fs/cifs/file.c

   1 /*
   2  *   fs/cifs/file.c
   3  *
   4  *   vfs operations that deal with files
   5  *
   6  *   Copyright (C) International Business Machines  Corp., 2002,2007
   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 <asm/div64.h>
  36 #include "cifsfs.h"
  37 #include "cifspdu.h"
  38 #include "cifsglob.h"
  39 #include "cifsproto.h"
  40 #include "cifs_unicode.h"
  41 #include "cifs_debug.h"
  42 #include "cifs_fs_sb.h"
  43
  44 static inline int cifs_convert_flags(unsigned int flags)
  45 {
  46         if ((flags & O_ACCMODE) == O_RDONLY)
  47                 return GENERIC_READ;
  48         else if ((flags & O_ACCMODE) == O_WRONLY)
  49                 return GENERIC_WRITE;
  50         else if ((flags & O_ACCMODE) == O_RDWR) {
  51                 /* GENERIC_ALL is too much permission to request
  52                    can cause unnecessary access denied on create */
  53                 /* return GENERIC_ALL; */
  54                 return (GENERIC_READ | GENERIC_WRITE);
  55         }
  56
  57         return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
  58                 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
  59                 FILE_READ_DATA);
  60 }
  61
  62 static inline fmode_t cifs_posix_convert_flags(unsigned int flags)
  63 {
  64         fmode_t posix_flags = 0;
  65
  66         if ((flags & O_ACCMODE) == O_RDONLY)
  67                 posix_flags = FMODE_READ;
  68         else if ((flags & O_ACCMODE) == O_WRONLY)
  69                 posix_flags = FMODE_WRITE;
  70         else if ((flags & O_ACCMODE) == O_RDWR) {
  71                 /* GENERIC_ALL is too much permission to request
  72                    can cause unnecessary access denied on create */
  73                 /* return GENERIC_ALL; */
  74                 posix_flags = FMODE_READ | FMODE_WRITE;
  75         }
  76         /* can not map O_CREAT or O_EXCL or O_TRUNC flags when
  77            reopening a file.  They had their effect on the original open */
  78         if (flags & O_APPEND)
  79                 posix_flags |= (fmode_t)O_APPEND;
  80         if (flags & O_DSYNC)
  81                 posix_flags |= (fmode_t)O_DSYNC;
  82         if (flags & __O_SYNC)
  83                 posix_flags |= (fmode_t)__O_SYNC;
  84         if (flags & O_DIRECTORY)
  85                 posix_flags |= (fmode_t)O_DIRECTORY;
  86         if (flags & O_NOFOLLOW)
  87                 posix_flags |= (fmode_t)O_NOFOLLOW;
  88         if (flags & O_DIRECT)
  89                 posix_flags |= (fmode_t)O_DIRECT;
  90
  91         return posix_flags;
  92 }
  93
  94 static 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
 108 /* all arguments to this function must be checked for validity in caller */
 109 static inline int
 110 cifs_posix_open_inode_helper(struct inode *inode, struct file *file,
 111                              struct cifsInodeInfo *pCifsInode,
 112                              struct cifsFileInfo *pCifsFile, __u32 oplock,
 113                              u16 netfid)
 114 {
 115
 116         write_lock(&GlobalSMBSeslock);
 117
 118         pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
 119         if (pCifsInode == NULL) {
 120                 write_unlock(&GlobalSMBSeslock);
 121                 return -EINVAL;
 122         }
 123
 124         if (pCifsInode->clientCanCacheRead) {
 125                 /* we have the inode open somewhere else
 126                    no need to discard cache data */
 127                 goto psx_client_can_cache;
 128         }
 129
 130         /* BB FIXME need to fix this check to move it earlier into posix_open
 131            BB  fIX following section BB FIXME */
 132
 133         /* if not oplocked, invalidate inode pages if mtime or file
 134            size changed */
 135 /*      temp = cifs_NTtimeToUnix(le64_to_cpu(buf->LastWriteTime));
 136         if (timespec_equal(&file->f_path.dentry->d_inode->i_mtime, &temp) &&
 137                            (file->f_path.dentry->d_inode->i_size ==
 138                             (loff_t)le64_to_cpu(buf->EndOfFile))) {
 139                 cFYI(1, ("inode unchanged on server"));
 140         } else {
 141                 if (file->f_path.dentry->d_inode->i_mapping) {
 142                         rc = filemap_write_and_wait(file->f_path.dentry->d_inode->i_mapping);
 143                         if (rc != 0)
 144                                 CIFS_I(file->f_path.dentry->d_inode)->write_behind_rc = rc;
 145                 }
 146                 cFYI(1, ("invalidating remote inode since open detected it "
 147                          "changed"));
 148                 invalidate_remote_inode(file->f_path.dentry->d_inode);
 149         } */
 150
 151 psx_client_can_cache:
 152         if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
 153                 pCifsInode->clientCanCacheAll = true;
 154                 pCifsInode->clientCanCacheRead = true;
 155                 cFYI(1, ("Exclusive Oplock granted on inode %p",
 156                          file->f_path.dentry->d_inode));
 157         } else if ((oplock & 0xF) == OPLOCK_READ)
 158                 pCifsInode->clientCanCacheRead = true;
 159
 160         /* will have to change the unlock if we reenable the
 161            filemap_fdatawrite (which does not seem necessary */
 162         write_unlock(&GlobalSMBSeslock);
 163         return 0;
 164 }
 165
 166 static struct cifsFileInfo *
 167 cifs_fill_filedata(struct file *file)
 168 {
 169         struct list_head *tmp;
 170         struct cifsFileInfo *pCifsFile = NULL;
 171         struct cifsInodeInfo *pCifsInode = NULL;
 172
 173         /* search inode for this file and fill in file->private_data */
 174         pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
 175         read_lock(&GlobalSMBSeslock);
 176         list_for_each(tmp, &pCifsInode->openFileList) {
 177                 pCifsFile = list_entry(tmp, struct cifsFileInfo, flist);
 178                 if ((pCifsFile->pfile == NULL) &&
 179                     (pCifsFile->pid == current->tgid)) {
 180                         /* mode set in cifs_create */
 181
 182                         /* needed for writepage */
 183                         pCifsFile->pfile = file;
 184                         file->private_data = pCifsFile;
 185                         break;
 186                 }
 187         }
 188         read_unlock(&GlobalSMBSeslock);
 189
 190         if (file->private_data != NULL) {
 191                 return pCifsFile;
 192         } else if ((file->f_flags & O_CREAT) && (file->f_flags & O_EXCL))
 193                         cERROR(1, ("could not find file instance for "
 194                                    "new file %p", file));
 195         return NULL;
 196 }
 197
 198 /* all arguments to this function must be checked for validity in caller */
 199 static inline int cifs_open_inode_helper(struct inode *inode, struct file *file,
 200         struct cifsInodeInfo *pCifsInode, struct cifsFileInfo *pCifsFile,
 201         struct cifsTconInfo *pTcon, int *oplock, FILE_ALL_INFO *buf,
 202         char *full_path, int xid)
 203 {
 204         struct timespec temp;
 205         int rc;
 206
 207         if (pCifsInode->clientCanCacheRead) {
 208                 /* we have the inode open somewhere else
 209                    no need to discard cache data */
 210                 goto client_can_cache;
 211         }
 212
 213         /* BB need same check in cifs_create too? */
 214         /* if not oplocked, invalidate inode pages if mtime or file
 215            size changed */
 216         temp = cifs_NTtimeToUnix(buf->LastWriteTime);
 217         if (timespec_equal(&file->f_path.dentry->d_inode->i_mtime, &temp) &&
 218                            (file->f_path.dentry->d_inode->i_size ==
 219                             (loff_t)le64_to_cpu(buf->EndOfFile))) {
 220                 cFYI(1, ("inode unchanged on server"));
 221         } else {
 222                 if (file->f_path.dentry->d_inode->i_mapping) {
 223                         /* BB no need to lock inode until after invalidate
 224                         since namei code should already have it locked? */
 225                         rc = filemap_write_and_wait(file->f_path.dentry->d_inode->i_mapping);
 226                         if (rc != 0)
 227                                 CIFS_I(file->f_path.dentry->d_inode)->write_behind_rc = rc;
 228                 }
 229                 cFYI(1, ("invalidating remote inode since open detected it "
 230                          "changed"));
 231                 invalidate_remote_inode(file->f_path.dentry->d_inode);
 232         }
 233
 234 client_can_cache:
 235         if (pTcon->unix_ext)
 236                 rc = cifs_get_inode_info_unix(&file->f_path.dentry->d_inode,
 237                         full_path, inode->i_sb, xid);
 238         else
 239                 rc = cifs_get_inode_info(&file->f_path.dentry->d_inode,
 240                         full_path, buf, inode->i_sb, xid, NULL);
 241
 242         if ((*oplock & 0xF) == OPLOCK_EXCLUSIVE) {
 243                 pCifsInode->clientCanCacheAll = true;
 244                 pCifsInode->clientCanCacheRead = true;
 245                 cFYI(1, ("Exclusive Oplock granted on inode %p",
 246                          file->f_path.dentry->d_inode));
 247         } else if ((*oplock & 0xF) == OPLOCK_READ)
 248                 pCifsInode->clientCanCacheRead = true;
 249
 250         return rc;
 251 }
 252
 253 int cifs_open(struct inode *inode, struct file *file)
 254 {
 255         int rc = -EACCES;
 256         int xid;
 257         __u32 oplock;
 258         struct cifs_sb_info *cifs_sb;
 259         struct cifsTconInfo *tcon;
 260         struct cifsFileInfo *pCifsFile;
 261         struct cifsInodeInfo *pCifsInode;
 262         char *full_path = NULL;
 263         int desiredAccess;
 264         int disposition;
 265         __u16 netfid;
 266         FILE_ALL_INFO *buf = NULL;
 267
 268         xid = GetXid();
 269
 270         cifs_sb = CIFS_SB(inode->i_sb);
 271         tcon = cifs_sb->tcon;
 272
 273         pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
 274         pCifsFile = cifs_fill_filedata(file);
 275         if (pCifsFile) {
 276                 rc = 0;
 277                 FreeXid(xid);
 278                 return rc;
 279         }
 280
 281         full_path = build_path_from_dentry(file->f_path.dentry);
 282         if (full_path == NULL) {
 283                 rc = -ENOMEM;
 284                 FreeXid(xid);
 285                 return rc;
 286         }
 287
 288         cFYI(1, ("inode = 0x%p file flags are 0x%x for %s",
 289                  inode, file->f_flags, full_path));
 290
 291         if (oplockEnabled)
 292                 oplock = REQ_OPLOCK;
 293         else
 294                 oplock = 0;
 295
 296         if (!tcon->broken_posix_open && tcon->unix_ext &&
 297             (tcon->ses->capabilities & CAP_UNIX) &&
 298             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 299                         le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 300                 int oflags = (int) cifs_posix_convert_flags(file->f_flags);
 301                 /* can not refresh inode info since size could be stale */
 302                 rc = cifs_posix_open(full_path, &inode, file->f_path.mnt,
 303                                      cifs_sb->mnt_file_mode /* ignored */,
 304                                      oflags, &oplock, &netfid, xid);
 305                 if (rc == 0) {
 306                         cFYI(1, ("posix open succeeded"));
 307                         /* no need for special case handling of setting mode
 308                            on read only files needed here */
 309
 310                         pCifsFile = cifs_fill_filedata(file);
 311                         cifs_posix_open_inode_helper(inode, file, pCifsInode,
 312                                                      pCifsFile, oplock, netfid);
 313                         goto out;
 314                 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
 315                         if (tcon->ses->serverNOS)
 316                                 cERROR(1, ("server %s of type %s returned"
 317                                            " unexpected error on SMB posix open"
 318                                            ", disabling posix open support."
 319                                            " Check if server update available.",
 320                                            tcon->ses->serverName,
 321                                            tcon->ses->serverNOS));
 322                         tcon->broken_posix_open = true;
 323                 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
 324                          (rc != -EOPNOTSUPP)) /* path not found or net err */
 325                         goto out;
 326                 /* else fallthrough to retry open the old way on network i/o
 327                    or DFS errors */
 328         }
 329
 330         desiredAccess = cifs_convert_flags(file->f_flags);
 331
 332 /*********************************************************************
 333  *  open flag mapping table:
 334  *
 335  *      POSIX Flag            CIFS Disposition
 336  *      ----------            ----------------
 337  *      O_CREAT               FILE_OPEN_IF
 338  *      O_CREAT | O_EXCL      FILE_CREATE
 339  *      O_CREAT | O_TRUNC     FILE_OVERWRITE_IF
 340  *      O_TRUNC               FILE_OVERWRITE
 341  *      none of the above     FILE_OPEN
 342  *
 343  *      Note that there is not a direct match between disposition
 344  *      FILE_SUPERSEDE (ie create whether or not file exists although
 345  *      O_CREAT | O_TRUNC is similar but truncates the existing
 346  *      file rather than creating a new file as FILE_SUPERSEDE does
 347  *      (which uses the attributes / metadata passed in on open call)
 348  *?
 349  *?  O_SYNC is a reasonable match to CIFS writethrough flag
 350  *?  and the read write flags match reasonably.  O_LARGEFILE
 351  *?  is irrelevant because largefile support is always used
 352  *?  by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
 353  *       O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
 354  *********************************************************************/
 355
 356         disposition = cifs_get_disposition(file->f_flags);
 357
 358         /* BB pass O_SYNC flag through on file attributes .. BB */
 359
 360         /* Also refresh inode by passing in file_info buf returned by SMBOpen
 361            and calling get_inode_info with returned buf (at least helps
 362            non-Unix server case) */
 363
 364         /* BB we can not do this if this is the second open of a file
 365            and the first handle has writebehind data, we might be
 366            able to simply do a filemap_fdatawrite/filemap_fdatawait first */
 367         buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
 368         if (!buf) {
 369                 rc = -ENOMEM;
 370                 goto out;
 371         }
 372
 373         if (cifs_sb->tcon->ses->capabilities & CAP_NT_SMBS)
 374                 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
 375                          desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
 376                          cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
 377                                  & CIFS_MOUNT_MAP_SPECIAL_CHR);
 378         else
 379                 rc = -EIO; /* no NT SMB support fall into legacy open below */
 380
 381         if (rc == -EIO) {
 382                 /* Old server, try legacy style OpenX */
 383                 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
 384                         desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
 385                         cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
 386                                 & CIFS_MOUNT_MAP_SPECIAL_CHR);
 387         }
 388         if (rc) {
 389                 cFYI(1, ("cifs_open returned 0x%x", rc));
 390                 goto out;
 391         }
 392
 393         pCifsFile = cifs_new_fileinfo(inode, netfid, file, file->f_path.mnt,
 394                                         file->f_flags);
 395         file->private_data = pCifsFile;
 396         if (file->private_data == NULL) {
 397                 rc = -ENOMEM;
 398                 goto out;
 399         }
 400
 401         rc = cifs_open_inode_helper(inode, file, pCifsInode, pCifsFile, tcon,
 402                                     &oplock, buf, full_path, xid);
 403
 404         if (oplock & CIFS_CREATE_ACTION) {
 405                 /* time to set mode which we can not set earlier due to
 406                    problems creating new read-only files */
 407                 if (tcon->unix_ext) {
 408                         struct cifs_unix_set_info_args args = {
 409                                 .mode   = inode->i_mode,
 410                                 .uid    = NO_CHANGE_64,
 411                                 .gid    = NO_CHANGE_64,
 412                                 .ctime  = NO_CHANGE_64,
 413                                 .atime  = NO_CHANGE_64,
 414                                 .mtime  = NO_CHANGE_64,
 415                                 .device = 0,
 416                         };
 417                         CIFSSMBUnixSetPathInfo(xid, tcon, full_path, &args,
 418                                                cifs_sb->local_nls,
 419                                                cifs_sb->mnt_cifs_flags &
 420                                                 CIFS_MOUNT_MAP_SPECIAL_CHR);
 421                 }
 422         }
 423
 424 out:
 425         kfree(buf);
 426         kfree(full_path);
 427         FreeXid(xid);
 428         return rc;
 429 }
 430
 431 /* Try to reacquire byte range locks that were released when session */
 432 /* to server was lost */
 433 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
 434 {
 435         int rc = 0;
 436
 437 /* BB list all locks open on this file and relock */
 438
 439         return rc;
 440 }
 441
 442 static int cifs_reopen_file(struct file *file, bool can_flush)
 443 {
 444         int rc = -EACCES;
 445         int xid;
 446         __u32 oplock;
 447         struct cifs_sb_info *cifs_sb;
 448         struct cifsTconInfo *tcon;
 449         struct cifsFileInfo *pCifsFile;
 450         struct cifsInodeInfo *pCifsInode;
 451         struct inode *inode;
 452         char *full_path = NULL;
 453         int desiredAccess;
 454         int disposition = FILE_OPEN;
 455         __u16 netfid;
 456
 457         if (file->private_data)
 458                 pCifsFile = (struct cifsFileInfo *)file->private_data;
 459         else
 460                 return -EBADF;
 461
 462         xid = GetXid();
 463         mutex_lock(&pCifsFile->fh_mutex);
 464         if (!pCifsFile->invalidHandle) {
 465                 mutex_unlock(&pCifsFile->fh_mutex);
 466                 rc = 0;
 467                 FreeXid(xid);
 468                 return rc;
 469         }
 470
 471         if (file->f_path.dentry == NULL) {
 472                 cERROR(1, ("no valid name if dentry freed"));
 473                 dump_stack();
 474                 rc = -EBADF;
 475                 goto reopen_error_exit;
 476         }
 477
 478         inode = file->f_path.dentry->d_inode;
 479         if (inode == NULL) {
 480                 cERROR(1, ("inode not valid"));
 481                 dump_stack();
 482                 rc = -EBADF;
 483                 goto reopen_error_exit;
 484         }
 485
 486         cifs_sb = CIFS_SB(inode->i_sb);
 487         tcon = cifs_sb->tcon;
 488
 489 /* can not grab rename sem here because various ops, including
 490    those that already have the rename sem can end up causing writepage
 491    to get called and if the server was down that means we end up here,
 492    and we can never tell if the caller already has the rename_sem */
 493         full_path = build_path_from_dentry(file->f_path.dentry);
 494         if (full_path == NULL) {
 495                 rc = -ENOMEM;
 496 reopen_error_exit:
 497                 mutex_unlock(&pCifsFile->fh_mutex);
 498                 FreeXid(xid);
 499                 return rc;
 500         }
 501
 502         cFYI(1, ("inode = 0x%p file flags 0x%x for %s",
 503                  inode, file->f_flags, full_path));
 504
 505         if (oplockEnabled)
 506                 oplock = REQ_OPLOCK;
 507         else
 508                 oplock = 0;
 509
 510         if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
 511             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 512                         le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 513                 int oflags = (int) cifs_posix_convert_flags(file->f_flags);
 514                 /* can not refresh inode info since size could be stale */
 515                 rc = cifs_posix_open(full_path, NULL, file->f_path.mnt,
 516                                      cifs_sb->mnt_file_mode /* ignored */,
 517                                      oflags, &oplock, &netfid, xid);
 518                 if (rc == 0) {
 519                         cFYI(1, ("posix reopen succeeded"));
 520                         goto reopen_success;
 521                 }
 522                 /* fallthrough to retry open the old way on errors, especially
 523                    in the reconnect path it is important to retry hard */
 524         }
 525
 526         desiredAccess = cifs_convert_flags(file->f_flags);
 527
 528         /* Can not refresh inode by passing in file_info buf to be returned
 529            by SMBOpen and then calling get_inode_info with returned buf
 530            since file might have write behind data that needs to be flushed
 531            and server version of file size can be stale. If we knew for sure
 532            that inode was not dirty locally we could do this */
 533
 534         rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
 535                          CREATE_NOT_DIR, &netfid, &oplock, NULL,
 536                          cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
 537                                 CIFS_MOUNT_MAP_SPECIAL_CHR);
 538         if (rc) {
 539                 mutex_unlock(&pCifsFile->fh_mutex);
 540                 cFYI(1, ("cifs_open returned 0x%x", rc));
 541                 cFYI(1, ("oplock: %d", oplock));
 542         } else {
 543 reopen_success:
 544                 pCifsFile->netfid = netfid;
 545                 pCifsFile->invalidHandle = false;
 546                 mutex_unlock(&pCifsFile->fh_mutex);
 547                 pCifsInode = CIFS_I(inode);
 548                 if (pCifsInode) {
 549                         if (can_flush) {
 550                                 rc = filemap_write_and_wait(inode->i_mapping);
 551                                 if (rc != 0)
 552                                         CIFS_I(inode)->write_behind_rc = rc;
 553                         /* temporarily disable caching while we
 554                            go to server to get inode info */
 555                                 pCifsInode->clientCanCacheAll = false;
 556                                 pCifsInode->clientCanCacheRead = false;
 557                                 if (tcon->unix_ext)
 558                                         rc = cifs_get_inode_info_unix(&inode,
 559                                                 full_path, inode->i_sb, xid);
 560                                 else
 561                                         rc = cifs_get_inode_info(&inode,
 562                                                 full_path, NULL, inode->i_sb,
 563                                                 xid, NULL);
 564                         } /* else we are writing out data to server already
 565                              and could deadlock if we tried to flush data, and
 566                              since we do not know if we have data that would
 567                              invalidate the current end of file on the server
 568                              we can not go to the server to get the new inod
 569                              info */
 570                         if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
 571                                 pCifsInode->clientCanCacheAll = true;
 572                                 pCifsInode->clientCanCacheRead = true;
 573                                 cFYI(1, ("Exclusive Oplock granted on inode %p",
 574                                          file->f_path.dentry->d_inode));
 575                         } else if ((oplock & 0xF) == OPLOCK_READ) {
 576                                 pCifsInode->clientCanCacheRead = true;
 577                                 pCifsInode->clientCanCacheAll = false;
 578                         } else {
 579                                 pCifsInode->clientCanCacheRead = false;
 580                                 pCifsInode->clientCanCacheAll = false;
 581                         }
 582                         cifs_relock_file(pCifsFile);
 583                 }
 584         }
 585         kfree(full_path);
 586         FreeXid(xid);
 587         return rc;
 588 }
 589
 590 int cifs_close(struct inode *inode, struct file *file)
 591 {
 592         int rc = 0;
 593         int xid, timeout;
 594         struct cifs_sb_info *cifs_sb;
 595         struct cifsTconInfo *pTcon;
 596         struct cifsFileInfo *pSMBFile =
 597                 (struct cifsFileInfo *)file->private_data;
 598
 599         xid = GetXid();
 600
 601         cifs_sb = CIFS_SB(inode->i_sb);
 602         pTcon = cifs_sb->tcon;
 603         if (pSMBFile) {
 604                 struct cifsLockInfo *li, *tmp;
 605                 write_lock(&GlobalSMBSeslock);
 606                 pSMBFile->closePend = true;
 607                 if (pTcon) {
 608                         /* no sense reconnecting to close a file that is
 609                            already closed */
 610                         if (!pTcon->need_reconnect) {
 611                                 write_unlock(&GlobalSMBSeslock);
 612                                 timeout = 2;
 613                                 while ((atomic_read(&pSMBFile->count) != 1)
 614                                         && (timeout <= 2048)) {
 615                                         /* Give write a better chance to get to
 616                                         server ahead of the close.  We do not
 617                                         want to add a wait_q here as it would
 618                                         increase the memory utilization as
 619                                         the struct would be in each open file,
 620                                         but this should give enough time to
 621                                         clear the socket */
 622                                         cFYI(DBG2,
 623                                                 ("close delay, write pending"));
 624                                         msleep(timeout);
 625                                         timeout *= 4;
 626                                 }
 627                                 if (!pTcon->need_reconnect &&
 628                                     !pSMBFile->invalidHandle)
 629                                         rc = CIFSSMBClose(xid, pTcon,
 630                                                   pSMBFile->netfid);
 631                         } else
 632                                 write_unlock(&GlobalSMBSeslock);
 633                 } else
 634                         write_unlock(&GlobalSMBSeslock);
 635
 636                 /* Delete any outstanding lock records.
 637                    We'll lose them when the file is closed anyway. */
 638                 mutex_lock(&pSMBFile->lock_mutex);
 639                 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
 640                         list_del(&li->llist);
 641                         kfree(li);
 642                 }
 643                 mutex_unlock(&pSMBFile->lock_mutex);
 644
 645                 write_lock(&GlobalSMBSeslock);
 646                 list_del(&pSMBFile->flist);
 647                 list_del(&pSMBFile->tlist);
 648                 write_unlock(&GlobalSMBSeslock);
 649                 cifsFileInfo_put(file->private_data);
 650                 file->private_data = NULL;
 651         } else
 652                 rc = -EBADF;
 653
 654         read_lock(&GlobalSMBSeslock);
 655         if (list_empty(&(CIFS_I(inode)->openFileList))) {
 656                 cFYI(1, ("closing last open instance for inode %p", inode));
 657                 /* if the file is not open we do not know if we can cache info
 658                    on this inode, much less write behind and read ahead */
 659                 CIFS_I(inode)->clientCanCacheRead = false;
 660                 CIFS_I(inode)->clientCanCacheAll  = false;
 661         }
 662         read_unlock(&GlobalSMBSeslock);
 663         if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
 664                 rc = CIFS_I(inode)->write_behind_rc;
 665         FreeXid(xid);
 666         return rc;
 667 }
 668
 669 int cifs_closedir(struct inode *inode, struct file *file)
 670 {
 671         int rc = 0;
 672         int xid;
 673         struct cifsFileInfo *pCFileStruct =
 674             (struct cifsFileInfo *)file->private_data;
 675         char *ptmp;
 676
 677         cFYI(1, ("Closedir inode = 0x%p", inode));
 678
 679         xid = GetXid();
 680
 681         if (pCFileStruct) {
 682                 struct cifsTconInfo *pTcon;
 683                 struct cifs_sb_info *cifs_sb =
 684                         CIFS_SB(file->f_path.dentry->d_sb);
 685
 686                 pTcon = cifs_sb->tcon;
 687
 688                 cFYI(1, ("Freeing private data in close dir"));
 689                 write_lock(&GlobalSMBSeslock);
 690                 if (!pCFileStruct->srch_inf.endOfSearch &&
 691                     !pCFileStruct->invalidHandle) {
 692                         pCFileStruct->invalidHandle = true;
 693                         write_unlock(&GlobalSMBSeslock);
 694                         rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
 695                         cFYI(1, ("Closing uncompleted readdir with rc %d",
 696                                  rc));
 697                         /* not much we can do if it fails anyway, ignore rc */
 698                         rc = 0;
 699                 } else
 700                         write_unlock(&GlobalSMBSeslock);
 701                 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
 702                 if (ptmp) {
 703                         cFYI(1, ("closedir free smb buf in srch struct"));
 704                         pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
 705                         if (pCFileStruct->srch_inf.smallBuf)
 706                                 cifs_small_buf_release(ptmp);
 707                         else
 708                                 cifs_buf_release(ptmp);
 709                 }
 710                 kfree(file->private_data);
 711                 file->private_data = NULL;
 712         }
 713         /* BB can we lock the filestruct while this is going on? */
 714         FreeXid(xid);
 715         return rc;
 716 }
 717
 718 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
 719                                 __u64 offset, __u8 lockType)
 720 {
 721         struct cifsLockInfo *li =
 722                 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
 723         if (li == NULL)
 724                 return -ENOMEM;
 725         li->offset = offset;
 726         li->length = len;
 727         li->type = lockType;
 728         mutex_lock(&fid->lock_mutex);
 729         list_add(&li->llist, &fid->llist);
 730         mutex_unlock(&fid->lock_mutex);
 731         return 0;
 732 }
 733
 734 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
 735 {
 736         int rc, xid;
 737         __u32 numLock = 0;
 738         __u32 numUnlock = 0;
 739         __u64 length;
 740         bool wait_flag = false;
 741         struct cifs_sb_info *cifs_sb;
 742         struct cifsTconInfo *tcon;
 743         __u16 netfid;
 744         __u8 lockType = LOCKING_ANDX_LARGE_FILES;
 745         bool posix_locking = 0;
 746
 747         length = 1 + pfLock->fl_end - pfLock->fl_start;
 748         rc = -EACCES;
 749         xid = GetXid();
 750
 751         cFYI(1, ("Lock parm: 0x%x flockflags: "
 752                  "0x%x flocktype: 0x%x start: %lld end: %lld",
 753                 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
 754                 pfLock->fl_end));
 755
 756         if (pfLock->fl_flags & FL_POSIX)
 757                 cFYI(1, ("Posix"));
 758         if (pfLock->fl_flags & FL_FLOCK)
 759                 cFYI(1, ("Flock"));
 760         if (pfLock->fl_flags & FL_SLEEP) {
 761                 cFYI(1, ("Blocking lock"));
 762                 wait_flag = true;
 763         }
 764         if (pfLock->fl_flags & FL_ACCESS)
 765                 cFYI(1, ("Process suspended by mandatory locking - "
 766                          "not implemented yet"));
 767         if (pfLock->fl_flags & FL_LEASE)
 768                 cFYI(1, ("Lease on file - not implemented yet"));
 769         if (pfLock->fl_flags &
 770             (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
 771                 cFYI(1, ("Unknown lock flags 0x%x", pfLock->fl_flags));
 772
 773         if (pfLock->fl_type == F_WRLCK) {
 774                 cFYI(1, ("F_WRLCK "));
 775                 numLock = 1;
 776         } else if (pfLock->fl_type == F_UNLCK) {
 777                 cFYI(1, ("F_UNLCK"));
 778                 numUnlock = 1;
 779                 /* Check if unlock includes more than
 780                 one lock range */
 781         } else if (pfLock->fl_type == F_RDLCK) {
 782                 cFYI(1, ("F_RDLCK"));
 783                 lockType |= LOCKING_ANDX_SHARED_LOCK;
 784                 numLock = 1;
 785         } else if (pfLock->fl_type == F_EXLCK) {
 786                 cFYI(1, ("F_EXLCK"));
 787                 numLock = 1;
 788         } else if (pfLock->fl_type == F_SHLCK) {
 789                 cFYI(1, ("F_SHLCK"));
 790                 lockType |= LOCKING_ANDX_SHARED_LOCK;
 791                 numLock = 1;
 792         } else
 793                 cFYI(1, ("Unknown type of lock"));
 794
 795         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
 796         tcon = cifs_sb->tcon;
 797
 798         if (file->private_data == NULL) {
 799                 rc = -EBADF;
 800                 FreeXid(xid);
 801                 return rc;
 802         }
 803         netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
 804
 805         if ((tcon->ses->capabilities & CAP_UNIX) &&
 806             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
 807             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
 808                 posix_locking = 1;
 809         /* BB add code here to normalize offset and length to
 810         account for negative length which we can not accept over the
 811         wire */
 812         if (IS_GETLK(cmd)) {
 813                 if (posix_locking) {
 814                         int posix_lock_type;
 815                         if (lockType & LOCKING_ANDX_SHARED_LOCK)
 816                                 posix_lock_type = CIFS_RDLCK;
 817                         else
 818                                 posix_lock_type = CIFS_WRLCK;
 819                         rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
 820                                         length, pfLock,
 821                                         posix_lock_type, wait_flag);
 822                         FreeXid(xid);
 823                         return rc;
 824                 }
 825
 826                 /* BB we could chain these into one lock request BB */
 827                 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
 828                                  0, 1, lockType, 0 /* wait flag */ );
 829                 if (rc == 0) {
 830                         rc = CIFSSMBLock(xid, tcon, netfid, length,
 831                                          pfLock->fl_start, 1 /* numUnlock */ ,
 832                                          0 /* numLock */ , lockType,
 833                                          0 /* wait flag */ );
 834                         pfLock->fl_type = F_UNLCK;
 835                         if (rc != 0)
 836                                 cERROR(1, ("Error unlocking previously locked "
 837                                            "range %d during test of lock", rc));
 838                         rc = 0;
 839
 840                 } else {
 841                         /* if rc == ERR_SHARING_VIOLATION ? */
 842                         rc = 0;
 843
 844                         if (lockType & LOCKING_ANDX_SHARED_LOCK) {
 845                                 pfLock->fl_type = F_WRLCK;
 846                         } else {
 847                                 rc = CIFSSMBLock(xid, tcon, netfid, length,
 848                                         pfLock->fl_start, 0, 1,
 849                                         lockType | LOCKING_ANDX_SHARED_LOCK,
 850                                         0 /* wait flag */);
 851                                 if (rc == 0) {
 852                                         rc = CIFSSMBLock(xid, tcon, netfid,
 853                                                 length, pfLock->fl_start, 1, 0,
 854                                                 lockType |
 855                                                 LOCKING_ANDX_SHARED_LOCK,
 856                                                 0 /* wait flag */);
 857                                         pfLock->fl_type = F_RDLCK;
 858                                         if (rc != 0)
 859                                                 cERROR(1, ("Error unlocking "
 860                                                 "previously locked range %d "
 861                                                 "during test of lock", rc));
 862                                         rc = 0;
 863                                 } else {
 864                                         pfLock->fl_type = F_WRLCK;
 865                                         rc = 0;
 866                                 }
 867                         }
 868                 }
 869
 870                 FreeXid(xid);
 871                 return rc;
 872         }
 873
 874         if (!numLock && !numUnlock) {
 875                 /* if no lock or unlock then nothing
 876                 to do since we do not know what it is */
 877                 FreeXid(xid);
 878                 return -EOPNOTSUPP;
 879         }
 880
 881         if (posix_locking) {
 882                 int posix_lock_type;
 883                 if (lockType & LOCKING_ANDX_SHARED_LOCK)
 884                         posix_lock_type = CIFS_RDLCK;
 885                 else
 886                         posix_lock_type = CIFS_WRLCK;
 887
 888                 if (numUnlock == 1)
 889                         posix_lock_type = CIFS_UNLCK;
 890
 891                 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
 892                                       length, pfLock,
 893                                       posix_lock_type, wait_flag);
 894         } else {
 895                 struct cifsFileInfo *fid =
 896                         (struct cifsFileInfo *)file->private_data;
 897
 898                 if (numLock) {
 899                         rc = CIFSSMBLock(xid, tcon, netfid, length,
 900                                         pfLock->fl_start,
 901                                         0, numLock, lockType, wait_flag);
 902
 903                         if (rc == 0) {
 904                                 /* For Windows locks we must store them. */
 905                                 rc = store_file_lock(fid, length,
 906                                                 pfLock->fl_start, lockType);
 907                         }
 908                 } else if (numUnlock) {
 909                         /* For each stored lock that this unlock overlaps
 910                            completely, unlock it. */
 911                         int stored_rc = 0;
 912                         struct cifsLockInfo *li, *tmp;
 913
 914                         rc = 0;
 915                         mutex_lock(&fid->lock_mutex);
 916                         list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
 917                                 if (pfLock->fl_start <= li->offset &&
 918                                                 (pfLock->fl_start + length) >=
 919                                                 (li->offset + li->length)) {
 920                                         stored_rc = CIFSSMBLock(xid, tcon,
 921                                                         netfid,
 922                                                         li->length, li->offset,
 923                                                         1, 0, li->type, false);
 924                                         if (stored_rc)
 925                                                 rc = stored_rc;
 926
 927                                         list_del(&li->llist);
 928                                         kfree(li);
 929                                 }
 930                         }
 931                         mutex_unlock(&fid->lock_mutex);
 932                 }
 933         }
 934
 935         if (pfLock->fl_flags & FL_POSIX)
 936                 posix_lock_file_wait(file, pfLock);
 937         FreeXid(xid);
 938         return rc;
 939 }
 940
 941 /*
 942  * Set the timeout on write requests past EOF. For some servers (Windows)
 943  * these calls can be very long.
 944  *
 945  * If we're writing >10M past the EOF we give a 180s timeout. Anything less
 946  * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
 947  * The 10M cutoff is totally arbitrary. A better scheme for this would be
 948  * welcome if someone wants to suggest one.
 949  *
 950  * We may be able to do a better job with this if there were some way to
 951  * declare that a file should be sparse.
 952  */
 953 static int
 954 cifs_write_timeout(struct cifsInodeInfo *cifsi, loff_t offset)
 955 {
 956         if (offset <= cifsi->server_eof)
 957                 return CIFS_STD_OP;
 958         else if (offset > (cifsi->server_eof + (10 * 1024 * 1024)))
 959                 return CIFS_VLONG_OP;
 960         else
 961                 return CIFS_LONG_OP;
 962 }
 963
 964 /* update the file size (if needed) after a write */
 965 static void
 966 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
 967                       unsigned int bytes_written)
 968 {
 969         loff_t end_of_write = offset + bytes_written;
 970
 971         if (end_of_write > cifsi->server_eof)
 972                 cifsi->server_eof = end_of_write;
 973 }
 974
 975 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
 976         size_t write_size, loff_t *poffset)
 977 {
 978         int rc = 0;
 979         unsigned int bytes_written = 0;
 980         unsigned int total_written;
 981         struct cifs_sb_info *cifs_sb;
 982         struct cifsTconInfo *pTcon;
 983         int xid, long_op;
 984         struct cifsFileInfo *open_file;
 985         struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
 986
 987         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
 988
 989         pTcon = cifs_sb->tcon;
 990
 991         /* cFYI(1,
 992            (" write %d bytes to offset %lld of %s", write_size,
 993            *poffset, file->f_path.dentry->d_name.name)); */
 994
 995         if (file->private_data == NULL)
 996                 return -EBADF;
 997         open_file = (struct cifsFileInfo *) file->private_data;
 998
 999         rc = generic_write_checks(file, poffset, &write_size, 0);
1000         if (rc)
1001                 return rc;
1002
1003         xid = GetXid();
1004
1005         long_op = cifs_write_timeout(cifsi, *poffset);
1006         for (total_written = 0; write_size > total_written;
1007              total_written += bytes_written) {
1008                 rc = -EAGAIN;
1009                 while (rc == -EAGAIN) {
1010                         if (file->private_data == NULL) {
1011                                 /* file has been closed on us */
1012                                 FreeXid(xid);
1013                         /* if we have gotten here we have written some data
1014                            and blocked, and the file has been freed on us while
1015                            we blocked so return what we managed to write */
1016                                 return total_written;
1017                         }
1018                         if (open_file->closePend) {
1019                                 FreeXid(xid);
1020                                 if (total_written)
1021                                         return total_written;
1022                                 else
1023                                         return -EBADF;
1024                         }
1025                         if (open_file->invalidHandle) {
1026                                 /* we could deadlock if we called
1027                                    filemap_fdatawait from here so tell
1028                                    reopen_file not to flush data to server
1029                                    now */
1030                                 rc = cifs_reopen_file(file, false);
1031                                 if (rc != 0)
1032                                         break;
1033                         }
1034
1035                         rc = CIFSSMBWrite(xid, pTcon,
1036                                 open_file->netfid,
1037                                 min_t(const int, cifs_sb->wsize,
1038                                       write_size - total_written),
1039                                 *poffset, &bytes_written,
1040                                 NULL, write_data + total_written, long_op);
1041                 }
1042                 if (rc || (bytes_written == 0)) {
1043                         if (total_written)
1044                                 break;
1045                         else {
1046                                 FreeXid(xid);
1047                                 return rc;
1048                         }
1049                 } else {
1050                         cifs_update_eof(cifsi, *poffset, bytes_written);
1051                         *poffset += bytes_written;
1052                 }
1053                 long_op = CIFS_STD_OP; /* subsequent writes fast -
1054                                     15 seconds is plenty */
1055         }
1056
1057         cifs_stats_bytes_written(pTcon, total_written);
1058
1059         /* since the write may have blocked check these pointers again */
1060         if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1061                 struct inode *inode = file->f_path.dentry->d_inode;
1062 /* Do not update local mtime - server will set its actual value on write
1063  *              inode->i_ctime = inode->i_mtime =
1064  *                      current_fs_time(inode->i_sb);*/
1065                 if (total_written > 0) {
1066                         spin_lock(&inode->i_lock);
1067                         if (*poffset > file->f_path.dentry->d_inode->i_size)
1068                                 i_size_write(file->f_path.dentry->d_inode,
1069                                         *poffset);
1070                         spin_unlock(&inode->i_lock);
1071                 }
1072                 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1073         }
1074         FreeXid(xid);
1075         return total_written;
1076 }
1077
1078 static ssize_t cifs_write(struct file *file, const char *write_data,
1079                           size_t write_size, loff_t *poffset)
1080 {
1081         int rc = 0;
1082         unsigned int bytes_written = 0;
1083         unsigned int total_written;
1084         struct cifs_sb_info *cifs_sb;
1085         struct cifsTconInfo *pTcon;
1086         int xid, long_op;
1087         struct cifsFileInfo *open_file;
1088         struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
1089
1090         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1091
1092         pTcon = cifs_sb->tcon;
1093
1094         cFYI(1, ("write %zd bytes to offset %lld of %s", write_size,
1095            *poffset, file->f_path.dentry->d_name.name));
1096
1097         if (file->private_data == NULL)
1098                 return -EBADF;
1099         open_file = (struct cifsFileInfo *)file->private_data;
1100
1101         xid = GetXid();
1102
1103         long_op = cifs_write_timeout(cifsi, *poffset);
1104         for (total_written = 0; write_size > total_written;
1105              total_written += bytes_written) {
1106                 rc = -EAGAIN;
1107                 while (rc == -EAGAIN) {
1108                         if (file->private_data == NULL) {
1109                                 /* file has been closed on us */
1110                                 FreeXid(xid);
1111                         /* if we have gotten here we have written some data
1112                            and blocked, and the file has been freed on us
1113                            while we blocked so return what we managed to
1114                            write */
1115                                 return total_written;
1116                         }
1117                         if (open_file->closePend) {
1118                                 FreeXid(xid);
1119                                 if (total_written)
1120                                         return total_written;
1121                                 else
1122                                         return -EBADF;
1123                         }
1124                         if (open_file->invalidHandle) {
1125                                 /* we could deadlock if we called
1126                                    filemap_fdatawait from here so tell
1127                                    reopen_file not to flush data to
1128                                    server now */
1129                                 rc = cifs_reopen_file(file, false);
1130                                 if (rc != 0)
1131                                         break;
1132                         }
1133                         if (experimEnabled || (pTcon->ses->server &&
1134                                 ((pTcon->ses->server->secMode &
1135                                 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1136                                 == 0))) {
1137                                 struct kvec iov[2];
1138                                 unsigned int len;
1139
1140                                 len = min((size_t)cifs_sb->wsize,
1141                                           write_size - total_written);
1142                                 /* iov[0] is reserved for smb header */
1143                                 iov[1].iov_base = (char *)write_data +
1144                                                   total_written;
1145                                 iov[1].iov_len = len;
1146                                 rc = CIFSSMBWrite2(xid, pTcon,
1147                                                 open_file->netfid, len,
1148                                                 *poffset, &bytes_written,
1149                                                 iov, 1, long_op);
1150                         } else
1151                                 rc = CIFSSMBWrite(xid, pTcon,
1152                                          open_file->netfid,
1153                                          min_t(const int, cifs_sb->wsize,
1154                                                write_size - total_written),
1155                                          *poffset, &bytes_written,
1156                                          write_data + total_written,
1157                                          NULL, long_op);
1158                 }
1159                 if (rc || (bytes_written == 0)) {
1160                         if (total_written)
1161                                 break;
1162                         else {
1163                                 FreeXid(xid);
1164                                 return rc;
1165                         }
1166                 } else {
1167                         cifs_update_eof(cifsi, *poffset, bytes_written);
1168                         *poffset += bytes_written;
1169                 }
1170                 long_op = CIFS_STD_OP; /* subsequent writes fast -
1171                                     15 seconds is plenty */
1172         }
1173
1174         cifs_stats_bytes_written(pTcon, total_written);
1175
1176         /* since the write may have blocked check these pointers again */
1177         if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1178 /*BB We could make this contingent on superblock ATIME flag too */
1179 /*              file->f_path.dentry->d_inode->i_ctime =
1180                 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
1181                 if (total_written > 0) {
1182                         spin_lock(&file->f_path.dentry->d_inode->i_lock);
1183                         if (*poffset > file->f_path.dentry->d_inode->i_size)
1184                                 i_size_write(file->f_path.dentry->d_inode,
1185                                              *poffset);
1186                         spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1187                 }
1188                 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1189         }
1190         FreeXid(xid);
1191         return total_written;
1192 }
1193
1194 #ifdef CONFIG_CIFS_EXPERIMENTAL
1195 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode)
1196 {
1197         struct cifsFileInfo *open_file = NULL;
1198
1199         read_lock(&GlobalSMBSeslock);
1200         /* we could simply get the first_list_entry since write-only entries
1201            are always at the end of the list but since the first entry might
1202            have a close pending, we go through the whole list */
1203         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1204                 if (open_file->closePend)
1205                         continue;
1206                 if (open_file->pfile && ((open_file->pfile->f_flags & O_RDWR) ||
1207                     (open_file->pfile->f_flags & O_RDONLY))) {
1208                         if (!open_file->invalidHandle) {
1209                                 /* found a good file */
1210                                 /* lock it so it will not be closed on us */
1211                                 cifsFileInfo_get(open_file);
1212                                 read_unlock(&GlobalSMBSeslock);
1213                                 return open_file;
1214                         } /* else might as well continue, and look for
1215                              another, or simply have the caller reopen it
1216                              again rather than trying to fix this handle */
1217                 } else /* write only file */
1218                         break; /* write only files are last so must be done */
1219         }
1220         read_unlock(&GlobalSMBSeslock);
1221         return NULL;
1222 }
1223 #endif
1224
1225 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1226 {
1227         struct cifsFileInfo *open_file;
1228         bool any_available = false;
1229         int rc;
1230
1231         /* Having a null inode here (because mapping->host was set to zero by
1232         the VFS or MM) should not happen but we had reports of on oops (due to
1233         it being zero) during stress testcases so we need to check for it */
1234
1235         if (cifs_inode == NULL) {
1236                 cERROR(1, ("Null inode passed to cifs_writeable_file"));
1237                 dump_stack();
1238                 return NULL;
1239         }
1240
1241         read_lock(&GlobalSMBSeslock);
1242 refind_writable:
1243         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1244                 if (open_file->closePend ||
1245                     (!any_available && open_file->pid != current->tgid))
1246                         continue;
1247
1248                 if (open_file->pfile &&
1249                     ((open_file->pfile->f_flags & O_RDWR) ||
1250                      (open_file->pfile->f_flags & O_WRONLY))) {
1251                         cifsFileInfo_get(open_file);
1252
1253                         if (!open_file->invalidHandle) {
1254                                 /* found a good writable file */
1255                                 read_unlock(&GlobalSMBSeslock);
1256                                 return open_file;
1257                         }
1258
1259                         read_unlock(&GlobalSMBSeslock);
1260                         /* Had to unlock since following call can block */
1261                         rc = cifs_reopen_file(open_file->pfile, false);
1262                         if (!rc) {
1263                                 if (!open_file->closePend)
1264                                         return open_file;
1265                                 else { /* start over in case this was deleted */
1266                                        /* since the list could be modified */
1267                                         read_lock(&GlobalSMBSeslock);
1268                                         cifsFileInfo_put(open_file);
1269                                         goto refind_writable;
1270                                 }
1271                         }
1272
1273                         /* if it fails, try another handle if possible -
1274                         (we can not do this if closePending since
1275                         loop could be modified - in which case we
1276                         have to start at the beginning of the list
1277                         again. Note that it would be bad
1278                         to hold up writepages here (rather than
1279                         in caller) with continuous retries */
1280                         cFYI(1, ("wp failed on reopen file"));
1281                         read_lock(&GlobalSMBSeslock);
1282                         /* can not use this handle, no write
1283                            pending on this one after all */
1284                         cifsFileInfo_put(open_file);
1285
1286                         if (open_file->closePend) /* list could have changed */
1287                                 goto refind_writable;
1288                         /* else we simply continue to the next entry. Thus
1289                            we do not loop on reopen errors.  If we
1290                            can not reopen the file, for example if we
1291                            reconnected to a server with another client
1292                            racing to delete or lock the file we would not
1293                            make progress if we restarted before the beginning
1294                            of the loop here. */
1295                 }
1296         }
1297         /* couldn't find useable FH with same pid, try any available */
1298         if (!any_available) {
1299                 any_available = true;
1300                 goto refind_writable;
1301         }
1302         read_unlock(&GlobalSMBSeslock);
1303         return NULL;
1304 }
1305
1306 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1307 {
1308         struct address_space *mapping = page->mapping;
1309         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1310         char *write_data;
1311         int rc = -EFAULT;
1312         int bytes_written = 0;
1313         struct cifs_sb_info *cifs_sb;
1314         struct cifsTconInfo *pTcon;
1315         struct inode *inode;
1316         struct cifsFileInfo *open_file;
1317
1318         if (!mapping || !mapping->host)
1319                 return -EFAULT;
1320
1321         inode = page->mapping->host;
1322         cifs_sb = CIFS_SB(inode->i_sb);
1323         pTcon = cifs_sb->tcon;
1324
1325         offset += (loff_t)from;
1326         write_data = kmap(page);
1327         write_data += from;
1328
1329         if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1330                 kunmap(page);
1331                 return -EIO;
1332         }
1333
1334         /* racing with truncate? */
1335         if (offset > mapping->host->i_size) {
1336                 kunmap(page);
1337                 return 0; /* don't care */
1338         }
1339
1340         /* check to make sure that we are not extending the file */
1341         if (mapping->host->i_size - offset < (loff_t)to)
1342                 to = (unsigned)(mapping->host->i_size - offset);
1343
1344         open_file = find_writable_file(CIFS_I(mapping->host));
1345         if (open_file) {
1346                 bytes_written = cifs_write(open_file->pfile, write_data,
1347                                            to-from, &offset);
1348                 cifsFileInfo_put(open_file);
1349                 /* Does mm or vfs already set times? */
1350                 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1351                 if ((bytes_written > 0) && (offset))
1352                         rc = 0;
1353                 else if (bytes_written < 0)
1354                         rc = bytes_written;
1355         } else {
1356                 cFYI(1, ("No writeable filehandles for inode"));
1357                 rc = -EIO;
1358         }
1359
1360         kunmap(page);
1361         return rc;
1362 }
1363
1364 static int cifs_writepages(struct address_space *mapping,
1365                            struct writeback_control *wbc)
1366 {
1367         struct backing_dev_info *bdi = mapping->backing_dev_info;
1368         unsigned int bytes_to_write;
1369         unsigned int bytes_written;
1370         struct cifs_sb_info *cifs_sb;
1371         int done = 0;
1372         pgoff_t end;
1373         pgoff_t index;
1374         int range_whole = 0;
1375         struct kvec *iov;
1376         int len;
1377         int n_iov = 0;
1378         pgoff_t next;
1379         int nr_pages;
1380         __u64 offset = 0;
1381         struct cifsFileInfo *open_file;
1382         struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1383         struct page *page;
1384         struct pagevec pvec;
1385         int rc = 0;
1386         int scanned = 0;
1387         int xid, long_op;
1388
1389         cifs_sb = CIFS_SB(mapping->host->i_sb);
1390
1391         /*
1392          * If wsize is smaller that the page cache size, default to writing
1393          * one page at a time via cifs_writepage
1394          */
1395         if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1396                 return generic_writepages(mapping, wbc);
1397
1398         if ((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1399                 if (cifs_sb->tcon->ses->server->secMode &
1400                                 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1401                         if (!experimEnabled)
1402                                 return generic_writepages(mapping, wbc);
1403
1404         iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1405         if (iov == NULL)
1406                 return generic_writepages(mapping, wbc);
1407
1408
1409         /*
1410          * BB: Is this meaningful for a non-block-device file system?
1411          * If it is, we should test it again after we do I/O
1412          */
1413         if (wbc->nonblocking && bdi_write_congested(bdi)) {
1414                 wbc->encountered_congestion = 1;
1415                 kfree(iov);
1416                 return 0;
1417         }
1418
1419         xid = GetXid();
1420
1421         pagevec_init(&pvec, 0);
1422         if (wbc->range_cyclic) {
1423                 index = mapping->writeback_index; /* Start from prev offset */
1424                 end = -1;
1425         } else {
1426                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1427                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1428                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1429                         range_whole = 1;
1430                 scanned = 1;
1431         }
1432 retry:
1433         while (!done && (index <= end) &&
1434                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1435                         PAGECACHE_TAG_DIRTY,
1436                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1437                 int first;
1438                 unsigned int i;
1439
1440                 first = -1;
1441                 next = 0;
1442                 n_iov = 0;
1443                 bytes_to_write = 0;
1444
1445                 for (i = 0; i < nr_pages; i++) {
1446                         page = pvec.pages[i];
1447                         /*
1448                          * At this point we hold neither mapping->tree_lock nor
1449                          * lock on the page itself: the page may be truncated or
1450                          * invalidated (changing page->mapping to NULL), or even
1451                          * swizzled back from swapper_space to tmpfs file
1452                          * mapping
1453                          */
1454
1455                         if (first < 0)
1456                                 lock_page(page);
1457                         else if (!trylock_page(page))
1458                                 break;
1459
1460                         if (unlikely(page->mapping != mapping)) {
1461                                 unlock_page(page);
1462                                 break;
1463                         }
1464
1465                         if (!wbc->range_cyclic && page->index > end) {
1466                                 done = 1;
1467                                 unlock_page(page);
1468                                 break;
1469                         }
1470
1471                         if (next && (page->index != next)) {
1472                                 /* Not next consecutive page */
1473                                 unlock_page(page);
1474                                 break;
1475                         }
1476
1477                         if (wbc->sync_mode != WB_SYNC_NONE)
1478                                 wait_on_page_writeback(page);
1479
1480                         if (PageWriteback(page) ||
1481                                         !clear_page_dirty_for_io(page)) {
1482                                 unlock_page(page);
1483                                 break;
1484                         }
1485
1486                         /*
1487                          * This actually clears the dirty bit in the radix tree.
1488                          * See cifs_writepage() for more commentary.
1489                          */
1490                         set_page_writeback(page);
1491
1492                         if (page_offset(page) >= mapping->host->i_size) {
1493                                 done = 1;
1494                                 unlock_page(page);
1495                                 end_page_writeback(page);
1496                                 break;
1497                         }
1498
1499                         /*
1500                          * BB can we get rid of this?  pages are held by pvec
1501                          */
1502                         page_cache_get(page);
1503
1504                         len = min(mapping->host->i_size - page_offset(page),
1505                                   (loff_t)PAGE_CACHE_SIZE);
1506
1507                         /* reserve iov[0] for the smb header */
1508                         n_iov++;
1509                         iov[n_iov].iov_base = kmap(page);
1510                         iov[n_iov].iov_len = len;
1511                         bytes_to_write += len;
1512
1513                         if (first < 0) {
1514                                 first = i;
1515                                 offset = page_offset(page);
1516                         }
1517                         next = page->index + 1;
1518                         if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1519                                 break;
1520                 }
1521                 if (n_iov) {
1522                         /* Search for a writable handle every time we call
1523                          * CIFSSMBWrite2.  We can't rely on the last handle
1524                          * we used to still be valid
1525                          */
1526                         open_file = find_writable_file(CIFS_I(mapping->host));
1527                         if (!open_file) {
1528                                 cERROR(1, ("No writable handles for inode"));
1529                                 rc = -EBADF;
1530                         } else {
1531                                 long_op = cifs_write_timeout(cifsi, offset);
1532                                 rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
1533                                                    open_file->netfid,
1534                                                    bytes_to_write, offset,
1535                                                    &bytes_written, iov, n_iov,
1536                                                    long_op);
1537                                 cifsFileInfo_put(open_file);
1538                                 cifs_update_eof(cifsi, offset, bytes_written);
1539
1540                                 if (rc || bytes_written < bytes_to_write) {
1541                                         cERROR(1, ("Write2 ret %d, wrote %d",
1542                                                   rc, bytes_written));
1543                                         /* BB what if continued retry is
1544                                            requested via mount flags? */
1545                                         if (rc == -ENOSPC)
1546                                                 set_bit(AS_ENOSPC, &mapping->flags);
1547                                         else
1548                                                 set_bit(AS_EIO, &mapping->flags);
1549                                 } else {
1550                                         cifs_stats_bytes_written(cifs_sb->tcon,
1551                                                                  bytes_written);
1552                                 }
1553                         }
1554                         for (i = 0; i < n_iov; i++) {
1555                                 page = pvec.pages[first + i];
1556                                 /* Should we also set page error on
1557                                 success rc but too little data written? */
1558                                 /* BB investigate retry logic on temporary
1559                                 server crash cases and how recovery works
1560                                 when page marked as error */
1561                                 if (rc)
1562                                         SetPageError(page);
1563                                 kunmap(page);
1564                                 unlock_page(page);
1565                                 end_page_writeback(page);
1566                                 page_cache_release(page);
1567                         }
1568                         if ((wbc->nr_to_write -= n_iov) <= 0)
1569                                 done = 1;
1570                         index = next;
1571                 } else
1572                         /* Need to re-find the pages we skipped */
1573                         index = pvec.pages[0]->index + 1;
1574
1575                 pagevec_release(&pvec);
1576         }
1577         if (!scanned && !done) {
1578                 /*
1579                  * We hit the last page and there is more work to be done: wrap
1580                  * back to the start of the file
1581                  */
1582                 scanned = 1;
1583                 index = 0;
1584                 goto retry;
1585         }
1586         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1587                 mapping->writeback_index = index;
1588
1589         FreeXid(xid);
1590         kfree(iov);
1591         return rc;
1592 }
1593
1594 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1595 {
1596         int rc = -EFAULT;
1597         int xid;
1598
1599         xid = GetXid();
1600 /* BB add check for wbc flags */
1601         page_cache_get(page);
1602         if (!PageUptodate(page))
1603                 cFYI(1, ("ppw - page not up to date"));
1604
1605         /*
1606          * Set the "writeback" flag, and clear "dirty" in the radix tree.
1607          *
1608          * A writepage() implementation always needs to do either this,
1609          * or re-dirty the page with "redirty_page_for_writepage()" in
1610          * the case of a failure.
1611          *
1612          * Just unlocking the page will cause the radix tree tag-bits
1613          * to fail to update with the state of the page correctly.
1614          */
1615         set_page_writeback(page);
1616         rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1617         SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1618         unlock_page(page);
1619         end_page_writeback(page);
1620         page_cache_release(page);
1621         FreeXid(xid);
1622         return rc;
1623 }
1624
1625 static int cifs_write_end(struct file *file, struct address_space *mapping,
1626                         loff_t pos, unsigned len, unsigned copied,
1627                         struct page *page, void *fsdata)
1628 {
1629         int rc;
1630         struct inode *inode = mapping->host;
1631
1632         cFYI(1, ("write_end for page %p from pos %lld with %d bytes",
1633                  page, pos, copied));
1634
1635         if (PageChecked(page)) {
1636                 if (copied == len)
1637                         SetPageUptodate(page);
1638                 ClearPageChecked(page);
1639         } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1640                 SetPageUptodate(page);
1641
1642         if (!PageUptodate(page)) {
1643                 char *page_data;
1644                 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1645                 int xid;
1646
1647                 xid = GetXid();
1648                 /* this is probably better than directly calling
1649                    partialpage_write since in this function the file handle is
1650                    known which we might as well leverage */
1651                 /* BB check if anything else missing out of ppw
1652                    such as updating last write time */
1653                 page_data = kmap(page);
1654                 rc = cifs_write(file, page_data + offset, copied, &pos);
1655                 /* if (rc < 0) should we set writebehind rc? */
1656                 kunmap(page);
1657
1658                 FreeXid(xid);
1659         } else {
1660                 rc = copied;
1661                 pos += copied;
1662                 set_page_dirty(page);
1663         }
1664
1665         if (rc > 0) {
1666                 spin_lock(&inode->i_lock);
1667                 if (pos > inode->i_size)
1668                         i_size_write(inode, pos);
1669                 spin_unlock(&inode->i_lock);
1670         }
1671
1672         unlock_page(page);
1673         page_cache_release(page);
1674
1675         return rc;
1676 }
1677
1678 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1679 {
1680         int xid;
1681         int rc = 0;
1682         struct cifsTconInfo *tcon;
1683         struct cifsFileInfo *smbfile =
1684                 (struct cifsFileInfo *)file->private_data;
1685         struct inode *inode = file->f_path.dentry->d_inode;
1686
1687         xid = GetXid();
1688
1689         cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1690                 dentry->d_name.name, datasync));
1691
1692         rc = filemap_write_and_wait(inode->i_mapping);
1693         if (rc == 0) {
1694                 rc = CIFS_I(inode)->write_behind_rc;
1695                 CIFS_I(inode)->write_behind_rc = 0;
1696                 tcon = CIFS_SB(inode->i_sb)->tcon;
1697                 if (!rc && tcon && smbfile &&
1698                    !(CIFS_SB(inode->i_sb)->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1699                         rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1700         }
1701
1702         FreeXid(xid);
1703         return rc;
1704 }
1705
1706 /* static void cifs_sync_page(struct page *page)
1707 {
1708         struct address_space *mapping;
1709         struct inode *inode;
1710         unsigned long index = page->index;
1711         unsigned int rpages = 0;
1712         int rc = 0;
1713
1714         cFYI(1, ("sync page %p",page));
1715         mapping = page->mapping;
1716         if (!mapping)
1717                 return 0;
1718         inode = mapping->host;
1719         if (!inode)
1720                 return; */
1721
1722 /*      fill in rpages then
1723         result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1724
1725 /*      cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1726
1727 #if 0
1728         if (rc < 0)
1729                 return rc;
1730         return 0;
1731 #endif
1732 } */
1733
1734 /*
1735  * As file closes, flush all cached write data for this inode checking
1736  * for write behind errors.
1737  */
1738 int cifs_flush(struct file *file, fl_owner_t id)
1739 {
1740         struct inode *inode = file->f_path.dentry->d_inode;
1741         int rc = 0;
1742
1743         /* Rather than do the steps manually:
1744            lock the inode for writing
1745            loop through pages looking for write behind data (dirty pages)
1746            coalesce into contiguous 16K (or smaller) chunks to write to server
1747            send to server (prefer in parallel)
1748            deal with writebehind errors
1749            unlock inode for writing
1750            filemapfdatawrite appears easier for the time being */
1751
1752         rc = filemap_fdatawrite(inode->i_mapping);
1753         /* reset wb rc if we were able to write out dirty pages */
1754         if (!rc) {
1755                 rc = CIFS_I(inode)->write_behind_rc;
1756                 CIFS_I(inode)->write_behind_rc = 0;
1757         }
1758
1759         cFYI(1, ("Flush inode %p file %p rc %d", inode, file, rc));
1760
1761         return rc;
1762 }
1763
1764 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1765         size_t read_size, loff_t *poffset)
1766 {
1767         int rc = -EACCES;
1768         unsigned int bytes_read = 0;
1769         unsigned int total_read = 0;
1770         unsigned int current_read_size;
1771         struct cifs_sb_info *cifs_sb;
1772         struct cifsTconInfo *pTcon;
1773         int xid;
1774         struct cifsFileInfo *open_file;
1775         char *smb_read_data;
1776         char __user *current_offset;
1777         struct smb_com_read_rsp *pSMBr;
1778
1779         xid = GetXid();
1780         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1781         pTcon = cifs_sb->tcon;
1782
1783         if (file->private_data == NULL) {
1784                 rc = -EBADF;
1785                 FreeXid(xid);
1786                 return rc;
1787         }
1788         open_file = (struct cifsFileInfo *)file->private_data;
1789
1790         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1791                 cFYI(1, ("attempting read on write only file instance"));
1792
1793         for (total_read = 0, current_offset = read_data;
1794              read_size > total_read;
1795              total_read += bytes_read, current_offset += bytes_read) {
1796                 current_read_size = min_t(const int, read_size - total_read,
1797                                           cifs_sb->rsize);
1798                 rc = -EAGAIN;
1799                 smb_read_data = NULL;
1800                 while (rc == -EAGAIN) {
1801                         int buf_type = CIFS_NO_BUFFER;
1802                         if ((open_file->invalidHandle) &&
1803                             (!open_file->closePend)) {
1804                                 rc = cifs_reopen_file(file, true);
1805                                 if (rc != 0)
1806                                         break;
1807                         }
1808                         rc = CIFSSMBRead(xid, pTcon,
1809                                          open_file->netfid,
1810                                          current_read_size, *poffset,
1811                                          &bytes_read, &smb_read_data,
1812                                          &buf_type);
1813                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1814                         if (smb_read_data) {
1815                                 if (copy_to_user(current_offset,
1816                                                 smb_read_data +
1817                                                 4 /* RFC1001 length field */ +
1818                                                 le16_to_cpu(pSMBr->DataOffset),
1819                                                 bytes_read))
1820                                         rc = -EFAULT;
1821
1822                                 if (buf_type == CIFS_SMALL_BUFFER)
1823                                         cifs_small_buf_release(smb_read_data);
1824                                 else if (buf_type == CIFS_LARGE_BUFFER)
1825                                         cifs_buf_release(smb_read_data);
1826                                 smb_read_data = NULL;
1827                         }
1828                 }
1829                 if (rc || (bytes_read == 0)) {
1830                         if (total_read) {
1831                                 break;
1832                         } else {
1833                                 FreeXid(xid);
1834                                 return rc;
1835                         }
1836                 } else {
1837                         cifs_stats_bytes_read(pTcon, bytes_read);
1838                         *poffset += bytes_read;
1839                 }
1840         }
1841         FreeXid(xid);
1842         return total_read;
1843 }
1844
1845
1846 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1847         loff_t *poffset)
1848 {
1849         int rc = -EACCES;
1850         unsigned int bytes_read = 0;
1851         unsigned int total_read;
1852         unsigned int current_read_size;
1853         struct cifs_sb_info *cifs_sb;
1854         struct cifsTconInfo *pTcon;
1855         int xid;
1856         char *current_offset;
1857         struct cifsFileInfo *open_file;
1858         int buf_type = CIFS_NO_BUFFER;
1859
1860         xid = GetXid();
1861         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1862         pTcon = cifs_sb->tcon;
1863
1864         if (file->private_data == NULL) {
1865                 rc = -EBADF;
1866                 FreeXid(xid);
1867                 return rc;
1868         }
1869         open_file = (struct cifsFileInfo *)file->private_data;
1870
1871         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1872                 cFYI(1, ("attempting read on write only file instance"));
1873
1874         for (total_read = 0, current_offset = read_data;
1875              read_size > total_read;
1876              total_read += bytes_read, current_offset += bytes_read) {
1877                 current_read_size = min_t(const int, read_size - total_read,
1878                                           cifs_sb->rsize);
1879                 /* For windows me and 9x we do not want to request more
1880                 than it negotiated since it will refuse the read then */
1881                 if ((pTcon->ses) &&
1882                         !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1883                         current_read_size = min_t(const int, current_read_size,
1884                                         pTcon->ses->server->maxBuf - 128);
1885                 }
1886                 rc = -EAGAIN;
1887                 while (rc == -EAGAIN) {
1888                         if ((open_file->invalidHandle) &&
1889                             (!open_file->closePend)) {
1890                                 rc = cifs_reopen_file(file, true);
1891                                 if (rc != 0)
1892                                         break;
1893                         }
1894                         rc = CIFSSMBRead(xid, pTcon,
1895                                          open_file->netfid,
1896                                          current_read_size, *poffset,
1897                                          &bytes_read, &current_offset,
1898                                          &buf_type);
1899                 }
1900                 if (rc || (bytes_read == 0)) {
1901                         if (total_read) {
1902                                 break;
1903                         } else {
1904                                 FreeXid(xid);
1905                                 return rc;
1906                         }
1907                 } else {
1908                         cifs_stats_bytes_read(pTcon, total_read);
1909                         *poffset += bytes_read;
1910                 }
1911         }
1912         FreeXid(xid);
1913         return total_read;
1914 }
1915
1916 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1917 {
1918         int rc, xid;
1919
1920         xid = GetXid();
1921         rc = cifs_revalidate_file(file);
1922         if (rc) {
1923                 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1924                 FreeXid(xid);
1925                 return rc;
1926         }
1927         rc = generic_file_mmap(file, vma);
1928         FreeXid(xid);
1929         return rc;
1930 }
1931
1932
1933 static void cifs_copy_cache_pages(struct address_space *mapping,
1934         struct list_head *pages, int bytes_read, char *data,
1935         struct pagevec *plru_pvec)
1936 {
1937         struct page *page;
1938         char *target;
1939
1940         while (bytes_read > 0) {
1941                 if (list_empty(pages))
1942                         break;
1943
1944                 page = list_entry(pages->prev, struct page, lru);
1945                 list_del(&page->lru);
1946
1947                 if (add_to_page_cache(page, mapping, page->index,
1948                                       GFP_KERNEL)) {
1949                         page_cache_release(page);
1950                         cFYI(1, ("Add page cache failed"));
1951                         data += PAGE_CACHE_SIZE;
1952                         bytes_read -= PAGE_CACHE_SIZE;
1953                         continue;
1954                 }
1955
1956                 target = kmap_atomic(page, KM_USER0);
1957
1958                 if (PAGE_CACHE_SIZE > bytes_read) {
1959                         memcpy(target, data, bytes_read);
1960                         /* zero the tail end of this partial page */
1961                         memset(target + bytes_read, 0,
1962                                PAGE_CACHE_SIZE - bytes_read);
1963                         bytes_read = 0;
1964                 } else {
1965                         memcpy(target, data, PAGE_CACHE_SIZE);
1966                         bytes_read -= PAGE_CACHE_SIZE;
1967                 }
1968                 kunmap_atomic(target, KM_USER0);
1969
1970                 flush_dcache_page(page);
1971                 SetPageUptodate(page);
1972                 unlock_page(page);
1973                 if (!pagevec_add(plru_pvec, page))
1974                         __pagevec_lru_add_file(plru_pvec);
1975                 data += PAGE_CACHE_SIZE;
1976         }
1977         return;
1978 }
1979
1980 static int cifs_readpages(struct file *file, struct address_space *mapping,
1981         struct list_head *page_list, unsigned num_pages)
1982 {
1983         int rc = -EACCES;
1984         int xid;
1985         loff_t offset;
1986         struct page *page;
1987         struct cifs_sb_info *cifs_sb;
1988         struct cifsTconInfo *pTcon;
1989         unsigned int bytes_read = 0;
1990         unsigned int read_size, i;
1991         char *smb_read_data = NULL;
1992         struct smb_com_read_rsp *pSMBr;
1993         struct pagevec lru_pvec;
1994         struct cifsFileInfo *open_file;
1995         int buf_type = CIFS_NO_BUFFER;
1996
1997         xid = GetXid();
1998         if (file->private_data == NULL) {
1999                 rc = -EBADF;
2000                 FreeXid(xid);
2001                 return rc;
2002         }
2003         open_file = (struct cifsFileInfo *)file->private_data;
2004         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2005         pTcon = cifs_sb->tcon;
2006
2007         pagevec_init(&lru_pvec, 0);
2008         cFYI(DBG2, ("rpages: num pages %d", num_pages));
2009         for (i = 0; i < num_pages; ) {
2010                 unsigned contig_pages;
2011                 struct page *tmp_page;
2012                 unsigned long expected_index;
2013
2014                 if (list_empty(page_list))
2015                         break;
2016
2017                 page = list_entry(page_list->prev, struct page, lru);
2018                 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2019
2020                 /* count adjacent pages that we will read into */
2021                 contig_pages = 0;
2022                 expected_index =
2023                         list_entry(page_list->prev, struct page, lru)->index;
2024                 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2025                         if (tmp_page->index == expected_index) {
2026                                 contig_pages++;
2027                                 expected_index++;
2028                         } else
2029                                 break;
2030                 }
2031                 if (contig_pages + i >  num_pages)
2032                         contig_pages = num_pages - i;
2033
2034                 /* for reads over a certain size could initiate async
2035                    read ahead */
2036
2037                 read_size = contig_pages * PAGE_CACHE_SIZE;
2038                 /* Read size needs to be in multiples of one page */
2039                 read_size = min_t(const unsigned int, read_size,
2040                                   cifs_sb->rsize & PAGE_CACHE_MASK);
2041                 cFYI(DBG2, ("rpages: read size 0x%x  contiguous pages %d",
2042                                 read_size, contig_pages));
2043                 rc = -EAGAIN;
2044                 while (rc == -EAGAIN) {
2045                         if ((open_file->invalidHandle) &&
2046                             (!open_file->closePend)) {
2047                                 rc = cifs_reopen_file(file, true);
2048                                 if (rc != 0)
2049                                         break;
2050                         }
2051
2052                         rc = CIFSSMBRead(xid, pTcon,
2053                                          open_file->netfid,
2054                                          read_size, offset,
2055                                          &bytes_read, &smb_read_data,
2056                                          &buf_type);
2057                         /* BB more RC checks ? */
2058                         if (rc == -EAGAIN) {
2059                                 if (smb_read_data) {
2060                                         if (buf_type == CIFS_SMALL_BUFFER)
2061                                                 cifs_small_buf_release(smb_read_data);
2062                                         else if (buf_type == CIFS_LARGE_BUFFER)
2063                                                 cifs_buf_release(smb_read_data);
2064                                         smb_read_data = NULL;
2065                                 }
2066                         }
2067                 }
2068                 if ((rc < 0) || (smb_read_data == NULL)) {
2069                         cFYI(1, ("Read error in readpages: %d", rc));
2070                         break;
2071                 } else if (bytes_read > 0) {
2072                         task_io_account_read(bytes_read);
2073                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2074                         cifs_copy_cache_pages(mapping, page_list, bytes_read,
2075                                 smb_read_data + 4 /* RFC1001 hdr */ +
2076                                 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
2077
2078                         i +=  bytes_read >> PAGE_CACHE_SHIFT;
2079                         cifs_stats_bytes_read(pTcon, bytes_read);
2080                         if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2081                                 i++; /* account for partial page */
2082
2083                                 /* server copy of file can have smaller size
2084                                    than client */
2085                                 /* BB do we need to verify this common case ?
2086                                    this case is ok - if we are at server EOF
2087                                    we will hit it on next read */
2088
2089                                 /* break; */
2090                         }
2091                 } else {
2092                         cFYI(1, ("No bytes read (%d) at offset %lld . "
2093                                  "Cleaning remaining pages from readahead list",
2094                                  bytes_read, offset));
2095                         /* BB turn off caching and do new lookup on
2096                            file size at server? */
2097                         break;
2098                 }
2099                 if (smb_read_data) {
2100                         if (buf_type == CIFS_SMALL_BUFFER)
2101                                 cifs_small_buf_release(smb_read_data);
2102                         else if (buf_type == CIFS_LARGE_BUFFER)
2103                                 cifs_buf_release(smb_read_data);
2104                         smb_read_data = NULL;
2105                 }
2106                 bytes_read = 0;
2107         }
2108
2109         pagevec_lru_add_file(&lru_pvec);
2110
2111 /* need to free smb_read_data buf before exit */
2112         if (smb_read_data) {
2113                 if (buf_type == CIFS_SMALL_BUFFER)
2114                         cifs_small_buf_release(smb_read_data);
2115                 else if (buf_type == CIFS_LARGE_BUFFER)
2116                         cifs_buf_release(smb_read_data);
2117                 smb_read_data = NULL;
2118         }
2119
2120         FreeXid(xid);
2121         return rc;
2122 }
2123
2124 static int cifs_readpage_worker(struct file *file, struct page *page,
2125         loff_t *poffset)
2126 {
2127         char *read_data;
2128         int rc;
2129
2130         page_cache_get(page);
2131         read_data = kmap(page);
2132         /* for reads over a certain size could initiate async read ahead */
2133
2134         rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2135
2136         if (rc < 0)
2137                 goto io_error;
2138         else
2139                 cFYI(1, ("Bytes read %d", rc));
2140
2141         file->f_path.dentry->d_inode->i_atime =
2142                 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2143
2144         if (PAGE_CACHE_SIZE > rc)
2145                 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2146
2147         flush_dcache_page(page);
2148         SetPageUptodate(page);
2149         rc = 0;
2150
2151 io_error:
2152         kunmap(page);
2153         page_cache_release(page);
2154         return rc;
2155 }
2156
2157 static int cifs_readpage(struct file *file, struct page *page)
2158 {
2159         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2160         int rc = -EACCES;
2161         int xid;
2162
2163         xid = GetXid();
2164
2165         if (file->private_data == NULL) {
2166                 rc = -EBADF;
2167                 FreeXid(xid);
2168                 return rc;
2169         }
2170
2171         cFYI(1, ("readpage %p at offset %d 0x%x\n",
2172                  page, (int)offset, (int)offset));
2173
2174         rc = cifs_readpage_worker(file, page, &offset);
2175
2176         unlock_page(page);
2177
2178         FreeXid(xid);
2179         return rc;
2180 }
2181
2182 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2183 {
2184         struct cifsFileInfo *open_file;
2185
2186         read_lock(&GlobalSMBSeslock);
2187         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2188                 if (open_file->closePend)
2189                         continue;
2190                 if (open_file->pfile &&
2191                     ((open_file->pfile->f_flags & O_RDWR) ||
2192                      (open_file->pfile->f_flags & O_WRONLY))) {
2193                         read_unlock(&GlobalSMBSeslock);
2194                         return 1;
2195                 }
2196         }
2197         read_unlock(&GlobalSMBSeslock);
2198         return 0;
2199 }
2200
2201 /* We do not want to update the file size from server for inodes
2202    open for write - to avoid races with writepage extending
2203    the file - in the future we could consider allowing
2204    refreshing the inode only on increases in the file size
2205    but this is tricky to do without racing with writebehind
2206    page caching in the current Linux kernel design */
2207 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2208 {
2209         if (!cifsInode)
2210                 return true;
2211
2212         if (is_inode_writable(cifsInode)) {
2213                 /* This inode is open for write at least once */
2214                 struct cifs_sb_info *cifs_sb;
2215
2216                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2217                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2218                         /* since no page cache to corrupt on directio
2219                         we can change size safely */
2220                         return true;
2221                 }
2222
2223                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2224                         return true;
2225
2226                 return false;
2227         } else
2228                 return true;
2229 }
2230
2231 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2232                         loff_t pos, unsigned len, unsigned flags,
2233                         struct page **pagep, void **fsdata)
2234 {
2235         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2236         loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2237         loff_t page_start = pos & PAGE_MASK;
2238         loff_t i_size;
2239         struct page *page;
2240         int rc = 0;
2241
2242         cFYI(1, ("write_begin from %lld len %d", (long long)pos, len));
2243
2244         page = grab_cache_page_write_begin(mapping, index, flags);
2245         if (!page) {
2246                 rc = -ENOMEM;
2247                 goto out;
2248         }
2249
2250         if (PageUptodate(page))
2251                 goto out;
2252
2253         /*
2254          * If we write a full page it will be up to date, no need to read from
2255          * the server. If the write is short, we'll end up doing a sync write
2256          * instead.
2257          */
2258         if (len == PAGE_CACHE_SIZE)
2259                 goto out;
2260
2261         /*
2262          * optimize away the read when we have an oplock, and we're not
2263          * expecting to use any of the data we'd be reading in. That
2264          * is, when the page lies beyond the EOF, or straddles the EOF
2265          * and the write will cover all of the existing data.
2266          */
2267         if (CIFS_I(mapping->host)->clientCanCacheRead) {
2268                 i_size = i_size_read(mapping->host);
2269                 if (page_start >= i_size ||
2270                     (offset == 0 && (pos + len) >= i_size)) {
2271                         zero_user_segments(page, 0, offset,
2272                                            offset + len,
2273                                            PAGE_CACHE_SIZE);
2274                         /*
2275                          * PageChecked means that the parts of the page
2276                          * to which we're not writing are considered up
2277                          * to date. Once the data is copied to the
2278                          * page, it can be set uptodate.
2279                          */
2280                         SetPageChecked(page);
2281                         goto out;
2282                 }
2283         }
2284
2285         if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2286                 /*
2287                  * might as well read a page, it is fast enough. If we get
2288                  * an error, we don't need to return it. cifs_write_end will
2289                  * do a sync write instead since PG_uptodate isn't set.
2290                  */
2291                 cifs_readpage_worker(file, page, &page_start);
2292         } else {
2293                 /* we could try using another file handle if there is one -
2294                    but how would we lock it to prevent close of that handle
2295                    racing with this read? In any case
2296                    this will be written out by write_end so is fine */
2297         }
2298 out:
2299         *pagep = page;
2300         return rc;
2301 }
2302
2303 static void
2304 cifs_oplock_break(struct slow_work *work)
2305 {
2306         struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2307                                                   oplock_break);
2308         struct inode *inode = cfile->pInode;
2309         struct cifsInodeInfo *cinode = CIFS_I(inode);
2310         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->mnt->mnt_sb);
2311         int rc, waitrc = 0;
2312
2313         if (inode && S_ISREG(inode->i_mode)) {
2314 #ifdef CONFIG_CIFS_EXPERIMENTAL
2315                 if (cinode->clientCanCacheAll == 0)
2316                         break_lease(inode, O_RDONLY);
2317                 else if (cinode->clientCanCacheRead == 0)
2318                         break_lease(inode, O_WRONLY);
2319 #endif
2320                 rc = filemap_fdatawrite(inode->i_mapping);
2321                 if (cinode->clientCanCacheRead == 0) {
2322                         waitrc = filemap_fdatawait(inode->i_mapping);
2323                         invalidate_remote_inode(inode);
2324                 }
2325                 if (!rc)
2326                         rc = waitrc;
2327                 if (rc)
2328                         cinode->write_behind_rc = rc;
2329                 cFYI(1, ("Oplock flush inode %p rc %d", inode, rc));
2330         }
2331
2332         /*
2333          * releasing stale oplock after recent reconnect of smb session using
2334          * a now incorrect file handle is not a data integrity issue but do
2335          * not bother sending an oplock release if session to server still is
2336          * disconnected since oplock already released by the server
2337          */
2338         if (!cfile->closePend && !cfile->oplock_break_cancelled) {
2339                 rc = CIFSSMBLock(0, cifs_sb->tcon, cfile->netfid, 0, 0, 0, 0,
2340                                  LOCKING_ANDX_OPLOCK_RELEASE, false);
2341                 cFYI(1, ("Oplock release rc = %d", rc));
2342         }
2343 }
2344
2345 static int
2346 cifs_oplock_break_get(struct slow_work *work)
2347 {
2348         struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2349                                                   oplock_break);
2350         mntget(cfile->mnt);
2351         cifsFileInfo_get(cfile);
2352         return 0;
2353 }
2354
2355 static void
2356 cifs_oplock_break_put(struct slow_work *work)
2357 {
2358         struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2359                                                   oplock_break);
2360         mntput(cfile->mnt);
2361         cifsFileInfo_put(cfile);
2362 }
2363
2364 const struct slow_work_ops cifs_oplock_break_ops = {
2365         .get_ref        = cifs_oplock_break_get,
2366         .put_ref        = cifs_oplock_break_put,
2367         .execute        = cifs_oplock_break,
2368 };
2369
2370 const struct address_space_operations cifs_addr_ops = {
2371         .readpage = cifs_readpage,
2372         .readpages = cifs_readpages,
2373         .writepage = cifs_writepage,
2374         .writepages = cifs_writepages,
2375         .write_begin = cifs_write_begin,
2376         .write_end = cifs_write_end,
2377         .set_page_dirty = __set_page_dirty_nobuffers,
2378         /* .sync_page = cifs_sync_page, */
2379         /* .direct_IO = */
2380 };
2381
2382 /*
2383  * cifs_readpages requires the server to support a buffer large enough to
2384  * contain the header plus one complete page of data.  Otherwise, we need
2385  * to leave cifs_readpages out of the address space operations.
2386  */
2387 const struct address_space_operations cifs_addr_ops_smallbuf = {
2388         .readpage = cifs_readpage,
2389         .writepage = cifs_writepage,
2390         .writepages = cifs_writepages,
2391         .write_begin = cifs_write_begin,
2392         .write_end = cifs_write_end,
2393         .set_page_dirty = __set_page_dirty_nobuffers,
2394         /* .sync_page = cifs_sync_page, */
2395         /* .direct_IO = */
2396 };