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_shared	= 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;
 121	if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
 122		/* This is a mountpoint, so d_revalidate will never
 123		 * have been called, so we need to refresh the
 124		 * inode (for close-open consistency) ourselves.
 125		 */
 126		__nfs_revalidate_inode(NFS_SERVER(inode), inode);
 127	}
 128out:
 129	put_rpccred(cred);
 130	return res;
 131}
 132
 133static int
 134nfs_closedir(struct inode *inode, struct file *filp)
 135{
 136	put_nfs_open_dir_context(file_inode(filp), filp->private_data);
 137	return 0;
 138}
 139
 140struct nfs_cache_array_entry {
 141	u64 cookie;
 142	u64 ino;
 143	struct qstr string;
 144	unsigned char d_type;
 145};
 146
 147struct nfs_cache_array {
 148	atomic_t refcount;
 149	int size;
 150	int eof_index;
 151	u64 last_cookie;
 152	struct nfs_cache_array_entry array[0];
 153};
 154
 155typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
 156typedef struct {
 157	struct file	*file;
 158	struct page	*page;
 159	struct dir_context *ctx;
 160	unsigned long	page_index;
 161	u64		*dir_cookie;
 162	u64		last_cookie;
 163	loff_t		current_index;
 164	decode_dirent_t	decode;
 165
 166	unsigned long	timestamp;
 167	unsigned long	gencount;
 168	unsigned int	cache_entry_index;
 169	unsigned int	plus:1;
 170	unsigned int	eof:1;
 171} nfs_readdir_descriptor_t;
 172
 173/*
 174 * The caller is responsible for calling nfs_readdir_release_array(page)
 175 */
 176static
 177struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
 178{
 179	void *ptr;
 180	if (page == NULL)
 181		return ERR_PTR(-EIO);
 182	ptr = kmap(page);
 183	if (ptr == NULL)
 184		return ERR_PTR(-ENOMEM);
 185	return ptr;
 186}
 187
 188static
 189void nfs_readdir_release_array(struct page *page)
 190{
 191	kunmap(page);
 192}
 193
 194/*
 195 * we are freeing strings created by nfs_add_to_readdir_array()
 196 */
 197static
 198void nfs_readdir_clear_array(struct page *page)
 199{
 200	struct nfs_cache_array *array;
 201	int i;
 202
 203	array = kmap_atomic(page);
 204	if (atomic_dec_and_test(&array->refcount))
 205		for (i = 0; i < array->size; i++)
 206			kfree(array->array[i].string.name);
 207	kunmap_atomic(array);
 208}
 209
 210static bool grab_page(struct page *page)
 211{
 212	struct nfs_cache_array *array = kmap_atomic(page);
 213	bool res = atomic_inc_not_zero(&array->refcount);
 214	kunmap_atomic(array);
 215	return res;
 216}
 217
 218/*
 219 * the caller is responsible for freeing qstr.name
 220 * when called by nfs_readdir_add_to_array, the strings will be freed in
 221 * nfs_clear_readdir_array()
 222 */
 223static
 224int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
 225{
 226	string->len = len;
 227	string->name = kmemdup(name, len, GFP_KERNEL);
 228	if (string->name == NULL)
 229		return -ENOMEM;
 230	/*
 231	 * Avoid a kmemleak false positive. The pointer to the name is stored
 232	 * in a page cache page which kmemleak does not scan.
 233	 */
 234	kmemleak_not_leak(string->name);
 235	string->hash = full_name_hash(NULL, name, len);
 236	return 0;
 237}
 238
 239static
 240int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
 241{
 242	struct nfs_cache_array *array = nfs_readdir_get_array(page);
 243	struct nfs_cache_array_entry *cache_entry;
 244	int ret;
 245
 246	if (IS_ERR(array))
 247		return PTR_ERR(array);
 248
 249	cache_entry = &array->array[array->size];
 250
 251	/* Check that this entry lies within the page bounds */
 252	ret = -ENOSPC;
 253	if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
 254		goto out;
 255
 256	cache_entry->cookie = entry->prev_cookie;
 257	cache_entry->ino = entry->ino;
 258	cache_entry->d_type = entry->d_type;
 259	ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
 260	if (ret)
 261		goto out;
 262	array->last_cookie = entry->cookie;
 263	array->size++;
 264	if (entry->eof != 0)
 265		array->eof_index = array->size;
 266out:
 267	nfs_readdir_release_array(page);
 268	return ret;
 269}
 270
 271static
 272int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
 273{
 274	loff_t diff = desc->ctx->pos - desc->current_index;
 275	unsigned int index;
 276
 277	if (diff < 0)
 278		goto out_eof;
 279	if (diff >= array->size) {
 280		if (array->eof_index >= 0)
 281			goto out_eof;
 282		return -EAGAIN;
 283	}
 284
 285	index = (unsigned int)diff;
 286	*desc->dir_cookie = array->array[index].cookie;
 287	desc->cache_entry_index = index;
 288	return 0;
 289out_eof:
 290	desc->eof = 1;
 291	return -EBADCOOKIE;
 292}
 293
 294static bool
 295nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
 296{
 297	if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
 298		return false;
 299	smp_rmb();
 300	return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
 301}
 302
 303static
 304int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
 305{
 306	int i;
 307	loff_t new_pos;
 308	int status = -EAGAIN;
 309
 310	for (i = 0; i < array->size; i++) {
 311		if (array->array[i].cookie == *desc->dir_cookie) {
 312			struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
 313			struct nfs_open_dir_context *ctx = desc->file->private_data;
 314
 315			new_pos = desc->current_index + i;
 316			if (ctx->attr_gencount != nfsi->attr_gencount ||
 317			    !nfs_readdir_inode_mapping_valid(nfsi)) {
 318				ctx->duped = 0;
 319				ctx->attr_gencount = nfsi->attr_gencount;
 320			} else if (new_pos < desc->ctx->pos) {
 321				if (ctx->duped > 0
 322				    && ctx->dup_cookie == *desc->dir_cookie) {
 323					if (printk_ratelimit()) {
 324						pr_notice("NFS: directory %pD2 contains a readdir loop."
 325								"Please contact your server vendor.  "
 326								"The file: %.*s has duplicate cookie %llu\n",
 327								desc->file, array->array[i].string.len,
 328								array->array[i].string.name, *desc->dir_cookie);
 
 
 329					}
 330					status = -ELOOP;
 331					goto out;
 332				}
 333				ctx->dup_cookie = *desc->dir_cookie;
 334				ctx->duped = -1;
 335			}
 336			desc->ctx->pos = new_pos;
 337			desc->cache_entry_index = i;
 338			return 0;
 339		}
 340	}
 341	if (array->eof_index >= 0) {
 342		status = -EBADCOOKIE;
 343		if (*desc->dir_cookie == array->last_cookie)
 344			desc->eof = 1;
 345	}
 346out:
 347	return status;
 348}
 349
 350static
 351int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
 352{
 353	struct nfs_cache_array *array;
 354	int status;
 355
 356	array = nfs_readdir_get_array(desc->page);
 357	if (IS_ERR(array)) {
 358		status = PTR_ERR(array);
 359		goto out;
 360	}
 361
 362	if (*desc->dir_cookie == 0)
 363		status = nfs_readdir_search_for_pos(array, desc);
 364	else
 365		status = nfs_readdir_search_for_cookie(array, desc);
 366
 367	if (status == -EAGAIN) {
 368		desc->last_cookie = array->last_cookie;
 369		desc->current_index += array->size;
 370		desc->page_index++;
 371	}
 372	nfs_readdir_release_array(desc->page);
 373out:
 374	return status;
 375}
 376
 377/* Fill a page with xdr information before transferring to the cache page */
 378static
 379int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
 380			struct nfs_entry *entry, struct file *file, struct inode *inode)
 381{
 382	struct nfs_open_dir_context *ctx = file->private_data;
 383	struct rpc_cred	*cred = ctx->cred;
 384	unsigned long	timestamp, gencount;
 385	int		error;
 386
 387 again:
 388	timestamp = jiffies;
 389	gencount = nfs_inc_attr_generation_counter();
 390	error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages,
 391					  NFS_SERVER(inode)->dtsize, desc->plus);
 392	if (error < 0) {
 393		/* We requested READDIRPLUS, but the server doesn't grok it */
 394		if (error == -ENOTSUPP && desc->plus) {
 395			NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
 396			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
 397			desc->plus = 0;
 398			goto again;
 399		}
 400		goto error;
 401	}
 402	desc->timestamp = timestamp;
 403	desc->gencount = gencount;
 404error:
 405	return error;
 406}
 407
 408static int xdr_decode(nfs_readdir_descriptor_t *desc,
 409		      struct nfs_entry *entry, struct xdr_stream *xdr)
 410{
 411	int error;
 412
 413	error = desc->decode(xdr, entry, desc->plus);
 414	if (error)
 415		return error;
 416	entry->fattr->time_start = desc->timestamp;
 417	entry->fattr->gencount = desc->gencount;
 418	return 0;
 419}
 420
 421/* Match file and dirent using either filehandle or fileid
 422 * Note: caller is responsible for checking the fsid
 423 */
 424static
 425int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
 426{
 427	struct inode *inode;
 428	struct nfs_inode *nfsi;
 429
 430	if (d_really_is_negative(dentry))
 431		return 0;
 432
 433	inode = d_inode(dentry);
 434	if (is_bad_inode(inode) || NFS_STALE(inode))
 435		return 0;
 436
 437	nfsi = NFS_I(inode);
 438	if (entry->fattr->fileid != nfsi->fileid)
 439		return 0;
 440	if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
 441		return 0;
 442	return 1;
 443}
 444
 445static
 446bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
 447{
 448	if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
 449		return false;
 450	if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
 451		return true;
 452	if (ctx->pos == 0)
 453		return true;
 454	return false;
 455}
 456
 457/*
 458 * This function is called by the lookup and getattr code to request the
 459 * use of readdirplus to accelerate any future lookups in the same
 460 * directory.
 461 */
 462void nfs_advise_use_readdirplus(struct inode *dir)
 463{
 464	struct nfs_inode *nfsi = NFS_I(dir);
 465
 466	if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
 467	    !list_empty(&nfsi->open_files))
 468		set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
 469}
 470
 471/*
 472 * This function is mainly for use by nfs_getattr().
 473 *
 474 * If this is an 'ls -l', we want to force use of readdirplus.
 475 * Do this by checking if there is an active file descriptor
 476 * and calling nfs_advise_use_readdirplus, then forcing a
 477 * cache flush.
 478 */
 479void nfs_force_use_readdirplus(struct inode *dir)
 480{
 481	struct nfs_inode *nfsi = NFS_I(dir);
 482
 483	if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
 484	    !list_empty(&nfsi->open_files)) {
 485		set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
 486		invalidate_mapping_pages(dir->i_mapping, 0, -1);
 487	}
 488}
 489
 490static
 491void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
 492{
 493	struct qstr filename = QSTR_INIT(entry->name, entry->len);
 494	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
 
 
 495	struct dentry *dentry;
 496	struct dentry *alias;
 497	struct inode *dir = d_inode(parent);
 498	struct inode *inode;
 499	int status;
 500
 501	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
 502		return;
 503	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
 504		return;
 505	if (filename.len == 0)
 506		return;
 507	/* Validate that the name doesn't contain any illegal '\0' */
 508	if (strnlen(filename.name, filename.len) != filename.len)
 509		return;
 510	/* ...or '/' */
 511	if (strnchr(filename.name, filename.len, '/'))
 512		return;
 513	if (filename.name[0] == '.') {
 514		if (filename.len == 1)
 515			return;
 516		if (filename.len == 2 && filename.name[1] == '.')
 517			return;
 518	}
 519	filename.hash = full_name_hash(parent, filename.name, filename.len);
 520
 521	dentry = d_lookup(parent, &filename);
 522again:
 523	if (!dentry) {
 524		dentry = d_alloc_parallel(parent, &filename, &wq);
 525		if (IS_ERR(dentry))
 526			return;
 527	}
 528	if (!d_in_lookup(dentry)) {
 529		/* Is there a mountpoint here? If so, just exit */
 530		if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
 531					&entry->fattr->fsid))
 532			goto out;
 533		if (nfs_same_file(dentry, entry)) {
 534			if (!entry->fh->size)
 535				goto out;
 536			nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
 537			status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
 538			if (!status)
 539				nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
 540			goto out;
 541		} else {
 542			d_invalidate(dentry);
 543			dput(dentry);
 544			dentry = NULL;
 545			goto again;
 546		}
 547	}
 548	if (!entry->fh->size) {
 549		d_lookup_done(dentry);
 
 
 
 
 
 550		goto out;
 551	}
 552
 553	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
 554	alias = d_splice_alias(inode, dentry);
 555	d_lookup_done(dentry);
 556	if (alias) {
 557		if (IS_ERR(alias))
 558			goto out;
 559		dput(dentry);
 560		dentry = alias;
 561	}
 562	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
 563out:
 564	dput(dentry);
 565}
 566
 567/* Perform conversion from xdr to cache array */
 568static
 569int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
 570				struct page **xdr_pages, struct page *page, unsigned int buflen)
 571{
 572	struct xdr_stream stream;
 573	struct xdr_buf buf;
 574	struct page *scratch;
 575	struct nfs_cache_array *array;
 576	unsigned int count = 0;
 577	int status;
 578
 579	scratch = alloc_page(GFP_KERNEL);
 580	if (scratch == NULL)
 581		return -ENOMEM;
 582
 583	if (buflen == 0)
 584		goto out_nopages;
 585
 586	xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
 587	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 588
 589	do {
 590		status = xdr_decode(desc, entry, &stream);
 591		if (status != 0) {
 592			if (status == -EAGAIN)
 593				status = 0;
 594			break;
 595		}
 596
 597		count++;
 598
 599		if (desc->plus != 0)
 600			nfs_prime_dcache(file_dentry(desc->file), entry);
 601
 602		status = nfs_readdir_add_to_array(entry, page);
 603		if (status != 0)
 604			break;
 605	} while (!entry->eof);
 606
 607out_nopages:
 608	if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
 609		array = nfs_readdir_get_array(page);
 610		if (!IS_ERR(array)) {
 611			array->eof_index = array->size;
 612			status = 0;
 613			nfs_readdir_release_array(page);
 614		} else
 615			status = PTR_ERR(array);
 616	}
 617
 618	put_page(scratch);
 619	return status;
 620}
 621
 622static
 623void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
 624{
 625	unsigned int i;
 626	for (i = 0; i < npages; i++)
 627		put_page(pages[i]);
 628}
 629
 
 
 
 
 
 
 
 630/*
 631 * nfs_readdir_large_page will allocate pages that must be freed with a call
 632 * to nfs_readdir_free_pagearray
 633 */
 634static
 635int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
 636{
 637	unsigned int i;
 638
 639	for (i = 0; i < npages; i++) {
 640		struct page *page = alloc_page(GFP_KERNEL);
 641		if (page == NULL)
 642			goto out_freepages;
 643		pages[i] = page;
 644	}
 645	return 0;
 646
 647out_freepages:
 648	nfs_readdir_free_pages(pages, i);
 649	return -ENOMEM;
 650}
 651
 652static
 653int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
 654{
 655	struct page *pages[NFS_MAX_READDIR_PAGES];
 
 656	struct nfs_entry entry;
 657	struct file	*file = desc->file;
 658	struct nfs_cache_array *array;
 659	int status = -ENOMEM;
 660	unsigned int array_size = ARRAY_SIZE(pages);
 661
 662	entry.prev_cookie = 0;
 663	entry.cookie = desc->last_cookie;
 664	entry.eof = 0;
 665	entry.fh = nfs_alloc_fhandle();
 666	entry.fattr = nfs_alloc_fattr();
 667	entry.server = NFS_SERVER(inode);
 668	if (entry.fh == NULL || entry.fattr == NULL)
 669		goto out;
 670
 671	entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
 672	if (IS_ERR(entry.label)) {
 673		status = PTR_ERR(entry.label);
 674		goto out;
 675	}
 676
 677	array = nfs_readdir_get_array(page);
 678	if (IS_ERR(array)) {
 679		status = PTR_ERR(array);
 680		goto out_label_free;
 681	}
 682	memset(array, 0, sizeof(struct nfs_cache_array));
 683	atomic_set(&array->refcount, 1);
 684	array->eof_index = -1;
 685
 686	status = nfs_readdir_alloc_pages(pages, array_size);
 687	if (status < 0)
 688		goto out_release_array;
 689	do {
 690		unsigned int pglen;
 691		status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
 692
 693		if (status < 0)
 694			break;
 695		pglen = status;
 696		status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
 697		if (status < 0) {
 698			if (status == -ENOSPC)
 699				status = 0;
 700			break;
 701		}
 702	} while (array->eof_index < 0);
 703
 704	nfs_readdir_free_pages(pages, array_size);
 705out_release_array:
 706	nfs_readdir_release_array(page);
 707out_label_free:
 708	nfs4_label_free(entry.label);
 709out:
 710	nfs_free_fattr(entry.fattr);
 711	nfs_free_fhandle(entry.fh);
 712	return status;
 713}
 714
 715/*
 716 * Now we cache directories properly, by converting xdr information
 717 * to an array that can be used for lookups later.  This results in
 718 * fewer cache pages, since we can store more information on each page.
 719 * We only need to convert from xdr once so future lookups are much simpler
 720 */
 721static
 722int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
 723{
 724	struct inode	*inode = file_inode(desc->file);
 725	int ret;
 726
 727	ret = nfs_readdir_xdr_to_array(desc, page, inode);
 728	if (ret < 0)
 729		goto error;
 730	SetPageUptodate(page);
 731
 732	if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
 733		/* Should never happen */
 734		nfs_zap_mapping(inode, inode->i_mapping);
 735	}
 736	unlock_page(page);
 737	return 0;
 738 error:
 739	unlock_page(page);
 740	return ret;
 741}
 742
 743static
 744void cache_page_release(nfs_readdir_descriptor_t *desc)
 745{
 746	nfs_readdir_clear_array(desc->page);
 747	put_page(desc->page);
 
 748	desc->page = NULL;
 749}
 750
 751static
 752struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
 753{
 754	struct page *page;
 755
 756	for (;;) {
 757		page = read_cache_page(desc->file->f_mapping,
 758			desc->page_index, (filler_t *)nfs_readdir_filler, desc);
 759		if (IS_ERR(page) || grab_page(page))
 760			break;
 761		put_page(page);
 762	}
 763	return page;
 764}
 765
 766/*
 767 * Returns 0 if desc->dir_cookie was found on page desc->page_index
 768 */
 769static
 770int find_cache_page(nfs_readdir_descriptor_t *desc)
 771{
 772	int res;
 773
 774	desc->page = get_cache_page(desc);
 775	if (IS_ERR(desc->page))
 776		return PTR_ERR(desc->page);
 777
 778	res = nfs_readdir_search_array(desc);
 779	if (res != 0)
 780		cache_page_release(desc);
 781	return res;
 782}
 783
 784/* Search for desc->dir_cookie from the beginning of the page cache */
 785static inline
 786int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
 787{
 788	int res;
 789
 790	if (desc->page_index == 0) {
 791		desc->current_index = 0;
 792		desc->last_cookie = 0;
 793	}
 794	do {
 795		res = find_cache_page(desc);
 796	} while (res == -EAGAIN);
 797	return res;
 798}
 799
 800/*
 801 * Once we've found the start of the dirent within a page: fill 'er up...
 802 */
 803static 
 804int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
 
 805{
 806	struct file	*file = desc->file;
 807	int i = 0;
 808	int res = 0;
 809	struct nfs_cache_array *array = NULL;
 810	struct nfs_open_dir_context *ctx = file->private_data;
 811
 812	array = nfs_readdir_get_array(desc->page);
 813	if (IS_ERR(array)) {
 814		res = PTR_ERR(array);
 815		goto out;
 816	}
 817
 818	for (i = desc->cache_entry_index; i < array->size; i++) {
 819		struct nfs_cache_array_entry *ent;
 820
 821		ent = &array->array[i];
 822		if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
 823		    nfs_compat_user_ino64(ent->ino), ent->d_type)) {
 
 824			desc->eof = 1;
 825			break;
 826		}
 827		desc->ctx->pos++;
 828		if (i < (array->size-1))
 829			*desc->dir_cookie = array->array[i+1].cookie;
 830		else
 831			*desc->dir_cookie = array->last_cookie;
 832		if (ctx->duped != 0)
 833			ctx->duped = 1;
 834	}
 835	if (array->eof_index >= 0)
 836		desc->eof = 1;
 837
 838	nfs_readdir_release_array(desc->page);
 839out:
 840	cache_page_release(desc);
 841	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
 842			(unsigned long long)*desc->dir_cookie, res);
 843	return res;
 844}
 845
 846/*
 847 * If we cannot find a cookie in our cache, we suspect that this is
 848 * because it points to a deleted file, so we ask the server to return
 849 * whatever it thinks is the next entry. We then feed this to filldir.
 850 * If all goes well, we should then be able to find our way round the
 851 * cache on the next call to readdir_search_pagecache();
 852 *
 853 * NOTE: we cannot add the anonymous page to the pagecache because
 854 *	 the data it contains might not be page aligned. Besides,
 855 *	 we should already have a complete representation of the
 856 *	 directory in the page cache by the time we get here.
 857 */
 858static inline
 859int uncached_readdir(nfs_readdir_descriptor_t *desc)
 
 860{
 861	struct page	*page = NULL;
 862	int		status;
 863	struct inode *inode = file_inode(desc->file);
 864	struct nfs_open_dir_context *ctx = desc->file->private_data;
 865
 866	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
 867			(unsigned long long)*desc->dir_cookie);
 868
 869	page = alloc_page(GFP_HIGHUSER);
 870	if (!page) {
 871		status = -ENOMEM;
 872		goto out;
 873	}
 874
 875	desc->page_index = 0;
 876	desc->last_cookie = *desc->dir_cookie;
 877	desc->page = page;
 878	ctx->duped = 0;
 879
 880	status = nfs_readdir_xdr_to_array(desc, page, inode);
 881	if (status < 0)
 882		goto out_release;
 883
 884	status = nfs_do_filldir(desc);
 885
 886 out:
 887	dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
 888			__func__, status);
 889	return status;
 890 out_release:
 891	cache_page_release(desc);
 892	goto out;
 893}
 894
 895/* The file offset position represents the dirent entry number.  A
 896   last cookie cache takes care of the common case of reading the
 897   whole directory.
 898 */
 899static int nfs_readdir(struct file *file, struct dir_context *ctx)
 900{
 901	struct dentry	*dentry = file_dentry(file);
 902	struct inode	*inode = d_inode(dentry);
 903	nfs_readdir_descriptor_t my_desc,
 904			*desc = &my_desc;
 905	struct nfs_open_dir_context *dir_ctx = file->private_data;
 906	int res = 0;
 907
 908	dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
 909			file, (long long)ctx->pos);
 
 910	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
 911
 912	/*
 913	 * ctx->pos points to the dirent entry number.
 914	 * *desc->dir_cookie has the cookie for the next entry. We have
 915	 * to either find the entry with the appropriate number or
 916	 * revalidate the cookie.
 917	 */
 918	memset(desc, 0, sizeof(*desc));
 919
 920	desc->file = file;
 921	desc->ctx = ctx;
 922	desc->dir_cookie = &dir_ctx->dir_cookie;
 923	desc->decode = NFS_PROTO(inode)->decode_dirent;
 924	desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
 925
 926	if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
 927		res = nfs_revalidate_mapping(inode, file->f_mapping);
 928	if (res < 0)
 929		goto out;
 930
 931	do {
 932		res = readdir_search_pagecache(desc);
 933
 934		if (res == -EBADCOOKIE) {
 935			res = 0;
 936			/* This means either end of directory */
 937			if (*desc->dir_cookie && desc->eof == 0) {
 938				/* Or that the server has 'lost' a cookie */
 939				res = uncached_readdir(desc);
 940				if (res == 0)
 941					continue;
 942			}
 943			break;
 944		}
 945		if (res == -ETOOSMALL && desc->plus) {
 946			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
 947			nfs_zap_caches(inode);
 948			desc->page_index = 0;
 949			desc->plus = 0;
 950			desc->eof = 0;
 951			continue;
 952		}
 953		if (res < 0)
 954			break;
 955
 956		res = nfs_do_filldir(desc);
 957		if (res < 0)
 958			break;
 959	} while (!desc->eof);
 960out:
 
 961	if (res > 0)
 962		res = 0;
 963	dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
 
 
 964	return res;
 965}
 966
 967static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
 968{
 
 
 969	struct nfs_open_dir_context *dir_ctx = filp->private_data;
 970
 971	dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
 972			filp, offset, whence);
 
 
 973
 974	switch (whence) {
 
 975		case 1:
 976			offset += filp->f_pos;
 977		case 0:
 978			if (offset >= 0)
 979				break;
 980		default:
 981			return -EINVAL;
 
 982	}
 983	if (offset != filp->f_pos) {
 984		filp->f_pos = offset;
 985		dir_ctx->dir_cookie = 0;
 986		dir_ctx->duped = 0;
 987	}
 
 
 988	return offset;
 989}
 990
 991/*
 992 * All directory operations under NFS are synchronous, so fsync()
 993 * is a dummy operation.
 994 */
 995static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
 996			 int datasync)
 997{
 998	struct inode *inode = file_inode(filp);
 999
1000	dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1001
1002	inode_lock(inode);
1003	nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
1004	inode_unlock(inode);
 
 
 
 
1005	return 0;
1006}
1007
1008/**
1009 * nfs_force_lookup_revalidate - Mark the directory as having changed
1010 * @dir - pointer to directory inode
1011 *
1012 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1013 * full lookup on all child dentries of 'dir' whenever a change occurs
1014 * on the server that might have invalidated our dcache.
1015 *
1016 * The caller should be holding dir->i_lock
1017 */
1018void nfs_force_lookup_revalidate(struct inode *dir)
1019{
1020	NFS_I(dir)->cache_change_attribute++;
1021}
1022EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1023
1024/*
1025 * A check for whether or not the parent directory has changed.
1026 * In the case it has, we assume that the dentries are untrustworthy
1027 * and may need to be looked up again.
1028 * If rcu_walk prevents us from performing a full check, return 0.
1029 */
1030static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1031			      int rcu_walk)
1032{
1033	if (IS_ROOT(dentry))
1034		return 1;
1035	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1036		return 0;
1037	if (!nfs_verify_change_attribute(dir, dentry->d_time))
1038		return 0;
1039	/* Revalidate nfsi->cache_change_attribute before we declare a match */
1040	if (nfs_mapping_need_revalidate_inode(dir)) {
1041		if (rcu_walk)
1042			return 0;
1043		if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1044			return 0;
1045	}
1046	if (!nfs_verify_change_attribute(dir, dentry->d_time))
1047		return 0;
1048	return 1;
1049}
1050
1051/*
 
 
 
 
 
 
 
 
 
 
 
 
 
1052 * Use intent information to check whether or not we're going to do
1053 * an O_EXCL create using this path component.
1054 */
1055static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1056{
1057	if (NFS_PROTO(dir)->version == 2)
1058		return 0;
1059	return flags & LOOKUP_EXCL;
1060}
1061
1062/*
1063 * Inode and filehandle revalidation for lookups.
1064 *
1065 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1066 * or if the intent information indicates that we're about to open this
1067 * particular file and the "nocto" mount flag is not set.
1068 *
1069 */
1070static
1071int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1072{
1073	struct nfs_server *server = NFS_SERVER(inode);
1074	int ret;
1075
1076	if (IS_AUTOMOUNT(inode))
1077		return 0;
1078	/* VFS wants an on-the-wire revalidation */
1079	if (flags & LOOKUP_REVAL)
1080		goto out_force;
1081	/* This is an open(2) */
1082	if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1083	    (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1084		goto out_force;
1085out:
1086	return (inode->i_nlink == 0) ? -ENOENT : 0;
 
 
 
 
1087out_force:
1088	if (flags & LOOKUP_RCU)
1089		return -ECHILD;
1090	ret = __nfs_revalidate_inode(server, inode);
1091	if (ret != 0)
1092		return ret;
1093	goto out;
1094}
1095
1096/*
1097 * We judge how long we want to trust negative
1098 * dentries by looking at the parent inode mtime.
1099 *
1100 * If parent mtime has changed, we revalidate, else we wait for a
1101 * period corresponding to the parent's attribute cache timeout value.
1102 *
1103 * If LOOKUP_RCU prevents us from performing a full check, return 1
1104 * suggesting a reval is needed.
1105 */
1106static inline
1107int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1108		       unsigned int flags)
1109{
1110	/* Don't revalidate a negative dentry if we're creating a new file */
1111	if (flags & LOOKUP_CREATE)
1112		return 0;
1113	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1114		return 1;
1115	return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1116}
1117
1118/*
1119 * This is called every time the dcache has a lookup hit,
1120 * and we should check whether we can really trust that
1121 * lookup.
1122 *
1123 * NOTE! The hit can be a negative hit too, don't assume
1124 * we have an inode!
1125 *
1126 * If the parent directory is seen to have changed, we throw out the
1127 * cached dentry and do a new lookup.
1128 */
1129static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1130{
1131	struct inode *dir;
1132	struct inode *inode;
1133	struct dentry *parent;
1134	struct nfs_fh *fhandle = NULL;
1135	struct nfs_fattr *fattr = NULL;
1136	struct nfs4_label *label = NULL;
1137	int error;
1138
1139	if (flags & LOOKUP_RCU) {
1140		parent = ACCESS_ONCE(dentry->d_parent);
1141		dir = d_inode_rcu(parent);
1142		if (!dir)
1143			return -ECHILD;
1144	} else {
1145		parent = dget_parent(dentry);
1146		dir = d_inode(parent);
1147	}
1148	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1149	inode = d_inode(dentry);
1150
1151	if (!inode) {
1152		if (nfs_neg_need_reval(dir, dentry, flags)) {
1153			if (flags & LOOKUP_RCU)
1154				return -ECHILD;
1155			goto out_bad;
1156		}
1157		goto out_valid;
1158	}
1159
1160	if (is_bad_inode(inode)) {
1161		if (flags & LOOKUP_RCU)
1162			return -ECHILD;
1163		dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1164				__func__, dentry);
1165		goto out_bad;
1166	}
1167
1168	if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1169		goto out_set_verifier;
1170
1171	/* Force a full look up iff the parent directory has changed */
1172	if (!nfs_is_exclusive_create(dir, flags) &&
1173	    nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1174
1175		if (nfs_lookup_verify_inode(inode, flags)) {
1176			if (flags & LOOKUP_RCU)
1177				return -ECHILD;
1178			goto out_zap_parent;
1179		}
1180		nfs_advise_use_readdirplus(dir);
1181		goto out_valid;
1182	}
1183
1184	if (flags & LOOKUP_RCU)
1185		return -ECHILD;
1186
1187	if (NFS_STALE(inode))
1188		goto out_bad;
1189
1190	error = -ENOMEM;
1191	fhandle = nfs_alloc_fhandle();
1192	fattr = nfs_alloc_fattr();
1193	if (fhandle == NULL || fattr == NULL)
1194		goto out_error;
1195
1196	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1197	if (IS_ERR(label))
1198		goto out_error;
1199
1200	trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1201	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1202	trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1203	if (error)
1204		goto out_bad;
1205	if (nfs_compare_fh(NFS_FH(inode), fhandle))
1206		goto out_bad;
1207	if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1208		goto out_bad;
1209
1210	nfs_setsecurity(inode, fattr, label);
1211
1212	nfs_free_fattr(fattr);
1213	nfs_free_fhandle(fhandle);
1214	nfs4_label_free(label);
1215
1216	/* set a readdirplus hint that we had a cache miss */
1217	nfs_force_use_readdirplus(dir);
1218
1219out_set_verifier:
1220	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1221 out_valid:
1222	if (flags & LOOKUP_RCU) {
1223		if (parent != ACCESS_ONCE(dentry->d_parent))
1224			return -ECHILD;
1225	} else
1226		dput(parent);
1227	dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1228			__func__, dentry);
1229	return 1;
1230out_zap_parent:
1231	nfs_zap_caches(dir);
1232 out_bad:
1233	WARN_ON(flags & LOOKUP_RCU);
1234	nfs_free_fattr(fattr);
1235	nfs_free_fhandle(fhandle);
1236	nfs4_label_free(label);
1237	nfs_mark_for_revalidate(dir);
1238	if (inode && S_ISDIR(inode->i_mode)) {
1239		/* Purge readdir caches. */
1240		nfs_zap_caches(inode);
1241		/*
1242		 * We can't d_drop the root of a disconnected tree:
1243		 * its d_hash is on the s_anon list and d_drop() would hide
1244		 * it from shrink_dcache_for_unmount(), leading to busy
1245		 * inodes on unmount and further oopses.
1246		 */
1247		if (IS_ROOT(dentry))
1248			goto out_valid;
 
 
 
1249	}
 
 
 
1250	dput(parent);
1251	dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1252			__func__, dentry);
 
1253	return 0;
1254out_error:
1255	WARN_ON(flags & LOOKUP_RCU);
1256	nfs_free_fattr(fattr);
1257	nfs_free_fhandle(fhandle);
1258	nfs4_label_free(label);
1259	dput(parent);
1260	dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1261			__func__, dentry, error);
 
1262	return error;
1263}
1264
1265/*
1266 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1267 * when we don't really care about the dentry name. This is called when a
1268 * pathwalk ends on a dentry that was not found via a normal lookup in the
1269 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1270 *
1271 * In this situation, we just want to verify that the inode itself is OK
1272 * since the dentry might have changed on the server.
1273 */
1274static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1275{
1276	struct inode *inode = d_inode(dentry);
1277	int error = 0;
1278
1279	/*
1280	 * I believe we can only get a negative dentry here in the case of a
1281	 * procfs-style symlink. Just assume it's correct for now, but we may
1282	 * eventually need to do something more here.
1283	 */
1284	if (!inode) {
1285		dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1286				__func__, dentry);
1287		return 1;
1288	}
1289
1290	if (is_bad_inode(inode)) {
1291		dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1292				__func__, dentry);
1293		return 0;
1294	}
1295
1296	if (nfs_mapping_need_revalidate_inode(inode))
1297		error = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1298	dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1299			__func__, inode->i_ino, error ? "invalid" : "valid");
1300	return !error;
1301}
1302
1303/*
1304 * This is called from dput() when d_count is going to 0.
1305 */
1306static int nfs_dentry_delete(const struct dentry *dentry)
1307{
1308	dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1309		dentry, dentry->d_flags);
 
1310
1311	/* Unhash any dentry with a stale inode */
1312	if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1313		return 1;
1314
1315	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1316		/* Unhash it, so that ->d_iput() would be called */
1317		return 1;
1318	}
1319	if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1320		/* Unhash it, so that ancestors of killed async unlink
1321		 * files will be cleaned up during umount */
1322		return 1;
1323	}
1324	return 0;
1325
1326}
1327
1328/* Ensure that we revalidate inode->i_nlink */
1329static void nfs_drop_nlink(struct inode *inode)
1330{
1331	spin_lock(&inode->i_lock);
1332	/* drop the inode if we're reasonably sure this is the last link */
1333	if (inode->i_nlink == 1)
1334		clear_nlink(inode);
1335	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1336	spin_unlock(&inode->i_lock);
1337}
1338
1339/*
1340 * Called when the dentry loses inode.
1341 * We use it to clean up silly-renamed files.
1342 */
1343static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1344{
1345	if (S_ISDIR(inode->i_mode))
1346		/* drop any readdir cache as it could easily be old */
1347		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1348
1349	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
 
1350		nfs_complete_unlink(dentry, inode);
1351		nfs_drop_nlink(inode);
1352	}
1353	iput(inode);
1354}
1355
1356static void nfs_d_release(struct dentry *dentry)
1357{
1358	/* free cached devname value, if it survived that far */
1359	if (unlikely(dentry->d_fsdata)) {
1360		if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1361			WARN_ON(1);
1362		else
1363			kfree(dentry->d_fsdata);
1364	}
1365}
1366
1367const struct dentry_operations nfs_dentry_operations = {
1368	.d_revalidate	= nfs_lookup_revalidate,
1369	.d_weak_revalidate	= nfs_weak_revalidate,
1370	.d_delete	= nfs_dentry_delete,
1371	.d_iput		= nfs_dentry_iput,
1372	.d_automount	= nfs_d_automount,
1373	.d_release	= nfs_d_release,
1374};
1375EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1376
1377struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1378{
1379	struct dentry *res;
 
1380	struct inode *inode = NULL;
1381	struct nfs_fh *fhandle = NULL;
1382	struct nfs_fattr *fattr = NULL;
1383	struct nfs4_label *label = NULL;
1384	int error;
1385
1386	dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
 
1387	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1388
1389	if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1390		return ERR_PTR(-ENAMETOOLONG);
 
1391
1392	/*
1393	 * If we're doing an exclusive create, optimize away the lookup
1394	 * but don't hash the dentry.
1395	 */
1396	if (nfs_is_exclusive_create(dir, flags))
1397		return NULL;
 
 
 
1398
1399	res = ERR_PTR(-ENOMEM);
1400	fhandle = nfs_alloc_fhandle();
1401	fattr = nfs_alloc_fattr();
1402	if (fhandle == NULL || fattr == NULL)
1403		goto out;
1404
1405	label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1406	if (IS_ERR(label))
1407		goto out;
1408
1409	trace_nfs_lookup_enter(dir, dentry, flags);
1410	error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1411	if (error == -ENOENT)
1412		goto no_entry;
1413	if (error < 0) {
1414		res = ERR_PTR(error);
1415		goto out_label;
1416	}
1417	inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1418	res = ERR_CAST(inode);
1419	if (IS_ERR(res))
1420		goto out_label;
1421
1422	/* Notify readdir to use READDIRPLUS */
1423	nfs_force_use_readdirplus(dir);
1424
1425no_entry:
1426	res = d_splice_alias(inode, dentry);
1427	if (res != NULL) {
1428		if (IS_ERR(res))
1429			goto out_label;
1430		dentry = res;
1431	}
1432	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1433out_label:
1434	trace_nfs_lookup_exit(dir, dentry, flags, error);
1435	nfs4_label_free(label);
1436out:
1437	nfs_free_fattr(fattr);
1438	nfs_free_fhandle(fhandle);
1439	return res;
1440}
1441EXPORT_SYMBOL_GPL(nfs_lookup);
1442
1443#if IS_ENABLED(CONFIG_NFS_V4)
1444static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1445
1446const struct dentry_operations nfs4_dentry_operations = {
1447	.d_revalidate	= nfs4_lookup_revalidate,
1448	.d_delete	= nfs_dentry_delete,
1449	.d_iput		= nfs_dentry_iput,
1450	.d_automount	= nfs_d_automount,
1451	.d_release	= nfs_d_release,
1452};
1453EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1454
1455static fmode_t flags_to_mode(int flags)
1456{
1457	fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1458	if ((flags & O_ACCMODE) != O_WRONLY)
1459		res |= FMODE_READ;
1460	if ((flags & O_ACCMODE) != O_RDONLY)
1461		res |= FMODE_WRITE;
1462	return res;
1463}
1464
1465static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1466{
1467	return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
 
 
 
 
 
 
 
 
 
 
 
1468}
1469
1470static int do_open(struct inode *inode, struct file *filp)
1471{
1472	nfs_fscache_open_file(inode, filp);
1473	return 0;
1474}
1475
1476static int nfs_finish_open(struct nfs_open_context *ctx,
1477			   struct dentry *dentry,
1478			   struct file *file, unsigned open_flags,
1479			   int *opened)
1480{
1481	int err;
1482
1483	err = finish_open(file, dentry, do_open, opened);
1484	if (err)
1485		goto out;
1486	nfs_file_set_open_context(file, ctx);
1487
 
 
 
 
 
 
 
 
 
 
 
 
 
1488out:
1489	return err;
 
1490}
1491
1492int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1493		    struct file *file, unsigned open_flags,
1494		    umode_t mode, int *opened)
1495{
1496	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1497	struct nfs_open_context *ctx;
1498	struct dentry *res;
1499	struct iattr attr = { .ia_valid = ATTR_OPEN };
1500	struct inode *inode;
1501	unsigned int lookup_flags = 0;
1502	bool switched = false;
1503	int err;
1504
1505	/* Expect a negative dentry */
1506	BUG_ON(d_inode(dentry));
1507
1508	dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1509			dir->i_sb->s_id, dir->i_ino, dentry);
1510
1511	err = nfs_check_flags(open_flags);
1512	if (err)
1513		return err;
1514
1515	/* NFS only supports OPEN on regular files */
1516	if ((open_flags & O_DIRECTORY)) {
1517		if (!d_in_lookup(dentry)) {
1518			/*
1519			 * Hashed negative dentry with O_DIRECTORY: dentry was
1520			 * revalidated and is fine, no need to perform lookup
1521			 * again
1522			 */
1523			return -ENOENT;
1524		}
1525		lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1526		goto no_open;
1527	}
1528
1529	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1530		return -ENAMETOOLONG;
1531
1532	if (open_flags & O_CREAT) {
1533		struct nfs_server *server = NFS_SERVER(dir);
1534
1535		if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1536			mode &= ~current_umask();
1537
1538		attr.ia_valid |= ATTR_MODE;
1539		attr.ia_mode = mode;
1540	}
1541	if (open_flags & O_TRUNC) {
1542		attr.ia_valid |= ATTR_SIZE;
1543		attr.ia_size = 0;
1544	}
1545
1546	if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1547		d_drop(dentry);
1548		switched = true;
1549		dentry = d_alloc_parallel(dentry->d_parent,
1550					  &dentry->d_name, &wq);
1551		if (IS_ERR(dentry))
1552			return PTR_ERR(dentry);
1553		if (unlikely(!d_in_lookup(dentry)))
1554			return finish_no_open(file, dentry);
1555	}
1556
1557	ctx = create_nfs_open_context(dentry, open_flags, file);
1558	err = PTR_ERR(ctx);
 
 
1559	if (IS_ERR(ctx))
1560		goto out;
1561
1562	trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1563	inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
 
 
 
 
 
 
 
 
 
 
1564	if (IS_ERR(inode)) {
1565		err = PTR_ERR(inode);
1566		trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1567		put_nfs_open_context(ctx);
1568		d_drop(dentry);
1569		switch (err) {
1570		case -ENOENT:
1571			d_add(dentry, NULL);
1572			nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1573			break;
1574		case -EISDIR:
1575		case -ENOTDIR:
1576			goto no_open;
1577		case -ELOOP:
1578			if (!(open_flags & O_NOFOLLOW))
1579				goto no_open;
1580			break;
 
 
 
1581			/* case -EINVAL: */
1582		default:
1583			break;
 
1584		}
1585		goto out;
1586	}
1587
1588	err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1589	trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1590	put_nfs_open_context(ctx);
1591out:
1592	if (unlikely(switched)) {
1593		d_lookup_done(dentry);
1594		dput(dentry);
1595	}
1596	return err;
1597
1598no_open:
1599	res = nfs_lookup(dir, dentry, lookup_flags);
1600	if (switched) {
1601		d_lookup_done(dentry);
1602		if (!res)
1603			res = dentry;
1604		else
1605			dput(dentry);
1606	}
1607	if (IS_ERR(res))
1608		return PTR_ERR(res);
1609	return finish_no_open(file, res);
 
 
1610}
1611EXPORT_SYMBOL_GPL(nfs_atomic_open);
1612
1613static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1614{
 
1615	struct inode *inode;
1616	int ret = 0;
 
 
1617
1618	if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1619		goto no_open;
1620	if (d_mountpoint(dentry))
1621		goto no_open;
1622	if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1623		goto no_open;
1624
1625	inode = d_inode(dentry);
 
1626
1627	/* We can't create new files in nfs_open_revalidate(), so we
1628	 * optimize away revalidation of negative dentries.
1629	 */
1630	if (inode == NULL) {
1631		struct dentry *parent;
1632		struct inode *dir;
1633
1634		if (flags & LOOKUP_RCU) {
1635			parent = ACCESS_ONCE(dentry->d_parent);
1636			dir = d_inode_rcu(parent);
1637			if (!dir)
1638				return -ECHILD;
1639		} else {
1640			parent = dget_parent(dentry);
1641			dir = d_inode(parent);
1642		}
1643		if (!nfs_neg_need_reval(dir, dentry, flags))
1644			ret = 1;
1645		else if (flags & LOOKUP_RCU)
1646			ret = -ECHILD;
1647		if (!(flags & LOOKUP_RCU))
1648			dput(parent);
1649		else if (parent != ACCESS_ONCE(dentry->d_parent))
1650			return -ECHILD;
1651		goto out;
1652	}
1653
1654	/* NFS only supports OPEN on regular files */
1655	if (!S_ISREG(inode->i_mode))
1656		goto no_open;
 
1657	/* We cannot do exclusive creation on a positive dentry */
1658	if (flags & LOOKUP_EXCL)
1659		goto no_open;
 
 
1660
1661	/* Let f_op->open() actually open (and revalidate) the file */
1662	ret = 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1663
 
 
 
 
1664out:
 
1665	return ret;
 
 
 
 
 
 
1666
 
 
1667no_open:
1668	return nfs_lookup_revalidate(dentry, flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1669}
1670
1671#endif /* CONFIG_NFSV4 */
1672
1673/*
1674 * Code common to create, mkdir, and mknod.
1675 */
1676int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1677				struct nfs_fattr *fattr,
1678				struct nfs4_label *label)
1679{
1680	struct dentry *parent = dget_parent(dentry);
1681	struct inode *dir = d_inode(parent);
1682	struct inode *inode;
1683	int error = -EACCES;
1684
1685	d_drop(dentry);
1686
1687	/* We may have been initialized further down */
1688	if (d_really_is_positive(dentry))
1689		goto out;
1690	if (fhandle->size == 0) {
1691		error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1692		if (error)
1693			goto out_error;
1694	}
1695	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1696	if (!(fattr->valid & NFS_ATTR_FATTR)) {
1697		struct nfs_server *server = NFS_SB(dentry->d_sb);
1698		error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1699		if (error < 0)
1700			goto out_error;
1701	}
1702	inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1703	error = PTR_ERR(inode);
1704	if (IS_ERR(inode))
1705		goto out_error;
1706	d_add(dentry, inode);
1707out:
1708	dput(parent);
1709	return 0;
1710out_error:
1711	nfs_mark_for_revalidate(dir);
1712	dput(parent);
1713	return error;
1714}
1715EXPORT_SYMBOL_GPL(nfs_instantiate);
1716
1717/*
1718 * Following a failed create operation, we drop the dentry rather
1719 * than retain a negative dentry. This avoids a problem in the event
1720 * that the operation succeeded on the server, but an error in the
1721 * reply path made it appear to have failed.
1722 */
1723int nfs_create(struct inode *dir, struct dentry *dentry,
1724		umode_t mode, bool excl)
1725{
1726	struct iattr attr;
1727	int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1728	int error;
 
1729
1730	dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1731			dir->i_sb->s_id, dir->i_ino, dentry);
1732
1733	attr.ia_mode = mode;
1734	attr.ia_valid = ATTR_MODE;
1735
1736	trace_nfs_create_enter(dir, dentry, open_flags);
1737	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1738	trace_nfs_create_exit(dir, dentry, open_flags, error);
 
1739	if (error != 0)
1740		goto out_err;
1741	return 0;
1742out_err:
1743	d_drop(dentry);
1744	return error;
1745}
1746EXPORT_SYMBOL_GPL(nfs_create);
1747
1748/*
1749 * See comments for nfs_proc_create regarding failed operations.
1750 */
1751int
1752nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1753{
1754	struct iattr attr;
1755	int status;
1756
1757	dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1758			dir->i_sb->s_id, dir->i_ino, dentry);
 
 
 
1759
1760	attr.ia_mode = mode;
1761	attr.ia_valid = ATTR_MODE;
1762
1763	trace_nfs_mknod_enter(dir, dentry);
1764	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1765	trace_nfs_mknod_exit(dir, dentry, status);
1766	if (status != 0)
1767		goto out_err;
1768	return 0;
1769out_err:
1770	d_drop(dentry);
1771	return status;
1772}
1773EXPORT_SYMBOL_GPL(nfs_mknod);
1774
1775/*
1776 * See comments for nfs_proc_create regarding failed operations.
1777 */
1778int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1779{
1780	struct iattr attr;
1781	int error;
1782
1783	dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1784			dir->i_sb->s_id, dir->i_ino, dentry);
1785
1786	attr.ia_valid = ATTR_MODE;
1787	attr.ia_mode = mode | S_IFDIR;
1788
1789	trace_nfs_mkdir_enter(dir, dentry);
1790	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1791	trace_nfs_mkdir_exit(dir, dentry, error);
1792	if (error != 0)
1793		goto out_err;
1794	return 0;
1795out_err:
1796	d_drop(dentry);
1797	return error;
1798}
1799EXPORT_SYMBOL_GPL(nfs_mkdir);
1800
1801static void nfs_dentry_handle_enoent(struct dentry *dentry)
1802{
1803	if (simple_positive(dentry))
1804		d_delete(dentry);
1805}
1806
1807int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1808{
1809	int error;
1810
1811	dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1812			dir->i_sb->s_id, dir->i_ino, dentry);
1813
1814	trace_nfs_rmdir_enter(dir, dentry);
1815	if (d_really_is_positive(dentry)) {
1816		down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1817		error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1818		/* Ensure the VFS deletes this inode */
1819		switch (error) {
1820		case 0:
1821			clear_nlink(d_inode(dentry));
1822			break;
1823		case -ENOENT:
1824			nfs_dentry_handle_enoent(dentry);
1825		}
1826		up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1827	} else
1828		error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1829	trace_nfs_rmdir_exit(dir, dentry, error);
1830
1831	return error;
1832}
1833EXPORT_SYMBOL_GPL(nfs_rmdir);
1834
1835/*
1836 * Remove a file after making sure there are no pending writes,
1837 * and after checking that the file has only one user. 
1838 *
1839 * We invalidate the attribute cache and free the inode prior to the operation
1840 * to avoid possible races if the server reuses the inode.
1841 */
1842static int nfs_safe_remove(struct dentry *dentry)
1843{
1844	struct inode *dir = d_inode(dentry->d_parent);
1845	struct inode *inode = d_inode(dentry);
1846	int error = -EBUSY;
1847		
1848	dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
 
1849
1850	/* If the dentry was sillyrenamed, we simply call d_delete() */
1851	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1852		error = 0;
1853		goto out;
1854	}
1855
1856	trace_nfs_remove_enter(dir, dentry);
1857	if (inode != NULL) {
1858		NFS_PROTO(inode)->return_delegation(inode);
1859		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
 
1860		if (error == 0)
1861			nfs_drop_nlink(inode);
 
1862	} else
1863		error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1864	if (error == -ENOENT)
1865		nfs_dentry_handle_enoent(dentry);
1866	trace_nfs_remove_exit(dir, dentry, error);
1867out:
1868	return error;
1869}
1870
1871/*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
1872 *  belongs to an active ".nfs..." file and we return -EBUSY.
1873 *
1874 *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
1875 */
1876int nfs_unlink(struct inode *dir, struct dentry *dentry)
1877{
1878	int error;
1879	int need_rehash = 0;
1880
1881	dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1882		dir->i_ino, dentry);
1883
1884	trace_nfs_unlink_enter(dir, dentry);
1885	spin_lock(&dentry->d_lock);
1886	if (d_count(dentry) > 1) {
1887		spin_unlock(&dentry->d_lock);
1888		/* Start asynchronous writeout of the inode */
1889		write_inode_now(d_inode(dentry), 0);
1890		error = nfs_sillyrename(dir, dentry);
1891		goto out;
1892	}
1893	if (!d_unhashed(dentry)) {
1894		__d_drop(dentry);
1895		need_rehash = 1;
1896	}
1897	spin_unlock(&dentry->d_lock);
1898	error = nfs_safe_remove(dentry);
1899	if (!error || error == -ENOENT) {
1900		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1901	} else if (need_rehash)
1902		d_rehash(dentry);
1903out:
1904	trace_nfs_unlink_exit(dir, dentry, error);
1905	return error;
1906}
1907EXPORT_SYMBOL_GPL(nfs_unlink);
1908
1909/*
1910 * To create a symbolic link, most file systems instantiate a new inode,
1911 * add a page to it containing the path, then write it out to the disk
1912 * using prepare_write/commit_write.
1913 *
1914 * Unfortunately the NFS client can't create the in-core inode first
1915 * because it needs a file handle to create an in-core inode (see
1916 * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
1917 * symlink request has completed on the server.
1918 *
1919 * So instead we allocate a raw page, copy the symname into it, then do
1920 * the SYMLINK request with the page as the buffer.  If it succeeds, we
1921 * now have a new file handle and can instantiate an in-core NFS inode
1922 * and move the raw page into its mapping.
1923 */
1924int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1925{
 
1926	struct page *page;
1927	char *kaddr;
1928	struct iattr attr;
1929	unsigned int pathlen = strlen(symname);
1930	int error;
1931
1932	dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1933		dir->i_ino, dentry, symname);
1934
1935	if (pathlen > PAGE_SIZE)
1936		return -ENAMETOOLONG;
1937
1938	attr.ia_mode = S_IFLNK | S_IRWXUGO;
1939	attr.ia_valid = ATTR_MODE;
1940
1941	page = alloc_page(GFP_USER);
1942	if (!page)
1943		return -ENOMEM;
1944
1945	kaddr = page_address(page);
1946	memcpy(kaddr, symname, pathlen);
1947	if (pathlen < PAGE_SIZE)
1948		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
 
1949
1950	trace_nfs_symlink_enter(dir, dentry);
1951	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1952	trace_nfs_symlink_exit(dir, dentry, error);
1953	if (error != 0) {
1954		dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1955			dir->i_sb->s_id, dir->i_ino,
1956			dentry, symname, error);
1957		d_drop(dentry);
1958		__free_page(page);
1959		return error;
1960	}
1961
1962	/*
1963	 * No big deal if we can't add this page to the page cache here.
1964	 * READLINK will get the missing page from the server if needed.
1965	 */
1966	if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
 
1967							GFP_KERNEL)) {
 
 
1968		SetPageUptodate(page);
1969		unlock_page(page);
1970		/*
1971		 * add_to_page_cache_lru() grabs an extra page refcount.
1972		 * Drop it here to avoid leaking this page later.
1973		 */
1974		put_page(page);
1975	} else
1976		__free_page(page);
1977
1978	return 0;
1979}
1980EXPORT_SYMBOL_GPL(nfs_symlink);
1981
1982int
1983nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1984{
1985	struct inode *inode = d_inode(old_dentry);
1986	int error;
1987
1988	dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1989		old_dentry, dentry);
 
1990
1991	trace_nfs_link_enter(inode, dir, dentry);
1992	NFS_PROTO(inode)->return_delegation(inode);
1993
1994	d_drop(dentry);
1995	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1996	if (error == 0) {
1997		ihold(inode);
1998		d_add(dentry, inode);
1999	}
2000	trace_nfs_link_exit(inode, dir, dentry, error);
2001	return error;
2002}
2003EXPORT_SYMBOL_GPL(nfs_link);
2004
2005/*
2006 * RENAME
2007 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2008 * different file handle for the same inode after a rename (e.g. when
2009 * moving to a different directory). A fail-safe method to do so would
2010 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2011 * rename the old file using the sillyrename stuff. This way, the original
2012 * file in old_dir will go away when the last process iput()s the inode.
2013 *
2014 * FIXED.
2015 * 
2016 * It actually works quite well. One needs to have the possibility for
2017 * at least one ".nfs..." file in each directory the file ever gets
2018 * moved or linked to which happens automagically with the new
2019 * implementation that only depends on the dcache stuff instead of
2020 * using the inode layer
2021 *
2022 * Unfortunately, things are a little more complicated than indicated
2023 * above. For a cross-directory move, we want to make sure we can get
2024 * rid of the old inode after the operation.  This means there must be
2025 * no pending writes (if it's a file), and the use count must be 1.
2026 * If these conditions are met, we can drop the dentries before doing
2027 * the rename.
2028 */
2029int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2030	       struct inode *new_dir, struct dentry *new_dentry,
2031	       unsigned int flags)
2032{
2033	struct inode *old_inode = d_inode(old_dentry);
2034	struct inode *new_inode = d_inode(new_dentry);
2035	struct dentry *dentry = NULL, *rehash = NULL;
2036	struct rpc_task *task;
2037	int error = -EBUSY;
2038
2039	if (flags)
2040		return -EINVAL;
2041
2042	dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2043		 old_dentry, new_dentry,
2044		 d_count(new_dentry));
2045
2046	trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2047	/*
2048	 * For non-directories, check whether the target is busy and if so,
2049	 * make a copy of the dentry and then do a silly-rename. If the
2050	 * silly-rename succeeds, the copied dentry is hashed and becomes
2051	 * the new target.
2052	 */
2053	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2054		/*
2055		 * To prevent any new references to the target during the
2056		 * rename, we unhash the dentry in advance.
2057		 */
2058		if (!d_unhashed(new_dentry)) {
2059			d_drop(new_dentry);
2060			rehash = new_dentry;
2061		}
2062
2063		if (d_count(new_dentry) > 2) {
2064			int err;
2065
2066			/* copy the target dentry's name */
2067			dentry = d_alloc(new_dentry->d_parent,
2068					 &new_dentry->d_name);
2069			if (!dentry)
2070				goto out;
2071
2072			/* silly-rename the existing target ... */
2073			err = nfs_sillyrename(new_dir, new_dentry);
2074			if (err)
2075				goto out;
2076
2077			new_dentry = dentry;
2078			rehash = NULL;
2079			new_inode = NULL;
2080		}
2081	}
2082
2083	NFS_PROTO(old_inode)->return_delegation(old_inode);
2084	if (new_inode != NULL)
2085		NFS_PROTO(new_inode)->return_delegation(new_inode);
2086
2087	task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2088	if (IS_ERR(task)) {
2089		error = PTR_ERR(task);
2090		goto out;
2091	}
2092
2093	error = rpc_wait_for_completion_task(task);
2094	if (error == 0)
2095		error = task->tk_status;
2096	rpc_put_task(task);
2097	nfs_mark_for_revalidate(old_inode);
2098out:
2099	if (rehash)
2100		d_rehash(rehash);
2101	trace_nfs_rename_exit(old_dir, old_dentry,
2102			new_dir, new_dentry, error);
2103	if (!error) {
2104		if (new_inode != NULL)
2105			nfs_drop_nlink(new_inode);
2106		d_move(old_dentry, new_dentry);
2107		nfs_set_verifier(new_dentry,
2108					nfs_save_change_attribute(new_dir));
2109	} else if (error == -ENOENT)
2110		nfs_dentry_handle_enoent(old_dentry);
2111
2112	/* new dentry created? */
2113	if (dentry)
2114		dput(dentry);
2115	return error;
2116}
2117EXPORT_SYMBOL_GPL(nfs_rename);
2118
2119static DEFINE_SPINLOCK(nfs_access_lru_lock);
2120static LIST_HEAD(nfs_access_lru_list);
2121static atomic_long_t nfs_access_nr_entries;
2122
2123static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2124module_param(nfs_access_max_cachesize, ulong, 0644);
2125MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2126
2127static void nfs_access_free_entry(struct nfs_access_entry *entry)
2128{
2129	put_rpccred(entry->cred);
2130	kfree_rcu(entry, rcu_head);
2131	smp_mb__before_atomic();
2132	atomic_long_dec(&nfs_access_nr_entries);
2133	smp_mb__after_atomic();
2134}
2135
2136static void nfs_access_free_list(struct list_head *head)
2137{
2138	struct nfs_access_entry *cache;
2139
2140	while (!list_empty(head)) {
2141		cache = list_entry(head->next, struct nfs_access_entry, lru);
2142		list_del(&cache->lru);
2143		nfs_access_free_entry(cache);
2144	}
2145}
2146
2147static unsigned long
2148nfs_do_access_cache_scan(unsigned int nr_to_scan)
2149{
2150	LIST_HEAD(head);
2151	struct nfs_inode *nfsi, *next;
2152	struct nfs_access_entry *cache;
2153	long freed = 0;
 
 
 
 
2154
2155	spin_lock(&nfs_access_lru_lock);
2156	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2157		struct inode *inode;
2158
2159		if (nr_to_scan-- == 0)
2160			break;
2161		inode = &nfsi->vfs_inode;
2162		spin_lock(&inode->i_lock);
2163		if (list_empty(&nfsi->access_cache_entry_lru))
2164			goto remove_lru_entry;
2165		cache = list_entry(nfsi->access_cache_entry_lru.next,
2166				struct nfs_access_entry, lru);
2167		list_move(&cache->lru, &head);
2168		rb_erase(&cache->rb_node, &nfsi->access_cache);
2169		freed++;
2170		if (!list_empty(&nfsi->access_cache_entry_lru))
2171			list_move_tail(&nfsi->access_cache_inode_lru,
2172					&nfs_access_lru_list);
2173		else {
2174remove_lru_entry:
2175			list_del_init(&nfsi->access_cache_inode_lru);
2176			smp_mb__before_atomic();
2177			clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2178			smp_mb__after_atomic();
2179		}
2180		spin_unlock(&inode->i_lock);
2181	}
2182	spin_unlock(&nfs_access_lru_lock);
2183	nfs_access_free_list(&head);
2184	return freed;
2185}
2186
2187unsigned long
2188nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2189{
2190	int nr_to_scan = sc->nr_to_scan;
2191	gfp_t gfp_mask = sc->gfp_mask;
2192
2193	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2194		return SHRINK_STOP;
2195	return nfs_do_access_cache_scan(nr_to_scan);
2196}
2197
2198
2199unsigned long
2200nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2201{
2202	return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2203}
2204
2205static void
2206nfs_access_cache_enforce_limit(void)
2207{
2208	long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2209	unsigned long diff;
2210	unsigned int nr_to_scan;
2211
2212	if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2213		return;
2214	nr_to_scan = 100;
2215	diff = nr_entries - nfs_access_max_cachesize;
2216	if (diff < nr_to_scan)
2217		nr_to_scan = diff;
2218	nfs_do_access_cache_scan(nr_to_scan);
2219}
2220
2221static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2222{
2223	struct rb_root *root_node = &nfsi->access_cache;
2224	struct rb_node *n;
2225	struct nfs_access_entry *entry;
2226
2227	/* Unhook entries from the cache */
2228	while ((n = rb_first(root_node)) != NULL) {
2229		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2230		rb_erase(n, root_node);
2231		list_move(&entry->lru, head);
2232	}
2233	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2234}
2235
2236void nfs_access_zap_cache(struct inode *inode)
2237{
2238	LIST_HEAD(head);
2239
2240	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2241		return;
2242	/* Remove from global LRU init */
2243	spin_lock(&nfs_access_lru_lock);
2244	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2245		list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2246
2247	spin_lock(&inode->i_lock);
2248	__nfs_access_zap_cache(NFS_I(inode), &head);
2249	spin_unlock(&inode->i_lock);
2250	spin_unlock(&nfs_access_lru_lock);
2251	nfs_access_free_list(&head);
2252}
2253EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2254
2255static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2256{
2257	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2258	struct nfs_access_entry *entry;
2259
2260	while (n != NULL) {
2261		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2262
2263		if (cred < entry->cred)
2264			n = n->rb_left;
2265		else if (cred > entry->cred)
2266			n = n->rb_right;
2267		else
2268			return entry;
2269	}
2270	return NULL;
2271}
2272
2273static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res, bool may_block)
2274{
2275	struct nfs_inode *nfsi = NFS_I(inode);
2276	struct nfs_access_entry *cache;
2277	bool retry = true;
2278	int err;
2279
2280	spin_lock(&inode->i_lock);
2281	for(;;) {
2282		if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2283			goto out_zap;
2284		cache = nfs_access_search_rbtree(inode, cred);
2285		err = -ENOENT;
2286		if (cache == NULL)
2287			goto out;
2288		/* Found an entry, is our attribute cache valid? */
2289		if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2290			break;
2291		err = -ECHILD;
2292		if (!may_block)
2293			goto out;
2294		if (!retry)
2295			goto out_zap;
2296		spin_unlock(&inode->i_lock);
2297		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2298		if (err)
2299			return err;
2300		spin_lock(&inode->i_lock);
2301		retry = false;
2302	}
2303	res->jiffies = cache->jiffies;
2304	res->cred = cache->cred;
2305	res->mask = cache->mask;
2306	list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2307	err = 0;
2308out:
2309	spin_unlock(&inode->i_lock);
2310	return err;
 
 
 
 
 
 
2311out_zap:
2312	spin_unlock(&inode->i_lock);
2313	nfs_access_zap_cache(inode);
2314	return -ENOENT;
2315}
2316
2317static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2318{
2319	/* Only check the most recently returned cache entry,
2320	 * but do it without locking.
2321	 */
2322	struct nfs_inode *nfsi = NFS_I(inode);
2323	struct nfs_access_entry *cache;
2324	int err = -ECHILD;
2325	struct list_head *lh;
2326
2327	rcu_read_lock();
2328	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2329		goto out;
2330	lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2331	cache = list_entry(lh, struct nfs_access_entry, lru);
2332	if (lh == &nfsi->access_cache_entry_lru ||
2333	    cred != cache->cred)
2334		cache = NULL;
2335	if (cache == NULL)
2336		goto out;
2337	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2338		goto out;
2339	res->jiffies = cache->jiffies;
2340	res->cred = cache->cred;
2341	res->mask = cache->mask;
2342	err = 0;
2343out:
2344	rcu_read_unlock();
2345	return err;
2346}
2347
2348static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2349{
2350	struct nfs_inode *nfsi = NFS_I(inode);
2351	struct rb_root *root_node = &nfsi->access_cache;
2352	struct rb_node **p = &root_node->rb_node;
2353	struct rb_node *parent = NULL;
2354	struct nfs_access_entry *entry;
2355
2356	spin_lock(&inode->i_lock);
2357	while (*p != NULL) {
2358		parent = *p;
2359		entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2360
2361		if (set->cred < entry->cred)
2362			p = &parent->rb_left;
2363		else if (set->cred > entry->cred)
2364			p = &parent->rb_right;
2365		else
2366			goto found;
2367	}
2368	rb_link_node(&set->rb_node, parent, p);
2369	rb_insert_color(&set->rb_node, root_node);
2370	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2371	spin_unlock(&inode->i_lock);
2372	return;
2373found:
2374	rb_replace_node(parent, &set->rb_node, root_node);
2375	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2376	list_del(&entry->lru);
2377	spin_unlock(&inode->i_lock);
2378	nfs_access_free_entry(entry);
2379}
2380
2381void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2382{
2383	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2384	if (cache == NULL)
2385		return;
2386	RB_CLEAR_NODE(&cache->rb_node);
2387	cache->jiffies = set->jiffies;
2388	cache->cred = get_rpccred(set->cred);
2389	cache->mask = set->mask;
2390
2391	/* The above field assignments must be visible
2392	 * before this item appears on the lru.  We cannot easily
2393	 * use rcu_assign_pointer, so just force the memory barrier.
2394	 */
2395	smp_wmb();
2396	nfs_access_add_rbtree(inode, cache);
2397
2398	/* Update accounting */
2399	smp_mb__before_atomic();
2400	atomic_long_inc(&nfs_access_nr_entries);
2401	smp_mb__after_atomic();
2402
2403	/* Add inode to global LRU list */
2404	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2405		spin_lock(&nfs_access_lru_lock);
2406		if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2407			list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2408					&nfs_access_lru_list);
2409		spin_unlock(&nfs_access_lru_lock);
2410	}
2411	nfs_access_cache_enforce_limit();
2412}
2413EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2414
2415void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2416{
2417	entry->mask = 0;
2418	if (access_result & NFS4_ACCESS_READ)
2419		entry->mask |= MAY_READ;
2420	if (access_result &
2421	    (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2422		entry->mask |= MAY_WRITE;
2423	if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2424		entry->mask |= MAY_EXEC;
2425}
2426EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2427
2428static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2429{
2430	struct nfs_access_entry cache;
2431	bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2432	int status;
2433
2434	trace_nfs_access_enter(inode);
2435
2436	status = nfs_access_get_cached_rcu(inode, cred, &cache);
2437	if (status != 0)
2438		status = nfs_access_get_cached(inode, cred, &cache, may_block);
2439	if (status == 0)
2440		goto out_cached;
2441
2442	status = -ECHILD;
2443	if (!may_block)
2444		goto out;
2445
2446	/* Be clever: ask server to check for all possible rights */
2447	cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2448	cache.cred = cred;
2449	cache.jiffies = jiffies;
2450	status = NFS_PROTO(inode)->access(inode, &cache);
2451	if (status != 0) {
2452		if (status == -ESTALE) {
2453			nfs_zap_caches(inode);
2454			if (!S_ISDIR(inode->i_mode))
2455				set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2456		}
2457		goto out;
2458	}
2459	nfs_access_add_cache(inode, &cache);
2460out_cached:
2461	if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2462		status = -EACCES;
2463out:
2464	trace_nfs_access_exit(inode, status);
2465	return status;
 
2466}
2467
2468static int nfs_open_permission_mask(int openflags)
2469{
2470	int mask = 0;
2471
2472	if (openflags & __FMODE_EXEC) {
2473		/* ONLY check exec rights */
2474		mask = MAY_EXEC;
2475	} else {
2476		if ((openflags & O_ACCMODE) != O_WRONLY)
2477			mask |= MAY_READ;
2478		if ((openflags & O_ACCMODE) != O_RDONLY)
2479			mask |= MAY_WRITE;
2480	}
2481
2482	return mask;
2483}
2484
2485int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2486{
2487	return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2488}
2489EXPORT_SYMBOL_GPL(nfs_may_open);
2490
2491static int nfs_execute_ok(struct inode *inode, int mask)
2492{
2493	struct nfs_server *server = NFS_SERVER(inode);
2494	int ret = 0;
2495
2496	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) {
2497		if (mask & MAY_NOT_BLOCK)
2498			return -ECHILD;
2499		ret = __nfs_revalidate_inode(server, inode);
2500	}
2501	if (ret == 0 && !execute_ok(inode))
2502		ret = -EACCES;
2503	return ret;
2504}
2505
2506int nfs_permission(struct inode *inode, int mask)
2507{
2508	struct rpc_cred *cred;
2509	int res = 0;
2510
 
 
 
2511	nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2512
2513	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2514		goto out;
2515	/* Is this sys_access() ? */
2516	if (mask & (MAY_ACCESS | MAY_CHDIR))
2517		goto force_lookup;
2518
2519	switch (inode->i_mode & S_IFMT) {
2520		case S_IFLNK:
2521			goto out;
2522		case S_IFREG:
2523			if ((mask & MAY_OPEN) &&
2524			   nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2525				return 0;
 
 
2526			break;
2527		case S_IFDIR:
2528			/*
2529			 * Optimize away all write operations, since the server
2530			 * will check permissions when we perform the op.
2531			 */
2532			if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2533				goto out;
2534	}
2535
2536force_lookup:
2537	if (!NFS_PROTO(inode)->access)
2538		goto out_notsup;
2539
2540	/* Always try fast lookups first */
2541	rcu_read_lock();
2542	cred = rpc_lookup_cred_nonblock();
2543	if (!IS_ERR(cred))
2544		res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2545	else
2546		res = PTR_ERR(cred);
2547	rcu_read_unlock();
2548	if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2549		/* Fast lookup failed, try the slow way */
2550		cred = rpc_lookup_cred();
2551		if (!IS_ERR(cred)) {
2552			res = nfs_do_access(inode, cred, mask);
2553			put_rpccred(cred);
2554		} else
2555			res = PTR_ERR(cred);
2556	}
2557out:
2558	if (!res && (mask & MAY_EXEC))
2559		res = nfs_execute_ok(inode, mask);
2560
2561	dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2562		inode->i_sb->s_id, inode->i_ino, mask, res);
2563	return res;
2564out_notsup:
2565	if (mask & MAY_NOT_BLOCK)
2566		return -ECHILD;
2567
2568	res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2569	if (res == 0)
2570		res = generic_permission(inode, mask);
2571	goto out;
2572}
2573EXPORT_SYMBOL_GPL(nfs_permission);
2574
2575/*
2576 * Local variables:
2577 *  version-control: t
2578 *  kept-new-versions: 5
2579 * End:
2580 */