Linux Audio

Check our new training course

Loading...
v4.17
   1/*
   2 *   fs/cifs/file.c
   3 *
   4 *   vfs operations that deal with files
   5 *
   6 *   Copyright (C) International Business Machines  Corp., 2002,2010
   7 *   Author(s): Steve French (sfrench@us.ibm.com)
   8 *              Jeremy Allison (jra@samba.org)
   9 *
  10 *   This library is free software; you can redistribute it and/or modify
  11 *   it under the terms of the GNU Lesser General Public License as published
  12 *   by the Free Software Foundation; either version 2.1 of the License, or
  13 *   (at your option) any later version.
  14 *
  15 *   This library is distributed in the hope that it will be useful,
  16 *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  17 *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
  18 *   the GNU Lesser General Public License for more details.
  19 *
  20 *   You should have received a copy of the GNU Lesser General Public License
  21 *   along with this library; if not, write to the Free Software
  22 *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  23 */
  24#include <linux/fs.h>
  25#include <linux/backing-dev.h>
  26#include <linux/stat.h>
  27#include <linux/fcntl.h>
  28#include <linux/pagemap.h>
  29#include <linux/pagevec.h>
  30#include <linux/writeback.h>
  31#include <linux/task_io_accounting_ops.h>
  32#include <linux/delay.h>
  33#include <linux/mount.h>
  34#include <linux/slab.h>
  35#include <linux/swap.h>
  36#include <asm/div64.h>
  37#include "cifsfs.h"
  38#include "cifspdu.h"
  39#include "cifsglob.h"
  40#include "cifsproto.h"
  41#include "cifs_unicode.h"
  42#include "cifs_debug.h"
  43#include "cifs_fs_sb.h"
  44#include "fscache.h"
  45#include "smbdirect.h"
  46
  47static inline int cifs_convert_flags(unsigned int flags)
  48{
  49	if ((flags & O_ACCMODE) == O_RDONLY)
  50		return GENERIC_READ;
  51	else if ((flags & O_ACCMODE) == O_WRONLY)
  52		return GENERIC_WRITE;
  53	else if ((flags & O_ACCMODE) == O_RDWR) {
  54		/* GENERIC_ALL is too much permission to request
  55		   can cause unnecessary access denied on create */
  56		/* return GENERIC_ALL; */
  57		return (GENERIC_READ | GENERIC_WRITE);
  58	}
  59
  60	return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
  61		FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
  62		FILE_READ_DATA);
  63}
  64
  65static u32 cifs_posix_convert_flags(unsigned int flags)
  66{
  67	u32 posix_flags = 0;
  68
  69	if ((flags & O_ACCMODE) == O_RDONLY)
  70		posix_flags = SMB_O_RDONLY;
  71	else if ((flags & O_ACCMODE) == O_WRONLY)
  72		posix_flags = SMB_O_WRONLY;
  73	else if ((flags & O_ACCMODE) == O_RDWR)
  74		posix_flags = SMB_O_RDWR;
  75
  76	if (flags & O_CREAT) {
  77		posix_flags |= SMB_O_CREAT;
  78		if (flags & O_EXCL)
  79			posix_flags |= SMB_O_EXCL;
  80	} else if (flags & O_EXCL)
  81		cifs_dbg(FYI, "Application %s pid %d has incorrectly set O_EXCL flag but not O_CREAT on file open. Ignoring O_EXCL\n",
  82			 current->comm, current->tgid);
  83
  84	if (flags & O_TRUNC)
  85		posix_flags |= SMB_O_TRUNC;
  86	/* be safe and imply O_SYNC for O_DSYNC */
  87	if (flags & O_DSYNC)
  88		posix_flags |= SMB_O_SYNC;
  89	if (flags & O_DIRECTORY)
  90		posix_flags |= SMB_O_DIRECTORY;
  91	if (flags & O_NOFOLLOW)
  92		posix_flags |= SMB_O_NOFOLLOW;
  93	if (flags & O_DIRECT)
  94		posix_flags |= SMB_O_DIRECT;
  95
  96	return posix_flags;
  97}
  98
  99static inline int cifs_get_disposition(unsigned int flags)
 100{
 101	if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
 102		return FILE_CREATE;
 103	else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
 104		return FILE_OVERWRITE_IF;
 105	else if ((flags & O_CREAT) == O_CREAT)
 106		return FILE_OPEN_IF;
 107	else if ((flags & O_TRUNC) == O_TRUNC)
 108		return FILE_OVERWRITE;
 109	else
 110		return FILE_OPEN;
 111}
 112
 113int cifs_posix_open(char *full_path, struct inode **pinode,
 114			struct super_block *sb, int mode, unsigned int f_flags,
 115			__u32 *poplock, __u16 *pnetfid, unsigned int xid)
 116{
 117	int rc;
 118	FILE_UNIX_BASIC_INFO *presp_data;
 119	__u32 posix_flags = 0;
 120	struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
 121	struct cifs_fattr fattr;
 122	struct tcon_link *tlink;
 123	struct cifs_tcon *tcon;
 124
 125	cifs_dbg(FYI, "posix open %s\n", full_path);
 126
 127	presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
 128	if (presp_data == NULL)
 129		return -ENOMEM;
 130
 131	tlink = cifs_sb_tlink(cifs_sb);
 132	if (IS_ERR(tlink)) {
 133		rc = PTR_ERR(tlink);
 134		goto posix_open_ret;
 135	}
 136
 137	tcon = tlink_tcon(tlink);
 138	mode &= ~current_umask();
 139
 140	posix_flags = cifs_posix_convert_flags(f_flags);
 141	rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
 142			     poplock, full_path, cifs_sb->local_nls,
 143			     cifs_remap(cifs_sb));
 
 144	cifs_put_tlink(tlink);
 145
 146	if (rc)
 147		goto posix_open_ret;
 148
 149	if (presp_data->Type == cpu_to_le32(-1))
 150		goto posix_open_ret; /* open ok, caller does qpathinfo */
 151
 152	if (!pinode)
 153		goto posix_open_ret; /* caller does not need info */
 154
 155	cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
 156
 157	/* get new inode and set it up */
 158	if (*pinode == NULL) {
 159		cifs_fill_uniqueid(sb, &fattr);
 160		*pinode = cifs_iget(sb, &fattr);
 161		if (!*pinode) {
 162			rc = -ENOMEM;
 163			goto posix_open_ret;
 164		}
 165	} else {
 166		cifs_fattr_to_inode(*pinode, &fattr);
 167	}
 168
 169posix_open_ret:
 170	kfree(presp_data);
 171	return rc;
 172}
 173
 174static int
 175cifs_nt_open(char *full_path, struct inode *inode, struct cifs_sb_info *cifs_sb,
 176	     struct cifs_tcon *tcon, unsigned int f_flags, __u32 *oplock,
 177	     struct cifs_fid *fid, unsigned int xid)
 178{
 179	int rc;
 180	int desired_access;
 181	int disposition;
 182	int create_options = CREATE_NOT_DIR;
 183	FILE_ALL_INFO *buf;
 184	struct TCP_Server_Info *server = tcon->ses->server;
 185	struct cifs_open_parms oparms;
 186
 187	if (!server->ops->open)
 188		return -ENOSYS;
 189
 190	desired_access = cifs_convert_flags(f_flags);
 191
 192/*********************************************************************
 193 *  open flag mapping table:
 194 *
 195 *	POSIX Flag            CIFS Disposition
 196 *	----------            ----------------
 197 *	O_CREAT               FILE_OPEN_IF
 198 *	O_CREAT | O_EXCL      FILE_CREATE
 199 *	O_CREAT | O_TRUNC     FILE_OVERWRITE_IF
 200 *	O_TRUNC               FILE_OVERWRITE
 201 *	none of the above     FILE_OPEN
 202 *
 203 *	Note that there is not a direct match between disposition
 204 *	FILE_SUPERSEDE (ie create whether or not file exists although
 205 *	O_CREAT | O_TRUNC is similar but truncates the existing
 206 *	file rather than creating a new file as FILE_SUPERSEDE does
 207 *	(which uses the attributes / metadata passed in on open call)
 208 *?
 209 *?  O_SYNC is a reasonable match to CIFS writethrough flag
 210 *?  and the read write flags match reasonably.  O_LARGEFILE
 211 *?  is irrelevant because largefile support is always used
 212 *?  by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
 213 *	 O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
 214 *********************************************************************/
 215
 216	disposition = cifs_get_disposition(f_flags);
 217
 218	/* BB pass O_SYNC flag through on file attributes .. BB */
 219
 220	buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
 221	if (!buf)
 222		return -ENOMEM;
 223
 224	if (backup_cred(cifs_sb))
 225		create_options |= CREATE_OPEN_BACKUP_INTENT;
 226
 227	/* O_SYNC also has bit for O_DSYNC so following check picks up either */
 228	if (f_flags & O_SYNC)
 229		create_options |= CREATE_WRITE_THROUGH;
 230
 231	if (f_flags & O_DIRECT)
 232		create_options |= CREATE_NO_BUFFER;
 233
 234	oparms.tcon = tcon;
 235	oparms.cifs_sb = cifs_sb;
 236	oparms.desired_access = desired_access;
 237	oparms.create_options = create_options;
 238	oparms.disposition = disposition;
 239	oparms.path = full_path;
 240	oparms.fid = fid;
 241	oparms.reconnect = false;
 242
 243	rc = server->ops->open(xid, &oparms, oplock, buf);
 244
 245	if (rc)
 246		goto out;
 247
 248	if (tcon->unix_ext)
 249		rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
 250					      xid);
 251	else
 252		rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
 253					 xid, fid);
 254
 255out:
 256	kfree(buf);
 257	return rc;
 258}
 259
 260static bool
 261cifs_has_mand_locks(struct cifsInodeInfo *cinode)
 262{
 263	struct cifs_fid_locks *cur;
 264	bool has_locks = false;
 265
 266	down_read(&cinode->lock_sem);
 267	list_for_each_entry(cur, &cinode->llist, llist) {
 268		if (!list_empty(&cur->locks)) {
 269			has_locks = true;
 270			break;
 271		}
 272	}
 273	up_read(&cinode->lock_sem);
 274	return has_locks;
 275}
 276
 277struct cifsFileInfo *
 278cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
 279		  struct tcon_link *tlink, __u32 oplock)
 280{
 281	struct dentry *dentry = file_dentry(file);
 282	struct inode *inode = d_inode(dentry);
 283	struct cifsInodeInfo *cinode = CIFS_I(inode);
 284	struct cifsFileInfo *cfile;
 285	struct cifs_fid_locks *fdlocks;
 286	struct cifs_tcon *tcon = tlink_tcon(tlink);
 287	struct TCP_Server_Info *server = tcon->ses->server;
 288
 289	cfile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
 290	if (cfile == NULL)
 291		return cfile;
 292
 293	fdlocks = kzalloc(sizeof(struct cifs_fid_locks), GFP_KERNEL);
 294	if (!fdlocks) {
 295		kfree(cfile);
 296		return NULL;
 297	}
 298
 299	INIT_LIST_HEAD(&fdlocks->locks);
 300	fdlocks->cfile = cfile;
 301	cfile->llist = fdlocks;
 302	down_write(&cinode->lock_sem);
 303	list_add(&fdlocks->llist, &cinode->llist);
 304	up_write(&cinode->lock_sem);
 305
 306	cfile->count = 1;
 307	cfile->pid = current->tgid;
 308	cfile->uid = current_fsuid();
 309	cfile->dentry = dget(dentry);
 310	cfile->f_flags = file->f_flags;
 311	cfile->invalidHandle = false;
 312	cfile->tlink = cifs_get_tlink(tlink);
 313	INIT_WORK(&cfile->oplock_break, cifs_oplock_break);
 314	mutex_init(&cfile->fh_mutex);
 315	spin_lock_init(&cfile->file_info_lock);
 316
 317	cifs_sb_active(inode->i_sb);
 318
 319	/*
 320	 * If the server returned a read oplock and we have mandatory brlocks,
 321	 * set oplock level to None.
 322	 */
 323	if (server->ops->is_read_op(oplock) && cifs_has_mand_locks(cinode)) {
 324		cifs_dbg(FYI, "Reset oplock val from read to None due to mand locks\n");
 325		oplock = 0;
 326	}
 327
 328	spin_lock(&tcon->open_file_lock);
 329	if (fid->pending_open->oplock != CIFS_OPLOCK_NO_CHANGE && oplock)
 330		oplock = fid->pending_open->oplock;
 331	list_del(&fid->pending_open->olist);
 332
 333	fid->purge_cache = false;
 334	server->ops->set_fid(cfile, fid, oplock);
 335
 336	list_add(&cfile->tlist, &tcon->openFileList);
 337
 
 
 338	/* if readable file instance put first in list*/
 339	if (file->f_mode & FMODE_READ)
 340		list_add(&cfile->flist, &cinode->openFileList);
 341	else
 342		list_add_tail(&cfile->flist, &cinode->openFileList);
 343	spin_unlock(&tcon->open_file_lock);
 344
 345	if (fid->purge_cache)
 346		cifs_zap_mapping(inode);
 347
 348	file->private_data = cfile;
 349	return cfile;
 350}
 351
 
 
 352struct cifsFileInfo *
 353cifsFileInfo_get(struct cifsFileInfo *cifs_file)
 354{
 355	spin_lock(&cifs_file->file_info_lock);
 356	cifsFileInfo_get_locked(cifs_file);
 357	spin_unlock(&cifs_file->file_info_lock);
 358	return cifs_file;
 359}
 360
 361/*
 362 * Release a reference on the file private data. This may involve closing
 363 * the filehandle out on the server. Must be called without holding
 364 * tcon->open_file_lock and cifs_file->file_info_lock.
 365 */
 366void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
 367{
 368	struct inode *inode = d_inode(cifs_file->dentry);
 369	struct cifs_tcon *tcon = tlink_tcon(cifs_file->tlink);
 370	struct TCP_Server_Info *server = tcon->ses->server;
 371	struct cifsInodeInfo *cifsi = CIFS_I(inode);
 372	struct super_block *sb = inode->i_sb;
 373	struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
 374	struct cifsLockInfo *li, *tmp;
 375	struct cifs_fid fid;
 376	struct cifs_pending_open open;
 377	bool oplock_break_cancelled;
 378
 379	spin_lock(&tcon->open_file_lock);
 380
 381	spin_lock(&cifs_file->file_info_lock);
 382	if (--cifs_file->count > 0) {
 383		spin_unlock(&cifs_file->file_info_lock);
 384		spin_unlock(&tcon->open_file_lock);
 385		return;
 386	}
 387	spin_unlock(&cifs_file->file_info_lock);
 388
 389	if (server->ops->get_lease_key)
 390		server->ops->get_lease_key(inode, &fid);
 391
 392	/* store open in pending opens to make sure we don't miss lease break */
 393	cifs_add_pending_open_locked(&fid, cifs_file->tlink, &open);
 394
 395	/* remove it from the lists */
 396	list_del(&cifs_file->flist);
 397	list_del(&cifs_file->tlist);
 398
 399	if (list_empty(&cifsi->openFileList)) {
 400		cifs_dbg(FYI, "closing last open instance for inode %p\n",
 401			 d_inode(cifs_file->dentry));
 402		/*
 403		 * In strict cache mode we need invalidate mapping on the last
 404		 * close  because it may cause a error when we open this file
 405		 * again and get at least level II oplock.
 406		 */
 407		if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
 408			set_bit(CIFS_INO_INVALID_MAPPING, &cifsi->flags);
 
 409		cifs_set_oplock_level(cifsi, 0);
 410	}
 
 411
 412	spin_unlock(&tcon->open_file_lock);
 413
 414	oplock_break_cancelled = cancel_work_sync(&cifs_file->oplock_break);
 415
 416	if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
 417		struct TCP_Server_Info *server = tcon->ses->server;
 418		unsigned int xid;
 419
 420		xid = get_xid();
 421		if (server->ops->close)
 422			server->ops->close(xid, tcon, &cifs_file->fid);
 423		_free_xid(xid);
 424	}
 425
 426	if (oplock_break_cancelled)
 427		cifs_done_oplock_break(cifsi);
 428
 429	cifs_del_pending_open(&open);
 430
 431	/*
 432	 * Delete any outstanding lock records. We'll lose them when the file
 433	 * is closed anyway.
 434	 */
 435	down_write(&cifsi->lock_sem);
 436	list_for_each_entry_safe(li, tmp, &cifs_file->llist->locks, llist) {
 437		list_del(&li->llist);
 438		cifs_del_lock_waiters(li);
 439		kfree(li);
 440	}
 441	list_del(&cifs_file->llist->llist);
 442	kfree(cifs_file->llist);
 443	up_write(&cifsi->lock_sem);
 444
 445	cifs_put_tlink(cifs_file->tlink);
 446	dput(cifs_file->dentry);
 447	cifs_sb_deactive(sb);
 448	kfree(cifs_file);
 449}
 450
 451int cifs_open(struct inode *inode, struct file *file)
 452
 453{
 454	int rc = -EACCES;
 455	unsigned int xid;
 456	__u32 oplock;
 457	struct cifs_sb_info *cifs_sb;
 458	struct TCP_Server_Info *server;
 459	struct cifs_tcon *tcon;
 460	struct tcon_link *tlink;
 461	struct cifsFileInfo *cfile = NULL;
 462	char *full_path = NULL;
 463	bool posix_open_ok = false;
 464	struct cifs_fid fid;
 465	struct cifs_pending_open open;
 466
 467	xid = get_xid();
 468
 469	cifs_sb = CIFS_SB(inode->i_sb);
 470	tlink = cifs_sb_tlink(cifs_sb);
 471	if (IS_ERR(tlink)) {
 472		free_xid(xid);
 473		return PTR_ERR(tlink);
 474	}
 475	tcon = tlink_tcon(tlink);
 476	server = tcon->ses->server;
 477
 478	full_path = build_path_from_dentry(file_dentry(file));
 479	if (full_path == NULL) {
 480		rc = -ENOMEM;
 481		goto out;
 482	}
 483
 484	cifs_dbg(FYI, "inode = 0x%p file flags are 0x%x for %s\n",
 485		 inode, file->f_flags, full_path);
 486
 487	if (file->f_flags & O_DIRECT &&
 488	    cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO) {
 489		if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_BRL)
 490			file->f_op = &cifs_file_direct_nobrl_ops;
 491		else
 492			file->f_op = &cifs_file_direct_ops;
 493	}
 494
 495	if (server->oplocks)
 496		oplock = REQ_OPLOCK;
 497	else
 498		oplock = 0;
 499
 500	if (!tcon->broken_posix_open && tcon->unix_ext &&
 501	    cap_unix(tcon->ses) && (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 502				le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 
 503		/* can not refresh inode info since size could be stale */
 504		rc = cifs_posix_open(full_path, &inode, inode->i_sb,
 505				cifs_sb->mnt_file_mode /* ignored */,
 506				file->f_flags, &oplock, &fid.netfid, xid);
 507		if (rc == 0) {
 508			cifs_dbg(FYI, "posix open succeeded\n");
 509			posix_open_ok = true;
 510		} else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
 511			if (tcon->ses->serverNOS)
 512				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",
 513					 tcon->ses->serverName,
 514					 tcon->ses->serverNOS);
 
 
 
 515			tcon->broken_posix_open = true;
 516		} else if ((rc != -EIO) && (rc != -EREMOTE) &&
 517			 (rc != -EOPNOTSUPP)) /* path not found or net err */
 518			goto out;
 519		/*
 520		 * Else fallthrough to retry open the old way on network i/o
 521		 * or DFS errors.
 522		 */
 523	}
 524
 525	if (server->ops->get_lease_key)
 526		server->ops->get_lease_key(inode, &fid);
 527
 528	cifs_add_pending_open(&fid, tlink, &open);
 529
 530	if (!posix_open_ok) {
 531		if (server->ops->get_lease_key)
 532			server->ops->get_lease_key(inode, &fid);
 533
 534		rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
 535				  file->f_flags, &oplock, &fid, xid);
 536		if (rc) {
 537			cifs_del_pending_open(&open);
 538			goto out;
 539		}
 540	}
 541
 542	cfile = cifs_new_fileinfo(&fid, file, tlink, oplock);
 543	if (cfile == NULL) {
 544		if (server->ops->close)
 545			server->ops->close(xid, tcon, &fid);
 546		cifs_del_pending_open(&open);
 547		rc = -ENOMEM;
 548		goto out;
 549	}
 550
 551	cifs_fscache_set_inode_cookie(inode, file);
 552
 553	if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
 554		/*
 555		 * Time to set mode which we can not set earlier due to
 556		 * problems creating new read-only files.
 557		 */
 558		struct cifs_unix_set_info_args args = {
 559			.mode	= inode->i_mode,
 560			.uid	= INVALID_UID, /* no change */
 561			.gid	= INVALID_GID, /* no change */
 562			.ctime	= NO_CHANGE_64,
 563			.atime	= NO_CHANGE_64,
 564			.mtime	= NO_CHANGE_64,
 565			.device	= 0,
 566		};
 567		CIFSSMBUnixSetFileInfo(xid, tcon, &args, fid.netfid,
 568				       cfile->pid);
 569	}
 570
 571out:
 572	kfree(full_path);
 573	free_xid(xid);
 574	cifs_put_tlink(tlink);
 575	return rc;
 576}
 577
 578static int cifs_push_posix_locks(struct cifsFileInfo *cfile);
 579
 580/*
 581 * Try to reacquire byte range locks that were released when session
 582 * to server was lost.
 583 */
 584static int
 585cifs_relock_file(struct cifsFileInfo *cfile)
 586{
 587	struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
 588	struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 589	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
 590	int rc = 0;
 591
 592	down_read_nested(&cinode->lock_sem, SINGLE_DEPTH_NESTING);
 593	if (cinode->can_cache_brlcks) {
 594		/* can cache locks - no need to relock */
 595		up_read(&cinode->lock_sem);
 596		return rc;
 597	}
 598
 599	if (cap_unix(tcon->ses) &&
 600	    (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
 601	    ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
 602		rc = cifs_push_posix_locks(cfile);
 603	else
 604		rc = tcon->ses->server->ops->push_mand_locks(cfile);
 605
 606	up_read(&cinode->lock_sem);
 607	return rc;
 608}
 609
 610static int
 611cifs_reopen_file(struct cifsFileInfo *cfile, bool can_flush)
 612{
 613	int rc = -EACCES;
 614	unsigned int xid;
 615	__u32 oplock;
 616	struct cifs_sb_info *cifs_sb;
 617	struct cifs_tcon *tcon;
 618	struct TCP_Server_Info *server;
 619	struct cifsInodeInfo *cinode;
 620	struct inode *inode;
 621	char *full_path = NULL;
 622	int desired_access;
 623	int disposition = FILE_OPEN;
 624	int create_options = CREATE_NOT_DIR;
 625	struct cifs_open_parms oparms;
 626
 627	xid = get_xid();
 628	mutex_lock(&cfile->fh_mutex);
 629	if (!cfile->invalidHandle) {
 630		mutex_unlock(&cfile->fh_mutex);
 631		rc = 0;
 632		free_xid(xid);
 633		return rc;
 634	}
 635
 636	inode = d_inode(cfile->dentry);
 637	cifs_sb = CIFS_SB(inode->i_sb);
 638	tcon = tlink_tcon(cfile->tlink);
 639	server = tcon->ses->server;
 640
 641	/*
 642	 * Can not grab rename sem here because various ops, including those
 643	 * that already have the rename sem can end up causing writepage to get
 644	 * called and if the server was down that means we end up here, and we
 645	 * can never tell if the caller already has the rename_sem.
 646	 */
 647	full_path = build_path_from_dentry(cfile->dentry);
 648	if (full_path == NULL) {
 649		rc = -ENOMEM;
 650		mutex_unlock(&cfile->fh_mutex);
 651		free_xid(xid);
 652		return rc;
 653	}
 654
 655	cifs_dbg(FYI, "inode = 0x%p file flags 0x%x for %s\n",
 656		 inode, cfile->f_flags, full_path);
 657
 658	if (tcon->ses->server->oplocks)
 659		oplock = REQ_OPLOCK;
 660	else
 661		oplock = 0;
 662
 663	if (tcon->unix_ext && cap_unix(tcon->ses) &&
 664	    (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 665				le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 
 666		/*
 667		 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
 668		 * original open. Must mask them off for a reopen.
 669		 */
 670		unsigned int oflags = cfile->f_flags &
 671						~(O_CREAT | O_EXCL | O_TRUNC);
 672
 673		rc = cifs_posix_open(full_path, NULL, inode->i_sb,
 674				     cifs_sb->mnt_file_mode /* ignored */,
 675				     oflags, &oplock, &cfile->fid.netfid, xid);
 676		if (rc == 0) {
 677			cifs_dbg(FYI, "posix reopen succeeded\n");
 678			oparms.reconnect = true;
 679			goto reopen_success;
 680		}
 681		/*
 682		 * fallthrough to retry open the old way on errors, especially
 683		 * in the reconnect path it is important to retry hard
 684		 */
 685	}
 686
 687	desired_access = cifs_convert_flags(cfile->f_flags);
 688
 689	if (backup_cred(cifs_sb))
 690		create_options |= CREATE_OPEN_BACKUP_INTENT;
 691
 692	if (server->ops->get_lease_key)
 693		server->ops->get_lease_key(inode, &cfile->fid);
 694
 695	oparms.tcon = tcon;
 696	oparms.cifs_sb = cifs_sb;
 697	oparms.desired_access = desired_access;
 698	oparms.create_options = create_options;
 699	oparms.disposition = disposition;
 700	oparms.path = full_path;
 701	oparms.fid = &cfile->fid;
 702	oparms.reconnect = true;
 703
 704	/*
 705	 * Can not refresh inode by passing in file_info buf to be returned by
 706	 * ops->open and then calling get_inode_info with returned buf since
 707	 * file might have write behind data that needs to be flushed and server
 708	 * version of file size can be stale. If we knew for sure that inode was
 709	 * not dirty locally we could do this.
 710	 */
 711	rc = server->ops->open(xid, &oparms, &oplock, NULL);
 712	if (rc == -ENOENT && oparms.reconnect == false) {
 713		/* durable handle timeout is expired - open the file again */
 714		rc = server->ops->open(xid, &oparms, &oplock, NULL);
 715		/* indicate that we need to relock the file */
 716		oparms.reconnect = true;
 717	}
 718
 719	if (rc) {
 720		mutex_unlock(&cfile->fh_mutex);
 721		cifs_dbg(FYI, "cifs_reopen returned 0x%x\n", rc);
 722		cifs_dbg(FYI, "oplock: %d\n", oplock);
 723		goto reopen_error_exit;
 724	}
 725
 726reopen_success:
 727	cfile->invalidHandle = false;
 728	mutex_unlock(&cfile->fh_mutex);
 729	cinode = CIFS_I(inode);
 
 730
 731	if (can_flush) {
 732		rc = filemap_write_and_wait(inode->i_mapping);
 733		mapping_set_error(inode->i_mapping, rc);
 734
 735		if (tcon->unix_ext)
 736			rc = cifs_get_inode_info_unix(&inode, full_path,
 737						      inode->i_sb, xid);
 738		else
 739			rc = cifs_get_inode_info(&inode, full_path, NULL,
 740						 inode->i_sb, xid, NULL);
 741	}
 742	/*
 743	 * Else we are writing out data to server already and could deadlock if
 744	 * we tried to flush data, and since we do not know if we have data that
 745	 * would invalidate the current end of file on the server we can not go
 746	 * to the server to get the new inode info.
 747	 */
 748
 749	/*
 750	 * If the server returned a read oplock and we have mandatory brlocks,
 751	 * set oplock level to None.
 752	 */
 753	if (server->ops->is_read_op(oplock) && cifs_has_mand_locks(cinode)) {
 754		cifs_dbg(FYI, "Reset oplock val from read to None due to mand locks\n");
 755		oplock = 0;
 756	}
 757
 758	server->ops->set_fid(cfile, &cfile->fid, oplock);
 759	if (oparms.reconnect)
 760		cifs_relock_file(cfile);
 761
 762reopen_error_exit:
 763	kfree(full_path);
 764	free_xid(xid);
 765	return rc;
 766}
 767
 768int cifs_close(struct inode *inode, struct file *file)
 769{
 770	if (file->private_data != NULL) {
 771		cifsFileInfo_put(file->private_data);
 772		file->private_data = NULL;
 773	}
 774
 775	/* return code from the ->release op is always ignored */
 776	return 0;
 777}
 778
 779void
 780cifs_reopen_persistent_handles(struct cifs_tcon *tcon)
 781{
 782	struct cifsFileInfo *open_file;
 783	struct list_head *tmp;
 784	struct list_head *tmp1;
 785	struct list_head tmp_list;
 786
 787	if (!tcon->use_persistent || !tcon->need_reopen_files)
 788		return;
 789
 790	tcon->need_reopen_files = false;
 791
 792	cifs_dbg(FYI, "Reopen persistent handles");
 793	INIT_LIST_HEAD(&tmp_list);
 794
 795	/* list all files open on tree connection, reopen resilient handles  */
 796	spin_lock(&tcon->open_file_lock);
 797	list_for_each(tmp, &tcon->openFileList) {
 798		open_file = list_entry(tmp, struct cifsFileInfo, tlist);
 799		if (!open_file->invalidHandle)
 800			continue;
 801		cifsFileInfo_get(open_file);
 802		list_add_tail(&open_file->rlist, &tmp_list);
 803	}
 804	spin_unlock(&tcon->open_file_lock);
 805
 806	list_for_each_safe(tmp, tmp1, &tmp_list) {
 807		open_file = list_entry(tmp, struct cifsFileInfo, rlist);
 808		if (cifs_reopen_file(open_file, false /* do not flush */))
 809			tcon->need_reopen_files = true;
 810		list_del_init(&open_file->rlist);
 811		cifsFileInfo_put(open_file);
 812	}
 813}
 814
 815int cifs_closedir(struct inode *inode, struct file *file)
 816{
 817	int rc = 0;
 818	unsigned int xid;
 819	struct cifsFileInfo *cfile = file->private_data;
 820	struct cifs_tcon *tcon;
 821	struct TCP_Server_Info *server;
 822	char *buf;
 823
 824	cifs_dbg(FYI, "Closedir inode = 0x%p\n", inode);
 825
 826	if (cfile == NULL)
 827		return rc;
 828
 829	xid = get_xid();
 830	tcon = tlink_tcon(cfile->tlink);
 831	server = tcon->ses->server;
 832
 833	cifs_dbg(FYI, "Freeing private data in close dir\n");
 834	spin_lock(&cfile->file_info_lock);
 835	if (server->ops->dir_needs_close(cfile)) {
 836		cfile->invalidHandle = true;
 837		spin_unlock(&cfile->file_info_lock);
 838		if (server->ops->close_dir)
 839			rc = server->ops->close_dir(xid, tcon, &cfile->fid);
 840		else
 841			rc = -ENOSYS;
 842		cifs_dbg(FYI, "Closing uncompleted readdir with rc %d\n", rc);
 843		/* not much we can do if it fails anyway, ignore rc */
 844		rc = 0;
 845	} else
 846		spin_unlock(&cfile->file_info_lock);
 847
 848	buf = cfile->srch_inf.ntwrk_buf_start;
 849	if (buf) {
 850		cifs_dbg(FYI, "closedir free smb buf in srch struct\n");
 851		cfile->srch_inf.ntwrk_buf_start = NULL;
 852		if (cfile->srch_inf.smallBuf)
 853			cifs_small_buf_release(buf);
 854		else
 855			cifs_buf_release(buf);
 856	}
 857
 858	cifs_put_tlink(cfile->tlink);
 859	kfree(file->private_data);
 860	file->private_data = NULL;
 861	/* BB can we lock the filestruct while this is going on? */
 862	free_xid(xid);
 863	return rc;
 864}
 865
 866static struct cifsLockInfo *
 867cifs_lock_init(__u64 offset, __u64 length, __u8 type)
 868{
 869	struct cifsLockInfo *lock =
 870		kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
 871	if (!lock)
 872		return lock;
 873	lock->offset = offset;
 874	lock->length = length;
 875	lock->type = type;
 876	lock->pid = current->tgid;
 877	INIT_LIST_HEAD(&lock->blist);
 878	init_waitqueue_head(&lock->block_q);
 879	return lock;
 880}
 881
 882void
 883cifs_del_lock_waiters(struct cifsLockInfo *lock)
 884{
 885	struct cifsLockInfo *li, *tmp;
 886	list_for_each_entry_safe(li, tmp, &lock->blist, blist) {
 887		list_del_init(&li->blist);
 888		wake_up(&li->block_q);
 889	}
 890}
 891
 892#define CIFS_LOCK_OP	0
 893#define CIFS_READ_OP	1
 894#define CIFS_WRITE_OP	2
 895
 896/* @rw_check : 0 - no op, 1 - read, 2 - write */
 897static bool
 898cifs_find_fid_lock_conflict(struct cifs_fid_locks *fdlocks, __u64 offset,
 899			    __u64 length, __u8 type, struct cifsFileInfo *cfile,
 900			    struct cifsLockInfo **conf_lock, int rw_check)
 901{
 902	struct cifsLockInfo *li;
 903	struct cifsFileInfo *cur_cfile = fdlocks->cfile;
 904	struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
 905
 906	list_for_each_entry(li, &fdlocks->locks, llist) {
 907		if (offset + length <= li->offset ||
 908		    offset >= li->offset + li->length)
 909			continue;
 910		if (rw_check != CIFS_LOCK_OP && current->tgid == li->pid &&
 911		    server->ops->compare_fids(cfile, cur_cfile)) {
 912			/* shared lock prevents write op through the same fid */
 913			if (!(li->type & server->vals->shared_lock_type) ||
 914			    rw_check != CIFS_WRITE_OP)
 915				continue;
 916		}
 917		if ((type & server->vals->shared_lock_type) &&
 918		    ((server->ops->compare_fids(cfile, cur_cfile) &&
 919		     current->tgid == li->pid) || type == li->type))
 920			continue;
 921		if (conf_lock)
 922			*conf_lock = li;
 923		return true;
 
 924	}
 925	return false;
 926}
 927
 928bool
 929cifs_find_lock_conflict(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
 930			__u8 type, struct cifsLockInfo **conf_lock,
 931			int rw_check)
 932{
 933	bool rc = false;
 934	struct cifs_fid_locks *cur;
 935	struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 936
 937	list_for_each_entry(cur, &cinode->llist, llist) {
 938		rc = cifs_find_fid_lock_conflict(cur, offset, length, type,
 939						 cfile, conf_lock, rw_check);
 
 940		if (rc)
 941			break;
 942	}
 
 943
 944	return rc;
 945}
 946
 947/*
 948 * Check if there is another lock that prevents us to set the lock (mandatory
 949 * style). If such a lock exists, update the flock structure with its
 950 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
 951 * or leave it the same if we can't. Returns 0 if we don't need to request to
 952 * the server or 1 otherwise.
 953 */
 954static int
 955cifs_lock_test(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
 956	       __u8 type, struct file_lock *flock)
 957{
 958	int rc = 0;
 959	struct cifsLockInfo *conf_lock;
 960	struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 961	struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
 962	bool exist;
 963
 964	down_read(&cinode->lock_sem);
 965
 966	exist = cifs_find_lock_conflict(cfile, offset, length, type,
 967					&conf_lock, CIFS_LOCK_OP);
 968	if (exist) {
 969		flock->fl_start = conf_lock->offset;
 970		flock->fl_end = conf_lock->offset + conf_lock->length - 1;
 971		flock->fl_pid = conf_lock->pid;
 972		if (conf_lock->type & server->vals->shared_lock_type)
 973			flock->fl_type = F_RDLCK;
 974		else
 975			flock->fl_type = F_WRLCK;
 976	} else if (!cinode->can_cache_brlcks)
 977		rc = 1;
 978	else
 979		flock->fl_type = F_UNLCK;
 980
 981	up_read(&cinode->lock_sem);
 982	return rc;
 983}
 984
 985static void
 986cifs_lock_add(struct cifsFileInfo *cfile, struct cifsLockInfo *lock)
 987{
 988	struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 989	down_write(&cinode->lock_sem);
 990	list_add_tail(&lock->llist, &cfile->llist->locks);
 991	up_write(&cinode->lock_sem);
 992}
 993
 994/*
 995 * Set the byte-range lock (mandatory style). Returns:
 996 * 1) 0, if we set the lock and don't need to request to the server;
 997 * 2) 1, if no locks prevent us but we need to request to the server;
 998 * 3) -EACCESS, if there is a lock that prevents us and wait is false.
 999 */
1000static int
1001cifs_lock_add_if(struct cifsFileInfo *cfile, struct cifsLockInfo *lock,
1002		 bool wait)
1003{
1004	struct cifsLockInfo *conf_lock;
1005	struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1006	bool exist;
1007	int rc = 0;
1008
1009try_again:
1010	exist = false;
1011	down_write(&cinode->lock_sem);
1012
1013	exist = cifs_find_lock_conflict(cfile, lock->offset, lock->length,
1014					lock->type, &conf_lock, CIFS_LOCK_OP);
1015	if (!exist && cinode->can_cache_brlcks) {
1016		list_add_tail(&lock->llist, &cfile->llist->locks);
1017		up_write(&cinode->lock_sem);
1018		return rc;
1019	}
1020
1021	if (!exist)
1022		rc = 1;
1023	else if (!wait)
1024		rc = -EACCES;
1025	else {
1026		list_add_tail(&lock->blist, &conf_lock->blist);
1027		up_write(&cinode->lock_sem);
1028		rc = wait_event_interruptible(lock->block_q,
1029					(lock->blist.prev == &lock->blist) &&
1030					(lock->blist.next == &lock->blist));
1031		if (!rc)
1032			goto try_again;
1033		down_write(&cinode->lock_sem);
1034		list_del_init(&lock->blist);
1035	}
1036
1037	up_write(&cinode->lock_sem);
1038	return rc;
1039}
1040
1041/*
1042 * Check if there is another lock that prevents us to set the lock (posix
1043 * style). If such a lock exists, update the flock structure with its
1044 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
1045 * or leave it the same if we can't. Returns 0 if we don't need to request to
1046 * the server or 1 otherwise.
1047 */
1048static int
1049cifs_posix_lock_test(struct file *file, struct file_lock *flock)
1050{
1051	int rc = 0;
1052	struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
1053	unsigned char saved_type = flock->fl_type;
1054
1055	if ((flock->fl_flags & FL_POSIX) == 0)
1056		return 1;
1057
1058	down_read(&cinode->lock_sem);
1059	posix_test_lock(file, flock);
1060
1061	if (flock->fl_type == F_UNLCK && !cinode->can_cache_brlcks) {
1062		flock->fl_type = saved_type;
1063		rc = 1;
1064	}
1065
1066	up_read(&cinode->lock_sem);
1067	return rc;
1068}
1069
1070/*
1071 * Set the byte-range lock (posix style). Returns:
1072 * 1) 0, if we set the lock and don't need to request to the server;
1073 * 2) 1, if we need to request to the server;
1074 * 3) <0, if the error occurs while setting the lock.
1075 */
1076static int
1077cifs_posix_lock_set(struct file *file, struct file_lock *flock)
1078{
1079	struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
1080	int rc = 1;
1081
1082	if ((flock->fl_flags & FL_POSIX) == 0)
1083		return rc;
1084
1085try_again:
1086	down_write(&cinode->lock_sem);
1087	if (!cinode->can_cache_brlcks) {
1088		up_write(&cinode->lock_sem);
1089		return rc;
1090	}
1091
1092	rc = posix_lock_file(file, flock, NULL);
1093	up_write(&cinode->lock_sem);
1094	if (rc == FILE_LOCK_DEFERRED) {
1095		rc = wait_event_interruptible(flock->fl_wait, !flock->fl_next);
1096		if (!rc)
1097			goto try_again;
1098		posix_unblock_lock(flock);
1099	}
1100	return rc;
1101}
1102
1103int
1104cifs_push_mandatory_locks(struct cifsFileInfo *cfile)
1105{
1106	unsigned int xid;
1107	int rc = 0, stored_rc;
1108	struct cifsLockInfo *li, *tmp;
1109	struct cifs_tcon *tcon;
 
1110	unsigned int num, max_num, max_buf;
1111	LOCKING_ANDX_RANGE *buf, *cur;
1112	static const int types[] = {
1113		LOCKING_ANDX_LARGE_FILES,
1114		LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES
1115	};
1116	int i;
1117
1118	xid = get_xid();
1119	tcon = tlink_tcon(cfile->tlink);
1120
 
 
 
 
 
 
 
1121	/*
1122	 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1123	 * and check it for zero before using.
1124	 */
1125	max_buf = tcon->ses->server->maxBuf;
1126	if (!max_buf) {
1127		free_xid(xid);
 
1128		return -EINVAL;
1129	}
1130
1131	max_num = (max_buf - sizeof(struct smb_hdr)) /
1132						sizeof(LOCKING_ANDX_RANGE);
1133	buf = kcalloc(max_num, sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1134	if (!buf) {
1135		free_xid(xid);
1136		return -ENOMEM;
 
1137	}
1138
1139	for (i = 0; i < 2; i++) {
1140		cur = buf;
1141		num = 0;
1142		list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1143			if (li->type != types[i])
1144				continue;
1145			cur->Pid = cpu_to_le16(li->pid);
1146			cur->LengthLow = cpu_to_le32((u32)li->length);
1147			cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1148			cur->OffsetLow = cpu_to_le32((u32)li->offset);
1149			cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1150			if (++num == max_num) {
1151				stored_rc = cifs_lockv(xid, tcon,
1152						       cfile->fid.netfid,
1153						       (__u8)li->type, 0, num,
1154						       buf);
1155				if (stored_rc)
1156					rc = stored_rc;
1157				cur = buf;
1158				num = 0;
1159			} else
1160				cur++;
1161		}
1162
1163		if (num) {
1164			stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1165					       (__u8)types[i], 0, num, buf);
1166			if (stored_rc)
1167				rc = stored_rc;
1168		}
1169	}
1170
 
 
 
1171	kfree(buf);
1172	free_xid(xid);
1173	return rc;
1174}
1175
1176static __u32
1177hash_lockowner(fl_owner_t owner)
1178{
1179	return cifs_lock_secret ^ hash32_ptr((const void *)owner);
1180}
1181
1182struct lock_to_push {
1183	struct list_head llist;
1184	__u64 offset;
1185	__u64 length;
1186	__u32 pid;
1187	__u16 netfid;
1188	__u8 type;
1189};
1190
1191static int
1192cifs_push_posix_locks(struct cifsFileInfo *cfile)
1193{
1194	struct inode *inode = d_inode(cfile->dentry);
1195	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1196	struct file_lock *flock;
1197	struct file_lock_context *flctx = inode->i_flctx;
1198	unsigned int count = 0, i;
1199	int rc = 0, xid, type;
1200	struct list_head locks_to_send, *el;
1201	struct lock_to_push *lck, *tmp;
1202	__u64 length;
1203
1204	xid = get_xid();
1205
1206	if (!flctx)
1207		goto out;
 
 
 
 
1208
1209	spin_lock(&flctx->flc_lock);
1210	list_for_each(el, &flctx->flc_posix) {
1211		count++;
 
1212	}
1213	spin_unlock(&flctx->flc_lock);
1214
1215	INIT_LIST_HEAD(&locks_to_send);
1216
1217	/*
1218	 * Allocating count locks is enough because no FL_POSIX locks can be
1219	 * added to the list while we are holding cinode->lock_sem that
1220	 * protects locking operations of this inode.
1221	 */
1222	for (i = 0; i < count; i++) {
1223		lck = kmalloc(sizeof(struct lock_to_push), GFP_KERNEL);
1224		if (!lck) {
1225			rc = -ENOMEM;
1226			goto err_out;
1227		}
1228		list_add_tail(&lck->llist, &locks_to_send);
1229	}
1230
1231	el = locks_to_send.next;
1232	spin_lock(&flctx->flc_lock);
1233	list_for_each_entry(flock, &flctx->flc_posix, fl_list) {
 
 
 
1234		if (el == &locks_to_send) {
1235			/*
1236			 * The list ended. We don't have enough allocated
1237			 * structures - something is really wrong.
1238			 */
1239			cifs_dbg(VFS, "Can't push all brlocks!\n");
1240			break;
1241		}
1242		length = 1 + flock->fl_end - flock->fl_start;
1243		if (flock->fl_type == F_RDLCK || flock->fl_type == F_SHLCK)
1244			type = CIFS_RDLCK;
1245		else
1246			type = CIFS_WRLCK;
1247		lck = list_entry(el, struct lock_to_push, llist);
1248		lck->pid = hash_lockowner(flock->fl_owner);
1249		lck->netfid = cfile->fid.netfid;
1250		lck->length = length;
1251		lck->type = type;
1252		lck->offset = flock->fl_start;
 
1253	}
1254	spin_unlock(&flctx->flc_lock);
1255
1256	list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
 
1257		int stored_rc;
1258
 
1259		stored_rc = CIFSSMBPosixLock(xid, tcon, lck->netfid, lck->pid,
1260					     lck->offset, lck->length, NULL,
1261					     lck->type, 0);
1262		if (stored_rc)
1263			rc = stored_rc;
1264		list_del(&lck->llist);
1265		kfree(lck);
1266	}
1267
1268out:
1269	free_xid(xid);
 
 
 
1270	return rc;
1271err_out:
1272	list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1273		list_del(&lck->llist);
1274		kfree(lck);
1275	}
1276	goto out;
1277}
1278
1279static int
1280cifs_push_locks(struct cifsFileInfo *cfile)
1281{
1282	struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
1283	struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1284	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1285	int rc = 0;
1286
1287	/* we are going to update can_cache_brlcks here - need a write access */
1288	down_write(&cinode->lock_sem);
1289	if (!cinode->can_cache_brlcks) {
1290		up_write(&cinode->lock_sem);
1291		return rc;
1292	}
1293
1294	if (cap_unix(tcon->ses) &&
1295	    (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1296	    ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1297		rc = cifs_push_posix_locks(cfile);
1298	else
1299		rc = tcon->ses->server->ops->push_mand_locks(cfile);
1300
1301	cinode->can_cache_brlcks = false;
1302	up_write(&cinode->lock_sem);
1303	return rc;
1304}
1305
1306static void
1307cifs_read_flock(struct file_lock *flock, __u32 *type, int *lock, int *unlock,
1308		bool *wait_flag, struct TCP_Server_Info *server)
1309{
1310	if (flock->fl_flags & FL_POSIX)
1311		cifs_dbg(FYI, "Posix\n");
1312	if (flock->fl_flags & FL_FLOCK)
1313		cifs_dbg(FYI, "Flock\n");
1314	if (flock->fl_flags & FL_SLEEP) {
1315		cifs_dbg(FYI, "Blocking lock\n");
1316		*wait_flag = true;
1317	}
1318	if (flock->fl_flags & FL_ACCESS)
1319		cifs_dbg(FYI, "Process suspended by mandatory locking - not implemented yet\n");
 
1320	if (flock->fl_flags & FL_LEASE)
1321		cifs_dbg(FYI, "Lease on file - not implemented yet\n");
1322	if (flock->fl_flags &
1323	    (~(FL_POSIX | FL_FLOCK | FL_SLEEP |
1324	       FL_ACCESS | FL_LEASE | FL_CLOSE)))
1325		cifs_dbg(FYI, "Unknown lock flags 0x%x\n", flock->fl_flags);
1326
1327	*type = server->vals->large_lock_type;
1328	if (flock->fl_type == F_WRLCK) {
1329		cifs_dbg(FYI, "F_WRLCK\n");
1330		*type |= server->vals->exclusive_lock_type;
1331		*lock = 1;
1332	} else if (flock->fl_type == F_UNLCK) {
1333		cifs_dbg(FYI, "F_UNLCK\n");
1334		*type |= server->vals->unlock_lock_type;
1335		*unlock = 1;
1336		/* Check if unlock includes more than one lock range */
1337	} else if (flock->fl_type == F_RDLCK) {
1338		cifs_dbg(FYI, "F_RDLCK\n");
1339		*type |= server->vals->shared_lock_type;
1340		*lock = 1;
1341	} else if (flock->fl_type == F_EXLCK) {
1342		cifs_dbg(FYI, "F_EXLCK\n");
1343		*type |= server->vals->exclusive_lock_type;
1344		*lock = 1;
1345	} else if (flock->fl_type == F_SHLCK) {
1346		cifs_dbg(FYI, "F_SHLCK\n");
1347		*type |= server->vals->shared_lock_type;
1348		*lock = 1;
1349	} else
1350		cifs_dbg(FYI, "Unknown type of lock\n");
 
 
 
 
 
 
 
 
 
1351}
1352
1353static int
1354cifs_getlk(struct file *file, struct file_lock *flock, __u32 type,
1355	   bool wait_flag, bool posix_lck, unsigned int xid)
1356{
1357	int rc = 0;
1358	__u64 length = 1 + flock->fl_end - flock->fl_start;
1359	struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1360	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1361	struct TCP_Server_Info *server = tcon->ses->server;
1362	__u16 netfid = cfile->fid.netfid;
1363
1364	if (posix_lck) {
1365		int posix_lock_type;
1366
1367		rc = cifs_posix_lock_test(file, flock);
1368		if (!rc)
1369			return rc;
1370
1371		if (type & server->vals->shared_lock_type)
1372			posix_lock_type = CIFS_RDLCK;
1373		else
1374			posix_lock_type = CIFS_WRLCK;
1375		rc = CIFSSMBPosixLock(xid, tcon, netfid,
1376				      hash_lockowner(flock->fl_owner),
1377				      flock->fl_start, length, flock,
1378				      posix_lock_type, wait_flag);
1379		return rc;
1380	}
1381
1382	rc = cifs_lock_test(cfile, flock->fl_start, length, type, flock);
1383	if (!rc)
1384		return rc;
1385
1386	/* BB we could chain these into one lock request BB */
1387	rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length, type,
1388				    1, 0, false);
1389	if (rc == 0) {
1390		rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1391					    type, 0, 1, false);
1392		flock->fl_type = F_UNLCK;
1393		if (rc != 0)
1394			cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1395				 rc);
1396		return 0;
1397	}
1398
1399	if (type & server->vals->shared_lock_type) {
1400		flock->fl_type = F_WRLCK;
1401		return 0;
1402	}
1403
1404	type &= ~server->vals->exclusive_lock_type;
1405
1406	rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1407				    type | server->vals->shared_lock_type,
1408				    1, 0, false);
1409	if (rc == 0) {
1410		rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1411			type | server->vals->shared_lock_type, 0, 1, false);
 
1412		flock->fl_type = F_RDLCK;
1413		if (rc != 0)
1414			cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1415				 rc);
1416	} else
1417		flock->fl_type = F_WRLCK;
1418
1419	return 0;
1420}
1421
1422void
1423cifs_move_llist(struct list_head *source, struct list_head *dest)
1424{
1425	struct list_head *li, *tmp;
1426	list_for_each_safe(li, tmp, source)
1427		list_move(li, dest);
1428}
1429
1430void
1431cifs_free_llist(struct list_head *llist)
1432{
1433	struct cifsLockInfo *li, *tmp;
1434	list_for_each_entry_safe(li, tmp, llist, llist) {
1435		cifs_del_lock_waiters(li);
1436		list_del(&li->llist);
1437		kfree(li);
1438	}
1439}
1440
1441int
1442cifs_unlock_range(struct cifsFileInfo *cfile, struct file_lock *flock,
1443		  unsigned int xid)
1444{
1445	int rc = 0, stored_rc;
1446	static const int types[] = {
1447		LOCKING_ANDX_LARGE_FILES,
1448		LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES
1449	};
1450	unsigned int i;
1451	unsigned int max_num, num, max_buf;
1452	LOCKING_ANDX_RANGE *buf, *cur;
1453	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1454	struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1455	struct cifsLockInfo *li, *tmp;
1456	__u64 length = 1 + flock->fl_end - flock->fl_start;
1457	struct list_head tmp_llist;
1458
1459	INIT_LIST_HEAD(&tmp_llist);
1460
1461	/*
1462	 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1463	 * and check it for zero before using.
1464	 */
1465	max_buf = tcon->ses->server->maxBuf;
1466	if (!max_buf)
1467		return -EINVAL;
1468
1469	max_num = (max_buf - sizeof(struct smb_hdr)) /
1470						sizeof(LOCKING_ANDX_RANGE);
1471	buf = kcalloc(max_num, sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1472	if (!buf)
1473		return -ENOMEM;
1474
1475	down_write(&cinode->lock_sem);
1476	for (i = 0; i < 2; i++) {
1477		cur = buf;
1478		num = 0;
1479		list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1480			if (flock->fl_start > li->offset ||
1481			    (flock->fl_start + length) <
1482			    (li->offset + li->length))
1483				continue;
1484			if (current->tgid != li->pid)
1485				continue;
1486			if (types[i] != li->type)
1487				continue;
1488			if (cinode->can_cache_brlcks) {
1489				/*
1490				 * We can cache brlock requests - simply remove
1491				 * a lock from the file's list.
1492				 */
1493				list_del(&li->llist);
1494				cifs_del_lock_waiters(li);
1495				kfree(li);
1496				continue;
1497			}
1498			cur->Pid = cpu_to_le16(li->pid);
1499			cur->LengthLow = cpu_to_le32((u32)li->length);
1500			cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1501			cur->OffsetLow = cpu_to_le32((u32)li->offset);
1502			cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1503			/*
1504			 * We need to save a lock here to let us add it again to
1505			 * the file's list if the unlock range request fails on
1506			 * the server.
1507			 */
1508			list_move(&li->llist, &tmp_llist);
1509			if (++num == max_num) {
1510				stored_rc = cifs_lockv(xid, tcon,
1511						       cfile->fid.netfid,
1512						       li->type, num, 0, buf);
1513				if (stored_rc) {
1514					/*
1515					 * We failed on the unlock range
1516					 * request - add all locks from the tmp
1517					 * list to the head of the file's list.
1518					 */
1519					cifs_move_llist(&tmp_llist,
1520							&cfile->llist->locks);
1521					rc = stored_rc;
1522				} else
1523					/*
1524					 * The unlock range request succeed -
1525					 * free the tmp list.
1526					 */
1527					cifs_free_llist(&tmp_llist);
1528				cur = buf;
1529				num = 0;
1530			} else
1531				cur++;
1532		}
1533		if (num) {
1534			stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1535					       types[i], num, 0, buf);
1536			if (stored_rc) {
1537				cifs_move_llist(&tmp_llist,
1538						&cfile->llist->locks);
1539				rc = stored_rc;
1540			} else
1541				cifs_free_llist(&tmp_llist);
1542		}
1543	}
1544
1545	up_write(&cinode->lock_sem);
1546	kfree(buf);
1547	return rc;
1548}
1549
1550static int
1551cifs_setlk(struct file *file, struct file_lock *flock, __u32 type,
1552	   bool wait_flag, bool posix_lck, int lock, int unlock,
1553	   unsigned int xid)
1554{
1555	int rc = 0;
1556	__u64 length = 1 + flock->fl_end - flock->fl_start;
1557	struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1558	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1559	struct TCP_Server_Info *server = tcon->ses->server;
1560	struct inode *inode = d_inode(cfile->dentry);
1561
1562	if (posix_lck) {
1563		int posix_lock_type;
1564
1565		rc = cifs_posix_lock_set(file, flock);
1566		if (!rc || rc < 0)
1567			return rc;
1568
1569		if (type & server->vals->shared_lock_type)
1570			posix_lock_type = CIFS_RDLCK;
1571		else
1572			posix_lock_type = CIFS_WRLCK;
1573
1574		if (unlock == 1)
1575			posix_lock_type = CIFS_UNLCK;
1576
1577		rc = CIFSSMBPosixLock(xid, tcon, cfile->fid.netfid,
1578				      hash_lockowner(flock->fl_owner),
1579				      flock->fl_start, length,
1580				      NULL, posix_lock_type, wait_flag);
1581		goto out;
1582	}
1583
1584	if (lock) {
1585		struct cifsLockInfo *lock;
1586
1587		lock = cifs_lock_init(flock->fl_start, length, type);
1588		if (!lock)
1589			return -ENOMEM;
1590
1591		rc = cifs_lock_add_if(cfile, lock, wait_flag);
1592		if (rc < 0) {
1593			kfree(lock);
1594			return rc;
1595		}
1596		if (!rc)
1597			goto out;
1598
1599		/*
1600		 * Windows 7 server can delay breaking lease from read to None
1601		 * if we set a byte-range lock on a file - break it explicitly
1602		 * before sending the lock to the server to be sure the next
1603		 * read won't conflict with non-overlapted locks due to
1604		 * pagereading.
1605		 */
1606		if (!CIFS_CACHE_WRITE(CIFS_I(inode)) &&
1607					CIFS_CACHE_READ(CIFS_I(inode))) {
1608			cifs_zap_mapping(inode);
1609			cifs_dbg(FYI, "Set no oplock for inode=%p due to mand locks\n",
1610				 inode);
1611			CIFS_I(inode)->oplock = 0;
1612		}
1613
1614		rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1615					    type, 1, 0, wait_flag);
1616		if (rc) {
1617			kfree(lock);
1618			return rc;
1619		}
1620
1621		cifs_lock_add(cfile, lock);
1622	} else if (unlock)
1623		rc = server->ops->mand_unlock_range(cfile, flock, xid);
1624
1625out:
1626	if (flock->fl_flags & FL_POSIX && !rc)
1627		rc = locks_lock_file_wait(file, flock);
1628	return rc;
1629}
1630
1631int cifs_lock(struct file *file, int cmd, struct file_lock *flock)
1632{
1633	int rc, xid;
1634	int lock = 0, unlock = 0;
1635	bool wait_flag = false;
1636	bool posix_lck = false;
1637	struct cifs_sb_info *cifs_sb;
1638	struct cifs_tcon *tcon;
1639	struct cifsInodeInfo *cinode;
1640	struct cifsFileInfo *cfile;
1641	__u16 netfid;
1642	__u32 type;
1643
1644	rc = -EACCES;
1645	xid = get_xid();
1646
1647	cifs_dbg(FYI, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld end: %lld\n",
1648		 cmd, flock->fl_flags, flock->fl_type,
1649		 flock->fl_start, flock->fl_end);
1650
1651	cfile = (struct cifsFileInfo *)file->private_data;
1652	tcon = tlink_tcon(cfile->tlink);
1653
1654	cifs_read_flock(flock, &type, &lock, &unlock, &wait_flag,
1655			tcon->ses->server);
1656
1657	cifs_sb = CIFS_FILE_SB(file);
1658	netfid = cfile->fid.netfid;
1659	cinode = CIFS_I(file_inode(file));
1660
1661	if (cap_unix(tcon->ses) &&
1662	    (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1663	    ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1664		posix_lck = true;
1665	/*
1666	 * BB add code here to normalize offset and length to account for
1667	 * negative length which we can not accept over the wire.
1668	 */
1669	if (IS_GETLK(cmd)) {
1670		rc = cifs_getlk(file, flock, type, wait_flag, posix_lck, xid);
1671		free_xid(xid);
1672		return rc;
1673	}
1674
1675	if (!lock && !unlock) {
1676		/*
1677		 * if no lock or unlock then nothing to do since we do not
1678		 * know what it is
1679		 */
1680		free_xid(xid);
1681		return -EOPNOTSUPP;
1682	}
1683
1684	rc = cifs_setlk(file, flock, type, wait_flag, posix_lck, lock, unlock,
1685			xid);
1686	free_xid(xid);
1687	return rc;
1688}
1689
1690/*
1691 * update the file size (if needed) after a write. Should be called with
1692 * the inode->i_lock held
1693 */
1694void
1695cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
1696		      unsigned int bytes_written)
1697{
1698	loff_t end_of_write = offset + bytes_written;
1699
1700	if (end_of_write > cifsi->server_eof)
1701		cifsi->server_eof = end_of_write;
1702}
1703
1704static ssize_t
1705cifs_write(struct cifsFileInfo *open_file, __u32 pid, const char *write_data,
1706	   size_t write_size, loff_t *offset)
1707{
1708	int rc = 0;
1709	unsigned int bytes_written = 0;
1710	unsigned int total_written;
1711	struct cifs_sb_info *cifs_sb;
1712	struct cifs_tcon *tcon;
1713	struct TCP_Server_Info *server;
1714	unsigned int xid;
1715	struct dentry *dentry = open_file->dentry;
1716	struct cifsInodeInfo *cifsi = CIFS_I(d_inode(dentry));
1717	struct cifs_io_parms io_parms;
1718
1719	cifs_sb = CIFS_SB(dentry->d_sb);
1720
1721	cifs_dbg(FYI, "write %zd bytes to offset %lld of %pd\n",
1722		 write_size, *offset, dentry);
1723
1724	tcon = tlink_tcon(open_file->tlink);
1725	server = tcon->ses->server;
1726
1727	if (!server->ops->sync_write)
1728		return -ENOSYS;
1729
1730	xid = get_xid();
1731
1732	for (total_written = 0; write_size > total_written;
1733	     total_written += bytes_written) {
1734		rc = -EAGAIN;
1735		while (rc == -EAGAIN) {
1736			struct kvec iov[2];
1737			unsigned int len;
1738
1739			if (open_file->invalidHandle) {
1740				/* we could deadlock if we called
1741				   filemap_fdatawait from here so tell
1742				   reopen_file not to flush data to
1743				   server now */
1744				rc = cifs_reopen_file(open_file, false);
1745				if (rc != 0)
1746					break;
1747			}
1748
1749			len = min(server->ops->wp_retry_size(d_inode(dentry)),
1750				  (unsigned int)write_size - total_written);
1751			/* iov[0] is reserved for smb header */
1752			iov[1].iov_base = (char *)write_data + total_written;
1753			iov[1].iov_len = len;
 
1754			io_parms.pid = pid;
1755			io_parms.tcon = tcon;
1756			io_parms.offset = *offset;
1757			io_parms.length = len;
1758			rc = server->ops->sync_write(xid, &open_file->fid,
1759					&io_parms, &bytes_written, iov, 1);
1760		}
1761		if (rc || (bytes_written == 0)) {
1762			if (total_written)
1763				break;
1764			else {
1765				free_xid(xid);
1766				return rc;
1767			}
1768		} else {
1769			spin_lock(&d_inode(dentry)->i_lock);
1770			cifs_update_eof(cifsi, *offset, bytes_written);
1771			spin_unlock(&d_inode(dentry)->i_lock);
1772			*offset += bytes_written;
1773		}
1774	}
1775
1776	cifs_stats_bytes_written(tcon, total_written);
1777
1778	if (total_written > 0) {
1779		spin_lock(&d_inode(dentry)->i_lock);
1780		if (*offset > d_inode(dentry)->i_size)
1781			i_size_write(d_inode(dentry), *offset);
1782		spin_unlock(&d_inode(dentry)->i_lock);
1783	}
1784	mark_inode_dirty_sync(d_inode(dentry));
1785	free_xid(xid);
1786	return total_written;
1787}
1788
1789struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1790					bool fsuid_only)
1791{
1792	struct cifsFileInfo *open_file = NULL;
1793	struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1794	struct cifs_tcon *tcon = cifs_sb_master_tcon(cifs_sb);
1795
1796	/* only filter by fsuid on multiuser mounts */
1797	if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1798		fsuid_only = false;
1799
1800	spin_lock(&tcon->open_file_lock);
1801	/* we could simply get the first_list_entry since write-only entries
1802	   are always at the end of the list but since the first entry might
1803	   have a close pending, we go through the whole list */
1804	list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1805		if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
1806			continue;
1807		if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1808			if (!open_file->invalidHandle) {
1809				/* found a good file */
1810				/* lock it so it will not be closed on us */
1811				cifsFileInfo_get(open_file);
1812				spin_unlock(&tcon->open_file_lock);
1813				return open_file;
1814			} /* else might as well continue, and look for
1815			     another, or simply have the caller reopen it
1816			     again rather than trying to fix this handle */
1817		} else /* write only file */
1818			break; /* write only files are last so must be done */
1819	}
1820	spin_unlock(&tcon->open_file_lock);
1821	return NULL;
1822}
1823
1824struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1825					bool fsuid_only)
1826{
1827	struct cifsFileInfo *open_file, *inv_file = NULL;
1828	struct cifs_sb_info *cifs_sb;
1829	struct cifs_tcon *tcon;
1830	bool any_available = false;
1831	int rc;
1832	unsigned int refind = 0;
1833
1834	/* Having a null inode here (because mapping->host was set to zero by
1835	the VFS or MM) should not happen but we had reports of on oops (due to
1836	it being zero) during stress testcases so we need to check for it */
1837
1838	if (cifs_inode == NULL) {
1839		cifs_dbg(VFS, "Null inode passed to cifs_writeable_file\n");
1840		dump_stack();
1841		return NULL;
1842	}
1843
1844	cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1845	tcon = cifs_sb_master_tcon(cifs_sb);
1846
1847	/* only filter by fsuid on multiuser mounts */
1848	if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1849		fsuid_only = false;
1850
1851	spin_lock(&tcon->open_file_lock);
1852refind_writable:
1853	if (refind > MAX_REOPEN_ATT) {
1854		spin_unlock(&tcon->open_file_lock);
1855		return NULL;
1856	}
1857	list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1858		if (!any_available && open_file->pid != current->tgid)
1859			continue;
1860		if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
1861			continue;
1862		if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1863			if (!open_file->invalidHandle) {
1864				/* found a good writable file */
1865				cifsFileInfo_get(open_file);
1866				spin_unlock(&tcon->open_file_lock);
1867				return open_file;
1868			} else {
1869				if (!inv_file)
1870					inv_file = open_file;
1871			}
1872		}
1873	}
1874	/* couldn't find useable FH with same pid, try any available */
1875	if (!any_available) {
1876		any_available = true;
1877		goto refind_writable;
1878	}
1879
1880	if (inv_file) {
1881		any_available = false;
1882		cifsFileInfo_get(inv_file);
1883	}
1884
1885	spin_unlock(&tcon->open_file_lock);
1886
1887	if (inv_file) {
1888		rc = cifs_reopen_file(inv_file, false);
1889		if (!rc)
1890			return inv_file;
1891		else {
1892			spin_lock(&tcon->open_file_lock);
1893			list_move_tail(&inv_file->flist,
1894					&cifs_inode->openFileList);
1895			spin_unlock(&tcon->open_file_lock);
1896			cifsFileInfo_put(inv_file);
 
1897			++refind;
1898			inv_file = NULL;
1899			spin_lock(&tcon->open_file_lock);
1900			goto refind_writable;
1901		}
1902	}
1903
1904	return NULL;
1905}
1906
1907static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1908{
1909	struct address_space *mapping = page->mapping;
1910	loff_t offset = (loff_t)page->index << PAGE_SHIFT;
1911	char *write_data;
1912	int rc = -EFAULT;
1913	int bytes_written = 0;
1914	struct inode *inode;
1915	struct cifsFileInfo *open_file;
1916
1917	if (!mapping || !mapping->host)
1918		return -EFAULT;
1919
1920	inode = page->mapping->host;
1921
1922	offset += (loff_t)from;
1923	write_data = kmap(page);
1924	write_data += from;
1925
1926	if ((to > PAGE_SIZE) || (from > to)) {
1927		kunmap(page);
1928		return -EIO;
1929	}
1930
1931	/* racing with truncate? */
1932	if (offset > mapping->host->i_size) {
1933		kunmap(page);
1934		return 0; /* don't care */
1935	}
1936
1937	/* check to make sure that we are not extending the file */
1938	if (mapping->host->i_size - offset < (loff_t)to)
1939		to = (unsigned)(mapping->host->i_size - offset);
1940
1941	open_file = find_writable_file(CIFS_I(mapping->host), false);
1942	if (open_file) {
1943		bytes_written = cifs_write(open_file, open_file->pid,
1944					   write_data, to - from, &offset);
1945		cifsFileInfo_put(open_file);
1946		/* Does mm or vfs already set times? */
1947		inode->i_atime = inode->i_mtime = current_time(inode);
1948		if ((bytes_written > 0) && (offset))
1949			rc = 0;
1950		else if (bytes_written < 0)
1951			rc = bytes_written;
1952	} else {
1953		cifs_dbg(FYI, "No writeable filehandles for inode\n");
1954		rc = -EIO;
1955	}
1956
1957	kunmap(page);
1958	return rc;
1959}
1960
1961static struct cifs_writedata *
1962wdata_alloc_and_fillpages(pgoff_t tofind, struct address_space *mapping,
1963			  pgoff_t end, pgoff_t *index,
1964			  unsigned int *found_pages)
1965{
1966	struct cifs_writedata *wdata;
1967
1968	wdata = cifs_writedata_alloc((unsigned int)tofind,
1969				     cifs_writev_complete);
1970	if (!wdata)
1971		return NULL;
1972
1973	*found_pages = find_get_pages_range_tag(mapping, index, end,
1974				PAGECACHE_TAG_DIRTY, tofind, wdata->pages);
1975	return wdata;
1976}
1977
1978static unsigned int
1979wdata_prepare_pages(struct cifs_writedata *wdata, unsigned int found_pages,
1980		    struct address_space *mapping,
1981		    struct writeback_control *wbc,
1982		    pgoff_t end, pgoff_t *index, pgoff_t *next, bool *done)
1983{
1984	unsigned int nr_pages = 0, i;
1985	struct page *page;
1986
1987	for (i = 0; i < found_pages; i++) {
1988		page = wdata->pages[i];
1989		/*
1990		 * At this point we hold neither the i_pages lock nor the
1991		 * page lock: the page may be truncated or invalidated
1992		 * (changing page->mapping to NULL), or even swizzled
1993		 * back from swapper_space to tmpfs file mapping
1994		 */
1995
1996		if (nr_pages == 0)
1997			lock_page(page);
1998		else if (!trylock_page(page))
1999			break;
2000
2001		if (unlikely(page->mapping != mapping)) {
2002			unlock_page(page);
2003			break;
2004		}
2005
2006		if (!wbc->range_cyclic && page->index > end) {
2007			*done = true;
2008			unlock_page(page);
2009			break;
2010		}
2011
2012		if (*next && (page->index != *next)) {
2013			/* Not next consecutive page */
2014			unlock_page(page);
2015			break;
2016		}
2017
2018		if (wbc->sync_mode != WB_SYNC_NONE)
2019			wait_on_page_writeback(page);
2020
2021		if (PageWriteback(page) ||
2022				!clear_page_dirty_for_io(page)) {
2023			unlock_page(page);
2024			break;
2025		}
2026
2027		/*
2028		 * This actually clears the dirty bit in the radix tree.
2029		 * See cifs_writepage() for more commentary.
2030		 */
2031		set_page_writeback(page);
2032		if (page_offset(page) >= i_size_read(mapping->host)) {
2033			*done = true;
2034			unlock_page(page);
2035			end_page_writeback(page);
2036			break;
2037		}
2038
2039		wdata->pages[i] = page;
2040		*next = page->index + 1;
2041		++nr_pages;
2042	}
2043
2044	/* reset index to refind any pages skipped */
2045	if (nr_pages == 0)
2046		*index = wdata->pages[0]->index + 1;
2047
2048	/* put any pages we aren't going to use */
2049	for (i = nr_pages; i < found_pages; i++) {
2050		put_page(wdata->pages[i]);
2051		wdata->pages[i] = NULL;
2052	}
2053
2054	return nr_pages;
2055}
2056
2057static int
2058wdata_send_pages(struct cifs_writedata *wdata, unsigned int nr_pages,
2059		 struct address_space *mapping, struct writeback_control *wbc)
2060{
2061	int rc = 0;
2062	struct TCP_Server_Info *server;
2063	unsigned int i;
2064
2065	wdata->sync_mode = wbc->sync_mode;
2066	wdata->nr_pages = nr_pages;
2067	wdata->offset = page_offset(wdata->pages[0]);
2068	wdata->pagesz = PAGE_SIZE;
2069	wdata->tailsz = min(i_size_read(mapping->host) -
2070			page_offset(wdata->pages[nr_pages - 1]),
2071			(loff_t)PAGE_SIZE);
2072	wdata->bytes = ((nr_pages - 1) * PAGE_SIZE) + wdata->tailsz;
2073
2074	if (wdata->cfile != NULL)
2075		cifsFileInfo_put(wdata->cfile);
2076	wdata->cfile = find_writable_file(CIFS_I(mapping->host), false);
2077	if (!wdata->cfile) {
2078		cifs_dbg(VFS, "No writable handles for inode\n");
2079		rc = -EBADF;
2080	} else {
2081		wdata->pid = wdata->cfile->pid;
2082		server = tlink_tcon(wdata->cfile->tlink)->ses->server;
2083		rc = server->ops->async_writev(wdata, cifs_writedata_release);
2084	}
2085
2086	for (i = 0; i < nr_pages; ++i)
2087		unlock_page(wdata->pages[i]);
2088
2089	return rc;
2090}
2091
2092static int cifs_writepages(struct address_space *mapping,
2093			   struct writeback_control *wbc)
2094{
2095	struct cifs_sb_info *cifs_sb = CIFS_SB(mapping->host->i_sb);
2096	struct TCP_Server_Info *server;
2097	bool done = false, scanned = false, range_whole = false;
2098	pgoff_t end, index;
2099	struct cifs_writedata *wdata;
 
2100	int rc = 0;
2101
2102	/*
2103	 * If wsize is smaller than the page cache size, default to writing
2104	 * one page at a time via cifs_writepage
2105	 */
2106	if (cifs_sb->wsize < PAGE_SIZE)
2107		return generic_writepages(mapping, wbc);
2108
2109	if (wbc->range_cyclic) {
2110		index = mapping->writeback_index; /* Start from prev offset */
2111		end = -1;
2112	} else {
2113		index = wbc->range_start >> PAGE_SHIFT;
2114		end = wbc->range_end >> PAGE_SHIFT;
2115		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2116			range_whole = true;
2117		scanned = true;
2118	}
2119	server = cifs_sb_master_tcon(cifs_sb)->ses->server;
2120retry:
2121	while (!done && index <= end) {
2122		unsigned int i, nr_pages, found_pages, wsize, credits;
2123		pgoff_t next = 0, tofind, saved_index = index;
 
2124
2125		rc = server->ops->wait_mtu_credits(server, cifs_sb->wsize,
2126						   &wsize, &credits);
2127		if (rc)
2128			break;
2129
2130		tofind = min((wsize / PAGE_SIZE) - 1, end - index) + 1;
2131
2132		wdata = wdata_alloc_and_fillpages(tofind, mapping, end, &index,
2133						  &found_pages);
2134		if (!wdata) {
2135			rc = -ENOMEM;
2136			add_credits_and_wake_if(server, credits, 0);
2137			break;
2138		}
2139
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2140		if (found_pages == 0) {
2141			kref_put(&wdata->refcount, cifs_writedata_release);
2142			add_credits_and_wake_if(server, credits, 0);
2143			break;
2144		}
2145
2146		nr_pages = wdata_prepare_pages(wdata, found_pages, mapping, wbc,
2147					       end, &index, &next, &done);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2148
2149		/* nothing to write? */
2150		if (nr_pages == 0) {
2151			kref_put(&wdata->refcount, cifs_writedata_release);
2152			add_credits_and_wake_if(server, credits, 0);
2153			continue;
2154		}
2155
2156		wdata->credits = credits;
 
 
 
2157
2158		rc = wdata_send_pages(wdata, nr_pages, mapping, wbc);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2159
2160		/* send failure -- clean up the mess */
2161		if (rc != 0) {
2162			add_credits_and_wake_if(server, wdata->credits, 0);
2163			for (i = 0; i < nr_pages; ++i) {
2164				if (rc == -EAGAIN)
2165					redirty_page_for_writepage(wbc,
2166							   wdata->pages[i]);
2167				else
2168					SetPageError(wdata->pages[i]);
2169				end_page_writeback(wdata->pages[i]);
2170				put_page(wdata->pages[i]);
2171			}
2172			if (rc != -EAGAIN)
2173				mapping_set_error(mapping, rc);
2174		}
2175		kref_put(&wdata->refcount, cifs_writedata_release);
2176
2177		if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN) {
2178			index = saved_index;
2179			continue;
2180		}
2181
2182		wbc->nr_to_write -= nr_pages;
2183		if (wbc->nr_to_write <= 0)
2184			done = true;
2185
2186		index = next;
2187	}
2188
2189	if (!scanned && !done) {
2190		/*
2191		 * We hit the last page and there is more work to be done: wrap
2192		 * back to the start of the file
2193		 */
2194		scanned = true;
2195		index = 0;
2196		goto retry;
2197	}
2198
2199	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2200		mapping->writeback_index = index;
2201
2202	return rc;
2203}
2204
2205static int
2206cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
2207{
2208	int rc;
2209	unsigned int xid;
2210
2211	xid = get_xid();
2212/* BB add check for wbc flags */
2213	get_page(page);
2214	if (!PageUptodate(page))
2215		cifs_dbg(FYI, "ppw - page not up to date\n");
2216
2217	/*
2218	 * Set the "writeback" flag, and clear "dirty" in the radix tree.
2219	 *
2220	 * A writepage() implementation always needs to do either this,
2221	 * or re-dirty the page with "redirty_page_for_writepage()" in
2222	 * the case of a failure.
2223	 *
2224	 * Just unlocking the page will cause the radix tree tag-bits
2225	 * to fail to update with the state of the page correctly.
2226	 */
2227	set_page_writeback(page);
2228retry_write:
2229	rc = cifs_partialpagewrite(page, 0, PAGE_SIZE);
2230	if (rc == -EAGAIN) {
2231		if (wbc->sync_mode == WB_SYNC_ALL)
2232			goto retry_write;
2233		redirty_page_for_writepage(wbc, page);
2234	} else if (rc != 0) {
2235		SetPageError(page);
2236		mapping_set_error(page->mapping, rc);
2237	} else {
2238		SetPageUptodate(page);
2239	}
2240	end_page_writeback(page);
2241	put_page(page);
2242	free_xid(xid);
2243	return rc;
2244}
2245
2246static int cifs_writepage(struct page *page, struct writeback_control *wbc)
2247{
2248	int rc = cifs_writepage_locked(page, wbc);
2249	unlock_page(page);
2250	return rc;
2251}
2252
2253static int cifs_write_end(struct file *file, struct address_space *mapping,
2254			loff_t pos, unsigned len, unsigned copied,
2255			struct page *page, void *fsdata)
2256{
2257	int rc;
2258	struct inode *inode = mapping->host;
2259	struct cifsFileInfo *cfile = file->private_data;
2260	struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
2261	__u32 pid;
2262
2263	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2264		pid = cfile->pid;
2265	else
2266		pid = current->tgid;
2267
2268	cifs_dbg(FYI, "write_end for page %p from pos %lld with %d bytes\n",
2269		 page, pos, copied);
2270
2271	if (PageChecked(page)) {
2272		if (copied == len)
2273			SetPageUptodate(page);
2274		ClearPageChecked(page);
2275	} else if (!PageUptodate(page) && copied == PAGE_SIZE)
2276		SetPageUptodate(page);
2277
2278	if (!PageUptodate(page)) {
2279		char *page_data;
2280		unsigned offset = pos & (PAGE_SIZE - 1);
2281		unsigned int xid;
2282
2283		xid = get_xid();
2284		/* this is probably better than directly calling
2285		   partialpage_write since in this function the file handle is
2286		   known which we might as well	leverage */
2287		/* BB check if anything else missing out of ppw
2288		   such as updating last write time */
2289		page_data = kmap(page);
2290		rc = cifs_write(cfile, pid, page_data + offset, copied, &pos);
2291		/* if (rc < 0) should we set writebehind rc? */
2292		kunmap(page);
2293
2294		free_xid(xid);
2295	} else {
2296		rc = copied;
2297		pos += copied;
2298		set_page_dirty(page);
2299	}
2300
2301	if (rc > 0) {
2302		spin_lock(&inode->i_lock);
2303		if (pos > inode->i_size)
2304			i_size_write(inode, pos);
2305		spin_unlock(&inode->i_lock);
2306	}
2307
2308	unlock_page(page);
2309	put_page(page);
2310
2311	return rc;
2312}
2313
2314int cifs_strict_fsync(struct file *file, loff_t start, loff_t end,
2315		      int datasync)
2316{
2317	unsigned int xid;
2318	int rc = 0;
2319	struct cifs_tcon *tcon;
2320	struct TCP_Server_Info *server;
2321	struct cifsFileInfo *smbfile = file->private_data;
2322	struct inode *inode = file_inode(file);
2323	struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2324
2325	rc = file_write_and_wait_range(file, start, end);
2326	if (rc)
2327		return rc;
2328	inode_lock(inode);
2329
2330	xid = get_xid();
2331
2332	cifs_dbg(FYI, "Sync file - name: %pD datasync: 0x%x\n",
2333		 file, datasync);
2334
2335	if (!CIFS_CACHE_READ(CIFS_I(inode))) {
2336		rc = cifs_zap_mapping(inode);
2337		if (rc) {
2338			cifs_dbg(FYI, "rc: %d during invalidate phase\n", rc);
2339			rc = 0; /* don't care about it in fsync */
2340		}
2341	}
2342
2343	tcon = tlink_tcon(smbfile->tlink);
2344	if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2345		server = tcon->ses->server;
2346		if (server->ops->flush)
2347			rc = server->ops->flush(xid, tcon, &smbfile->fid);
2348		else
2349			rc = -ENOSYS;
2350	}
2351
2352	free_xid(xid);
2353	inode_unlock(inode);
2354	return rc;
2355}
2356
2357int cifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2358{
2359	unsigned int xid;
2360	int rc = 0;
2361	struct cifs_tcon *tcon;
2362	struct TCP_Server_Info *server;
2363	struct cifsFileInfo *smbfile = file->private_data;
2364	struct cifs_sb_info *cifs_sb = CIFS_FILE_SB(file);
2365	struct inode *inode = file->f_mapping->host;
2366
2367	rc = file_write_and_wait_range(file, start, end);
2368	if (rc)
2369		return rc;
2370	inode_lock(inode);
2371
2372	xid = get_xid();
2373
2374	cifs_dbg(FYI, "Sync file - name: %pD datasync: 0x%x\n",
2375		 file, datasync);
2376
2377	tcon = tlink_tcon(smbfile->tlink);
2378	if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2379		server = tcon->ses->server;
2380		if (server->ops->flush)
2381			rc = server->ops->flush(xid, tcon, &smbfile->fid);
2382		else
2383			rc = -ENOSYS;
2384	}
2385
2386	free_xid(xid);
2387	inode_unlock(inode);
2388	return rc;
2389}
2390
2391/*
2392 * As file closes, flush all cached write data for this inode checking
2393 * for write behind errors.
2394 */
2395int cifs_flush(struct file *file, fl_owner_t id)
2396{
2397	struct inode *inode = file_inode(file);
2398	int rc = 0;
2399
2400	if (file->f_mode & FMODE_WRITE)
2401		rc = filemap_write_and_wait(inode->i_mapping);
2402
2403	cifs_dbg(FYI, "Flush inode %p file %p rc %d\n", inode, file, rc);
2404
2405	return rc;
2406}
2407
2408static int
2409cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
2410{
2411	int rc = 0;
2412	unsigned long i;
2413
2414	for (i = 0; i < num_pages; i++) {
2415		pages[i] = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2416		if (!pages[i]) {
2417			/*
2418			 * save number of pages we have already allocated and
2419			 * return with ENOMEM error
2420			 */
2421			num_pages = i;
2422			rc = -ENOMEM;
2423			break;
2424		}
2425	}
2426
2427	if (rc) {
2428		for (i = 0; i < num_pages; i++)
2429			put_page(pages[i]);
2430	}
2431	return rc;
2432}
2433
2434static inline
2435size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
2436{
2437	size_t num_pages;
2438	size_t clen;
2439
2440	clen = min_t(const size_t, len, wsize);
2441	num_pages = DIV_ROUND_UP(clen, PAGE_SIZE);
2442
2443	if (cur_len)
2444		*cur_len = clen;
2445
2446	return num_pages;
2447}
2448
2449static void
2450cifs_uncached_writedata_release(struct kref *refcount)
2451{
2452	int i;
2453	struct cifs_writedata *wdata = container_of(refcount,
2454					struct cifs_writedata, refcount);
2455
2456	kref_put(&wdata->ctx->refcount, cifs_aio_ctx_release);
2457	for (i = 0; i < wdata->nr_pages; i++)
2458		put_page(wdata->pages[i]);
2459	cifs_writedata_release(refcount);
 
 
2460}
2461
2462static void collect_uncached_write_data(struct cifs_aio_ctx *ctx);
2463
2464static void
2465cifs_uncached_writev_complete(struct work_struct *work)
2466{
 
2467	struct cifs_writedata *wdata = container_of(work,
2468					struct cifs_writedata, work);
2469	struct inode *inode = d_inode(wdata->cfile->dentry);
2470	struct cifsInodeInfo *cifsi = CIFS_I(inode);
2471
2472	spin_lock(&inode->i_lock);
2473	cifs_update_eof(cifsi, wdata->offset, wdata->bytes);
2474	if (cifsi->server_eof > inode->i_size)
2475		i_size_write(inode, cifsi->server_eof);
2476	spin_unlock(&inode->i_lock);
2477
2478	complete(&wdata->done);
2479	collect_uncached_write_data(wdata->ctx);
2480	/* the below call can possibly free the last ref to aio ctx */
2481	kref_put(&wdata->refcount, cifs_uncached_writedata_release);
 
 
 
 
2482}
2483
 
2484static int
2485wdata_fill_from_iovec(struct cifs_writedata *wdata, struct iov_iter *from,
2486		      size_t *len, unsigned long *num_pages)
2487{
2488	size_t save_len, copied, bytes, cur_len = *len;
2489	unsigned long i, nr_pages = *num_pages;
2490
2491	save_len = cur_len;
2492	for (i = 0; i < nr_pages; i++) {
2493		bytes = min_t(const size_t, cur_len, PAGE_SIZE);
2494		copied = copy_page_from_iter(wdata->pages[i], 0, bytes, from);
2495		cur_len -= copied;
2496		/*
2497		 * If we didn't copy as much as we expected, then that
2498		 * may mean we trod into an unmapped area. Stop copying
2499		 * at that point. On the next pass through the big
2500		 * loop, we'll likely end up getting a zero-length
2501		 * write and bailing out of it.
2502		 */
2503		if (copied < bytes)
2504			break;
2505	}
2506	cur_len = save_len - cur_len;
2507	*len = cur_len;
2508
2509	/*
2510	 * If we have no data to send, then that probably means that
2511	 * the copy above failed altogether. That's most likely because
2512	 * the address in the iovec was bogus. Return -EFAULT and let
2513	 * the caller free anything we allocated and bail out.
2514	 */
2515	if (!cur_len)
2516		return -EFAULT;
2517
2518	/*
2519	 * i + 1 now represents the number of pages we actually used in
2520	 * the copy phase above.
2521	 */
2522	*num_pages = i + 1;
2523	return 0;
2524}
2525
2526static int
2527cifs_write_from_iter(loff_t offset, size_t len, struct iov_iter *from,
2528		     struct cifsFileInfo *open_file,
2529		     struct cifs_sb_info *cifs_sb, struct list_head *wdata_list,
2530		     struct cifs_aio_ctx *ctx)
2531{
2532	int rc = 0;
2533	size_t cur_len;
2534	unsigned long nr_pages, num_pages, i;
2535	struct cifs_writedata *wdata;
2536	struct iov_iter saved_from = *from;
2537	loff_t saved_offset = offset;
 
 
 
 
 
2538	pid_t pid;
2539	struct TCP_Server_Info *server;
 
 
 
 
 
 
 
 
 
 
 
 
 
2540
2541	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2542		pid = open_file->pid;
2543	else
2544		pid = current->tgid;
2545
2546	server = tlink_tcon(open_file->tlink)->ses->server;
2547
2548	do {
2549		unsigned int wsize, credits;
2550
2551		rc = server->ops->wait_mtu_credits(server, cifs_sb->wsize,
2552						   &wsize, &credits);
2553		if (rc)
2554			break;
2555
2556		nr_pages = get_numpages(wsize, len, &cur_len);
2557		wdata = cifs_writedata_alloc(nr_pages,
2558					     cifs_uncached_writev_complete);
2559		if (!wdata) {
2560			rc = -ENOMEM;
2561			add_credits_and_wake_if(server, credits, 0);
2562			break;
2563		}
2564
2565		rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
2566		if (rc) {
2567			kfree(wdata);
2568			add_credits_and_wake_if(server, credits, 0);
2569			break;
2570		}
2571
2572		num_pages = nr_pages;
2573		rc = wdata_fill_from_iovec(wdata, from, &cur_len, &num_pages);
2574		if (rc) {
2575			for (i = 0; i < nr_pages; i++)
2576				put_page(wdata->pages[i]);
2577			kfree(wdata);
2578			add_credits_and_wake_if(server, credits, 0);
2579			break;
2580		}
2581
2582		/*
2583		 * Bring nr_pages down to the number of pages we actually used,
2584		 * and free any pages that we didn't use.
2585		 */
2586		for ( ; nr_pages > num_pages; nr_pages--)
2587			put_page(wdata->pages[nr_pages - 1]);
2588
2589		wdata->sync_mode = WB_SYNC_ALL;
2590		wdata->nr_pages = nr_pages;
2591		wdata->offset = (__u64)offset;
2592		wdata->cfile = cifsFileInfo_get(open_file);
2593		wdata->pid = pid;
2594		wdata->bytes = cur_len;
2595		wdata->pagesz = PAGE_SIZE;
2596		wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
2597		wdata->credits = credits;
2598		wdata->ctx = ctx;
2599		kref_get(&ctx->refcount);
2600
2601		if (!wdata->cfile->invalidHandle ||
2602		    !(rc = cifs_reopen_file(wdata->cfile, false)))
2603			rc = server->ops->async_writev(wdata,
2604					cifs_uncached_writedata_release);
2605		if (rc) {
2606			add_credits_and_wake_if(server, wdata->credits, 0);
2607			kref_put(&wdata->refcount,
2608				 cifs_uncached_writedata_release);
2609			if (rc == -EAGAIN) {
2610				*from = saved_from;
2611				iov_iter_advance(from, offset - saved_offset);
2612				continue;
2613			}
2614			break;
2615		}
2616
2617		list_add_tail(&wdata->list, wdata_list);
2618		offset += cur_len;
2619		len -= cur_len;
2620	} while (len > 0);
2621
2622	return rc;
2623}
2624
2625static void collect_uncached_write_data(struct cifs_aio_ctx *ctx)
2626{
2627	struct cifs_writedata *wdata, *tmp;
2628	struct cifs_tcon *tcon;
2629	struct cifs_sb_info *cifs_sb;
2630	struct dentry *dentry = ctx->cfile->dentry;
2631	unsigned int i;
2632	int rc;
2633
2634	tcon = tlink_tcon(ctx->cfile->tlink);
2635	cifs_sb = CIFS_SB(dentry->d_sb);
2636
2637	mutex_lock(&ctx->aio_mutex);
2638
2639	if (list_empty(&ctx->list)) {
2640		mutex_unlock(&ctx->aio_mutex);
2641		return;
2642	}
2643
2644	rc = ctx->rc;
2645	/*
2646	 * Wait for and collect replies for any successful sends in order of
2647	 * increasing offset. Once an error is hit, then return without waiting
2648	 * for any more replies.
2649	 */
2650restart_loop:
2651	list_for_each_entry_safe(wdata, tmp, &ctx->list, list) {
2652		if (!rc) {
2653			if (!try_wait_for_completion(&wdata->done)) {
2654				mutex_unlock(&ctx->aio_mutex);
2655				return;
2656			}
2657
2658			if (wdata->result)
2659				rc = wdata->result;
2660			else
2661				ctx->total_len += wdata->bytes;
2662
2663			/* resend call if it's a retryable error */
2664			if (rc == -EAGAIN) {
2665				struct list_head tmp_list;
2666				struct iov_iter tmp_from = ctx->iter;
2667
2668				INIT_LIST_HEAD(&tmp_list);
2669				list_del_init(&wdata->list);
2670
2671				iov_iter_advance(&tmp_from,
2672						 wdata->offset - ctx->pos);
2673
2674				rc = cifs_write_from_iter(wdata->offset,
2675						wdata->bytes, &tmp_from,
2676						ctx->cfile, cifs_sb, &tmp_list,
2677						ctx);
2678
2679				list_splice(&tmp_list, &ctx->list);
2680
2681				kref_put(&wdata->refcount,
2682					 cifs_uncached_writedata_release);
2683				goto restart_loop;
2684			}
2685		}
2686		list_del_init(&wdata->list);
2687		kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2688	}
2689
2690	for (i = 0; i < ctx->npages; i++)
2691		put_page(ctx->bv[i].bv_page);
2692
2693	cifs_stats_bytes_written(tcon, ctx->total_len);
2694	set_bit(CIFS_INO_INVALID_MAPPING, &CIFS_I(dentry->d_inode)->flags);
2695
2696	ctx->rc = (rc == 0) ? ctx->total_len : rc;
2697
2698	mutex_unlock(&ctx->aio_mutex);
2699
2700	if (ctx->iocb && ctx->iocb->ki_complete)
2701		ctx->iocb->ki_complete(ctx->iocb, ctx->rc, 0);
2702	else
2703		complete(&ctx->done);
2704}
2705
2706ssize_t cifs_user_writev(struct kiocb *iocb, struct iov_iter *from)
 
2707{
2708	struct file *file = iocb->ki_filp;
2709	ssize_t total_written = 0;
2710	struct cifsFileInfo *cfile;
2711	struct cifs_tcon *tcon;
2712	struct cifs_sb_info *cifs_sb;
2713	struct cifs_aio_ctx *ctx;
2714	struct iov_iter saved_from = *from;
2715	int rc;
2716
2717	/*
2718	 * BB - optimize the way when signing is disabled. We can drop this
2719	 * extra memory-to-memory copying and use iovec buffers for constructing
2720	 * write request.
2721	 */
2722
2723	rc = generic_write_checks(iocb, from);
2724	if (rc <= 0)
2725		return rc;
2726
2727	cifs_sb = CIFS_FILE_SB(file);
2728	cfile = file->private_data;
2729	tcon = tlink_tcon(cfile->tlink);
2730
2731	if (!tcon->ses->server->ops->async_writev)
2732		return -ENOSYS;
2733
2734	ctx = cifs_aio_ctx_alloc();
2735	if (!ctx)
2736		return -ENOMEM;
2737
2738	ctx->cfile = cifsFileInfo_get(cfile);
2739
2740	if (!is_sync_kiocb(iocb))
2741		ctx->iocb = iocb;
2742
2743	ctx->pos = iocb->ki_pos;
2744
2745	rc = setup_aio_ctx_iter(ctx, from, WRITE);
2746	if (rc) {
2747		kref_put(&ctx->refcount, cifs_aio_ctx_release);
2748		return rc;
2749	}
2750
2751	/* grab a lock here due to read response handlers can access ctx */
2752	mutex_lock(&ctx->aio_mutex);
2753
2754	rc = cifs_write_from_iter(iocb->ki_pos, ctx->len, &saved_from,
2755				  cfile, cifs_sb, &ctx->list, ctx);
2756
2757	/*
2758	 * If at least one write was successfully sent, then discard any rc
2759	 * value from the later writes. If the other write succeeds, then
2760	 * we'll end up returning whatever was written. If it fails, then
2761	 * we'll get a new rc value from that.
2762	 */
2763	if (!list_empty(&ctx->list))
2764		rc = 0;
2765
2766	mutex_unlock(&ctx->aio_mutex);
2767
2768	if (rc) {
2769		kref_put(&ctx->refcount, cifs_aio_ctx_release);
2770		return rc;
2771	}
2772
2773	if (!is_sync_kiocb(iocb)) {
2774		kref_put(&ctx->refcount, cifs_aio_ctx_release);
2775		return -EIOCBQUEUED;
2776	}
2777
2778	rc = wait_for_completion_killable(&ctx->done);
2779	if (rc) {
2780		mutex_lock(&ctx->aio_mutex);
2781		ctx->rc = rc = -EINTR;
2782		total_written = ctx->total_len;
2783		mutex_unlock(&ctx->aio_mutex);
2784	} else {
2785		rc = ctx->rc;
2786		total_written = ctx->total_len;
2787	}
2788
2789	kref_put(&ctx->refcount, cifs_aio_ctx_release);
2790
2791	if (unlikely(!total_written))
2792		return rc;
2793
2794	iocb->ki_pos += total_written;
2795	return total_written;
2796}
2797
2798static ssize_t
2799cifs_writev(struct kiocb *iocb, struct iov_iter *from)
2800{
2801	struct file *file = iocb->ki_filp;
2802	struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
2803	struct inode *inode = file->f_mapping->host;
2804	struct cifsInodeInfo *cinode = CIFS_I(inode);
2805	struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
2806	ssize_t rc;
2807
2808	inode_lock(inode);
2809	/*
2810	 * We need to hold the sem to be sure nobody modifies lock list
2811	 * with a brlock that prevents writing.
2812	 */
2813	down_read(&cinode->lock_sem);
2814
2815	rc = generic_write_checks(iocb, from);
2816	if (rc <= 0)
2817		goto out;
2818
2819	if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(from),
2820				     server->vals->exclusive_lock_type, NULL,
2821				     CIFS_WRITE_OP))
2822		rc = __generic_file_write_iter(iocb, from);
2823	else
2824		rc = -EACCES;
2825out:
2826	up_read(&cinode->lock_sem);
2827	inode_unlock(inode);
2828
2829	if (rc > 0)
2830		rc = generic_write_sync(iocb, rc);
2831	return rc;
2832}
2833
2834ssize_t
2835cifs_strict_writev(struct kiocb *iocb, struct iov_iter *from)
2836{
2837	struct inode *inode = file_inode(iocb->ki_filp);
2838	struct cifsInodeInfo *cinode = CIFS_I(inode);
2839	struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2840	struct cifsFileInfo *cfile = (struct cifsFileInfo *)
2841						iocb->ki_filp->private_data;
2842	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
2843	ssize_t written;
2844
2845	written = cifs_get_writer(cinode);
2846	if (written)
2847		return written;
2848
2849	if (CIFS_CACHE_WRITE(cinode)) {
2850		if (cap_unix(tcon->ses) &&
2851		(CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability))
2852		  && ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0)) {
2853			written = generic_file_write_iter(iocb, from);
2854			goto out;
2855		}
2856		written = cifs_writev(iocb, from);
2857		goto out;
2858	}
2859	/*
2860	 * For non-oplocked files in strict cache mode we need to write the data
2861	 * to the server exactly from the pos to pos+len-1 rather than flush all
2862	 * affected pages because it may cause a error with mandatory locks on
2863	 * these pages but not on the region from pos to ppos+len-1.
2864	 */
2865	written = cifs_user_writev(iocb, from);
2866	if (written > 0 && CIFS_CACHE_READ(cinode)) {
2867		/*
2868		 * Windows 7 server can delay breaking level2 oplock if a write
2869		 * request comes - break it on the client to prevent reading
2870		 * an old data.
2871		 */
2872		cifs_zap_mapping(inode);
2873		cifs_dbg(FYI, "Set no oplock for inode=%p after a write operation\n",
2874			 inode);
2875		cinode->oplock = 0;
2876	}
2877out:
2878	cifs_put_writer(cinode);
2879	return written;
2880}
2881
2882static struct cifs_readdata *
2883cifs_readdata_alloc(unsigned int nr_pages, work_func_t complete)
2884{
2885	struct cifs_readdata *rdata;
2886
2887	rdata = kzalloc(sizeof(*rdata) + (sizeof(struct page *) * nr_pages),
2888			GFP_KERNEL);
2889	if (rdata != NULL) {
2890		kref_init(&rdata->refcount);
2891		INIT_LIST_HEAD(&rdata->list);
2892		init_completion(&rdata->done);
2893		INIT_WORK(&rdata->work, complete);
 
2894	}
2895
2896	return rdata;
2897}
2898
2899void
2900cifs_readdata_release(struct kref *refcount)
2901{
2902	struct cifs_readdata *rdata = container_of(refcount,
2903					struct cifs_readdata, refcount);
2904#ifdef CONFIG_CIFS_SMB_DIRECT
2905	if (rdata->mr) {
2906		smbd_deregister_mr(rdata->mr);
2907		rdata->mr = NULL;
2908	}
2909#endif
2910	if (rdata->cfile)
2911		cifsFileInfo_put(rdata->cfile);
2912
2913	kfree(rdata);
2914}
2915
2916static int
2917cifs_read_allocate_pages(struct cifs_readdata *rdata, unsigned int nr_pages)
2918{
2919	int rc = 0;
2920	struct page *page;
2921	unsigned int i;
2922
2923	for (i = 0; i < nr_pages; i++) {
2924		page = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2925		if (!page) {
2926			rc = -ENOMEM;
2927			break;
2928		}
2929		rdata->pages[i] = page;
2930	}
2931
2932	if (rc) {
2933		for (i = 0; i < nr_pages; i++) {
2934			put_page(rdata->pages[i]);
2935			rdata->pages[i] = NULL;
2936		}
2937	}
2938	return rc;
2939}
2940
2941static void
2942cifs_uncached_readdata_release(struct kref *refcount)
2943{
 
2944	struct cifs_readdata *rdata = container_of(refcount,
2945					struct cifs_readdata, refcount);
2946	unsigned int i;
2947
2948	kref_put(&rdata->ctx->refcount, cifs_aio_ctx_release);
2949	for (i = 0; i < rdata->nr_pages; i++) {
2950		put_page(rdata->pages[i]);
2951		rdata->pages[i] = NULL;
2952	}
2953	cifs_readdata_release(refcount);
2954}
2955
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2956/**
2957 * cifs_readdata_to_iov - copy data from pages in response to an iovec
2958 * @rdata:	the readdata response with list of pages holding data
2959 * @iter:	destination for our data
 
 
 
2960 *
2961 * This function copies data from a list of pages in a readdata response into
2962 * an array of iovecs. It will first calculate where the data should go
2963 * based on the info in the readdata and then copy the data into that spot.
2964 */
2965static int
2966cifs_readdata_to_iov(struct cifs_readdata *rdata, struct iov_iter *iter)
 
2967{
2968	size_t remaining = rdata->got_bytes;
2969	unsigned int i;
2970
2971	for (i = 0; i < rdata->nr_pages; i++) {
2972		struct page *page = rdata->pages[i];
2973		size_t copy = min_t(size_t, remaining, PAGE_SIZE);
2974		size_t written;
2975
2976		if (unlikely(iter->type & ITER_PIPE)) {
2977			void *addr = kmap_atomic(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2978
2979			written = copy_to_iter(addr, copy, iter);
2980			kunmap_atomic(addr);
2981		} else
2982			written = copy_page_to_iter(page, 0, copy, iter);
2983		remaining -= written;
2984		if (written < copy && iov_iter_count(iter) > 0)
2985			break;
2986	}
2987	return remaining ? -EFAULT : 0;
2988}
2989
2990static void collect_uncached_read_data(struct cifs_aio_ctx *ctx);
 
2991
2992static void
2993cifs_uncached_readv_complete(struct work_struct *work)
2994{
2995	struct cifs_readdata *rdata = container_of(work,
2996						struct cifs_readdata, work);
2997
 
 
 
 
 
 
 
 
2998	complete(&rdata->done);
2999	collect_uncached_read_data(rdata->ctx);
3000	/* the below call can possibly free the last ref to aio ctx */
3001	kref_put(&rdata->refcount, cifs_uncached_readdata_release);
3002}
3003
3004static int
3005uncached_fill_pages(struct TCP_Server_Info *server,
3006		    struct cifs_readdata *rdata, struct iov_iter *iter,
3007		    unsigned int len)
3008{
3009	int result = 0;
3010	unsigned int i;
3011	unsigned int nr_pages = rdata->nr_pages;
3012
3013	rdata->got_bytes = 0;
3014	rdata->tailsz = PAGE_SIZE;
3015	for (i = 0; i < nr_pages; i++) {
3016		struct page *page = rdata->pages[i];
3017		size_t n;
3018
3019		if (len <= 0) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3020			/* no need to hold page hostage */
3021			rdata->pages[i] = NULL;
3022			rdata->nr_pages--;
3023			put_page(page);
3024			continue;
3025		}
3026		n = len;
3027		if (len >= PAGE_SIZE) {
3028			/* enough data to fill the page */
3029			n = PAGE_SIZE;
3030			len -= n;
3031		} else {
3032			zero_user(page, len, PAGE_SIZE - len);
3033			rdata->tailsz = len;
3034			len = 0;
3035		}
3036		if (iter)
3037			result = copy_page_from_iter(page, 0, n, iter);
3038#ifdef CONFIG_CIFS_SMB_DIRECT
3039		else if (rdata->mr)
3040			result = n;
3041#endif
3042		else
3043			result = cifs_read_page_from_socket(server, page, n);
3044		if (result < 0)
3045			break;
3046
3047		rdata->got_bytes += result;
3048	}
3049
3050	return rdata->got_bytes > 0 && result != -ECONNABORTED ?
3051						rdata->got_bytes : result;
3052}
3053
3054static int
3055cifs_uncached_read_into_pages(struct TCP_Server_Info *server,
3056			      struct cifs_readdata *rdata, unsigned int len)
3057{
3058	return uncached_fill_pages(server, rdata, NULL, len);
3059}
 
 
 
 
 
 
 
 
 
3060
3061static int
3062cifs_uncached_copy_into_pages(struct TCP_Server_Info *server,
3063			      struct cifs_readdata *rdata,
3064			      struct iov_iter *iter)
3065{
3066	return uncached_fill_pages(server, rdata, iter, iter->count);
3067}
3068
3069static int
3070cifs_send_async_read(loff_t offset, size_t len, struct cifsFileInfo *open_file,
3071		     struct cifs_sb_info *cifs_sb, struct list_head *rdata_list,
3072		     struct cifs_aio_ctx *ctx)
3073{
3074	struct cifs_readdata *rdata;
3075	unsigned int npages, rsize, credits;
3076	size_t cur_len;
3077	int rc;
3078	pid_t pid;
3079	struct TCP_Server_Info *server;
3080
3081	server = tlink_tcon(open_file->tlink)->ses->server;
 
 
 
3082
3083	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3084		pid = open_file->pid;
3085	else
3086		pid = current->tgid;
3087
3088	do {
3089		rc = server->ops->wait_mtu_credits(server, cifs_sb->rsize,
3090						   &rsize, &credits);
3091		if (rc)
3092			break;
3093
3094		cur_len = min_t(const size_t, len, rsize);
 
3095		npages = DIV_ROUND_UP(cur_len, PAGE_SIZE);
3096
3097		/* allocate a readdata struct */
3098		rdata = cifs_readdata_alloc(npages,
3099					    cifs_uncached_readv_complete);
3100		if (!rdata) {
3101			add_credits_and_wake_if(server, credits, 0);
3102			rc = -ENOMEM;
3103			break;
3104		}
3105
3106		rc = cifs_read_allocate_pages(rdata, npages);
3107		if (rc)
3108			goto error;
3109
3110		rdata->cfile = cifsFileInfo_get(open_file);
3111		rdata->nr_pages = npages;
3112		rdata->offset = offset;
3113		rdata->bytes = cur_len;
3114		rdata->pid = pid;
3115		rdata->pagesz = PAGE_SIZE;
3116		rdata->read_into_pages = cifs_uncached_read_into_pages;
3117		rdata->copy_into_pages = cifs_uncached_copy_into_pages;
3118		rdata->credits = credits;
3119		rdata->ctx = ctx;
3120		kref_get(&ctx->refcount);
3121
3122		if (!rdata->cfile->invalidHandle ||
3123		    !(rc = cifs_reopen_file(rdata->cfile, true)))
3124			rc = server->ops->async_readv(rdata);
3125error:
3126		if (rc) {
3127			add_credits_and_wake_if(server, rdata->credits, 0);
3128			kref_put(&rdata->refcount,
3129				 cifs_uncached_readdata_release);
3130			if (rc == -EAGAIN)
3131				continue;
3132			break;
3133		}
3134
3135		list_add_tail(&rdata->list, rdata_list);
3136		offset += cur_len;
3137		len -= cur_len;
3138	} while (len > 0);
3139
3140	return rc;
3141}
3142
3143static void
3144collect_uncached_read_data(struct cifs_aio_ctx *ctx)
3145{
3146	struct cifs_readdata *rdata, *tmp;
3147	struct iov_iter *to = &ctx->iter;
3148	struct cifs_sb_info *cifs_sb;
3149	struct cifs_tcon *tcon;
3150	unsigned int i;
3151	int rc;
3152
3153	tcon = tlink_tcon(ctx->cfile->tlink);
3154	cifs_sb = CIFS_SB(ctx->cfile->dentry->d_sb);
3155
3156	mutex_lock(&ctx->aio_mutex);
3157
3158	if (list_empty(&ctx->list)) {
3159		mutex_unlock(&ctx->aio_mutex);
3160		return;
3161	}
3162
3163	rc = ctx->rc;
3164	/* the loop below should proceed in the order of increasing offsets */
3165again:
3166	list_for_each_entry_safe(rdata, tmp, &ctx->list, list) {
3167		if (!rc) {
3168			if (!try_wait_for_completion(&rdata->done)) {
3169				mutex_unlock(&ctx->aio_mutex);
3170				return;
3171			}
3172
3173			if (rdata->result == -EAGAIN) {
3174				/* resend call if it's a retryable error */
3175				struct list_head tmp_list;
3176				unsigned int got_bytes = rdata->got_bytes;
3177
3178				list_del_init(&rdata->list);
3179				INIT_LIST_HEAD(&tmp_list);
3180
3181				/*
3182				 * Got a part of data and then reconnect has
3183				 * happened -- fill the buffer and continue
3184				 * reading.
3185				 */
3186				if (got_bytes && got_bytes < rdata->bytes) {
3187					rc = cifs_readdata_to_iov(rdata, to);
3188					if (rc) {
3189						kref_put(&rdata->refcount,
3190						cifs_uncached_readdata_release);
3191						continue;
3192					}
3193				}
3194
3195				rc = cifs_send_async_read(
3196						rdata->offset + got_bytes,
3197						rdata->bytes - got_bytes,
3198						rdata->cfile, cifs_sb,
3199						&tmp_list, ctx);
3200
3201				list_splice(&tmp_list, &ctx->list);
3202
3203				kref_put(&rdata->refcount,
3204					 cifs_uncached_readdata_release);
3205				goto again;
3206			} else if (rdata->result)
3207				rc = rdata->result;
3208			else
3209				rc = cifs_readdata_to_iov(rdata, to);
 
 
 
 
3210
3211			/* if there was a short read -- discard anything left */
3212			if (rdata->got_bytes && rdata->got_bytes < rdata->bytes)
3213				rc = -ENODATA;
 
 
3214		}
3215		list_del_init(&rdata->list);
3216		kref_put(&rdata->refcount, cifs_uncached_readdata_release);
3217	}
3218
3219	for (i = 0; i < ctx->npages; i++) {
3220		if (ctx->should_dirty)
3221			set_page_dirty(ctx->bv[i].bv_page);
3222		put_page(ctx->bv[i].bv_page);
3223	}
3224
3225	ctx->total_len = ctx->len - iov_iter_count(to);
3226
3227	cifs_stats_bytes_read(tcon, ctx->total_len);
3228
3229	/* mask nodata case */
3230	if (rc == -ENODATA)
3231		rc = 0;
3232
3233	ctx->rc = (rc == 0) ? ctx->total_len : rc;
3234
3235	mutex_unlock(&ctx->aio_mutex);
3236
3237	if (ctx->iocb && ctx->iocb->ki_complete)
3238		ctx->iocb->ki_complete(ctx->iocb, ctx->rc, 0);
3239	else
3240		complete(&ctx->done);
3241}
3242
3243ssize_t cifs_user_readv(struct kiocb *iocb, struct iov_iter *to)
 
3244{
3245	struct file *file = iocb->ki_filp;
3246	ssize_t rc;
3247	size_t len;
3248	ssize_t total_read = 0;
3249	loff_t offset = iocb->ki_pos;
3250	struct cifs_sb_info *cifs_sb;
3251	struct cifs_tcon *tcon;
3252	struct cifsFileInfo *cfile;
3253	struct cifs_aio_ctx *ctx;
3254
3255	len = iov_iter_count(to);
3256	if (!len)
3257		return 0;
3258
3259	cifs_sb = CIFS_FILE_SB(file);
3260	cfile = file->private_data;
3261	tcon = tlink_tcon(cfile->tlink);
3262
3263	if (!tcon->ses->server->ops->async_readv)
3264		return -ENOSYS;
3265
3266	if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3267		cifs_dbg(FYI, "attempting read on write only file instance\n");
3268
3269	ctx = cifs_aio_ctx_alloc();
3270	if (!ctx)
3271		return -ENOMEM;
3272
3273	ctx->cfile = cifsFileInfo_get(cfile);
3274
3275	if (!is_sync_kiocb(iocb))
3276		ctx->iocb = iocb;
3277
3278	if (to->type == ITER_IOVEC)
3279		ctx->should_dirty = true;
3280
3281	rc = setup_aio_ctx_iter(ctx, to, READ);
3282	if (rc) {
3283		kref_put(&ctx->refcount, cifs_aio_ctx_release);
3284		return rc;
3285	}
3286
3287	len = ctx->len;
3288
3289	/* grab a lock here due to read response handlers can access ctx */
3290	mutex_lock(&ctx->aio_mutex);
3291
3292	rc = cifs_send_async_read(offset, len, cfile, cifs_sb, &ctx->list, ctx);
3293
3294	/* if at least one read request send succeeded, then reset rc */
3295	if (!list_empty(&ctx->list))
3296		rc = 0;
3297
3298	mutex_unlock(&ctx->aio_mutex);
3299
3300	if (rc) {
3301		kref_put(&ctx->refcount, cifs_aio_ctx_release);
3302		return rc;
3303	}
3304
3305	if (!is_sync_kiocb(iocb)) {
3306		kref_put(&ctx->refcount, cifs_aio_ctx_release);
3307		return -EIOCBQUEUED;
3308	}
3309
3310	rc = wait_for_completion_killable(&ctx->done);
3311	if (rc) {
3312		mutex_lock(&ctx->aio_mutex);
3313		ctx->rc = rc = -EINTR;
3314		total_read = ctx->total_len;
3315		mutex_unlock(&ctx->aio_mutex);
3316	} else {
3317		rc = ctx->rc;
3318		total_read = ctx->total_len;
3319	}
3320
3321	kref_put(&ctx->refcount, cifs_aio_ctx_release);
3322
3323	if (total_read) {
3324		iocb->ki_pos += total_read;
3325		return total_read;
3326	}
3327	return rc;
3328}
3329
3330ssize_t
3331cifs_strict_readv(struct kiocb *iocb, struct iov_iter *to)
3332{
3333	struct inode *inode = file_inode(iocb->ki_filp);
3334	struct cifsInodeInfo *cinode = CIFS_I(inode);
3335	struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
3336	struct cifsFileInfo *cfile = (struct cifsFileInfo *)
3337						iocb->ki_filp->private_data;
3338	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
3339	int rc = -EACCES;
3340
3341	/*
3342	 * In strict cache mode we need to read from the server all the time
3343	 * if we don't have level II oplock because the server can delay mtime
3344	 * change - so we can't make a decision about inode invalidating.
3345	 * And we can also fail with pagereading if there are mandatory locks
3346	 * on pages affected by this read but not on the region from pos to
3347	 * pos+len-1.
3348	 */
3349	if (!CIFS_CACHE_READ(cinode))
3350		return cifs_user_readv(iocb, to);
3351
3352	if (cap_unix(tcon->ses) &&
3353	    (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
3354	    ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
3355		return generic_file_read_iter(iocb, to);
3356
3357	/*
3358	 * We need to hold the sem to be sure nobody modifies lock list
3359	 * with a brlock that prevents reading.
3360	 */
3361	down_read(&cinode->lock_sem);
3362	if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(to),
3363				     tcon->ses->server->vals->shared_lock_type,
3364				     NULL, CIFS_READ_OP))
3365		rc = generic_file_read_iter(iocb, to);
3366	up_read(&cinode->lock_sem);
3367	return rc;
3368}
3369
3370static ssize_t
3371cifs_read(struct file *file, char *read_data, size_t read_size, loff_t *offset)
3372{
3373	int rc = -EACCES;
3374	unsigned int bytes_read = 0;
3375	unsigned int total_read;
3376	unsigned int current_read_size;
3377	unsigned int rsize;
3378	struct cifs_sb_info *cifs_sb;
3379	struct cifs_tcon *tcon;
3380	struct TCP_Server_Info *server;
3381	unsigned int xid;
3382	char *cur_offset;
3383	struct cifsFileInfo *open_file;
3384	struct cifs_io_parms io_parms;
3385	int buf_type = CIFS_NO_BUFFER;
3386	__u32 pid;
3387
3388	xid = get_xid();
3389	cifs_sb = CIFS_FILE_SB(file);
3390
3391	/* FIXME: set up handlers for larger reads and/or convert to async */
3392	rsize = min_t(unsigned int, cifs_sb->rsize, CIFSMaxBufSize);
3393
3394	if (file->private_data == NULL) {
3395		rc = -EBADF;
3396		free_xid(xid);
3397		return rc;
3398	}
3399	open_file = file->private_data;
3400	tcon = tlink_tcon(open_file->tlink);
3401	server = tcon->ses->server;
3402
3403	if (!server->ops->sync_read) {
3404		free_xid(xid);
3405		return -ENOSYS;
3406	}
3407
3408	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3409		pid = open_file->pid;
3410	else
3411		pid = current->tgid;
3412
3413	if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3414		cifs_dbg(FYI, "attempting read on write only file instance\n");
3415
3416	for (total_read = 0, cur_offset = read_data; read_size > total_read;
3417	     total_read += bytes_read, cur_offset += bytes_read) {
3418		do {
3419			current_read_size = min_t(uint, read_size - total_read,
3420						  rsize);
3421			/*
3422			 * For windows me and 9x we do not want to request more
3423			 * than it negotiated since it will refuse the read
3424			 * then.
3425			 */
3426			if ((tcon->ses) && !(tcon->ses->capabilities &
3427				tcon->ses->server->vals->cap_large_files)) {
3428				current_read_size = min_t(uint,
3429					current_read_size, CIFSMaxBufSize);
3430			}
3431			if (open_file->invalidHandle) {
3432				rc = cifs_reopen_file(open_file, true);
3433				if (rc != 0)
3434					break;
3435			}
 
3436			io_parms.pid = pid;
3437			io_parms.tcon = tcon;
3438			io_parms.offset = *offset;
3439			io_parms.length = current_read_size;
3440			rc = server->ops->sync_read(xid, &open_file->fid, &io_parms,
3441						    &bytes_read, &cur_offset,
3442						    &buf_type);
3443		} while (rc == -EAGAIN);
3444
3445		if (rc || (bytes_read == 0)) {
3446			if (total_read) {
3447				break;
3448			} else {
3449				free_xid(xid);
3450				return rc;
3451			}
3452		} else {
3453			cifs_stats_bytes_read(tcon, total_read);
3454			*offset += bytes_read;
3455		}
3456	}
3457	free_xid(xid);
3458	return total_read;
3459}
3460
3461/*
3462 * If the page is mmap'ed into a process' page tables, then we need to make
3463 * sure that it doesn't change while being written back.
3464 */
3465static vm_fault_t
3466cifs_page_mkwrite(struct vm_fault *vmf)
3467{
3468	struct page *page = vmf->page;
3469
3470	lock_page(page);
3471	return VM_FAULT_LOCKED;
3472}
3473
3474static const struct vm_operations_struct cifs_file_vm_ops = {
3475	.fault = filemap_fault,
3476	.map_pages = filemap_map_pages,
3477	.page_mkwrite = cifs_page_mkwrite,
3478};
3479
3480int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
3481{
3482	int xid, rc = 0;
3483	struct inode *inode = file_inode(file);
3484
3485	xid = get_xid();
3486
3487	if (!CIFS_CACHE_READ(CIFS_I(inode)))
3488		rc = cifs_zap_mapping(inode);
3489	if (!rc)
3490		rc = generic_file_mmap(file, vma);
3491	if (!rc)
3492		vma->vm_ops = &cifs_file_vm_ops;
3493
3494	free_xid(xid);
 
 
 
3495	return rc;
3496}
3497
3498int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
3499{
3500	int rc, xid;
3501
3502	xid = get_xid();
3503
3504	rc = cifs_revalidate_file(file);
3505	if (rc)
3506		cifs_dbg(FYI, "Validation prior to mmap failed, error=%d\n",
3507			 rc);
3508	if (!rc)
3509		rc = generic_file_mmap(file, vma);
3510	if (!rc)
 
3511		vma->vm_ops = &cifs_file_vm_ops;
3512
3513	free_xid(xid);
3514	return rc;
3515}
3516
3517static void
3518cifs_readv_complete(struct work_struct *work)
3519{
3520	unsigned int i, got_bytes;
3521	struct cifs_readdata *rdata = container_of(work,
3522						struct cifs_readdata, work);
 
3523
3524	got_bytes = rdata->got_bytes;
3525	for (i = 0; i < rdata->nr_pages; i++) {
3526		struct page *page = rdata->pages[i];
3527
3528		lru_cache_add_file(page);
3529
3530		if (rdata->result == 0 ||
3531		    (rdata->result == -EAGAIN && got_bytes)) {
3532			flush_dcache_page(page);
3533			SetPageUptodate(page);
3534		}
3535
3536		unlock_page(page);
3537
3538		if (rdata->result == 0 ||
3539		    (rdata->result == -EAGAIN && got_bytes))
3540			cifs_readpage_to_fscache(rdata->mapping->host, page);
3541
3542		got_bytes -= min_t(unsigned int, PAGE_SIZE, got_bytes);
3543
3544		put_page(page);
3545		rdata->pages[i] = NULL;
3546	}
3547	kref_put(&rdata->refcount, cifs_readdata_release);
3548}
3549
3550static int
3551readpages_fill_pages(struct TCP_Server_Info *server,
3552		     struct cifs_readdata *rdata, struct iov_iter *iter,
3553		     unsigned int len)
3554{
3555	int result = 0;
3556	unsigned int i;
3557	u64 eof;
3558	pgoff_t eof_index;
3559	unsigned int nr_pages = rdata->nr_pages;
3560
3561	/* determine the eof that the server (probably) has */
3562	eof = CIFS_I(rdata->mapping->host)->server_eof;
3563	eof_index = eof ? (eof - 1) >> PAGE_SHIFT : 0;
3564	cifs_dbg(FYI, "eof=%llu eof_index=%lu\n", eof, eof_index);
3565
3566	rdata->got_bytes = 0;
3567	rdata->tailsz = PAGE_SIZE;
3568	for (i = 0; i < nr_pages; i++) {
3569		struct page *page = rdata->pages[i];
3570		size_t n = PAGE_SIZE;
3571
3572		if (len >= PAGE_SIZE) {
3573			len -= PAGE_SIZE;
3574		} else if (len > 0) {
 
 
 
 
 
3575			/* enough for partial page, fill and zero the rest */
3576			zero_user(page, len, PAGE_SIZE - len);
3577			n = rdata->tailsz = len;
3578			len = 0;
 
 
 
 
 
 
 
 
3579		} else if (page->index > eof_index) {
3580			/*
3581			 * The VFS will not try to do readahead past the
3582			 * i_size, but it's possible that we have outstanding
3583			 * writes with gaps in the middle and the i_size hasn't
3584			 * caught up yet. Populate those with zeroed out pages
3585			 * to prevent the VFS from repeatedly attempting to
3586			 * fill them until the writes are flushed.
3587			 */
3588			zero_user(page, 0, PAGE_SIZE);
 
3589			lru_cache_add_file(page);
3590			flush_dcache_page(page);
3591			SetPageUptodate(page);
3592			unlock_page(page);
3593			put_page(page);
3594			rdata->pages[i] = NULL;
3595			rdata->nr_pages--;
3596			continue;
3597		} else {
3598			/* no need to hold page hostage */
 
3599			lru_cache_add_file(page);
3600			unlock_page(page);
3601			put_page(page);
3602			rdata->pages[i] = NULL;
3603			rdata->nr_pages--;
3604			continue;
3605		}
3606
3607		if (iter)
3608			result = copy_page_from_iter(page, 0, n, iter);
3609#ifdef CONFIG_CIFS_SMB_DIRECT
3610		else if (rdata->mr)
3611			result = n;
3612#endif
3613		else
3614			result = cifs_read_page_from_socket(server, page, n);
3615		if (result < 0)
3616			break;
3617
3618		rdata->got_bytes += result;
3619	}
3620
3621	return rdata->got_bytes > 0 && result != -ECONNABORTED ?
3622						rdata->got_bytes : result;
3623}
3624
3625static int
3626cifs_readpages_read_into_pages(struct TCP_Server_Info *server,
3627			       struct cifs_readdata *rdata, unsigned int len)
3628{
3629	return readpages_fill_pages(server, rdata, NULL, len);
3630}
3631
3632static int
3633cifs_readpages_copy_into_pages(struct TCP_Server_Info *server,
3634			       struct cifs_readdata *rdata,
3635			       struct iov_iter *iter)
3636{
3637	return readpages_fill_pages(server, rdata, iter, iter->count);
3638}
3639
3640static int
3641readpages_get_pages(struct address_space *mapping, struct list_head *page_list,
3642		    unsigned int rsize, struct list_head *tmplist,
3643		    unsigned int *nr_pages, loff_t *offset, unsigned int *bytes)
3644{
3645	struct page *page, *tpage;
3646	unsigned int expected_index;
3647	int rc;
3648	gfp_t gfp = readahead_gfp_mask(mapping);
3649
3650	INIT_LIST_HEAD(tmplist);
3651
3652	page = list_entry(page_list->prev, struct page, lru);
3653
3654	/*
3655	 * Lock the page and put it in the cache. Since no one else
3656	 * should have access to this page, we're safe to simply set
3657	 * PG_locked without checking it first.
3658	 */
3659	__SetPageLocked(page);
3660	rc = add_to_page_cache_locked(page, mapping,
3661				      page->index, gfp);
3662
3663	/* give up if we can't stick it in the cache */
3664	if (rc) {
3665		__ClearPageLocked(page);
3666		return rc;
3667	}
3668
3669	/* move first page to the tmplist */
3670	*offset = (loff_t)page->index << PAGE_SHIFT;
3671	*bytes = PAGE_SIZE;
3672	*nr_pages = 1;
3673	list_move_tail(&page->lru, tmplist);
3674
3675	/* now try and add more pages onto the request */
3676	expected_index = page->index + 1;
3677	list_for_each_entry_safe_reverse(page, tpage, page_list, lru) {
3678		/* discontinuity ? */
3679		if (page->index != expected_index)
3680			break;
3681
3682		/* would this page push the read over the rsize? */
3683		if (*bytes + PAGE_SIZE > rsize)
3684			break;
3685
3686		__SetPageLocked(page);
3687		if (add_to_page_cache_locked(page, mapping, page->index, gfp)) {
3688			__ClearPageLocked(page);
3689			break;
3690		}
3691		list_move_tail(&page->lru, tmplist);
3692		(*bytes) += PAGE_SIZE;
3693		expected_index++;
3694		(*nr_pages)++;
3695	}
3696	return rc;
3697}
3698
3699static int cifs_readpages(struct file *file, struct address_space *mapping,
3700	struct list_head *page_list, unsigned num_pages)
3701{
3702	int rc;
3703	struct list_head tmplist;
3704	struct cifsFileInfo *open_file = file->private_data;
3705	struct cifs_sb_info *cifs_sb = CIFS_FILE_SB(file);
3706	struct TCP_Server_Info *server;
3707	pid_t pid;
3708
3709	/*
 
 
 
 
 
 
 
 
 
3710	 * Reads as many pages as possible from fscache. Returns -ENOBUFS
3711	 * immediately if the cookie is negative
3712	 *
3713	 * After this point, every page in the list might have PG_fscache set,
3714	 * so we will need to clean that up off of every page we don't use.
3715	 */
3716	rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
3717					 &num_pages);
3718	if (rc == 0)
3719		return rc;
3720
3721	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3722		pid = open_file->pid;
3723	else
3724		pid = current->tgid;
3725
3726	rc = 0;
3727	server = tlink_tcon(open_file->tlink)->ses->server;
3728
3729	cifs_dbg(FYI, "%s: file=%p mapping=%p num_pages=%u\n",
3730		 __func__, file, mapping, num_pages);
3731
3732	/*
3733	 * Start with the page at end of list and move it to private
3734	 * list. Do the same with any following pages until we hit
3735	 * the rsize limit, hit an index discontinuity, or run out of
3736	 * pages. Issue the async read and then start the loop again
3737	 * until the list is empty.
3738	 *
3739	 * Note that list order is important. The page_list is in
3740	 * the order of declining indexes. When we put the pages in
3741	 * the rdata->pages, then we want them in increasing order.
3742	 */
3743	while (!list_empty(page_list)) {
3744		unsigned int i, nr_pages, bytes, rsize;
 
 
3745		loff_t offset;
3746		struct page *page, *tpage;
3747		struct cifs_readdata *rdata;
3748		unsigned credits;
3749
3750		rc = server->ops->wait_mtu_credits(server, cifs_sb->rsize,
3751						   &rsize, &credits);
3752		if (rc)
3753			break;
3754
3755		/*
3756		 * Give up immediately if rsize is too small to read an entire
3757		 * page. The VFS will fall back to readpage. We should never
3758		 * reach this point however since we set ra_pages to 0 when the
3759		 * rsize is smaller than a cache page.
3760		 */
3761		if (unlikely(rsize < PAGE_SIZE)) {
3762			add_credits_and_wake_if(server, credits, 0);
3763			return 0;
3764		}
3765
3766		rc = readpages_get_pages(mapping, page_list, rsize, &tmplist,
3767					 &nr_pages, &offset, &bytes);
3768		if (rc) {
3769			add_credits_and_wake_if(server, credits, 0);
3770			break;
3771		}
3772
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3773		rdata = cifs_readdata_alloc(nr_pages, cifs_readv_complete);
3774		if (!rdata) {
3775			/* best to give up if we're out of mem */
3776			list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3777				list_del(&page->lru);
3778				lru_cache_add_file(page);
3779				unlock_page(page);
3780				put_page(page);
3781			}
3782			rc = -ENOMEM;
3783			add_credits_and_wake_if(server, credits, 0);
3784			break;
3785		}
3786
3787		rdata->cfile = cifsFileInfo_get(open_file);
3788		rdata->mapping = mapping;
3789		rdata->offset = offset;
3790		rdata->bytes = bytes;
3791		rdata->pid = pid;
3792		rdata->pagesz = PAGE_SIZE;
3793		rdata->read_into_pages = cifs_readpages_read_into_pages;
3794		rdata->copy_into_pages = cifs_readpages_copy_into_pages;
3795		rdata->credits = credits;
3796
3797		list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3798			list_del(&page->lru);
3799			rdata->pages[rdata->nr_pages++] = page;
3800		}
3801
3802		if (!rdata->cfile->invalidHandle ||
3803		    !(rc = cifs_reopen_file(rdata->cfile, true)))
3804			rc = server->ops->async_readv(rdata);
3805		if (rc) {
3806			add_credits_and_wake_if(server, rdata->credits, 0);
3807			for (i = 0; i < rdata->nr_pages; i++) {
3808				page = rdata->pages[i];
3809				lru_cache_add_file(page);
3810				unlock_page(page);
3811				put_page(page);
3812			}
3813			/* Fallback to the readpage in error/reconnect cases */
3814			kref_put(&rdata->refcount, cifs_readdata_release);
3815			break;
3816		}
3817
3818		kref_put(&rdata->refcount, cifs_readdata_release);
3819	}
3820
3821	/* Any pages that have been shown to fscache but didn't get added to
3822	 * the pagecache must be uncached before they get returned to the
3823	 * allocator.
3824	 */
3825	cifs_fscache_readpages_cancel(mapping->host, page_list);
3826	return rc;
3827}
3828
3829/*
3830 * cifs_readpage_worker must be called with the page pinned
3831 */
3832static int cifs_readpage_worker(struct file *file, struct page *page,
3833	loff_t *poffset)
3834{
3835	char *read_data;
3836	int rc;
3837
3838	/* Is the page cached? */
3839	rc = cifs_readpage_from_fscache(file_inode(file), page);
3840	if (rc == 0)
3841		goto read_complete;
3842
 
3843	read_data = kmap(page);
3844	/* for reads over a certain size could initiate async read ahead */
3845
3846	rc = cifs_read(file, read_data, PAGE_SIZE, poffset);
3847
3848	if (rc < 0)
3849		goto io_error;
3850	else
3851		cifs_dbg(FYI, "Bytes read %d\n", rc);
3852
3853	file_inode(file)->i_atime =
3854		current_time(file_inode(file));
3855
3856	if (PAGE_SIZE > rc)
3857		memset(read_data + rc, 0, PAGE_SIZE - rc);
3858
3859	flush_dcache_page(page);
3860	SetPageUptodate(page);
3861
3862	/* send this page to the cache */
3863	cifs_readpage_to_fscache(file_inode(file), page);
3864
3865	rc = 0;
3866
3867io_error:
3868	kunmap(page);
3869	unlock_page(page);
3870
3871read_complete:
3872	return rc;
3873}
3874
3875static int cifs_readpage(struct file *file, struct page *page)
3876{
3877	loff_t offset = (loff_t)page->index << PAGE_SHIFT;
3878	int rc = -EACCES;
3879	unsigned int xid;
3880
3881	xid = get_xid();
3882
3883	if (file->private_data == NULL) {
3884		rc = -EBADF;
3885		free_xid(xid);
3886		return rc;
3887	}
3888
3889	cifs_dbg(FYI, "readpage %p at offset %d 0x%x\n",
3890		 page, (int)offset, (int)offset);
3891
3892	rc = cifs_readpage_worker(file, page, &offset);
3893
3894	free_xid(xid);
 
 
3895	return rc;
3896}
3897
3898static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
3899{
3900	struct cifsFileInfo *open_file;
3901	struct cifs_tcon *tcon =
3902		cifs_sb_master_tcon(CIFS_SB(cifs_inode->vfs_inode.i_sb));
3903
3904	spin_lock(&tcon->open_file_lock);
3905	list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
3906		if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
3907			spin_unlock(&tcon->open_file_lock);
3908			return 1;
3909		}
3910	}
3911	spin_unlock(&tcon->open_file_lock);
3912	return 0;
3913}
3914
3915/* We do not want to update the file size from server for inodes
3916   open for write - to avoid races with writepage extending
3917   the file - in the future we could consider allowing
3918   refreshing the inode only on increases in the file size
3919   but this is tricky to do without racing with writebehind
3920   page caching in the current Linux kernel design */
3921bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
3922{
3923	if (!cifsInode)
3924		return true;
3925
3926	if (is_inode_writable(cifsInode)) {
3927		/* This inode is open for write at least once */
3928		struct cifs_sb_info *cifs_sb;
3929
3930		cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
3931		if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
3932			/* since no page cache to corrupt on directio
3933			we can change size safely */
3934			return true;
3935		}
3936
3937		if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
3938			return true;
3939
3940		return false;
3941	} else
3942		return true;
3943}
3944
3945static int cifs_write_begin(struct file *file, struct address_space *mapping,
3946			loff_t pos, unsigned len, unsigned flags,
3947			struct page **pagep, void **fsdata)
3948{
3949	int oncethru = 0;
3950	pgoff_t index = pos >> PAGE_SHIFT;
3951	loff_t offset = pos & (PAGE_SIZE - 1);
3952	loff_t page_start = pos & PAGE_MASK;
3953	loff_t i_size;
3954	struct page *page;
3955	int rc = 0;
3956
3957	cifs_dbg(FYI, "write_begin from %lld len %d\n", (long long)pos, len);
3958
3959start:
3960	page = grab_cache_page_write_begin(mapping, index, flags);
3961	if (!page) {
3962		rc = -ENOMEM;
3963		goto out;
3964	}
3965
3966	if (PageUptodate(page))
3967		goto out;
3968
3969	/*
3970	 * If we write a full page it will be up to date, no need to read from
3971	 * the server. If the write is short, we'll end up doing a sync write
3972	 * instead.
3973	 */
3974	if (len == PAGE_SIZE)
3975		goto out;
3976
3977	/*
3978	 * optimize away the read when we have an oplock, and we're not
3979	 * expecting to use any of the data we'd be reading in. That
3980	 * is, when the page lies beyond the EOF, or straddles the EOF
3981	 * and the write will cover all of the existing data.
3982	 */
3983	if (CIFS_CACHE_READ(CIFS_I(mapping->host))) {
3984		i_size = i_size_read(mapping->host);
3985		if (page_start >= i_size ||
3986		    (offset == 0 && (pos + len) >= i_size)) {
3987			zero_user_segments(page, 0, offset,
3988					   offset + len,
3989					   PAGE_SIZE);
3990			/*
3991			 * PageChecked means that the parts of the page
3992			 * to which we're not writing are considered up
3993			 * to date. Once the data is copied to the
3994			 * page, it can be set uptodate.
3995			 */
3996			SetPageChecked(page);
3997			goto out;
3998		}
3999	}
4000
4001	if ((file->f_flags & O_ACCMODE) != O_WRONLY && !oncethru) {
4002		/*
4003		 * might as well read a page, it is fast enough. If we get
4004		 * an error, we don't need to return it. cifs_write_end will
4005		 * do a sync write instead since PG_uptodate isn't set.
4006		 */
4007		cifs_readpage_worker(file, page, &page_start);
4008		put_page(page);
4009		oncethru = 1;
4010		goto start;
4011	} else {
4012		/* we could try using another file handle if there is one -
4013		   but how would we lock it to prevent close of that handle
4014		   racing with this read? In any case
4015		   this will be written out by write_end so is fine */
4016	}
4017out:
4018	*pagep = page;
4019	return rc;
4020}
4021
4022static int cifs_release_page(struct page *page, gfp_t gfp)
4023{
4024	if (PagePrivate(page))
4025		return 0;
4026
4027	return cifs_fscache_release_page(page, gfp);
4028}
4029
4030static void cifs_invalidate_page(struct page *page, unsigned int offset,
4031				 unsigned int length)
4032{
4033	struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
4034
4035	if (offset == 0 && length == PAGE_SIZE)
4036		cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
4037}
4038
4039static int cifs_launder_page(struct page *page)
4040{
4041	int rc = 0;
4042	loff_t range_start = page_offset(page);
4043	loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1);
4044	struct writeback_control wbc = {
4045		.sync_mode = WB_SYNC_ALL,
4046		.nr_to_write = 0,
4047		.range_start = range_start,
4048		.range_end = range_end,
4049	};
4050
4051	cifs_dbg(FYI, "Launder page: %p\n", page);
4052
4053	if (clear_page_dirty_for_io(page))
4054		rc = cifs_writepage_locked(page, &wbc);
4055
4056	cifs_fscache_invalidate_page(page, page->mapping->host);
4057	return rc;
4058}
4059
4060void cifs_oplock_break(struct work_struct *work)
4061{
4062	struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
4063						  oplock_break);
4064	struct inode *inode = d_inode(cfile->dentry);
4065	struct cifsInodeInfo *cinode = CIFS_I(inode);
4066	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
4067	struct TCP_Server_Info *server = tcon->ses->server;
4068	int rc = 0;
4069
4070	wait_on_bit(&cinode->flags, CIFS_INODE_PENDING_WRITERS,
4071			TASK_UNINTERRUPTIBLE);
4072
4073	server->ops->downgrade_oplock(server, cinode,
4074		test_bit(CIFS_INODE_DOWNGRADE_OPLOCK_TO_L2, &cinode->flags));
4075
4076	if (!CIFS_CACHE_WRITE(cinode) && CIFS_CACHE_READ(cinode) &&
4077						cifs_has_mand_locks(cinode)) {
4078		cifs_dbg(FYI, "Reset oplock to None for inode=%p due to mand locks\n",
4079			 inode);
4080		cinode->oplock = 0;
4081	}
4082
4083	if (inode && S_ISREG(inode->i_mode)) {
4084		if (CIFS_CACHE_READ(cinode))
4085			break_lease(inode, O_RDONLY);
4086		else
4087			break_lease(inode, O_WRONLY);
4088		rc = filemap_fdatawrite(inode->i_mapping);
4089		if (!CIFS_CACHE_READ(cinode)) {
4090			rc = filemap_fdatawait(inode->i_mapping);
4091			mapping_set_error(inode->i_mapping, rc);
4092			cifs_zap_mapping(inode);
4093		}
4094		cifs_dbg(FYI, "Oplock flush inode %p rc %d\n", inode, rc);
4095	}
4096
4097	rc = cifs_push_locks(cfile);
4098	if (rc)
4099		cifs_dbg(VFS, "Push locks rc = %d\n", rc);
4100
4101	/*
4102	 * releasing stale oplock after recent reconnect of smb session using
4103	 * a now incorrect file handle is not a data integrity issue but do
4104	 * not bother sending an oplock release if session to server still is
4105	 * disconnected since oplock already released by the server
4106	 */
4107	if (!cfile->oplock_break_cancelled) {
4108		rc = tcon->ses->server->ops->oplock_response(tcon, &cfile->fid,
4109							     cinode);
4110		cifs_dbg(FYI, "Oplock release rc = %d\n", rc);
 
 
4111	}
4112	cifs_done_oplock_break(cinode);
4113}
4114
4115/*
4116 * The presence of cifs_direct_io() in the address space ops vector
4117 * allowes open() O_DIRECT flags which would have failed otherwise.
4118 *
4119 * In the non-cached mode (mount with cache=none), we shunt off direct read and write requests
4120 * so this method should never be called.
4121 *
4122 * Direct IO is not yet supported in the cached mode. 
4123 */
4124static ssize_t
4125cifs_direct_io(struct kiocb *iocb, struct iov_iter *iter)
4126{
4127        /*
4128         * FIXME
4129         * Eventually need to support direct IO for non forcedirectio mounts
4130         */
4131        return -EINVAL;
4132}
4133
4134
4135const struct address_space_operations cifs_addr_ops = {
4136	.readpage = cifs_readpage,
4137	.readpages = cifs_readpages,
4138	.writepage = cifs_writepage,
4139	.writepages = cifs_writepages,
4140	.write_begin = cifs_write_begin,
4141	.write_end = cifs_write_end,
4142	.set_page_dirty = __set_page_dirty_nobuffers,
4143	.releasepage = cifs_release_page,
4144	.direct_IO = cifs_direct_io,
4145	.invalidatepage = cifs_invalidate_page,
4146	.launder_page = cifs_launder_page,
4147};
4148
4149/*
4150 * cifs_readpages requires the server to support a buffer large enough to
4151 * contain the header plus one complete page of data.  Otherwise, we need
4152 * to leave cifs_readpages out of the address space operations.
4153 */
4154const struct address_space_operations cifs_addr_ops_smallbuf = {
4155	.readpage = cifs_readpage,
4156	.writepage = cifs_writepage,
4157	.writepages = cifs_writepages,
4158	.write_begin = cifs_write_begin,
4159	.write_end = cifs_write_end,
4160	.set_page_dirty = __set_page_dirty_nobuffers,
4161	.releasepage = cifs_release_page,
4162	.invalidatepage = cifs_invalidate_page,
4163	.launder_page = cifs_launder_page,
4164};
v3.5.6
   1/*
   2 *   fs/cifs/file.c
   3 *
   4 *   vfs operations that deal with files
   5 *
   6 *   Copyright (C) International Business Machines  Corp., 2002,2010
   7 *   Author(s): Steve French (sfrench@us.ibm.com)
   8 *              Jeremy Allison (jra@samba.org)
   9 *
  10 *   This library is free software; you can redistribute it and/or modify
  11 *   it under the terms of the GNU Lesser General Public License as published
  12 *   by the Free Software Foundation; either version 2.1 of the License, or
  13 *   (at your option) any later version.
  14 *
  15 *   This library is distributed in the hope that it will be useful,
  16 *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  17 *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
  18 *   the GNU Lesser General Public License for more details.
  19 *
  20 *   You should have received a copy of the GNU Lesser General Public License
  21 *   along with this library; if not, write to the Free Software
  22 *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  23 */
  24#include <linux/fs.h>
  25#include <linux/backing-dev.h>
  26#include <linux/stat.h>
  27#include <linux/fcntl.h>
  28#include <linux/pagemap.h>
  29#include <linux/pagevec.h>
  30#include <linux/writeback.h>
  31#include <linux/task_io_accounting_ops.h>
  32#include <linux/delay.h>
  33#include <linux/mount.h>
  34#include <linux/slab.h>
  35#include <linux/swap.h>
  36#include <asm/div64.h>
  37#include "cifsfs.h"
  38#include "cifspdu.h"
  39#include "cifsglob.h"
  40#include "cifsproto.h"
  41#include "cifs_unicode.h"
  42#include "cifs_debug.h"
  43#include "cifs_fs_sb.h"
  44#include "fscache.h"
 
  45
  46static inline int cifs_convert_flags(unsigned int flags)
  47{
  48	if ((flags & O_ACCMODE) == O_RDONLY)
  49		return GENERIC_READ;
  50	else if ((flags & O_ACCMODE) == O_WRONLY)
  51		return GENERIC_WRITE;
  52	else if ((flags & O_ACCMODE) == O_RDWR) {
  53		/* GENERIC_ALL is too much permission to request
  54		   can cause unnecessary access denied on create */
  55		/* return GENERIC_ALL; */
  56		return (GENERIC_READ | GENERIC_WRITE);
  57	}
  58
  59	return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
  60		FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
  61		FILE_READ_DATA);
  62}
  63
  64static u32 cifs_posix_convert_flags(unsigned int flags)
  65{
  66	u32 posix_flags = 0;
  67
  68	if ((flags & O_ACCMODE) == O_RDONLY)
  69		posix_flags = SMB_O_RDONLY;
  70	else if ((flags & O_ACCMODE) == O_WRONLY)
  71		posix_flags = SMB_O_WRONLY;
  72	else if ((flags & O_ACCMODE) == O_RDWR)
  73		posix_flags = SMB_O_RDWR;
  74
  75	if (flags & O_CREAT)
  76		posix_flags |= SMB_O_CREAT;
  77	if (flags & O_EXCL)
  78		posix_flags |= SMB_O_EXCL;
 
 
 
 
  79	if (flags & O_TRUNC)
  80		posix_flags |= SMB_O_TRUNC;
  81	/* be safe and imply O_SYNC for O_DSYNC */
  82	if (flags & O_DSYNC)
  83		posix_flags |= SMB_O_SYNC;
  84	if (flags & O_DIRECTORY)
  85		posix_flags |= SMB_O_DIRECTORY;
  86	if (flags & O_NOFOLLOW)
  87		posix_flags |= SMB_O_NOFOLLOW;
  88	if (flags & O_DIRECT)
  89		posix_flags |= SMB_O_DIRECT;
  90
  91	return posix_flags;
  92}
  93
  94static inline int cifs_get_disposition(unsigned int flags)
  95{
  96	if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
  97		return FILE_CREATE;
  98	else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
  99		return FILE_OVERWRITE_IF;
 100	else if ((flags & O_CREAT) == O_CREAT)
 101		return FILE_OPEN_IF;
 102	else if ((flags & O_TRUNC) == O_TRUNC)
 103		return FILE_OVERWRITE;
 104	else
 105		return FILE_OPEN;
 106}
 107
 108int cifs_posix_open(char *full_path, struct inode **pinode,
 109			struct super_block *sb, int mode, unsigned int f_flags,
 110			__u32 *poplock, __u16 *pnetfid, int xid)
 111{
 112	int rc;
 113	FILE_UNIX_BASIC_INFO *presp_data;
 114	__u32 posix_flags = 0;
 115	struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
 116	struct cifs_fattr fattr;
 117	struct tcon_link *tlink;
 118	struct cifs_tcon *tcon;
 119
 120	cFYI(1, "posix open %s", full_path);
 121
 122	presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
 123	if (presp_data == NULL)
 124		return -ENOMEM;
 125
 126	tlink = cifs_sb_tlink(cifs_sb);
 127	if (IS_ERR(tlink)) {
 128		rc = PTR_ERR(tlink);
 129		goto posix_open_ret;
 130	}
 131
 132	tcon = tlink_tcon(tlink);
 133	mode &= ~current_umask();
 134
 135	posix_flags = cifs_posix_convert_flags(f_flags);
 136	rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
 137			     poplock, full_path, cifs_sb->local_nls,
 138			     cifs_sb->mnt_cifs_flags &
 139					CIFS_MOUNT_MAP_SPECIAL_CHR);
 140	cifs_put_tlink(tlink);
 141
 142	if (rc)
 143		goto posix_open_ret;
 144
 145	if (presp_data->Type == cpu_to_le32(-1))
 146		goto posix_open_ret; /* open ok, caller does qpathinfo */
 147
 148	if (!pinode)
 149		goto posix_open_ret; /* caller does not need info */
 150
 151	cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
 152
 153	/* get new inode and set it up */
 154	if (*pinode == NULL) {
 155		cifs_fill_uniqueid(sb, &fattr);
 156		*pinode = cifs_iget(sb, &fattr);
 157		if (!*pinode) {
 158			rc = -ENOMEM;
 159			goto posix_open_ret;
 160		}
 161	} else {
 162		cifs_fattr_to_inode(*pinode, &fattr);
 163	}
 164
 165posix_open_ret:
 166	kfree(presp_data);
 167	return rc;
 168}
 169
 170static int
 171cifs_nt_open(char *full_path, struct inode *inode, struct cifs_sb_info *cifs_sb,
 172	     struct cifs_tcon *tcon, unsigned int f_flags, __u32 *poplock,
 173	     __u16 *pnetfid, int xid)
 174{
 175	int rc;
 176	int desiredAccess;
 177	int disposition;
 178	int create_options = CREATE_NOT_DIR;
 179	FILE_ALL_INFO *buf;
 
 
 180
 181	desiredAccess = cifs_convert_flags(f_flags);
 
 
 
 182
 183/*********************************************************************
 184 *  open flag mapping table:
 185 *
 186 *	POSIX Flag            CIFS Disposition
 187 *	----------            ----------------
 188 *	O_CREAT               FILE_OPEN_IF
 189 *	O_CREAT | O_EXCL      FILE_CREATE
 190 *	O_CREAT | O_TRUNC     FILE_OVERWRITE_IF
 191 *	O_TRUNC               FILE_OVERWRITE
 192 *	none of the above     FILE_OPEN
 193 *
 194 *	Note that there is not a direct match between disposition
 195 *	FILE_SUPERSEDE (ie create whether or not file exists although
 196 *	O_CREAT | O_TRUNC is similar but truncates the existing
 197 *	file rather than creating a new file as FILE_SUPERSEDE does
 198 *	(which uses the attributes / metadata passed in on open call)
 199 *?
 200 *?  O_SYNC is a reasonable match to CIFS writethrough flag
 201 *?  and the read write flags match reasonably.  O_LARGEFILE
 202 *?  is irrelevant because largefile support is always used
 203 *?  by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
 204 *	 O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
 205 *********************************************************************/
 206
 207	disposition = cifs_get_disposition(f_flags);
 208
 209	/* BB pass O_SYNC flag through on file attributes .. BB */
 210
 211	buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
 212	if (!buf)
 213		return -ENOMEM;
 214
 215	if (backup_cred(cifs_sb))
 216		create_options |= CREATE_OPEN_BACKUP_INTENT;
 217
 218	if (tcon->ses->capabilities & CAP_NT_SMBS)
 219		rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
 220			 desiredAccess, create_options, pnetfid, poplock, buf,
 221			 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
 222				 & CIFS_MOUNT_MAP_SPECIAL_CHR);
 223	else
 224		rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
 225			desiredAccess, CREATE_NOT_DIR, pnetfid, poplock, buf,
 226			cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
 227				& CIFS_MOUNT_MAP_SPECIAL_CHR);
 
 
 
 
 
 
 
 228
 229	if (rc)
 230		goto out;
 231
 232	if (tcon->unix_ext)
 233		rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
 234					      xid);
 235	else
 236		rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
 237					 xid, pnetfid);
 238
 239out:
 240	kfree(buf);
 241	return rc;
 242}
 243
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 244struct cifsFileInfo *
 245cifs_new_fileinfo(__u16 fileHandle, struct file *file,
 246		  struct tcon_link *tlink, __u32 oplock)
 247{
 248	struct dentry *dentry = file->f_path.dentry;
 249	struct inode *inode = dentry->d_inode;
 250	struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
 251	struct cifsFileInfo *pCifsFile;
 252
 253	pCifsFile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
 254	if (pCifsFile == NULL)
 255		return pCifsFile;
 256
 257	pCifsFile->count = 1;
 258	pCifsFile->netfid = fileHandle;
 259	pCifsFile->pid = current->tgid;
 260	pCifsFile->uid = current_fsuid();
 261	pCifsFile->dentry = dget(dentry);
 262	pCifsFile->f_flags = file->f_flags;
 263	pCifsFile->invalidHandle = false;
 264	pCifsFile->tlink = cifs_get_tlink(tlink);
 265	mutex_init(&pCifsFile->fh_mutex);
 266	INIT_WORK(&pCifsFile->oplock_break, cifs_oplock_break);
 267	INIT_LIST_HEAD(&pCifsFile->llist);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 268
 269	spin_lock(&cifs_file_list_lock);
 270	list_add(&pCifsFile->tlist, &(tlink_tcon(tlink)->openFileList));
 271	/* if readable file instance put first in list*/
 272	if (file->f_mode & FMODE_READ)
 273		list_add(&pCifsFile->flist, &pCifsInode->openFileList);
 274	else
 275		list_add_tail(&pCifsFile->flist, &pCifsInode->openFileList);
 276	spin_unlock(&cifs_file_list_lock);
 277
 278	cifs_set_oplock_level(pCifsInode, oplock);
 279	pCifsInode->can_cache_brlcks = pCifsInode->clientCanCacheAll;
 280
 281	file->private_data = pCifsFile;
 282	return pCifsFile;
 283}
 284
 285static void cifs_del_lock_waiters(struct cifsLockInfo *lock);
 286
 287struct cifsFileInfo *
 288cifsFileInfo_get(struct cifsFileInfo *cifs_file)
 289{
 290	spin_lock(&cifs_file_list_lock);
 291	cifsFileInfo_get_locked(cifs_file);
 292	spin_unlock(&cifs_file_list_lock);
 293	return cifs_file;
 294}
 295
 296/*
 297 * Release a reference on the file private data. This may involve closing
 298 * the filehandle out on the server. Must be called without holding
 299 * cifs_file_list_lock.
 300 */
 301void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
 302{
 303	struct inode *inode = cifs_file->dentry->d_inode;
 304	struct cifs_tcon *tcon = tlink_tcon(cifs_file->tlink);
 
 305	struct cifsInodeInfo *cifsi = CIFS_I(inode);
 306	struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
 
 307	struct cifsLockInfo *li, *tmp;
 
 
 
 
 
 308
 309	spin_lock(&cifs_file_list_lock);
 310	if (--cifs_file->count > 0) {
 311		spin_unlock(&cifs_file_list_lock);
 
 312		return;
 313	}
 
 
 
 
 
 
 
 314
 315	/* remove it from the lists */
 316	list_del(&cifs_file->flist);
 317	list_del(&cifs_file->tlist);
 318
 319	if (list_empty(&cifsi->openFileList)) {
 320		cFYI(1, "closing last open instance for inode %p",
 321			cifs_file->dentry->d_inode);
 322
 323		/* in strict cache mode we need invalidate mapping on the last
 324		   close  because it may cause a error when we open this file
 325		   again and get at least level II oplock */
 
 326		if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
 327			CIFS_I(inode)->invalid_mapping = true;
 328
 329		cifs_set_oplock_level(cifsi, 0);
 330	}
 331	spin_unlock(&cifs_file_list_lock);
 332
 333	cancel_work_sync(&cifs_file->oplock_break);
 
 
 334
 335	if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
 336		int xid, rc;
 
 337
 338		xid = GetXid();
 339		rc = CIFSSMBClose(xid, tcon, cifs_file->netfid);
 340		FreeXid(xid);
 
 341	}
 342
 343	/* Delete any outstanding lock records. We'll lose them when the file
 
 
 
 
 
 
 344	 * is closed anyway.
 345	 */
 346	mutex_lock(&cifsi->lock_mutex);
 347	list_for_each_entry_safe(li, tmp, &cifs_file->llist, llist) {
 348		list_del(&li->llist);
 349		cifs_del_lock_waiters(li);
 350		kfree(li);
 351	}
 352	mutex_unlock(&cifsi->lock_mutex);
 
 
 353
 354	cifs_put_tlink(cifs_file->tlink);
 355	dput(cifs_file->dentry);
 
 356	kfree(cifs_file);
 357}
 358
 359int cifs_open(struct inode *inode, struct file *file)
 
 360{
 361	int rc = -EACCES;
 362	int xid;
 363	__u32 oplock;
 364	struct cifs_sb_info *cifs_sb;
 
 365	struct cifs_tcon *tcon;
 366	struct tcon_link *tlink;
 367	struct cifsFileInfo *pCifsFile = NULL;
 368	char *full_path = NULL;
 369	bool posix_open_ok = false;
 370	__u16 netfid;
 
 371
 372	xid = GetXid();
 373
 374	cifs_sb = CIFS_SB(inode->i_sb);
 375	tlink = cifs_sb_tlink(cifs_sb);
 376	if (IS_ERR(tlink)) {
 377		FreeXid(xid);
 378		return PTR_ERR(tlink);
 379	}
 380	tcon = tlink_tcon(tlink);
 
 381
 382	full_path = build_path_from_dentry(file->f_path.dentry);
 383	if (full_path == NULL) {
 384		rc = -ENOMEM;
 385		goto out;
 386	}
 387
 388	cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
 389		 inode, file->f_flags, full_path);
 390
 391	if (tcon->ses->server->oplocks)
 
 
 
 
 
 
 
 
 392		oplock = REQ_OPLOCK;
 393	else
 394		oplock = 0;
 395
 396	if (!tcon->broken_posix_open && tcon->unix_ext &&
 397	    (tcon->ses->capabilities & CAP_UNIX) &&
 398	    (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 399			le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 400		/* can not refresh inode info since size could be stale */
 401		rc = cifs_posix_open(full_path, &inode, inode->i_sb,
 402				cifs_sb->mnt_file_mode /* ignored */,
 403				file->f_flags, &oplock, &netfid, xid);
 404		if (rc == 0) {
 405			cFYI(1, "posix open succeeded");
 406			posix_open_ok = true;
 407		} else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
 408			if (tcon->ses->serverNOS)
 409				cERROR(1, "server %s of type %s returned"
 410					   " unexpected error on SMB posix open"
 411					   ", disabling posix open support."
 412					   " Check if server update available.",
 413					   tcon->ses->serverName,
 414					   tcon->ses->serverNOS);
 415			tcon->broken_posix_open = true;
 416		} else if ((rc != -EIO) && (rc != -EREMOTE) &&
 417			 (rc != -EOPNOTSUPP)) /* path not found or net err */
 418			goto out;
 419		/* else fallthrough to retry open the old way on network i/o
 420		   or DFS errors */
 
 
 421	}
 422
 
 
 
 
 
 423	if (!posix_open_ok) {
 
 
 
 424		rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
 425				  file->f_flags, &oplock, &netfid, xid);
 426		if (rc)
 
 427			goto out;
 
 428	}
 429
 430	pCifsFile = cifs_new_fileinfo(netfid, file, tlink, oplock);
 431	if (pCifsFile == NULL) {
 432		CIFSSMBClose(xid, tcon, netfid);
 
 
 433		rc = -ENOMEM;
 434		goto out;
 435	}
 436
 437	cifs_fscache_set_inode_cookie(inode, file);
 438
 439	if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
 440		/* time to set mode which we can not set earlier due to
 441		   problems creating new read-only files */
 
 
 442		struct cifs_unix_set_info_args args = {
 443			.mode	= inode->i_mode,
 444			.uid	= NO_CHANGE_64,
 445			.gid	= NO_CHANGE_64,
 446			.ctime	= NO_CHANGE_64,
 447			.atime	= NO_CHANGE_64,
 448			.mtime	= NO_CHANGE_64,
 449			.device	= 0,
 450		};
 451		CIFSSMBUnixSetFileInfo(xid, tcon, &args, netfid,
 452					pCifsFile->pid);
 453	}
 454
 455out:
 456	kfree(full_path);
 457	FreeXid(xid);
 458	cifs_put_tlink(tlink);
 459	return rc;
 460}
 461
 462/* Try to reacquire byte range locks that were released when session */
 463/* to server was lost */
 464static int cifs_relock_file(struct cifsFileInfo *cifsFile)
 
 
 
 
 
 465{
 
 
 
 466	int rc = 0;
 467
 468/* BB list all locks open on this file and relock */
 
 
 
 
 
 
 
 
 
 
 
 
 469
 
 470	return rc;
 471}
 472
 473static int cifs_reopen_file(struct cifsFileInfo *pCifsFile, bool can_flush)
 
 474{
 475	int rc = -EACCES;
 476	int xid;
 477	__u32 oplock;
 478	struct cifs_sb_info *cifs_sb;
 479	struct cifs_tcon *tcon;
 480	struct cifsInodeInfo *pCifsInode;
 
 481	struct inode *inode;
 482	char *full_path = NULL;
 483	int desiredAccess;
 484	int disposition = FILE_OPEN;
 485	int create_options = CREATE_NOT_DIR;
 486	__u16 netfid;
 487
 488	xid = GetXid();
 489	mutex_lock(&pCifsFile->fh_mutex);
 490	if (!pCifsFile->invalidHandle) {
 491		mutex_unlock(&pCifsFile->fh_mutex);
 492		rc = 0;
 493		FreeXid(xid);
 494		return rc;
 495	}
 496
 497	inode = pCifsFile->dentry->d_inode;
 498	cifs_sb = CIFS_SB(inode->i_sb);
 499	tcon = tlink_tcon(pCifsFile->tlink);
 
 500
 501/* can not grab rename sem here because various ops, including
 502   those that already have the rename sem can end up causing writepage
 503   to get called and if the server was down that means we end up here,
 504   and we can never tell if the caller already has the rename_sem */
 505	full_path = build_path_from_dentry(pCifsFile->dentry);
 
 
 506	if (full_path == NULL) {
 507		rc = -ENOMEM;
 508		mutex_unlock(&pCifsFile->fh_mutex);
 509		FreeXid(xid);
 510		return rc;
 511	}
 512
 513	cFYI(1, "inode = 0x%p file flags 0x%x for %s",
 514		 inode, pCifsFile->f_flags, full_path);
 515
 516	if (tcon->ses->server->oplocks)
 517		oplock = REQ_OPLOCK;
 518	else
 519		oplock = 0;
 520
 521	if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
 522	    (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 523			le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 524
 525		/*
 526		 * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
 527		 * original open. Must mask them off for a reopen.
 528		 */
 529		unsigned int oflags = pCifsFile->f_flags &
 530						~(O_CREAT | O_EXCL | O_TRUNC);
 531
 532		rc = cifs_posix_open(full_path, NULL, inode->i_sb,
 533				cifs_sb->mnt_file_mode /* ignored */,
 534				oflags, &oplock, &netfid, xid);
 535		if (rc == 0) {
 536			cFYI(1, "posix reopen succeeded");
 
 537			goto reopen_success;
 538		}
 539		/* fallthrough to retry open the old way on errors, especially
 540		   in the reconnect path it is important to retry hard */
 
 
 541	}
 542
 543	desiredAccess = cifs_convert_flags(pCifsFile->f_flags);
 544
 545	if (backup_cred(cifs_sb))
 546		create_options |= CREATE_OPEN_BACKUP_INTENT;
 547
 548	/* Can not refresh inode by passing in file_info buf to be returned
 549	   by SMBOpen and then calling get_inode_info with returned buf
 550	   since file might have write behind data that needs to be flushed
 551	   and server version of file size can be stale. If we knew for sure
 552	   that inode was not dirty locally we could do this */
 553
 554	rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
 555			 create_options, &netfid, &oplock, NULL,
 556			 cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
 557				CIFS_MOUNT_MAP_SPECIAL_CHR);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 558	if (rc) {
 559		mutex_unlock(&pCifsFile->fh_mutex);
 560		cFYI(1, "cifs_open returned 0x%x", rc);
 561		cFYI(1, "oplock: %d", oplock);
 562		goto reopen_error_exit;
 563	}
 564
 565reopen_success:
 566	pCifsFile->netfid = netfid;
 567	pCifsFile->invalidHandle = false;
 568	mutex_unlock(&pCifsFile->fh_mutex);
 569	pCifsInode = CIFS_I(inode);
 570
 571	if (can_flush) {
 572		rc = filemap_write_and_wait(inode->i_mapping);
 573		mapping_set_error(inode->i_mapping, rc);
 574
 575		if (tcon->unix_ext)
 576			rc = cifs_get_inode_info_unix(&inode,
 577				full_path, inode->i_sb, xid);
 578		else
 579			rc = cifs_get_inode_info(&inode,
 580				full_path, NULL, inode->i_sb,
 581				xid, NULL);
 582	} /* else we are writing out data to server already
 583	     and could deadlock if we tried to flush data, and
 584	     since we do not know if we have data that would
 585	     invalidate the current end of file on the server
 586	     we can not go to the server to get the new inod
 587	     info */
 588
 589	cifs_set_oplock_level(pCifsInode, oplock);
 
 
 
 
 
 
 
 590
 591	cifs_relock_file(pCifsFile);
 
 
 592
 593reopen_error_exit:
 594	kfree(full_path);
 595	FreeXid(xid);
 596	return rc;
 597}
 598
 599int cifs_close(struct inode *inode, struct file *file)
 600{
 601	if (file->private_data != NULL) {
 602		cifsFileInfo_put(file->private_data);
 603		file->private_data = NULL;
 604	}
 605
 606	/* return code from the ->release op is always ignored */
 607	return 0;
 608}
 609
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 610int cifs_closedir(struct inode *inode, struct file *file)
 611{
 612	int rc = 0;
 613	int xid;
 614	struct cifsFileInfo *pCFileStruct = file->private_data;
 615	char *ptmp;
 616
 617	cFYI(1, "Closedir inode = 0x%p", inode);
 618
 619	xid = GetXid();
 620
 621	if (pCFileStruct) {
 622		struct cifs_tcon *pTcon = tlink_tcon(pCFileStruct->tlink);
 623
 624		cFYI(1, "Freeing private data in close dir");
 625		spin_lock(&cifs_file_list_lock);
 626		if (!pCFileStruct->srch_inf.endOfSearch &&
 627		    !pCFileStruct->invalidHandle) {
 628			pCFileStruct->invalidHandle = true;
 629			spin_unlock(&cifs_file_list_lock);
 630			rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
 631			cFYI(1, "Closing uncompleted readdir with rc %d",
 632				 rc);
 633			/* not much we can do if it fails anyway, ignore rc */
 634			rc = 0;
 635		} else
 636			spin_unlock(&cifs_file_list_lock);
 637		ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
 638		if (ptmp) {
 639			cFYI(1, "closedir free smb buf in srch struct");
 640			pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
 641			if (pCFileStruct->srch_inf.smallBuf)
 642				cifs_small_buf_release(ptmp);
 643			else
 644				cifs_buf_release(ptmp);
 645		}
 646		cifs_put_tlink(pCFileStruct->tlink);
 647		kfree(file->private_data);
 648		file->private_data = NULL;
 
 
 649	}
 
 
 
 
 650	/* BB can we lock the filestruct while this is going on? */
 651	FreeXid(xid);
 652	return rc;
 653}
 654
 655static struct cifsLockInfo *
 656cifs_lock_init(__u64 offset, __u64 length, __u8 type)
 657{
 658	struct cifsLockInfo *lock =
 659		kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
 660	if (!lock)
 661		return lock;
 662	lock->offset = offset;
 663	lock->length = length;
 664	lock->type = type;
 665	lock->pid = current->tgid;
 666	INIT_LIST_HEAD(&lock->blist);
 667	init_waitqueue_head(&lock->block_q);
 668	return lock;
 669}
 670
 671static void
 672cifs_del_lock_waiters(struct cifsLockInfo *lock)
 673{
 674	struct cifsLockInfo *li, *tmp;
 675	list_for_each_entry_safe(li, tmp, &lock->blist, blist) {
 676		list_del_init(&li->blist);
 677		wake_up(&li->block_q);
 678	}
 679}
 680
 
 
 
 
 
 681static bool
 682cifs_find_fid_lock_conflict(struct cifsFileInfo *cfile, __u64 offset,
 683			    __u64 length, __u8 type, struct cifsFileInfo *cur,
 684			    struct cifsLockInfo **conf_lock)
 685{
 686	struct cifsLockInfo *li;
 
 687	struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
 688
 689	list_for_each_entry(li, &cfile->llist, llist) {
 690		if (offset + length <= li->offset ||
 691		    offset >= li->offset + li->length)
 692			continue;
 693		else if ((type & server->vals->shared_lock_type) &&
 694			 ((server->ops->compare_fids(cur, cfile) &&
 695			   current->tgid == li->pid) || type == li->type))
 
 
 
 
 
 
 
 696			continue;
 697		else {
 698			*conf_lock = li;
 699			return true;
 700		}
 701	}
 702	return false;
 703}
 704
 705static bool
 706cifs_find_lock_conflict(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
 707			__u8 type, struct cifsLockInfo **conf_lock)
 
 708{
 709	bool rc = false;
 710	struct cifsFileInfo *fid, *tmp;
 711	struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
 712
 713	spin_lock(&cifs_file_list_lock);
 714	list_for_each_entry_safe(fid, tmp, &cinode->openFileList, flist) {
 715		rc = cifs_find_fid_lock_conflict(fid, offset, length, type,
 716						 cfile, conf_lock);
 717		if (rc)
 718			break;
 719	}
 720	spin_unlock(&cifs_file_list_lock);
 721
 722	return rc;
 723}
 724
 725/*
 726 * Check if there is another lock that prevents us to set the lock (mandatory
 727 * style). If such a lock exists, update the flock structure with its
 728 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
 729 * or leave it the same if we can't. Returns 0 if we don't need to request to
 730 * the server or 1 otherwise.
 731 */
 732static int
 733cifs_lock_test(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
 734	       __u8 type, struct file_lock *flock)
 735{
 736	int rc = 0;
 737	struct cifsLockInfo *conf_lock;
 738	struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
 739	struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
 740	bool exist;
 741
 742	mutex_lock(&cinode->lock_mutex);
 743
 744	exist = cifs_find_lock_conflict(cfile, offset, length, type,
 745					&conf_lock);
 746	if (exist) {
 747		flock->fl_start = conf_lock->offset;
 748		flock->fl_end = conf_lock->offset + conf_lock->length - 1;
 749		flock->fl_pid = conf_lock->pid;
 750		if (conf_lock->type & server->vals->shared_lock_type)
 751			flock->fl_type = F_RDLCK;
 752		else
 753			flock->fl_type = F_WRLCK;
 754	} else if (!cinode->can_cache_brlcks)
 755		rc = 1;
 756	else
 757		flock->fl_type = F_UNLCK;
 758
 759	mutex_unlock(&cinode->lock_mutex);
 760	return rc;
 761}
 762
 763static void
 764cifs_lock_add(struct cifsFileInfo *cfile, struct cifsLockInfo *lock)
 765{
 766	struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
 767	mutex_lock(&cinode->lock_mutex);
 768	list_add_tail(&lock->llist, &cfile->llist);
 769	mutex_unlock(&cinode->lock_mutex);
 770}
 771
 772/*
 773 * Set the byte-range lock (mandatory style). Returns:
 774 * 1) 0, if we set the lock and don't need to request to the server;
 775 * 2) 1, if no locks prevent us but we need to request to the server;
 776 * 3) -EACCESS, if there is a lock that prevents us and wait is false.
 777 */
 778static int
 779cifs_lock_add_if(struct cifsFileInfo *cfile, struct cifsLockInfo *lock,
 780		 bool wait)
 781{
 782	struct cifsLockInfo *conf_lock;
 783	struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
 784	bool exist;
 785	int rc = 0;
 786
 787try_again:
 788	exist = false;
 789	mutex_lock(&cinode->lock_mutex);
 790
 791	exist = cifs_find_lock_conflict(cfile, lock->offset, lock->length,
 792					lock->type, &conf_lock);
 793	if (!exist && cinode->can_cache_brlcks) {
 794		list_add_tail(&lock->llist, &cfile->llist);
 795		mutex_unlock(&cinode->lock_mutex);
 796		return rc;
 797	}
 798
 799	if (!exist)
 800		rc = 1;
 801	else if (!wait)
 802		rc = -EACCES;
 803	else {
 804		list_add_tail(&lock->blist, &conf_lock->blist);
 805		mutex_unlock(&cinode->lock_mutex);
 806		rc = wait_event_interruptible(lock->block_q,
 807					(lock->blist.prev == &lock->blist) &&
 808					(lock->blist.next == &lock->blist));
 809		if (!rc)
 810			goto try_again;
 811		mutex_lock(&cinode->lock_mutex);
 812		list_del_init(&lock->blist);
 813	}
 814
 815	mutex_unlock(&cinode->lock_mutex);
 816	return rc;
 817}
 818
 819/*
 820 * Check if there is another lock that prevents us to set the lock (posix
 821 * style). If such a lock exists, update the flock structure with its
 822 * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
 823 * or leave it the same if we can't. Returns 0 if we don't need to request to
 824 * the server or 1 otherwise.
 825 */
 826static int
 827cifs_posix_lock_test(struct file *file, struct file_lock *flock)
 828{
 829	int rc = 0;
 830	struct cifsInodeInfo *cinode = CIFS_I(file->f_path.dentry->d_inode);
 831	unsigned char saved_type = flock->fl_type;
 832
 833	if ((flock->fl_flags & FL_POSIX) == 0)
 834		return 1;
 835
 836	mutex_lock(&cinode->lock_mutex);
 837	posix_test_lock(file, flock);
 838
 839	if (flock->fl_type == F_UNLCK && !cinode->can_cache_brlcks) {
 840		flock->fl_type = saved_type;
 841		rc = 1;
 842	}
 843
 844	mutex_unlock(&cinode->lock_mutex);
 845	return rc;
 846}
 847
 848/*
 849 * Set the byte-range lock (posix style). Returns:
 850 * 1) 0, if we set the lock and don't need to request to the server;
 851 * 2) 1, if we need to request to the server;
 852 * 3) <0, if the error occurs while setting the lock.
 853 */
 854static int
 855cifs_posix_lock_set(struct file *file, struct file_lock *flock)
 856{
 857	struct cifsInodeInfo *cinode = CIFS_I(file->f_path.dentry->d_inode);
 858	int rc = 1;
 859
 860	if ((flock->fl_flags & FL_POSIX) == 0)
 861		return rc;
 862
 863try_again:
 864	mutex_lock(&cinode->lock_mutex);
 865	if (!cinode->can_cache_brlcks) {
 866		mutex_unlock(&cinode->lock_mutex);
 867		return rc;
 868	}
 869
 870	rc = posix_lock_file(file, flock, NULL);
 871	mutex_unlock(&cinode->lock_mutex);
 872	if (rc == FILE_LOCK_DEFERRED) {
 873		rc = wait_event_interruptible(flock->fl_wait, !flock->fl_next);
 874		if (!rc)
 875			goto try_again;
 876		locks_delete_block(flock);
 877	}
 878	return rc;
 879}
 880
 881static int
 882cifs_push_mandatory_locks(struct cifsFileInfo *cfile)
 883{
 884	int xid, rc = 0, stored_rc;
 
 885	struct cifsLockInfo *li, *tmp;
 886	struct cifs_tcon *tcon;
 887	struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
 888	unsigned int num, max_num, max_buf;
 889	LOCKING_ANDX_RANGE *buf, *cur;
 890	int types[] = {LOCKING_ANDX_LARGE_FILES,
 891		       LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES};
 
 
 892	int i;
 893
 894	xid = GetXid();
 895	tcon = tlink_tcon(cfile->tlink);
 896
 897	mutex_lock(&cinode->lock_mutex);
 898	if (!cinode->can_cache_brlcks) {
 899		mutex_unlock(&cinode->lock_mutex);
 900		FreeXid(xid);
 901		return rc;
 902	}
 903
 904	/*
 905	 * Accessing maxBuf is racy with cifs_reconnect - need to store value
 906	 * and check it for zero before using.
 907	 */
 908	max_buf = tcon->ses->server->maxBuf;
 909	if (!max_buf) {
 910		mutex_unlock(&cinode->lock_mutex);
 911		FreeXid(xid);
 912		return -EINVAL;
 913	}
 914
 915	max_num = (max_buf - sizeof(struct smb_hdr)) /
 916						sizeof(LOCKING_ANDX_RANGE);
 917	buf = kzalloc(max_num * sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
 918	if (!buf) {
 919		mutex_unlock(&cinode->lock_mutex);
 920		FreeXid(xid);
 921		return rc;
 922	}
 923
 924	for (i = 0; i < 2; i++) {
 925		cur = buf;
 926		num = 0;
 927		list_for_each_entry_safe(li, tmp, &cfile->llist, llist) {
 928			if (li->type != types[i])
 929				continue;
 930			cur->Pid = cpu_to_le16(li->pid);
 931			cur->LengthLow = cpu_to_le32((u32)li->length);
 932			cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
 933			cur->OffsetLow = cpu_to_le32((u32)li->offset);
 934			cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
 935			if (++num == max_num) {
 936				stored_rc = cifs_lockv(xid, tcon, cfile->netfid,
 
 937						       (__u8)li->type, 0, num,
 938						       buf);
 939				if (stored_rc)
 940					rc = stored_rc;
 941				cur = buf;
 942				num = 0;
 943			} else
 944				cur++;
 945		}
 946
 947		if (num) {
 948			stored_rc = cifs_lockv(xid, tcon, cfile->netfid,
 949					       (__u8)types[i], 0, num, buf);
 950			if (stored_rc)
 951				rc = stored_rc;
 952		}
 953	}
 954
 955	cinode->can_cache_brlcks = false;
 956	mutex_unlock(&cinode->lock_mutex);
 957
 958	kfree(buf);
 959	FreeXid(xid);
 960	return rc;
 961}
 962
 963/* copied from fs/locks.c with a name change */
 964#define cifs_for_each_lock(inode, lockp) \
 965	for (lockp = &inode->i_flock; *lockp != NULL; \
 966	     lockp = &(*lockp)->fl_next)
 
 967
 968struct lock_to_push {
 969	struct list_head llist;
 970	__u64 offset;
 971	__u64 length;
 972	__u32 pid;
 973	__u16 netfid;
 974	__u8 type;
 975};
 976
 977static int
 978cifs_push_posix_locks(struct cifsFileInfo *cfile)
 979{
 980	struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
 981	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
 982	struct file_lock *flock, **before;
 983	unsigned int count = 0, i = 0;
 
 984	int rc = 0, xid, type;
 985	struct list_head locks_to_send, *el;
 986	struct lock_to_push *lck, *tmp;
 987	__u64 length;
 988
 989	xid = GetXid();
 990
 991	mutex_lock(&cinode->lock_mutex);
 992	if (!cinode->can_cache_brlcks) {
 993		mutex_unlock(&cinode->lock_mutex);
 994		FreeXid(xid);
 995		return rc;
 996	}
 997
 998	lock_flocks();
 999	cifs_for_each_lock(cfile->dentry->d_inode, before) {
1000		if ((*before)->fl_flags & FL_POSIX)
1001			count++;
1002	}
1003	unlock_flocks();
1004
1005	INIT_LIST_HEAD(&locks_to_send);
1006
1007	/*
1008	 * Allocating count locks is enough because no FL_POSIX locks can be
1009	 * added to the list while we are holding cinode->lock_mutex that
1010	 * protects locking operations of this inode.
1011	 */
1012	for (; i < count; i++) {
1013		lck = kmalloc(sizeof(struct lock_to_push), GFP_KERNEL);
1014		if (!lck) {
1015			rc = -ENOMEM;
1016			goto err_out;
1017		}
1018		list_add_tail(&lck->llist, &locks_to_send);
1019	}
1020
1021	el = locks_to_send.next;
1022	lock_flocks();
1023	cifs_for_each_lock(cfile->dentry->d_inode, before) {
1024		flock = *before;
1025		if ((flock->fl_flags & FL_POSIX) == 0)
1026			continue;
1027		if (el == &locks_to_send) {
1028			/*
1029			 * The list ended. We don't have enough allocated
1030			 * structures - something is really wrong.
1031			 */
1032			cERROR(1, "Can't push all brlocks!");
1033			break;
1034		}
1035		length = 1 + flock->fl_end - flock->fl_start;
1036		if (flock->fl_type == F_RDLCK || flock->fl_type == F_SHLCK)
1037			type = CIFS_RDLCK;
1038		else
1039			type = CIFS_WRLCK;
1040		lck = list_entry(el, struct lock_to_push, llist);
1041		lck->pid = flock->fl_pid;
1042		lck->netfid = cfile->netfid;
1043		lck->length = length;
1044		lck->type = type;
1045		lck->offset = flock->fl_start;
1046		el = el->next;
1047	}
1048	unlock_flocks();
1049
1050	list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1051		struct file_lock tmp_lock;
1052		int stored_rc;
1053
1054		tmp_lock.fl_start = lck->offset;
1055		stored_rc = CIFSSMBPosixLock(xid, tcon, lck->netfid, lck->pid,
1056					     0, lck->length, &tmp_lock,
1057					     lck->type, 0);
1058		if (stored_rc)
1059			rc = stored_rc;
1060		list_del(&lck->llist);
1061		kfree(lck);
1062	}
1063
1064out:
1065	cinode->can_cache_brlcks = false;
1066	mutex_unlock(&cinode->lock_mutex);
1067
1068	FreeXid(xid);
1069	return rc;
1070err_out:
1071	list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1072		list_del(&lck->llist);
1073		kfree(lck);
1074	}
1075	goto out;
1076}
1077
1078static int
1079cifs_push_locks(struct cifsFileInfo *cfile)
1080{
1081	struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
 
1082	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
 
 
 
 
 
 
 
 
1083
1084	if ((tcon->ses->capabilities & CAP_UNIX) &&
1085	    (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1086	    ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1087		return cifs_push_posix_locks(cfile);
 
 
1088
1089	return cifs_push_mandatory_locks(cfile);
 
 
1090}
1091
1092static void
1093cifs_read_flock(struct file_lock *flock, __u32 *type, int *lock, int *unlock,
1094		bool *wait_flag, struct TCP_Server_Info *server)
1095{
1096	if (flock->fl_flags & FL_POSIX)
1097		cFYI(1, "Posix");
1098	if (flock->fl_flags & FL_FLOCK)
1099		cFYI(1, "Flock");
1100	if (flock->fl_flags & FL_SLEEP) {
1101		cFYI(1, "Blocking lock");
1102		*wait_flag = true;
1103	}
1104	if (flock->fl_flags & FL_ACCESS)
1105		cFYI(1, "Process suspended by mandatory locking - "
1106			"not implemented yet");
1107	if (flock->fl_flags & FL_LEASE)
1108		cFYI(1, "Lease on file - not implemented yet");
1109	if (flock->fl_flags &
1110	    (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
1111		cFYI(1, "Unknown lock flags 0x%x", flock->fl_flags);
 
1112
1113	*type = server->vals->large_lock_type;
1114	if (flock->fl_type == F_WRLCK) {
1115		cFYI(1, "F_WRLCK ");
1116		*type |= server->vals->exclusive_lock_type;
1117		*lock = 1;
1118	} else if (flock->fl_type == F_UNLCK) {
1119		cFYI(1, "F_UNLCK");
1120		*type |= server->vals->unlock_lock_type;
1121		*unlock = 1;
1122		/* Check if unlock includes more than one lock range */
1123	} else if (flock->fl_type == F_RDLCK) {
1124		cFYI(1, "F_RDLCK");
1125		*type |= server->vals->shared_lock_type;
1126		*lock = 1;
1127	} else if (flock->fl_type == F_EXLCK) {
1128		cFYI(1, "F_EXLCK");
1129		*type |= server->vals->exclusive_lock_type;
1130		*lock = 1;
1131	} else if (flock->fl_type == F_SHLCK) {
1132		cFYI(1, "F_SHLCK");
1133		*type |= server->vals->shared_lock_type;
1134		*lock = 1;
1135	} else
1136		cFYI(1, "Unknown type of lock");
1137}
1138
1139static int
1140cifs_mandatory_lock(int xid, struct cifsFileInfo *cfile, __u64 offset,
1141		    __u64 length, __u32 type, int lock, int unlock, bool wait)
1142{
1143	return CIFSSMBLock(xid, tlink_tcon(cfile->tlink), cfile->netfid,
1144			   current->tgid, length, offset, unlock, lock,
1145			   (__u8)type, wait, 0);
1146}
1147
1148static int
1149cifs_getlk(struct file *file, struct file_lock *flock, __u32 type,
1150	   bool wait_flag, bool posix_lck, int xid)
1151{
1152	int rc = 0;
1153	__u64 length = 1 + flock->fl_end - flock->fl_start;
1154	struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1155	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1156	struct TCP_Server_Info *server = tcon->ses->server;
1157	__u16 netfid = cfile->netfid;
1158
1159	if (posix_lck) {
1160		int posix_lock_type;
1161
1162		rc = cifs_posix_lock_test(file, flock);
1163		if (!rc)
1164			return rc;
1165
1166		if (type & server->vals->shared_lock_type)
1167			posix_lock_type = CIFS_RDLCK;
1168		else
1169			posix_lock_type = CIFS_WRLCK;
1170		rc = CIFSSMBPosixLock(xid, tcon, netfid, current->tgid,
1171				      1 /* get */, length, flock,
 
1172				      posix_lock_type, wait_flag);
1173		return rc;
1174	}
1175
1176	rc = cifs_lock_test(cfile, flock->fl_start, length, type, flock);
1177	if (!rc)
1178		return rc;
1179
1180	/* BB we could chain these into one lock request BB */
1181	rc = cifs_mandatory_lock(xid, cfile, flock->fl_start, length, type,
1182				 1, 0, false);
1183	if (rc == 0) {
1184		rc = cifs_mandatory_lock(xid, cfile, flock->fl_start, length,
1185					 type, 0, 1, false);
1186		flock->fl_type = F_UNLCK;
1187		if (rc != 0)
1188			cERROR(1, "Error unlocking previously locked "
1189				  "range %d during test of lock", rc);
1190		return 0;
1191	}
1192
1193	if (type & server->vals->shared_lock_type) {
1194		flock->fl_type = F_WRLCK;
1195		return 0;
1196	}
1197
1198	rc = cifs_mandatory_lock(xid, cfile, flock->fl_start, length,
1199				 type | server->vals->shared_lock_type, 1, 0,
1200				 false);
 
 
1201	if (rc == 0) {
1202		rc = cifs_mandatory_lock(xid, cfile, flock->fl_start, length,
1203					 type | server->vals->shared_lock_type,
1204					 0, 1, false);
1205		flock->fl_type = F_RDLCK;
1206		if (rc != 0)
1207			cERROR(1, "Error unlocking previously locked "
1208				  "range %d during test of lock", rc);
1209	} else
1210		flock->fl_type = F_WRLCK;
1211
1212	return 0;
1213}
1214
1215static void
1216cifs_move_llist(struct list_head *source, struct list_head *dest)
1217{
1218	struct list_head *li, *tmp;
1219	list_for_each_safe(li, tmp, source)
1220		list_move(li, dest);
1221}
1222
1223static void
1224cifs_free_llist(struct list_head *llist)
1225{
1226	struct cifsLockInfo *li, *tmp;
1227	list_for_each_entry_safe(li, tmp, llist, llist) {
1228		cifs_del_lock_waiters(li);
1229		list_del(&li->llist);
1230		kfree(li);
1231	}
1232}
1233
1234static int
1235cifs_unlock_range(struct cifsFileInfo *cfile, struct file_lock *flock, int xid)
 
1236{
1237	int rc = 0, stored_rc;
1238	int types[] = {LOCKING_ANDX_LARGE_FILES,
1239		       LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES};
 
 
1240	unsigned int i;
1241	unsigned int max_num, num, max_buf;
1242	LOCKING_ANDX_RANGE *buf, *cur;
1243	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1244	struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
1245	struct cifsLockInfo *li, *tmp;
1246	__u64 length = 1 + flock->fl_end - flock->fl_start;
1247	struct list_head tmp_llist;
1248
1249	INIT_LIST_HEAD(&tmp_llist);
1250
1251	/*
1252	 * Accessing maxBuf is racy with cifs_reconnect - need to store value
1253	 * and check it for zero before using.
1254	 */
1255	max_buf = tcon->ses->server->maxBuf;
1256	if (!max_buf)
1257		return -EINVAL;
1258
1259	max_num = (max_buf - sizeof(struct smb_hdr)) /
1260						sizeof(LOCKING_ANDX_RANGE);
1261	buf = kzalloc(max_num * sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1262	if (!buf)
1263		return -ENOMEM;
1264
1265	mutex_lock(&cinode->lock_mutex);
1266	for (i = 0; i < 2; i++) {
1267		cur = buf;
1268		num = 0;
1269		list_for_each_entry_safe(li, tmp, &cfile->llist, llist) {
1270			if (flock->fl_start > li->offset ||
1271			    (flock->fl_start + length) <
1272			    (li->offset + li->length))
1273				continue;
1274			if (current->tgid != li->pid)
1275				continue;
1276			if (types[i] != li->type)
1277				continue;
1278			if (cinode->can_cache_brlcks) {
1279				/*
1280				 * We can cache brlock requests - simply remove
1281				 * a lock from the file's list.
1282				 */
1283				list_del(&li->llist);
1284				cifs_del_lock_waiters(li);
1285				kfree(li);
1286				continue;
1287			}
1288			cur->Pid = cpu_to_le16(li->pid);
1289			cur->LengthLow = cpu_to_le32((u32)li->length);
1290			cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1291			cur->OffsetLow = cpu_to_le32((u32)li->offset);
1292			cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1293			/*
1294			 * We need to save a lock here to let us add it again to
1295			 * the file's list if the unlock range request fails on
1296			 * the server.
1297			 */
1298			list_move(&li->llist, &tmp_llist);
1299			if (++num == max_num) {
1300				stored_rc = cifs_lockv(xid, tcon, cfile->netfid,
 
1301						       li->type, num, 0, buf);
1302				if (stored_rc) {
1303					/*
1304					 * We failed on the unlock range
1305					 * request - add all locks from the tmp
1306					 * list to the head of the file's list.
1307					 */
1308					cifs_move_llist(&tmp_llist,
1309							&cfile->llist);
1310					rc = stored_rc;
1311				} else
1312					/*
1313					 * The unlock range request succeed -
1314					 * free the tmp list.
1315					 */
1316					cifs_free_llist(&tmp_llist);
1317				cur = buf;
1318				num = 0;
1319			} else
1320				cur++;
1321		}
1322		if (num) {
1323			stored_rc = cifs_lockv(xid, tcon, cfile->netfid,
1324					       types[i], num, 0, buf);
1325			if (stored_rc) {
1326				cifs_move_llist(&tmp_llist, &cfile->llist);
 
1327				rc = stored_rc;
1328			} else
1329				cifs_free_llist(&tmp_llist);
1330		}
1331	}
1332
1333	mutex_unlock(&cinode->lock_mutex);
1334	kfree(buf);
1335	return rc;
1336}
1337
1338static int
1339cifs_setlk(struct file *file,  struct file_lock *flock, __u32 type,
1340	   bool wait_flag, bool posix_lck, int lock, int unlock, int xid)
 
1341{
1342	int rc = 0;
1343	__u64 length = 1 + flock->fl_end - flock->fl_start;
1344	struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1345	struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1346	struct TCP_Server_Info *server = tcon->ses->server;
1347	__u16 netfid = cfile->netfid;
1348
1349	if (posix_lck) {
1350		int posix_lock_type;
1351
1352		rc = cifs_posix_lock_set(file, flock);
1353		if (!rc || rc < 0)
1354			return rc;
1355
1356		if (type & server->vals->shared_lock_type)
1357			posix_lock_type = CIFS_RDLCK;
1358		else
1359			posix_lock_type = CIFS_WRLCK;
1360
1361		if (unlock == 1)
1362			posix_lock_type = CIFS_UNLCK;
1363
1364		rc = CIFSSMBPosixLock(xid, tcon, netfid, current->tgid,
1365				      0 /* set */, length, flock,
1366				      posix_lock_type, wait_flag);
 
1367		goto out;
1368	}
1369
1370	if (lock) {
1371		struct cifsLockInfo *lock;
1372
1373		lock = cifs_lock_init(flock->fl_start, length, type);
1374		if (!lock)
1375			return -ENOMEM;
1376
1377		rc = cifs_lock_add_if(cfile, lock, wait_flag);
1378		if (rc < 0)
1379			kfree(lock);
1380		if (rc <= 0)
 
 
1381			goto out;
1382
1383		rc = cifs_mandatory_lock(xid, cfile, flock->fl_start, length,
1384					 type, 1, 0, wait_flag);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1385		if (rc) {
1386			kfree(lock);
1387			goto out;
1388		}
1389
1390		cifs_lock_add(cfile, lock);
1391	} else if (unlock)
1392		rc = cifs_unlock_range(cfile, flock, xid);
1393
1394out:
1395	if (flock->fl_flags & FL_POSIX)
1396		posix_lock_file_wait(file, flock);
1397	return rc;
1398}
1399
1400int cifs_lock(struct file *file, int cmd, struct file_lock *flock)
1401{
1402	int rc, xid;
1403	int lock = 0, unlock = 0;
1404	bool wait_flag = false;
1405	bool posix_lck = false;
1406	struct cifs_sb_info *cifs_sb;
1407	struct cifs_tcon *tcon;
1408	struct cifsInodeInfo *cinode;
1409	struct cifsFileInfo *cfile;
1410	__u16 netfid;
1411	__u32 type;
1412
1413	rc = -EACCES;
1414	xid = GetXid();
1415
1416	cFYI(1, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld "
1417		"end: %lld", cmd, flock->fl_flags, flock->fl_type,
1418		flock->fl_start, flock->fl_end);
1419
1420	cfile = (struct cifsFileInfo *)file->private_data;
1421	tcon = tlink_tcon(cfile->tlink);
1422
1423	cifs_read_flock(flock, &type, &lock, &unlock, &wait_flag,
1424			tcon->ses->server);
1425
1426	cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1427	netfid = cfile->netfid;
1428	cinode = CIFS_I(file->f_path.dentry->d_inode);
1429
1430	if ((tcon->ses->capabilities & CAP_UNIX) &&
1431	    (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1432	    ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1433		posix_lck = true;
1434	/*
1435	 * BB add code here to normalize offset and length to account for
1436	 * negative length which we can not accept over the wire.
1437	 */
1438	if (IS_GETLK(cmd)) {
1439		rc = cifs_getlk(file, flock, type, wait_flag, posix_lck, xid);
1440		FreeXid(xid);
1441		return rc;
1442	}
1443
1444	if (!lock && !unlock) {
1445		/*
1446		 * if no lock or unlock then nothing to do since we do not
1447		 * know what it is
1448		 */
1449		FreeXid(xid);
1450		return -EOPNOTSUPP;
1451	}
1452
1453	rc = cifs_setlk(file, flock, type, wait_flag, posix_lck, lock, unlock,
1454			xid);
1455	FreeXid(xid);
1456	return rc;
1457}
1458
1459/*
1460 * update the file size (if needed) after a write. Should be called with
1461 * the inode->i_lock held
1462 */
1463void
1464cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
1465		      unsigned int bytes_written)
1466{
1467	loff_t end_of_write = offset + bytes_written;
1468
1469	if (end_of_write > cifsi->server_eof)
1470		cifsi->server_eof = end_of_write;
1471}
1472
1473static ssize_t cifs_write(struct cifsFileInfo *open_file, __u32 pid,
1474			  const char *write_data, size_t write_size,
1475			  loff_t *poffset)
1476{
1477	int rc = 0;
1478	unsigned int bytes_written = 0;
1479	unsigned int total_written;
1480	struct cifs_sb_info *cifs_sb;
1481	struct cifs_tcon *pTcon;
1482	int xid;
 
1483	struct dentry *dentry = open_file->dentry;
1484	struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
1485	struct cifs_io_parms io_parms;
1486
1487	cifs_sb = CIFS_SB(dentry->d_sb);
1488
1489	cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
1490	   *poffset, dentry->d_name.name);
 
 
 
1491
1492	pTcon = tlink_tcon(open_file->tlink);
 
1493
1494	xid = GetXid();
1495
1496	for (total_written = 0; write_size > total_written;
1497	     total_written += bytes_written) {
1498		rc = -EAGAIN;
1499		while (rc == -EAGAIN) {
1500			struct kvec iov[2];
1501			unsigned int len;
1502
1503			if (open_file->invalidHandle) {
1504				/* we could deadlock if we called
1505				   filemap_fdatawait from here so tell
1506				   reopen_file not to flush data to
1507				   server now */
1508				rc = cifs_reopen_file(open_file, false);
1509				if (rc != 0)
1510					break;
1511			}
1512
1513			len = min((size_t)cifs_sb->wsize,
1514				  write_size - total_written);
1515			/* iov[0] is reserved for smb header */
1516			iov[1].iov_base = (char *)write_data + total_written;
1517			iov[1].iov_len = len;
1518			io_parms.netfid = open_file->netfid;
1519			io_parms.pid = pid;
1520			io_parms.tcon = pTcon;
1521			io_parms.offset = *poffset;
1522			io_parms.length = len;
1523			rc = CIFSSMBWrite2(xid, &io_parms, &bytes_written, iov,
1524					   1, 0);
1525		}
1526		if (rc || (bytes_written == 0)) {
1527			if (total_written)
1528				break;
1529			else {
1530				FreeXid(xid);
1531				return rc;
1532			}
1533		} else {
1534			spin_lock(&dentry->d_inode->i_lock);
1535			cifs_update_eof(cifsi, *poffset, bytes_written);
1536			spin_unlock(&dentry->d_inode->i_lock);
1537			*poffset += bytes_written;
1538		}
1539	}
1540
1541	cifs_stats_bytes_written(pTcon, total_written);
1542
1543	if (total_written > 0) {
1544		spin_lock(&dentry->d_inode->i_lock);
1545		if (*poffset > dentry->d_inode->i_size)
1546			i_size_write(dentry->d_inode, *poffset);
1547		spin_unlock(&dentry->d_inode->i_lock);
1548	}
1549	mark_inode_dirty_sync(dentry->d_inode);
1550	FreeXid(xid);
1551	return total_written;
1552}
1553
1554struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1555					bool fsuid_only)
1556{
1557	struct cifsFileInfo *open_file = NULL;
1558	struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
 
1559
1560	/* only filter by fsuid on multiuser mounts */
1561	if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1562		fsuid_only = false;
1563
1564	spin_lock(&cifs_file_list_lock);
1565	/* we could simply get the first_list_entry since write-only entries
1566	   are always at the end of the list but since the first entry might
1567	   have a close pending, we go through the whole list */
1568	list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1569		if (fsuid_only && open_file->uid != current_fsuid())
1570			continue;
1571		if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1572			if (!open_file->invalidHandle) {
1573				/* found a good file */
1574				/* lock it so it will not be closed on us */
1575				cifsFileInfo_get_locked(open_file);
1576				spin_unlock(&cifs_file_list_lock);
1577				return open_file;
1578			} /* else might as well continue, and look for
1579			     another, or simply have the caller reopen it
1580			     again rather than trying to fix this handle */
1581		} else /* write only file */
1582			break; /* write only files are last so must be done */
1583	}
1584	spin_unlock(&cifs_file_list_lock);
1585	return NULL;
1586}
1587
1588struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1589					bool fsuid_only)
1590{
1591	struct cifsFileInfo *open_file, *inv_file = NULL;
1592	struct cifs_sb_info *cifs_sb;
 
1593	bool any_available = false;
1594	int rc;
1595	unsigned int refind = 0;
1596
1597	/* Having a null inode here (because mapping->host was set to zero by
1598	the VFS or MM) should not happen but we had reports of on oops (due to
1599	it being zero) during stress testcases so we need to check for it */
1600
1601	if (cifs_inode == NULL) {
1602		cERROR(1, "Null inode passed to cifs_writeable_file");
1603		dump_stack();
1604		return NULL;
1605	}
1606
1607	cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
 
1608
1609	/* only filter by fsuid on multiuser mounts */
1610	if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1611		fsuid_only = false;
1612
1613	spin_lock(&cifs_file_list_lock);
1614refind_writable:
1615	if (refind > MAX_REOPEN_ATT) {
1616		spin_unlock(&cifs_file_list_lock);
1617		return NULL;
1618	}
1619	list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1620		if (!any_available && open_file->pid != current->tgid)
1621			continue;
1622		if (fsuid_only && open_file->uid != current_fsuid())
1623			continue;
1624		if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1625			if (!open_file->invalidHandle) {
1626				/* found a good writable file */
1627				cifsFileInfo_get_locked(open_file);
1628				spin_unlock(&cifs_file_list_lock);
1629				return open_file;
1630			} else {
1631				if (!inv_file)
1632					inv_file = open_file;
1633			}
1634		}
1635	}
1636	/* couldn't find useable FH with same pid, try any available */
1637	if (!any_available) {
1638		any_available = true;
1639		goto refind_writable;
1640	}
1641
1642	if (inv_file) {
1643		any_available = false;
1644		cifsFileInfo_get_locked(inv_file);
1645	}
1646
1647	spin_unlock(&cifs_file_list_lock);
1648
1649	if (inv_file) {
1650		rc = cifs_reopen_file(inv_file, false);
1651		if (!rc)
1652			return inv_file;
1653		else {
1654			spin_lock(&cifs_file_list_lock);
1655			list_move_tail(&inv_file->flist,
1656					&cifs_inode->openFileList);
1657			spin_unlock(&cifs_file_list_lock);
1658			cifsFileInfo_put(inv_file);
1659			spin_lock(&cifs_file_list_lock);
1660			++refind;
 
 
1661			goto refind_writable;
1662		}
1663	}
1664
1665	return NULL;
1666}
1667
1668static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1669{
1670	struct address_space *mapping = page->mapping;
1671	loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1672	char *write_data;
1673	int rc = -EFAULT;
1674	int bytes_written = 0;
1675	struct inode *inode;
1676	struct cifsFileInfo *open_file;
1677
1678	if (!mapping || !mapping->host)
1679		return -EFAULT;
1680
1681	inode = page->mapping->host;
1682
1683	offset += (loff_t)from;
1684	write_data = kmap(page);
1685	write_data += from;
1686
1687	if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1688		kunmap(page);
1689		return -EIO;
1690	}
1691
1692	/* racing with truncate? */
1693	if (offset > mapping->host->i_size) {
1694		kunmap(page);
1695		return 0; /* don't care */
1696	}
1697
1698	/* check to make sure that we are not extending the file */
1699	if (mapping->host->i_size - offset < (loff_t)to)
1700		to = (unsigned)(mapping->host->i_size - offset);
1701
1702	open_file = find_writable_file(CIFS_I(mapping->host), false);
1703	if (open_file) {
1704		bytes_written = cifs_write(open_file, open_file->pid,
1705					   write_data, to - from, &offset);
1706		cifsFileInfo_put(open_file);
1707		/* Does mm or vfs already set times? */
1708		inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1709		if ((bytes_written > 0) && (offset))
1710			rc = 0;
1711		else if (bytes_written < 0)
1712			rc = bytes_written;
1713	} else {
1714		cFYI(1, "No writeable filehandles for inode");
1715		rc = -EIO;
1716	}
1717
1718	kunmap(page);
1719	return rc;
1720}
1721
1722/*
1723 * Marshal up the iov array, reserving the first one for the header. Also,
1724 * set wdata->bytes.
1725 */
1726static void
1727cifs_writepages_marshal_iov(struct kvec *iov, struct cifs_writedata *wdata)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1728{
1729	int i;
1730	struct inode *inode = wdata->cfile->dentry->d_inode;
1731	loff_t size = i_size_read(inode);
1732
1733	/* marshal up the pages into iov array */
1734	wdata->bytes = 0;
1735	for (i = 0; i < wdata->nr_pages; i++) {
1736		iov[i + 1].iov_len = min(size - page_offset(wdata->pages[i]),
1737					(loff_t)PAGE_CACHE_SIZE);
1738		iov[i + 1].iov_base = kmap(wdata->pages[i]);
1739		wdata->bytes += iov[i + 1].iov_len;
 
 
 
 
 
 
 
 
 
 
 
 
1740	}
 
 
 
 
 
1741}
1742
1743static int cifs_writepages(struct address_space *mapping,
1744			   struct writeback_control *wbc)
1745{
1746	struct cifs_sb_info *cifs_sb = CIFS_SB(mapping->host->i_sb);
 
1747	bool done = false, scanned = false, range_whole = false;
1748	pgoff_t end, index;
1749	struct cifs_writedata *wdata;
1750	struct page *page;
1751	int rc = 0;
1752
1753	/*
1754	 * If wsize is smaller than the page cache size, default to writing
1755	 * one page at a time via cifs_writepage
1756	 */
1757	if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1758		return generic_writepages(mapping, wbc);
1759
1760	if (wbc->range_cyclic) {
1761		index = mapping->writeback_index; /* Start from prev offset */
1762		end = -1;
1763	} else {
1764		index = wbc->range_start >> PAGE_CACHE_SHIFT;
1765		end = wbc->range_end >> PAGE_CACHE_SHIFT;
1766		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1767			range_whole = true;
1768		scanned = true;
1769	}
 
1770retry:
1771	while (!done && index <= end) {
1772		unsigned int i, nr_pages, found_pages;
1773		pgoff_t next = 0, tofind;
1774		struct page **pages;
1775
1776		tofind = min((cifs_sb->wsize / PAGE_CACHE_SIZE) - 1,
1777				end - index) + 1;
 
 
1778
1779		wdata = cifs_writedata_alloc((unsigned int)tofind,
1780					     cifs_writev_complete);
 
 
1781		if (!wdata) {
1782			rc = -ENOMEM;
 
1783			break;
1784		}
1785
1786		/*
1787		 * find_get_pages_tag seems to return a max of 256 on each
1788		 * iteration, so we must call it several times in order to
1789		 * fill the array or the wsize is effectively limited to
1790		 * 256 * PAGE_CACHE_SIZE.
1791		 */
1792		found_pages = 0;
1793		pages = wdata->pages;
1794		do {
1795			nr_pages = find_get_pages_tag(mapping, &index,
1796							PAGECACHE_TAG_DIRTY,
1797							tofind, pages);
1798			found_pages += nr_pages;
1799			tofind -= nr_pages;
1800			pages += nr_pages;
1801		} while (nr_pages && tofind && index <= end);
1802
1803		if (found_pages == 0) {
1804			kref_put(&wdata->refcount, cifs_writedata_release);
 
1805			break;
1806		}
1807
1808		nr_pages = 0;
1809		for (i = 0; i < found_pages; i++) {
1810			page = wdata->pages[i];
1811			/*
1812			 * At this point we hold neither mapping->tree_lock nor
1813			 * lock on the page itself: the page may be truncated or
1814			 * invalidated (changing page->mapping to NULL), or even
1815			 * swizzled back from swapper_space to tmpfs file
1816			 * mapping
1817			 */
1818
1819			if (nr_pages == 0)
1820				lock_page(page);
1821			else if (!trylock_page(page))
1822				break;
1823
1824			if (unlikely(page->mapping != mapping)) {
1825				unlock_page(page);
1826				break;
1827			}
1828
1829			if (!wbc->range_cyclic && page->index > end) {
1830				done = true;
1831				unlock_page(page);
1832				break;
1833			}
1834
1835			if (next && (page->index != next)) {
1836				/* Not next consecutive page */
1837				unlock_page(page);
1838				break;
1839			}
1840
1841			if (wbc->sync_mode != WB_SYNC_NONE)
1842				wait_on_page_writeback(page);
1843
1844			if (PageWriteback(page) ||
1845					!clear_page_dirty_for_io(page)) {
1846				unlock_page(page);
1847				break;
1848			}
1849
1850			/*
1851			 * This actually clears the dirty bit in the radix tree.
1852			 * See cifs_writepage() for more commentary.
1853			 */
1854			set_page_writeback(page);
1855
1856			if (page_offset(page) >= mapping->host->i_size) {
1857				done = true;
1858				unlock_page(page);
1859				end_page_writeback(page);
1860				break;
1861			}
1862
1863			wdata->pages[i] = page;
1864			next = page->index + 1;
1865			++nr_pages;
1866		}
1867
1868		/* reset index to refind any pages skipped */
1869		if (nr_pages == 0)
1870			index = wdata->pages[0]->index + 1;
1871
1872		/* put any pages we aren't going to use */
1873		for (i = nr_pages; i < found_pages; i++) {
1874			page_cache_release(wdata->pages[i]);
1875			wdata->pages[i] = NULL;
1876		}
1877
1878		/* nothing to write? */
1879		if (nr_pages == 0) {
1880			kref_put(&wdata->refcount, cifs_writedata_release);
 
1881			continue;
1882		}
1883
1884		wdata->sync_mode = wbc->sync_mode;
1885		wdata->nr_pages = nr_pages;
1886		wdata->offset = page_offset(wdata->pages[0]);
1887		wdata->marshal_iov = cifs_writepages_marshal_iov;
1888
1889		do {
1890			if (wdata->cfile != NULL)
1891				cifsFileInfo_put(wdata->cfile);
1892			wdata->cfile = find_writable_file(CIFS_I(mapping->host),
1893							  false);
1894			if (!wdata->cfile) {
1895				cERROR(1, "No writable handles for inode");
1896				rc = -EBADF;
1897				break;
1898			}
1899			wdata->pid = wdata->cfile->pid;
1900			rc = cifs_async_writev(wdata);
1901		} while (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN);
1902
1903		for (i = 0; i < nr_pages; ++i)
1904			unlock_page(wdata->pages[i]);
1905
1906		/* send failure -- clean up the mess */
1907		if (rc != 0) {
 
1908			for (i = 0; i < nr_pages; ++i) {
1909				if (rc == -EAGAIN)
1910					redirty_page_for_writepage(wbc,
1911							   wdata->pages[i]);
1912				else
1913					SetPageError(wdata->pages[i]);
1914				end_page_writeback(wdata->pages[i]);
1915				page_cache_release(wdata->pages[i]);
1916			}
1917			if (rc != -EAGAIN)
1918				mapping_set_error(mapping, rc);
1919		}
1920		kref_put(&wdata->refcount, cifs_writedata_release);
1921
 
 
 
 
 
1922		wbc->nr_to_write -= nr_pages;
1923		if (wbc->nr_to_write <= 0)
1924			done = true;
1925
1926		index = next;
1927	}
1928
1929	if (!scanned && !done) {
1930		/*
1931		 * We hit the last page and there is more work to be done: wrap
1932		 * back to the start of the file
1933		 */
1934		scanned = true;
1935		index = 0;
1936		goto retry;
1937	}
1938
1939	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1940		mapping->writeback_index = index;
1941
1942	return rc;
1943}
1944
1945static int
1946cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
1947{
1948	int rc;
1949	int xid;
1950
1951	xid = GetXid();
1952/* BB add check for wbc flags */
1953	page_cache_get(page);
1954	if (!PageUptodate(page))
1955		cFYI(1, "ppw - page not up to date");
1956
1957	/*
1958	 * Set the "writeback" flag, and clear "dirty" in the radix tree.
1959	 *
1960	 * A writepage() implementation always needs to do either this,
1961	 * or re-dirty the page with "redirty_page_for_writepage()" in
1962	 * the case of a failure.
1963	 *
1964	 * Just unlocking the page will cause the radix tree tag-bits
1965	 * to fail to update with the state of the page correctly.
1966	 */
1967	set_page_writeback(page);
1968retry_write:
1969	rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1970	if (rc == -EAGAIN && wbc->sync_mode == WB_SYNC_ALL)
1971		goto retry_write;
1972	else if (rc == -EAGAIN)
1973		redirty_page_for_writepage(wbc, page);
1974	else if (rc != 0)
1975		SetPageError(page);
1976	else
 
1977		SetPageUptodate(page);
 
1978	end_page_writeback(page);
1979	page_cache_release(page);
1980	FreeXid(xid);
1981	return rc;
1982}
1983
1984static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1985{
1986	int rc = cifs_writepage_locked(page, wbc);
1987	unlock_page(page);
1988	return rc;
1989}
1990
1991static int cifs_write_end(struct file *file, struct address_space *mapping,
1992			loff_t pos, unsigned len, unsigned copied,
1993			struct page *page, void *fsdata)
1994{
1995	int rc;
1996	struct inode *inode = mapping->host;
1997	struct cifsFileInfo *cfile = file->private_data;
1998	struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
1999	__u32 pid;
2000
2001	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2002		pid = cfile->pid;
2003	else
2004		pid = current->tgid;
2005
2006	cFYI(1, "write_end for page %p from pos %lld with %d bytes",
2007		 page, pos, copied);
2008
2009	if (PageChecked(page)) {
2010		if (copied == len)
2011			SetPageUptodate(page);
2012		ClearPageChecked(page);
2013	} else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
2014		SetPageUptodate(page);
2015
2016	if (!PageUptodate(page)) {
2017		char *page_data;
2018		unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
2019		int xid;
2020
2021		xid = GetXid();
2022		/* this is probably better than directly calling
2023		   partialpage_write since in this function the file handle is
2024		   known which we might as well	leverage */
2025		/* BB check if anything else missing out of ppw
2026		   such as updating last write time */
2027		page_data = kmap(page);
2028		rc = cifs_write(cfile, pid, page_data + offset, copied, &pos);
2029		/* if (rc < 0) should we set writebehind rc? */
2030		kunmap(page);
2031
2032		FreeXid(xid);
2033	} else {
2034		rc = copied;
2035		pos += copied;
2036		set_page_dirty(page);
2037	}
2038
2039	if (rc > 0) {
2040		spin_lock(&inode->i_lock);
2041		if (pos > inode->i_size)
2042			i_size_write(inode, pos);
2043		spin_unlock(&inode->i_lock);
2044	}
2045
2046	unlock_page(page);
2047	page_cache_release(page);
2048
2049	return rc;
2050}
2051
2052int cifs_strict_fsync(struct file *file, loff_t start, loff_t end,
2053		      int datasync)
2054{
2055	int xid;
2056	int rc = 0;
2057	struct cifs_tcon *tcon;
 
2058	struct cifsFileInfo *smbfile = file->private_data;
2059	struct inode *inode = file->f_path.dentry->d_inode;
2060	struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2061
2062	rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2063	if (rc)
2064		return rc;
2065	mutex_lock(&inode->i_mutex);
2066
2067	xid = GetXid();
2068
2069	cFYI(1, "Sync file - name: %s datasync: 0x%x",
2070		file->f_path.dentry->d_name.name, datasync);
2071
2072	if (!CIFS_I(inode)->clientCanCacheRead) {
2073		rc = cifs_invalidate_mapping(inode);
2074		if (rc) {
2075			cFYI(1, "rc: %d during invalidate phase", rc);
2076			rc = 0; /* don't care about it in fsync */
2077		}
2078	}
2079
2080	tcon = tlink_tcon(smbfile->tlink);
2081	if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
2082		rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
 
 
 
 
 
2083
2084	FreeXid(xid);
2085	mutex_unlock(&inode->i_mutex);
2086	return rc;
2087}
2088
2089int cifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2090{
2091	int xid;
2092	int rc = 0;
2093	struct cifs_tcon *tcon;
 
2094	struct cifsFileInfo *smbfile = file->private_data;
2095	struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2096	struct inode *inode = file->f_mapping->host;
2097
2098	rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2099	if (rc)
2100		return rc;
2101	mutex_lock(&inode->i_mutex);
2102
2103	xid = GetXid();
2104
2105	cFYI(1, "Sync file - name: %s datasync: 0x%x",
2106		file->f_path.dentry->d_name.name, datasync);
2107
2108	tcon = tlink_tcon(smbfile->tlink);
2109	if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
2110		rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
 
 
 
 
 
2111
2112	FreeXid(xid);
2113	mutex_unlock(&inode->i_mutex);
2114	return rc;
2115}
2116
2117/*
2118 * As file closes, flush all cached write data for this inode checking
2119 * for write behind errors.
2120 */
2121int cifs_flush(struct file *file, fl_owner_t id)
2122{
2123	struct inode *inode = file->f_path.dentry->d_inode;
2124	int rc = 0;
2125
2126	if (file->f_mode & FMODE_WRITE)
2127		rc = filemap_write_and_wait(inode->i_mapping);
2128
2129	cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
2130
2131	return rc;
2132}
2133
2134static int
2135cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
2136{
2137	int rc = 0;
2138	unsigned long i;
2139
2140	for (i = 0; i < num_pages; i++) {
2141		pages[i] = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2142		if (!pages[i]) {
2143			/*
2144			 * save number of pages we have already allocated and
2145			 * return with ENOMEM error
2146			 */
2147			num_pages = i;
2148			rc = -ENOMEM;
2149			break;
2150		}
2151	}
2152
2153	if (rc) {
2154		for (i = 0; i < num_pages; i++)
2155			put_page(pages[i]);
2156	}
2157	return rc;
2158}
2159
2160static inline
2161size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
2162{
2163	size_t num_pages;
2164	size_t clen;
2165
2166	clen = min_t(const size_t, len, wsize);
2167	num_pages = DIV_ROUND_UP(clen, PAGE_SIZE);
2168
2169	if (cur_len)
2170		*cur_len = clen;
2171
2172	return num_pages;
2173}
2174
2175static void
2176cifs_uncached_marshal_iov(struct kvec *iov, struct cifs_writedata *wdata)
2177{
2178	int i;
2179	size_t bytes = wdata->bytes;
 
2180
2181	/* marshal up the pages into iov array */
2182	for (i = 0; i < wdata->nr_pages; i++) {
2183		iov[i + 1].iov_len = min_t(size_t, bytes, PAGE_SIZE);
2184		iov[i + 1].iov_base = kmap(wdata->pages[i]);
2185		bytes -= iov[i + 1].iov_len;
2186	}
2187}
2188
 
 
2189static void
2190cifs_uncached_writev_complete(struct work_struct *work)
2191{
2192	int i;
2193	struct cifs_writedata *wdata = container_of(work,
2194					struct cifs_writedata, work);
2195	struct inode *inode = wdata->cfile->dentry->d_inode;
2196	struct cifsInodeInfo *cifsi = CIFS_I(inode);
2197
2198	spin_lock(&inode->i_lock);
2199	cifs_update_eof(cifsi, wdata->offset, wdata->bytes);
2200	if (cifsi->server_eof > inode->i_size)
2201		i_size_write(inode, cifsi->server_eof);
2202	spin_unlock(&inode->i_lock);
2203
2204	complete(&wdata->done);
2205
2206	if (wdata->result != -EAGAIN) {
2207		for (i = 0; i < wdata->nr_pages; i++)
2208			put_page(wdata->pages[i]);
2209	}
2210
2211	kref_put(&wdata->refcount, cifs_writedata_release);
2212}
2213
2214/* attempt to send write to server, retry on any -EAGAIN errors */
2215static int
2216cifs_uncached_retry_writev(struct cifs_writedata *wdata)
 
2217{
2218	int rc;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2219
2220	do {
2221		if (wdata->cfile->invalidHandle) {
2222			rc = cifs_reopen_file(wdata->cfile, false);
2223			if (rc != 0)
2224				continue;
2225		}
2226		rc = cifs_async_writev(wdata);
2227	} while (rc == -EAGAIN);
2228
2229	return rc;
 
 
 
 
 
2230}
2231
2232static ssize_t
2233cifs_iovec_write(struct file *file, const struct iovec *iov,
2234		 unsigned long nr_segs, loff_t *poffset)
 
 
2235{
2236	unsigned long nr_pages, i;
2237	size_t copied, len, cur_len;
2238	ssize_t total_written = 0;
2239	loff_t offset;
2240	struct iov_iter it;
2241	struct cifsFileInfo *open_file;
2242	struct cifs_tcon *tcon;
2243	struct cifs_sb_info *cifs_sb;
2244	struct cifs_writedata *wdata, *tmp;
2245	struct list_head wdata_list;
2246	int rc;
2247	pid_t pid;
2248
2249	len = iov_length(iov, nr_segs);
2250	if (!len)
2251		return 0;
2252
2253	rc = generic_write_checks(file, poffset, &len, 0);
2254	if (rc)
2255		return rc;
2256
2257	INIT_LIST_HEAD(&wdata_list);
2258	cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2259	open_file = file->private_data;
2260	tcon = tlink_tcon(open_file->tlink);
2261	offset = *poffset;
2262
2263	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2264		pid = open_file->pid;
2265	else
2266		pid = current->tgid;
2267
2268	iov_iter_init(&it, iov, nr_segs, len, 0);
 
2269	do {
2270		size_t save_len;
2271
2272		nr_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
 
 
 
 
 
2273		wdata = cifs_writedata_alloc(nr_pages,
2274					     cifs_uncached_writev_complete);
2275		if (!wdata) {
2276			rc = -ENOMEM;
 
2277			break;
2278		}
2279
2280		rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
2281		if (rc) {
2282			kfree(wdata);
 
2283			break;
2284		}
2285
2286		save_len = cur_len;
2287		for (i = 0; i < nr_pages; i++) {
2288			copied = min_t(const size_t, cur_len, PAGE_SIZE);
2289			copied = iov_iter_copy_from_user(wdata->pages[i], &it,
2290							 0, copied);
2291			cur_len -= copied;
2292			iov_iter_advance(&it, copied);
 
2293		}
2294		cur_len = save_len - cur_len;
 
 
 
 
 
 
2295
2296		wdata->sync_mode = WB_SYNC_ALL;
2297		wdata->nr_pages = nr_pages;
2298		wdata->offset = (__u64)offset;
2299		wdata->cfile = cifsFileInfo_get(open_file);
2300		wdata->pid = pid;
2301		wdata->bytes = cur_len;
2302		wdata->marshal_iov = cifs_uncached_marshal_iov;
2303		rc = cifs_uncached_retry_writev(wdata);
 
 
 
 
 
 
 
 
2304		if (rc) {
2305			kref_put(&wdata->refcount, cifs_writedata_release);
 
 
 
 
 
 
 
2306			break;
2307		}
2308
2309		list_add_tail(&wdata->list, &wdata_list);
2310		offset += cur_len;
2311		len -= cur_len;
2312	} while (len > 0);
2313
2314	/*
2315	 * If at least one write was successfully sent, then discard any rc
2316	 * value from the later writes. If the other write succeeds, then
2317	 * we'll end up returning whatever was written. If it fails, then
2318	 * we'll get a new rc value from that.
2319	 */
2320	if (!list_empty(&wdata_list))
2321		rc = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
2322
 
2323	/*
2324	 * Wait for and collect replies for any successful sends in order of
2325	 * increasing offset. Once an error is hit or we get a fatal signal
2326	 * while waiting, then return without waiting for any more replies.
2327	 */
2328restart_loop:
2329	list_for_each_entry_safe(wdata, tmp, &wdata_list, list) {
2330		if (!rc) {
2331			/* FIXME: freezable too? */
2332			rc = wait_for_completion_killable(&wdata->done);
2333			if (rc)
2334				rc = -EINTR;
2335			else if (wdata->result)
 
2336				rc = wdata->result;
2337			else
2338				total_written += wdata->bytes;
2339
2340			/* resend call if it's a retryable error */
2341			if (rc == -EAGAIN) {
2342				rc = cifs_uncached_retry_writev(wdata);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2343				goto restart_loop;
2344			}
2345		}
2346		list_del_init(&wdata->list);
2347		kref_put(&wdata->refcount, cifs_writedata_release);
2348	}
2349
2350	if (total_written > 0)
2351		*poffset += total_written;
 
 
 
2352
2353	cifs_stats_bytes_written(tcon, total_written);
2354	return total_written ? total_written : (ssize_t)rc;
 
 
 
 
 
 
2355}
2356
2357ssize_t cifs_user_writev(struct kiocb *iocb, const struct iovec *iov,
2358				unsigned long nr_segs, loff_t pos)
2359{
2360	ssize_t written;
2361	struct inode *inode;
2362
2363	inode = iocb->ki_filp->f_path.dentry->d_inode;
 
 
 
 
2364
2365	/*
2366	 * BB - optimize the way when signing is disabled. We can drop this
2367	 * extra memory-to-memory copying and use iovec buffers for constructing
2368	 * write request.
2369	 */
2370
2371	written = cifs_iovec_write(iocb->ki_filp, iov, nr_segs, &pos);
2372	if (written > 0) {
2373		CIFS_I(inode)->invalid_mapping = true;
2374		iocb->ki_pos = pos;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2375	}
2376
2377	return written;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2378}
2379
2380ssize_t cifs_strict_writev(struct kiocb *iocb, const struct iovec *iov,
2381			   unsigned long nr_segs, loff_t pos)
2382{
2383	struct inode *inode;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2384
2385	inode = iocb->ki_filp->f_path.dentry->d_inode;
 
 
 
 
 
 
 
 
2386
2387	if (CIFS_I(inode)->clientCanCacheAll)
2388		return generic_file_aio_write(iocb, iov, nr_segs, pos);
 
 
2389
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2390	/*
2391	 * In strict cache mode we need to write the data to the server exactly
2392	 * from the pos to pos+len-1 rather than flush all affected pages
2393	 * because it may cause a error with mandatory locks on these pages but
2394	 * not on the region from pos to ppos+len-1.
2395	 */
2396
2397	return cifs_user_writev(iocb, iov, nr_segs, pos);
 
 
 
 
 
 
 
 
 
 
 
 
 
2398}
2399
2400static struct cifs_readdata *
2401cifs_readdata_alloc(unsigned int nr_vecs, work_func_t complete)
2402{
2403	struct cifs_readdata *rdata;
2404
2405	rdata = kzalloc(sizeof(*rdata) +
2406			sizeof(struct kvec) * nr_vecs, GFP_KERNEL);
2407	if (rdata != NULL) {
2408		kref_init(&rdata->refcount);
2409		INIT_LIST_HEAD(&rdata->list);
2410		init_completion(&rdata->done);
2411		INIT_WORK(&rdata->work, complete);
2412		INIT_LIST_HEAD(&rdata->pages);
2413	}
 
2414	return rdata;
2415}
2416
2417void
2418cifs_readdata_release(struct kref *refcount)
2419{
2420	struct cifs_readdata *rdata = container_of(refcount,
2421					struct cifs_readdata, refcount);
2422
 
 
 
 
 
2423	if (rdata->cfile)
2424		cifsFileInfo_put(rdata->cfile);
2425
2426	kfree(rdata);
2427}
2428
2429static int
2430cifs_read_allocate_pages(struct list_head *list, unsigned int npages)
2431{
2432	int rc = 0;
2433	struct page *page, *tpage;
2434	unsigned int i;
2435
2436	for (i = 0; i < npages; i++) {
2437		page = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2438		if (!page) {
2439			rc = -ENOMEM;
2440			break;
2441		}
2442		list_add(&page->lru, list);
2443	}
2444
2445	if (rc) {
2446		list_for_each_entry_safe(page, tpage, list, lru) {
2447			list_del(&page->lru);
2448			put_page(page);
2449		}
2450	}
2451	return rc;
2452}
2453
2454static void
2455cifs_uncached_readdata_release(struct kref *refcount)
2456{
2457	struct page *page, *tpage;
2458	struct cifs_readdata *rdata = container_of(refcount,
2459					struct cifs_readdata, refcount);
 
2460
2461	list_for_each_entry_safe(page, tpage, &rdata->pages, lru) {
2462		list_del(&page->lru);
2463		put_page(page);
 
2464	}
2465	cifs_readdata_release(refcount);
2466}
2467
2468static int
2469cifs_retry_async_readv(struct cifs_readdata *rdata)
2470{
2471	int rc;
2472
2473	do {
2474		if (rdata->cfile->invalidHandle) {
2475			rc = cifs_reopen_file(rdata->cfile, true);
2476			if (rc != 0)
2477				continue;
2478		}
2479		rc = cifs_async_readv(rdata);
2480	} while (rc == -EAGAIN);
2481
2482	return rc;
2483}
2484
2485/**
2486 * cifs_readdata_to_iov - copy data from pages in response to an iovec
2487 * @rdata:	the readdata response with list of pages holding data
2488 * @iov:	vector in which we should copy the data
2489 * @nr_segs:	number of segments in vector
2490 * @offset:	offset into file of the first iovec
2491 * @copied:	used to return the amount of data copied to the iov
2492 *
2493 * This function copies data from a list of pages in a readdata response into
2494 * an array of iovecs. It will first calculate where the data should go
2495 * based on the info in the readdata and then copy the data into that spot.
2496 */
2497static ssize_t
2498cifs_readdata_to_iov(struct cifs_readdata *rdata, const struct iovec *iov,
2499			unsigned long nr_segs, loff_t offset, ssize_t *copied)
2500{
2501	int rc = 0;
2502	struct iov_iter ii;
2503	size_t pos = rdata->offset - offset;
2504	struct page *page, *tpage;
2505	ssize_t remaining = rdata->bytes;
2506	unsigned char *pdata;
 
2507
2508	/* set up iov_iter and advance to the correct offset */
2509	iov_iter_init(&ii, iov, nr_segs, iov_length(iov, nr_segs), 0);
2510	iov_iter_advance(&ii, pos);
2511
2512	*copied = 0;
2513	list_for_each_entry_safe(page, tpage, &rdata->pages, lru) {
2514		ssize_t copy;
2515
2516		/* copy a whole page or whatever's left */
2517		copy = min_t(ssize_t, remaining, PAGE_SIZE);
2518
2519		/* ...but limit it to whatever space is left in the iov */
2520		copy = min_t(ssize_t, copy, iov_iter_count(&ii));
2521
2522		/* go while there's data to be copied and no errors */
2523		if (copy && !rc) {
2524			pdata = kmap(page);
2525			rc = memcpy_toiovecend(ii.iov, pdata, ii.iov_offset,
2526						(int)copy);
2527			kunmap(page);
2528			if (!rc) {
2529				*copied += copy;
2530				remaining -= copy;
2531				iov_iter_advance(&ii, copy);
2532			}
2533		}
2534
2535		list_del(&page->lru);
2536		put_page(page);
 
 
 
 
 
2537	}
 
 
2538
2539	return rc;
2540}
2541
2542static void
2543cifs_uncached_readv_complete(struct work_struct *work)
2544{
2545	struct cifs_readdata *rdata = container_of(work,
2546						struct cifs_readdata, work);
2547
2548	/* if the result is non-zero then the pages weren't kmapped */
2549	if (rdata->result == 0) {
2550		struct page *page;
2551
2552		list_for_each_entry(page, &rdata->pages, lru)
2553			kunmap(page);
2554	}
2555
2556	complete(&rdata->done);
 
 
2557	kref_put(&rdata->refcount, cifs_uncached_readdata_release);
2558}
2559
2560static int
2561cifs_uncached_read_marshal_iov(struct cifs_readdata *rdata,
2562				unsigned int remaining)
 
2563{
2564	int len = 0;
2565	struct page *page, *tpage;
 
 
 
 
 
 
 
2566
2567	rdata->nr_iov = 1;
2568	list_for_each_entry_safe(page, tpage, &rdata->pages, lru) {
2569		if (remaining >= PAGE_SIZE) {
2570			/* enough data to fill the page */
2571			rdata->iov[rdata->nr_iov].iov_base = kmap(page);
2572			rdata->iov[rdata->nr_iov].iov_len = PAGE_SIZE;
2573			cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
2574				rdata->nr_iov, page->index,
2575				rdata->iov[rdata->nr_iov].iov_base,
2576				rdata->iov[rdata->nr_iov].iov_len);
2577			++rdata->nr_iov;
2578			len += PAGE_SIZE;
2579			remaining -= PAGE_SIZE;
2580		} else if (remaining > 0) {
2581			/* enough for partial page, fill and zero the rest */
2582			rdata->iov[rdata->nr_iov].iov_base = kmap(page);
2583			rdata->iov[rdata->nr_iov].iov_len = remaining;
2584			cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
2585				rdata->nr_iov, page->index,
2586				rdata->iov[rdata->nr_iov].iov_base,
2587				rdata->iov[rdata->nr_iov].iov_len);
2588			memset(rdata->iov[rdata->nr_iov].iov_base + remaining,
2589				'\0', PAGE_SIZE - remaining);
2590			++rdata->nr_iov;
2591			len += remaining;
2592			remaining = 0;
2593		} else {
2594			/* no need to hold page hostage */
2595			list_del(&page->lru);
 
2596			put_page(page);
 
2597		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2598	}
2599
2600	return len;
 
2601}
2602
2603static ssize_t
2604cifs_iovec_read(struct file *file, const struct iovec *iov,
2605		 unsigned long nr_segs, loff_t *poffset)
2606{
2607	ssize_t rc;
2608	size_t len, cur_len;
2609	ssize_t total_read = 0;
2610	loff_t offset = *poffset;
2611	unsigned int npages;
2612	struct cifs_sb_info *cifs_sb;
2613	struct cifs_tcon *tcon;
2614	struct cifsFileInfo *open_file;
2615	struct cifs_readdata *rdata, *tmp;
2616	struct list_head rdata_list;
2617	pid_t pid;
2618
2619	if (!nr_segs)
2620		return 0;
 
 
 
 
 
2621
2622	len = iov_length(iov, nr_segs);
2623	if (!len)
2624		return 0;
 
 
 
 
 
 
 
 
2625
2626	INIT_LIST_HEAD(&rdata_list);
2627	cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2628	open_file = file->private_data;
2629	tcon = tlink_tcon(open_file->tlink);
2630
2631	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2632		pid = open_file->pid;
2633	else
2634		pid = current->tgid;
2635
2636	if ((file->f_flags & O_ACCMODE) == O_WRONLY)
2637		cFYI(1, "attempting read on write only file instance");
 
 
 
2638
2639	do {
2640		cur_len = min_t(const size_t, len - total_read, cifs_sb->rsize);
2641		npages = DIV_ROUND_UP(cur_len, PAGE_SIZE);
2642
2643		/* allocate a readdata struct */
2644		rdata = cifs_readdata_alloc(npages,
2645					    cifs_uncached_readv_complete);
2646		if (!rdata) {
 
2647			rc = -ENOMEM;
2648			goto error;
2649		}
2650
2651		rc = cifs_read_allocate_pages(&rdata->pages, npages);
2652		if (rc)
2653			goto error;
2654
2655		rdata->cfile = cifsFileInfo_get(open_file);
 
2656		rdata->offset = offset;
2657		rdata->bytes = cur_len;
2658		rdata->pid = pid;
2659		rdata->marshal_iov = cifs_uncached_read_marshal_iov;
2660
2661		rc = cifs_retry_async_readv(rdata);
 
 
 
 
 
 
 
2662error:
2663		if (rc) {
 
2664			kref_put(&rdata->refcount,
2665				 cifs_uncached_readdata_release);
 
 
2666			break;
2667		}
2668
2669		list_add_tail(&rdata->list, &rdata_list);
2670		offset += cur_len;
2671		len -= cur_len;
2672	} while (len > 0);
2673
2674	/* if at least one read request send succeeded, then reset rc */
2675	if (!list_empty(&rdata_list))
2676		rc = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2677
 
2678	/* the loop below should proceed in the order of increasing offsets */
2679restart_loop:
2680	list_for_each_entry_safe(rdata, tmp, &rdata_list, list) {
2681		if (!rc) {
2682			ssize_t copied;
 
 
 
 
 
 
 
 
2683
2684			/* FIXME: freezable sleep too? */
2685			rc = wait_for_completion_killable(&rdata->done);
2686			if (rc)
2687				rc = -EINTR;
2688			else if (rdata->result)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2689				rc = rdata->result;
2690			else {
2691				rc = cifs_readdata_to_iov(rdata, iov,
2692							nr_segs, *poffset,
2693							&copied);
2694				total_read += copied;
2695			}
2696
2697			/* resend call if it's a retryable error */
2698			if (rc == -EAGAIN) {
2699				rc = cifs_retry_async_readv(rdata);
2700				goto restart_loop;
2701			}
2702		}
2703		list_del_init(&rdata->list);
2704		kref_put(&rdata->refcount, cifs_uncached_readdata_release);
2705	}
2706
2707	cifs_stats_bytes_read(tcon, total_read);
2708	*poffset += total_read;
 
 
 
 
 
 
 
 
 
 
 
 
 
2709
2710	return total_read ? total_read : rc;
 
 
 
 
 
2711}
2712
2713ssize_t cifs_user_readv(struct kiocb *iocb, const struct iovec *iov,
2714			       unsigned long nr_segs, loff_t pos)
2715{
2716	ssize_t read;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2717
2718	read = cifs_iovec_read(iocb->ki_filp, iov, nr_segs, &pos);
2719	if (read > 0)
2720		iocb->ki_pos = pos;
 
 
 
 
 
 
 
2721
2722	return read;
 
 
 
 
 
 
2723}
2724
2725ssize_t cifs_strict_readv(struct kiocb *iocb, const struct iovec *iov,
2726			  unsigned long nr_segs, loff_t pos)
2727{
2728	struct inode *inode;
2729
2730	inode = iocb->ki_filp->f_path.dentry->d_inode;
2731
2732	if (CIFS_I(inode)->clientCanCacheRead)
2733		return generic_file_aio_read(iocb, iov, nr_segs, pos);
 
2734
2735	/*
2736	 * In strict cache mode we need to read from the server all the time
2737	 * if we don't have level II oplock because the server can delay mtime
2738	 * change - so we can't make a decision about inode invalidating.
2739	 * And we can also fail with pagereading if there are mandatory locks
2740	 * on pages affected by this read but not on the region from pos to
2741	 * pos+len-1.
2742	 */
 
 
 
 
 
 
 
2743
2744	return cifs_user_readv(iocb, iov, nr_segs, pos);
 
 
 
 
 
 
 
 
 
 
2745}
2746
2747static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
2748			 loff_t *poffset)
2749{
2750	int rc = -EACCES;
2751	unsigned int bytes_read = 0;
2752	unsigned int total_read;
2753	unsigned int current_read_size;
2754	unsigned int rsize;
2755	struct cifs_sb_info *cifs_sb;
2756	struct cifs_tcon *pTcon;
2757	int xid;
2758	char *current_offset;
 
2759	struct cifsFileInfo *open_file;
2760	struct cifs_io_parms io_parms;
2761	int buf_type = CIFS_NO_BUFFER;
2762	__u32 pid;
2763
2764	xid = GetXid();
2765	cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2766
2767	/* FIXME: set up handlers for larger reads and/or convert to async */
2768	rsize = min_t(unsigned int, cifs_sb->rsize, CIFSMaxBufSize);
2769
2770	if (file->private_data == NULL) {
2771		rc = -EBADF;
2772		FreeXid(xid);
2773		return rc;
2774	}
2775	open_file = file->private_data;
2776	pTcon = tlink_tcon(open_file->tlink);
 
 
 
 
 
 
2777
2778	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2779		pid = open_file->pid;
2780	else
2781		pid = current->tgid;
2782
2783	if ((file->f_flags & O_ACCMODE) == O_WRONLY)
2784		cFYI(1, "attempting read on write only file instance");
2785
2786	for (total_read = 0, current_offset = read_data;
2787	     read_size > total_read;
2788	     total_read += bytes_read, current_offset += bytes_read) {
2789		current_read_size = min_t(uint, read_size - total_read, rsize);
2790
2791		/* For windows me and 9x we do not want to request more
2792		than it negotiated since it will refuse the read then */
2793		if ((pTcon->ses) &&
2794			!(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
2795			current_read_size = min_t(uint, current_read_size,
2796					CIFSMaxBufSize);
2797		}
2798		rc = -EAGAIN;
2799		while (rc == -EAGAIN) {
 
2800			if (open_file->invalidHandle) {
2801				rc = cifs_reopen_file(open_file, true);
2802				if (rc != 0)
2803					break;
2804			}
2805			io_parms.netfid = open_file->netfid;
2806			io_parms.pid = pid;
2807			io_parms.tcon = pTcon;
2808			io_parms.offset = *poffset;
2809			io_parms.length = current_read_size;
2810			rc = CIFSSMBRead(xid, &io_parms, &bytes_read,
2811					 &current_offset, &buf_type);
2812		}
 
 
2813		if (rc || (bytes_read == 0)) {
2814			if (total_read) {
2815				break;
2816			} else {
2817				FreeXid(xid);
2818				return rc;
2819			}
2820		} else {
2821			cifs_stats_bytes_read(pTcon, total_read);
2822			*poffset += bytes_read;
2823		}
2824	}
2825	FreeXid(xid);
2826	return total_read;
2827}
2828
2829/*
2830 * If the page is mmap'ed into a process' page tables, then we need to make
2831 * sure that it doesn't change while being written back.
2832 */
2833static int
2834cifs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2835{
2836	struct page *page = vmf->page;
2837
2838	lock_page(page);
2839	return VM_FAULT_LOCKED;
2840}
2841
2842static struct vm_operations_struct cifs_file_vm_ops = {
2843	.fault = filemap_fault,
 
2844	.page_mkwrite = cifs_page_mkwrite,
2845};
2846
2847int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
2848{
2849	int rc, xid;
2850	struct inode *inode = file->f_path.dentry->d_inode;
2851
2852	xid = GetXid();
2853
2854	if (!CIFS_I(inode)->clientCanCacheRead) {
2855		rc = cifs_invalidate_mapping(inode);
2856		if (rc)
2857			return rc;
2858	}
 
2859
2860	rc = generic_file_mmap(file, vma);
2861	if (rc == 0)
2862		vma->vm_ops = &cifs_file_vm_ops;
2863	FreeXid(xid);
2864	return rc;
2865}
2866
2867int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
2868{
2869	int rc, xid;
2870
2871	xid = GetXid();
 
2872	rc = cifs_revalidate_file(file);
2873	if (rc) {
2874		cFYI(1, "Validation prior to mmap failed, error=%d", rc);
2875		FreeXid(xid);
2876		return rc;
2877	}
2878	rc = generic_file_mmap(file, vma);
2879	if (rc == 0)
2880		vma->vm_ops = &cifs_file_vm_ops;
2881	FreeXid(xid);
 
2882	return rc;
2883}
2884
2885static void
2886cifs_readv_complete(struct work_struct *work)
2887{
 
2888	struct cifs_readdata *rdata = container_of(work,
2889						struct cifs_readdata, work);
2890	struct page *page, *tpage;
2891
2892	list_for_each_entry_safe(page, tpage, &rdata->pages, lru) {
2893		list_del(&page->lru);
 
 
2894		lru_cache_add_file(page);
2895
2896		if (rdata->result == 0) {
2897			kunmap(page);
2898			flush_dcache_page(page);
2899			SetPageUptodate(page);
2900		}
2901
2902		unlock_page(page);
2903
2904		if (rdata->result == 0)
 
2905			cifs_readpage_to_fscache(rdata->mapping->host, page);
2906
2907		page_cache_release(page);
 
 
 
2908	}
2909	kref_put(&rdata->refcount, cifs_readdata_release);
2910}
2911
2912static int
2913cifs_readpages_marshal_iov(struct cifs_readdata *rdata, unsigned int remaining)
 
 
2914{
2915	int len = 0;
2916	struct page *page, *tpage;
2917	u64 eof;
2918	pgoff_t eof_index;
 
2919
2920	/* determine the eof that the server (probably) has */
2921	eof = CIFS_I(rdata->mapping->host)->server_eof;
2922	eof_index = eof ? (eof - 1) >> PAGE_CACHE_SHIFT : 0;
2923	cFYI(1, "eof=%llu eof_index=%lu", eof, eof_index);
2924
2925	rdata->nr_iov = 1;
2926	list_for_each_entry_safe(page, tpage, &rdata->pages, lru) {
2927		if (remaining >= PAGE_CACHE_SIZE) {
2928			/* enough data to fill the page */
2929			rdata->iov[rdata->nr_iov].iov_base = kmap(page);
2930			rdata->iov[rdata->nr_iov].iov_len = PAGE_CACHE_SIZE;
2931			cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
2932				rdata->nr_iov, page->index,
2933				rdata->iov[rdata->nr_iov].iov_base,
2934				rdata->iov[rdata->nr_iov].iov_len);
2935			++rdata->nr_iov;
2936			len += PAGE_CACHE_SIZE;
2937			remaining -= PAGE_CACHE_SIZE;
2938		} else if (remaining > 0) {
2939			/* enough for partial page, fill and zero the rest */
2940			rdata->iov[rdata->nr_iov].iov_base = kmap(page);
2941			rdata->iov[rdata->nr_iov].iov_len = remaining;
2942			cFYI(1, "%u: idx=%lu iov_base=%p iov_len=%zu",
2943				rdata->nr_iov, page->index,
2944				rdata->iov[rdata->nr_iov].iov_base,
2945				rdata->iov[rdata->nr_iov].iov_len);
2946			memset(rdata->iov[rdata->nr_iov].iov_base + remaining,
2947				'\0', PAGE_CACHE_SIZE - remaining);
2948			++rdata->nr_iov;
2949			len += remaining;
2950			remaining = 0;
2951		} else if (page->index > eof_index) {
2952			/*
2953			 * The VFS will not try to do readahead past the
2954			 * i_size, but it's possible that we have outstanding
2955			 * writes with gaps in the middle and the i_size hasn't
2956			 * caught up yet. Populate those with zeroed out pages
2957			 * to prevent the VFS from repeatedly attempting to
2958			 * fill them until the writes are flushed.
2959			 */
2960			zero_user(page, 0, PAGE_CACHE_SIZE);
2961			list_del(&page->lru);
2962			lru_cache_add_file(page);
2963			flush_dcache_page(page);
2964			SetPageUptodate(page);
2965			unlock_page(page);
2966			page_cache_release(page);
 
 
 
2967		} else {
2968			/* no need to hold page hostage */
2969			list_del(&page->lru);
2970			lru_cache_add_file(page);
2971			unlock_page(page);
2972			page_cache_release(page);
 
 
 
2973		}
 
 
 
 
 
 
 
 
 
 
 
 
 
2974	}
2975
2976	return len;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2977}
2978
2979static int cifs_readpages(struct file *file, struct address_space *mapping,
2980	struct list_head *page_list, unsigned num_pages)
2981{
2982	int rc;
2983	struct list_head tmplist;
2984	struct cifsFileInfo *open_file = file->private_data;
2985	struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2986	unsigned int rsize = cifs_sb->rsize;
2987	pid_t pid;
2988
2989	/*
2990	 * Give up immediately if rsize is too small to read an entire page.
2991	 * The VFS will fall back to readpage. We should never reach this
2992	 * point however since we set ra_pages to 0 when the rsize is smaller
2993	 * than a cache page.
2994	 */
2995	if (unlikely(rsize < PAGE_CACHE_SIZE))
2996		return 0;
2997
2998	/*
2999	 * Reads as many pages as possible from fscache. Returns -ENOBUFS
3000	 * immediately if the cookie is negative
 
 
 
3001	 */
3002	rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
3003					 &num_pages);
3004	if (rc == 0)
3005		return rc;
3006
3007	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3008		pid = open_file->pid;
3009	else
3010		pid = current->tgid;
3011
3012	rc = 0;
3013	INIT_LIST_HEAD(&tmplist);
3014
3015	cFYI(1, "%s: file=%p mapping=%p num_pages=%u", __func__, file,
3016		mapping, num_pages);
3017
3018	/*
3019	 * Start with the page at end of list and move it to private
3020	 * list. Do the same with any following pages until we hit
3021	 * the rsize limit, hit an index discontinuity, or run out of
3022	 * pages. Issue the async read and then start the loop again
3023	 * until the list is empty.
3024	 *
3025	 * Note that list order is important. The page_list is in
3026	 * the order of declining indexes. When we put the pages in
3027	 * the rdata->pages, then we want them in increasing order.
3028	 */
3029	while (!list_empty(page_list)) {
3030		unsigned int bytes = PAGE_CACHE_SIZE;
3031		unsigned int expected_index;
3032		unsigned int nr_pages = 1;
3033		loff_t offset;
3034		struct page *page, *tpage;
3035		struct cifs_readdata *rdata;
 
3036
3037		page = list_entry(page_list->prev, struct page, lru);
 
 
 
3038
3039		/*
3040		 * Lock the page and put it in the cache. Since no one else
3041		 * should have access to this page, we're safe to simply set
3042		 * PG_locked without checking it first.
 
3043		 */
3044		__set_page_locked(page);
3045		rc = add_to_page_cache_locked(page, mapping,
3046					      page->index, GFP_KERNEL);
 
3047
3048		/* give up if we can't stick it in the cache */
 
3049		if (rc) {
3050			__clear_page_locked(page);
3051			break;
3052		}
3053
3054		/* move first page to the tmplist */
3055		offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
3056		list_move_tail(&page->lru, &tmplist);
3057
3058		/* now try and add more pages onto the request */
3059		expected_index = page->index + 1;
3060		list_for_each_entry_safe_reverse(page, tpage, page_list, lru) {
3061			/* discontinuity ? */
3062			if (page->index != expected_index)
3063				break;
3064
3065			/* would this page push the read over the rsize? */
3066			if (bytes + PAGE_CACHE_SIZE > rsize)
3067				break;
3068
3069			__set_page_locked(page);
3070			if (add_to_page_cache_locked(page, mapping,
3071						page->index, GFP_KERNEL)) {
3072				__clear_page_locked(page);
3073				break;
3074			}
3075			list_move_tail(&page->lru, &tmplist);
3076			bytes += PAGE_CACHE_SIZE;
3077			expected_index++;
3078			nr_pages++;
3079		}
3080
3081		rdata = cifs_readdata_alloc(nr_pages, cifs_readv_complete);
3082		if (!rdata) {
3083			/* best to give up if we're out of mem */
3084			list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3085				list_del(&page->lru);
3086				lru_cache_add_file(page);
3087				unlock_page(page);
3088				page_cache_release(page);
3089			}
3090			rc = -ENOMEM;
 
3091			break;
3092		}
3093
3094		rdata->cfile = cifsFileInfo_get(open_file);
3095		rdata->mapping = mapping;
3096		rdata->offset = offset;
3097		rdata->bytes = bytes;
3098		rdata->pid = pid;
3099		rdata->marshal_iov = cifs_readpages_marshal_iov;
3100		list_splice_init(&tmplist, &rdata->pages);
 
 
 
 
 
 
 
3101
3102		rc = cifs_retry_async_readv(rdata);
3103		if (rc != 0) {
3104			list_for_each_entry_safe(page, tpage, &rdata->pages,
3105						 lru) {
3106				list_del(&page->lru);
 
 
3107				lru_cache_add_file(page);
3108				unlock_page(page);
3109				page_cache_release(page);
3110			}
 
3111			kref_put(&rdata->refcount, cifs_readdata_release);
3112			break;
3113		}
3114
3115		kref_put(&rdata->refcount, cifs_readdata_release);
3116	}
3117
 
 
 
 
 
3118	return rc;
3119}
3120
 
 
 
3121static int cifs_readpage_worker(struct file *file, struct page *page,
3122	loff_t *poffset)
3123{
3124	char *read_data;
3125	int rc;
3126
3127	/* Is the page cached? */
3128	rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
3129	if (rc == 0)
3130		goto read_complete;
3131
3132	page_cache_get(page);
3133	read_data = kmap(page);
3134	/* for reads over a certain size could initiate async read ahead */
3135
3136	rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
3137
3138	if (rc < 0)
3139		goto io_error;
3140	else
3141		cFYI(1, "Bytes read %d", rc);
3142
3143	file->f_path.dentry->d_inode->i_atime =
3144		current_fs_time(file->f_path.dentry->d_inode->i_sb);
3145
3146	if (PAGE_CACHE_SIZE > rc)
3147		memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
3148
3149	flush_dcache_page(page);
3150	SetPageUptodate(page);
3151
3152	/* send this page to the cache */
3153	cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
3154
3155	rc = 0;
3156
3157io_error:
3158	kunmap(page);
3159	page_cache_release(page);
3160
3161read_complete:
3162	return rc;
3163}
3164
3165static int cifs_readpage(struct file *file, struct page *page)
3166{
3167	loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
3168	int rc = -EACCES;
3169	int xid;
3170
3171	xid = GetXid();
3172
3173	if (file->private_data == NULL) {
3174		rc = -EBADF;
3175		FreeXid(xid);
3176		return rc;
3177	}
3178
3179	cFYI(1, "readpage %p at offset %d 0x%x\n",
3180		 page, (int)offset, (int)offset);
3181
3182	rc = cifs_readpage_worker(file, page, &offset);
3183
3184	unlock_page(page);
3185
3186	FreeXid(xid);
3187	return rc;
3188}
3189
3190static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
3191{
3192	struct cifsFileInfo *open_file;
 
 
3193
3194	spin_lock(&cifs_file_list_lock);
3195	list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
3196		if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
3197			spin_unlock(&cifs_file_list_lock);
3198			return 1;
3199		}
3200	}
3201	spin_unlock(&cifs_file_list_lock);
3202	return 0;
3203}
3204
3205/* We do not want to update the file size from server for inodes
3206   open for write - to avoid races with writepage extending
3207   the file - in the future we could consider allowing
3208   refreshing the inode only on increases in the file size
3209   but this is tricky to do without racing with writebehind
3210   page caching in the current Linux kernel design */
3211bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
3212{
3213	if (!cifsInode)
3214		return true;
3215
3216	if (is_inode_writable(cifsInode)) {
3217		/* This inode is open for write at least once */
3218		struct cifs_sb_info *cifs_sb;
3219
3220		cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
3221		if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
3222			/* since no page cache to corrupt on directio
3223			we can change size safely */
3224			return true;
3225		}
3226
3227		if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
3228			return true;
3229
3230		return false;
3231	} else
3232		return true;
3233}
3234
3235static int cifs_write_begin(struct file *file, struct address_space *mapping,
3236			loff_t pos, unsigned len, unsigned flags,
3237			struct page **pagep, void **fsdata)
3238{
3239	pgoff_t index = pos >> PAGE_CACHE_SHIFT;
3240	loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
 
3241	loff_t page_start = pos & PAGE_MASK;
3242	loff_t i_size;
3243	struct page *page;
3244	int rc = 0;
3245
3246	cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
3247
 
3248	page = grab_cache_page_write_begin(mapping, index, flags);
3249	if (!page) {
3250		rc = -ENOMEM;
3251		goto out;
3252	}
3253
3254	if (PageUptodate(page))
3255		goto out;
3256
3257	/*
3258	 * If we write a full page it will be up to date, no need to read from
3259	 * the server. If the write is short, we'll end up doing a sync write
3260	 * instead.
3261	 */
3262	if (len == PAGE_CACHE_SIZE)
3263		goto out;
3264
3265	/*
3266	 * optimize away the read when we have an oplock, and we're not
3267	 * expecting to use any of the data we'd be reading in. That
3268	 * is, when the page lies beyond the EOF, or straddles the EOF
3269	 * and the write will cover all of the existing data.
3270	 */
3271	if (CIFS_I(mapping->host)->clientCanCacheRead) {
3272		i_size = i_size_read(mapping->host);
3273		if (page_start >= i_size ||
3274		    (offset == 0 && (pos + len) >= i_size)) {
3275			zero_user_segments(page, 0, offset,
3276					   offset + len,
3277					   PAGE_CACHE_SIZE);
3278			/*
3279			 * PageChecked means that the parts of the page
3280			 * to which we're not writing are considered up
3281			 * to date. Once the data is copied to the
3282			 * page, it can be set uptodate.
3283			 */
3284			SetPageChecked(page);
3285			goto out;
3286		}
3287	}
3288
3289	if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
3290		/*
3291		 * might as well read a page, it is fast enough. If we get
3292		 * an error, we don't need to return it. cifs_write_end will
3293		 * do a sync write instead since PG_uptodate isn't set.
3294		 */
3295		cifs_readpage_worker(file, page, &page_start);
 
 
 
3296	} else {
3297		/* we could try using another file handle if there is one -
3298		   but how would we lock it to prevent close of that handle
3299		   racing with this read? In any case
3300		   this will be written out by write_end so is fine */
3301	}
3302out:
3303	*pagep = page;
3304	return rc;
3305}
3306
3307static int cifs_release_page(struct page *page, gfp_t gfp)
3308{
3309	if (PagePrivate(page))
3310		return 0;
3311
3312	return cifs_fscache_release_page(page, gfp);
3313}
3314
3315static void cifs_invalidate_page(struct page *page, unsigned long offset)
 
3316{
3317	struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
3318
3319	if (offset == 0)
3320		cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
3321}
3322
3323static int cifs_launder_page(struct page *page)
3324{
3325	int rc = 0;
3326	loff_t range_start = page_offset(page);
3327	loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
3328	struct writeback_control wbc = {
3329		.sync_mode = WB_SYNC_ALL,
3330		.nr_to_write = 0,
3331		.range_start = range_start,
3332		.range_end = range_end,
3333	};
3334
3335	cFYI(1, "Launder page: %p", page);
3336
3337	if (clear_page_dirty_for_io(page))
3338		rc = cifs_writepage_locked(page, &wbc);
3339
3340	cifs_fscache_invalidate_page(page, page->mapping->host);
3341	return rc;
3342}
3343
3344void cifs_oplock_break(struct work_struct *work)
3345{
3346	struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
3347						  oplock_break);
3348	struct inode *inode = cfile->dentry->d_inode;
3349	struct cifsInodeInfo *cinode = CIFS_I(inode);
 
 
3350	int rc = 0;
3351
 
 
 
 
 
 
 
 
 
 
 
 
 
3352	if (inode && S_ISREG(inode->i_mode)) {
3353		if (cinode->clientCanCacheRead)
3354			break_lease(inode, O_RDONLY);
3355		else
3356			break_lease(inode, O_WRONLY);
3357		rc = filemap_fdatawrite(inode->i_mapping);
3358		if (cinode->clientCanCacheRead == 0) {
3359			rc = filemap_fdatawait(inode->i_mapping);
3360			mapping_set_error(inode->i_mapping, rc);
3361			invalidate_remote_inode(inode);
3362		}
3363		cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
3364	}
3365
3366	rc = cifs_push_locks(cfile);
3367	if (rc)
3368		cERROR(1, "Push locks rc = %d", rc);
3369
3370	/*
3371	 * releasing stale oplock after recent reconnect of smb session using
3372	 * a now incorrect file handle is not a data integrity issue but do
3373	 * not bother sending an oplock release if session to server still is
3374	 * disconnected since oplock already released by the server
3375	 */
3376	if (!cfile->oplock_break_cancelled) {
3377		rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid,
3378				 current->tgid, 0, 0, 0, 0,
3379				 LOCKING_ANDX_OPLOCK_RELEASE, false,
3380				 cinode->clientCanCacheRead ? 1 : 0);
3381		cFYI(1, "Oplock release rc = %d", rc);
3382	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3383}
3384
 
3385const struct address_space_operations cifs_addr_ops = {
3386	.readpage = cifs_readpage,
3387	.readpages = cifs_readpages,
3388	.writepage = cifs_writepage,
3389	.writepages = cifs_writepages,
3390	.write_begin = cifs_write_begin,
3391	.write_end = cifs_write_end,
3392	.set_page_dirty = __set_page_dirty_nobuffers,
3393	.releasepage = cifs_release_page,
 
3394	.invalidatepage = cifs_invalidate_page,
3395	.launder_page = cifs_launder_page,
3396};
3397
3398/*
3399 * cifs_readpages requires the server to support a buffer large enough to
3400 * contain the header plus one complete page of data.  Otherwise, we need
3401 * to leave cifs_readpages out of the address space operations.
3402 */
3403const struct address_space_operations cifs_addr_ops_smallbuf = {
3404	.readpage = cifs_readpage,
3405	.writepage = cifs_writepage,
3406	.writepages = cifs_writepages,
3407	.write_begin = cifs_write_begin,
3408	.write_end = cifs_write_end,
3409	.set_page_dirty = __set_page_dirty_nobuffers,
3410	.releasepage = cifs_release_page,
3411	.invalidatepage = cifs_invalidate_page,
3412	.launder_page = cifs_launder_page,
3413};