1/*
   2 *  linux/fs/nfs/dir.c
   3 *
   4 *  Copyright (C) 1992  Rick Sladkey
   5 *
   6 *  nfs directory handling functions
   7 *
   8 * 10 Apr 1996	Added silly rename for unlink	--okir
   9 * 28 Sep 1996	Improved directory cache --okir
  10 * 23 Aug 1997  Claus Heine claus@momo.math.rwth-aachen.de 
  11 *              Re-implemented silly rename for unlink, newly implemented
  12 *              silly rename for nfs_rename() following the suggestions
  13 *              of Olaf Kirch (okir) found in this file.
  14 *              Following Linus comments on my original hack, this version
  15 *              depends only on the dcache stuff and doesn't touch the inode
  16 *              layer (iput() and friends).
  17 *  6 Jun 1999	Cache readdir lookups in the page cache. -DaveM
  18 */
  19
  20#include <linux/module.h>
  21#include <linux/time.h>
  22#include <linux/errno.h>
  23#include <linux/stat.h>
  24#include <linux/fcntl.h>
  25#include <linux/string.h>
  26#include <linux/kernel.h>
  27#include <linux/slab.h>
  28#include <linux/mm.h>
  29#include <linux/sunrpc/clnt.h>
  30#include <linux/nfs_fs.h>
  31#include <linux/nfs_mount.h>
  32#include <linux/pagemap.h>
  33#include <linux/pagevec.h>
  34#include <linux/namei.h>
  35#include <linux/mount.h>
  36#include <linux/swap.h>
  37#include <linux/sched.h>
  38#include <linux/kmemleak.h>
  39#include <linux/xattr.h>
  40
  41#include "delegation.h"
  42#include "iostat.h"
  43#include "internal.h"
  44#include "fscache.h"
  45
  46#include "nfstrace.h"
  47
  48/* #define NFS_DEBUG_VERBOSE 1 */
  49
  50static int nfs_opendir(struct inode *, struct file *);
  51static int nfs_closedir(struct inode *, struct file *);
  52static int nfs_readdir(struct file *, struct dir_context *);
  53static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
  54static loff_t nfs_llseek_dir(struct file *, loff_t, int);
  55static void nfs_readdir_clear_array(struct page*);
  56
  57const struct file_operations nfs_dir_operations = {
  58	.llseek		= nfs_llseek_dir,
  59	.read		= generic_read_dir,
  60	.iterate	= nfs_readdir,
  61	.open		= nfs_opendir,
  62	.release	= nfs_closedir,
  63	.fsync		= nfs_fsync_dir,
  64};
  65
  66const struct address_space_operations nfs_dir_aops = {
  67	.freepage = nfs_readdir_clear_array,
  68};
  69
  70static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
  71{
  72	struct nfs_inode *nfsi = NFS_I(dir);
  73	struct nfs_open_dir_context *ctx;
  74	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
  75	if (ctx != NULL) {
  76		ctx->duped = 0;
  77		ctx->attr_gencount = nfsi->attr_gencount;
  78		ctx->dir_cookie = 0;
  79		ctx->dup_cookie = 0;
  80		ctx->cred = get_rpccred(cred);
  81		spin_lock(&dir->i_lock);
  82		list_add(&ctx->list, &nfsi->open_files);
  83		spin_unlock(&dir->i_lock);
  84		return ctx;
  85	}
  86	return  ERR_PTR(-ENOMEM);
  87}
  88
  89static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
  90{
  91	spin_lock(&dir->i_lock);
  92	list_del(&ctx->list);
  93	spin_unlock(&dir->i_lock);
  94	put_rpccred(ctx->cred);
  95	kfree(ctx);
  96}
  97
  98/*
  99 * Open file
 100 */
 101static int
 102nfs_opendir(struct inode *inode, struct file *filp)
 103{
 104	int res = 0;
 105	struct nfs_open_dir_context *ctx;
 106	struct rpc_cred *cred;
 107
 108	dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
 109
 110	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 111
 112	cred = rpc_lookup_cred();
 113	if (IS_ERR(cred))
 114		return PTR_ERR(cred);
 115	ctx = alloc_nfs_open_dir_context(inode, cred);
 116	if (IS_ERR(ctx)) {
 117		res = PTR_ERR(ctx);
 118		goto out;
 119	}
 120	filp->private_data = ctx;
 
 
 
 
 
 
 
 121out:
 122	put_rpccred(cred);
 123	return res;
 124}
 125
 126static int
 127nfs_closedir(struct inode *inode, struct file *filp)
 128{
 129	put_nfs_open_dir_context(file_inode(filp), filp->private_data);
 130	return 0;
 131}
 132
 133struct nfs_cache_array_entry {
 134	u64 cookie;
 135	u64 ino;
 136	struct qstr string;
 137	unsigned char d_type;
 138};
 139
 140struct nfs_cache_array {
 
 141	int size;
 142	int eof_index;
 143	u64 last_cookie;
 144	struct nfs_cache_array_entry array[0];
 145};
 146
 147typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, bool);
 148typedef struct {
 149	struct file	*file;
 150	struct page	*page;
 151	struct dir_context *ctx;
 152	unsigned long	page_index;
 153	u64		*dir_cookie;
 154	u64		last_cookie;
 155	loff_t		current_index;
 156	decode_dirent_t	decode;
 157
 158	unsigned long	timestamp;
 159	unsigned long	gencount;
 160	unsigned int	cache_entry_index;
 161	bool plus;
 162	bool eof;
 163} nfs_readdir_descriptor_t;
 164
 165/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 166 * we are freeing strings created by nfs_add_to_readdir_array()
 167 */
 168static
 169void nfs_readdir_clear_array(struct page *page)
 170{
 171	struct nfs_cache_array *array;
 172	int i;
 173
 174	array = kmap_atomic(page);
 175	for (i = 0; i < array->size; i++)
 176		kfree(array->array[i].string.name);
 
 
 
 
 
 
 
 
 177	kunmap_atomic(array);
 
 178}
 179
 180/*
 181 * the caller is responsible for freeing qstr.name
 182 * when called by nfs_readdir_add_to_array, the strings will be freed in
 183 * nfs_clear_readdir_array()
 184 */
 185static
 186int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
 187{
 188	string->len = len;
 189	string->name = kmemdup(name, len, GFP_KERNEL);
 190	if (string->name == NULL)
 191		return -ENOMEM;
 192	/*
 193	 * Avoid a kmemleak false positive. The pointer to the name is stored
 194	 * in a page cache page which kmemleak does not scan.
 195	 */
 196	kmemleak_not_leak(string->name);
 197	string->hash = full_name_hash(NULL, name, len);
 198	return 0;
 199}
 200
 201static
 202int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
 203{
 204	struct nfs_cache_array *array = kmap(page);
 205	struct nfs_cache_array_entry *cache_entry;
 206	int ret;
 207
 
 
 
 208	cache_entry = &array->array[array->size];
 209
 210	/* Check that this entry lies within the page bounds */
 211	ret = -ENOSPC;
 212	if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
 213		goto out;
 214
 215	cache_entry->cookie = entry->prev_cookie;
 216	cache_entry->ino = entry->ino;
 217	cache_entry->d_type = entry->d_type;
 218	ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
 219	if (ret)
 220		goto out;
 221	array->last_cookie = entry->cookie;
 222	array->size++;
 223	if (entry->eof != 0)
 224		array->eof_index = array->size;
 225out:
 226	kunmap(page);
 227	return ret;
 228}
 229
 230static
 231int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
 232{
 233	loff_t diff = desc->ctx->pos - desc->current_index;
 234	unsigned int index;
 235
 236	if (diff < 0)
 237		goto out_eof;
 238	if (diff >= array->size) {
 239		if (array->eof_index >= 0)
 240			goto out_eof;
 241		return -EAGAIN;
 242	}
 243
 244	index = (unsigned int)diff;
 245	*desc->dir_cookie = array->array[index].cookie;
 246	desc->cache_entry_index = index;
 247	return 0;
 248out_eof:
 249	desc->eof = true;
 250	return -EBADCOOKIE;
 251}
 252
 253static bool
 254nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
 255{
 256	if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
 257		return false;
 258	smp_rmb();
 259	return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
 260}
 261
 262static
 263int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
 264{
 265	int i;
 266	loff_t new_pos;
 267	int status = -EAGAIN;
 268
 269	for (i = 0; i < array->size; i++) {
 270		if (array->array[i].cookie == *desc->dir_cookie) {
 271			struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
 272			struct nfs_open_dir_context *ctx = desc->file->private_data;
 273
 274			new_pos = desc->current_index + i;
 275			if (ctx->attr_gencount != nfsi->attr_gencount ||
 276			    !nfs_readdir_inode_mapping_valid(nfsi)) {
 277				ctx->duped = 0;
 278				ctx->attr_gencount = nfsi->attr_gencount;
 279			} else if (new_pos < desc->ctx->pos) {
 280				if (ctx->duped > 0
 281				    && ctx->dup_cookie == *desc->dir_cookie) {
 282					if (printk_ratelimit()) {
 283						pr_notice("NFS: directory %pD2 contains a readdir loop."
 284								"Please contact your server vendor.  "
 285								"The file: %.*s has duplicate cookie %llu\n",
 286								desc->file, array->array[i].string.len,
 287								array->array[i].string.name, *desc->dir_cookie);
 288					}
 289					status = -ELOOP;
 290					goto out;
 291				}
 292				ctx->dup_cookie = *desc->dir_cookie;
 293				ctx->duped = -1;
 294			}
 295			desc->ctx->pos = new_pos;
 296			desc->cache_entry_index = i;
 297			return 0;
 298		}
 299	}
 300	if (array->eof_index >= 0) {
 301		status = -EBADCOOKIE;
 302		if (*desc->dir_cookie == array->last_cookie)
 303			desc->eof = true;
 304	}
 305out:
 306	return status;
 307}
 308
 309static
 310int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
 311{
 312	struct nfs_cache_array *array;
 313	int status;
 314
 315	array = kmap(desc->page);
 
 
 
 
 316
 317	if (*desc->dir_cookie == 0)
 318		status = nfs_readdir_search_for_pos(array, desc);
 319	else
 320		status = nfs_readdir_search_for_cookie(array, desc);
 321
 322	if (status == -EAGAIN) {
 323		desc->last_cookie = array->last_cookie;
 324		desc->current_index += array->size;
 325		desc->page_index++;
 326	}
 327	kunmap(desc->page);
 
 328	return status;
 329}
 330
 331/* Fill a page with xdr information before transferring to the cache page */
 332static
 333int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
 334			struct nfs_entry *entry, struct file *file, struct inode *inode)
 335{
 336	struct nfs_open_dir_context *ctx = file->private_data;
 337	struct rpc_cred	*cred = ctx->cred;
 338	unsigned long	timestamp, gencount;
 339	int		error;
 340
 341 again:
 342	timestamp = jiffies;
 343	gencount = nfs_inc_attr_generation_counter();
 344	error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages,
 345					  NFS_SERVER(inode)->dtsize, desc->plus);
 346	if (error < 0) {
 347		/* We requested READDIRPLUS, but the server doesn't grok it */
 348		if (error == -ENOTSUPP && desc->plus) {
 349			NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
 350			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
 351			desc->plus = false;
 352			goto again;
 353		}
 354		goto error;
 355	}
 356	desc->timestamp = timestamp;
 357	desc->gencount = gencount;
 358error:
 359	return error;
 360}
 361
 362static int xdr_decode(nfs_readdir_descriptor_t *desc,
 363		      struct nfs_entry *entry, struct xdr_stream *xdr)
 364{
 365	int error;
 366
 367	error = desc->decode(xdr, entry, desc->plus);
 368	if (error)
 369		return error;
 370	entry->fattr->time_start = desc->timestamp;
 371	entry->fattr->gencount = desc->gencount;
 372	return 0;
 373}
 374
 375/* Match file and dirent using either filehandle or fileid
 376 * Note: caller is responsible for checking the fsid
 377 */
 378static
 379int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
 380{
 381	struct inode *inode;
 382	struct nfs_inode *nfsi;
 383
 384	if (d_really_is_negative(dentry))
 385		return 0;
 386
 387	inode = d_inode(dentry);
 388	if (is_bad_inode(inode) || NFS_STALE(inode))
 389		return 0;
 390
 391	nfsi = NFS_I(inode);
 392	if (entry->fattr->fileid != nfsi->fileid)
 393		return 0;
 394	if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
 395		return 0;
 396	return 1;
 397}
 398
 399static
 400bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
 401{
 402	if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
 403		return false;
 404	if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
 405		return true;
 406	if (ctx->pos == 0)
 407		return true;
 408	return false;
 409}
 410
 411/*
 412 * This function is called by the lookup and getattr code to request the
 413 * use of readdirplus to accelerate any future lookups in the same
 414 * directory.
 415 */
 416void nfs_advise_use_readdirplus(struct inode *dir)
 417{
 418	struct nfs_inode *nfsi = NFS_I(dir);
 419
 420	if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
 421	    !list_empty(&nfsi->open_files))
 422		set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
 423}
 424
 425/*
 426 * This function is mainly for use by nfs_getattr().
 427 *
 428 * If this is an 'ls -l', we want to force use of readdirplus.
 429 * Do this by checking if there is an active file descriptor
 430 * and calling nfs_advise_use_readdirplus, then forcing a
 431 * cache flush.
 432 */
 433void nfs_force_use_readdirplus(struct inode *dir)
 434{
 435	struct nfs_inode *nfsi = NFS_I(dir);
 436
 437	if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
 438	    !list_empty(&nfsi->open_files)) {
 439		set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
 440		invalidate_mapping_pages(dir->i_mapping, 0, -1);
 441	}
 442}
 443
 444static
 445void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
 446{
 447	struct qstr filename = QSTR_INIT(entry->name, entry->len);
 448	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
 449	struct dentry *dentry;
 450	struct dentry *alias;
 451	struct inode *dir = d_inode(parent);
 452	struct inode *inode;
 453	int status;
 454
 455	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
 456		return;
 457	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
 458		return;
 459	if (filename.len == 0)
 460		return;
 461	/* Validate that the name doesn't contain any illegal '\0' */
 462	if (strnlen(filename.name, filename.len) != filename.len)
 463		return;
 464	/* ...or '/' */
 465	if (strnchr(filename.name, filename.len, '/'))
 466		return;
 467	if (filename.name[0] == '.') {
 468		if (filename.len == 1)
 469			return;
 470		if (filename.len == 2 && filename.name[1] == '.')
 471			return;
 472	}
 473	filename.hash = full_name_hash(parent, filename.name, filename.len);
 474
 475	dentry = d_lookup(parent, &filename);
 476again:
 477	if (!dentry) {
 478		dentry = d_alloc_parallel(parent, &filename, &wq);
 479		if (IS_ERR(dentry))
 480			return;
 481	}
 482	if (!d_in_lookup(dentry)) {
 483		/* Is there a mountpoint here? If so, just exit */
 484		if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
 485					&entry->fattr->fsid))
 486			goto out;
 487		if (nfs_same_file(dentry, entry)) {
 488			if (!entry->fh->size)
 489				goto out;
 490			nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
 491			status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
 492			if (!status)
 493				nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
 494			goto out;
 495		} else {
 496			d_invalidate(dentry);
 497			dput(dentry);
 498			dentry = NULL;
 499			goto again;
 500		}
 501	}
 502	if (!entry->fh->size) {
 503		d_lookup_done(dentry);
 504		goto out;
 505	}
 506
 507	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
 508	alias = d_splice_alias(inode, dentry);
 509	d_lookup_done(dentry);
 510	if (alias) {
 511		if (IS_ERR(alias))
 512			goto out;
 513		dput(dentry);
 514		dentry = alias;
 515	}
 516	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
 517out:
 518	dput(dentry);
 519}
 520
 521/* Perform conversion from xdr to cache array */
 522static
 523int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
 524				struct page **xdr_pages, struct page *page, unsigned int buflen)
 525{
 526	struct xdr_stream stream;
 527	struct xdr_buf buf;
 528	struct page *scratch;
 529	struct nfs_cache_array *array;
 530	unsigned int count = 0;
 531	int status;
 532
 533	scratch = alloc_page(GFP_KERNEL);
 534	if (scratch == NULL)
 535		return -ENOMEM;
 536
 537	if (buflen == 0)
 538		goto out_nopages;
 539
 540	xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
 541	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 542
 543	do {
 544		status = xdr_decode(desc, entry, &stream);
 545		if (status != 0) {
 546			if (status == -EAGAIN)
 547				status = 0;
 548			break;
 549		}
 550
 551		count++;
 552
 553		if (desc->plus)
 554			nfs_prime_dcache(file_dentry(desc->file), entry);
 555
 556		status = nfs_readdir_add_to_array(entry, page);
 557		if (status != 0)
 558			break;
 559	} while (!entry->eof);
 560
 561out_nopages:
 562	if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
 563		array = kmap(page);
 564		array->eof_index = array->size;
 565		status = 0;
 566		kunmap(page);
 
 
 
 567	}
 568
 569	put_page(scratch);
 570	return status;
 571}
 572
 573static
 574void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
 575{
 576	unsigned int i;
 577	for (i = 0; i < npages; i++)
 578		put_page(pages[i]);
 579}
 580
 581/*
 582 * nfs_readdir_large_page will allocate pages that must be freed with a call
 583 * to nfs_readdir_free_pagearray
 584 */
 585static
 586int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
 587{
 588	unsigned int i;
 589
 590	for (i = 0; i < npages; i++) {
 591		struct page *page = alloc_page(GFP_KERNEL);
 592		if (page == NULL)
 593			goto out_freepages;
 594		pages[i] = page;
 595	}
 596	return 0;
 597
 598out_freepages:
 599	nfs_readdir_free_pages(pages, i);
 600	return -ENOMEM;
 601}
 602
 603static
 604int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
 605{
 606	struct page *pages[NFS_MAX_READDIR_PAGES];
 607	struct nfs_entry entry;
 608	struct file	*file = desc->file;
 609	struct nfs_cache_array *array;
 610	int status = -ENOMEM;
 611	unsigned int array_size = ARRAY_SIZE(pages);
 612
 613	entry.prev_cookie = 0;
 614	entry.cookie = desc->last_cookie;
 615	entry.eof = 0;
 616	entry.fh = nfs_alloc_fhandle();
 617	entry.fattr = nfs_alloc_fattr();
 618	entry.server = NFS_SERVER(inode);
 619	if (entry.fh == NULL || entry.fattr == NULL)
 620		goto out;
 621
 622	entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
 623	if (IS_ERR(entry.label)) {
 624		status = PTR_ERR(entry.label);
 625		goto out;
 626	}
 627
 628	array = kmap(page);
 
 
 
 
 629	memset(array, 0, sizeof(struct nfs_cache_array));
 
 630	array->eof_index = -1;
 631
 632	status = nfs_readdir_alloc_pages(pages, array_size);
 633	if (status < 0)
 634		goto out_release_array;
 635	do {
 636		unsigned int pglen;
 637		status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
 638
 639		if (status < 0)
 640			break;
 641		pglen = status;
 642		status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
 643		if (status < 0) {
 644			if (status == -ENOSPC)
 645				status = 0;
 646			break;
 647		}
 648	} while (array->eof_index < 0);
 649
 650	nfs_readdir_free_pages(pages, array_size);
 651out_release_array:
 652	kunmap(page);
 
 653	nfs4_label_free(entry.label);
 654out:
 655	nfs_free_fattr(entry.fattr);
 656	nfs_free_fhandle(entry.fh);
 657	return status;
 658}
 659
 660/*
 661 * Now we cache directories properly, by converting xdr information
 662 * to an array that can be used for lookups later.  This results in
 663 * fewer cache pages, since we can store more information on each page.
 664 * We only need to convert from xdr once so future lookups are much simpler
 665 */
 666static
 667int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
 668{
 669	struct inode	*inode = file_inode(desc->file);
 670	int ret;
 671
 672	ret = nfs_readdir_xdr_to_array(desc, page, inode);
 673	if (ret < 0)
 674		goto error;
 675	SetPageUptodate(page);
 676
 677	if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
 678		/* Should never happen */
 679		nfs_zap_mapping(inode, inode->i_mapping);
 680	}
 681	unlock_page(page);
 682	return 0;
 683 error:
 684	unlock_page(page);
 685	return ret;
 686}
 687
 688static
 689void cache_page_release(nfs_readdir_descriptor_t *desc)
 690{
 691	if (!desc->page->mapping)
 692		nfs_readdir_clear_array(desc->page);
 693	put_page(desc->page);
 694	desc->page = NULL;
 695}
 696
 697static
 698struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
 699{
 700	return read_cache_page(desc->file->f_mapping,
 
 
 
 701			desc->page_index, (filler_t *)nfs_readdir_filler, desc);
 
 
 
 
 
 702}
 703
 704/*
 705 * Returns 0 if desc->dir_cookie was found on page desc->page_index
 706 */
 707static
 708int find_cache_page(nfs_readdir_descriptor_t *desc)
 709{
 710	int res;
 711
 712	desc->page = get_cache_page(desc);
 713	if (IS_ERR(desc->page))
 714		return PTR_ERR(desc->page);
 715
 716	res = nfs_readdir_search_array(desc);
 717	if (res != 0)
 718		cache_page_release(desc);
 719	return res;
 720}
 721
 722/* Search for desc->dir_cookie from the beginning of the page cache */
 723static inline
 724int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
 725{
 726	int res;
 727
 728	if (desc->page_index == 0) {
 729		desc->current_index = 0;
 730		desc->last_cookie = 0;
 731	}
 732	do {
 733		res = find_cache_page(desc);
 734	} while (res == -EAGAIN);
 735	return res;
 736}
 737
 738/*
 739 * Once we've found the start of the dirent within a page: fill 'er up...
 740 */
 741static 
 742int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
 743{
 744	struct file	*file = desc->file;
 745	int i = 0;
 746	int res = 0;
 747	struct nfs_cache_array *array = NULL;
 748	struct nfs_open_dir_context *ctx = file->private_data;
 749
 750	array = kmap(desc->page);
 
 
 
 
 
 751	for (i = desc->cache_entry_index; i < array->size; i++) {
 752		struct nfs_cache_array_entry *ent;
 753
 754		ent = &array->array[i];
 755		if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
 756		    nfs_compat_user_ino64(ent->ino), ent->d_type)) {
 757			desc->eof = true;
 758			break;
 759		}
 760		desc->ctx->pos++;
 761		if (i < (array->size-1))
 762			*desc->dir_cookie = array->array[i+1].cookie;
 763		else
 764			*desc->dir_cookie = array->last_cookie;
 765		if (ctx->duped != 0)
 766			ctx->duped = 1;
 767	}
 768	if (array->eof_index >= 0)
 769		desc->eof = true;
 770
 771	kunmap(desc->page);
 
 772	cache_page_release(desc);
 773	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
 774			(unsigned long long)*desc->dir_cookie, res);
 775	return res;
 776}
 777
 778/*
 779 * If we cannot find a cookie in our cache, we suspect that this is
 780 * because it points to a deleted file, so we ask the server to return
 781 * whatever it thinks is the next entry. We then feed this to filldir.
 782 * If all goes well, we should then be able to find our way round the
 783 * cache on the next call to readdir_search_pagecache();
 784 *
 785 * NOTE: we cannot add the anonymous page to the pagecache because
 786 *	 the data it contains might not be page aligned. Besides,
 787 *	 we should already have a complete representation of the
 788 *	 directory in the page cache by the time we get here.
 789 */
 790static inline
 791int uncached_readdir(nfs_readdir_descriptor_t *desc)
 792{
 793	struct page	*page = NULL;
 794	int		status;
 795	struct inode *inode = file_inode(desc->file);
 796	struct nfs_open_dir_context *ctx = desc->file->private_data;
 797
 798	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
 799			(unsigned long long)*desc->dir_cookie);
 800
 801	page = alloc_page(GFP_HIGHUSER);
 802	if (!page) {
 803		status = -ENOMEM;
 804		goto out;
 805	}
 806
 807	desc->page_index = 0;
 808	desc->last_cookie = *desc->dir_cookie;
 809	desc->page = page;
 810	ctx->duped = 0;
 811
 812	status = nfs_readdir_xdr_to_array(desc, page, inode);
 813	if (status < 0)
 814		goto out_release;
 815
 816	status = nfs_do_filldir(desc);
 817
 818 out:
 819	dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
 820			__func__, status);
 821	return status;
 822 out_release:
 823	cache_page_release(desc);
 824	goto out;
 825}
 826
 827/* The file offset position represents the dirent entry number.  A
 828   last cookie cache takes care of the common case of reading the
 829   whole directory.
 830 */
 831static int nfs_readdir(struct file *file, struct dir_context *ctx)
 832{
 833	struct dentry	*dentry = file_dentry(file);
 834	struct inode	*inode = d_inode(dentry);
 835	nfs_readdir_descriptor_t my_desc,
 836			*desc = &my_desc;
 837	struct nfs_open_dir_context *dir_ctx = file->private_data;
 838	int res = 0;
 839
 840	dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
 841			file, (long long)ctx->pos);
 842	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
 843
 844	/*
 845	 * ctx->pos points to the dirent entry number.
 846	 * *desc->dir_cookie has the cookie for the next entry. We have
 847	 * to either find the entry with the appropriate number or
 848	 * revalidate the cookie.
 849	 */
 850	memset(desc, 0, sizeof(*desc));
 851
 852	desc->file = file;
 853	desc->ctx = ctx;
 854	desc->dir_cookie = &dir_ctx->dir_cookie;
 855	desc->decode = NFS_PROTO(inode)->decode_dirent;
 856	desc->plus = nfs_use_readdirplus(inode, ctx);
 857
 858	if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
 859		res = nfs_revalidate_mapping(inode, file->f_mapping);
 860	if (res < 0)
 861		goto out;
 862
 863	do {
 864		res = readdir_search_pagecache(desc);
 865
 866		if (res == -EBADCOOKIE) {
 867			res = 0;
 868			/* This means either end of directory */
 869			if (*desc->dir_cookie && !desc->eof) {
 870				/* Or that the server has 'lost' a cookie */
 871				res = uncached_readdir(desc);
 872				if (res == 0)
 873					continue;
 874			}
 875			break;
 876		}
 877		if (res == -ETOOSMALL && desc->plus) {
 878			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
 879			nfs_zap_caches(inode);
 880			desc->page_index = 0;
 881			desc->plus = false;
 882			desc->eof = false;
 883			continue;
 884		}
 885		if (res < 0)
 886			break;
 887
 888		res = nfs_do_filldir(desc);
 889		if (res < 0)
 890			break;
 891	} while (!desc->eof);
 892out:
 893	if (res > 0)
 894		res = 0;
 895	dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
 896	return res;
 897}
 898
 899static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
 900{
 901	struct inode *inode = file_inode(filp);
 902	struct nfs_open_dir_context *dir_ctx = filp->private_data;
 903
 904	dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
 905			filp, offset, whence);
 906
 907	inode_lock(inode);
 908	switch (whence) {
 909		case 1:
 910			offset += filp->f_pos;
 911		case 0:
 912			if (offset >= 0)
 913				break;
 914		default:
 915			offset = -EINVAL;
 916			goto out;
 917	}
 918	if (offset != filp->f_pos) {
 919		filp->f_pos = offset;
 920		dir_ctx->dir_cookie = 0;
 921		dir_ctx->duped = 0;
 922	}
 923out:
 924	inode_unlock(inode);
 925	return offset;
 926}
 927
 928/*
 929 * All directory operations under NFS are synchronous, so fsync()
 930 * is a dummy operation.
 931 */
 932static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
 933			 int datasync)
 934{
 935	struct inode *inode = file_inode(filp);
 936
 937	dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
 938
 939	inode_lock(inode);
 940	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
 941	inode_unlock(inode);
 942	return 0;
 943}
 944
 945/**
 946 * nfs_force_lookup_revalidate - Mark the directory as having changed
 947 * @dir - pointer to directory inode
 948 *
 949 * This forces the revalidation code in nfs_lookup_revalidate() to do a
 950 * full lookup on all child dentries of 'dir' whenever a change occurs
 951 * on the server that might have invalidated our dcache.
 952 *
 953 * The caller should be holding dir->i_lock
 954 */
 955void nfs_force_lookup_revalidate(struct inode *dir)
 956{
 957	NFS_I(dir)->cache_change_attribute++;
 958}
 959EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
 960
 961/*
 962 * A check for whether or not the parent directory has changed.
 963 * In the case it has, we assume that the dentries are untrustworthy
 964 * and may need to be looked up again.
 965 * If rcu_walk prevents us from performing a full check, return 0.
 966 */
 967static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
 968			      int rcu_walk)
 969{
 970	if (IS_ROOT(dentry))
 971		return 1;
 972	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
 973		return 0;
 974	if (!nfs_verify_change_attribute(dir, dentry->d_time))
 975		return 0;
 976	/* Revalidate nfsi->cache_change_attribute before we declare a match */
 977	if (nfs_mapping_need_revalidate_inode(dir)) {
 978		if (rcu_walk)
 979			return 0;
 980		if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
 981			return 0;
 982	}
 983	if (!nfs_verify_change_attribute(dir, dentry->d_time))
 984		return 0;
 985	return 1;
 986}
 987
 988/*
 989 * Use intent information to check whether or not we're going to do
 990 * an O_EXCL create using this path component.
 991 */
 992static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
 993{
 994	if (NFS_PROTO(dir)->version == 2)
 995		return 0;
 996	return flags & LOOKUP_EXCL;
 997}
 998
 999/*
1000 * Inode and filehandle revalidation for lookups.
1001 *
1002 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1003 * or if the intent information indicates that we're about to open this
1004 * particular file and the "nocto" mount flag is not set.
1005 *
1006 */
1007static
1008int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1009{
1010	struct nfs_server *server = NFS_SERVER(inode);
1011	int ret;
1012
1013	if (IS_AUTOMOUNT(inode))
1014		return 0;
1015	/* VFS wants an on-the-wire revalidation */
1016	if (flags & LOOKUP_REVAL)
1017		goto out_force;
1018	/* This is an open(2) */
1019	if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1020	    (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1021		goto out_force;
1022out:
1023	return (inode->i_nlink == 0) ? -ENOENT : 0;
1024out_force:
1025	if (flags & LOOKUP_RCU)
1026		return -ECHILD;
1027	ret = __nfs_revalidate_inode(server, inode);
1028	if (ret != 0)
1029		return ret;
1030	goto out;
1031}
1032
1033/*
1034 * We judge how long we want to trust negative
1035 * dentries by looking at the parent inode mtime.
1036 *
1037 * If parent mtime has changed, we revalidate, else we wait for a
1038 * period corresponding to the parent's attribute cache timeout value.
1039 *
1040 * If LOOKUP_RCU prevents us from performing a full check, return 1
1041 * suggesting a reval is needed.
1042 */
1043static inline
1044int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1045		       unsigned int flags)
1046{
1047	/* Don't revalidate a negative dentry if we're creating a new file */
1048	if (flags & LOOKUP_CREATE)
1049		return 0;
1050	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1051		return 1;
1052	return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1053}
1054
1055/*
1056 * This is called every time the dcache has a lookup hit,
1057 * and we should check whether we can really trust that
1058 * lookup.
1059 *
1060 * NOTE! The hit can be a negative hit too, don't assume
1061 * we have an inode!
1062 *
1063 * If the parent directory is seen to have changed, we throw out the
1064 * cached dentry and do a new lookup.
1065 */
1066static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1067{
1068	struct inode *dir;
1069	struct inode *inode;
1070	struct dentry *parent;
1071	struct nfs_fh *fhandle = NULL;
1072	struct nfs_fattr *fattr = NULL;
1073	struct nfs4_label *label = NULL;
1074	int error;
1075
1076	if (flags & LOOKUP_RCU) {
1077		parent = READ_ONCE(dentry->d_parent);
1078		dir = d_inode_rcu(parent);
1079		if (!dir)
1080			return -ECHILD;
1081	} else {
1082		parent = dget_parent(dentry);
1083		dir = d_inode(parent);
1084	}
1085	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1086	inode = d_inode(dentry);
1087
1088	if (!inode) {
1089		if (nfs_neg_need_reval(dir, dentry, flags)) {
1090			if (flags & LOOKUP_RCU)
1091				return -ECHILD;
1092			goto out_bad;
1093		}
1094		goto out_valid;
1095	}
1096
1097	if (is_bad_inode(inode)) {
1098		if (flags & LOOKUP_RCU)
1099			return -ECHILD;
1100		dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1101				__func__, dentry);
1102		goto out_bad;
1103	}
1104
1105	if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1106		goto out_set_verifier;
1107
1108	/* Force a full look up iff the parent directory has changed */
1109	if (!nfs_is_exclusive_create(dir, flags) &&
1110	    nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1111		error = nfs_lookup_verify_inode(inode, flags);
1112		if (error) {
1113			if (flags & LOOKUP_RCU)
1114				return -ECHILD;
1115			if (error == -ESTALE)
1116				goto out_zap_parent;
1117			goto out_error;
1118		}
1119		nfs_advise_use_readdirplus(dir);
1120		goto out_valid;
1121	}
1122
1123	if (flags & LOOKUP_RCU)
1124		return -ECHILD;
1125
1126	if (NFS_STALE(inode))
1127		goto out_bad;
1128
1129	error = -ENOMEM;
1130	fhandle = nfs_alloc_fhandle();
1131	fattr = nfs_alloc_fattr();
1132	if (fhandle == NULL || fattr == NULL)
1133		goto out_error;
1134
1135	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1136	if (IS_ERR(label))
1137		goto out_error;
1138
1139	trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1140	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1141	trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1142	if (error == -ESTALE || error == -ENOENT)
1143		goto out_bad;
1144	if (error)
1145		goto out_error;
1146	if (nfs_compare_fh(NFS_FH(inode), fhandle))
1147		goto out_bad;
1148	if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1149		goto out_bad;
1150
1151	nfs_setsecurity(inode, fattr, label);
1152
1153	nfs_free_fattr(fattr);
1154	nfs_free_fhandle(fhandle);
1155	nfs4_label_free(label);
1156
1157	/* set a readdirplus hint that we had a cache miss */
1158	nfs_force_use_readdirplus(dir);
1159
1160out_set_verifier:
1161	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1162 out_valid:
1163	if (flags & LOOKUP_RCU) {
1164		if (parent != READ_ONCE(dentry->d_parent))
1165			return -ECHILD;
1166	} else
1167		dput(parent);
1168	dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1169			__func__, dentry);
1170	return 1;
1171out_zap_parent:
1172	nfs_zap_caches(dir);
1173 out_bad:
1174	WARN_ON(flags & LOOKUP_RCU);
1175	nfs_free_fattr(fattr);
1176	nfs_free_fhandle(fhandle);
1177	nfs4_label_free(label);
1178	nfs_mark_for_revalidate(dir);
1179	if (inode && S_ISDIR(inode->i_mode)) {
1180		/* Purge readdir caches. */
1181		nfs_zap_caches(inode);
1182		/*
1183		 * We can't d_drop the root of a disconnected tree:
1184		 * its d_hash is on the s_anon list and d_drop() would hide
1185		 * it from shrink_dcache_for_unmount(), leading to busy
1186		 * inodes on unmount and further oopses.
1187		 */
1188		if (IS_ROOT(dentry))
1189			goto out_valid;
1190	}
1191	dput(parent);
1192	dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1193			__func__, dentry);
1194	return 0;
1195out_error:
1196	WARN_ON(flags & LOOKUP_RCU);
1197	nfs_free_fattr(fattr);
1198	nfs_free_fhandle(fhandle);
1199	nfs4_label_free(label);
1200	dput(parent);
1201	dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1202			__func__, dentry, error);
1203	return error;
1204}
1205
1206/*
1207 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1208 * when we don't really care about the dentry name. This is called when a
1209 * pathwalk ends on a dentry that was not found via a normal lookup in the
1210 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1211 *
1212 * In this situation, we just want to verify that the inode itself is OK
1213 * since the dentry might have changed on the server.
1214 */
1215static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1216{
1217	struct inode *inode = d_inode(dentry);
1218	int error = 0;
1219
1220	/*
1221	 * I believe we can only get a negative dentry here in the case of a
1222	 * procfs-style symlink. Just assume it's correct for now, but we may
1223	 * eventually need to do something more here.
1224	 */
1225	if (!inode) {
1226		dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1227				__func__, dentry);
1228		return 1;
1229	}
1230
1231	if (is_bad_inode(inode)) {
1232		dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1233				__func__, dentry);
1234		return 0;
1235	}
1236
1237	error = nfs_lookup_verify_inode(inode, flags);
 
1238	dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1239			__func__, inode->i_ino, error ? "invalid" : "valid");
1240	return !error;
1241}
1242
1243/*
1244 * This is called from dput() when d_count is going to 0.
1245 */
1246static int nfs_dentry_delete(const struct dentry *dentry)
1247{
1248	dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1249		dentry, dentry->d_flags);
1250
1251	/* Unhash any dentry with a stale inode */
1252	if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1253		return 1;
1254
1255	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1256		/* Unhash it, so that ->d_iput() would be called */
1257		return 1;
1258	}
1259	if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1260		/* Unhash it, so that ancestors of killed async unlink
1261		 * files will be cleaned up during umount */
1262		return 1;
1263	}
1264	return 0;
1265
1266}
1267
1268/* Ensure that we revalidate inode->i_nlink */
1269static void nfs_drop_nlink(struct inode *inode)
1270{
1271	spin_lock(&inode->i_lock);
1272	/* drop the inode if we're reasonably sure this is the last link */
1273	if (inode->i_nlink == 1)
1274		clear_nlink(inode);
1275	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE
1276		| NFS_INO_INVALID_CTIME
1277		| NFS_INO_INVALID_OTHER;
1278	spin_unlock(&inode->i_lock);
1279}
1280
1281/*
1282 * Called when the dentry loses inode.
1283 * We use it to clean up silly-renamed files.
1284 */
1285static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1286{
1287	if (S_ISDIR(inode->i_mode))
1288		/* drop any readdir cache as it could easily be old */
1289		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1290
1291	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1292		nfs_complete_unlink(dentry, inode);
1293		nfs_drop_nlink(inode);
1294	}
1295	iput(inode);
1296}
1297
1298static void nfs_d_release(struct dentry *dentry)
1299{
1300	/* free cached devname value, if it survived that far */
1301	if (unlikely(dentry->d_fsdata)) {
1302		if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1303			WARN_ON(1);
1304		else
1305			kfree(dentry->d_fsdata);
1306	}
1307}
1308
1309const struct dentry_operations nfs_dentry_operations = {
1310	.d_revalidate	= nfs_lookup_revalidate,
1311	.d_weak_revalidate	= nfs_weak_revalidate,
1312	.d_delete	= nfs_dentry_delete,
1313	.d_iput		= nfs_dentry_iput,
1314	.d_automount	= nfs_d_automount,
1315	.d_release	= nfs_d_release,
1316};
1317EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1318
1319struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1320{
1321	struct dentry *res;
1322	struct inode *inode = NULL;
1323	struct nfs_fh *fhandle = NULL;
1324	struct nfs_fattr *fattr = NULL;
1325	struct nfs4_label *label = NULL;
1326	int error;
1327
1328	dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1329	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1330
1331	if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1332		return ERR_PTR(-ENAMETOOLONG);
1333
1334	/*
1335	 * If we're doing an exclusive create, optimize away the lookup
1336	 * but don't hash the dentry.
1337	 */
1338	if (nfs_is_exclusive_create(dir, flags))
1339		return NULL;
1340
1341	res = ERR_PTR(-ENOMEM);
1342	fhandle = nfs_alloc_fhandle();
1343	fattr = nfs_alloc_fattr();
1344	if (fhandle == NULL || fattr == NULL)
1345		goto out;
1346
1347	label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1348	if (IS_ERR(label))
1349		goto out;
1350
1351	trace_nfs_lookup_enter(dir, dentry, flags);
1352	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1353	if (error == -ENOENT)
1354		goto no_entry;
1355	if (error < 0) {
1356		res = ERR_PTR(error);
1357		goto out_label;
1358	}
1359	inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1360	res = ERR_CAST(inode);
1361	if (IS_ERR(res))
1362		goto out_label;
1363
1364	/* Notify readdir to use READDIRPLUS */
1365	nfs_force_use_readdirplus(dir);
1366
1367no_entry:
1368	res = d_splice_alias(inode, dentry);
1369	if (res != NULL) {
1370		if (IS_ERR(res))
1371			goto out_label;
1372		dentry = res;
1373	}
1374	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1375out_label:
1376	trace_nfs_lookup_exit(dir, dentry, flags, error);
1377	nfs4_label_free(label);
1378out:
1379	nfs_free_fattr(fattr);
1380	nfs_free_fhandle(fhandle);
1381	return res;
1382}
1383EXPORT_SYMBOL_GPL(nfs_lookup);
1384
1385#if IS_ENABLED(CONFIG_NFS_V4)
1386static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1387
1388const struct dentry_operations nfs4_dentry_operations = {
1389	.d_revalidate	= nfs4_lookup_revalidate,
1390	.d_weak_revalidate	= nfs_weak_revalidate,
1391	.d_delete	= nfs_dentry_delete,
1392	.d_iput		= nfs_dentry_iput,
1393	.d_automount	= nfs_d_automount,
1394	.d_release	= nfs_d_release,
1395};
1396EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1397
1398static fmode_t flags_to_mode(int flags)
1399{
1400	fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1401	if ((flags & O_ACCMODE) != O_WRONLY)
1402		res |= FMODE_READ;
1403	if ((flags & O_ACCMODE) != O_RDONLY)
1404		res |= FMODE_WRITE;
1405	return res;
1406}
1407
1408static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1409{
1410	return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1411}
1412
1413static int do_open(struct inode *inode, struct file *filp)
1414{
1415	nfs_fscache_open_file(inode, filp);
1416	return 0;
1417}
1418
1419static int nfs_finish_open(struct nfs_open_context *ctx,
1420			   struct dentry *dentry,
1421			   struct file *file, unsigned open_flags,
1422			   int *opened)
1423{
1424	int err;
1425
1426	err = finish_open(file, dentry, do_open, opened);
1427	if (err)
1428		goto out;
1429	if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1430		nfs_file_set_open_context(file, ctx);
1431	else
1432		err = -ESTALE;
1433out:
1434	return err;
1435}
1436
1437int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1438		    struct file *file, unsigned open_flags,
1439		    umode_t mode, int *opened)
1440{
1441	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1442	struct nfs_open_context *ctx;
1443	struct dentry *res;
1444	struct iattr attr = { .ia_valid = ATTR_OPEN };
1445	struct inode *inode;
1446	unsigned int lookup_flags = 0;
1447	bool switched = false;
1448	int err;
1449
1450	/* Expect a negative dentry */
1451	BUG_ON(d_inode(dentry));
1452
1453	dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1454			dir->i_sb->s_id, dir->i_ino, dentry);
1455
1456	err = nfs_check_flags(open_flags);
1457	if (err)
1458		return err;
1459
1460	/* NFS only supports OPEN on regular files */
1461	if ((open_flags & O_DIRECTORY)) {
1462		if (!d_in_lookup(dentry)) {
1463			/*
1464			 * Hashed negative dentry with O_DIRECTORY: dentry was
1465			 * revalidated and is fine, no need to perform lookup
1466			 * again
1467			 */
1468			return -ENOENT;
1469		}
1470		lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1471		goto no_open;
1472	}
1473
1474	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1475		return -ENAMETOOLONG;
1476
1477	if (open_flags & O_CREAT) {
1478		struct nfs_server *server = NFS_SERVER(dir);
1479
1480		if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1481			mode &= ~current_umask();
1482
1483		attr.ia_valid |= ATTR_MODE;
1484		attr.ia_mode = mode;
1485	}
1486	if (open_flags & O_TRUNC) {
1487		attr.ia_valid |= ATTR_SIZE;
1488		attr.ia_size = 0;
1489	}
1490
1491	if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1492		d_drop(dentry);
1493		switched = true;
1494		dentry = d_alloc_parallel(dentry->d_parent,
1495					  &dentry->d_name, &wq);
1496		if (IS_ERR(dentry))
1497			return PTR_ERR(dentry);
1498		if (unlikely(!d_in_lookup(dentry)))
1499			return finish_no_open(file, dentry);
1500	}
1501
1502	ctx = create_nfs_open_context(dentry, open_flags, file);
1503	err = PTR_ERR(ctx);
1504	if (IS_ERR(ctx))
1505		goto out;
1506
1507	trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1508	inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
1509	if (IS_ERR(inode)) {
1510		err = PTR_ERR(inode);
1511		trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1512		put_nfs_open_context(ctx);
1513		d_drop(dentry);
1514		switch (err) {
1515		case -ENOENT:
1516			d_splice_alias(NULL, dentry);
1517			nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1518			break;
1519		case -EISDIR:
1520		case -ENOTDIR:
1521			goto no_open;
1522		case -ELOOP:
1523			if (!(open_flags & O_NOFOLLOW))
1524				goto no_open;
1525			break;
1526			/* case -EINVAL: */
1527		default:
1528			break;
1529		}
1530		goto out;
1531	}
1532
1533	err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1534	trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1535	put_nfs_open_context(ctx);
1536out:
1537	if (unlikely(switched)) {
1538		d_lookup_done(dentry);
1539		dput(dentry);
1540	}
1541	return err;
1542
1543no_open:
1544	res = nfs_lookup(dir, dentry, lookup_flags);
1545	if (switched) {
1546		d_lookup_done(dentry);
1547		if (!res)
1548			res = dentry;
1549		else
1550			dput(dentry);
1551	}
1552	if (IS_ERR(res))
1553		return PTR_ERR(res);
1554	return finish_no_open(file, res);
1555}
1556EXPORT_SYMBOL_GPL(nfs_atomic_open);
1557
1558static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1559{
1560	struct inode *inode;
1561	int ret = 0;
1562
1563	if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1564		goto no_open;
1565	if (d_mountpoint(dentry))
1566		goto no_open;
1567	if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1568		goto no_open;
1569
1570	inode = d_inode(dentry);
1571
1572	/* We can't create new files in nfs_open_revalidate(), so we
1573	 * optimize away revalidation of negative dentries.
1574	 */
1575	if (inode == NULL) {
1576		struct dentry *parent;
1577		struct inode *dir;
1578
1579		if (flags & LOOKUP_RCU) {
1580			parent = READ_ONCE(dentry->d_parent);
1581			dir = d_inode_rcu(parent);
1582			if (!dir)
1583				return -ECHILD;
1584		} else {
1585			parent = dget_parent(dentry);
1586			dir = d_inode(parent);
1587		}
1588		if (!nfs_neg_need_reval(dir, dentry, flags))
1589			ret = 1;
1590		else if (flags & LOOKUP_RCU)
1591			ret = -ECHILD;
1592		if (!(flags & LOOKUP_RCU))
1593			dput(parent);
1594		else if (parent != READ_ONCE(dentry->d_parent))
1595			return -ECHILD;
1596		goto out;
1597	}
1598
1599	/* NFS only supports OPEN on regular files */
1600	if (!S_ISREG(inode->i_mode))
1601		goto no_open;
1602	/* We cannot do exclusive creation on a positive dentry */
1603	if (flags & LOOKUP_EXCL)
1604		goto no_open;
1605
1606	/* Let f_op->open() actually open (and revalidate) the file */
1607	ret = 1;
1608
1609out:
1610	return ret;
1611
1612no_open:
1613	return nfs_lookup_revalidate(dentry, flags);
1614}
1615
1616#endif /* CONFIG_NFSV4 */
1617
1618/*
1619 * Code common to create, mkdir, and mknod.
1620 */
1621int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1622				struct nfs_fattr *fattr,
1623				struct nfs4_label *label)
1624{
1625	struct dentry *parent = dget_parent(dentry);
1626	struct inode *dir = d_inode(parent);
1627	struct inode *inode;
1628	int error = -EACCES;
1629
1630	d_drop(dentry);
1631
1632	/* We may have been initialized further down */
1633	if (d_really_is_positive(dentry))
1634		goto out;
1635	if (fhandle->size == 0) {
1636		error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1637		if (error)
1638			goto out_error;
1639	}
1640	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1641	if (!(fattr->valid & NFS_ATTR_FATTR)) {
1642		struct nfs_server *server = NFS_SB(dentry->d_sb);
1643		error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1644		if (error < 0)
1645			goto out_error;
1646	}
1647	inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1648	error = PTR_ERR(inode);
1649	if (IS_ERR(inode))
1650		goto out_error;
1651	d_add(dentry, inode);
1652out:
1653	dput(parent);
1654	return 0;
1655out_error:
1656	nfs_mark_for_revalidate(dir);
1657	dput(parent);
1658	return error;
1659}
1660EXPORT_SYMBOL_GPL(nfs_instantiate);
1661
1662/*
1663 * Following a failed create operation, we drop the dentry rather
1664 * than retain a negative dentry. This avoids a problem in the event
1665 * that the operation succeeded on the server, but an error in the
1666 * reply path made it appear to have failed.
1667 */
1668int nfs_create(struct inode *dir, struct dentry *dentry,
1669		umode_t mode, bool excl)
1670{
1671	struct iattr attr;
1672	int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1673	int error;
1674
1675	dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1676			dir->i_sb->s_id, dir->i_ino, dentry);
1677
1678	attr.ia_mode = mode;
1679	attr.ia_valid = ATTR_MODE;
1680
1681	trace_nfs_create_enter(dir, dentry, open_flags);
1682	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1683	trace_nfs_create_exit(dir, dentry, open_flags, error);
1684	if (error != 0)
1685		goto out_err;
1686	return 0;
1687out_err:
1688	d_drop(dentry);
1689	return error;
1690}
1691EXPORT_SYMBOL_GPL(nfs_create);
1692
1693/*
1694 * See comments for nfs_proc_create regarding failed operations.
1695 */
1696int
1697nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1698{
1699	struct iattr attr;
1700	int status;
1701
1702	dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1703			dir->i_sb->s_id, dir->i_ino, dentry);
1704
1705	attr.ia_mode = mode;
1706	attr.ia_valid = ATTR_MODE;
1707
1708	trace_nfs_mknod_enter(dir, dentry);
1709	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1710	trace_nfs_mknod_exit(dir, dentry, status);
1711	if (status != 0)
1712		goto out_err;
1713	return 0;
1714out_err:
1715	d_drop(dentry);
1716	return status;
1717}
1718EXPORT_SYMBOL_GPL(nfs_mknod);
1719
1720/*
1721 * See comments for nfs_proc_create regarding failed operations.
1722 */
1723int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1724{
1725	struct iattr attr;
1726	int error;
1727
1728	dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1729			dir->i_sb->s_id, dir->i_ino, dentry);
1730
1731	attr.ia_valid = ATTR_MODE;
1732	attr.ia_mode = mode | S_IFDIR;
1733
1734	trace_nfs_mkdir_enter(dir, dentry);
1735	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1736	trace_nfs_mkdir_exit(dir, dentry, error);
1737	if (error != 0)
1738		goto out_err;
1739	return 0;
1740out_err:
1741	d_drop(dentry);
1742	return error;
1743}
1744EXPORT_SYMBOL_GPL(nfs_mkdir);
1745
1746static void nfs_dentry_handle_enoent(struct dentry *dentry)
1747{
1748	if (simple_positive(dentry))
1749		d_delete(dentry);
1750}
1751
1752int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1753{
1754	int error;
1755
1756	dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1757			dir->i_sb->s_id, dir->i_ino, dentry);
1758
1759	trace_nfs_rmdir_enter(dir, dentry);
1760	if (d_really_is_positive(dentry)) {
1761		down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1762		error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1763		/* Ensure the VFS deletes this inode */
1764		switch (error) {
1765		case 0:
1766			clear_nlink(d_inode(dentry));
1767			break;
1768		case -ENOENT:
1769			nfs_dentry_handle_enoent(dentry);
1770		}
1771		up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1772	} else
1773		error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1774	trace_nfs_rmdir_exit(dir, dentry, error);
1775
1776	return error;
1777}
1778EXPORT_SYMBOL_GPL(nfs_rmdir);
1779
1780/*
1781 * Remove a file after making sure there are no pending writes,
1782 * and after checking that the file has only one user. 
1783 *
1784 * We invalidate the attribute cache and free the inode prior to the operation
1785 * to avoid possible races if the server reuses the inode.
1786 */
1787static int nfs_safe_remove(struct dentry *dentry)
1788{
1789	struct inode *dir = d_inode(dentry->d_parent);
1790	struct inode *inode = d_inode(dentry);
1791	int error = -EBUSY;
1792		
1793	dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1794
1795	/* If the dentry was sillyrenamed, we simply call d_delete() */
1796	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1797		error = 0;
1798		goto out;
1799	}
1800
1801	trace_nfs_remove_enter(dir, dentry);
1802	if (inode != NULL) {
1803		error = NFS_PROTO(dir)->remove(dir, dentry);
 
1804		if (error == 0)
1805			nfs_drop_nlink(inode);
1806	} else
1807		error = NFS_PROTO(dir)->remove(dir, dentry);
1808	if (error == -ENOENT)
1809		nfs_dentry_handle_enoent(dentry);
1810	trace_nfs_remove_exit(dir, dentry, error);
1811out:
1812	return error;
1813}
1814
1815/*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1816 *  belongs to an active ".nfs..." file and we return -EBUSY.
1817 *
1818 *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1819 */
1820int nfs_unlink(struct inode *dir, struct dentry *dentry)
1821{
1822	int error;
1823	int need_rehash = 0;
1824
1825	dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1826		dir->i_ino, dentry);
1827
1828	trace_nfs_unlink_enter(dir, dentry);
1829	spin_lock(&dentry->d_lock);
1830	if (d_count(dentry) > 1) {
1831		spin_unlock(&dentry->d_lock);
1832		/* Start asynchronous writeout of the inode */
1833		write_inode_now(d_inode(dentry), 0);
1834		error = nfs_sillyrename(dir, dentry);
1835		goto out;
1836	}
1837	if (!d_unhashed(dentry)) {
1838		__d_drop(dentry);
1839		need_rehash = 1;
1840	}
1841	spin_unlock(&dentry->d_lock);
1842	error = nfs_safe_remove(dentry);
1843	if (!error || error == -ENOENT) {
1844		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1845	} else if (need_rehash)
1846		d_rehash(dentry);
1847out:
1848	trace_nfs_unlink_exit(dir, dentry, error);
1849	return error;
1850}
1851EXPORT_SYMBOL_GPL(nfs_unlink);
1852
1853/*
1854 * To create a symbolic link, most file systems instantiate a new inode,
1855 * add a page to it containing the path, then write it out to the disk
1856 * using prepare_write/commit_write.
1857 *
1858 * Unfortunately the NFS client can't create the in-core inode first
1859 * because it needs a file handle to create an in-core inode (see
1860 * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1861 * symlink request has completed on the server.
1862 *
1863 * So instead we allocate a raw page, copy the symname into it, then do
1864 * the SYMLINK request with the page as the buffer.  If it succeeds, we
1865 * now have a new file handle and can instantiate an in-core NFS inode
1866 * and move the raw page into its mapping.
1867 */
1868int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1869{
1870	struct page *page;
1871	char *kaddr;
1872	struct iattr attr;
1873	unsigned int pathlen = strlen(symname);
1874	int error;
1875
1876	dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1877		dir->i_ino, dentry, symname);
1878
1879	if (pathlen > PAGE_SIZE)
1880		return -ENAMETOOLONG;
1881
1882	attr.ia_mode = S_IFLNK | S_IRWXUGO;
1883	attr.ia_valid = ATTR_MODE;
1884
1885	page = alloc_page(GFP_USER);
1886	if (!page)
1887		return -ENOMEM;
1888
1889	kaddr = page_address(page);
1890	memcpy(kaddr, symname, pathlen);
1891	if (pathlen < PAGE_SIZE)
1892		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1893
1894	trace_nfs_symlink_enter(dir, dentry);
1895	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1896	trace_nfs_symlink_exit(dir, dentry, error);
1897	if (error != 0) {
1898		dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1899			dir->i_sb->s_id, dir->i_ino,
1900			dentry, symname, error);
1901		d_drop(dentry);
1902		__free_page(page);
1903		return error;
1904	}
1905
1906	/*
1907	 * No big deal if we can't add this page to the page cache here.
1908	 * READLINK will get the missing page from the server if needed.
1909	 */
1910	if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
1911							GFP_KERNEL)) {
1912		SetPageUptodate(page);
1913		unlock_page(page);
1914		/*
1915		 * add_to_page_cache_lru() grabs an extra page refcount.
1916		 * Drop it here to avoid leaking this page later.
1917		 */
1918		put_page(page);
1919	} else
1920		__free_page(page);
1921
1922	return 0;
1923}
1924EXPORT_SYMBOL_GPL(nfs_symlink);
1925
1926int
1927nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1928{
1929	struct inode *inode = d_inode(old_dentry);
1930	int error;
1931
1932	dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1933		old_dentry, dentry);
1934
1935	trace_nfs_link_enter(inode, dir, dentry);
 
 
1936	d_drop(dentry);
1937	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1938	if (error == 0) {
1939		ihold(inode);
1940		d_add(dentry, inode);
1941	}
1942	trace_nfs_link_exit(inode, dir, dentry, error);
1943	return error;
1944}
1945EXPORT_SYMBOL_GPL(nfs_link);
1946
1947/*
1948 * RENAME
1949 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1950 * different file handle for the same inode after a rename (e.g. when
1951 * moving to a different directory). A fail-safe method to do so would
1952 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1953 * rename the old file using the sillyrename stuff. This way, the original
1954 * file in old_dir will go away when the last process iput()s the inode.
1955 *
1956 * FIXED.
1957 * 
1958 * It actually works quite well. One needs to have the possibility for
1959 * at least one ".nfs..." file in each directory the file ever gets
1960 * moved or linked to which happens automagically with the new
1961 * implementation that only depends on the dcache stuff instead of
1962 * using the inode layer
1963 *
1964 * Unfortunately, things are a little more complicated than indicated
1965 * above. For a cross-directory move, we want to make sure we can get
1966 * rid of the old inode after the operation.  This means there must be
1967 * no pending writes (if it's a file), and the use count must be 1.
1968 * If these conditions are met, we can drop the dentries before doing
1969 * the rename.
1970 */
1971int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1972	       struct inode *new_dir, struct dentry *new_dentry,
1973	       unsigned int flags)
1974{
1975	struct inode *old_inode = d_inode(old_dentry);
1976	struct inode *new_inode = d_inode(new_dentry);
1977	struct dentry *dentry = NULL, *rehash = NULL;
1978	struct rpc_task *task;
1979	int error = -EBUSY;
1980
1981	if (flags)
1982		return -EINVAL;
1983
1984	dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
1985		 old_dentry, new_dentry,
1986		 d_count(new_dentry));
1987
1988	trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
1989	/*
1990	 * For non-directories, check whether the target is busy and if so,
1991	 * make a copy of the dentry and then do a silly-rename. If the
1992	 * silly-rename succeeds, the copied dentry is hashed and becomes
1993	 * the new target.
1994	 */
1995	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1996		/*
1997		 * To prevent any new references to the target during the
1998		 * rename, we unhash the dentry in advance.
1999		 */
2000		if (!d_unhashed(new_dentry)) {
2001			d_drop(new_dentry);
2002			rehash = new_dentry;
2003		}
2004
2005		if (d_count(new_dentry) > 2) {
2006			int err;
2007
2008			/* copy the target dentry's name */
2009			dentry = d_alloc(new_dentry->d_parent,
2010					 &new_dentry->d_name);
2011			if (!dentry)
2012				goto out;
2013
2014			/* silly-rename the existing target ... */
2015			err = nfs_sillyrename(new_dir, new_dentry);
2016			if (err)
2017				goto out;
2018
2019			new_dentry = dentry;
2020			rehash = NULL;
2021			new_inode = NULL;
2022		}
2023	}
2024
 
 
 
 
2025	task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2026	if (IS_ERR(task)) {
2027		error = PTR_ERR(task);
2028		goto out;
2029	}
2030
2031	error = rpc_wait_for_completion_task(task);
2032	if (error != 0) {
2033		((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2034		/* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2035		smp_wmb();
2036	} else
2037		error = task->tk_status;
2038	rpc_put_task(task);
2039	nfs_mark_for_revalidate(old_inode);
2040out:
2041	if (rehash)
2042		d_rehash(rehash);
2043	trace_nfs_rename_exit(old_dir, old_dentry,
2044			new_dir, new_dentry, error);
2045	if (!error) {
2046		if (new_inode != NULL)
2047			nfs_drop_nlink(new_inode);
2048		/*
2049		 * The d_move() should be here instead of in an async RPC completion
2050		 * handler because we need the proper locks to move the dentry.  If
2051		 * we're interrupted by a signal, the async RPC completion handler
2052		 * should mark the directories for revalidation.
2053		 */
2054		d_move(old_dentry, new_dentry);
2055		nfs_set_verifier(old_dentry,
2056					nfs_save_change_attribute(new_dir));
2057	} else if (error == -ENOENT)
2058		nfs_dentry_handle_enoent(old_dentry);
2059
2060	/* new dentry created? */
2061	if (dentry)
2062		dput(dentry);
2063	return error;
2064}
2065EXPORT_SYMBOL_GPL(nfs_rename);
2066
2067static DEFINE_SPINLOCK(nfs_access_lru_lock);
2068static LIST_HEAD(nfs_access_lru_list);
2069static atomic_long_t nfs_access_nr_entries;
2070
2071static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2072module_param(nfs_access_max_cachesize, ulong, 0644);
2073MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2074
2075static void nfs_access_free_entry(struct nfs_access_entry *entry)
2076{
2077	put_rpccred(entry->cred);
2078	kfree_rcu(entry, rcu_head);
2079	smp_mb__before_atomic();
2080	atomic_long_dec(&nfs_access_nr_entries);
2081	smp_mb__after_atomic();
2082}
2083
2084static void nfs_access_free_list(struct list_head *head)
2085{
2086	struct nfs_access_entry *cache;
2087
2088	while (!list_empty(head)) {
2089		cache = list_entry(head->next, struct nfs_access_entry, lru);
2090		list_del(&cache->lru);
2091		nfs_access_free_entry(cache);
2092	}
2093}
2094
2095static unsigned long
2096nfs_do_access_cache_scan(unsigned int nr_to_scan)
2097{
2098	LIST_HEAD(head);
2099	struct nfs_inode *nfsi, *next;
2100	struct nfs_access_entry *cache;
2101	long freed = 0;
2102
2103	spin_lock(&nfs_access_lru_lock);
2104	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2105		struct inode *inode;
2106
2107		if (nr_to_scan-- == 0)
2108			break;
2109		inode = &nfsi->vfs_inode;
2110		spin_lock(&inode->i_lock);
2111		if (list_empty(&nfsi->access_cache_entry_lru))
2112			goto remove_lru_entry;
2113		cache = list_entry(nfsi->access_cache_entry_lru.next,
2114				struct nfs_access_entry, lru);
2115		list_move(&cache->lru, &head);
2116		rb_erase(&cache->rb_node, &nfsi->access_cache);
2117		freed++;
2118		if (!list_empty(&nfsi->access_cache_entry_lru))
2119			list_move_tail(&nfsi->access_cache_inode_lru,
2120					&nfs_access_lru_list);
2121		else {
2122remove_lru_entry:
2123			list_del_init(&nfsi->access_cache_inode_lru);
2124			smp_mb__before_atomic();
2125			clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2126			smp_mb__after_atomic();
2127		}
2128		spin_unlock(&inode->i_lock);
2129	}
2130	spin_unlock(&nfs_access_lru_lock);
2131	nfs_access_free_list(&head);
2132	return freed;
2133}
2134
2135unsigned long
2136nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2137{
2138	int nr_to_scan = sc->nr_to_scan;
2139	gfp_t gfp_mask = sc->gfp_mask;
2140
2141	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2142		return SHRINK_STOP;
2143	return nfs_do_access_cache_scan(nr_to_scan);
2144}
2145
2146
2147unsigned long
2148nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2149{
2150	return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2151}
2152
2153static void
2154nfs_access_cache_enforce_limit(void)
2155{
2156	long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2157	unsigned long diff;
2158	unsigned int nr_to_scan;
2159
2160	if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2161		return;
2162	nr_to_scan = 100;
2163	diff = nr_entries - nfs_access_max_cachesize;
2164	if (diff < nr_to_scan)
2165		nr_to_scan = diff;
2166	nfs_do_access_cache_scan(nr_to_scan);
2167}
2168
2169static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2170{
2171	struct rb_root *root_node = &nfsi->access_cache;
2172	struct rb_node *n;
2173	struct nfs_access_entry *entry;
2174
2175	/* Unhook entries from the cache */
2176	while ((n = rb_first(root_node)) != NULL) {
2177		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2178		rb_erase(n, root_node);
2179		list_move(&entry->lru, head);
2180	}
2181	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2182}
2183
2184void nfs_access_zap_cache(struct inode *inode)
2185{
2186	LIST_HEAD(head);
2187
2188	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2189		return;
2190	/* Remove from global LRU init */
2191	spin_lock(&nfs_access_lru_lock);
2192	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2193		list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2194
2195	spin_lock(&inode->i_lock);
2196	__nfs_access_zap_cache(NFS_I(inode), &head);
2197	spin_unlock(&inode->i_lock);
2198	spin_unlock(&nfs_access_lru_lock);
2199	nfs_access_free_list(&head);
2200}
2201EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2202
2203static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2204{
2205	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2206	struct nfs_access_entry *entry;
2207
2208	while (n != NULL) {
2209		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2210
2211		if (cred < entry->cred)
2212			n = n->rb_left;
2213		else if (cred > entry->cred)
2214			n = n->rb_right;
2215		else
2216			return entry;
2217	}
2218	return NULL;
2219}
2220
2221static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res, bool may_block)
2222{
2223	struct nfs_inode *nfsi = NFS_I(inode);
2224	struct nfs_access_entry *cache;
2225	bool retry = true;
2226	int err;
2227
2228	spin_lock(&inode->i_lock);
2229	for(;;) {
2230		if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2231			goto out_zap;
2232		cache = nfs_access_search_rbtree(inode, cred);
2233		err = -ENOENT;
2234		if (cache == NULL)
2235			goto out;
2236		/* Found an entry, is our attribute cache valid? */
2237		if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2238			break;
2239		err = -ECHILD;
2240		if (!may_block)
2241			goto out;
2242		if (!retry)
2243			goto out_zap;
2244		spin_unlock(&inode->i_lock);
2245		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2246		if (err)
2247			return err;
2248		spin_lock(&inode->i_lock);
2249		retry = false;
2250	}
 
2251	res->cred = cache->cred;
2252	res->mask = cache->mask;
2253	list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2254	err = 0;
2255out:
2256	spin_unlock(&inode->i_lock);
2257	return err;
2258out_zap:
2259	spin_unlock(&inode->i_lock);
2260	nfs_access_zap_cache(inode);
2261	return -ENOENT;
2262}
2263
2264static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2265{
2266	/* Only check the most recently returned cache entry,
2267	 * but do it without locking.
2268	 */
2269	struct nfs_inode *nfsi = NFS_I(inode);
2270	struct nfs_access_entry *cache;
2271	int err = -ECHILD;
2272	struct list_head *lh;
2273
2274	rcu_read_lock();
2275	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2276		goto out;
2277	lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2278	cache = list_entry(lh, struct nfs_access_entry, lru);
2279	if (lh == &nfsi->access_cache_entry_lru ||
2280	    cred != cache->cred)
2281		cache = NULL;
2282	if (cache == NULL)
2283		goto out;
2284	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2285		goto out;
 
2286	res->cred = cache->cred;
2287	res->mask = cache->mask;
2288	err = 0;
2289out:
2290	rcu_read_unlock();
2291	return err;
2292}
2293
2294static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2295{
2296	struct nfs_inode *nfsi = NFS_I(inode);
2297	struct rb_root *root_node = &nfsi->access_cache;
2298	struct rb_node **p = &root_node->rb_node;
2299	struct rb_node *parent = NULL;
2300	struct nfs_access_entry *entry;
2301
2302	spin_lock(&inode->i_lock);
2303	while (*p != NULL) {
2304		parent = *p;
2305		entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2306
2307		if (set->cred < entry->cred)
2308			p = &parent->rb_left;
2309		else if (set->cred > entry->cred)
2310			p = &parent->rb_right;
2311		else
2312			goto found;
2313	}
2314	rb_link_node(&set->rb_node, parent, p);
2315	rb_insert_color(&set->rb_node, root_node);
2316	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2317	spin_unlock(&inode->i_lock);
2318	return;
2319found:
2320	rb_replace_node(parent, &set->rb_node, root_node);
2321	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2322	list_del(&entry->lru);
2323	spin_unlock(&inode->i_lock);
2324	nfs_access_free_entry(entry);
2325}
2326
2327void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2328{
2329	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2330	if (cache == NULL)
2331		return;
2332	RB_CLEAR_NODE(&cache->rb_node);
 
2333	cache->cred = get_rpccred(set->cred);
2334	cache->mask = set->mask;
2335
2336	/* The above field assignments must be visible
2337	 * before this item appears on the lru.  We cannot easily
2338	 * use rcu_assign_pointer, so just force the memory barrier.
2339	 */
2340	smp_wmb();
2341	nfs_access_add_rbtree(inode, cache);
2342
2343	/* Update accounting */
2344	smp_mb__before_atomic();
2345	atomic_long_inc(&nfs_access_nr_entries);
2346	smp_mb__after_atomic();
2347
2348	/* Add inode to global LRU list */
2349	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2350		spin_lock(&nfs_access_lru_lock);
2351		if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2352			list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2353					&nfs_access_lru_list);
2354		spin_unlock(&nfs_access_lru_lock);
2355	}
2356	nfs_access_cache_enforce_limit();
2357}
2358EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2359
2360#define NFS_MAY_READ (NFS_ACCESS_READ)
2361#define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2362		NFS_ACCESS_EXTEND | \
2363		NFS_ACCESS_DELETE)
2364#define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2365		NFS_ACCESS_EXTEND)
2366#define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2367#define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2368#define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2369static int
2370nfs_access_calc_mask(u32 access_result, umode_t umode)
2371{
2372	int mask = 0;
2373
2374	if (access_result & NFS_MAY_READ)
2375		mask |= MAY_READ;
2376	if (S_ISDIR(umode)) {
2377		if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2378			mask |= MAY_WRITE;
2379		if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2380			mask |= MAY_EXEC;
2381	} else if (S_ISREG(umode)) {
2382		if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2383			mask |= MAY_WRITE;
2384		if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2385			mask |= MAY_EXEC;
2386	} else if (access_result & NFS_MAY_WRITE)
2387			mask |= MAY_WRITE;
2388	return mask;
2389}
2390
2391void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2392{
2393	entry->mask = access_result;
 
 
 
 
 
 
 
2394}
2395EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2396
2397static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2398{
2399	struct nfs_access_entry cache;
2400	bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2401	int cache_mask;
2402	int status;
2403
2404	trace_nfs_access_enter(inode);
2405
2406	status = nfs_access_get_cached_rcu(inode, cred, &cache);
2407	if (status != 0)
2408		status = nfs_access_get_cached(inode, cred, &cache, may_block);
2409	if (status == 0)
2410		goto out_cached;
2411
2412	status = -ECHILD;
2413	if (!may_block)
2414		goto out;
2415
2416	/*
2417	 * Determine which access bits we want to ask for...
2418	 */
2419	cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2420	if (S_ISDIR(inode->i_mode))
2421		cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2422	else
2423		cache.mask |= NFS_ACCESS_EXECUTE;
2424	cache.cred = cred;
 
2425	status = NFS_PROTO(inode)->access(inode, &cache);
2426	if (status != 0) {
2427		if (status == -ESTALE) {
2428			nfs_zap_caches(inode);
2429			if (!S_ISDIR(inode->i_mode))
2430				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2431		}
2432		goto out;
2433	}
2434	nfs_access_add_cache(inode, &cache);
2435out_cached:
2436	cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2437	if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2438		status = -EACCES;
2439out:
2440	trace_nfs_access_exit(inode, status);
2441	return status;
2442}
2443
2444static int nfs_open_permission_mask(int openflags)
2445{
2446	int mask = 0;
2447
2448	if (openflags & __FMODE_EXEC) {
2449		/* ONLY check exec rights */
2450		mask = MAY_EXEC;
2451	} else {
2452		if ((openflags & O_ACCMODE) != O_WRONLY)
2453			mask |= MAY_READ;
2454		if ((openflags & O_ACCMODE) != O_RDONLY)
2455			mask |= MAY_WRITE;
2456	}
2457
2458	return mask;
2459}
2460
2461int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2462{
2463	return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2464}
2465EXPORT_SYMBOL_GPL(nfs_may_open);
2466
2467static int nfs_execute_ok(struct inode *inode, int mask)
2468{
2469	struct nfs_server *server = NFS_SERVER(inode);
2470	int ret = 0;
2471
2472	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) {
2473		if (mask & MAY_NOT_BLOCK)
2474			return -ECHILD;
2475		ret = __nfs_revalidate_inode(server, inode);
2476	}
2477	if (ret == 0 && !execute_ok(inode))
2478		ret = -EACCES;
2479	return ret;
2480}
2481
2482int nfs_permission(struct inode *inode, int mask)
2483{
2484	struct rpc_cred *cred;
2485	int res = 0;
2486
2487	nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2488
2489	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2490		goto out;
2491	/* Is this sys_access() ? */
2492	if (mask & (MAY_ACCESS | MAY_CHDIR))
2493		goto force_lookup;
2494
2495	switch (inode->i_mode & S_IFMT) {
2496		case S_IFLNK:
2497			goto out;
2498		case S_IFREG:
2499			if ((mask & MAY_OPEN) &&
2500			   nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2501				return 0;
2502			break;
2503		case S_IFDIR:
2504			/*
2505			 * Optimize away all write operations, since the server
2506			 * will check permissions when we perform the op.
2507			 */
2508			if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2509				goto out;
2510	}
2511
2512force_lookup:
2513	if (!NFS_PROTO(inode)->access)
2514		goto out_notsup;
2515
2516	/* Always try fast lookups first */
2517	rcu_read_lock();
2518	cred = rpc_lookup_cred_nonblock();
2519	if (!IS_ERR(cred))
2520		res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2521	else
2522		res = PTR_ERR(cred);
2523	rcu_read_unlock();
2524	if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2525		/* Fast lookup failed, try the slow way */
2526		cred = rpc_lookup_cred();
2527		if (!IS_ERR(cred)) {
2528			res = nfs_do_access(inode, cred, mask);
2529			put_rpccred(cred);
2530		} else
2531			res = PTR_ERR(cred);
2532	}
2533out:
2534	if (!res && (mask & MAY_EXEC))
2535		res = nfs_execute_ok(inode, mask);
2536
2537	dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2538		inode->i_sb->s_id, inode->i_ino, mask, res);
2539	return res;
2540out_notsup:
2541	if (mask & MAY_NOT_BLOCK)
2542		return -ECHILD;
2543
2544	res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2545	if (res == 0)
2546		res = generic_permission(inode, mask);
2547	goto out;
2548}
2549EXPORT_SYMBOL_GPL(nfs_permission);
2550
2551/*
2552 * Local variables:
2553 *  version-control: t
2554 *  kept-new-versions: 5
2555 * End:
2556 */