015		kn->flags |= KERNFS_HAS_RELEASE;
1016
1017	rc = kernfs_add_one(kn);
1018	if (rc) {
1019		kernfs_put(kn);
1020		return ERR_PTR(rc);
1021	}
1022	return kn;
1023}