Linux Audio

Check our new training course

Loading...
Note: File does not exist in v6.8.
   1// SPDX-License-Identifier: LGPL-2.1
   2/*
   3 *   fs/cifs/file.c
   4 *
   5 *   vfs operations that deal with files
   6 *
   7 *   Copyright (C) International Business Machines  Corp., 2002,2010
   8 *   Author(s): Steve French (sfrench@us.ibm.com)
   9 *              Jeremy Allison (jra@samba.org)
  10 *
  11 */
  12#include <linux/fs.h>
  13#include <linux/backing-dev.h>
  14#include <linux/stat.h>
  15#include <linux/fcntl.h>
  16#include <linux/pagemap.h>
  17#include <linux/pagevec.h>
  18#include <linux/writeback.h>
  19#include <linux/task_io_accounting_ops.h>
  20#include <linux/delay.h>
  21#include <linux/mount.h>
  22#include <linux/slab.h>
  23#include <linux/swap.h>
  24#include <linux/mm.h>
  25#include <asm/div64.h>
  26#include "cifsfs.h"
  27#include "cifspdu.h"
  28#include "cifsglob.h"
  29#include "cifsproto.h"
  30#include "cifs_unicode.h"
  31#include "cifs_debug.h"
  32#include "cifs_fs_sb.h"
  33#include "fscache.h"
  34#include "smbdirect.h"
  35#include "fs_context.h"
  36#include "cifs_ioctl.h"
  37
  38static inline int cifs_convert_flags(unsigned int flags)
  39{
  40	if ((flags & O_ACCMODE) == O_RDONLY)
  41		return GENERIC_READ;
  42	else if ((flags & O_ACCMODE) == O_WRONLY)
  43		return GENERIC_WRITE;
  44	else if ((flags & O_ACCMODE) == O_RDWR) {
  45		/* GENERIC_ALL is too much permission to request
  46		   can cause unnecessary access denied on create */
  47		/* return GENERIC_ALL; */
  48		return (GENERIC_READ | GENERIC_WRITE);
  49	}
  50
  51	return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
  52		FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
  53		FILE_READ_DATA);
  54}
  55
  56static u32 cifs_posix_convert_flags(unsigned int flags)
  57{
  58	u32 posix_flags = 0;
  59
  60	if ((flags & O_ACCMODE) == O_RDONLY)
  61		posix_flags = SMB_O_RDONLY;
  62	else if ((flags & O_ACCMODE) == O_WRONLY)
  63		posix_flags = SMB_O_WRONLY;
  64	else if ((flags & O_ACCMODE) == O_RDWR)
  65		posix_flags = SMB_O_RDWR;
  66
  67	if (flags & O_CREAT) {
  68		posix_flags |= SMB_O_CREAT;
  69		if (flags & O_EXCL)
  70			posix_flags |= SMB_O_EXCL;
  71	} else if (flags & O_EXCL)
  72		cifs_dbg(FYI, "Application %s pid %d has incorrectly set O_EXCL flag but not O_CREAT on file open. Ignoring O_EXCL\n",
  73			 current->comm, current->tgid);
  74
  75	if (flags & O_TRUNC)
  76		posix_flags |= SMB_O_TRUNC;
  77	/* be safe and imply O_SYNC for O_DSYNC */
  78	if (flags & O_DSYNC)
  79		posix_flags |= SMB_O_SYNC;
  80	if (flags & O_DIRECTORY)
  81		posix_flags |= SMB_O_DIRECTORY;
  82	if (flags & O_NOFOLLOW)
  83		posix_flags |= SMB_O_NOFOLLOW;
  84	if (flags & O_DIRECT)
  85		posix_flags |= SMB_O_DIRECT;
  86
  87	return posix_flags;
  88}
  89
  90static inline int cifs_get_disposition(unsigned int flags)
  91{
  92	if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
  93		return FILE_CREATE;
  94	else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
  95		return FILE_OVERWRITE_IF;
  96	else if ((flags & O_CREAT) == O_CREAT)
  97		return FILE_OPEN_IF;
  98	else if ((flags & O_TRUNC) == O_TRUNC)
  99		return FILE_OVERWRITE;
 100	else
 101		return FILE_OPEN;
 102}
 103
 104int cifs_posix_open(const char *full_path, struct inode **pinode,
 105			struct super_block *sb, int mode, unsigned int f_flags,
 106			__u32 *poplock, __u16 *pnetfid, unsigned int xid)
 107{
 108	int rc;
 109	FILE_UNIX_BASIC_INFO *presp_data;
 110	__u32 posix_flags = 0;
 111	struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
 112	struct cifs_fattr fattr;
 113	struct tcon_link *tlink;
 114	struct cifs_tcon *tcon;
 115
 116	cifs_dbg(FYI, "posix open %s\n", full_path);
 117
 118	presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
 119	if (presp_data == NULL)
 120		return -ENOMEM;
 121
 122	tlink = cifs_sb_tlink(cifs_sb);
 123	if (IS_ERR(tlink)) {
 124		rc = PTR_ERR(tlink);
 125		goto posix_open_ret;
 126	}
 127
 128	tcon = tlink_tcon(tlink);
 129	mode &= ~current_umask();
 130
 131	posix_flags = cifs_posix_convert_flags(f_flags);
 132	rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
 133			     poplock, full_path, cifs_sb->local_nls,
 134			     cifs_remap(cifs_sb));
 135	cifs_put_tlink(tlink);
 136
 137	if (rc)
 138		goto posix_open_ret;
 139
 140	if (presp_data->Type == cpu_to_le32(-1))
 141		goto posix_open_ret; /* open ok, caller does qpathinfo */
 142
 143	if (!pinode)
 144		goto posix_open_ret; /* caller does not need info */
 145
 146	cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
 147
 148	/* get new inode and set it up */
 149	if (*pinode == NULL) {
 150		cifs_fill_uniqueid(sb, &fattr);
 151		*pinode = cifs_iget(sb, &fattr);
 152		if (!*pinode) {
 153			rc = -ENOMEM;
 154			goto posix_open_ret;
 155		}
 156	} else {
 157		cifs_revalidate_mapping(*pinode);
 158		rc = cifs_fattr_to_inode(*pinode, &fattr);
 159	}
 160
 161posix_open_ret:
 162	kfree(presp_data);
 163	return rc;
 164}
 165
 166static int
 167cifs_nt_open(const char *full_path, struct inode *inode, struct cifs_sb_info *cifs_sb,
 168	     struct cifs_tcon *tcon, unsigned int f_flags, __u32 *oplock,
 169	     struct cifs_fid *fid, unsigned int xid)
 170{
 171	int rc;
 172	int desired_access;
 173	int disposition;
 174	int create_options = CREATE_NOT_DIR;
 175	FILE_ALL_INFO *buf;
 176	struct TCP_Server_Info *server = tcon->ses->server;
 177	struct cifs_open_parms oparms;
 178
 179	if (!server->ops->open)
 180		return -ENOSYS;
 181
 182	desired_access = cifs_convert_flags(f_flags);
 183
 184/*********************************************************************
 185 *  open flag mapping table:
 186 *
 187 *	POSIX Flag            CIFS Disposition
 188 *	----------            ----------------
 189 *	O_CREAT               FILE_OPEN_IF
 190 *	O_CREAT | O_EXCL      FILE_CREATE
 191 *	O_CREAT | O_TRUNC     FILE_OVERWRITE_IF
 192 *	O_TRUNC               FILE_OVERWRITE
 193 *	none of the above     FILE_OPEN
 194 *
 195 *	Note that there is not a direct match between disposition
 196 *	FILE_SUPERSEDE (ie create whether or not file exists although
 197 *	O_CREAT | O_TRUNC is similar but truncates the existing
 198 *	file rather than creating a new file as FILE_SUPERSEDE does
 199 *	(which uses the attributes / metadata passed in on open call)
 200 *?
 201 *?  O_SYNC is a reasonable match to CIFS writethrough flag
 202 *?  and the read write flags match reasonably.  O_LARGEFILE
 203 *?  is irrelevant because largefile support is always used
 204 *?  by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
 205 *	 O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
 206 *********************************************************************/
 207
 208	disposition = cifs_get_disposition(f_flags);
 209
 210	/* BB pass O_SYNC flag through on file attributes .. BB */
 211
 212	buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
 213	if (!buf)
 214		return -ENOMEM;
 215
 216	/* O_SYNC also has bit for O_DSYNC so following check picks up either */
 217	if (f_flags & O_SYNC)
 218		create_options |= CREATE_WRITE_THROUGH;
 219
 220	if (f_flags & O_DIRECT)
 221		create_options |= CREATE_NO_BUFFER;
 222
 223	oparms.tcon = tcon;
 224	oparms.cifs_sb = cifs_sb;
 225	oparms.desired_access = desired_access;
 226	oparms.create_options = cifs_create_options(cifs_sb, create_options);
 227	oparms.disposition = disposition;
 228	oparms.path = full_path;
 229	oparms.fid = fid;
 230	oparms.reconnect = false;
 231
 232	rc = server->ops->open(xid, &oparms, oplock, buf);
 233
 234	if (rc)
 235		goto out;
 236
 237	/* TODO: Add support for calling posix query info but with passing in fid */
 238	if (tcon->unix_ext)
 239		rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
 240					      xid);
 241	else
 242		rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
 243					 xid, fid);
 244
 245	if (rc) {
 246		server->ops->close(xid, tcon, fid);
 247		if (rc == -ESTALE)
 248			rc = -EOPENSTALE;
 249	}
 250
 251out:
 252	kfree(buf);
 253	return rc;
 254}
 255
 256static bool
 257cifs_has_mand_locks(struct cifsInodeInfo *cinode)
 258{
 259	struct cifs_fid_locks *cur;
 260	bool has_locks = false;
 261
 262	down_read(&cinode->lock_sem);
 263	list_for_each_entry(cur, &cinode->llist, llist) {
 264		if (!list_empty(&cur->locks)) {
 265			has_locks = true;
 266			break;
 267		}
 268	}
 269	up_read(&cinode->lock_sem);
 270	return has_locks;
 271}
 272
 273void
 274cifs_down_write(struct rw_semaphore *sem)
 275{
 276	while (!down_write_trylock(sem))
 277		msleep(10);
 278}
 279
 280static void cifsFileInfo_put_work(struct work_struct *work);
 281
 282struct cifsFileInfo *
 283cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
 284		  struct tcon_link *tlink, __u32 oplock)
 285{
 286	struct dentry *dentry = file_dentry(file);
 287	struct inode *inode = d_inode(dentry);
 288	struct cifsInodeInfo *cinode = CIFS_I(inode);
 289	struct cifsFileInfo *cfile;
 290	struct cifs_fid_locks *fdlocks;
 291	struct cifs_tcon *tcon = tlink_tcon(tlink);
 292	struct TCP_Server_Info *server = tcon->ses->server;
 293
 294	cfile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
 295	if (cfile == NULL)
 296		return cfile;
 297
 298	fdlocks = kzalloc(sizeof(struct cifs_fid_locks), GFP_KERNEL);
 299	if (!fdlocks) {
 300		kfree(cfile);
 301		return NULL;
 302	}
 303
 304	INIT_LIST_HEAD(&fdlocks->locks);
 305	fdlocks->cfile = cfile;
 306	cfile->llist = fdlocks;
 307
 308	cfile->count = 1;
 309	cfile->pid = current->tgid;
 310	cfile->uid = current_fsuid();
 311	cfile->dentry = dget(dentry);
 312	cfile->f_flags = file->f_flags;
 313	cfile->invalidHandle = false;
 314	cfile->deferred_close_scheduled = false;
 315	cfile->tlink = cifs_get_tlink(tlink);
 316	INIT_WORK(&cfile->oplock_break, cifs_oplock_break);
 317	INIT_WORK(&cfile->put, cifsFileInfo_put_work);
 318	INIT_DELAYED_WORK(&cfile->deferred, smb2_deferred_work_close);
 319	mutex_init(&cfile->fh_mutex);
 320	spin_lock_init(&cfile->file_info_lock);
 321
 322	cifs_sb_active(inode->i_sb);
 323
 324	/*
 325	 * If the server returned a read oplock and we have mandatory brlocks,
 326	 * set oplock level to None.
 327	 */
 328	if (server->ops->is_read_op(oplock) && cifs_has_mand_locks(cinode)) {
 329		cifs_dbg(FYI, "Reset oplock val from read to None due to mand locks\n");
 330		oplock = 0;
 331	}
 332
 333	cifs_down_write(&cinode->lock_sem);
 334	list_add(&fdlocks->llist, &cinode->llist);
 335	up_write(&cinode->lock_sem);
 336
 337	spin_lock(&tcon->open_file_lock);
 338	if (fid->pending_open->oplock != CIFS_OPLOCK_NO_CHANGE && oplock)
 339		oplock = fid->pending_open->oplock;
 340	list_del(&fid->pending_open->olist);
 341
 342	fid->purge_cache = false;
 343	server->ops->set_fid(cfile, fid, oplock);
 344
 345	list_add(&cfile->tlist, &tcon->openFileList);
 346	atomic_inc(&tcon->num_local_opens);
 347
 348	/* if readable file instance put first in list*/
 349	spin_lock(&cinode->open_file_lock);
 350	if (file->f_mode & FMODE_READ)
 351		list_add(&cfile->flist, &cinode->openFileList);
 352	else
 353		list_add_tail(&cfile->flist, &cinode->openFileList);
 354	spin_unlock(&cinode->open_file_lock);
 355	spin_unlock(&tcon->open_file_lock);
 356
 357	if (fid->purge_cache)
 358		cifs_zap_mapping(inode);
 359
 360	file->private_data = cfile;
 361	return cfile;
 362}
 363
 364struct cifsFileInfo *
 365cifsFileInfo_get(struct cifsFileInfo *cifs_file)
 366{
 367	spin_lock(&cifs_file->file_info_lock);
 368	cifsFileInfo_get_locked(cifs_file);
 369	spin_unlock(&cifs_file->file_info_lock);
 370	return cifs_file;
 371}
 372
 373static void cifsFileInfo_put_final(struct cifsFileInfo *cifs_file)
 374{
 375	struct inode *inode = d_inode(cifs_file->dentry);
 376	struct cifsInodeInfo *cifsi = CIFS_I(inode);
 377	struct cifsLockInfo *li, *tmp;
 378	struct super_block *sb = inode->i_sb;
 379
 380	/*
 381	 * Delete any outstanding lock records. We'll lose them when the file
 382	 * is closed anyway.
 383	 */
 384	cifs_down_write(&cifsi->lock_sem);
 385	list_for_each_entry_safe(li, tmp, &cifs_file->llist->locks, llist) {
 386		list_del(&li->llist);
 387		cifs_del_lock_waiters(li);
 388		kfree(li);
 389	}
 390	list_del(&cifs_file->llist->llist);
 391	kfree(cifs_file->llist);
 392	up_write(&cifsi->lock_sem);
 393
 394	cifs_put_tlink(cifs_file->tlink);
 395	dput(cifs_file->dentry);
 396	cifs_sb_deactive(sb);
 397	kfree(cifs_file);
 398}
 399
 400static void cifsFileInfo_put_work(struct work_struct *work)
 401{
 402	struct cifsFileInfo *cifs_file = container_of(work,
 403			struct cifsFileInfo, put);
 404
 405	cifsFileInfo_put_final(cifs_file);
 406}
 407
 408/**
 409 * cifsFileInfo_put - release a reference of file priv data
 410 *
 411 * Always potentially wait for oplock handler. See _cifsFileInfo_put().
 412 *
 413 * @cifs_file:	cifs/smb3 specific info (eg refcounts) for an open file
 414 */
 415void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
 416{
 417	_cifsFileInfo_put(cifs_file, true, true);
 418}
 419
 420/**
 421 * _cifsFileInfo_put - release a reference of file priv data
 422 *
 423 * This may involve closing the filehandle @cifs_file out on the
 424 * server. Must be called without holding tcon->open_file_lock,
 425 * cinode->open_file_lock and cifs_file->file_info_lock.
 426 *
 427 * If @wait_for_oplock_handler is true and we are releasing the last
 428 * reference, wait for any running oplock break handler of the file
 429 * and cancel any pending one.
 430 *
 431 * @cifs_file:	cifs/smb3 specific info (eg refcounts) for an open file
 432 * @wait_oplock_handler: must be false if called from oplock_break_handler
 433 * @offload:	not offloaded on close and oplock breaks
 434 *
 435 */
 436void _cifsFileInfo_put(struct cifsFileInfo *cifs_file,
 437		       bool wait_oplock_handler, bool offload)
 438{
 439	struct inode *inode = d_inode(cifs_file->dentry);
 440	struct cifs_tcon *tcon = tlink_tcon(cifs_file->tlink);
 441	struct TCP_Server_Info *server = tcon->ses->server;
 442	struct cifsInodeInfo *cifsi = CIFS_I(inode);
 443	struct super_block *sb = inode->i_sb;
 444	struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
 445	struct cifs_fid fid;
 446	struct cifs_pending_open open;
 447	bool oplock_break_cancelled;
 448
 449	spin_lock(&tcon->open_file_lock);
 450	spin_lock(&cifsi->open_file_lock);
 451	spin_lock(&cifs_file->file_info_lock);
 452	if (--cifs_file->count > 0) {
 453		spin_unlock(&cifs_file->file_info_lock);
 454		spin_unlock(&cifsi->open_file_lock);
 455		spin_unlock(&tcon->open_file_lock);
 456		return;
 457	}
 458	spin_unlock(&cifs_file->file_info_lock);
 459
 460	if (server->ops->get_lease_key)
 461		server->ops->get_lease_key(inode, &fid);
 462
 463	/* store open in pending opens to make sure we don't miss lease break */
 464	cifs_add_pending_open_locked(&fid, cifs_file->tlink, &open);
 465
 466	/* remove it from the lists */
 467	list_del(&cifs_file->flist);
 468	list_del(&cifs_file->tlist);
 469	atomic_dec(&tcon->num_local_opens);
 470
 471	if (list_empty(&cifsi->openFileList)) {
 472		cifs_dbg(FYI, "closing last open instance for inode %p\n",
 473			 d_inode(cifs_file->dentry));
 474		/*
 475		 * In strict cache mode we need invalidate mapping on the last
 476		 * close  because it may cause a error when we open this file
 477		 * again and get at least level II oplock.
 478		 */
 479		if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
 480			set_bit(CIFS_INO_INVALID_MAPPING, &cifsi->flags);
 481		cifs_set_oplock_level(cifsi, 0);
 482	}
 483
 484	spin_unlock(&cifsi->open_file_lock);
 485	spin_unlock(&tcon->open_file_lock);
 486
 487	oplock_break_cancelled = wait_oplock_handler ?
 488		cancel_work_sync(&cifs_file->oplock_break) : false;
 489
 490	if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
 491		struct TCP_Server_Info *server = tcon->ses->server;
 492		unsigned int xid;
 493
 494		xid = get_xid();
 495		if (server->ops->close_getattr)
 496			server->ops->close_getattr(xid, tcon, cifs_file);
 497		else if (server->ops->close)
 498			server->ops->close(xid, tcon, &cifs_file->fid);
 499		_free_xid(xid);
 500	}
 501
 502	if (oplock_break_cancelled)
 503		cifs_done_oplock_break(cifsi);
 504
 505	cifs_del_pending_open(&open);
 506
 507	if (offload)
 508		queue_work(fileinfo_put_wq, &cifs_file->put);
 509	else
 510		cifsFileInfo_put_final(cifs_file);
 511}
 512
 513int cifs_open(struct inode *inode, struct file *file)
 514
 515{
 516	int rc = -EACCES;
 517	unsigned int xid;
 518	__u32 oplock;
 519	struct cifs_sb_info *cifs_sb;
 520	struct TCP_Server_Info *server;
 521	struct cifs_tcon *tcon;
 522	struct tcon_link *tlink;
 523	struct cifsFileInfo *cfile = NULL;
 524	void *page;
 525	const char *full_path;
 526	bool posix_open_ok = false;
 527	struct cifs_fid fid;
 528	struct cifs_pending_open open;
 529
 530	xid = get_xid();
 531
 532	cifs_sb = CIFS_SB(inode->i_sb);
 533	if (unlikely(cifs_forced_shutdown(cifs_sb))) {
 534		free_xid(xid);
 535		return -EIO;
 536	}
 537
 538	tlink = cifs_sb_tlink(cifs_sb);
 539	if (IS_ERR(tlink)) {
 540		free_xid(xid);
 541		return PTR_ERR(tlink);
 542	}
 543	tcon = tlink_tcon(tlink);
 544	server = tcon->ses->server;
 545
 546	page = alloc_dentry_path();
 547	full_path = build_path_from_dentry(file_dentry(file), page);
 548	if (IS_ERR(full_path)) {
 549		rc = PTR_ERR(full_path);
 550		goto out;
 551	}
 552
 553	cifs_dbg(FYI, "inode = 0x%p file flags are 0x%x for %s\n",
 554		 inode, file->f_flags, full_path);
 555
 556	if (file->f_flags & O_DIRECT &&
 557	    cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO) {
 558		if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_BRL)
 559			file->f_op = &cifs_file_direct_nobrl_ops;
 560		else
 561			file->f_op = &cifs_file_direct_ops;
 562	}
 563
 564	/* Get the cached handle as SMB2 close is deferred */
 565	rc = cifs_get_readable_path(tcon, full_path, &cfile);
 566	if (rc == 0) {
 567		if (file->f_flags == cfile->f_flags) {
 568			file->private_data = cfile;
 569			spin_lock(&CIFS_I(inode)->deferred_lock);
 570			cifs_del_deferred_close(cfile);
 571			spin_unlock(&CIFS_I(inode)->deferred_lock);
 572			goto out;
 573		} else {
 574			_cifsFileInfo_put(cfile, true, false);
 575		}
 576	}
 577
 578	if (server->oplocks)
 579		oplock = REQ_OPLOCK;
 580	else
 581		oplock = 0;
 582
 583	if (!tcon->broken_posix_open && tcon->unix_ext &&
 584	    cap_unix(tcon->ses) && (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 585				le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 586		/* can not refresh inode info since size could be stale */
 587		rc = cifs_posix_open(full_path, &inode, inode->i_sb,
 588				cifs_sb->ctx->file_mode /* ignored */,
 589				file->f_flags, &oplock, &fid.netfid, xid);
 590		if (rc == 0) {
 591			cifs_dbg(FYI, "posix open succeeded\n");
 592			posix_open_ok = true;
 593		} else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
 594			if (tcon->ses->serverNOS)
 595				cifs_dbg(VFS, "server %s of type %s returned unexpected error on SMB posix open, disabling posix open support. Check if server update available.\n",
 596					 tcon->ses->ip_addr,
 597					 tcon->ses->serverNOS);
 598			tcon->broken_posix_open = true;
 599		} else if ((rc != -EIO) && (rc != -EREMOTE) &&
 600			 (rc != -EOPNOTSUPP)) /* path not found or net err */
 601			goto out;
 602		/*
 603		 * Else fallthrough to retry open the old way on network i/o
 604		 * or DFS errors.
 605		 */
 606	}
 607
 608	if (server->ops->get_lease_key)
 609		server->ops->get_lease_key(inode, &fid);
 610
 611	cifs_add_pending_open(&fid, tlink, &open);
 612
 613	if (!posix_open_ok) {
 614		if (server->ops->get_lease_key)
 615			server->ops->get_lease_key(inode, &fid);
 616
 617		rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
 618				  file->f_flags, &oplock, &fid, xid);
 619		if (rc) {
 620			cifs_del_pending_open(&open);
 621			goto out;
 622		}
 623	}
 624
 625	cfile = cifs_new_fileinfo(&fid, file, tlink, oplock);
 626	if (cfile == NULL) {
 627		if (server->ops->close)
 628			server->ops->close(xid, tcon, &fid);
 629		cifs_del_pending_open(&open);
 630		rc = -ENOMEM;
 631		goto out;
 632	}
 633
 634	cifs_fscache_set_inode_cookie(inode, file);
 635
 636	if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
 637		/*
 638		 * Time to set mode which we can not set earlier due to
 639		 * problems creating new read-only files.
 640		 */
 641		struct cifs_unix_set_info_args args = {
 642			.mode	= inode->i_mode,
 643			.uid	= INVALID_UID, /* no change */
 644			.gid	= INVALID_GID, /* no change */
 645			.ctime	= NO_CHANGE_64,
 646			.atime	= NO_CHANGE_64,
 647			.mtime	= NO_CHANGE_64,
 648			.device	= 0,
 649		};
 650		CIFSSMBUnixSetFileInfo(xid, tcon, &args, fid.netfid,
 651				       cfile->pid);
 652	}
 653
 654out:
 655	free_dentry_path(page);
 656	free_xid(xid);
 657	cifs_put_tlink(tlink);
 658	return rc;
 659}
 660
 661static int cifs_push_posix_locks(struct cifsFileInfo *cfile);
 662
 663/*
 664 * Try to reacquire byte range locks that were released when session
 665 * to server was lost.
 666 */
 667static int
 668cifs_relock_file(struct cifsFileInfo *cfile)
 669{
 670	struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
 671	struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 672	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
 673	int rc = 0;
 674
 675	down_read_nested(&cinode->lock_sem, SINGLE_DEPTH_NESTING);
 676	if (cinode->can_cache_brlcks) {
 677		/* can cache locks - no need to relock */
 678		up_read(&cinode->lock_sem);
 679		return rc;
 680	}
 681
 682	if (cap_unix(tcon->ses) &&
 683	    (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
 684	    ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
 685		rc = cifs_push_posix_locks(cfile);
 686	else
 687		rc = tcon->ses->server->ops->push_mand_locks(cfile);
 688
 689	up_read(&cinode->lock_sem);
 690	return rc;
 691}
 692
 693static int
 694cifs_reopen_file(struct cifsFileInfo *cfile, bool can_flush)
 695{
 696	int rc = -EACCES;
 697	unsigned int xid;
 698	__u32 oplock;
 699	struct cifs_sb_info *cifs_sb;
 700	struct cifs_tcon *tcon;
 701	struct TCP_Server_Info *server;
 702	struct cifsInodeInfo *cinode;
 703	struct inode *inode;
 704	void *page;
 705	const char *full_path;
 706	int desired_access;
 707	int disposition = FILE_OPEN;
 708	int create_options = CREATE_NOT_DIR;
 709	struct cifs_open_parms oparms;
 710
 711	xid = get_xid();
 712	mutex_lock(&cfile->fh_mutex);
 713	if (!cfile->invalidHandle) {
 714		mutex_unlock(&cfile->fh_mutex);
 715		free_xid(xid);
 716		return 0;
 717	}
 718
 719	inode = d_inode(cfile->dentry);
 720	cifs_sb = CIFS_SB(inode->i_sb);
 721	tcon = tlink_tcon(cfile->tlink);
 722	server = tcon->ses->server;
 723
 724	/*
 725	 * Can not grab rename sem here because various ops, including those
 726	 * that already have the rename sem can end up causing writepage to get
 727	 * called and if the server was down that means we end up here, and we
 728	 * can never tell if the caller already has the rename_sem.
 729	 */
 730	page = alloc_dentry_path();
 731	full_path = build_path_from_dentry(cfile->dentry, page);
 732	if (IS_ERR(full_path)) {
 733		mutex_unlock(&cfile->fh_mutex);
 734		free_dentry_path(page);
 735		free_xid(xid);
 736		return PTR_ERR(full_path);
 737	}
 738
 739	cifs_dbg(FYI, "inode = 0x%p file flags 0x%x for %s\n",
 740		 inode, cfile->f_flags, full_path);
 741
 742	if (tcon->ses->server->oplocks)
 743		oplock = REQ_OPLOCK;
 744	else
 745		oplock = 0;
 746
 747	if (tcon->unix_ext && cap_unix(tcon->ses) &&
 748	    (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 749				le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 750		/*
 751		 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
 752		 * original open. Must mask them off for a reopen.
 753		 */
 754		unsigned int oflags = cfile->f_flags &
 755						~(O_CREAT | O_EXCL | O_TRUNC);
 756
 757		rc = cifs_posix_open(full_path, NULL, inode->i_sb,
 758				     cifs_sb->ctx->file_mode /* ignored */,
 759				     oflags, &oplock, &cfile->fid.netfid, xid);
 760		if (rc == 0) {
 761			cifs_dbg(FYI, "posix reopen succeeded\n");
 762			oparms.reconnect = true;
 763			goto reopen_success;
 764		}
 765		/*
 766		 * fallthrough to retry open the old way on errors, especially
 767		 * in the reconnect path it is important to retry hard
 768		 */
 769	}
 770
 771	desired_access = cifs_convert_flags(cfile->f_flags);
 772
 773	/* O_SYNC also has bit for O_DSYNC so following check picks up either */
 774	if (cfile->f_flags & O_SYNC)
 775		create_options |= CREATE_WRITE_THROUGH;
 776
 777	if (cfile->f_flags & O_DIRECT)
 778		create_options |= CREATE_NO_BUFFER;
 779
 780	if (server->ops->get_lease_key)
 781		server->ops->get_lease_key(inode, &cfile->fid);
 782
 783	oparms.tcon = tcon;
 784	oparms.cifs_sb = cifs_sb;
 785	oparms.desired_access = desired_access;
 786	oparms.create_options = cifs_create_options(cifs_sb, create_options);
 787	oparms.disposition = disposition;
 788	oparms.path = full_path;
 789	oparms.fid = &cfile->fid;
 790	oparms.reconnect = true;
 791
 792	/*
 793	 * Can not refresh inode by passing in file_info buf to be returned by
 794	 * ops->open and then calling get_inode_info with returned buf since
 795	 * file might have write behind data that needs to be flushed and server
 796	 * version of file size can be stale. If we knew for sure that inode was
 797	 * not dirty locally we could do this.
 798	 */
 799	rc = server->ops->open(xid, &oparms, &oplock, NULL);
 800	if (rc == -ENOENT && oparms.reconnect == false) {
 801		/* durable handle timeout is expired - open the file again */
 802		rc = server->ops->open(xid, &oparms, &oplock, NULL);
 803		/* indicate that we need to relock the file */
 804		oparms.reconnect = true;
 805	}
 806
 807	if (rc) {
 808		mutex_unlock(&cfile->fh_mutex);
 809		cifs_dbg(FYI, "cifs_reopen returned 0x%x\n", rc);
 810		cifs_dbg(FYI, "oplock: %d\n", oplock);
 811		goto reopen_error_exit;
 812	}
 813
 814reopen_success:
 815	cfile->invalidHandle = false;
 816	mutex_unlock(&cfile->fh_mutex);
 817	cinode = CIFS_I(inode);
 818
 819	if (can_flush) {
 820		rc = filemap_write_and_wait(inode->i_mapping);
 821		if (!is_interrupt_error(rc))
 822			mapping_set_error(inode->i_mapping, rc);
 823
 824		if (tcon->posix_extensions)
 825			rc = smb311_posix_get_inode_info(&inode, full_path, inode->i_sb, xid);
 826		else if (tcon->unix_ext)
 827			rc = cifs_get_inode_info_unix(&inode, full_path,
 828						      inode->i_sb, xid);
 829		else
 830			rc = cifs_get_inode_info(&inode, full_path, NULL,
 831						 inode->i_sb, xid, NULL);
 832	}
 833	/*
 834	 * Else we are writing out data to server already and could deadlock if
 835	 * we tried to flush data, and since we do not know if we have data that
 836	 * would invalidate the current end of file on the server we can not go
 837	 * to the server to get the new inode info.
 838	 */
 839
 840	/*
 841	 * If the server returned a read oplock and we have mandatory brlocks,
 842	 * set oplock level to None.
 843	 */
 844	if (server->ops->is_read_op(oplock) && cifs_has_mand_locks(cinode)) {
 845		cifs_dbg(FYI, "Reset oplock val from read to None due to mand locks\n");
 846		oplock = 0;
 847	}
 848
 849	server->ops->set_fid(cfile, &cfile->fid, oplock);
 850	if (oparms.reconnect)
 851		cifs_relock_file(cfile);
 852
 853reopen_error_exit:
 854	free_dentry_path(page);
 855	free_xid(xid);
 856	return rc;
 857}
 858
 859void smb2_deferred_work_close(struct work_struct *work)
 860{
 861	struct cifsFileInfo *cfile = container_of(work,
 862			struct cifsFileInfo, deferred.work);
 863
 864	spin_lock(&CIFS_I(d_inode(cfile->dentry))->deferred_lock);
 865	cifs_del_deferred_close(cfile);
 866	cfile->deferred_close_scheduled = false;
 867	spin_unlock(&CIFS_I(d_inode(cfile->dentry))->deferred_lock);
 868	_cifsFileInfo_put(cfile, true, false);
 869}
 870
 871int cifs_close(struct inode *inode, struct file *file)
 872{
 873	struct cifsFileInfo *cfile;
 874	struct cifsInodeInfo *cinode = CIFS_I(inode);
 875	struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
 876	struct cifs_deferred_close *dclose;
 877
 878	if (file->private_data != NULL) {
 879		cfile = file->private_data;
 880		file->private_data = NULL;
 881		dclose = kmalloc(sizeof(struct cifs_deferred_close), GFP_KERNEL);
 882		if ((cinode->oplock == CIFS_CACHE_RHW_FLG) &&
 883		    cinode->lease_granted &&
 884		    !test_bit(CIFS_INO_CLOSE_ON_LOCK, &cinode->flags) &&
 885		    dclose) {
 886			if (test_bit(CIFS_INO_MODIFIED_ATTR, &cinode->flags))
 887				inode->i_ctime = inode->i_mtime = current_time(inode);
 888			spin_lock(&cinode->deferred_lock);
 889			cifs_add_deferred_close(cfile, dclose);
 890			if (cfile->deferred_close_scheduled &&
 891			    delayed_work_pending(&cfile->deferred)) {
 892				/*
 893				 * If there is no pending work, mod_delayed_work queues new work.
 894				 * So, Increase the ref count to avoid use-after-free.
 895				 */
 896				if (!mod_delayed_work(deferredclose_wq,
 897						&cfile->deferred, cifs_sb->ctx->acregmax))
 898					cifsFileInfo_get(cfile);
 899			} else {
 900				/* Deferred close for files */
 901				queue_delayed_work(deferredclose_wq,
 902						&cfile->deferred, cifs_sb->ctx->acregmax);
 903				cfile->deferred_close_scheduled = true;
 904				spin_unlock(&cinode->deferred_lock);
 905				return 0;
 906			}
 907			spin_unlock(&cinode->deferred_lock);
 908			_cifsFileInfo_put(cfile, true, false);
 909		} else {
 910			_cifsFileInfo_put(cfile, true, false);
 911			kfree(dclose);
 912		}
 913	}
 914
 915	/* return code from the ->release op is always ignored */
 916	return 0;
 917}
 918
 919void
 920cifs_reopen_persistent_handles(struct cifs_tcon *tcon)
 921{
 922	struct cifsFileInfo *open_file;
 923	struct list_head *tmp;
 924	struct list_head *tmp1;
 925	struct list_head tmp_list;
 926
 927	if (!tcon->use_persistent || !tcon->need_reopen_files)
 928		return;
 929
 930	tcon->need_reopen_files = false;
 931
 932	cifs_dbg(FYI, "Reopen persistent handles\n");
 933	INIT_LIST_HEAD(&tmp_list);
 934
 935	/* list all files open on tree connection, reopen resilient handles  */
 936	spin_lock(&tcon->open_file_lock);
 937	list_for_each(tmp, &tcon->openFileList) {
 938		open_file = list_entry(tmp, struct cifsFileInfo, tlist);
 939		if (!open_file->invalidHandle)
 940			continue;
 941		cifsFileInfo_get(open_file);
 942		list_add_tail(&open_file->rlist, &tmp_list);
 943	}
 944	spin_unlock(&tcon->open_file_lock);
 945
 946	list_for_each_safe(tmp, tmp1, &tmp_list) {
 947		open_file = list_entry(tmp, struct cifsFileInfo, rlist);
 948		if (cifs_reopen_file(open_file, false /* do not flush */))
 949			tcon->need_reopen_files = true;
 950		list_del_init(&open_file->rlist);
 951		cifsFileInfo_put(open_file);
 952	}
 953}
 954
 955int cifs_closedir(struct inode *inode, struct file *file)
 956{
 957	int rc = 0;
 958	unsigned int xid;
 959	struct cifsFileInfo *cfile = file->private_data;
 960	struct cifs_tcon *tcon;
 961	struct TCP_Server_Info *server;
 962	char *buf;
 963
 964	cifs_dbg(FYI, "Closedir inode = 0x%p\n", inode);
 965
 966	if (cfile == NULL)
 967		return rc;
 968
 969	xid = get_xid();
 970	tcon = tlink_tcon(cfile->tlink);
 971	server = tcon->ses->server;
 972
 973	cifs_dbg(FYI, "Freeing private data in close dir\n");
 974	spin_lock(&cfile->file_info_lock);
 975	if (server->ops->dir_needs_close(cfile)) {
 976		cfile->invalidHandle = true;
 977		spin_unlock(&cfile->file_info_lock);
 978		if (server->ops->close_dir)
 979			rc = server->ops->close_dir(xid, tcon, &cfile->fid);
 980		else
 981			rc = -ENOSYS;
 982		cifs_dbg(FYI, "Closing uncompleted readdir with rc %d\n", rc);
 983		/* not much we can do if it fails anyway, ignore rc */
 984		rc = 0;
 985	} else
 986		spin_unlock(&cfile->file_info_lock);
 987
 988	buf = cfile->srch_inf.ntwrk_buf_start;
 989	if (buf) {
 990		cifs_dbg(FYI, "closedir free smb buf in srch struct\n");
 991		cfile->srch_inf.ntwrk_buf_start = NULL;
 992		if (cfile->srch_inf.smallBuf)
 993			cifs_small_buf_release(buf);
 994		else
 995			cifs_buf_release(buf);
 996	}
 997
 998	cifs_put_tlink(cfile->tlink);
 999	kfree(file->private_data);
1000	file->private_data = NULL;
1001	/* BB can we lock the filestruct while this is going on? */
1002	free_xid(xid);
1003	return rc;
1004}
1005
1006static struct cifsLockInfo *
1007cifs_lock_init(__u64 offset, __u64 length, __u8 type, __u16 flags)
1008{
1009	struct cifsLockInfo *lock =
1010		kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
1011	if (!lock)
1012		return lock;
1013	lock->offset = offset;
1014	lock->length = length;
1015	lock->type = type;
1016	lock->pid = current->tgid;
1017	lock->flags = flags;
1018	INIT_LIST_HEAD(&lock->blist);
1019	init_waitqueue_head(&lock->block_q);
1020	return lock;
1021}
1022
1023void
1024cifs_del_lock_waiters(struct cifsLockInfo *lock)
1025{
1026	struct cifsLockInfo *li, *tmp;
1027	list_for_each_entry_safe(li, tmp, &lock->blist, blist) {
1028		list_del_init(&li->blist);
1029		wake_up(&li->block_q);
1030	}
1031}
1032
1033#define CIFS_LOCK_OP	0
1034#define CIFS_READ_OP	1
1035#define CIFS_WRITE_OP	2
1036
1037/* @rw_check : 0 - no op, 1 - read, 2 - write */
1038static bool
1039cifs_find_fid_lock_conflict(struct cifs_fid_locks *fdlocks, __u64 offset,
1040			    __u64 length, __u8 type, __u16 flags,
1041			    struct cifsFileInfo *cfile,
1042			    struct cifsLockInfo **conf_lock, int rw_check)
1043{
1044	struct cifsLockInfo *li;
1045	struct cifsFileInfo *cur_cfile = fdlocks->cfile;
1046	struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
1047
1048	list_for_each_entry(li, &fdlocks->locks, llist) {
1049		if (offset + length <= li->offset ||
1050		    offset >= li->offset + li->length)
1051			continue;
1052		if (rw_check != CIFS_LOCK_OP && current->tgid == li->pid &&
1053		    server->ops->compare_fids(cfile, cur_cfile)) {
1054			/* shared lock prevents write op through the same fid */
1055			if (!(li->type & server->vals->shared_lock_type) ||
1056			    rw_check != CIFS_WRITE_OP)
1057				continue;
1058		}
1059		if ((type & server->vals->shared_lock_type) &&
1060		    ((server->ops->compare_fids(cfile, cur_cfile) &&
1061		     current->tgid == li->pid) || type == li->type))
1062			continue;
1063		if (rw_check == CIFS_LOCK_OP &&
1064		    (flags & FL_OFDLCK) && (li->flags & FL_OFDLCK) &&
1065		    server->ops->compare_fids(cfile, cur_cfile))
1066			continue;
1067		if (conf_lock)
1068			*conf_lock = li;
1069		return true;
1070	}
1071	return false;
1072}
1073
1074bool
1075cifs_find_lock_conflict(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
1076			__u8 type, __u16 flags,
1077			struct cifsLockInfo **conf_lock, int rw_check)
1078{
1079	bool rc = false;
1080	struct cifs_fid_locks *cur;
1081	struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1082
1083	list_for_each_entry(cur, &cinode->llist, llist) {
1084		rc = cifs_find_fid_lock_conflict(cur, offset, length, type,
1085						 flags, cfile, conf_lock,
1086						 rw_check);
1087		if (rc)
1088			break;
1089	}
1090
1091	return rc;
1092}
1093
1094/*
1095 * Check if there is another lock that prevents us to set the lock (mandatory
1096 * style). If such a lock exists, update the flock structure with its
1097 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
1098 * or leave it the same if we can't. Returns 0 if we don't need to request to
1099 * the server or 1 otherwise.
1100 */
1101static int
1102cifs_lock_test(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
1103	       __u8 type, struct file_lock *flock)
1104{
1105	int rc = 0;
1106	struct cifsLockInfo *conf_lock;
1107	struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1108	struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
1109	bool exist;
1110
1111	down_read(&cinode->lock_sem);
1112
1113	exist = cifs_find_lock_conflict(cfile, offset, length, type,
1114					flock->fl_flags, &conf_lock,
1115					CIFS_LOCK_OP);
1116	if (exist) {
1117		flock->fl_start = conf_lock->offset;
1118		flock->fl_end = conf_lock->offset + conf_lock->length - 1;
1119		flock->fl_pid = conf_lock->pid;
1120		if (conf_lock->type & server->vals->shared_lock_type)
1121			flock->fl_type = F_RDLCK;
1122		else
1123			flock->fl_type = F_WRLCK;
1124	} else if (!cinode->can_cache_brlcks)
1125		rc = 1;
1126	else
1127		flock->fl_type = F_UNLCK;
1128
1129	up_read(&cinode->lock_sem);
1130	return rc;
1131}
1132
1133static void
1134cifs_lock_add(struct cifsFileInfo *cfile, struct cifsLockInfo *lock)
1135{
1136	struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1137	cifs_down_write(&cinode->lock_sem);
1138	list_add_tail(&lock->llist, &cfile->llist->locks);
1139	up_write(&cinode->lock_sem);
1140}
1141
1142/*
1143 * Set the byte-range lock (mandatory style). Returns:
1144 * 1) 0, if we set the lock and don't need to request to the server;
1145 * 2) 1, if no locks prevent us but we need to request to the server;
1146 * 3) -EACCES, if there is a lock that prevents us and wait is false.
1147 */
1148static int
1149cifs_lock_add_if(struct cifsFileInfo *cfile, struct cifsLockInfo *lock,
1150		 bool wait)
1151{
1152	struct cifsLockInfo *conf_lock;
1153	struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1154	bool exist;
1155	int rc = 0;
1156
1157try_again:
1158	exist = false;
1159	cifs_down_write(&cinode->lock_sem);
1160
1161	exist = cifs_find_lock_conflict(cfile, lock->offset, lock->length,
1162					lock->type, lock->flags, &conf_lock,
1163					CIFS_LOCK_OP);
1164	if (!exist && cinode->can_cache_brlcks) {
1165		list_add_tail(&lock->llist, &cfile->llist->locks);
1166		up_write(&cinode->lock_sem);
1167		return rc;
1168	}
1169
1170	if (!exist)
1171		rc = 1;
1172	else if (!wait)
1173		rc = -EACCES;
1174	else {
1175		list_add_tail(&lock->blist, &conf_lock->blist);
1176		up_write(&cinode->lock_sem);
1177		rc = wait_event_interruptible(lock->block_q,
1178					(lock->blist.prev == &lock->blist) &&
1179					(lock->blist.next == &lock->blist));
1180		if (!rc)
1181			goto try_again;
1182		cifs_down_write(&cinode->lock_sem);
1183		list_del_init(&lock->blist);
1184	}
1185
1186	up_write(&cinode->lock_sem);
1187	return rc;
1188}
1189
1190/*
1191 * Check if there is another lock that prevents us to set the lock (posix
1192 * style). If such a lock exists, update the flock structure with its
1193 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
1194 * or leave it the same if we can't. Returns 0 if we don't need to request to
1195 * the server or 1 otherwise.
1196 */
1197static int
1198cifs_posix_lock_test(struct file *file, struct file_lock *flock)
1199{
1200	int rc = 0;
1201	struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
1202	unsigned char saved_type = flock->fl_type;
1203
1204	if ((flock->fl_flags & FL_POSIX) == 0)
1205		return 1;
1206
1207	down_read(&cinode->lock_sem);
1208	posix_test_lock(file, flock);
1209
1210	if (flock->fl_type == F_UNLCK && !cinode->can_cache_brlcks) {
1211		flock->fl_type = saved_type;
1212		rc = 1;
1213	}
1214
1215	up_read(&cinode->lock_sem);
1216	return rc;
1217}
1218
1219/*
1220 * Set the byte-range lock (posix style). Returns:
1221 * 1) <0, if the error occurs while setting the lock;
1222 * 2) 0, if we set the lock and don't need to request to the server;
1223 * 3) FILE_LOCK_DEFERRED, if we will wait for some other file_lock;
1224 * 4) FILE_LOCK_DEFERRED + 1, if we need to request to the server.
1225 */
1226static int
1227cifs_posix_lock_set(struct file *file, struct file_lock *flock)
1228{
1229	struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
1230	int rc = FILE_LOCK_DEFERRED + 1;
1231
1232	if ((flock->fl_flags & FL_POSIX) == 0)
1233		return rc;
1234
1235	cifs_down_write(&cinode->lock_sem);
1236	if (!cinode->can_cache_brlcks) {
1237		up_write(&cinode->lock_sem);
1238		return rc;
1239	}
1240
1241	rc = posix_lock_file(file, flock, NULL);
1242	up_write(&cinode->lock_sem);
1243	return rc;
1244}
1245
1246int
1247cifs_push_mandatory_locks(struct cifsFileInfo *cfile)
1248{
1249	unsigned int xid;
1250	int rc = 0, stored_rc;
1251	struct cifsLockInfo *li, *tmp;
1252	struct cifs_tcon *tcon;
1253	unsigned int num, max_num, max_buf;
1254	LOCKING_ANDX_RANGE *buf, *cur;
1255	static const int types[] = {
1256		LOCKING_ANDX_LARGE_FILES,
1257		LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES
1258	};
1259	int i;
1260
1261	xid = get_xid();
1262	tcon = tlink_tcon(cfile->tlink);
1263
1264	/*
1265	 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1266	 * and check it before using.
1267	 */
1268	max_buf = tcon->ses->server->maxBuf;
1269	if (max_buf < (sizeof(struct smb_hdr) + sizeof(LOCKING_ANDX_RANGE))) {
1270		free_xid(xid);
1271		return -EINVAL;
1272	}
1273
1274	BUILD_BUG_ON(sizeof(struct smb_hdr) + sizeof(LOCKING_ANDX_RANGE) >
1275		     PAGE_SIZE);
1276	max_buf = min_t(unsigned int, max_buf - sizeof(struct smb_hdr),
1277			PAGE_SIZE);
1278	max_num = (max_buf - sizeof(struct smb_hdr)) /
1279						sizeof(LOCKING_ANDX_RANGE);
1280	buf = kcalloc(max_num, sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1281	if (!buf) {
1282		free_xid(xid);
1283		return -ENOMEM;
1284	}
1285
1286	for (i = 0; i < 2; i++) {
1287		cur = buf;
1288		num = 0;
1289		list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1290			if (li->type != types[i])
1291				continue;
1292			cur->Pid = cpu_to_le16(li->pid);
1293			cur->LengthLow = cpu_to_le32((u32)li->length);
1294			cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1295			cur->OffsetLow = cpu_to_le32((u32)li->offset);
1296			cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1297			if (++num == max_num) {
1298				stored_rc = cifs_lockv(xid, tcon,
1299						       cfile->fid.netfid,
1300						       (__u8)li->type, 0, num,
1301						       buf);
1302				if (stored_rc)
1303					rc = stored_rc;
1304				cur = buf;
1305				num = 0;
1306			} else
1307				cur++;
1308		}
1309
1310		if (num) {
1311			stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1312					       (__u8)types[i], 0, num, buf);
1313			if (stored_rc)
1314				rc = stored_rc;
1315		}
1316	}
1317
1318	kfree(buf);
1319	free_xid(xid);
1320	return rc;
1321}
1322
1323static __u32
1324hash_lockowner(fl_owner_t owner)
1325{
1326	return cifs_lock_secret ^ hash32_ptr((const void *)owner);
1327}
1328
1329struct lock_to_push {
1330	struct list_head llist;
1331	__u64 offset;
1332	__u64 length;
1333	__u32 pid;
1334	__u16 netfid;
1335	__u8 type;
1336};
1337
1338static int
1339cifs_push_posix_locks(struct cifsFileInfo *cfile)
1340{
1341	struct inode *inode = d_inode(cfile->dentry);
1342	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1343	struct file_lock *flock;
1344	struct file_lock_context *flctx = inode->i_flctx;
1345	unsigned int count = 0, i;
1346	int rc = 0, xid, type;
1347	struct list_head locks_to_send, *el;
1348	struct lock_to_push *lck, *tmp;
1349	__u64 length;
1350
1351	xid = get_xid();
1352
1353	if (!flctx)
1354		goto out;
1355
1356	spin_lock(&flctx->flc_lock);
1357	list_for_each(el, &flctx->flc_posix) {
1358		count++;
1359	}
1360	spin_unlock(&flctx->flc_lock);
1361
1362	INIT_LIST_HEAD(&locks_to_send);
1363
1364	/*
1365	 * Allocating count locks is enough because no FL_POSIX locks can be
1366	 * added to the list while we are holding cinode->lock_sem that
1367	 * protects locking operations of this inode.
1368	 */
1369	for (i = 0; i < count; i++) {
1370		lck = kmalloc(sizeof(struct lock_to_push), GFP_KERNEL);
1371		if (!lck) {
1372			rc = -ENOMEM;
1373			goto err_out;
1374		}
1375		list_add_tail(&lck->llist, &locks_to_send);
1376	}
1377
1378	el = locks_to_send.next;
1379	spin_lock(&flctx->flc_lock);
1380	list_for_each_entry(flock, &flctx->flc_posix, fl_list) {
1381		if (el == &locks_to_send) {
1382			/*
1383			 * The list ended. We don't have enough allocated
1384			 * structures - something is really wrong.
1385			 */
1386			cifs_dbg(VFS, "Can't push all brlocks!\n");
1387			break;
1388		}
1389		length = 1 + flock->fl_end - flock->fl_start;
1390		if (flock->fl_type == F_RDLCK || flock->fl_type == F_SHLCK)
1391			type = CIFS_RDLCK;
1392		else
1393			type = CIFS_WRLCK;
1394		lck = list_entry(el, struct lock_to_push, llist);
1395		lck->pid = hash_lockowner(flock->fl_owner);
1396		lck->netfid = cfile->fid.netfid;
1397		lck->length = length;
1398		lck->type = type;
1399		lck->offset = flock->fl_start;
1400	}
1401	spin_unlock(&flctx->flc_lock);
1402
1403	list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1404		int stored_rc;
1405
1406		stored_rc = CIFSSMBPosixLock(xid, tcon, lck->netfid, lck->pid,
1407					     lck->offset, lck->length, NULL,
1408					     lck->type, 0);
1409		if (stored_rc)
1410			rc = stored_rc;
1411		list_del(&lck->llist);
1412		kfree(lck);
1413	}
1414
1415out:
1416	free_xid(xid);
1417	return rc;
1418err_out:
1419	list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1420		list_del(&lck->llist);
1421		kfree(lck);
1422	}
1423	goto out;
1424}
1425
1426static int
1427cifs_push_locks(struct cifsFileInfo *cfile)
1428{
1429	struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
1430	struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1431	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1432	int rc = 0;
1433
1434	/* we are going to update can_cache_brlcks here - need a write access */
1435	cifs_down_write(&cinode->lock_sem);
1436	if (!cinode->can_cache_brlcks) {
1437		up_write(&cinode->lock_sem);
1438		return rc;
1439	}
1440
1441	if (cap_unix(tcon->ses) &&
1442	    (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1443	    ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1444		rc = cifs_push_posix_locks(cfile);
1445	else
1446		rc = tcon->ses->server->ops->push_mand_locks(cfile);
1447
1448	cinode->can_cache_brlcks = false;
1449	up_write(&cinode->lock_sem);
1450	return rc;
1451}
1452
1453static void
1454cifs_read_flock(struct file_lock *flock, __u32 *type, int *lock, int *unlock,
1455		bool *wait_flag, struct TCP_Server_Info *server)
1456{
1457	if (flock->fl_flags & FL_POSIX)
1458		cifs_dbg(FYI, "Posix\n");
1459	if (flock->fl_flags & FL_FLOCK)
1460		cifs_dbg(FYI, "Flock\n");
1461	if (flock->fl_flags & FL_SLEEP) {
1462		cifs_dbg(FYI, "Blocking lock\n");
1463		*wait_flag = true;
1464	}
1465	if (flock->fl_flags & FL_ACCESS)
1466		cifs_dbg(FYI, "Process suspended by mandatory locking - not implemented yet\n");
1467	if (flock->fl_flags & FL_LEASE)
1468		cifs_dbg(FYI, "Lease on file - not implemented yet\n");
1469	if (flock->fl_flags &
1470	    (~(FL_POSIX | FL_FLOCK | FL_SLEEP |
1471	       FL_ACCESS | FL_LEASE | FL_CLOSE | FL_OFDLCK)))
1472		cifs_dbg(FYI, "Unknown lock flags 0x%x\n", flock->fl_flags);
1473
1474	*type = server->vals->large_lock_type;
1475	if (flock->fl_type == F_WRLCK) {
1476		cifs_dbg(FYI, "F_WRLCK\n");
1477		*type |= server->vals->exclusive_lock_type;
1478		*lock = 1;
1479	} else if (flock->fl_type == F_UNLCK) {
1480		cifs_dbg(FYI, "F_UNLCK\n");
1481		*type |= server->vals->unlock_lock_type;
1482		*unlock = 1;
1483		/* Check if unlock includes more than one lock range */
1484	} else if (flock->fl_type == F_RDLCK) {
1485		cifs_dbg(FYI, "F_RDLCK\n");
1486		*type |= server->vals->shared_lock_type;
1487		*lock = 1;
1488	} else if (flock->fl_type == F_EXLCK) {
1489		cifs_dbg(FYI, "F_EXLCK\n");
1490		*type |= server->vals->exclusive_lock_type;
1491		*lock = 1;
1492	} else if (flock->fl_type == F_SHLCK) {
1493		cifs_dbg(FYI, "F_SHLCK\n");
1494		*type |= server->vals->shared_lock_type;
1495		*lock = 1;
1496	} else
1497		cifs_dbg(FYI, "Unknown type of lock\n");
1498}
1499
1500static int
1501cifs_getlk(struct file *file, struct file_lock *flock, __u32 type,
1502	   bool wait_flag, bool posix_lck, unsigned int xid)
1503{
1504	int rc = 0;
1505	__u64 length = 1 + flock->fl_end - flock->fl_start;
1506	struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1507	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1508	struct TCP_Server_Info *server = tcon->ses->server;
1509	__u16 netfid = cfile->fid.netfid;
1510
1511	if (posix_lck) {
1512		int posix_lock_type;
1513
1514		rc = cifs_posix_lock_test(file, flock);
1515		if (!rc)
1516			return rc;
1517
1518		if (type & server->vals->shared_lock_type)
1519			posix_lock_type = CIFS_RDLCK;
1520		else
1521			posix_lock_type = CIFS_WRLCK;
1522		rc = CIFSSMBPosixLock(xid, tcon, netfid,
1523				      hash_lockowner(flock->fl_owner),
1524				      flock->fl_start, length, flock,
1525				      posix_lock_type, wait_flag);
1526		return rc;
1527	}
1528
1529	rc = cifs_lock_test(cfile, flock->fl_start, length, type, flock);
1530	if (!rc)
1531		return rc;
1532
1533	/* BB we could chain these into one lock request BB */
1534	rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length, type,
1535				    1, 0, false);
1536	if (rc == 0) {
1537		rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1538					    type, 0, 1, false);
1539		flock->fl_type = F_UNLCK;
1540		if (rc != 0)
1541			cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1542				 rc);
1543		return 0;
1544	}
1545
1546	if (type & server->vals->shared_lock_type) {
1547		flock->fl_type = F_WRLCK;
1548		return 0;
1549	}
1550
1551	type &= ~server->vals->exclusive_lock_type;
1552
1553	rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1554				    type | server->vals->shared_lock_type,
1555				    1, 0, false);
1556	if (rc == 0) {
1557		rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1558			type | server->vals->shared_lock_type, 0, 1, false);
1559		flock->fl_type = F_RDLCK;
1560		if (rc != 0)
1561			cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1562				 rc);
1563	} else
1564		flock->fl_type = F_WRLCK;
1565
1566	return 0;
1567}
1568
1569void
1570cifs_move_llist(struct list_head *source, struct list_head *dest)
1571{
1572	struct list_head *li, *tmp;
1573	list_for_each_safe(li, tmp, source)
1574		list_move(li, dest);
1575}
1576
1577void
1578cifs_free_llist(struct list_head *llist)
1579{
1580	struct cifsLockInfo *li, *tmp;
1581	list_for_each_entry_safe(li, tmp, llist, llist) {
1582		cifs_del_lock_waiters(li);
1583		list_del(&li->llist);
1584		kfree(li);
1585	}
1586}
1587
1588int
1589cifs_unlock_range(struct cifsFileInfo *cfile, struct file_lock *flock,
1590		  unsigned int xid)
1591{
1592	int rc = 0, stored_rc;
1593	static const int types[] = {
1594		LOCKING_ANDX_LARGE_FILES,
1595		LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES
1596	};
1597	unsigned int i;
1598	unsigned int max_num, num, max_buf;
1599	LOCKING_ANDX_RANGE *buf, *cur;
1600	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1601	struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1602	struct cifsLockInfo *li, *tmp;
1603	__u64 length = 1 + flock->fl_end - flock->fl_start;
1604	struct list_head tmp_llist;
1605
1606	INIT_LIST_HEAD(&tmp_llist);
1607
1608	/*
1609	 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1610	 * and check it before using.
1611	 */
1612	max_buf = tcon->ses->server->maxBuf;
1613	if (max_buf < (sizeof(struct smb_hdr) + sizeof(LOCKING_ANDX_RANGE)))
1614		return -EINVAL;
1615
1616	BUILD_BUG_ON(sizeof(struct smb_hdr) + sizeof(LOCKING_ANDX_RANGE) >
1617		     PAGE_SIZE);
1618	max_buf = min_t(unsigned int, max_buf - sizeof(struct smb_hdr),
1619			PAGE_SIZE);
1620	max_num = (max_buf - sizeof(struct smb_hdr)) /
1621						sizeof(LOCKING_ANDX_RANGE);
1622	buf = kcalloc(max_num, sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1623	if (!buf)
1624		return -ENOMEM;
1625
1626	cifs_down_write(&cinode->lock_sem);
1627	for (i = 0; i < 2; i++) {
1628		cur = buf;
1629		num = 0;
1630		list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1631			if (flock->fl_start > li->offset ||
1632			    (flock->fl_start + length) <
1633			    (li->offset + li->length))
1634				continue;
1635			if (current->tgid != li->pid)
1636				continue;
1637			if (types[i] != li->type)
1638				continue;
1639			if (cinode->can_cache_brlcks) {
1640				/*
1641				 * We can cache brlock requests - simply remove
1642				 * a lock from the file's list.
1643				 */
1644				list_del(&li->llist);
1645				cifs_del_lock_waiters(li);
1646				kfree(li);
1647				continue;
1648			}
1649			cur->Pid = cpu_to_le16(li->pid);
1650			cur->LengthLow = cpu_to_le32((u32)li->length);
1651			cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1652			cur->OffsetLow = cpu_to_le32((u32)li->offset);
1653			cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1654			/*
1655			 * We need to save a lock here to let us add it again to
1656			 * the file's list if the unlock range request fails on
1657			 * the server.
1658			 */
1659			list_move(&li->llist, &tmp_llist);
1660			if (++num == max_num) {
1661				stored_rc = cifs_lockv(xid, tcon,
1662						       cfile->fid.netfid,
1663						       li->type, num, 0, buf);
1664				if (stored_rc) {
1665					/*
1666					 * We failed on the unlock range
1667					 * request - add all locks from the tmp
1668					 * list to the head of the file's list.
1669					 */
1670					cifs_move_llist(&tmp_llist,
1671							&cfile->llist->locks);
1672					rc = stored_rc;
1673				} else
1674					/*
1675					 * The unlock range request succeed -
1676					 * free the tmp list.
1677					 */
1678					cifs_free_llist(&tmp_llist);
1679				cur = buf;
1680				num = 0;
1681			} else
1682				cur++;
1683		}
1684		if (num) {
1685			stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1686					       types[i], num, 0, buf);
1687			if (stored_rc) {
1688				cifs_move_llist(&tmp_llist,
1689						&cfile->llist->locks);
1690				rc = stored_rc;
1691			} else
1692				cifs_free_llist(&tmp_llist);
1693		}
1694	}
1695
1696	up_write(&cinode->lock_sem);
1697	kfree(buf);
1698	return rc;
1699}
1700
1701static int
1702cifs_setlk(struct file *file, struct file_lock *flock, __u32 type,
1703	   bool wait_flag, bool posix_lck, int lock, int unlock,
1704	   unsigned int xid)
1705{
1706	int rc = 0;
1707	__u64 length = 1 + flock->fl_end - flock->fl_start;
1708	struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1709	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1710	struct TCP_Server_Info *server = tcon->ses->server;
1711	struct inode *inode = d_inode(cfile->dentry);
1712
1713	if (posix_lck) {
1714		int posix_lock_type;
1715
1716		rc = cifs_posix_lock_set(file, flock);
1717		if (rc <= FILE_LOCK_DEFERRED)
1718			return rc;
1719
1720		if (type & server->vals->shared_lock_type)
1721			posix_lock_type = CIFS_RDLCK;
1722		else
1723			posix_lock_type = CIFS_WRLCK;
1724
1725		if (unlock == 1)
1726			posix_lock_type = CIFS_UNLCK;
1727
1728		rc = CIFSSMBPosixLock(xid, tcon, cfile->fid.netfid,
1729				      hash_lockowner(flock->fl_owner),
1730				      flock->fl_start, length,
1731				      NULL, posix_lock_type, wait_flag);
1732		goto out;
1733	}
1734
1735	if (lock) {
1736		struct cifsLockInfo *lock;
1737
1738		lock = cifs_lock_init(flock->fl_start, length, type,
1739				      flock->fl_flags);
1740		if (!lock)
1741			return -ENOMEM;
1742
1743		rc = cifs_lock_add_if(cfile, lock, wait_flag);
1744		if (rc < 0) {
1745			kfree(lock);
1746			return rc;
1747		}
1748		if (!rc)
1749			goto out;
1750
1751		/*
1752		 * Windows 7 server can delay breaking lease from read to None
1753		 * if we set a byte-range lock on a file - break it explicitly
1754		 * before sending the lock to the server to be sure the next
1755		 * read won't conflict with non-overlapted locks due to
1756		 * pagereading.
1757		 */
1758		if (!CIFS_CACHE_WRITE(CIFS_I(inode)) &&
1759					CIFS_CACHE_READ(CIFS_I(inode))) {
1760			cifs_zap_mapping(inode);
1761			cifs_dbg(FYI, "Set no oplock for inode=%p due to mand locks\n",
1762				 inode);
1763			CIFS_I(inode)->oplock = 0;
1764		}
1765
1766		rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1767					    type, 1, 0, wait_flag);
1768		if (rc) {
1769			kfree(lock);
1770			return rc;
1771		}
1772
1773		cifs_lock_add(cfile, lock);
1774	} else if (unlock)
1775		rc = server->ops->mand_unlock_range(cfile, flock, xid);
1776
1777out:
1778	if ((flock->fl_flags & FL_POSIX) || (flock->fl_flags & FL_FLOCK)) {
1779		/*
1780		 * If this is a request to remove all locks because we
1781		 * are closing the file, it doesn't matter if the
1782		 * unlocking failed as both cifs.ko and the SMB server
1783		 * remove the lock on file close
1784		 */
1785		if (rc) {
1786			cifs_dbg(VFS, "%s failed rc=%d\n", __func__, rc);
1787			if (!(flock->fl_flags & FL_CLOSE))
1788				return rc;
1789		}
1790		rc = locks_lock_file_wait(file, flock);
1791	}
1792	return rc;
1793}
1794
1795int cifs_flock(struct file *file, int cmd, struct file_lock *fl)
1796{
1797	int rc, xid;
1798	int lock = 0, unlock = 0;
1799	bool wait_flag = false;
1800	bool posix_lck = false;
1801	struct cifs_sb_info *cifs_sb;
1802	struct cifs_tcon *tcon;
1803	struct cifsFileInfo *cfile;
1804	__u32 type;
1805
1806	rc = -EACCES;
1807	xid = get_xid();
1808
1809	if (!(fl->fl_flags & FL_FLOCK))
1810		return -ENOLCK;
1811
1812	cfile = (struct cifsFileInfo *)file->private_data;
1813	tcon = tlink_tcon(cfile->tlink);
1814
1815	cifs_read_flock(fl, &type, &lock, &unlock, &wait_flag,
1816			tcon->ses->server);
1817	cifs_sb = CIFS_FILE_SB(file);
1818
1819	if (cap_unix(tcon->ses) &&
1820	    (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1821	    ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1822		posix_lck = true;
1823
1824	if (!lock && !unlock) {
1825		/*
1826		 * if no lock or unlock then nothing to do since we do not
1827		 * know what it is
1828		 */
1829		free_xid(xid);
1830		return -EOPNOTSUPP;
1831	}
1832
1833	rc = cifs_setlk(file, fl, type, wait_flag, posix_lck, lock, unlock,
1834			xid);
1835	free_xid(xid);
1836	return rc;
1837
1838
1839}
1840
1841int cifs_lock(struct file *file, int cmd, struct file_lock *flock)
1842{
1843	int rc, xid;
1844	int lock = 0, unlock = 0;
1845	bool wait_flag = false;
1846	bool posix_lck = false;
1847	struct cifs_sb_info *cifs_sb;
1848	struct cifs_tcon *tcon;
1849	struct cifsFileInfo *cfile;
1850	__u32 type;
1851
1852	rc = -EACCES;
1853	xid = get_xid();
1854
1855	cifs_dbg(FYI, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld end: %lld\n",
1856		 cmd, flock->fl_flags, flock->fl_type,
1857		 flock->fl_start, flock->fl_end);
1858
1859	cfile = (struct cifsFileInfo *)file->private_data;
1860	tcon = tlink_tcon(cfile->tlink);
1861
1862	cifs_read_flock(flock, &type, &lock, &unlock, &wait_flag,
1863			tcon->ses->server);
1864	cifs_sb = CIFS_FILE_SB(file);
1865	set_bit(CIFS_INO_CLOSE_ON_LOCK, &CIFS_I(d_inode(cfile->dentry))->flags);
1866
1867	if (cap_unix(tcon->ses) &&
1868	    (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1869	    ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1870		posix_lck = true;
1871	/*
1872	 * BB add code here to normalize offset and length to account for
1873	 * negative length which we can not accept over the wire.
1874	 */
1875	if (IS_GETLK(cmd)) {
1876		rc = cifs_getlk(file, flock, type, wait_flag, posix_lck, xid);
1877		free_xid(xid);
1878		return rc;
1879	}
1880
1881	if (!lock && !unlock) {
1882		/*
1883		 * if no lock or unlock then nothing to do since we do not
1884		 * know what it is
1885		 */
1886		free_xid(xid);
1887		return -EOPNOTSUPP;
1888	}
1889
1890	rc = cifs_setlk(file, flock, type, wait_flag, posix_lck, lock, unlock,
1891			xid);
1892	free_xid(xid);
1893	return rc;
1894}
1895
1896/*
1897 * update the file size (if needed) after a write. Should be called with
1898 * the inode->i_lock held
1899 */
1900void
1901cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
1902		      unsigned int bytes_written)
1903{
1904	loff_t end_of_write = offset + bytes_written;
1905
1906	if (end_of_write > cifsi->server_eof)
1907		cifsi->server_eof = end_of_write;
1908}
1909
1910static ssize_t
1911cifs_write(struct cifsFileInfo *open_file, __u32 pid, const char *write_data,
1912	   size_t write_size, loff_t *offset)
1913{
1914	int rc = 0;
1915	unsigned int bytes_written = 0;
1916	unsigned int total_written;
1917	struct cifs_tcon *tcon;
1918	struct TCP_Server_Info *server;
1919	unsigned int xid;
1920	struct dentry *dentry = open_file->dentry;
1921	struct cifsInodeInfo *cifsi = CIFS_I(d_inode(dentry));
1922	struct cifs_io_parms io_parms = {0};
1923
1924	cifs_dbg(FYI, "write %zd bytes to offset %lld of %pd\n",
1925		 write_size, *offset, dentry);
1926
1927	tcon = tlink_tcon(open_file->tlink);
1928	server = tcon->ses->server;
1929
1930	if (!server->ops->sync_write)
1931		return -ENOSYS;
1932
1933	xid = get_xid();
1934
1935	for (total_written = 0; write_size > total_written;
1936	     total_written += bytes_written) {
1937		rc = -EAGAIN;
1938		while (rc == -EAGAIN) {
1939			struct kvec iov[2];
1940			unsigned int len;
1941
1942			if (open_file->invalidHandle) {
1943				/* we could deadlock if we called
1944				   filemap_fdatawait from here so tell
1945				   reopen_file not to flush data to
1946				   server now */
1947				rc = cifs_reopen_file(open_file, false);
1948				if (rc != 0)
1949					break;
1950			}
1951
1952			len = min(server->ops->wp_retry_size(d_inode(dentry)),
1953				  (unsigned int)write_size - total_written);
1954			/* iov[0] is reserved for smb header */
1955			iov[1].iov_base = (char *)write_data + total_written;
1956			iov[1].iov_len = len;
1957			io_parms.pid = pid;
1958			io_parms.tcon = tcon;
1959			io_parms.offset = *offset;
1960			io_parms.length = len;
1961			rc = server->ops->sync_write(xid, &open_file->fid,
1962					&io_parms, &bytes_written, iov, 1);
1963		}
1964		if (rc || (bytes_written == 0)) {
1965			if (total_written)
1966				break;
1967			else {
1968				free_xid(xid);
1969				return rc;
1970			}
1971		} else {
1972			spin_lock(&d_inode(dentry)->i_lock);
1973			cifs_update_eof(cifsi, *offset, bytes_written);
1974			spin_unlock(&d_inode(dentry)->i_lock);
1975			*offset += bytes_written;
1976		}
1977	}
1978
1979	cifs_stats_bytes_written(tcon, total_written);
1980
1981	if (total_written > 0) {
1982		spin_lock(&d_inode(dentry)->i_lock);
1983		if (*offset > d_inode(dentry)->i_size) {
1984			i_size_write(d_inode(dentry), *offset);
1985			d_inode(dentry)->i_blocks = (512 - 1 + *offset) >> 9;
1986		}
1987		spin_unlock(&d_inode(dentry)->i_lock);
1988	}
1989	mark_inode_dirty_sync(d_inode(dentry));
1990	free_xid(xid);
1991	return total_written;
1992}
1993
1994struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1995					bool fsuid_only)
1996{
1997	struct cifsFileInfo *open_file = NULL;
1998	struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1999
2000	/* only filter by fsuid on multiuser mounts */
2001	if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
2002		fsuid_only = false;
2003
2004	spin_lock(&cifs_inode->open_file_lock);
2005	/* we could simply get the first_list_entry since write-only entries
2006	   are always at the end of the list but since the first entry might
2007	   have a close pending, we go through the whole list */
2008	list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2009		if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
2010			continue;
2011		if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
2012			if ((!open_file->invalidHandle)) {
2013				/* found a good file */
2014				/* lock it so it will not be closed on us */
2015				cifsFileInfo_get(open_file);
2016				spin_unlock(&cifs_inode->open_file_lock);
2017				return open_file;
2018			} /* else might as well continue, and look for
2019			     another, or simply have the caller reopen it
2020			     again rather than trying to fix this handle */
2021		} else /* write only file */
2022			break; /* write only files are last so must be done */
2023	}
2024	spin_unlock(&cifs_inode->open_file_lock);
2025	return NULL;
2026}
2027
2028/* Return -EBADF if no handle is found and general rc otherwise */
2029int
2030cifs_get_writable_file(struct cifsInodeInfo *cifs_inode, int flags,
2031		       struct cifsFileInfo **ret_file)
2032{
2033	struct cifsFileInfo *open_file, *inv_file = NULL;
2034	struct cifs_sb_info *cifs_sb;
2035	bool any_available = false;
2036	int rc = -EBADF;
2037	unsigned int refind = 0;
2038	bool fsuid_only = flags & FIND_WR_FSUID_ONLY;
2039	bool with_delete = flags & FIND_WR_WITH_DELETE;
2040	*ret_file = NULL;
2041
2042	/*
2043	 * Having a null inode here (because mapping->host was set to zero by
2044	 * the VFS or MM) should not happen but we had reports of on oops (due
2045	 * to it being zero) during stress testcases so we need to check for it
2046	 */
2047
2048	if (cifs_inode == NULL) {
2049		cifs_dbg(VFS, "Null inode passed to cifs_writeable_file\n");
2050		dump_stack();
2051		return rc;
2052	}
2053
2054	cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
2055
2056	/* only filter by fsuid on multiuser mounts */
2057	if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
2058		fsuid_only = false;
2059
2060	spin_lock(&cifs_inode->open_file_lock);
2061refind_writable:
2062	if (refind > MAX_REOPEN_ATT) {
2063		spin_unlock(&cifs_inode->open_file_lock);
2064		return rc;
2065	}
2066	list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2067		if (!any_available && open_file->pid != current->tgid)
2068			continue;
2069		if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
2070			continue;
2071		if (with_delete && !(open_file->fid.access & DELETE))
2072			continue;
2073		if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2074			if (!open_file->invalidHandle) {
2075				/* found a good writable file */
2076				cifsFileInfo_get(open_file);
2077				spin_unlock(&cifs_inode->open_file_lock);
2078				*ret_file = open_file;
2079				return 0;
2080			} else {
2081				if (!inv_file)
2082					inv_file = open_file;
2083			}
2084		}
2085	}
2086	/* couldn't find useable FH with same pid, try any available */
2087	if (!any_available) {
2088		any_available = true;
2089		goto refind_writable;
2090	}
2091
2092	if (inv_file) {
2093		any_available = false;
2094		cifsFileInfo_get(inv_file);
2095	}
2096
2097	spin_unlock(&cifs_inode->open_file_lock);
2098
2099	if (inv_file) {
2100		rc = cifs_reopen_file(inv_file, false);
2101		if (!rc) {
2102			*ret_file = inv_file;
2103			return 0;
2104		}
2105
2106		spin_lock(&cifs_inode->open_file_lock);
2107		list_move_tail(&inv_file->flist, &cifs_inode->openFileList);
2108		spin_unlock(&cifs_inode->open_file_lock);
2109		cifsFileInfo_put(inv_file);
2110		++refind;
2111		inv_file = NULL;
2112		spin_lock(&cifs_inode->open_file_lock);
2113		goto refind_writable;
2114	}
2115
2116	return rc;
2117}
2118
2119struct cifsFileInfo *
2120find_writable_file(struct cifsInodeInfo *cifs_inode, int flags)
2121{
2122	struct cifsFileInfo *cfile;
2123	int rc;
2124
2125	rc = cifs_get_writable_file(cifs_inode, flags, &cfile);
2126	if (rc)
2127		cifs_dbg(FYI, "Couldn't find writable handle rc=%d\n", rc);
2128
2129	return cfile;
2130}
2131
2132int
2133cifs_get_writable_path(struct cifs_tcon *tcon, const char *name,
2134		       int flags,
2135		       struct cifsFileInfo **ret_file)
2136{
2137	struct cifsFileInfo *cfile;
2138	void *page = alloc_dentry_path();
2139
2140	*ret_file = NULL;
2141
2142	spin_lock(&tcon->open_file_lock);
2143	list_for_each_entry(cfile, &tcon->openFileList, tlist) {
2144		struct cifsInodeInfo *cinode;
2145		const char *full_path = build_path_from_dentry(cfile->dentry, page);
2146		if (IS_ERR(full_path)) {
2147			spin_unlock(&tcon->open_file_lock);
2148			free_dentry_path(page);
2149			return PTR_ERR(full_path);
2150		}
2151		if (strcmp(full_path, name))
2152			continue;
2153
2154		cinode = CIFS_I(d_inode(cfile->dentry));
2155		spin_unlock(&tcon->open_file_lock);
2156		free_dentry_path(page);
2157		return cifs_get_writable_file(cinode, flags, ret_file);
2158	}
2159
2160	spin_unlock(&tcon->open_file_lock);
2161	free_dentry_path(page);
2162	return -ENOENT;
2163}
2164
2165int
2166cifs_get_readable_path(struct cifs_tcon *tcon, const char *name,
2167		       struct cifsFileInfo **ret_file)
2168{
2169	struct cifsFileInfo *cfile;
2170	void *page = alloc_dentry_path();
2171
2172	*ret_file = NULL;
2173
2174	spin_lock(&tcon->open_file_lock);
2175	list_for_each_entry(cfile, &tcon->openFileList, tlist) {
2176		struct cifsInodeInfo *cinode;
2177		const char *full_path = build_path_from_dentry(cfile->dentry, page);
2178		if (IS_ERR(full_path)) {
2179			spin_unlock(&tcon->open_file_lock);
2180			free_dentry_path(page);
2181			return PTR_ERR(full_path);
2182		}
2183		if (strcmp(full_path, name))
2184			continue;
2185
2186		cinode = CIFS_I(d_inode(cfile->dentry));
2187		spin_unlock(&tcon->open_file_lock);
2188		free_dentry_path(page);
2189		*ret_file = find_readable_file(cinode, 0);
2190		return *ret_file ? 0 : -ENOENT;
2191	}
2192
2193	spin_unlock(&tcon->open_file_lock);
2194	free_dentry_path(page);
2195	return -ENOENT;
2196}
2197
2198static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
2199{
2200	struct address_space *mapping = page->mapping;
2201	loff_t offset = (loff_t)page->index << PAGE_SHIFT;
2202	char *write_data;
2203	int rc = -EFAULT;
2204	int bytes_written = 0;
2205	struct inode *inode;
2206	struct cifsFileInfo *open_file;
2207
2208	if (!mapping || !mapping->host)
2209		return -EFAULT;
2210
2211	inode = page->mapping->host;
2212
2213	offset += (loff_t)from;
2214	write_data = kmap(page);
2215	write_data += from;
2216
2217	if ((to > PAGE_SIZE) || (from > to)) {
2218		kunmap(page);
2219		return -EIO;
2220	}
2221
2222	/* racing with truncate? */
2223	if (offset > mapping->host->i_size) {
2224		kunmap(page);
2225		return 0; /* don't care */
2226	}
2227
2228	/* check to make sure that we are not extending the file */
2229	if (mapping->host->i_size - offset < (loff_t)to)
2230		to = (unsigned)(mapping->host->i_size - offset);
2231
2232	rc = cifs_get_writable_file(CIFS_I(mapping->host), FIND_WR_ANY,
2233				    &open_file);
2234	if (!rc) {
2235		bytes_written = cifs_write(open_file, open_file->pid,
2236					   write_data, to - from, &offset);
2237		cifsFileInfo_put(open_file);
2238		/* Does mm or vfs already set times? */
2239		inode->i_atime = inode->i_mtime = current_time(inode);
2240		if ((bytes_written > 0) && (offset))
2241			rc = 0;
2242		else if (bytes_written < 0)
2243			rc = bytes_written;
2244		else
2245			rc = -EFAULT;
2246	} else {
2247		cifs_dbg(FYI, "No writable handle for write page rc=%d\n", rc);
2248		if (!is_retryable_error(rc))
2249			rc = -EIO;
2250	}
2251
2252	kunmap(page);
2253	return rc;
2254}
2255
2256static struct cifs_writedata *
2257wdata_alloc_and_fillpages(pgoff_t tofind, struct address_space *mapping,
2258			  pgoff_t end, pgoff_t *index,
2259			  unsigned int *found_pages)
2260{
2261	struct cifs_writedata *wdata;
2262
2263	wdata = cifs_writedata_alloc((unsigned int)tofind,
2264				     cifs_writev_complete);
2265	if (!wdata)
2266		return NULL;
2267
2268	*found_pages = find_get_pages_range_tag(mapping, index, end,
2269				PAGECACHE_TAG_DIRTY, tofind, wdata->pages);
2270	return wdata;
2271}
2272
2273static unsigned int
2274wdata_prepare_pages(struct cifs_writedata *wdata, unsigned int found_pages,
2275		    struct address_space *mapping,
2276		    struct writeback_control *wbc,
2277		    pgoff_t end, pgoff_t *index, pgoff_t *next, bool *done)
2278{
2279	unsigned int nr_pages = 0, i;
2280	struct page *page;
2281
2282	for (i = 0; i < found_pages; i++) {
2283		page = wdata->pages[i];
2284		/*
2285		 * At this point we hold neither the i_pages lock nor the
2286		 * page lock: the page may be truncated or invalidated
2287		 * (changing page->mapping to NULL), or even swizzled
2288		 * back from swapper_space to tmpfs file mapping
2289		 */
2290
2291		if (nr_pages == 0)
2292			lock_page(page);
2293		else if (!trylock_page(page))
2294			break;
2295
2296		if (unlikely(page->mapping != mapping)) {
2297			unlock_page(page);
2298			break;
2299		}
2300
2301		if (!wbc->range_cyclic && page->index > end) {
2302			*done = true;
2303			unlock_page(page);
2304			break;
2305		}
2306
2307		if (*next && (page->index != *next)) {
2308			/* Not next consecutive page */
2309			unlock_page(page);
2310			break;
2311		}
2312
2313		if (wbc->sync_mode != WB_SYNC_NONE)
2314			wait_on_page_writeback(page);
2315
2316		if (PageWriteback(page) ||
2317				!clear_page_dirty_for_io(page)) {
2318			unlock_page(page);
2319			break;
2320		}
2321
2322		/*
2323		 * This actually clears the dirty bit in the radix tree.
2324		 * See cifs_writepage() for more commentary.
2325		 */
2326		set_page_writeback(page);
2327		if (page_offset(page) >= i_size_read(mapping->host)) {
2328			*done = true;
2329			unlock_page(page);
2330			end_page_writeback(page);
2331			break;
2332		}
2333
2334		wdata->pages[i] = page;
2335		*next = page->index + 1;
2336		++nr_pages;
2337	}
2338
2339	/* reset index to refind any pages skipped */
2340	if (nr_pages == 0)
2341		*index = wdata->pages[0]->index + 1;
2342
2343	/* put any pages we aren't going to use */
2344	for (i = nr_pages; i < found_pages; i++) {
2345		put_page(wdata->pages[i]);
2346		wdata->pages[i] = NULL;
2347	}
2348
2349	return nr_pages;
2350}
2351
2352static int
2353wdata_send_pages(struct cifs_writedata *wdata, unsigned int nr_pages,
2354		 struct address_space *mapping, struct writeback_control *wbc)
2355{
2356	int rc;
2357
2358	wdata->sync_mode = wbc->sync_mode;
2359	wdata->nr_pages = nr_pages;
2360	wdata->offset = page_offset(wdata->pages[0]);
2361	wdata->pagesz = PAGE_SIZE;
2362	wdata->tailsz = min(i_size_read(mapping->host) -
2363			page_offset(wdata->pages[nr_pages - 1]),
2364			(loff_t)PAGE_SIZE);
2365	wdata->bytes = ((nr_pages - 1) * PAGE_SIZE) + wdata->tailsz;
2366	wdata->pid = wdata->cfile->pid;
2367
2368	rc = adjust_credits(wdata->server, &wdata->credits, wdata->bytes);
2369	if (rc)
2370		return rc;
2371
2372	if (wdata->cfile->invalidHandle)
2373		rc = -EAGAIN;
2374	else
2375		rc = wdata->server->ops->async_writev(wdata,
2376						      cifs_writedata_release);
2377
2378	return rc;
2379}
2380
2381static int cifs_writepages(struct address_space *mapping,
2382			   struct writeback_control *wbc)
2383{
2384	struct inode *inode = mapping->host;
2385	struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2386	struct TCP_Server_Info *server;
2387	bool done = false, scanned = false, range_whole = false;
2388	pgoff_t end, index;
2389	struct cifs_writedata *wdata;
2390	struct cifsFileInfo *cfile = NULL;
2391	int rc = 0;
2392	int saved_rc = 0;
2393	unsigned int xid;
2394
2395	/*
2396	 * If wsize is smaller than the page cache size, default to writing
2397	 * one page at a time via cifs_writepage
2398	 */
2399	if (cifs_sb->ctx->wsize < PAGE_SIZE)
2400		return generic_writepages(mapping, wbc);
2401
2402	xid = get_xid();
2403	if (wbc->range_cyclic) {
2404		index = mapping->writeback_index; /* Start from prev offset */
2405		end = -1;
2406	} else {
2407		index = wbc->range_start >> PAGE_SHIFT;
2408		end = wbc->range_end >> PAGE_SHIFT;
2409		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2410			range_whole = true;
2411		scanned = true;
2412	}
2413	server = cifs_pick_channel(cifs_sb_master_tcon(cifs_sb)->ses);
2414
2415retry:
2416	while (!done && index <= end) {
2417		unsigned int i, nr_pages, found_pages, wsize;
2418		pgoff_t next = 0, tofind, saved_index = index;
2419		struct cifs_credits credits_on_stack;
2420		struct cifs_credits *credits = &credits_on_stack;
2421		int get_file_rc = 0;
2422
2423		if (cfile)
2424			cifsFileInfo_put(cfile);
2425
2426		rc = cifs_get_writable_file(CIFS_I(inode), FIND_WR_ANY, &cfile);
2427
2428		/* in case of an error store it to return later */
2429		if (rc)
2430			get_file_rc = rc;
2431
2432		rc = server->ops->wait_mtu_credits(server, cifs_sb->ctx->wsize,
2433						   &wsize, credits);
2434		if (rc != 0) {
2435			done = true;
2436			break;
2437		}
2438
2439		tofind = min((wsize / PAGE_SIZE) - 1, end - index) + 1;
2440
2441		wdata = wdata_alloc_and_fillpages(tofind, mapping, end, &index,
2442						  &found_pages);
2443		if (!wdata) {
2444			rc = -ENOMEM;
2445			done = true;
2446			add_credits_and_wake_if(server, credits, 0);
2447			break;
2448		}
2449
2450		if (found_pages == 0) {
2451			kref_put(&wdata->refcount, cifs_writedata_release);
2452			add_credits_and_wake_if(server, credits, 0);
2453			break;
2454		}
2455
2456		nr_pages = wdata_prepare_pages(wdata, found_pages, mapping, wbc,
2457					       end, &index, &next, &done);
2458
2459		/* nothing to write? */
2460		if (nr_pages == 0) {
2461			kref_put(&wdata->refcount, cifs_writedata_release);
2462			add_credits_and_wake_if(server, credits, 0);
2463			continue;
2464		}
2465
2466		wdata->credits = credits_on_stack;
2467		wdata->cfile = cfile;
2468		wdata->server = server;
2469		cfile = NULL;
2470
2471		if (!wdata->cfile) {
2472			cifs_dbg(VFS, "No writable handle in writepages rc=%d\n",
2473				 get_file_rc);
2474			if (is_retryable_error(get_file_rc))
2475				rc = get_file_rc;
2476			else
2477				rc = -EBADF;
2478		} else
2479			rc = wdata_send_pages(wdata, nr_pages, mapping, wbc);
2480
2481		for (i = 0; i < nr_pages; ++i)
2482			unlock_page(wdata->pages[i]);
2483
2484		/* send failure -- clean up the mess */
2485		if (rc != 0) {
2486			add_credits_and_wake_if(server, &wdata->credits, 0);
2487			for (i = 0; i < nr_pages; ++i) {
2488				if (is_retryable_error(rc))
2489					redirty_page_for_writepage(wbc,
2490							   wdata->pages[i]);
2491				else
2492					SetPageError(wdata->pages[i]);
2493				end_page_writeback(wdata->pages[i]);
2494				put_page(wdata->pages[i]);
2495			}
2496			if (!is_retryable_error(rc))
2497				mapping_set_error(mapping, rc);
2498		}
2499		kref_put(&wdata->refcount, cifs_writedata_release);
2500
2501		if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN) {
2502			index = saved_index;
2503			continue;
2504		}
2505
2506		/* Return immediately if we received a signal during writing */
2507		if (is_interrupt_error(rc)) {
2508			done = true;
2509			break;
2510		}
2511
2512		if (rc != 0 && saved_rc == 0)
2513			saved_rc = rc;
2514
2515		wbc->nr_to_write -= nr_pages;
2516		if (wbc->nr_to_write <= 0)
2517			done = true;
2518
2519		index = next;
2520	}
2521
2522	if (!scanned && !done) {
2523		/*
2524		 * We hit the last page and there is more work to be done: wrap
2525		 * back to the start of the file
2526		 */
2527		scanned = true;
2528		index = 0;
2529		goto retry;
2530	}
2531
2532	if (saved_rc != 0)
2533		rc = saved_rc;
2534
2535	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2536		mapping->writeback_index = index;
2537
2538	if (cfile)
2539		cifsFileInfo_put(cfile);
2540	free_xid(xid);
2541	/* Indication to update ctime and mtime as close is deferred */
2542	set_bit(CIFS_INO_MODIFIED_ATTR, &CIFS_I(inode)->flags);
2543	return rc;
2544}
2545
2546static int
2547cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
2548{
2549	int rc;
2550	unsigned int xid;
2551
2552	xid = get_xid();
2553/* BB add check for wbc flags */
2554	get_page(page);
2555	if (!PageUptodate(page))
2556		cifs_dbg(FYI, "ppw - page not up to date\n");
2557
2558	/*
2559	 * Set the "writeback" flag, and clear "dirty" in the radix tree.
2560	 *
2561	 * A writepage() implementation always needs to do either this,
2562	 * or re-dirty the page with "redirty_page_for_writepage()" in
2563	 * the case of a failure.
2564	 *
2565	 * Just unlocking the page will cause the radix tree tag-bits
2566	 * to fail to update with the state of the page correctly.
2567	 */
2568	set_page_writeback(page);
2569retry_write:
2570	rc = cifs_partialpagewrite(page, 0, PAGE_SIZE);
2571	if (is_retryable_error(rc)) {
2572		if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN)
2573			goto retry_write;
2574		redirty_page_for_writepage(wbc, page);
2575	} else if (rc != 0) {
2576		SetPageError(page);
2577		mapping_set_error(page->mapping, rc);
2578	} else {
2579		SetPageUptodate(page);
2580	}
2581	end_page_writeback(page);
2582	put_page(page);
2583	free_xid(xid);
2584	return rc;
2585}
2586
2587static int cifs_writepage(struct page *page, struct writeback_control *wbc)
2588{
2589	int rc = cifs_writepage_locked(page, wbc);
2590	unlock_page(page);
2591	return rc;
2592}
2593
2594static int cifs_write_end(struct file *file, struct address_space *mapping,
2595			loff_t pos, unsigned len, unsigned copied,
2596			struct page *page, void *fsdata)
2597{
2598	int rc;
2599	struct inode *inode = mapping->host;
2600	struct cifsFileInfo *cfile = file->private_data;
2601	struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
2602	__u32 pid;
2603
2604	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2605		pid = cfile->pid;
2606	else
2607		pid = current->tgid;
2608
2609	cifs_dbg(FYI, "write_end for page %p from pos %lld with %d bytes\n",
2610		 page, pos, copied);
2611
2612	if (PageChecked(page)) {
2613		if (copied == len)
2614			SetPageUptodate(page);
2615		ClearPageChecked(page);
2616	} else if (!PageUptodate(page) && copied == PAGE_SIZE)
2617		SetPageUptodate(page);
2618
2619	if (!PageUptodate(page)) {
2620		char *page_data;
2621		unsigned offset = pos & (PAGE_SIZE - 1);
2622		unsigned int xid;
2623
2624		xid = get_xid();
2625		/* this is probably better than directly calling
2626		   partialpage_write since in this function the file handle is
2627		   known which we might as well	leverage */
2628		/* BB check if anything else missing out of ppw
2629		   such as updating last write time */
2630		page_data = kmap(page);
2631		rc = cifs_write(cfile, pid, page_data + offset, copied, &pos);
2632		/* if (rc < 0) should we set writebehind rc? */
2633		kunmap(page);
2634
2635		free_xid(xid);
2636	} else {
2637		rc = copied;
2638		pos += copied;
2639		set_page_dirty(page);
2640	}
2641
2642	if (rc > 0) {
2643		spin_lock(&inode->i_lock);
2644		if (pos > inode->i_size) {
2645			i_size_write(inode, pos);
2646			inode->i_blocks = (512 - 1 + pos) >> 9;
2647		}
2648		spin_unlock(&inode->i_lock);
2649	}
2650
2651	unlock_page(page);
2652	put_page(page);
2653	/* Indication to update ctime and mtime as close is deferred */
2654	set_bit(CIFS_INO_MODIFIED_ATTR, &CIFS_I(inode)->flags);
2655
2656	return rc;
2657}
2658
2659int cifs_strict_fsync(struct file *file, loff_t start, loff_t end,
2660		      int datasync)
2661{
2662	unsigned int xid;
2663	int rc = 0;
2664	struct cifs_tcon *tcon;
2665	struct TCP_Server_Info *server;
2666	struct cifsFileInfo *smbfile = file->private_data;
2667	struct inode *inode = file_inode(file);
2668	struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2669
2670	rc = file_write_and_wait_range(file, start, end);
2671	if (rc) {
2672		trace_cifs_fsync_err(inode->i_ino, rc);
2673		return rc;
2674	}
2675
2676	xid = get_xid();
2677
2678	cifs_dbg(FYI, "Sync file - name: %pD datasync: 0x%x\n",
2679		 file, datasync);
2680
2681	if (!CIFS_CACHE_READ(CIFS_I(inode))) {
2682		rc = cifs_zap_mapping(inode);
2683		if (rc) {
2684			cifs_dbg(FYI, "rc: %d during invalidate phase\n", rc);
2685			rc = 0; /* don't care about it in fsync */
2686		}
2687	}
2688
2689	tcon = tlink_tcon(smbfile->tlink);
2690	if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2691		server = tcon->ses->server;
2692		if (server->ops->flush)
2693			rc = server->ops->flush(xid, tcon, &smbfile->fid);
2694		else
2695			rc = -ENOSYS;
2696	}
2697
2698	free_xid(xid);
2699	return rc;
2700}
2701
2702int cifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2703{
2704	unsigned int xid;
2705	int rc = 0;
2706	struct cifs_tcon *tcon;
2707	struct TCP_Server_Info *server;
2708	struct cifsFileInfo *smbfile = file->private_data;
2709	struct cifs_sb_info *cifs_sb = CIFS_FILE_SB(file);
2710
2711	rc = file_write_and_wait_range(file, start, end);
2712	if (rc) {
2713		trace_cifs_fsync_err(file_inode(file)->i_ino, rc);
2714		return rc;
2715	}
2716
2717	xid = get_xid();
2718
2719	cifs_dbg(FYI, "Sync file - name: %pD datasync: 0x%x\n",
2720		 file, datasync);
2721
2722	tcon = tlink_tcon(smbfile->tlink);
2723	if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2724		server = tcon->ses->server;
2725		if (server->ops->flush)
2726			rc = server->ops->flush(xid, tcon, &smbfile->fid);
2727		else
2728			rc = -ENOSYS;
2729	}
2730
2731	free_xid(xid);
2732	return rc;
2733}
2734
2735/*
2736 * As file closes, flush all cached write data for this inode checking
2737 * for write behind errors.
2738 */
2739int cifs_flush(struct file *file, fl_owner_t id)
2740{
2741	struct inode *inode = file_inode(file);
2742	int rc = 0;
2743
2744	if (file->f_mode & FMODE_WRITE)
2745		rc = filemap_write_and_wait(inode->i_mapping);
2746
2747	cifs_dbg(FYI, "Flush inode %p file %p rc %d\n", inode, file, rc);
2748	if (rc)
2749		trace_cifs_flush_err(inode->i_ino, rc);
2750	return rc;
2751}
2752
2753static int
2754cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
2755{
2756	int rc = 0;
2757	unsigned long i;
2758
2759	for (i = 0; i < num_pages; i++) {
2760		pages[i] = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2761		if (!pages[i]) {
2762			/*
2763			 * save number of pages we have already allocated and
2764			 * return with ENOMEM error
2765			 */
2766			num_pages = i;
2767			rc = -ENOMEM;
2768			break;
2769		}
2770	}
2771
2772	if (rc) {
2773		for (i = 0; i < num_pages; i++)
2774			put_page(pages[i]);
2775	}
2776	return rc;
2777}
2778
2779static inline
2780size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
2781{
2782	size_t num_pages;
2783	size_t clen;
2784
2785	clen = min_t(const size_t, len, wsize);
2786	num_pages = DIV_ROUND_UP(clen, PAGE_SIZE);
2787
2788	if (cur_len)
2789		*cur_len = clen;
2790
2791	return num_pages;
2792}
2793
2794static void
2795cifs_uncached_writedata_release(struct kref *refcount)
2796{
2797	int i;
2798	struct cifs_writedata *wdata = container_of(refcount,
2799					struct cifs_writedata, refcount);
2800
2801	kref_put(&wdata->ctx->refcount, cifs_aio_ctx_release);
2802	for (i = 0; i < wdata->nr_pages; i++)
2803		put_page(wdata->pages[i]);
2804	cifs_writedata_release(refcount);
2805}
2806
2807static void collect_uncached_write_data(struct cifs_aio_ctx *ctx);
2808
2809static void
2810cifs_uncached_writev_complete(struct work_struct *work)
2811{
2812	struct cifs_writedata *wdata = container_of(work,
2813					struct cifs_writedata, work);
2814	struct inode *inode = d_inode(wdata->cfile->dentry);
2815	struct cifsInodeInfo *cifsi = CIFS_I(inode);
2816
2817	spin_lock(&inode->i_lock);
2818	cifs_update_eof(cifsi, wdata->offset, wdata->bytes);
2819	if (cifsi->server_eof > inode->i_size)
2820		i_size_write(inode, cifsi->server_eof);
2821	spin_unlock(&inode->i_lock);
2822
2823	complete(&wdata->done);
2824	collect_uncached_write_data(wdata->ctx);
2825	/* the below call can possibly free the last ref to aio ctx */
2826	kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2827}
2828
2829static int
2830wdata_fill_from_iovec(struct cifs_writedata *wdata, struct iov_iter *from,
2831		      size_t *len, unsigned long *num_pages)
2832{
2833	size_t save_len, copied, bytes, cur_len = *len;
2834	unsigned long i, nr_pages = *num_pages;
2835
2836	save_len = cur_len;
2837	for (i = 0; i < nr_pages; i++) {
2838		bytes = min_t(const size_t, cur_len, PAGE_SIZE);
2839		copied = copy_page_from_iter(wdata->pages[i], 0, bytes, from);
2840		cur_len -= copied;
2841		/*
2842		 * If we didn't copy as much as we expected, then that
2843		 * may mean we trod into an unmapped area. Stop copying
2844		 * at that point. On the next pass through the big
2845		 * loop, we'll likely end up getting a zero-length
2846		 * write and bailing out of it.
2847		 */
2848		if (copied < bytes)
2849			break;
2850	}
2851	cur_len = save_len - cur_len;
2852	*len = cur_len;
2853
2854	/*
2855	 * If we have no data to send, then that probably means that
2856	 * the copy above failed altogether. That's most likely because
2857	 * the address in the iovec was bogus. Return -EFAULT and let
2858	 * the caller free anything we allocated and bail out.
2859	 */
2860	if (!cur_len)
2861		return -EFAULT;
2862
2863	/*
2864	 * i + 1 now represents the number of pages we actually used in
2865	 * the copy phase above.
2866	 */
2867	*num_pages = i + 1;
2868	return 0;
2869}
2870
2871static int
2872cifs_resend_wdata(struct cifs_writedata *wdata, struct list_head *wdata_list,
2873	struct cifs_aio_ctx *ctx)
2874{
2875	unsigned int wsize;
2876	struct cifs_credits credits;
2877	int rc;
2878	struct TCP_Server_Info *server = wdata->server;
2879
2880	do {
2881		if (wdata->cfile->invalidHandle) {
2882			rc = cifs_reopen_file(wdata->cfile, false);
2883			if (rc == -EAGAIN)
2884				continue;
2885			else if (rc)
2886				break;
2887		}
2888
2889
2890		/*
2891		 * Wait for credits to resend this wdata.
2892		 * Note: we are attempting to resend the whole wdata not in
2893		 * segments
2894		 */
2895		do {
2896			rc = server->ops->wait_mtu_credits(server, wdata->bytes,
2897						&wsize, &credits);
2898			if (rc)
2899				goto fail;
2900
2901			if (wsize < wdata->bytes) {
2902				add_credits_and_wake_if(server, &credits, 0);
2903				msleep(1000);
2904			}
2905		} while (wsize < wdata->bytes);
2906		wdata->credits = credits;
2907
2908		rc = adjust_credits(server, &wdata->credits, wdata->bytes);
2909
2910		if (!rc) {
2911			if (wdata->cfile->invalidHandle)
2912				rc = -EAGAIN;
2913			else {
2914#ifdef CONFIG_CIFS_SMB_DIRECT
2915				if (wdata->mr) {
2916					wdata->mr->need_invalidate = true;
2917					smbd_deregister_mr(wdata->mr);
2918					wdata->mr = NULL;
2919				}
2920#endif
2921				rc = server->ops->async_writev(wdata,
2922					cifs_uncached_writedata_release);
2923			}
2924		}
2925
2926		/* If the write was successfully sent, we are done */
2927		if (!rc) {
2928			list_add_tail(&wdata->list, wdata_list);
2929			return 0;
2930		}
2931
2932		/* Roll back credits and retry if needed */
2933		add_credits_and_wake_if(server, &wdata->credits, 0);
2934	} while (rc == -EAGAIN);
2935
2936fail:
2937	kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2938	return rc;
2939}
2940
2941static int
2942cifs_write_from_iter(loff_t offset, size_t len, struct iov_iter *from,
2943		     struct cifsFileInfo *open_file,
2944		     struct cifs_sb_info *cifs_sb, struct list_head *wdata_list,
2945		     struct cifs_aio_ctx *ctx)
2946{
2947	int rc = 0;
2948	size_t cur_len;
2949	unsigned long nr_pages, num_pages, i;
2950	struct cifs_writedata *wdata;
2951	struct iov_iter saved_from = *from;
2952	loff_t saved_offset = offset;
2953	pid_t pid;
2954	struct TCP_Server_Info *server;
2955	struct page **pagevec;
2956	size_t start;
2957	unsigned int xid;
2958
2959	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2960		pid = open_file->pid;
2961	else
2962		pid = current->tgid;
2963
2964	server = cifs_pick_channel(tlink_tcon(open_file->tlink)->ses);
2965	xid = get_xid();
2966
2967	do {
2968		unsigned int wsize;
2969		struct cifs_credits credits_on_stack;
2970		struct cifs_credits *credits = &credits_on_stack;
2971
2972		if (open_file->invalidHandle) {
2973			rc = cifs_reopen_file(open_file, false);
2974			if (rc == -EAGAIN)
2975				continue;
2976			else if (rc)
2977				break;
2978		}
2979
2980		rc = server->ops->wait_mtu_credits(server, cifs_sb->ctx->wsize,
2981						   &wsize, credits);
2982		if (rc)
2983			break;
2984
2985		cur_len = min_t(const size_t, len, wsize);
2986
2987		if (ctx->direct_io) {
2988			ssize_t result;
2989
2990			result = iov_iter_get_pages_alloc(
2991				from, &pagevec, cur_len, &start);
2992			if (result < 0) {
2993				cifs_dbg(VFS,
2994					 "direct_writev couldn't get user pages (rc=%zd) iter type %d iov_offset %zd count %zd\n",
2995					 result, iov_iter_type(from),
2996					 from->iov_offset, from->count);
2997				dump_stack();
2998
2999				rc = result;
3000				add_credits_and_wake_if(server, credits, 0);
3001				break;
3002			}
3003			cur_len = (size_t)result;
3004			iov_iter_advance(from, cur_len);
3005
3006			nr_pages =
3007				(cur_len + start + PAGE_SIZE - 1) / PAGE_SIZE;
3008
3009			wdata = cifs_writedata_direct_alloc(pagevec,
3010					     cifs_uncached_writev_complete);
3011			if (!wdata) {
3012				rc = -ENOMEM;
3013				add_credits_and_wake_if(server, credits, 0);
3014				break;
3015			}
3016
3017
3018			wdata->page_offset = start;
3019			wdata->tailsz =
3020				nr_pages > 1 ?
3021					cur_len - (PAGE_SIZE - start) -
3022					(nr_pages - 2) * PAGE_SIZE :
3023					cur_len;
3024		} else {
3025			nr_pages = get_numpages(wsize, len, &cur_len);
3026			wdata = cifs_writedata_alloc(nr_pages,
3027					     cifs_uncached_writev_complete);
3028			if (!wdata) {
3029				rc = -ENOMEM;
3030				add_credits_and_wake_if(server, credits, 0);
3031				break;
3032			}
3033
3034			rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
3035			if (rc) {
3036				kvfree(wdata->pages);
3037				kfree(wdata);
3038				add_credits_and_wake_if(server, credits, 0);
3039				break;
3040			}
3041
3042			num_pages = nr_pages;
3043			rc = wdata_fill_from_iovec(
3044				wdata, from, &cur_len, &num_pages);
3045			if (rc) {
3046				for (i = 0; i < nr_pages; i++)
3047					put_page(wdata->pages[i]);
3048				kvfree(wdata->pages);
3049				kfree(wdata);
3050				add_credits_and_wake_if(server, credits, 0);
3051				break;
3052			}
3053
3054			/*
3055			 * Bring nr_pages down to the number of pages we
3056			 * actually used, and free any pages that we didn't use.
3057			 */
3058			for ( ; nr_pages > num_pages; nr_pages--)
3059				put_page(wdata->pages[nr_pages - 1]);
3060
3061			wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
3062		}
3063
3064		wdata->sync_mode = WB_SYNC_ALL;
3065		wdata->nr_pages = nr_pages;
3066		wdata->offset = (__u64)offset;
3067		wdata->cfile = cifsFileInfo_get(open_file);
3068		wdata->server = server;
3069		wdata->pid = pid;
3070		wdata->bytes = cur_len;
3071		wdata->pagesz = PAGE_SIZE;
3072		wdata->credits = credits_on_stack;
3073		wdata->ctx = ctx;
3074		kref_get(&ctx->refcount);
3075
3076		rc = adjust_credits(server, &wdata->credits, wdata->bytes);
3077
3078		if (!rc) {
3079			if (wdata->cfile->invalidHandle)
3080				rc = -EAGAIN;
3081			else
3082				rc = server->ops->async_writev(wdata,
3083					cifs_uncached_writedata_release);
3084		}
3085
3086		if (rc) {
3087			add_credits_and_wake_if(server, &wdata->credits, 0);
3088			kref_put(&wdata->refcount,
3089				 cifs_uncached_writedata_release);
3090			if (rc == -EAGAIN) {
3091				*from = saved_from;
3092				iov_iter_advance(from, offset - saved_offset);
3093				continue;
3094			}
3095			break;
3096		}
3097
3098		list_add_tail(&wdata->list, wdata_list);
3099		offset += cur_len;
3100		len -= cur_len;
3101	} while (len > 0);
3102
3103	free_xid(xid);
3104	return rc;
3105}
3106
3107static void collect_uncached_write_data(struct cifs_aio_ctx *ctx)
3108{
3109	struct cifs_writedata *wdata, *tmp;
3110	struct cifs_tcon *tcon;
3111	struct cifs_sb_info *cifs_sb;
3112	struct dentry *dentry = ctx->cfile->dentry;
3113	ssize_t rc;
3114
3115	tcon = tlink_tcon(ctx->cfile->tlink);
3116	cifs_sb = CIFS_SB(dentry->d_sb);
3117
3118	mutex_lock(&ctx->aio_mutex);
3119
3120	if (list_empty(&ctx->list)) {
3121		mutex_unlock(&ctx->aio_mutex);
3122		return;
3123	}
3124
3125	rc = ctx->rc;
3126	/*
3127	 * Wait for and collect replies for any successful sends in order of
3128	 * increasing offset. Once an error is hit, then return without waiting
3129	 * for any more replies.
3130	 */
3131restart_loop:
3132	list_for_each_entry_safe(wdata, tmp, &ctx->list, list) {
3133		if (!rc) {
3134			if (!try_wait_for_completion(&wdata->done)) {
3135				mutex_unlock(&ctx->aio_mutex);
3136				return;
3137			}
3138
3139			if (wdata->result)
3140				rc = wdata->result;
3141			else
3142				ctx->total_len += wdata->bytes;
3143
3144			/* resend call if it's a retryable error */
3145			if (rc == -EAGAIN) {
3146				struct list_head tmp_list;
3147				struct iov_iter tmp_from = ctx->iter;
3148
3149				INIT_LIST_HEAD(&tmp_list);
3150				list_del_init(&wdata->list);
3151
3152				if (ctx->direct_io)
3153					rc = cifs_resend_wdata(
3154						wdata, &tmp_list, ctx);
3155				else {
3156					iov_iter_advance(&tmp_from,
3157						 wdata->offset - ctx->pos);
3158
3159					rc = cifs_write_from_iter(wdata->offset,
3160						wdata->bytes, &tmp_from,
3161						ctx->cfile, cifs_sb, &tmp_list,
3162						ctx);
3163
3164					kref_put(&wdata->refcount,
3165						cifs_uncached_writedata_release);
3166				}
3167
3168				list_splice(&tmp_list, &ctx->list);
3169				goto restart_loop;
3170			}
3171		}
3172		list_del_init(&wdata->list);
3173		kref_put(&wdata->refcount, cifs_uncached_writedata_release);
3174	}
3175
3176	cifs_stats_bytes_written(tcon, ctx->total_len);
3177	set_bit(CIFS_INO_INVALID_MAPPING, &CIFS_I(dentry->d_inode)->flags);
3178
3179	ctx->rc = (rc == 0) ? ctx->total_len : rc;
3180
3181	mutex_unlock(&ctx->aio_mutex);
3182
3183	if (ctx->iocb && ctx->iocb->ki_complete)
3184		ctx->iocb->ki_complete(ctx->iocb, ctx->rc, 0);
3185	else
3186		complete(&ctx->done);
3187}
3188
3189static ssize_t __cifs_writev(
3190	struct kiocb *iocb, struct iov_iter *from, bool direct)
3191{
3192	struct file *file = iocb->ki_filp;
3193	ssize_t total_written = 0;
3194	struct cifsFileInfo *cfile;
3195	struct cifs_tcon *tcon;
3196	struct cifs_sb_info *cifs_sb;
3197	struct cifs_aio_ctx *ctx;
3198	struct iov_iter saved_from = *from;
3199	size_t len = iov_iter_count(from);
3200	int rc;
3201
3202	/*
3203	 * iov_iter_get_pages_alloc doesn't work with ITER_KVEC.
3204	 * In this case, fall back to non-direct write function.
3205	 * this could be improved by getting pages directly in ITER_KVEC
3206	 */
3207	if (direct && iov_iter_is_kvec(from)) {
3208		cifs_dbg(FYI, "use non-direct cifs_writev for kvec I/O\n");
3209		direct = false;
3210	}
3211
3212	rc = generic_write_checks(iocb, from);
3213	if (rc <= 0)
3214		return rc;
3215
3216	cifs_sb = CIFS_FILE_SB(file);
3217	cfile = file->private_data;
3218	tcon = tlink_tcon(cfile->tlink);
3219
3220	if (!tcon->ses->server->ops->async_writev)
3221		return -ENOSYS;
3222
3223	ctx = cifs_aio_ctx_alloc();
3224	if (!ctx)
3225		return -ENOMEM;
3226
3227	ctx->cfile = cifsFileInfo_get(cfile);
3228
3229	if (!is_sync_kiocb(iocb))
3230		ctx->iocb = iocb;
3231
3232	ctx->pos = iocb->ki_pos;
3233
3234	if (direct) {
3235		ctx->direct_io = true;
3236		ctx->iter = *from;
3237		ctx->len = len;
3238	} else {
3239		rc = setup_aio_ctx_iter(ctx, from, WRITE);
3240		if (rc) {
3241			kref_put(&ctx->refcount, cifs_aio_ctx_release);
3242			return rc;
3243		}
3244	}
3245
3246	/* grab a lock here due to read response handlers can access ctx */
3247	mutex_lock(&ctx->aio_mutex);
3248
3249	rc = cifs_write_from_iter(iocb->ki_pos, ctx->len, &saved_from,
3250				  cfile, cifs_sb, &ctx->list, ctx);
3251
3252	/*
3253	 * If at least one write was successfully sent, then discard any rc
3254	 * value from the later writes. If the other write succeeds, then
3255	 * we'll end up returning whatever was written. If it fails, then
3256	 * we'll get a new rc value from that.
3257	 */
3258	if (!list_empty(&ctx->list))
3259		rc = 0;
3260
3261	mutex_unlock(&ctx->aio_mutex);
3262
3263	if (rc) {
3264		kref_put(&ctx->refcount, cifs_aio_ctx_release);
3265		return rc;
3266	}
3267
3268	if (!is_sync_kiocb(iocb)) {
3269		kref_put(&ctx->refcount, cifs_aio_ctx_release);
3270		return -EIOCBQUEUED;
3271	}
3272
3273	rc = wait_for_completion_killable(&ctx->done);
3274	if (rc) {
3275		mutex_lock(&ctx->aio_mutex);
3276		ctx->rc = rc = -EINTR;
3277		total_written = ctx->total_len;
3278		mutex_unlock(&ctx->aio_mutex);
3279	} else {
3280		rc = ctx->rc;
3281		total_written = ctx->total_len;
3282	}
3283
3284	kref_put(&ctx->refcount, cifs_aio_ctx_release);
3285
3286	if (unlikely(!total_written))
3287		return rc;
3288
3289	iocb->ki_pos += total_written;
3290	return total_written;
3291}
3292
3293ssize_t cifs_direct_writev(struct kiocb *iocb, struct iov_iter *from)
3294{
3295	return __cifs_writev(iocb, from, true);
3296}
3297
3298ssize_t cifs_user_writev(struct kiocb *iocb, struct iov_iter *from)
3299{
3300	return __cifs_writev(iocb, from, false);
3301}
3302
3303static ssize_t
3304cifs_writev(struct kiocb *iocb, struct iov_iter *from)
3305{
3306	struct file *file = iocb->ki_filp;
3307	struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
3308	struct inode *inode = file->f_mapping->host;
3309	struct cifsInodeInfo *cinode = CIFS_I(inode);
3310	struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
3311	ssize_t rc;
3312
3313	inode_lock(inode);
3314	/*
3315	 * We need to hold the sem to be sure nobody modifies lock list
3316	 * with a brlock that prevents writing.
3317	 */
3318	down_read(&cinode->lock_sem);
3319
3320	rc = generic_write_checks(iocb, from);
3321	if (rc <= 0)
3322		goto out;
3323
3324	if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(from),
3325				     server->vals->exclusive_lock_type, 0,
3326				     NULL, CIFS_WRITE_OP))
3327		rc = __generic_file_write_iter(iocb, from);
3328	else
3329		rc = -EACCES;
3330out:
3331	up_read(&cinode->lock_sem);
3332	inode_unlock(inode);
3333
3334	if (rc > 0)
3335		rc = generic_write_sync(iocb, rc);
3336	return rc;
3337}
3338
3339ssize_t
3340cifs_strict_writev(struct kiocb *iocb, struct iov_iter *from)
3341{
3342	struct inode *inode = file_inode(iocb->ki_filp);
3343	struct cifsInodeInfo *cinode = CIFS_I(inode);
3344	struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
3345	struct cifsFileInfo *cfile = (struct cifsFileInfo *)
3346						iocb->ki_filp->private_data;
3347	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
3348	ssize_t written;
3349
3350	written = cifs_get_writer(cinode);
3351	if (written)
3352		return written;
3353
3354	if (CIFS_CACHE_WRITE(cinode)) {
3355		if (cap_unix(tcon->ses) &&
3356		(CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability))
3357		  && ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0)) {
3358			written = generic_file_write_iter(iocb, from);
3359			goto out;
3360		}
3361		written = cifs_writev(iocb, from);
3362		goto out;
3363	}
3364	/*
3365	 * For non-oplocked files in strict cache mode we need to write the data
3366	 * to the server exactly from the pos to pos+len-1 rather than flush all
3367	 * affected pages because it may cause a error with mandatory locks on
3368	 * these pages but not on the region from pos to ppos+len-1.
3369	 */
3370	written = cifs_user_writev(iocb, from);
3371	if (CIFS_CACHE_READ(cinode)) {
3372		/*
3373		 * We have read level caching and we have just sent a write
3374		 * request to the server thus making data in the cache stale.
3375		 * Zap the cache and set oplock/lease level to NONE to avoid
3376		 * reading stale data from the cache. All subsequent read
3377		 * operations will read new data from the server.
3378		 */
3379		cifs_zap_mapping(inode);
3380		cifs_dbg(FYI, "Set Oplock/Lease to NONE for inode=%p after write\n",
3381			 inode);
3382		cinode->oplock = 0;
3383	}
3384out:
3385	cifs_put_writer(cinode);
3386	return written;
3387}
3388
3389static struct cifs_readdata *
3390cifs_readdata_direct_alloc(struct page **pages, work_func_t complete)
3391{
3392	struct cifs_readdata *rdata;
3393
3394	rdata = kzalloc(sizeof(*rdata), GFP_KERNEL);
3395	if (rdata != NULL) {
3396		rdata->pages = pages;
3397		kref_init(&rdata->refcount);
3398		INIT_LIST_HEAD(&rdata->list);
3399		init_completion(&rdata->done);
3400		INIT_WORK(&rdata->work, complete);
3401	}
3402
3403	return rdata;
3404}
3405
3406static struct cifs_readdata *
3407cifs_readdata_alloc(unsigned int nr_pages, work_func_t complete)
3408{
3409	struct page **pages =
3410		kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
3411	struct cifs_readdata *ret = NULL;
3412
3413	if (pages) {
3414		ret = cifs_readdata_direct_alloc(pages, complete);
3415		if (!ret)
3416			kfree(pages);
3417	}
3418
3419	return ret;
3420}
3421
3422void
3423cifs_readdata_release(struct kref *refcount)
3424{
3425	struct cifs_readdata *rdata = container_of(refcount,
3426					struct cifs_readdata, refcount);
3427#ifdef CONFIG_CIFS_SMB_DIRECT
3428	if (rdata->mr) {
3429		smbd_deregister_mr(rdata->mr);
3430		rdata->mr = NULL;
3431	}
3432#endif
3433	if (rdata->cfile)
3434		cifsFileInfo_put(rdata->cfile);
3435
3436	kvfree(rdata->pages);
3437	kfree(rdata);
3438}
3439
3440static int
3441cifs_read_allocate_pages(struct cifs_readdata *rdata, unsigned int nr_pages)
3442{
3443	int rc = 0;
3444	struct page *page;
3445	unsigned int i;
3446
3447	for (i = 0; i < nr_pages; i++) {
3448		page = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
3449		if (!page) {
3450			rc = -ENOMEM;
3451			break;
3452		}
3453		rdata->pages[i] = page;
3454	}
3455
3456	if (rc) {
3457		unsigned int nr_page_failed = i;
3458
3459		for (i = 0; i < nr_page_failed; i++) {
3460			put_page(rdata->pages[i]);
3461			rdata->pages[i] = NULL;
3462		}
3463	}
3464	return rc;
3465}
3466
3467static void
3468cifs_uncached_readdata_release(struct kref *refcount)
3469{
3470	struct cifs_readdata *rdata = container_of(refcount,
3471					struct cifs_readdata, refcount);
3472	unsigned int i;
3473
3474	kref_put(&rdata->ctx->refcount, cifs_aio_ctx_release);
3475	for (i = 0; i < rdata->nr_pages; i++) {
3476		put_page(rdata->pages[i]);
3477	}
3478	cifs_readdata_release(refcount);
3479}
3480
3481/**
3482 * cifs_readdata_to_iov - copy data from pages in response to an iovec
3483 * @rdata:	the readdata response with list of pages holding data
3484 * @iter:	destination for our data
3485 *
3486 * This function copies data from a list of pages in a readdata response into
3487 * an array of iovecs. It will first calculate where the data should go
3488 * based on the info in the readdata and then copy the data into that spot.
3489 */
3490static int
3491cifs_readdata_to_iov(struct cifs_readdata *rdata, struct iov_iter *iter)
3492{
3493	size_t remaining = rdata->got_bytes;
3494	unsigned int i;
3495
3496	for (i = 0; i < rdata->nr_pages; i++) {
3497		struct page *page = rdata->pages[i];
3498		size_t copy = min_t(size_t, remaining, PAGE_SIZE);
3499		size_t written;
3500
3501		if (unlikely(iov_iter_is_pipe(iter))) {
3502			void *addr = kmap_atomic(page);
3503
3504			written = copy_to_iter(addr, copy, iter);
3505			kunmap_atomic(addr);
3506		} else
3507			written = copy_page_to_iter(page, 0, copy, iter);
3508		remaining -= written;
3509		if (written < copy && iov_iter_count(iter) > 0)
3510			break;
3511	}
3512	return remaining ? -EFAULT : 0;
3513}
3514
3515static void collect_uncached_read_data(struct cifs_aio_ctx *ctx);
3516
3517static void
3518cifs_uncached_readv_complete(struct work_struct *work)
3519{
3520	struct cifs_readdata *rdata = container_of(work,
3521						struct cifs_readdata, work);
3522
3523	complete(&rdata->done);
3524	collect_uncached_read_data(rdata->ctx);
3525	/* the below call can possibly free the last ref to aio ctx */
3526	kref_put(&rdata->refcount, cifs_uncached_readdata_release);
3527}
3528
3529static int
3530uncached_fill_pages(struct TCP_Server_Info *server,
3531		    struct cifs_readdata *rdata, struct iov_iter *iter,
3532		    unsigned int len)
3533{
3534	int result = 0;
3535	unsigned int i;
3536	unsigned int nr_pages = rdata->nr_pages;
3537	unsigned int page_offset = rdata->page_offset;
3538
3539	rdata->got_bytes = 0;
3540	rdata->tailsz = PAGE_SIZE;
3541	for (i = 0; i < nr_pages; i++) {
3542		struct page *page = rdata->pages[i];
3543		size_t n;
3544		unsigned int segment_size = rdata->pagesz;
3545
3546		if (i == 0)
3547			segment_size -= page_offset;
3548		else
3549			page_offset = 0;
3550
3551
3552		if (len <= 0) {
3553			/* no need to hold page hostage */
3554			rdata->pages[i] = NULL;
3555			rdata->nr_pages--;
3556			put_page(page);
3557			continue;
3558		}
3559
3560		n = len;
3561		if (len >= segment_size)
3562			/* enough data to fill the page */
3563			n = segment_size;
3564		else
3565			rdata->tailsz = len;
3566		len -= n;
3567
3568		if (iter)
3569			result = copy_page_from_iter(
3570					page, page_offset, n, iter);
3571#ifdef CONFIG_CIFS_SMB_DIRECT
3572		else if (rdata->mr)
3573			result = n;
3574#endif
3575		else
3576			result = cifs_read_page_from_socket(
3577					server, page, page_offset, n);
3578		if (result < 0)
3579			break;
3580
3581		rdata->got_bytes += result;
3582	}
3583
3584	return rdata->got_bytes > 0 && result != -ECONNABORTED ?
3585						rdata->got_bytes : result;
3586}
3587
3588static int
3589cifs_uncached_read_into_pages(struct TCP_Server_Info *server,
3590			      struct cifs_readdata *rdata, unsigned int len)
3591{
3592	return uncached_fill_pages(server, rdata, NULL, len);
3593}
3594
3595static int
3596cifs_uncached_copy_into_pages(struct TCP_Server_Info *server,
3597			      struct cifs_readdata *rdata,
3598			      struct iov_iter *iter)
3599{
3600	return uncached_fill_pages(server, rdata, iter, iter->count);
3601}
3602
3603static int cifs_resend_rdata(struct cifs_readdata *rdata,
3604			struct list_head *rdata_list,
3605			struct cifs_aio_ctx *ctx)
3606{
3607	unsigned int rsize;
3608	struct cifs_credits credits;
3609	int rc;
3610	struct TCP_Server_Info *server;
3611
3612	/* XXX: should we pick a new channel here? */
3613	server = rdata->server;
3614
3615	do {
3616		if (rdata->cfile->invalidHandle) {
3617			rc = cifs_reopen_file(rdata->cfile, true);
3618			if (rc == -EAGAIN)
3619				continue;
3620			else if (rc)
3621				break;
3622		}
3623
3624		/*
3625		 * Wait for credits to resend this rdata.
3626		 * Note: we are attempting to resend the whole rdata not in
3627		 * segments
3628		 */
3629		do {
3630			rc = server->ops->wait_mtu_credits(server, rdata->bytes,
3631						&rsize, &credits);
3632
3633			if (rc)
3634				goto fail;
3635
3636			if (rsize < rdata->bytes) {
3637				add_credits_and_wake_if(server, &credits, 0);
3638				msleep(1000);
3639			}
3640		} while (rsize < rdata->bytes);
3641		rdata->credits = credits;
3642
3643		rc = adjust_credits(server, &rdata->credits, rdata->bytes);
3644		if (!rc) {
3645			if (rdata->cfile->invalidHandle)
3646				rc = -EAGAIN;
3647			else {
3648#ifdef CONFIG_CIFS_SMB_DIRECT
3649				if (rdata->mr) {
3650					rdata->mr->need_invalidate = true;
3651					smbd_deregister_mr(rdata->mr);
3652					rdata->mr = NULL;
3653				}
3654#endif
3655				rc = server->ops->async_readv(rdata);
3656			}
3657		}
3658
3659		/* If the read was successfully sent, we are done */
3660		if (!rc) {
3661			/* Add to aio pending list */
3662			list_add_tail(&rdata->list, rdata_list);
3663			return 0;
3664		}
3665
3666		/* Roll back credits and retry if needed */
3667		add_credits_and_wake_if(server, &rdata->credits, 0);
3668	} while (rc == -EAGAIN);
3669
3670fail:
3671	kref_put(&rdata->refcount, cifs_uncached_readdata_release);
3672	return rc;
3673}
3674
3675static int
3676cifs_send_async_read(loff_t offset, size_t len, struct cifsFileInfo *open_file,
3677		     struct cifs_sb_info *cifs_sb, struct list_head *rdata_list,
3678		     struct cifs_aio_ctx *ctx)
3679{
3680	struct cifs_readdata *rdata;
3681	unsigned int npages, rsize;
3682	struct cifs_credits credits_on_stack;
3683	struct cifs_credits *credits = &credits_on_stack;
3684	size_t cur_len;
3685	int rc;
3686	pid_t pid;
3687	struct TCP_Server_Info *server;
3688	struct page **pagevec;
3689	size_t start;
3690	struct iov_iter direct_iov = ctx->iter;
3691
3692	server = cifs_pick_channel(tlink_tcon(open_file->tlink)->ses);
3693
3694	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3695		pid = open_file->pid;
3696	else
3697		pid = current->tgid;
3698
3699	if (ctx->direct_io)
3700		iov_iter_advance(&direct_iov, offset - ctx->pos);
3701
3702	do {
3703		if (open_file->invalidHandle) {
3704			rc = cifs_reopen_file(open_file, true);
3705			if (rc == -EAGAIN)
3706				continue;
3707			else if (rc)
3708				break;
3709		}
3710
3711		rc = server->ops->wait_mtu_credits(server, cifs_sb->ctx->rsize,
3712						   &rsize, credits);
3713		if (rc)
3714			break;
3715
3716		cur_len = min_t(const size_t, len, rsize);
3717
3718		if (ctx->direct_io) {
3719			ssize_t result;
3720
3721			result = iov_iter_get_pages_alloc(
3722					&direct_iov, &pagevec,
3723					cur_len, &start);
3724			if (result < 0) {
3725				cifs_dbg(VFS,
3726					 "Couldn't get user pages (rc=%zd) iter type %d iov_offset %zd count %zd\n",
3727					 result, iov_iter_type(&direct_iov),
3728					 direct_iov.iov_offset,
3729					 direct_iov.count);
3730				dump_stack();
3731
3732				rc = result;
3733				add_credits_and_wake_if(server, credits, 0);
3734				break;
3735			}
3736			cur_len = (size_t)result;
3737			iov_iter_advance(&direct_iov, cur_len);
3738
3739			rdata = cifs_readdata_direct_alloc(
3740					pagevec, cifs_uncached_readv_complete);
3741			if (!rdata) {
3742				add_credits_and_wake_if(server, credits, 0);
3743				rc = -ENOMEM;
3744				break;
3745			}
3746
3747			npages = (cur_len + start + PAGE_SIZE-1) / PAGE_SIZE;
3748			rdata->page_offset = start;
3749			rdata->tailsz = npages > 1 ?
3750				cur_len-(PAGE_SIZE-start)-(npages-2)*PAGE_SIZE :
3751				cur_len;
3752
3753		} else {
3754
3755			npages = DIV_ROUND_UP(cur_len, PAGE_SIZE);
3756			/* allocate a readdata struct */
3757			rdata = cifs_readdata_alloc(npages,
3758					    cifs_uncached_readv_complete);
3759			if (!rdata) {
3760				add_credits_and_wake_if(server, credits, 0);
3761				rc = -ENOMEM;
3762				break;
3763			}
3764
3765			rc = cifs_read_allocate_pages(rdata, npages);
3766			if (rc) {
3767				kvfree(rdata->pages);
3768				kfree(rdata);
3769				add_credits_and_wake_if(server, credits, 0);
3770				break;
3771			}
3772
3773			rdata->tailsz = PAGE_SIZE;
3774		}
3775
3776		rdata->server = server;
3777		rdata->cfile = cifsFileInfo_get(open_file);
3778		rdata->nr_pages = npages;
3779		rdata->offset = offset;
3780		rdata->bytes = cur_len;
3781		rdata->pid = pid;
3782		rdata->pagesz = PAGE_SIZE;
3783		rdata->read_into_pages = cifs_uncached_read_into_pages;
3784		rdata->copy_into_pages = cifs_uncached_copy_into_pages;
3785		rdata->credits = credits_on_stack;
3786		rdata->ctx = ctx;
3787		kref_get(&ctx->refcount);
3788
3789		rc = adjust_credits(server, &rdata->credits, rdata->bytes);
3790
3791		if (!rc) {
3792			if (rdata->cfile->invalidHandle)
3793				rc = -EAGAIN;
3794			else
3795				rc = server->ops->async_readv(rdata);
3796		}
3797
3798		if (rc) {
3799			add_credits_and_wake_if(server, &rdata->credits, 0);
3800			kref_put(&rdata->refcount,
3801				cifs_uncached_readdata_release);
3802			if (rc == -EAGAIN) {
3803				iov_iter_revert(&direct_iov, cur_len);
3804				continue;
3805			}
3806			break;
3807		}
3808
3809		list_add_tail(&rdata->list, rdata_list);
3810		offset += cur_len;
3811		len -= cur_len;
3812	} while (len > 0);
3813
3814	return rc;
3815}
3816
3817static void
3818collect_uncached_read_data(struct cifs_aio_ctx *ctx)
3819{
3820	struct cifs_readdata *rdata, *tmp;
3821	struct iov_iter *to = &ctx->iter;
3822	struct cifs_sb_info *cifs_sb;
3823	int rc;
3824
3825	cifs_sb = CIFS_SB(ctx->cfile->dentry->d_sb);
3826
3827	mutex_lock(&ctx->aio_mutex);
3828
3829	if (list_empty(&ctx->list)) {
3830		mutex_unlock(&ctx->aio_mutex);
3831		return;
3832	}
3833
3834	rc = ctx->rc;
3835	/* the loop below should proceed in the order of increasing offsets */
3836again:
3837	list_for_each_entry_safe(rdata, tmp, &ctx->list, list) {
3838		if (!rc) {
3839			if (!try_wait_for_completion(&rdata->done)) {
3840				mutex_unlock(&ctx->aio_mutex);
3841				return;
3842			}
3843
3844			if (rdata->result == -EAGAIN) {
3845				/* resend call if it's a retryable error */
3846				struct list_head tmp_list;
3847				unsigned int got_bytes = rdata->got_bytes;
3848
3849				list_del_init(&rdata->list);
3850				INIT_LIST_HEAD(&tmp_list);
3851
3852				/*
3853				 * Got a part of data and then reconnect has
3854				 * happened -- fill the buffer and continue
3855				 * reading.
3856				 */
3857				if (got_bytes && got_bytes < rdata->bytes) {
3858					rc = 0;
3859					if (!ctx->direct_io)
3860						rc = cifs_readdata_to_iov(rdata, to);
3861					if (rc) {
3862						kref_put(&rdata->refcount,
3863							cifs_uncached_readdata_release);
3864						continue;
3865					}
3866				}
3867
3868				if (ctx->direct_io) {
3869					/*
3870					 * Re-use rdata as this is a
3871					 * direct I/O
3872					 */
3873					rc = cifs_resend_rdata(
3874						rdata,
3875						&tmp_list, ctx);
3876				} else {
3877					rc = cifs_send_async_read(
3878						rdata->offset + got_bytes,
3879						rdata->bytes - got_bytes,
3880						rdata->cfile, cifs_sb,
3881						&tmp_list, ctx);
3882
3883					kref_put(&rdata->refcount,
3884						cifs_uncached_readdata_release);
3885				}
3886
3887				list_splice(&tmp_list, &ctx->list);
3888
3889				goto again;
3890			} else if (rdata->result)
3891				rc = rdata->result;
3892			else if (!ctx->direct_io)
3893				rc = cifs_readdata_to_iov(rdata, to);
3894
3895			/* if there was a short read -- discard anything left */
3896			if (rdata->got_bytes && rdata->got_bytes < rdata->bytes)
3897				rc = -ENODATA;
3898
3899			ctx->total_len += rdata->got_bytes;
3900		}
3901		list_del_init(&rdata->list);
3902		kref_put(&rdata->refcount, cifs_uncached_readdata_release);
3903	}
3904
3905	if (!ctx->direct_io)
3906		ctx->total_len = ctx->len - iov_iter_count(to);
3907
3908	/* mask nodata case */
3909	if (rc == -ENODATA)
3910		rc = 0;
3911
3912	ctx->rc = (rc == 0) ? (ssize_t)ctx->total_len : rc;
3913
3914	mutex_unlock(&ctx->aio_mutex);
3915
3916	if (ctx->iocb && ctx->iocb->ki_complete)
3917		ctx->iocb->ki_complete(ctx->iocb, ctx->rc, 0);
3918	else
3919		complete(&ctx->done);
3920}
3921
3922static ssize_t __cifs_readv(
3923	struct kiocb *iocb, struct iov_iter *to, bool direct)
3924{
3925	size_t len;
3926	struct file *file = iocb->ki_filp;
3927	struct cifs_sb_info *cifs_sb;
3928	struct cifsFileInfo *cfile;
3929	struct cifs_tcon *tcon;
3930	ssize_t rc, total_read = 0;
3931	loff_t offset = iocb->ki_pos;
3932	struct cifs_aio_ctx *ctx;
3933
3934	/*
3935	 * iov_iter_get_pages_alloc() doesn't work with ITER_KVEC,
3936	 * fall back to data copy read path
3937	 * this could be improved by getting pages directly in ITER_KVEC
3938	 */
3939	if (direct && iov_iter_is_kvec(to)) {
3940		cifs_dbg(FYI, "use non-direct cifs_user_readv for kvec I/O\n");
3941		direct = false;
3942	}
3943
3944	len = iov_iter_count(to);
3945	if (!len)
3946		return 0;
3947
3948	cifs_sb = CIFS_FILE_SB(file);
3949	cfile = file->private_data;
3950	tcon = tlink_tcon(cfile->tlink);
3951
3952	if (!tcon->ses->server->ops->async_readv)
3953		return -ENOSYS;
3954
3955	if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3956		cifs_dbg(FYI, "attempting read on write only file instance\n");
3957
3958	ctx = cifs_aio_ctx_alloc();
3959	if (!ctx)
3960		return -ENOMEM;
3961
3962	ctx->cfile = cifsFileInfo_get(cfile);
3963
3964	if (!is_sync_kiocb(iocb))
3965		ctx->iocb = iocb;
3966
3967	if (iter_is_iovec(to))
3968		ctx->should_dirty = true;
3969
3970	if (direct) {
3971		ctx->pos = offset;
3972		ctx->direct_io = true;
3973		ctx->iter = *to;
3974		ctx->len = len;
3975	} else {
3976		rc = setup_aio_ctx_iter(ctx, to, READ);
3977		if (rc) {
3978			kref_put(&ctx->refcount, cifs_aio_ctx_release);
3979			return rc;
3980		}
3981		len = ctx->len;
3982	}
3983
3984	/* grab a lock here due to read response handlers can access ctx */
3985	mutex_lock(&ctx->aio_mutex);
3986
3987	rc = cifs_send_async_read(offset, len, cfile, cifs_sb, &ctx->list, ctx);
3988
3989	/* if at least one read request send succeeded, then reset rc */
3990	if (!list_empty(&ctx->list))
3991		rc = 0;
3992
3993	mutex_unlock(&ctx->aio_mutex);
3994
3995	if (rc) {
3996		kref_put(&ctx->refcount, cifs_aio_ctx_release);
3997		return rc;
3998	}
3999
4000	if (!is_sync_kiocb(iocb)) {
4001		kref_put(&ctx->refcount, cifs_aio_ctx_release);
4002		return -EIOCBQUEUED;
4003	}
4004
4005	rc = wait_for_completion_killable(&ctx->done);
4006	if (rc) {
4007		mutex_lock(&ctx->aio_mutex);
4008		ctx->rc = rc = -EINTR;
4009		total_read = ctx->total_len;
4010		mutex_unlock(&ctx->aio_mutex);
4011	} else {
4012		rc = ctx->rc;
4013		total_read = ctx->total_len;
4014	}
4015
4016	kref_put(&ctx->refcount, cifs_aio_ctx_release);
4017
4018	if (total_read) {
4019		iocb->ki_pos += total_read;
4020		return total_read;
4021	}
4022	return rc;
4023}
4024
4025ssize_t cifs_direct_readv(struct kiocb *iocb, struct iov_iter *to)
4026{
4027	return __cifs_readv(iocb, to, true);
4028}
4029
4030ssize_t cifs_user_readv(struct kiocb *iocb, struct iov_iter *to)
4031{
4032	return __cifs_readv(iocb, to, false);
4033}
4034
4035ssize_t
4036cifs_strict_readv(struct kiocb *iocb, struct iov_iter *to)
4037{
4038	struct inode *inode = file_inode(iocb->ki_filp);
4039	struct cifsInodeInfo *cinode = CIFS_I(inode);
4040	struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
4041	struct cifsFileInfo *cfile = (struct cifsFileInfo *)
4042						iocb->ki_filp->private_data;
4043	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
4044	int rc = -EACCES;
4045
4046	/*
4047	 * In strict cache mode we need to read from the server all the time
4048	 * if we don't have level II oplock because the server can delay mtime
4049	 * change - so we can't make a decision about inode invalidating.
4050	 * And we can also fail with pagereading if there are mandatory locks
4051	 * on pages affected by this read but not on the region from pos to
4052	 * pos+len-1.
4053	 */
4054	if (!CIFS_CACHE_READ(cinode))
4055		return cifs_user_readv(iocb, to);
4056
4057	if (cap_unix(tcon->ses) &&
4058	    (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
4059	    ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
4060		return generic_file_read_iter(iocb, to);
4061
4062	/*
4063	 * We need to hold the sem to be sure nobody modifies lock list
4064	 * with a brlock that prevents reading.
4065	 */
4066	down_read(&cinode->lock_sem);
4067	if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(to),
4068				     tcon->ses->server->vals->shared_lock_type,
4069				     0, NULL, CIFS_READ_OP))
4070		rc = generic_file_read_iter(iocb, to);
4071	up_read(&cinode->lock_sem);
4072	return rc;
4073}
4074
4075static ssize_t
4076cifs_read(struct file *file, char *read_data, size_t read_size, loff_t *offset)
4077{
4078	int rc = -EACCES;
4079	unsigned int bytes_read = 0;
4080	unsigned int total_read;
4081	unsigned int current_read_size;
4082	unsigned int rsize;
4083	struct cifs_sb_info *cifs_sb;
4084	struct cifs_tcon *tcon;
4085	struct TCP_Server_Info *server;
4086	unsigned int xid;
4087	char *cur_offset;
4088	struct cifsFileInfo *open_file;
4089	struct cifs_io_parms io_parms = {0};
4090	int buf_type = CIFS_NO_BUFFER;
4091	__u32 pid;
4092
4093	xid = get_xid();
4094	cifs_sb = CIFS_FILE_SB(file);
4095
4096	/* FIXME: set up handlers for larger reads and/or convert to async */
4097	rsize = min_t(unsigned int, cifs_sb->ctx->rsize, CIFSMaxBufSize);
4098
4099	if (file->private_data == NULL) {
4100		rc = -EBADF;
4101		free_xid(xid);
4102		return rc;
4103	}
4104	open_file = file->private_data;
4105	tcon = tlink_tcon(open_file->tlink);
4106	server = cifs_pick_channel(tcon->ses);
4107
4108	if (!server->ops->sync_read) {
4109		free_xid(xid);
4110		return -ENOSYS;
4111	}
4112
4113	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
4114		pid = open_file->pid;
4115	else
4116		pid = current->tgid;
4117
4118	if ((file->f_flags & O_ACCMODE) == O_WRONLY)
4119		cifs_dbg(FYI, "attempting read on write only file instance\n");
4120
4121	for (total_read = 0, cur_offset = read_data; read_size > total_read;
4122	     total_read += bytes_read, cur_offset += bytes_read) {
4123		do {
4124			current_read_size = min_t(uint, read_size - total_read,
4125						  rsize);
4126			/*
4127			 * For windows me and 9x we do not want to request more
4128			 * than it negotiated since it will refuse the read
4129			 * then.
4130			 */
4131			if (!(tcon->ses->capabilities &
4132				tcon->ses->server->vals->cap_large_files)) {
4133				current_read_size = min_t(uint,
4134					current_read_size, CIFSMaxBufSize);
4135			}
4136			if (open_file->invalidHandle) {
4137				rc = cifs_reopen_file(open_file, true);
4138				if (rc != 0)
4139					break;
4140			}
4141			io_parms.pid = pid;
4142			io_parms.tcon = tcon;
4143			io_parms.offset = *offset;
4144			io_parms.length = current_read_size;
4145			io_parms.server = server;
4146			rc = server->ops->sync_read(xid, &open_file->fid, &io_parms,
4147						    &bytes_read, &cur_offset,
4148						    &buf_type);
4149		} while (rc == -EAGAIN);
4150
4151		if (rc || (bytes_read == 0)) {
4152			if (total_read) {
4153				break;
4154			} else {
4155				free_xid(xid);
4156				return rc;
4157			}
4158		} else {
4159			cifs_stats_bytes_read(tcon, total_read);
4160			*offset += bytes_read;
4161		}
4162	}
4163	free_xid(xid);
4164	return total_read;
4165}
4166
4167/*
4168 * If the page is mmap'ed into a process' page tables, then we need to make
4169 * sure that it doesn't change while being written back.
4170 */
4171static vm_fault_t
4172cifs_page_mkwrite(struct vm_fault *vmf)
4173{
4174	struct page *page = vmf->page;
4175
4176	lock_page(page);
4177	return VM_FAULT_LOCKED;
4178}
4179
4180static const struct vm_operations_struct cifs_file_vm_ops = {
4181	.fault = filemap_fault,
4182	.map_pages = filemap_map_pages,
4183	.page_mkwrite = cifs_page_mkwrite,
4184};
4185
4186int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
4187{
4188	int xid, rc = 0;
4189	struct inode *inode = file_inode(file);
4190
4191	xid = get_xid();
4192
4193	if (!CIFS_CACHE_READ(CIFS_I(inode)))
4194		rc = cifs_zap_mapping(inode);
4195	if (!rc)
4196		rc = generic_file_mmap(file, vma);
4197	if (!rc)
4198		vma->vm_ops = &cifs_file_vm_ops;
4199
4200	free_xid(xid);
4201	return rc;
4202}
4203
4204int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
4205{
4206	int rc, xid;
4207
4208	xid = get_xid();
4209
4210	rc = cifs_revalidate_file(file);
4211	if (rc)
4212		cifs_dbg(FYI, "Validation prior to mmap failed, error=%d\n",
4213			 rc);
4214	if (!rc)
4215		rc = generic_file_mmap(file, vma);
4216	if (!rc)
4217		vma->vm_ops = &cifs_file_vm_ops;
4218
4219	free_xid(xid);
4220	return rc;
4221}
4222
4223static void
4224cifs_readv_complete(struct work_struct *work)
4225{
4226	unsigned int i, got_bytes;
4227	struct cifs_readdata *rdata = container_of(work,
4228						struct cifs_readdata, work);
4229
4230	got_bytes = rdata->got_bytes;
4231	for (i = 0; i < rdata->nr_pages; i++) {
4232		struct page *page = rdata->pages[i];
4233
4234		lru_cache_add(page);
4235
4236		if (rdata->result == 0 ||
4237		    (rdata->result == -EAGAIN && got_bytes)) {
4238			flush_dcache_page(page);
4239			SetPageUptodate(page);
4240		}
4241
4242		unlock_page(page);
4243
4244		if (rdata->result == 0 ||
4245		    (rdata->result == -EAGAIN && got_bytes))
4246			cifs_readpage_to_fscache(rdata->mapping->host, page);
4247
4248		got_bytes -= min_t(unsigned int, PAGE_SIZE, got_bytes);
4249
4250		put_page(page);
4251		rdata->pages[i] = NULL;
4252	}
4253	kref_put(&rdata->refcount, cifs_readdata_release);
4254}
4255
4256static int
4257readpages_fill_pages(struct TCP_Server_Info *server,
4258		     struct cifs_readdata *rdata, struct iov_iter *iter,
4259		     unsigned int len)
4260{
4261	int result = 0;
4262	unsigned int i;
4263	u64 eof;
4264	pgoff_t eof_index;
4265	unsigned int nr_pages = rdata->nr_pages;
4266	unsigned int page_offset = rdata->page_offset;
4267
4268	/* determine the eof that the server (probably) has */
4269	eof = CIFS_I(rdata->mapping->host)->server_eof;
4270	eof_index = eof ? (eof - 1) >> PAGE_SHIFT : 0;
4271	cifs_dbg(FYI, "eof=%llu eof_index=%lu\n", eof, eof_index);
4272
4273	rdata->got_bytes = 0;
4274	rdata->tailsz = PAGE_SIZE;
4275	for (i = 0; i < nr_pages; i++) {
4276		struct page *page = rdata->pages[i];
4277		unsigned int to_read = rdata->pagesz;
4278		size_t n;
4279
4280		if (i == 0)
4281			to_read -= page_offset;
4282		else
4283			page_offset = 0;
4284
4285		n = to_read;
4286
4287		if (len >= to_read) {
4288			len -= to_read;
4289		} else if (len > 0) {
4290			/* enough for partial page, fill and zero the rest */
4291			zero_user(page, len + page_offset, to_read - len);
4292			n = rdata->tailsz = len;
4293			len = 0;
4294		} else if (page->index > eof_index) {
4295			/*
4296			 * The VFS will not try to do readahead past the
4297			 * i_size, but it's possible that we have outstanding
4298			 * writes with gaps in the middle and the i_size hasn't
4299			 * caught up yet. Populate those with zeroed out pages
4300			 * to prevent the VFS from repeatedly attempting to
4301			 * fill them until the writes are flushed.
4302			 */
4303			zero_user(page, 0, PAGE_SIZE);
4304			lru_cache_add(page);
4305			flush_dcache_page(page);
4306			SetPageUptodate(page);
4307			unlock_page(page);
4308			put_page(page);
4309			rdata->pages[i] = NULL;
4310			rdata->nr_pages--;
4311			continue;
4312		} else {
4313			/* no need to hold page hostage */
4314			lru_cache_add(page);
4315			unlock_page(page);
4316			put_page(page);
4317			rdata->pages[i] = NULL;
4318			rdata->nr_pages--;
4319			continue;
4320		}
4321
4322		if (iter)
4323			result = copy_page_from_iter(
4324					page, page_offset, n, iter);
4325#ifdef CONFIG_CIFS_SMB_DIRECT
4326		else if (rdata->mr)
4327			result = n;
4328#endif
4329		else
4330			result = cifs_read_page_from_socket(
4331					server, page, page_offset, n);
4332		if (result < 0)
4333			break;
4334
4335		rdata->got_bytes += result;
4336	}
4337
4338	return rdata->got_bytes > 0 && result != -ECONNABORTED ?
4339						rdata->got_bytes : result;
4340}
4341
4342static int
4343cifs_readpages_read_into_pages(struct TCP_Server_Info *server,
4344			       struct cifs_readdata *rdata, unsigned int len)
4345{
4346	return readpages_fill_pages(server, rdata, NULL, len);
4347}
4348
4349static int
4350cifs_readpages_copy_into_pages(struct TCP_Server_Info *server,
4351			       struct cifs_readdata *rdata,
4352			       struct iov_iter *iter)
4353{
4354	return readpages_fill_pages(server, rdata, iter, iter->count);
4355}
4356
4357static int
4358readpages_get_pages(struct address_space *mapping, struct list_head *page_list,
4359		    unsigned int rsize, struct list_head *tmplist,
4360		    unsigned int *nr_pages, loff_t *offset, unsigned int *bytes)
4361{
4362	struct page *page, *tpage;
4363	unsigned int expected_index;
4364	int rc;
4365	gfp_t gfp = readahead_gfp_mask(mapping);
4366
4367	INIT_LIST_HEAD(tmplist);
4368
4369	page = lru_to_page(page_list);
4370
4371	/*
4372	 * Lock the page and put it in the cache. Since no one else
4373	 * should have access to this page, we're safe to simply set
4374	 * PG_locked without checking it first.
4375	 */
4376	__SetPageLocked(page);
4377	rc = add_to_page_cache_locked(page, mapping,
4378				      page->index, gfp);
4379
4380	/* give up if we can't stick it in the cache */
4381	if (rc) {
4382		__ClearPageLocked(page);
4383		return rc;
4384	}
4385
4386	/* move first page to the tmplist */
4387	*offset = (loff_t)page->index << PAGE_SHIFT;
4388	*bytes = PAGE_SIZE;
4389	*nr_pages = 1;
4390	list_move_tail(&page->lru, tmplist);
4391
4392	/* now try and add more pages onto the request */
4393	expected_index = page->index + 1;
4394	list_for_each_entry_safe_reverse(page, tpage, page_list, lru) {
4395		/* discontinuity ? */
4396		if (page->index != expected_index)
4397			break;
4398
4399		/* would this page push the read over the rsize? */
4400		if (*bytes + PAGE_SIZE > rsize)
4401			break;
4402
4403		__SetPageLocked(page);
4404		rc = add_to_page_cache_locked(page, mapping, page->index, gfp);
4405		if (rc) {
4406			__ClearPageLocked(page);
4407			break;
4408		}
4409		list_move_tail(&page->lru, tmplist);
4410		(*bytes) += PAGE_SIZE;
4411		expected_index++;
4412		(*nr_pages)++;
4413	}
4414	return rc;
4415}
4416
4417static int cifs_readpages(struct file *file, struct address_space *mapping,
4418	struct list_head *page_list, unsigned num_pages)
4419{
4420	int rc;
4421	int err = 0;
4422	struct list_head tmplist;
4423	struct cifsFileInfo *open_file = file->private_data;
4424	struct cifs_sb_info *cifs_sb = CIFS_FILE_SB(file);
4425	struct TCP_Server_Info *server;
4426	pid_t pid;
4427	unsigned int xid;
4428
4429	xid = get_xid();
4430	/*
4431	 * Reads as many pages as possible from fscache. Returns -ENOBUFS
4432	 * immediately if the cookie is negative
4433	 *
4434	 * After this point, every page in the list might have PG_fscache set,
4435	 * so we will need to clean that up off of every page we don't use.
4436	 */
4437	rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
4438					 &num_pages);
4439	if (rc == 0) {
4440		free_xid(xid);
4441		return rc;
4442	}
4443
4444	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
4445		pid = open_file->pid;
4446	else
4447		pid = current->tgid;
4448
4449	rc = 0;
4450	server = cifs_pick_channel(tlink_tcon(open_file->tlink)->ses);
4451
4452	cifs_dbg(FYI, "%s: file=%p mapping=%p num_pages=%u\n",
4453		 __func__, file, mapping, num_pages);
4454
4455	/*
4456	 * Start with the page at end of list and move it to private
4457	 * list. Do the same with any following pages until we hit
4458	 * the rsize limit, hit an index discontinuity, or run out of
4459	 * pages. Issue the async read and then start the loop again
4460	 * until the list is empty.
4461	 *
4462	 * Note that list order is important. The page_list is in
4463	 * the order of declining indexes. When we put the pages in
4464	 * the rdata->pages, then we want them in increasing order.
4465	 */
4466	while (!list_empty(page_list) && !err) {
4467		unsigned int i, nr_pages, bytes, rsize;
4468		loff_t offset;
4469		struct page *page, *tpage;
4470		struct cifs_readdata *rdata;
4471		struct cifs_credits credits_on_stack;
4472		struct cifs_credits *credits = &credits_on_stack;
4473
4474		if (open_file->invalidHandle) {
4475			rc = cifs_reopen_file(open_file, true);
4476			if (rc == -EAGAIN)
4477				continue;
4478			else if (rc)
4479				break;
4480		}
4481
4482		rc = server->ops->wait_mtu_credits(server, cifs_sb->ctx->rsize,
4483						   &rsize, credits);
4484		if (rc)
4485			break;
4486
4487		/*
4488		 * Give up immediately if rsize is too small to read an entire
4489		 * page. The VFS will fall back to readpage. We should never
4490		 * reach this point however since we set ra_pages to 0 when the
4491		 * rsize is smaller than a cache page.
4492		 */
4493		if (unlikely(rsize < PAGE_SIZE)) {
4494			add_credits_and_wake_if(server, credits, 0);
4495			free_xid(xid);
4496			return 0;
4497		}
4498
4499		nr_pages = 0;
4500		err = readpages_get_pages(mapping, page_list, rsize, &tmplist,
4501					 &nr_pages, &offset, &bytes);
4502		if (!nr_pages) {
4503			add_credits_and_wake_if(server, credits, 0);
4504			break;
4505		}
4506
4507		rdata = cifs_readdata_alloc(nr_pages, cifs_readv_complete);
4508		if (!rdata) {
4509			/* best to give up if we're out of mem */
4510			list_for_each_entry_safe(page, tpage, &tmplist, lru) {
4511				list_del(&page->lru);
4512				lru_cache_add(page);
4513				unlock_page(page);
4514				put_page(page);
4515			}
4516			rc = -ENOMEM;
4517			add_credits_and_wake_if(server, credits, 0);
4518			break;
4519		}
4520
4521		rdata->cfile = cifsFileInfo_get(open_file);
4522		rdata->server = server;
4523		rdata->mapping = mapping;
4524		rdata->offset = offset;
4525		rdata->bytes = bytes;
4526		rdata->pid = pid;
4527		rdata->pagesz = PAGE_SIZE;
4528		rdata->tailsz = PAGE_SIZE;
4529		rdata->read_into_pages = cifs_readpages_read_into_pages;
4530		rdata->copy_into_pages = cifs_readpages_copy_into_pages;
4531		rdata->credits = credits_on_stack;
4532
4533		list_for_each_entry_safe(page, tpage, &tmplist, lru) {
4534			list_del(&page->lru);
4535			rdata->pages[rdata->nr_pages++] = page;
4536		}
4537
4538		rc = adjust_credits(server, &rdata->credits, rdata->bytes);
4539
4540		if (!rc) {
4541			if (rdata->cfile->invalidHandle)
4542				rc = -EAGAIN;
4543			else
4544				rc = server->ops->async_readv(rdata);
4545		}
4546
4547		if (rc) {
4548			add_credits_and_wake_if(server, &rdata->credits, 0);
4549			for (i = 0; i < rdata->nr_pages; i++) {
4550				page = rdata->pages[i];
4551				lru_cache_add(page);
4552				unlock_page(page);
4553				put_page(page);
4554			}
4555			/* Fallback to the readpage in error/reconnect cases */
4556			kref_put(&rdata->refcount, cifs_readdata_release);
4557			break;
4558		}
4559
4560		kref_put(&rdata->refcount, cifs_readdata_release);
4561	}
4562
4563	/* Any pages that have been shown to fscache but didn't get added to
4564	 * the pagecache must be uncached before they get returned to the
4565	 * allocator.
4566	 */
4567	cifs_fscache_readpages_cancel(mapping->host, page_list);
4568	free_xid(xid);
4569	return rc;
4570}
4571
4572/*
4573 * cifs_readpage_worker must be called with the page pinned
4574 */
4575static int cifs_readpage_worker(struct file *file, struct page *page,
4576	loff_t *poffset)
4577{
4578	char *read_data;
4579	int rc;
4580
4581	/* Is the page cached? */
4582	rc = cifs_readpage_from_fscache(file_inode(file), page);
4583	if (rc == 0)
4584		goto read_complete;
4585
4586	read_data = kmap(page);
4587	/* for reads over a certain size could initiate async read ahead */
4588
4589	rc = cifs_read(file, read_data, PAGE_SIZE, poffset);
4590
4591	if (rc < 0)
4592		goto io_error;
4593	else
4594		cifs_dbg(FYI, "Bytes read %d\n", rc);
4595
4596	/* we do not want atime to be less than mtime, it broke some apps */
4597	file_inode(file)->i_atime = current_time(file_inode(file));
4598	if (timespec64_compare(&(file_inode(file)->i_atime), &(file_inode(file)->i_mtime)))
4599		file_inode(file)->i_atime = file_inode(file)->i_mtime;
4600	else
4601		file_inode(file)->i_atime = current_time(file_inode(file));
4602
4603	if (PAGE_SIZE > rc)
4604		memset(read_data + rc, 0, PAGE_SIZE - rc);
4605
4606	flush_dcache_page(page);
4607	SetPageUptodate(page);
4608
4609	/* send this page to the cache */
4610	cifs_readpage_to_fscache(file_inode(file), page);
4611
4612	rc = 0;
4613
4614io_error:
4615	kunmap(page);
4616	unlock_page(page);
4617
4618read_complete:
4619	return rc;
4620}
4621
4622static int cifs_readpage(struct file *file, struct page *page)
4623{
4624	loff_t offset = page_file_offset(page);
4625	int rc = -EACCES;
4626	unsigned int xid;
4627
4628	xid = get_xid();
4629
4630	if (file->private_data == NULL) {
4631		rc = -EBADF;
4632		free_xid(xid);
4633		return rc;
4634	}
4635
4636	cifs_dbg(FYI, "readpage %p at offset %d 0x%x\n",
4637		 page, (int)offset, (int)offset);
4638
4639	rc = cifs_readpage_worker(file, page, &offset);
4640
4641	free_xid(xid);
4642	return rc;
4643}
4644
4645static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
4646{
4647	struct cifsFileInfo *open_file;
4648
4649	spin_lock(&cifs_inode->open_file_lock);
4650	list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
4651		if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
4652			spin_unlock(&cifs_inode->open_file_lock);
4653			return 1;
4654		}
4655	}
4656	spin_unlock(&cifs_inode->open_file_lock);
4657	return 0;
4658}
4659
4660/* We do not want to update the file size from server for inodes
4661   open for write - to avoid races with writepage extending
4662   the file - in the future we could consider allowing
4663   refreshing the inode only on increases in the file size
4664   but this is tricky to do without racing with writebehind
4665   page caching in the current Linux kernel design */
4666bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
4667{
4668	if (!cifsInode)
4669		return true;
4670
4671	if (is_inode_writable(cifsInode)) {
4672		/* This inode is open for write at least once */
4673		struct cifs_sb_info *cifs_sb;
4674
4675		cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
4676		if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
4677			/* since no page cache to corrupt on directio
4678			we can change size safely */
4679			return true;
4680		}
4681
4682		if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
4683			return true;
4684
4685		return false;
4686	} else
4687		return true;
4688}
4689
4690static int cifs_write_begin(struct file *file, struct address_space *mapping,
4691			loff_t pos, unsigned len, unsigned flags,
4692			struct page **pagep, void **fsdata)
4693{
4694	int oncethru = 0;
4695	pgoff_t index = pos >> PAGE_SHIFT;
4696	loff_t offset = pos & (PAGE_SIZE - 1);
4697	loff_t page_start = pos & PAGE_MASK;
4698	loff_t i_size;
4699	struct page *page;
4700	int rc = 0;
4701
4702	cifs_dbg(FYI, "write_begin from %lld len %d\n", (long long)pos, len);
4703
4704start:
4705	page = grab_cache_page_write_begin(mapping, index, flags);
4706	if (!page) {
4707		rc = -ENOMEM;
4708		goto out;
4709	}
4710
4711	if (PageUptodate(page))
4712		goto out;
4713
4714	/*
4715	 * If we write a full page it will be up to date, no need to read from
4716	 * the server. If the write is short, we'll end up doing a sync write
4717	 * instead.
4718	 */
4719	if (len == PAGE_SIZE)
4720		goto out;
4721
4722	/*
4723	 * optimize away the read when we have an oplock, and we're not
4724	 * expecting to use any of the data we'd be reading in. That
4725	 * is, when the page lies beyond the EOF, or straddles the EOF
4726	 * and the write will cover all of the existing data.
4727	 */
4728	if (CIFS_CACHE_READ(CIFS_I(mapping->host))) {
4729		i_size = i_size_read(mapping->host);
4730		if (page_start >= i_size ||
4731		    (offset == 0 && (pos + len) >= i_size)) {
4732			zero_user_segments(page, 0, offset,
4733					   offset + len,
4734					   PAGE_SIZE);
4735			/*
4736			 * PageChecked means that the parts of the page
4737			 * to which we're not writing are considered up
4738			 * to date. Once the data is copied to the
4739			 * page, it can be set uptodate.
4740			 */
4741			SetPageChecked(page);
4742			goto out;
4743		}
4744	}
4745
4746	if ((file->f_flags & O_ACCMODE) != O_WRONLY && !oncethru) {
4747		/*
4748		 * might as well read a page, it is fast enough. If we get
4749		 * an error, we don't need to return it. cifs_write_end will
4750		 * do a sync write instead since PG_uptodate isn't set.
4751		 */
4752		cifs_readpage_worker(file, page, &page_start);
4753		put_page(page);
4754		oncethru = 1;
4755		goto start;
4756	} else {
4757		/* we could try using another file handle if there is one -
4758		   but how would we lock it to prevent close of that handle
4759		   racing with this read? In any case
4760		   this will be written out by write_end so is fine */
4761	}
4762out:
4763	*pagep = page;
4764	return rc;
4765}
4766
4767static int cifs_release_page(struct page *page, gfp_t gfp)
4768{
4769	if (PagePrivate(page))
4770		return 0;
4771
4772	return cifs_fscache_release_page(page, gfp);
4773}
4774
4775static void cifs_invalidate_page(struct page *page, unsigned int offset,
4776				 unsigned int length)
4777{
4778	struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
4779
4780	if (offset == 0 && length == PAGE_SIZE)
4781		cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
4782}
4783
4784static int cifs_launder_page(struct page *page)
4785{
4786	int rc = 0;
4787	loff_t range_start = page_offset(page);
4788	loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1);
4789	struct writeback_control wbc = {
4790		.sync_mode = WB_SYNC_ALL,
4791		.nr_to_write = 0,
4792		.range_start = range_start,
4793		.range_end = range_end,
4794	};
4795
4796	cifs_dbg(FYI, "Launder page: %p\n", page);
4797
4798	if (clear_page_dirty_for_io(page))
4799		rc = cifs_writepage_locked(page, &wbc);
4800
4801	cifs_fscache_invalidate_page(page, page->mapping->host);
4802	return rc;
4803}
4804
4805void cifs_oplock_break(struct work_struct *work)
4806{
4807	struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
4808						  oplock_break);
4809	struct inode *inode = d_inode(cfile->dentry);
4810	struct cifsInodeInfo *cinode = CIFS_I(inode);
4811	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
4812	struct TCP_Server_Info *server = tcon->ses->server;
4813	int rc = 0;
4814	bool purge_cache = false;
4815	bool is_deferred = false;
4816	struct cifs_deferred_close *dclose;
4817
4818	wait_on_bit(&cinode->flags, CIFS_INODE_PENDING_WRITERS,
4819			TASK_UNINTERRUPTIBLE);
4820
4821	server->ops->downgrade_oplock(server, cinode, cfile->oplock_level,
4822				      cfile->oplock_epoch, &purge_cache);
4823
4824	if (!CIFS_CACHE_WRITE(cinode) && CIFS_CACHE_READ(cinode) &&
4825						cifs_has_mand_locks(cinode)) {
4826		cifs_dbg(FYI, "Reset oplock to None for inode=%p due to mand locks\n",
4827			 inode);
4828		cinode->oplock = 0;
4829	}
4830
4831	if (inode && S_ISREG(inode->i_mode)) {
4832		if (CIFS_CACHE_READ(cinode))
4833			break_lease(inode, O_RDONLY);
4834		else
4835			break_lease(inode, O_WRONLY);
4836		rc = filemap_fdatawrite(inode->i_mapping);
4837		if (!CIFS_CACHE_READ(cinode) || purge_cache) {
4838			rc = filemap_fdatawait(inode->i_mapping);
4839			mapping_set_error(inode->i_mapping, rc);
4840			cifs_zap_mapping(inode);
4841		}
4842		cifs_dbg(FYI, "Oplock flush inode %p rc %d\n", inode, rc);
4843		if (CIFS_CACHE_WRITE(cinode))
4844			goto oplock_break_ack;
4845	}
4846
4847	rc = cifs_push_locks(cfile);
4848	if (rc)
4849		cifs_dbg(VFS, "Push locks rc = %d\n", rc);
4850
4851oplock_break_ack:
4852	/*
4853	 * When oplock break is received and there are no active
4854	 * file handles but cached, then schedule deferred close immediately.
4855	 * So, new open will not use cached handle.
4856	 */
4857	spin_lock(&CIFS_I(inode)->deferred_lock);
4858	is_deferred = cifs_is_deferred_close(cfile, &dclose);
4859	spin_unlock(&CIFS_I(inode)->deferred_lock);
4860	if (is_deferred &&
4861	    cfile->deferred_close_scheduled &&
4862	    delayed_work_pending(&cfile->deferred)) {
4863		if (cancel_delayed_work(&cfile->deferred)) {
4864			_cifsFileInfo_put(cfile, false, false);
4865			goto oplock_break_done;
4866		}
4867	}
4868	/*
4869	 * releasing stale oplock after recent reconnect of smb session using
4870	 * a now incorrect file handle is not a data integrity issue but do
4871	 * not bother sending an oplock release if session to server still is
4872	 * disconnected since oplock already released by the server
4873	 */
4874	if (!cfile->oplock_break_cancelled) {
4875		rc = tcon->ses->server->ops->oplock_response(tcon, &cfile->fid,
4876							     cinode);
4877		cifs_dbg(FYI, "Oplock release rc = %d\n", rc);
4878	}
4879oplock_break_done:
4880	_cifsFileInfo_put(cfile, false /* do not wait for ourself */, false);
4881	cifs_done_oplock_break(cinode);
4882}
4883
4884/*
4885 * The presence of cifs_direct_io() in the address space ops vector
4886 * allowes open() O_DIRECT flags which would have failed otherwise.
4887 *
4888 * In the non-cached mode (mount with cache=none), we shunt off direct read and write requests
4889 * so this method should never be called.
4890 *
4891 * Direct IO is not yet supported in the cached mode. 
4892 */
4893static ssize_t
4894cifs_direct_io(struct kiocb *iocb, struct iov_iter *iter)
4895{
4896        /*
4897         * FIXME
4898         * Eventually need to support direct IO for non forcedirectio mounts
4899         */
4900        return -EINVAL;
4901}
4902
4903static int cifs_swap_activate(struct swap_info_struct *sis,
4904			      struct file *swap_file, sector_t *span)
4905{
4906	struct cifsFileInfo *cfile = swap_file->private_data;
4907	struct inode *inode = swap_file->f_mapping->host;
4908	unsigned long blocks;
4909	long long isize;
4910
4911	cifs_dbg(FYI, "swap activate\n");
4912
4913	spin_lock(&inode->i_lock);
4914	blocks = inode->i_blocks;
4915	isize = inode->i_size;
4916	spin_unlock(&inode->i_lock);
4917	if (blocks*512 < isize) {
4918		pr_warn("swap activate: swapfile has holes\n");
4919		return -EINVAL;
4920	}
4921	*span = sis->pages;
4922
4923	pr_warn_once("Swap support over SMB3 is experimental\n");
4924
4925	/*
4926	 * TODO: consider adding ACL (or documenting how) to prevent other
4927	 * users (on this or other systems) from reading it
4928	 */
4929
4930
4931	/* TODO: add sk_set_memalloc(inet) or similar */
4932
4933	if (cfile)
4934		cfile->swapfile = true;
4935	/*
4936	 * TODO: Since file already open, we can't open with DENY_ALL here
4937	 * but we could add call to grab a byte range lock to prevent others
4938	 * from reading or writing the file
4939	 */
4940
4941	return 0;
4942}
4943
4944static void cifs_swap_deactivate(struct file *file)
4945{
4946	struct cifsFileInfo *cfile = file->private_data;
4947
4948	cifs_dbg(FYI, "swap deactivate\n");
4949
4950	/* TODO: undo sk_set_memalloc(inet) will eventually be needed */
4951
4952	if (cfile)
4953		cfile->swapfile = false;
4954
4955	/* do we need to unpin (or unlock) the file */
4956}
4957
4958const struct address_space_operations cifs_addr_ops = {
4959	.readpage = cifs_readpage,
4960	.readpages = cifs_readpages,
4961	.writepage = cifs_writepage,
4962	.writepages = cifs_writepages,
4963	.write_begin = cifs_write_begin,
4964	.write_end = cifs_write_end,
4965	.set_page_dirty = __set_page_dirty_nobuffers,
4966	.releasepage = cifs_release_page,
4967	.direct_IO = cifs_direct_io,
4968	.invalidatepage = cifs_invalidate_page,
4969	.launder_page = cifs_launder_page,
4970	/*
4971	 * TODO: investigate and if useful we could add an cifs_migratePage
4972	 * helper (under an CONFIG_MIGRATION) in the future, and also
4973	 * investigate and add an is_dirty_writeback helper if needed
4974	 */
4975	.swap_activate = cifs_swap_activate,
4976	.swap_deactivate = cifs_swap_deactivate,
4977};
4978
4979/*
4980 * cifs_readpages requires the server to support a buffer large enough to
4981 * contain the header plus one complete page of data.  Otherwise, we need
4982 * to leave cifs_readpages out of the address space operations.
4983 */
4984const struct address_space_operations cifs_addr_ops_smallbuf = {
4985	.readpage = cifs_readpage,
4986	.writepage = cifs_writepage,
4987	.writepages = cifs_writepages,
4988	.write_begin = cifs_write_begin,
4989	.write_end = cifs_write_end,
4990	.set_page_dirty = __set_page_dirty_nobuffers,
4991	.releasepage = cifs_release_page,
4992	.invalidatepage = cifs_invalidate_page,
4993	.launder_page = cifs_launder_page,
4994};