Linux Audio

Check our new training course

Loading...
v5.9
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/* CacheFiles path walking and related routines
   3 *
   4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
   5 * Written by David Howells (dhowells@redhat.com)
 
 
 
 
 
   6 */
   7
   8#include <linux/module.h>
   9#include <linux/sched.h>
  10#include <linux/file.h>
  11#include <linux/fs.h>
  12#include <linux/fsnotify.h>
  13#include <linux/quotaops.h>
  14#include <linux/xattr.h>
  15#include <linux/mount.h>
  16#include <linux/namei.h>
  17#include <linux/security.h>
  18#include <linux/slab.h>
  19#include "internal.h"
  20
  21#define CACHEFILES_KEYBUF_SIZE 512
  22
  23/*
  24 * dump debugging info about an object
  25 */
  26static noinline
  27void __cachefiles_printk_object(struct cachefiles_object *object,
  28				const char *prefix)
 
  29{
  30	struct fscache_cookie *cookie;
  31	const u8 *k;
  32	unsigned loop;
  33
  34	pr_err("%sobject: OBJ%x\n", prefix, object->fscache.debug_id);
  35	pr_err("%sobjstate=%s fl=%lx wbusy=%x ev=%lx[%lx]\n",
  36	       prefix, object->fscache.state->name,
  37	       object->fscache.flags, work_busy(&object->fscache.work),
  38	       object->fscache.events, object->fscache.event_mask);
  39	pr_err("%sops=%u inp=%u exc=%u\n",
  40	       prefix, object->fscache.n_ops, object->fscache.n_in_progress,
  41	       object->fscache.n_exclusive);
  42	pr_err("%sparent=%p\n",
  43	       prefix, object->fscache.parent);
  44
  45	spin_lock(&object->fscache.lock);
  46	cookie = object->fscache.cookie;
  47	if (cookie) {
  48		pr_err("%scookie=%p [pr=%p nd=%p fl=%lx]\n",
  49		       prefix,
  50		       object->fscache.cookie,
  51		       object->fscache.cookie->parent,
  52		       object->fscache.cookie->netfs_data,
  53		       object->fscache.cookie->flags);
  54		pr_err("%skey=[%u] '", prefix, cookie->key_len);
  55		k = (cookie->key_len <= sizeof(cookie->inline_key)) ?
  56			cookie->inline_key : cookie->key;
  57		for (loop = 0; loop < cookie->key_len; loop++)
  58			pr_cont("%02x", k[loop]);
  59		pr_cont("'\n");
  60	} else {
  61		pr_err("%scookie=NULL\n", prefix);
 
  62	}
  63	spin_unlock(&object->fscache.lock);
 
 
 
 
 
 
 
  64}
  65
  66/*
  67 * dump debugging info about a pair of objects
  68 */
  69static noinline void cachefiles_printk_object(struct cachefiles_object *object,
  70					      struct cachefiles_object *xobject)
  71{
 
 
 
  72	if (object)
  73		__cachefiles_printk_object(object, "");
  74	if (xobject)
  75		__cachefiles_printk_object(xobject, "x");
 
  76}
  77
  78/*
  79 * mark the owner of a dentry, if there is one, to indicate that that dentry
  80 * has been preemptively deleted
  81 * - the caller must hold the i_mutex on the dentry's parent as required to
  82 *   call vfs_unlink(), vfs_rmdir() or vfs_rename()
  83 */
  84static void cachefiles_mark_object_buried(struct cachefiles_cache *cache,
  85					  struct dentry *dentry,
  86					  enum fscache_why_object_killed why)
  87{
  88	struct cachefiles_object *object;
  89	struct rb_node *p;
  90
  91	_enter(",'%pd'", dentry);
  92
  93	write_lock(&cache->active_lock);
  94
  95	p = cache->active_nodes.rb_node;
  96	while (p) {
  97		object = rb_entry(p, struct cachefiles_object, active_node);
  98		if (object->dentry > dentry)
  99			p = p->rb_left;
 100		else if (object->dentry < dentry)
 101			p = p->rb_right;
 102		else
 103			goto found_dentry;
 104	}
 105
 106	write_unlock(&cache->active_lock);
 107	trace_cachefiles_mark_buried(NULL, dentry, why);
 108	_leave(" [no owner]");
 109	return;
 110
 111	/* found the dentry for  */
 112found_dentry:
 113	kdebug("preemptive burial: OBJ%x [%s] %p",
 114	       object->fscache.debug_id,
 115	       object->fscache.state->name,
 116	       dentry);
 117
 118	trace_cachefiles_mark_buried(object, dentry, why);
 119
 120	if (fscache_object_is_live(&object->fscache)) {
 121		pr_err("\n");
 122		pr_err("Error: Can't preemptively bury live object\n");
 123		cachefiles_printk_object(object, NULL);
 124	} else {
 125		if (why != FSCACHE_OBJECT_IS_STALE)
 126			fscache_object_mark_killed(&object->fscache, why);
 127	}
 128
 129	write_unlock(&cache->active_lock);
 130	_leave(" [owner marked]");
 131}
 132
 133/*
 134 * record the fact that an object is now active
 135 */
 136static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
 137					 struct cachefiles_object *object)
 138{
 139	struct cachefiles_object *xobject;
 140	struct rb_node **_p, *_parent = NULL;
 141	struct dentry *dentry;
 142
 143	_enter(",%p", object);
 144
 145try_again:
 146	write_lock(&cache->active_lock);
 147
 148	dentry = object->dentry;
 149	trace_cachefiles_mark_active(object, dentry);
 150
 151	if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
 152		pr_err("Error: Object already active\n");
 153		cachefiles_printk_object(object, NULL);
 154		BUG();
 155	}
 156
 
 157	_p = &cache->active_nodes.rb_node;
 158	while (*_p) {
 159		_parent = *_p;
 160		xobject = rb_entry(_parent,
 161				   struct cachefiles_object, active_node);
 162
 163		ASSERT(xobject != object);
 164
 165		if (xobject->dentry > dentry)
 166			_p = &(*_p)->rb_left;
 167		else if (xobject->dentry < dentry)
 168			_p = &(*_p)->rb_right;
 169		else
 170			goto wait_for_old_object;
 171	}
 172
 173	rb_link_node(&object->active_node, _parent, _p);
 174	rb_insert_color(&object->active_node, &cache->active_nodes);
 175
 176	write_unlock(&cache->active_lock);
 177	_leave(" = 0");
 178	return 0;
 179
 180	/* an old object from a previous incarnation is hogging the slot - we
 181	 * need to wait for it to be destroyed */
 182wait_for_old_object:
 183	trace_cachefiles_wait_active(object, dentry, xobject);
 184	clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
 185
 186	if (fscache_object_is_live(&xobject->fscache)) {
 187		pr_err("\n");
 188		pr_err("Error: Unexpected object collision\n");
 189		cachefiles_printk_object(object, xobject);
 
 190	}
 191	atomic_inc(&xobject->usage);
 192	write_unlock(&cache->active_lock);
 193
 194	if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
 195		wait_queue_head_t *wq;
 196
 197		signed long timeout = 60 * HZ;
 198		wait_queue_entry_t wait;
 199		bool requeue;
 200
 201		/* if the object we're waiting for is queued for processing,
 202		 * then just put ourselves on the queue behind it */
 203		if (work_pending(&xobject->fscache.work)) {
 204			_debug("queue OBJ%x behind OBJ%x immediately",
 205			       object->fscache.debug_id,
 206			       xobject->fscache.debug_id);
 207			goto requeue;
 208		}
 209
 210		/* otherwise we sleep until either the object we're waiting for
 211		 * is done, or the fscache_object is congested */
 212		wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
 213		init_wait(&wait);
 214		requeue = false;
 215		do {
 216			prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
 217			if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
 218				break;
 219
 220			requeue = fscache_object_sleep_till_congested(&timeout);
 221		} while (timeout > 0 && !requeue);
 222		finish_wait(wq, &wait);
 223
 224		if (requeue &&
 225		    test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
 226			_debug("queue OBJ%x behind OBJ%x after wait",
 227			       object->fscache.debug_id,
 228			       xobject->fscache.debug_id);
 229			goto requeue;
 230		}
 231
 232		if (timeout <= 0) {
 233			pr_err("\n");
 234			pr_err("Error: Overlong wait for old active object to go away\n");
 235			cachefiles_printk_object(object, xobject);
 236			goto requeue;
 237		}
 238	}
 239
 240	ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
 241
 242	cache->cache.ops->put_object(&xobject->fscache,
 243		(enum fscache_obj_ref_trace)cachefiles_obj_put_wait_retry);
 244	goto try_again;
 245
 246requeue:
 247	cache->cache.ops->put_object(&xobject->fscache,
 248		(enum fscache_obj_ref_trace)cachefiles_obj_put_wait_timeo);
 249	_leave(" = -ETIMEDOUT");
 250	return -ETIMEDOUT;
 251}
 252
 253/*
 254 * Mark an object as being inactive.
 255 */
 256void cachefiles_mark_object_inactive(struct cachefiles_cache *cache,
 257				     struct cachefiles_object *object,
 258				     blkcnt_t i_blocks)
 259{
 260	struct dentry *dentry = object->dentry;
 261	struct inode *inode = d_backing_inode(dentry);
 262
 263	trace_cachefiles_mark_inactive(object, dentry, inode);
 264
 265	write_lock(&cache->active_lock);
 266	rb_erase(&object->active_node, &cache->active_nodes);
 267	clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
 268	write_unlock(&cache->active_lock);
 269
 270	wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
 271
 272	/* This object can now be culled, so we need to let the daemon know
 273	 * that there is something it can remove if it needs to.
 274	 */
 275	atomic_long_add(i_blocks, &cache->b_released);
 
 276	if (atomic_inc_return(&cache->f_released))
 277		cachefiles_state_changed(cache);
 278}
 279
 280/*
 281 * delete an object representation from the cache
 282 * - file backed objects are unlinked
 283 * - directory backed objects are stuffed into the graveyard for userspace to
 284 *   delete
 285 * - unlocks the directory mutex
 286 */
 287static int cachefiles_bury_object(struct cachefiles_cache *cache,
 288				  struct cachefiles_object *object,
 289				  struct dentry *dir,
 290				  struct dentry *rep,
 291				  bool preemptive,
 292				  enum fscache_why_object_killed why)
 293{
 294	struct dentry *grave, *trap;
 295	struct path path, path_to_graveyard;
 296	char nbuffer[8 + 8 + 1];
 297	int ret;
 298
 299	_enter(",'%pd','%pd'", dir, rep);
 300
 301	_debug("remove %p from %p", rep, dir);
 302
 303	/* non-directories can just be unlinked */
 304	if (!d_is_dir(rep)) {
 305		_debug("unlink stale object");
 306
 307		path.mnt = cache->mnt;
 308		path.dentry = dir;
 309		ret = security_path_unlink(&path, rep);
 310		if (ret < 0) {
 311			cachefiles_io_error(cache, "Unlink security error");
 312		} else {
 313			trace_cachefiles_unlink(object, rep, why);
 314			ret = vfs_unlink(d_inode(dir), rep, NULL);
 315
 316			if (preemptive)
 317				cachefiles_mark_object_buried(cache, rep, why);
 318		}
 319
 320		inode_unlock(d_inode(dir));
 321
 322		if (ret == -EIO)
 323			cachefiles_io_error(cache, "Unlink failed");
 324
 325		_leave(" = %d", ret);
 326		return ret;
 327	}
 328
 329	/* directories have to be moved to the graveyard */
 330	_debug("move stale object to graveyard");
 331	inode_unlock(d_inode(dir));
 332
 333try_again:
 334	/* first step is to make up a grave dentry in the graveyard */
 335	sprintf(nbuffer, "%08x%08x",
 336		(uint32_t) ktime_get_real_seconds(),
 337		(uint32_t) atomic_inc_return(&cache->gravecounter));
 338
 339	/* do the multiway lock magic */
 340	trap = lock_rename(cache->graveyard, dir);
 341
 342	/* do some checks before getting the grave dentry */
 343	if (rep->d_parent != dir || IS_DEADDIR(d_inode(rep))) {
 344		/* the entry was probably culled when we dropped the parent dir
 345		 * lock */
 346		unlock_rename(cache->graveyard, dir);
 347		_leave(" = 0 [culled?]");
 348		return 0;
 349	}
 350
 351	if (!d_can_lookup(cache->graveyard)) {
 352		unlock_rename(cache->graveyard, dir);
 353		cachefiles_io_error(cache, "Graveyard no longer a directory");
 354		return -EIO;
 355	}
 356
 357	if (trap == rep) {
 358		unlock_rename(cache->graveyard, dir);
 359		cachefiles_io_error(cache, "May not make directory loop");
 360		return -EIO;
 361	}
 362
 363	if (d_mountpoint(rep)) {
 364		unlock_rename(cache->graveyard, dir);
 365		cachefiles_io_error(cache, "Mountpoint in cache");
 366		return -EIO;
 367	}
 368
 369	grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
 370	if (IS_ERR(grave)) {
 371		unlock_rename(cache->graveyard, dir);
 372
 373		if (PTR_ERR(grave) == -ENOMEM) {
 374			_leave(" = -ENOMEM");
 375			return -ENOMEM;
 376		}
 377
 378		cachefiles_io_error(cache, "Lookup error %ld",
 379				    PTR_ERR(grave));
 380		return -EIO;
 381	}
 382
 383	if (d_is_positive(grave)) {
 384		unlock_rename(cache->graveyard, dir);
 385		dput(grave);
 386		grave = NULL;
 387		cond_resched();
 388		goto try_again;
 389	}
 390
 391	if (d_mountpoint(grave)) {
 392		unlock_rename(cache->graveyard, dir);
 393		dput(grave);
 394		cachefiles_io_error(cache, "Mountpoint in graveyard");
 395		return -EIO;
 396	}
 397
 398	/* target should not be an ancestor of source */
 399	if (trap == grave) {
 400		unlock_rename(cache->graveyard, dir);
 401		dput(grave);
 402		cachefiles_io_error(cache, "May not make directory loop");
 403		return -EIO;
 404	}
 405
 406	/* attempt the rename */
 407	path.mnt = cache->mnt;
 408	path.dentry = dir;
 409	path_to_graveyard.mnt = cache->mnt;
 410	path_to_graveyard.dentry = cache->graveyard;
 411	ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
 412	if (ret < 0) {
 413		cachefiles_io_error(cache, "Rename security error %d", ret);
 414	} else {
 415		trace_cachefiles_rename(object, rep, grave, why);
 416		ret = vfs_rename(d_inode(dir), rep,
 417				 d_inode(cache->graveyard), grave, NULL, 0);
 418		if (ret != 0 && ret != -ENOMEM)
 419			cachefiles_io_error(cache,
 420					    "Rename failed with error %d", ret);
 421
 422		if (preemptive)
 423			cachefiles_mark_object_buried(cache, rep, why);
 424	}
 425
 426	unlock_rename(cache->graveyard, dir);
 427	dput(grave);
 428	_leave(" = 0");
 429	return 0;
 430}
 431
 432/*
 433 * delete an object representation from the cache
 434 */
 435int cachefiles_delete_object(struct cachefiles_cache *cache,
 436			     struct cachefiles_object *object)
 437{
 438	struct dentry *dir;
 439	int ret;
 440
 441	_enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
 442
 443	ASSERT(object->dentry);
 444	ASSERT(d_backing_inode(object->dentry));
 445	ASSERT(object->dentry->d_parent);
 446
 447	dir = dget_parent(object->dentry);
 448
 449	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 450
 451	if (test_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->fscache.flags)) {
 452		/* object allocation for the same key preemptively deleted this
 453		 * object's file so that it could create its own file */
 454		_debug("object preemptively buried");
 455		inode_unlock(d_inode(dir));
 456		ret = 0;
 457	} else {
 458		/* we need to check that our parent is _still_ our parent - it
 459		 * may have been renamed */
 460		if (dir == object->dentry->d_parent) {
 461			ret = cachefiles_bury_object(cache, object, dir,
 462						     object->dentry, false,
 463						     FSCACHE_OBJECT_WAS_RETIRED);
 464		} else {
 465			/* it got moved, presumably by cachefilesd culling it,
 466			 * so it's no longer in the key path and we can ignore
 467			 * it */
 468			inode_unlock(d_inode(dir));
 469			ret = 0;
 470		}
 471	}
 472
 473	dput(dir);
 474	_leave(" = %d", ret);
 475	return ret;
 476}
 477
 478/*
 479 * walk from the parent object to the child object through the backing
 480 * filesystem, creating directories as we go
 481 */
 482int cachefiles_walk_to_object(struct cachefiles_object *parent,
 483			      struct cachefiles_object *object,
 484			      const char *key,
 485			      struct cachefiles_xattr *auxdata)
 486{
 487	struct cachefiles_cache *cache;
 488	struct dentry *dir, *next = NULL;
 489	struct inode *inode;
 490	struct path path;
 491	unsigned long start;
 492	const char *name;
 493	int ret, nlen;
 494
 495	_enter("OBJ%x{%p},OBJ%x,%s,",
 496	       parent->fscache.debug_id, parent->dentry,
 497	       object->fscache.debug_id, key);
 498
 499	cache = container_of(parent->fscache.cache,
 500			     struct cachefiles_cache, cache);
 501	path.mnt = cache->mnt;
 502
 503	ASSERT(parent->dentry);
 504	ASSERT(d_backing_inode(parent->dentry));
 505
 506	if (!(d_is_dir(parent->dentry))) {
 507		// TODO: convert file to dir
 508		_leave("looking up in none directory");
 509		return -ENOBUFS;
 510	}
 511
 512	dir = dget(parent->dentry);
 513
 514advance:
 515	/* attempt to transit the first directory component */
 516	name = key;
 517	nlen = strlen(key);
 518
 519	/* key ends in a double NUL */
 520	key = key + nlen + 1;
 521	if (!*key)
 522		key = NULL;
 523
 524lookup_again:
 525	/* search the current directory for the element name */
 526	_debug("lookup '%s'", name);
 527
 528	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 529
 530	start = jiffies;
 531	next = lookup_one_len(name, dir, nlen);
 532	cachefiles_hist(cachefiles_lookup_histogram, start);
 533	if (IS_ERR(next)) {
 534		trace_cachefiles_lookup(object, next, NULL);
 535		goto lookup_error;
 536	}
 537
 538	inode = d_backing_inode(next);
 539	trace_cachefiles_lookup(object, next, inode);
 540	_debug("next -> %p %s", next, inode ? "positive" : "negative");
 541
 542	if (!key)
 543		object->new = !inode;
 544
 545	/* if this element of the path doesn't exist, then the lookup phase
 546	 * failed, and we can release any readers in the certain knowledge that
 547	 * there's nothing for them to actually read */
 548	if (d_is_negative(next))
 549		fscache_object_lookup_negative(&object->fscache);
 550
 551	/* we need to create the object if it's negative */
 552	if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
 553		/* index objects and intervening tree levels must be subdirs */
 554		if (d_is_negative(next)) {
 555			ret = cachefiles_has_space(cache, 1, 0);
 556			if (ret < 0)
 557				goto no_space_error;
 558
 559			path.dentry = dir;
 560			ret = security_path_mkdir(&path, next, 0);
 561			if (ret < 0)
 562				goto create_error;
 563			start = jiffies;
 564			ret = vfs_mkdir(d_inode(dir), next, 0);
 565			cachefiles_hist(cachefiles_mkdir_histogram, start);
 566			if (!key)
 567				trace_cachefiles_mkdir(object, next, ret);
 568			if (ret < 0)
 569				goto create_error;
 570
 571			if (unlikely(d_unhashed(next))) {
 572				dput(next);
 573				inode_unlock(d_inode(dir));
 574				goto lookup_again;
 575			}
 576			ASSERT(d_backing_inode(next));
 577
 578			_debug("mkdir -> %p{%p{ino=%lu}}",
 579			       next, d_backing_inode(next), d_backing_inode(next)->i_ino);
 580
 581		} else if (!d_can_lookup(next)) {
 582			pr_err("inode %lu is not a directory\n",
 583			       d_backing_inode(next)->i_ino);
 584			ret = -ENOBUFS;
 585			goto error;
 586		}
 587
 588	} else {
 589		/* non-index objects start out life as files */
 590		if (d_is_negative(next)) {
 591			ret = cachefiles_has_space(cache, 1, 0);
 592			if (ret < 0)
 593				goto no_space_error;
 594
 595			path.dentry = dir;
 596			ret = security_path_mknod(&path, next, S_IFREG, 0);
 597			if (ret < 0)
 598				goto create_error;
 599			start = jiffies;
 600			ret = vfs_create(d_inode(dir), next, S_IFREG, true);
 601			cachefiles_hist(cachefiles_create_histogram, start);
 602			trace_cachefiles_create(object, next, ret);
 603			if (ret < 0)
 604				goto create_error;
 605
 606			ASSERT(d_backing_inode(next));
 607
 608			_debug("create -> %p{%p{ino=%lu}}",
 609			       next, d_backing_inode(next), d_backing_inode(next)->i_ino);
 610
 611		} else if (!d_can_lookup(next) &&
 612			   !d_is_reg(next)
 613			   ) {
 614			pr_err("inode %lu is not a file or directory\n",
 615			       d_backing_inode(next)->i_ino);
 616			ret = -ENOBUFS;
 617			goto error;
 618		}
 619	}
 620
 621	/* process the next component */
 622	if (key) {
 623		_debug("advance");
 624		inode_unlock(d_inode(dir));
 625		dput(dir);
 626		dir = next;
 627		next = NULL;
 628		goto advance;
 629	}
 630
 631	/* we've found the object we were looking for */
 632	object->dentry = next;
 633
 634	/* if we've found that the terminal object exists, then we need to
 635	 * check its attributes and delete it if it's out of date */
 636	if (!object->new) {
 637		_debug("validate '%pd'", next);
 638
 639		ret = cachefiles_check_object_xattr(object, auxdata);
 640		if (ret == -ESTALE) {
 641			/* delete the object (the deleter drops the directory
 642			 * mutex) */
 643			object->dentry = NULL;
 644
 645			ret = cachefiles_bury_object(cache, object, dir, next,
 646						     true,
 647						     FSCACHE_OBJECT_IS_STALE);
 648			dput(next);
 649			next = NULL;
 650
 651			if (ret < 0)
 652				goto delete_error;
 653
 654			_debug("redo lookup");
 655			fscache_object_retrying_stale(&object->fscache);
 656			goto lookup_again;
 657		}
 658	}
 659
 660	/* note that we're now using this object */
 661	ret = cachefiles_mark_object_active(cache, object);
 662
 663	inode_unlock(d_inode(dir));
 664	dput(dir);
 665	dir = NULL;
 666
 667	if (ret == -ETIMEDOUT)
 668		goto mark_active_timed_out;
 669
 670	_debug("=== OBTAINED_OBJECT ===");
 671
 672	if (object->new) {
 673		/* attach data to a newly constructed terminal object */
 674		ret = cachefiles_set_object_xattr(object, auxdata);
 675		if (ret < 0)
 676			goto check_error;
 677	} else {
 678		/* always update the atime on an object we've just looked up
 679		 * (this is used to keep track of culling, and atimes are only
 680		 * updated by read, write and readdir but not lookup or
 681		 * open) */
 682		path.dentry = next;
 683		touch_atime(&path);
 684	}
 685
 686	/* open a file interface onto a data file */
 687	if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
 688		if (d_is_reg(object->dentry)) {
 689			const struct address_space_operations *aops;
 690
 691			ret = -EPERM;
 692			aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
 693			if (!aops->bmap)
 694				goto check_error;
 695			if (object->dentry->d_sb->s_blocksize > PAGE_SIZE)
 696				goto check_error;
 697
 698			object->backer = object->dentry;
 699		} else {
 700			BUG(); // TODO: open file in data-class subdir
 701		}
 702	}
 703
 704	object->new = 0;
 705	fscache_obtained_object(&object->fscache);
 706
 707	_leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
 708	return 0;
 709
 710no_space_error:
 711	fscache_object_mark_killed(&object->fscache, FSCACHE_OBJECT_NO_SPACE);
 712create_error:
 713	_debug("create error %d", ret);
 714	if (ret == -EIO)
 715		cachefiles_io_error(cache, "Create/mkdir failed");
 716	goto error;
 717
 718mark_active_timed_out:
 719	_debug("mark active timed out");
 720	goto release_dentry;
 721
 722check_error:
 723	_debug("check error %d", ret);
 724	cachefiles_mark_object_inactive(
 725		cache, object, d_backing_inode(object->dentry)->i_blocks);
 726release_dentry:
 727	dput(object->dentry);
 728	object->dentry = NULL;
 729	goto error_out;
 730
 731delete_error:
 732	_debug("delete error %d", ret);
 733	goto error_out2;
 734
 735lookup_error:
 736	_debug("lookup error %ld", PTR_ERR(next));
 737	ret = PTR_ERR(next);
 738	if (ret == -EIO)
 739		cachefiles_io_error(cache, "Lookup failed");
 740	next = NULL;
 741error:
 742	inode_unlock(d_inode(dir));
 743	dput(next);
 744error_out2:
 745	dput(dir);
 746error_out:
 747	_leave(" = error %d", -ret);
 748	return ret;
 749}
 750
 751/*
 752 * get a subdirectory
 753 */
 754struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
 755					struct dentry *dir,
 756					const char *dirname)
 757{
 758	struct dentry *subdir;
 759	unsigned long start;
 760	struct path path;
 761	int ret;
 762
 763	_enter(",,%s", dirname);
 764
 765	/* search the current directory for the element name */
 766	inode_lock(d_inode(dir));
 767
 768retry:
 769	start = jiffies;
 770	subdir = lookup_one_len(dirname, dir, strlen(dirname));
 771	cachefiles_hist(cachefiles_lookup_histogram, start);
 772	if (IS_ERR(subdir)) {
 773		if (PTR_ERR(subdir) == -ENOMEM)
 774			goto nomem_d_alloc;
 775		goto lookup_error;
 776	}
 777
 778	_debug("subdir -> %p %s",
 779	       subdir, d_backing_inode(subdir) ? "positive" : "negative");
 780
 781	/* we need to create the subdir if it doesn't exist yet */
 782	if (d_is_negative(subdir)) {
 783		ret = cachefiles_has_space(cache, 1, 0);
 784		if (ret < 0)
 785			goto mkdir_error;
 786
 787		_debug("attempt mkdir");
 788
 789		path.mnt = cache->mnt;
 790		path.dentry = dir;
 791		ret = security_path_mkdir(&path, subdir, 0700);
 792		if (ret < 0)
 793			goto mkdir_error;
 794		ret = vfs_mkdir(d_inode(dir), subdir, 0700);
 795		if (ret < 0)
 796			goto mkdir_error;
 797
 798		if (unlikely(d_unhashed(subdir))) {
 799			dput(subdir);
 800			goto retry;
 801		}
 802		ASSERT(d_backing_inode(subdir));
 803
 804		_debug("mkdir -> %p{%p{ino=%lu}}",
 805		       subdir,
 806		       d_backing_inode(subdir),
 807		       d_backing_inode(subdir)->i_ino);
 808	}
 809
 810	inode_unlock(d_inode(dir));
 811
 812	/* we need to make sure the subdir is a directory */
 813	ASSERT(d_backing_inode(subdir));
 814
 815	if (!d_can_lookup(subdir)) {
 816		pr_err("%s is not a directory\n", dirname);
 817		ret = -EIO;
 818		goto check_error;
 819	}
 820
 821	ret = -EPERM;
 822	if (!(d_backing_inode(subdir)->i_opflags & IOP_XATTR) ||
 
 823	    !d_backing_inode(subdir)->i_op->lookup ||
 824	    !d_backing_inode(subdir)->i_op->mkdir ||
 825	    !d_backing_inode(subdir)->i_op->create ||
 826	    !d_backing_inode(subdir)->i_op->rename ||
 
 827	    !d_backing_inode(subdir)->i_op->rmdir ||
 828	    !d_backing_inode(subdir)->i_op->unlink)
 829		goto check_error;
 830
 831	_leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
 832	return subdir;
 833
 834check_error:
 835	dput(subdir);
 836	_leave(" = %d [check]", ret);
 837	return ERR_PTR(ret);
 838
 839mkdir_error:
 840	inode_unlock(d_inode(dir));
 841	dput(subdir);
 842	pr_err("mkdir %s failed with error %d\n", dirname, ret);
 843	return ERR_PTR(ret);
 844
 845lookup_error:
 846	inode_unlock(d_inode(dir));
 847	ret = PTR_ERR(subdir);
 848	pr_err("Lookup %s failed with error %d\n", dirname, ret);
 849	return ERR_PTR(ret);
 850
 851nomem_d_alloc:
 852	inode_unlock(d_inode(dir));
 853	_leave(" = -ENOMEM");
 854	return ERR_PTR(-ENOMEM);
 855}
 856
 857/*
 858 * find out if an object is in use or not
 859 * - if finds object and it's not in use:
 860 *   - returns a pointer to the object and a reference on it
 861 *   - returns with the directory locked
 862 */
 863static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
 864					      struct dentry *dir,
 865					      char *filename)
 866{
 867	struct cachefiles_object *object;
 868	struct rb_node *_n;
 869	struct dentry *victim;
 870	unsigned long start;
 871	int ret;
 872
 873	//_enter(",%pd/,%s",
 874	//       dir, filename);
 875
 876	/* look up the victim */
 877	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 878
 879	start = jiffies;
 880	victim = lookup_one_len(filename, dir, strlen(filename));
 881	cachefiles_hist(cachefiles_lookup_histogram, start);
 882	if (IS_ERR(victim))
 883		goto lookup_error;
 884
 885	//_debug("victim -> %p %s",
 886	//       victim, d_backing_inode(victim) ? "positive" : "negative");
 887
 888	/* if the object is no longer there then we probably retired the object
 889	 * at the netfs's request whilst the cull was in progress
 890	 */
 891	if (d_is_negative(victim)) {
 892		inode_unlock(d_inode(dir));
 893		dput(victim);
 894		_leave(" = -ENOENT [absent]");
 895		return ERR_PTR(-ENOENT);
 896	}
 897
 898	/* check to see if we're using this object */
 899	read_lock(&cache->active_lock);
 900
 901	_n = cache->active_nodes.rb_node;
 902
 903	while (_n) {
 904		object = rb_entry(_n, struct cachefiles_object, active_node);
 905
 906		if (object->dentry > victim)
 907			_n = _n->rb_left;
 908		else if (object->dentry < victim)
 909			_n = _n->rb_right;
 910		else
 911			goto object_in_use;
 912	}
 913
 914	read_unlock(&cache->active_lock);
 915
 916	//_leave(" = %p", victim);
 917	return victim;
 918
 919object_in_use:
 920	read_unlock(&cache->active_lock);
 921	inode_unlock(d_inode(dir));
 922	dput(victim);
 923	//_leave(" = -EBUSY [in use]");
 924	return ERR_PTR(-EBUSY);
 925
 926lookup_error:
 927	inode_unlock(d_inode(dir));
 928	ret = PTR_ERR(victim);
 929	if (ret == -ENOENT) {
 930		/* file or dir now absent - probably retired by netfs */
 931		_leave(" = -ESTALE [absent]");
 932		return ERR_PTR(-ESTALE);
 933	}
 934
 935	if (ret == -EIO) {
 936		cachefiles_io_error(cache, "Lookup failed");
 937	} else if (ret != -ENOMEM) {
 938		pr_err("Internal error: %d\n", ret);
 939		ret = -EIO;
 940	}
 941
 942	_leave(" = %d", ret);
 943	return ERR_PTR(ret);
 944}
 945
 946/*
 947 * cull an object if it's not in use
 948 * - called only by cache manager daemon
 949 */
 950int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
 951		    char *filename)
 952{
 953	struct dentry *victim;
 954	int ret;
 955
 956	_enter(",%pd/,%s", dir, filename);
 957
 958	victim = cachefiles_check_active(cache, dir, filename);
 959	if (IS_ERR(victim))
 960		return PTR_ERR(victim);
 961
 962	_debug("victim -> %p %s",
 963	       victim, d_backing_inode(victim) ? "positive" : "negative");
 964
 965	/* okay... the victim is not being used so we can cull it
 966	 * - start by marking it as stale
 967	 */
 968	_debug("victim is cullable");
 969
 970	ret = cachefiles_remove_object_xattr(cache, victim);
 971	if (ret < 0)
 972		goto error_unlock;
 973
 974	/*  actually remove the victim (drops the dir mutex) */
 975	_debug("bury");
 976
 977	ret = cachefiles_bury_object(cache, NULL, dir, victim, false,
 978				     FSCACHE_OBJECT_WAS_CULLED);
 979	if (ret < 0)
 980		goto error;
 981
 982	dput(victim);
 983	_leave(" = 0");
 984	return 0;
 985
 986error_unlock:
 987	inode_unlock(d_inode(dir));
 988error:
 989	dput(victim);
 990	if (ret == -ENOENT) {
 991		/* file or dir now absent - probably retired by netfs */
 992		_leave(" = -ESTALE [absent]");
 993		return -ESTALE;
 994	}
 995
 996	if (ret != -ENOMEM) {
 997		pr_err("Internal error: %d\n", ret);
 998		ret = -EIO;
 999	}
1000
1001	_leave(" = %d", ret);
1002	return ret;
1003}
1004
1005/*
1006 * find out if an object is in use or not
1007 * - called only by cache manager daemon
1008 * - returns -EBUSY or 0 to indicate whether an object is in use or not
1009 */
1010int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
1011			    char *filename)
1012{
1013	struct dentry *victim;
1014
1015	//_enter(",%pd/,%s",
1016	//       dir, filename);
1017
1018	victim = cachefiles_check_active(cache, dir, filename);
1019	if (IS_ERR(victim))
1020		return PTR_ERR(victim);
1021
1022	inode_unlock(d_inode(dir));
1023	dput(victim);
1024	//_leave(" = 0");
1025	return 0;
1026}
v4.6
 
   1/* CacheFiles path walking and related routines
   2 *
   3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
   4 * Written by David Howells (dhowells@redhat.com)
   5 *
   6 * This program is free software; you can redistribute it and/or
   7 * modify it under the terms of the GNU General Public Licence
   8 * as published by the Free Software Foundation; either version
   9 * 2 of the Licence, or (at your option) any later version.
  10 */
  11
  12#include <linux/module.h>
  13#include <linux/sched.h>
  14#include <linux/file.h>
  15#include <linux/fs.h>
  16#include <linux/fsnotify.h>
  17#include <linux/quotaops.h>
  18#include <linux/xattr.h>
  19#include <linux/mount.h>
  20#include <linux/namei.h>
  21#include <linux/security.h>
  22#include <linux/slab.h>
  23#include "internal.h"
  24
  25#define CACHEFILES_KEYBUF_SIZE 512
  26
  27/*
  28 * dump debugging info about an object
  29 */
  30static noinline
  31void __cachefiles_printk_object(struct cachefiles_object *object,
  32				const char *prefix,
  33				u8 *keybuf)
  34{
  35	struct fscache_cookie *cookie;
  36	unsigned keylen, loop;
 
  37
  38	pr_err("%sobject: OBJ%x\n", prefix, object->fscache.debug_id);
  39	pr_err("%sobjstate=%s fl=%lx wbusy=%x ev=%lx[%lx]\n",
  40	       prefix, object->fscache.state->name,
  41	       object->fscache.flags, work_busy(&object->fscache.work),
  42	       object->fscache.events, object->fscache.event_mask);
  43	pr_err("%sops=%u inp=%u exc=%u\n",
  44	       prefix, object->fscache.n_ops, object->fscache.n_in_progress,
  45	       object->fscache.n_exclusive);
  46	pr_err("%sparent=%p\n",
  47	       prefix, object->fscache.parent);
  48
  49	spin_lock(&object->fscache.lock);
  50	cookie = object->fscache.cookie;
  51	if (cookie) {
  52		pr_err("%scookie=%p [pr=%p nd=%p fl=%lx]\n",
  53		       prefix,
  54		       object->fscache.cookie,
  55		       object->fscache.cookie->parent,
  56		       object->fscache.cookie->netfs_data,
  57		       object->fscache.cookie->flags);
  58		if (keybuf && cookie->def)
  59			keylen = cookie->def->get_key(cookie->netfs_data, keybuf,
  60						      CACHEFILES_KEYBUF_SIZE);
  61		else
  62			keylen = 0;
 
  63	} else {
  64		pr_err("%scookie=NULL\n", prefix);
  65		keylen = 0;
  66	}
  67	spin_unlock(&object->fscache.lock);
  68
  69	if (keylen) {
  70		pr_err("%skey=[%u] '", prefix, keylen);
  71		for (loop = 0; loop < keylen; loop++)
  72			pr_cont("%02x", keybuf[loop]);
  73		pr_cont("'\n");
  74	}
  75}
  76
  77/*
  78 * dump debugging info about a pair of objects
  79 */
  80static noinline void cachefiles_printk_object(struct cachefiles_object *object,
  81					      struct cachefiles_object *xobject)
  82{
  83	u8 *keybuf;
  84
  85	keybuf = kmalloc(CACHEFILES_KEYBUF_SIZE, GFP_NOIO);
  86	if (object)
  87		__cachefiles_printk_object(object, "", keybuf);
  88	if (xobject)
  89		__cachefiles_printk_object(xobject, "x", keybuf);
  90	kfree(keybuf);
  91}
  92
  93/*
  94 * mark the owner of a dentry, if there is one, to indicate that that dentry
  95 * has been preemptively deleted
  96 * - the caller must hold the i_mutex on the dentry's parent as required to
  97 *   call vfs_unlink(), vfs_rmdir() or vfs_rename()
  98 */
  99static void cachefiles_mark_object_buried(struct cachefiles_cache *cache,
 100					  struct dentry *dentry,
 101					  enum fscache_why_object_killed why)
 102{
 103	struct cachefiles_object *object;
 104	struct rb_node *p;
 105
 106	_enter(",'%pd'", dentry);
 107
 108	write_lock(&cache->active_lock);
 109
 110	p = cache->active_nodes.rb_node;
 111	while (p) {
 112		object = rb_entry(p, struct cachefiles_object, active_node);
 113		if (object->dentry > dentry)
 114			p = p->rb_left;
 115		else if (object->dentry < dentry)
 116			p = p->rb_right;
 117		else
 118			goto found_dentry;
 119	}
 120
 121	write_unlock(&cache->active_lock);
 
 122	_leave(" [no owner]");
 123	return;
 124
 125	/* found the dentry for  */
 126found_dentry:
 127	kdebug("preemptive burial: OBJ%x [%s] %p",
 128	       object->fscache.debug_id,
 129	       object->fscache.state->name,
 130	       dentry);
 131
 
 
 132	if (fscache_object_is_live(&object->fscache)) {
 133		pr_err("\n");
 134		pr_err("Error: Can't preemptively bury live object\n");
 135		cachefiles_printk_object(object, NULL);
 136	} else {
 137		if (why != FSCACHE_OBJECT_IS_STALE)
 138			fscache_object_mark_killed(&object->fscache, why);
 139	}
 140
 141	write_unlock(&cache->active_lock);
 142	_leave(" [owner marked]");
 143}
 144
 145/*
 146 * record the fact that an object is now active
 147 */
 148static int cachefiles_mark_object_active(struct cachefiles_cache *cache,
 149					 struct cachefiles_object *object)
 150{
 151	struct cachefiles_object *xobject;
 152	struct rb_node **_p, *_parent = NULL;
 153	struct dentry *dentry;
 154
 155	_enter(",%p", object);
 156
 157try_again:
 158	write_lock(&cache->active_lock);
 159
 
 
 
 160	if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
 161		pr_err("Error: Object already active\n");
 162		cachefiles_printk_object(object, NULL);
 163		BUG();
 164	}
 165
 166	dentry = object->dentry;
 167	_p = &cache->active_nodes.rb_node;
 168	while (*_p) {
 169		_parent = *_p;
 170		xobject = rb_entry(_parent,
 171				   struct cachefiles_object, active_node);
 172
 173		ASSERT(xobject != object);
 174
 175		if (xobject->dentry > dentry)
 176			_p = &(*_p)->rb_left;
 177		else if (xobject->dentry < dentry)
 178			_p = &(*_p)->rb_right;
 179		else
 180			goto wait_for_old_object;
 181	}
 182
 183	rb_link_node(&object->active_node, _parent, _p);
 184	rb_insert_color(&object->active_node, &cache->active_nodes);
 185
 186	write_unlock(&cache->active_lock);
 187	_leave(" = 0");
 188	return 0;
 189
 190	/* an old object from a previous incarnation is hogging the slot - we
 191	 * need to wait for it to be destroyed */
 192wait_for_old_object:
 
 
 
 193	if (fscache_object_is_live(&xobject->fscache)) {
 194		pr_err("\n");
 195		pr_err("Error: Unexpected object collision\n");
 196		cachefiles_printk_object(object, xobject);
 197		BUG();
 198	}
 199	atomic_inc(&xobject->usage);
 200	write_unlock(&cache->active_lock);
 201
 202	if (test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
 203		wait_queue_head_t *wq;
 204
 205		signed long timeout = 60 * HZ;
 206		wait_queue_t wait;
 207		bool requeue;
 208
 209		/* if the object we're waiting for is queued for processing,
 210		 * then just put ourselves on the queue behind it */
 211		if (work_pending(&xobject->fscache.work)) {
 212			_debug("queue OBJ%x behind OBJ%x immediately",
 213			       object->fscache.debug_id,
 214			       xobject->fscache.debug_id);
 215			goto requeue;
 216		}
 217
 218		/* otherwise we sleep until either the object we're waiting for
 219		 * is done, or the fscache_object is congested */
 220		wq = bit_waitqueue(&xobject->flags, CACHEFILES_OBJECT_ACTIVE);
 221		init_wait(&wait);
 222		requeue = false;
 223		do {
 224			prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
 225			if (!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags))
 226				break;
 227
 228			requeue = fscache_object_sleep_till_congested(&timeout);
 229		} while (timeout > 0 && !requeue);
 230		finish_wait(wq, &wait);
 231
 232		if (requeue &&
 233		    test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags)) {
 234			_debug("queue OBJ%x behind OBJ%x after wait",
 235			       object->fscache.debug_id,
 236			       xobject->fscache.debug_id);
 237			goto requeue;
 238		}
 239
 240		if (timeout <= 0) {
 241			pr_err("\n");
 242			pr_err("Error: Overlong wait for old active object to go away\n");
 243			cachefiles_printk_object(object, xobject);
 244			goto requeue;
 245		}
 246	}
 247
 248	ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &xobject->flags));
 249
 250	cache->cache.ops->put_object(&xobject->fscache);
 
 251	goto try_again;
 252
 253requeue:
 254	clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
 255	cache->cache.ops->put_object(&xobject->fscache);
 256	_leave(" = -ETIMEDOUT");
 257	return -ETIMEDOUT;
 258}
 259
 260/*
 261 * Mark an object as being inactive.
 262 */
 263void cachefiles_mark_object_inactive(struct cachefiles_cache *cache,
 264				     struct cachefiles_object *object)
 
 265{
 
 
 
 
 
 266	write_lock(&cache->active_lock);
 267	rb_erase(&object->active_node, &cache->active_nodes);
 268	clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
 269	write_unlock(&cache->active_lock);
 270
 271	wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
 272
 273	/* This object can now be culled, so we need to let the daemon know
 274	 * that there is something it can remove if it needs to.
 275	 */
 276	atomic_long_add(d_backing_inode(object->dentry)->i_blocks,
 277			&cache->b_released);
 278	if (atomic_inc_return(&cache->f_released))
 279		cachefiles_state_changed(cache);
 280}
 281
 282/*
 283 * delete an object representation from the cache
 284 * - file backed objects are unlinked
 285 * - directory backed objects are stuffed into the graveyard for userspace to
 286 *   delete
 287 * - unlocks the directory mutex
 288 */
 289static int cachefiles_bury_object(struct cachefiles_cache *cache,
 
 290				  struct dentry *dir,
 291				  struct dentry *rep,
 292				  bool preemptive,
 293				  enum fscache_why_object_killed why)
 294{
 295	struct dentry *grave, *trap;
 296	struct path path, path_to_graveyard;
 297	char nbuffer[8 + 8 + 1];
 298	int ret;
 299
 300	_enter(",'%pd','%pd'", dir, rep);
 301
 302	_debug("remove %p from %p", rep, dir);
 303
 304	/* non-directories can just be unlinked */
 305	if (!d_is_dir(rep)) {
 306		_debug("unlink stale object");
 307
 308		path.mnt = cache->mnt;
 309		path.dentry = dir;
 310		ret = security_path_unlink(&path, rep);
 311		if (ret < 0) {
 312			cachefiles_io_error(cache, "Unlink security error");
 313		} else {
 
 314			ret = vfs_unlink(d_inode(dir), rep, NULL);
 315
 316			if (preemptive)
 317				cachefiles_mark_object_buried(cache, rep, why);
 318		}
 319
 320		inode_unlock(d_inode(dir));
 321
 322		if (ret == -EIO)
 323			cachefiles_io_error(cache, "Unlink failed");
 324
 325		_leave(" = %d", ret);
 326		return ret;
 327	}
 328
 329	/* directories have to be moved to the graveyard */
 330	_debug("move stale object to graveyard");
 331	inode_unlock(d_inode(dir));
 332
 333try_again:
 334	/* first step is to make up a grave dentry in the graveyard */
 335	sprintf(nbuffer, "%08x%08x",
 336		(uint32_t) get_seconds(),
 337		(uint32_t) atomic_inc_return(&cache->gravecounter));
 338
 339	/* do the multiway lock magic */
 340	trap = lock_rename(cache->graveyard, dir);
 341
 342	/* do some checks before getting the grave dentry */
 343	if (rep->d_parent != dir) {
 344		/* the entry was probably culled when we dropped the parent dir
 345		 * lock */
 346		unlock_rename(cache->graveyard, dir);
 347		_leave(" = 0 [culled?]");
 348		return 0;
 349	}
 350
 351	if (!d_can_lookup(cache->graveyard)) {
 352		unlock_rename(cache->graveyard, dir);
 353		cachefiles_io_error(cache, "Graveyard no longer a directory");
 354		return -EIO;
 355	}
 356
 357	if (trap == rep) {
 358		unlock_rename(cache->graveyard, dir);
 359		cachefiles_io_error(cache, "May not make directory loop");
 360		return -EIO;
 361	}
 362
 363	if (d_mountpoint(rep)) {
 364		unlock_rename(cache->graveyard, dir);
 365		cachefiles_io_error(cache, "Mountpoint in cache");
 366		return -EIO;
 367	}
 368
 369	grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
 370	if (IS_ERR(grave)) {
 371		unlock_rename(cache->graveyard, dir);
 372
 373		if (PTR_ERR(grave) == -ENOMEM) {
 374			_leave(" = -ENOMEM");
 375			return -ENOMEM;
 376		}
 377
 378		cachefiles_io_error(cache, "Lookup error %ld",
 379				    PTR_ERR(grave));
 380		return -EIO;
 381	}
 382
 383	if (d_is_positive(grave)) {
 384		unlock_rename(cache->graveyard, dir);
 385		dput(grave);
 386		grave = NULL;
 387		cond_resched();
 388		goto try_again;
 389	}
 390
 391	if (d_mountpoint(grave)) {
 392		unlock_rename(cache->graveyard, dir);
 393		dput(grave);
 394		cachefiles_io_error(cache, "Mountpoint in graveyard");
 395		return -EIO;
 396	}
 397
 398	/* target should not be an ancestor of source */
 399	if (trap == grave) {
 400		unlock_rename(cache->graveyard, dir);
 401		dput(grave);
 402		cachefiles_io_error(cache, "May not make directory loop");
 403		return -EIO;
 404	}
 405
 406	/* attempt the rename */
 407	path.mnt = cache->mnt;
 408	path.dentry = dir;
 409	path_to_graveyard.mnt = cache->mnt;
 410	path_to_graveyard.dentry = cache->graveyard;
 411	ret = security_path_rename(&path, rep, &path_to_graveyard, grave, 0);
 412	if (ret < 0) {
 413		cachefiles_io_error(cache, "Rename security error %d", ret);
 414	} else {
 
 415		ret = vfs_rename(d_inode(dir), rep,
 416				 d_inode(cache->graveyard), grave, NULL, 0);
 417		if (ret != 0 && ret != -ENOMEM)
 418			cachefiles_io_error(cache,
 419					    "Rename failed with error %d", ret);
 420
 421		if (preemptive)
 422			cachefiles_mark_object_buried(cache, rep, why);
 423	}
 424
 425	unlock_rename(cache->graveyard, dir);
 426	dput(grave);
 427	_leave(" = 0");
 428	return 0;
 429}
 430
 431/*
 432 * delete an object representation from the cache
 433 */
 434int cachefiles_delete_object(struct cachefiles_cache *cache,
 435			     struct cachefiles_object *object)
 436{
 437	struct dentry *dir;
 438	int ret;
 439
 440	_enter(",OBJ%x{%p}", object->fscache.debug_id, object->dentry);
 441
 442	ASSERT(object->dentry);
 443	ASSERT(d_backing_inode(object->dentry));
 444	ASSERT(object->dentry->d_parent);
 445
 446	dir = dget_parent(object->dentry);
 447
 448	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 449
 450	if (test_bit(FSCACHE_OBJECT_KILLED_BY_CACHE, &object->fscache.flags)) {
 451		/* object allocation for the same key preemptively deleted this
 452		 * object's file so that it could create its own file */
 453		_debug("object preemptively buried");
 454		inode_unlock(d_inode(dir));
 455		ret = 0;
 456	} else {
 457		/* we need to check that our parent is _still_ our parent - it
 458		 * may have been renamed */
 459		if (dir == object->dentry->d_parent) {
 460			ret = cachefiles_bury_object(cache, dir,
 461						     object->dentry, false,
 462						     FSCACHE_OBJECT_WAS_RETIRED);
 463		} else {
 464			/* it got moved, presumably by cachefilesd culling it,
 465			 * so it's no longer in the key path and we can ignore
 466			 * it */
 467			inode_unlock(d_inode(dir));
 468			ret = 0;
 469		}
 470	}
 471
 472	dput(dir);
 473	_leave(" = %d", ret);
 474	return ret;
 475}
 476
 477/*
 478 * walk from the parent object to the child object through the backing
 479 * filesystem, creating directories as we go
 480 */
 481int cachefiles_walk_to_object(struct cachefiles_object *parent,
 482			      struct cachefiles_object *object,
 483			      const char *key,
 484			      struct cachefiles_xattr *auxdata)
 485{
 486	struct cachefiles_cache *cache;
 487	struct dentry *dir, *next = NULL;
 
 488	struct path path;
 489	unsigned long start;
 490	const char *name;
 491	int ret, nlen;
 492
 493	_enter("OBJ%x{%p},OBJ%x,%s,",
 494	       parent->fscache.debug_id, parent->dentry,
 495	       object->fscache.debug_id, key);
 496
 497	cache = container_of(parent->fscache.cache,
 498			     struct cachefiles_cache, cache);
 499	path.mnt = cache->mnt;
 500
 501	ASSERT(parent->dentry);
 502	ASSERT(d_backing_inode(parent->dentry));
 503
 504	if (!(d_is_dir(parent->dentry))) {
 505		// TODO: convert file to dir
 506		_leave("looking up in none directory");
 507		return -ENOBUFS;
 508	}
 509
 510	dir = dget(parent->dentry);
 511
 512advance:
 513	/* attempt to transit the first directory component */
 514	name = key;
 515	nlen = strlen(key);
 516
 517	/* key ends in a double NUL */
 518	key = key + nlen + 1;
 519	if (!*key)
 520		key = NULL;
 521
 522lookup_again:
 523	/* search the current directory for the element name */
 524	_debug("lookup '%s'", name);
 525
 526	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 527
 528	start = jiffies;
 529	next = lookup_one_len(name, dir, nlen);
 530	cachefiles_hist(cachefiles_lookup_histogram, start);
 531	if (IS_ERR(next))
 
 532		goto lookup_error;
 
 533
 534	_debug("next -> %p %s", next, d_backing_inode(next) ? "positive" : "negative");
 
 
 535
 536	if (!key)
 537		object->new = !d_backing_inode(next);
 538
 539	/* if this element of the path doesn't exist, then the lookup phase
 540	 * failed, and we can release any readers in the certain knowledge that
 541	 * there's nothing for them to actually read */
 542	if (d_is_negative(next))
 543		fscache_object_lookup_negative(&object->fscache);
 544
 545	/* we need to create the object if it's negative */
 546	if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
 547		/* index objects and intervening tree levels must be subdirs */
 548		if (d_is_negative(next)) {
 549			ret = cachefiles_has_space(cache, 1, 0);
 550			if (ret < 0)
 551				goto no_space_error;
 552
 553			path.dentry = dir;
 554			ret = security_path_mkdir(&path, next, 0);
 555			if (ret < 0)
 556				goto create_error;
 557			start = jiffies;
 558			ret = vfs_mkdir(d_inode(dir), next, 0);
 559			cachefiles_hist(cachefiles_mkdir_histogram, start);
 
 
 560			if (ret < 0)
 561				goto create_error;
 562
 
 
 
 
 
 563			ASSERT(d_backing_inode(next));
 564
 565			_debug("mkdir -> %p{%p{ino=%lu}}",
 566			       next, d_backing_inode(next), d_backing_inode(next)->i_ino);
 567
 568		} else if (!d_can_lookup(next)) {
 569			pr_err("inode %lu is not a directory\n",
 570			       d_backing_inode(next)->i_ino);
 571			ret = -ENOBUFS;
 572			goto error;
 573		}
 574
 575	} else {
 576		/* non-index objects start out life as files */
 577		if (d_is_negative(next)) {
 578			ret = cachefiles_has_space(cache, 1, 0);
 579			if (ret < 0)
 580				goto no_space_error;
 581
 582			path.dentry = dir;
 583			ret = security_path_mknod(&path, next, S_IFREG, 0);
 584			if (ret < 0)
 585				goto create_error;
 586			start = jiffies;
 587			ret = vfs_create(d_inode(dir), next, S_IFREG, true);
 588			cachefiles_hist(cachefiles_create_histogram, start);
 
 589			if (ret < 0)
 590				goto create_error;
 591
 592			ASSERT(d_backing_inode(next));
 593
 594			_debug("create -> %p{%p{ino=%lu}}",
 595			       next, d_backing_inode(next), d_backing_inode(next)->i_ino);
 596
 597		} else if (!d_can_lookup(next) &&
 598			   !d_is_reg(next)
 599			   ) {
 600			pr_err("inode %lu is not a file or directory\n",
 601			       d_backing_inode(next)->i_ino);
 602			ret = -ENOBUFS;
 603			goto error;
 604		}
 605	}
 606
 607	/* process the next component */
 608	if (key) {
 609		_debug("advance");
 610		inode_unlock(d_inode(dir));
 611		dput(dir);
 612		dir = next;
 613		next = NULL;
 614		goto advance;
 615	}
 616
 617	/* we've found the object we were looking for */
 618	object->dentry = next;
 619
 620	/* if we've found that the terminal object exists, then we need to
 621	 * check its attributes and delete it if it's out of date */
 622	if (!object->new) {
 623		_debug("validate '%pd'", next);
 624
 625		ret = cachefiles_check_object_xattr(object, auxdata);
 626		if (ret == -ESTALE) {
 627			/* delete the object (the deleter drops the directory
 628			 * mutex) */
 629			object->dentry = NULL;
 630
 631			ret = cachefiles_bury_object(cache, dir, next, true,
 
 632						     FSCACHE_OBJECT_IS_STALE);
 633			dput(next);
 634			next = NULL;
 635
 636			if (ret < 0)
 637				goto delete_error;
 638
 639			_debug("redo lookup");
 640			fscache_object_retrying_stale(&object->fscache);
 641			goto lookup_again;
 642		}
 643	}
 644
 645	/* note that we're now using this object */
 646	ret = cachefiles_mark_object_active(cache, object);
 647
 648	inode_unlock(d_inode(dir));
 649	dput(dir);
 650	dir = NULL;
 651
 652	if (ret == -ETIMEDOUT)
 653		goto mark_active_timed_out;
 654
 655	_debug("=== OBTAINED_OBJECT ===");
 656
 657	if (object->new) {
 658		/* attach data to a newly constructed terminal object */
 659		ret = cachefiles_set_object_xattr(object, auxdata);
 660		if (ret < 0)
 661			goto check_error;
 662	} else {
 663		/* always update the atime on an object we've just looked up
 664		 * (this is used to keep track of culling, and atimes are only
 665		 * updated by read, write and readdir but not lookup or
 666		 * open) */
 667		path.dentry = next;
 668		touch_atime(&path);
 669	}
 670
 671	/* open a file interface onto a data file */
 672	if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
 673		if (d_is_reg(object->dentry)) {
 674			const struct address_space_operations *aops;
 675
 676			ret = -EPERM;
 677			aops = d_backing_inode(object->dentry)->i_mapping->a_ops;
 678			if (!aops->bmap)
 679				goto check_error;
 680			if (object->dentry->d_sb->s_blocksize > PAGE_SIZE)
 681				goto check_error;
 682
 683			object->backer = object->dentry;
 684		} else {
 685			BUG(); // TODO: open file in data-class subdir
 686		}
 687	}
 688
 689	object->new = 0;
 690	fscache_obtained_object(&object->fscache);
 691
 692	_leave(" = 0 [%lu]", d_backing_inode(object->dentry)->i_ino);
 693	return 0;
 694
 695no_space_error:
 696	fscache_object_mark_killed(&object->fscache, FSCACHE_OBJECT_NO_SPACE);
 697create_error:
 698	_debug("create error %d", ret);
 699	if (ret == -EIO)
 700		cachefiles_io_error(cache, "Create/mkdir failed");
 701	goto error;
 702
 703mark_active_timed_out:
 704	_debug("mark active timed out");
 705	goto release_dentry;
 706
 707check_error:
 708	_debug("check error %d", ret);
 709	cachefiles_mark_object_inactive(cache, object);
 
 710release_dentry:
 711	dput(object->dentry);
 712	object->dentry = NULL;
 713	goto error_out;
 714
 715delete_error:
 716	_debug("delete error %d", ret);
 717	goto error_out2;
 718
 719lookup_error:
 720	_debug("lookup error %ld", PTR_ERR(next));
 721	ret = PTR_ERR(next);
 722	if (ret == -EIO)
 723		cachefiles_io_error(cache, "Lookup failed");
 724	next = NULL;
 725error:
 726	inode_unlock(d_inode(dir));
 727	dput(next);
 728error_out2:
 729	dput(dir);
 730error_out:
 731	_leave(" = error %d", -ret);
 732	return ret;
 733}
 734
 735/*
 736 * get a subdirectory
 737 */
 738struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
 739					struct dentry *dir,
 740					const char *dirname)
 741{
 742	struct dentry *subdir;
 743	unsigned long start;
 744	struct path path;
 745	int ret;
 746
 747	_enter(",,%s", dirname);
 748
 749	/* search the current directory for the element name */
 750	inode_lock(d_inode(dir));
 751
 
 752	start = jiffies;
 753	subdir = lookup_one_len(dirname, dir, strlen(dirname));
 754	cachefiles_hist(cachefiles_lookup_histogram, start);
 755	if (IS_ERR(subdir)) {
 756		if (PTR_ERR(subdir) == -ENOMEM)
 757			goto nomem_d_alloc;
 758		goto lookup_error;
 759	}
 760
 761	_debug("subdir -> %p %s",
 762	       subdir, d_backing_inode(subdir) ? "positive" : "negative");
 763
 764	/* we need to create the subdir if it doesn't exist yet */
 765	if (d_is_negative(subdir)) {
 766		ret = cachefiles_has_space(cache, 1, 0);
 767		if (ret < 0)
 768			goto mkdir_error;
 769
 770		_debug("attempt mkdir");
 771
 772		path.mnt = cache->mnt;
 773		path.dentry = dir;
 774		ret = security_path_mkdir(&path, subdir, 0700);
 775		if (ret < 0)
 776			goto mkdir_error;
 777		ret = vfs_mkdir(d_inode(dir), subdir, 0700);
 778		if (ret < 0)
 779			goto mkdir_error;
 780
 
 
 
 
 781		ASSERT(d_backing_inode(subdir));
 782
 783		_debug("mkdir -> %p{%p{ino=%lu}}",
 784		       subdir,
 785		       d_backing_inode(subdir),
 786		       d_backing_inode(subdir)->i_ino);
 787	}
 788
 789	inode_unlock(d_inode(dir));
 790
 791	/* we need to make sure the subdir is a directory */
 792	ASSERT(d_backing_inode(subdir));
 793
 794	if (!d_can_lookup(subdir)) {
 795		pr_err("%s is not a directory\n", dirname);
 796		ret = -EIO;
 797		goto check_error;
 798	}
 799
 800	ret = -EPERM;
 801	if (!d_backing_inode(subdir)->i_op->setxattr ||
 802	    !d_backing_inode(subdir)->i_op->getxattr ||
 803	    !d_backing_inode(subdir)->i_op->lookup ||
 804	    !d_backing_inode(subdir)->i_op->mkdir ||
 805	    !d_backing_inode(subdir)->i_op->create ||
 806	    (!d_backing_inode(subdir)->i_op->rename &&
 807	     !d_backing_inode(subdir)->i_op->rename2) ||
 808	    !d_backing_inode(subdir)->i_op->rmdir ||
 809	    !d_backing_inode(subdir)->i_op->unlink)
 810		goto check_error;
 811
 812	_leave(" = [%lu]", d_backing_inode(subdir)->i_ino);
 813	return subdir;
 814
 815check_error:
 816	dput(subdir);
 817	_leave(" = %d [check]", ret);
 818	return ERR_PTR(ret);
 819
 820mkdir_error:
 821	inode_unlock(d_inode(dir));
 822	dput(subdir);
 823	pr_err("mkdir %s failed with error %d\n", dirname, ret);
 824	return ERR_PTR(ret);
 825
 826lookup_error:
 827	inode_unlock(d_inode(dir));
 828	ret = PTR_ERR(subdir);
 829	pr_err("Lookup %s failed with error %d\n", dirname, ret);
 830	return ERR_PTR(ret);
 831
 832nomem_d_alloc:
 833	inode_unlock(d_inode(dir));
 834	_leave(" = -ENOMEM");
 835	return ERR_PTR(-ENOMEM);
 836}
 837
 838/*
 839 * find out if an object is in use or not
 840 * - if finds object and it's not in use:
 841 *   - returns a pointer to the object and a reference on it
 842 *   - returns with the directory locked
 843 */
 844static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
 845					      struct dentry *dir,
 846					      char *filename)
 847{
 848	struct cachefiles_object *object;
 849	struct rb_node *_n;
 850	struct dentry *victim;
 851	unsigned long start;
 852	int ret;
 853
 854	//_enter(",%pd/,%s",
 855	//       dir, filename);
 856
 857	/* look up the victim */
 858	inode_lock_nested(d_inode(dir), I_MUTEX_PARENT);
 859
 860	start = jiffies;
 861	victim = lookup_one_len(filename, dir, strlen(filename));
 862	cachefiles_hist(cachefiles_lookup_histogram, start);
 863	if (IS_ERR(victim))
 864		goto lookup_error;
 865
 866	//_debug("victim -> %p %s",
 867	//       victim, d_backing_inode(victim) ? "positive" : "negative");
 868
 869	/* if the object is no longer there then we probably retired the object
 870	 * at the netfs's request whilst the cull was in progress
 871	 */
 872	if (d_is_negative(victim)) {
 873		inode_unlock(d_inode(dir));
 874		dput(victim);
 875		_leave(" = -ENOENT [absent]");
 876		return ERR_PTR(-ENOENT);
 877	}
 878
 879	/* check to see if we're using this object */
 880	read_lock(&cache->active_lock);
 881
 882	_n = cache->active_nodes.rb_node;
 883
 884	while (_n) {
 885		object = rb_entry(_n, struct cachefiles_object, active_node);
 886
 887		if (object->dentry > victim)
 888			_n = _n->rb_left;
 889		else if (object->dentry < victim)
 890			_n = _n->rb_right;
 891		else
 892			goto object_in_use;
 893	}
 894
 895	read_unlock(&cache->active_lock);
 896
 897	//_leave(" = %p", victim);
 898	return victim;
 899
 900object_in_use:
 901	read_unlock(&cache->active_lock);
 902	inode_unlock(d_inode(dir));
 903	dput(victim);
 904	//_leave(" = -EBUSY [in use]");
 905	return ERR_PTR(-EBUSY);
 906
 907lookup_error:
 908	inode_unlock(d_inode(dir));
 909	ret = PTR_ERR(victim);
 910	if (ret == -ENOENT) {
 911		/* file or dir now absent - probably retired by netfs */
 912		_leave(" = -ESTALE [absent]");
 913		return ERR_PTR(-ESTALE);
 914	}
 915
 916	if (ret == -EIO) {
 917		cachefiles_io_error(cache, "Lookup failed");
 918	} else if (ret != -ENOMEM) {
 919		pr_err("Internal error: %d\n", ret);
 920		ret = -EIO;
 921	}
 922
 923	_leave(" = %d", ret);
 924	return ERR_PTR(ret);
 925}
 926
 927/*
 928 * cull an object if it's not in use
 929 * - called only by cache manager daemon
 930 */
 931int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
 932		    char *filename)
 933{
 934	struct dentry *victim;
 935	int ret;
 936
 937	_enter(",%pd/,%s", dir, filename);
 938
 939	victim = cachefiles_check_active(cache, dir, filename);
 940	if (IS_ERR(victim))
 941		return PTR_ERR(victim);
 942
 943	_debug("victim -> %p %s",
 944	       victim, d_backing_inode(victim) ? "positive" : "negative");
 945
 946	/* okay... the victim is not being used so we can cull it
 947	 * - start by marking it as stale
 948	 */
 949	_debug("victim is cullable");
 950
 951	ret = cachefiles_remove_object_xattr(cache, victim);
 952	if (ret < 0)
 953		goto error_unlock;
 954
 955	/*  actually remove the victim (drops the dir mutex) */
 956	_debug("bury");
 957
 958	ret = cachefiles_bury_object(cache, dir, victim, false,
 959				     FSCACHE_OBJECT_WAS_CULLED);
 960	if (ret < 0)
 961		goto error;
 962
 963	dput(victim);
 964	_leave(" = 0");
 965	return 0;
 966
 967error_unlock:
 968	inode_unlock(d_inode(dir));
 969error:
 970	dput(victim);
 971	if (ret == -ENOENT) {
 972		/* file or dir now absent - probably retired by netfs */
 973		_leave(" = -ESTALE [absent]");
 974		return -ESTALE;
 975	}
 976
 977	if (ret != -ENOMEM) {
 978		pr_err("Internal error: %d\n", ret);
 979		ret = -EIO;
 980	}
 981
 982	_leave(" = %d", ret);
 983	return ret;
 984}
 985
 986/*
 987 * find out if an object is in use or not
 988 * - called only by cache manager daemon
 989 * - returns -EBUSY or 0 to indicate whether an object is in use or not
 990 */
 991int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
 992			    char *filename)
 993{
 994	struct dentry *victim;
 995
 996	//_enter(",%pd/,%s",
 997	//       dir, filename);
 998
 999	victim = cachefiles_check_active(cache, dir, filename);
1000	if (IS_ERR(victim))
1001		return PTR_ERR(victim);
1002
1003	inode_unlock(d_inode(dir));
1004	dput(victim);
1005	//_leave(" = 0");
1006	return 0;
1007}