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., 2002,2010
   7  *   Author(s): Steve French ([email protected])
   8  *              Jeremy Allison ([email protected])
   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 #include "fscache.h"
  44
  45 static inline int cifs_convert_flags(unsigned int flags)
  46 {
  47         if ((flags & O_ACCMODE) == O_RDONLY)
  48                 return GENERIC_READ;
  49         else if ((flags & O_ACCMODE) == O_WRONLY)
  50                 return GENERIC_WRITE;
  51         else if ((flags & O_ACCMODE) == O_RDWR) {
  52                 /* GENERIC_ALL is too much permission to request
  53                    can cause unnecessary access denied on create */
  54                 /* return GENERIC_ALL; */
  55                 return (GENERIC_READ | GENERIC_WRITE);
  56         }
  57
  58         return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
  59                 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
  60                 FILE_READ_DATA);
  61 }
  62
  63 static inline fmode_t cifs_posix_convert_flags(unsigned int flags)
  64 {
  65         fmode_t posix_flags = 0;
  66
  67         if ((flags & O_ACCMODE) == O_RDONLY)
  68                 posix_flags = FMODE_READ;
  69         else if ((flags & O_ACCMODE) == O_WRONLY)
  70                 posix_flags = FMODE_WRITE;
  71         else if ((flags & O_ACCMODE) == O_RDWR) {
  72                 /* GENERIC_ALL is too much permission to request
  73                    can cause unnecessary access denied on create */
  74                 /* return GENERIC_ALL; */
  75                 posix_flags = FMODE_READ | FMODE_WRITE;
  76         }
  77         /* can not map O_CREAT or O_EXCL or O_TRUNC flags when
  78            reopening a file.  They had their effect on the original open */
  79         if (flags & O_APPEND)
  80                 posix_flags |= (fmode_t)O_APPEND;
  81         if (flags & O_DSYNC)
  82                 posix_flags |= (fmode_t)O_DSYNC;
  83         if (flags & __O_SYNC)
  84                 posix_flags |= (fmode_t)__O_SYNC;
  85         if (flags & O_DIRECTORY)
  86                 posix_flags |= (fmode_t)O_DIRECTORY;
  87         if (flags & O_NOFOLLOW)
  88                 posix_flags |= (fmode_t)O_NOFOLLOW;
  89         if (flags & O_DIRECT)
  90                 posix_flags |= (fmode_t)O_DIRECT;
  91
  92         return posix_flags;
  93 }
  94
  95 static inline int cifs_get_disposition(unsigned int flags)
  96 {
  97         if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
  98                 return FILE_CREATE;
  99         else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
 100                 return FILE_OVERWRITE_IF;
 101         else if ((flags & O_CREAT) == O_CREAT)
 102                 return FILE_OPEN_IF;
 103         else if ((flags & O_TRUNC) == O_TRUNC)
 104                 return FILE_OVERWRITE;
 105         else
 106                 return FILE_OPEN;
 107 }
 108
 109 /* all arguments to this function must be checked for validity in caller */
 110 static inline int
 111 cifs_posix_open_inode_helper(struct inode *inode, struct file *file,
 112                              struct cifsInodeInfo *pCifsInode, __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 /* all arguments to this function must be checked for validity in caller */
 167 static inline int cifs_open_inode_helper(struct inode *inode,
 168         struct cifsTconInfo *pTcon, __u32 oplock, FILE_ALL_INFO *buf,
 169         char *full_path, int xid)
 170 {
 171         struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
 172         struct timespec temp;
 173         int rc;
 174
 175         if (pCifsInode->clientCanCacheRead) {
 176                 /* we have the inode open somewhere else
 177                    no need to discard cache data */
 178                 goto client_can_cache;
 179         }
 180
 181         /* BB need same check in cifs_create too? */
 182         /* if not oplocked, invalidate inode pages if mtime or file
 183            size changed */
 184         temp = cifs_NTtimeToUnix(buf->LastWriteTime);
 185         if (timespec_equal(&inode->i_mtime, &temp) &&
 186                            (inode->i_size ==
 187                             (loff_t)le64_to_cpu(buf->EndOfFile))) {
 188                 cFYI(1, "inode unchanged on server");
 189         } else {
 190                 if (inode->i_mapping) {
 191                         /* BB no need to lock inode until after invalidate
 192                         since namei code should already have it locked? */
 193                         rc = filemap_write_and_wait(inode->i_mapping);
 194                         if (rc != 0)
 195                                 pCifsInode->write_behind_rc = rc;
 196                 }
 197                 cFYI(1, "invalidating remote inode since open detected it "
 198                          "changed");
 199                 invalidate_remote_inode(inode);
 200         }
 201
 202 client_can_cache:
 203         if (pTcon->unix_ext)
 204                 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
 205                                               xid);
 206         else
 207                 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
 208                                          xid, NULL);
 209
 210         if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
 211                 pCifsInode->clientCanCacheAll = true;
 212                 pCifsInode->clientCanCacheRead = true;
 213                 cFYI(1, "Exclusive Oplock granted on inode %p", inode);
 214         } else if ((oplock & 0xF) == OPLOCK_READ)
 215                 pCifsInode->clientCanCacheRead = true;
 216
 217         return rc;
 218 }
 219
 220 int cifs_open(struct inode *inode, struct file *file)
 221 {
 222         int rc = -EACCES;
 223         int xid;
 224         __u32 oplock;
 225         struct cifs_sb_info *cifs_sb;
 226         struct cifsTconInfo *tcon;
 227         struct cifsFileInfo *pCifsFile = NULL;
 228         struct cifsInodeInfo *pCifsInode;
 229         char *full_path = NULL;
 230         int desiredAccess;
 231         int disposition;
 232         __u16 netfid;
 233         FILE_ALL_INFO *buf = NULL;
 234
 235         xid = GetXid();
 236
 237         cifs_sb = CIFS_SB(inode->i_sb);
 238         tcon = cifs_sb->tcon;
 239
 240         pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
 241
 242         full_path = build_path_from_dentry(file->f_path.dentry);
 243         if (full_path == NULL) {
 244                 rc = -ENOMEM;
 245                 goto out;
 246         }
 247
 248         cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
 249                  inode, file->f_flags, full_path);
 250
 251         if (oplockEnabled)
 252                 oplock = REQ_OPLOCK;
 253         else
 254                 oplock = 0;
 255
 256         if (!tcon->broken_posix_open && tcon->unix_ext &&
 257             (tcon->ses->capabilities & CAP_UNIX) &&
 258             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 259                         le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 260                 int oflags = (int) cifs_posix_convert_flags(file->f_flags);
 261                 oflags |= SMB_O_CREAT;
 262                 /* can not refresh inode info since size could be stale */
 263                 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
 264                                 cifs_sb->mnt_file_mode /* ignored */,
 265                                 oflags, &oplock, &netfid, xid);
 266                 if (rc == 0) {
 267                         cFYI(1, "posix open succeeded");
 268                         /* no need for special case handling of setting mode
 269                            on read only files needed here */
 270
 271                         rc = cifs_posix_open_inode_helper(inode, file,
 272                                         pCifsInode, oplock, netfid);
 273                         if (rc != 0) {
 274                                 CIFSSMBClose(xid, tcon, netfid);
 275                                 goto out;
 276                         }
 277
 278                         pCifsFile = cifs_new_fileinfo(inode, netfid, file,
 279                                                         file->f_path.mnt,
 280                                                         oflags, oplock);
 281                         if (pCifsFile == NULL) {
 282                                 CIFSSMBClose(xid, tcon, netfid);
 283                                 rc = -ENOMEM;
 284                         }
 285
 286                         cifs_fscache_set_inode_cookie(inode, file);
 287
 288                         goto out;
 289                 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
 290                         if (tcon->ses->serverNOS)
 291                                 cERROR(1, "server %s of type %s returned"
 292                                            " unexpected error on SMB posix open"
 293                                            ", disabling posix open support."
 294                                            " Check if server update available.",
 295                                            tcon->ses->serverName,
 296                                            tcon->ses->serverNOS);
 297                         tcon->broken_posix_open = true;
 298                 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
 299                          (rc != -EOPNOTSUPP)) /* path not found or net err */
 300                         goto out;
 301                 /* else fallthrough to retry open the old way on network i/o
 302                    or DFS errors */
 303         }
 304
 305         desiredAccess = cifs_convert_flags(file->f_flags);
 306
 307 /*********************************************************************
 308  *  open flag mapping table:
 309  *
 310  *      POSIX Flag            CIFS Disposition
 311  *      ----------            ----------------
 312  *      O_CREAT               FILE_OPEN_IF
 313  *      O_CREAT | O_EXCL      FILE_CREATE
 314  *      O_CREAT | O_TRUNC     FILE_OVERWRITE_IF
 315  *      O_TRUNC               FILE_OVERWRITE
 316  *      none of the above     FILE_OPEN
 317  *
 318  *      Note that there is not a direct match between disposition
 319  *      FILE_SUPERSEDE (ie create whether or not file exists although
 320  *      O_CREAT | O_TRUNC is similar but truncates the existing
 321  *      file rather than creating a new file as FILE_SUPERSEDE does
 322  *      (which uses the attributes / metadata passed in on open call)
 323  *?
 324  *?  O_SYNC is a reasonable match to CIFS writethrough flag
 325  *?  and the read write flags match reasonably.  O_LARGEFILE
 326  *?  is irrelevant because largefile support is always used
 327  *?  by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
 328  *       O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
 329  *********************************************************************/
 330
 331         disposition = cifs_get_disposition(file->f_flags);
 332
 333         /* BB pass O_SYNC flag through on file attributes .. BB */
 334
 335         /* Also refresh inode by passing in file_info buf returned by SMBOpen
 336            and calling get_inode_info with returned buf (at least helps
 337            non-Unix server case) */
 338
 339         /* BB we can not do this if this is the second open of a file
 340            and the first handle has writebehind data, we might be
 341            able to simply do a filemap_fdatawrite/filemap_fdatawait first */
 342         buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
 343         if (!buf) {
 344                 rc = -ENOMEM;
 345                 goto out;
 346         }
 347
 348         if (cifs_sb->tcon->ses->capabilities & CAP_NT_SMBS)
 349                 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
 350                          desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
 351                          cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
 352                                  & CIFS_MOUNT_MAP_SPECIAL_CHR);
 353         else
 354                 rc = -EIO; /* no NT SMB support fall into legacy open below */
 355
 356         if (rc == -EIO) {
 357                 /* Old server, try legacy style OpenX */
 358                 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
 359                         desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
 360                         cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
 361                                 & CIFS_MOUNT_MAP_SPECIAL_CHR);
 362         }
 363         if (rc) {
 364                 cFYI(1, "cifs_open returned 0x%x", rc);
 365                 goto out;
 366         }
 367
 368         rc = cifs_open_inode_helper(inode, tcon, oplock, buf, full_path, xid);
 369         if (rc != 0)
 370                 goto out;
 371
 372         pCifsFile = cifs_new_fileinfo(inode, netfid, file, file->f_path.mnt,
 373                                         file->f_flags, oplock);
 374         if (pCifsFile == NULL) {
 375                 rc = -ENOMEM;
 376                 goto out;
 377         }
 378
 379         cifs_fscache_set_inode_cookie(inode, file);
 380
 381         if (oplock & CIFS_CREATE_ACTION) {
 382                 /* time to set mode which we can not set earlier due to
 383                    problems creating new read-only files */
 384                 if (tcon->unix_ext) {
 385                         struct cifs_unix_set_info_args args = {
 386                                 .mode   = inode->i_mode,
 387                                 .uid    = NO_CHANGE_64,
 388                                 .gid    = NO_CHANGE_64,
 389                                 .ctime  = NO_CHANGE_64,
 390                                 .atime  = NO_CHANGE_64,
 391                                 .mtime  = NO_CHANGE_64,
 392                                 .device = 0,
 393                         };
 394                         CIFSSMBUnixSetPathInfo(xid, tcon, full_path, &args,
 395                                                cifs_sb->local_nls,
 396                                                cifs_sb->mnt_cifs_flags &
 397                                                 CIFS_MOUNT_MAP_SPECIAL_CHR);
 398                 }
 399         }
 400
 401 out:
 402         kfree(buf);
 403         kfree(full_path);
 404         FreeXid(xid);
 405         return rc;
 406 }
 407
 408 /* Try to reacquire byte range locks that were released when session */
 409 /* to server was lost */
 410 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
 411 {
 412         int rc = 0;
 413
 414 /* BB list all locks open on this file and relock */
 415
 416         return rc;
 417 }
 418
 419 static int cifs_reopen_file(struct file *file, bool can_flush)
 420 {
 421         int rc = -EACCES;
 422         int xid;
 423         __u32 oplock;
 424         struct cifs_sb_info *cifs_sb;
 425         struct cifsTconInfo *tcon;
 426         struct cifsFileInfo *pCifsFile;
 427         struct cifsInodeInfo *pCifsInode;
 428         struct inode *inode;
 429         char *full_path = NULL;
 430         int desiredAccess;
 431         int disposition = FILE_OPEN;
 432         __u16 netfid;
 433
 434         if (file->private_data)
 435                 pCifsFile = file->private_data;
 436         else
 437                 return -EBADF;
 438
 439         xid = GetXid();
 440         mutex_lock(&pCifsFile->fh_mutex);
 441         if (!pCifsFile->invalidHandle) {
 442                 mutex_unlock(&pCifsFile->fh_mutex);
 443                 rc = 0;
 444                 FreeXid(xid);
 445                 return rc;
 446         }
 447
 448         if (file->f_path.dentry == NULL) {
 449                 cERROR(1, "no valid name if dentry freed");
 450                 dump_stack();
 451                 rc = -EBADF;
 452                 goto reopen_error_exit;
 453         }
 454
 455         inode = file->f_path.dentry->d_inode;
 456         if (inode == NULL) {
 457                 cERROR(1, "inode not valid");
 458                 dump_stack();
 459                 rc = -EBADF;
 460                 goto reopen_error_exit;
 461         }
 462
 463         cifs_sb = CIFS_SB(inode->i_sb);
 464         tcon = pCifsFile->tcon;
 465
 466 /* can not grab rename sem here because various ops, including
 467    those that already have the rename sem can end up causing writepage
 468    to get called and if the server was down that means we end up here,
 469    and we can never tell if the caller already has the rename_sem */
 470         full_path = build_path_from_dentry(file->f_path.dentry);
 471         if (full_path == NULL) {
 472                 rc = -ENOMEM;
 473 reopen_error_exit:
 474                 mutex_unlock(&pCifsFile->fh_mutex);
 475                 FreeXid(xid);
 476                 return rc;
 477         }
 478
 479         cFYI(1, "inode = 0x%p file flags 0x%x for %s",
 480                  inode, file->f_flags, full_path);
 481
 482         if (oplockEnabled)
 483                 oplock = REQ_OPLOCK;
 484         else
 485                 oplock = 0;
 486
 487         if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
 488             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 489                         le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 490                 int oflags = (int) cifs_posix_convert_flags(file->f_flags);
 491                 /* can not refresh inode info since size could be stale */
 492                 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
 493                                 cifs_sb->mnt_file_mode /* ignored */,
 494                                 oflags, &oplock, &netfid, xid);
 495                 if (rc == 0) {
 496                         cFYI(1, "posix reopen succeeded");
 497                         goto reopen_success;
 498                 }
 499                 /* fallthrough to retry open the old way on errors, especially
 500                    in the reconnect path it is important to retry hard */
 501         }
 502
 503         desiredAccess = cifs_convert_flags(file->f_flags);
 504
 505         /* Can not refresh inode by passing in file_info buf to be returned
 506            by SMBOpen and then calling get_inode_info with returned buf
 507            since file might have write behind data that needs to be flushed
 508            and server version of file size can be stale. If we knew for sure
 509            that inode was not dirty locally we could do this */
 510
 511         rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
 512                          CREATE_NOT_DIR, &netfid, &oplock, NULL,
 513                          cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
 514                                 CIFS_MOUNT_MAP_SPECIAL_CHR);
 515         if (rc) {
 516                 mutex_unlock(&pCifsFile->fh_mutex);
 517                 cFYI(1, "cifs_open returned 0x%x", rc);
 518                 cFYI(1, "oplock: %d", oplock);
 519         } else {
 520 reopen_success:
 521                 pCifsFile->netfid = netfid;
 522                 pCifsFile->invalidHandle = false;
 523                 mutex_unlock(&pCifsFile->fh_mutex);
 524                 pCifsInode = CIFS_I(inode);
 525                 if (pCifsInode) {
 526                         if (can_flush) {
 527                                 rc = filemap_write_and_wait(inode->i_mapping);
 528                                 if (rc != 0)
 529                                         CIFS_I(inode)->write_behind_rc = rc;
 530                         /* temporarily disable caching while we
 531                            go to server to get inode info */
 532                                 pCifsInode->clientCanCacheAll = false;
 533                                 pCifsInode->clientCanCacheRead = false;
 534                                 if (tcon->unix_ext)
 535                                         rc = cifs_get_inode_info_unix(&inode,
 536                                                 full_path, inode->i_sb, xid);
 537                                 else
 538                                         rc = cifs_get_inode_info(&inode,
 539                                                 full_path, NULL, inode->i_sb,
 540                                                 xid, NULL);
 541                         } /* else we are writing out data to server already
 542                              and could deadlock if we tried to flush data, and
 543                              since we do not know if we have data that would
 544                              invalidate the current end of file on the server
 545                              we can not go to the server to get the new inod
 546                              info */
 547                         if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
 548                                 pCifsInode->clientCanCacheAll = true;
 549                                 pCifsInode->clientCanCacheRead = true;
 550                                 cFYI(1, "Exclusive Oplock granted on inode %p",
 551                                          file->f_path.dentry->d_inode);
 552                         } else if ((oplock & 0xF) == OPLOCK_READ) {
 553                                 pCifsInode->clientCanCacheRead = true;
 554                                 pCifsInode->clientCanCacheAll = false;
 555                         } else {
 556                                 pCifsInode->clientCanCacheRead = false;
 557                                 pCifsInode->clientCanCacheAll = false;
 558                         }
 559                         cifs_relock_file(pCifsFile);
 560                 }
 561         }
 562         kfree(full_path);
 563         FreeXid(xid);
 564         return rc;
 565 }
 566
 567 int cifs_close(struct inode *inode, struct file *file)
 568 {
 569         int rc = 0;
 570         int xid, timeout;
 571         struct cifs_sb_info *cifs_sb;
 572         struct cifsTconInfo *pTcon;
 573         struct cifsFileInfo *pSMBFile = file->private_data;
 574
 575         xid = GetXid();
 576
 577         cifs_sb = CIFS_SB(inode->i_sb);
 578         pTcon = pSMBFile->tcon;
 579         if (pSMBFile) {
 580                 struct cifsLockInfo *li, *tmp;
 581                 write_lock(&GlobalSMBSeslock);
 582                 pSMBFile->closePend = true;
 583                 if (pTcon) {
 584                         /* no sense reconnecting to close a file that is
 585                            already closed */
 586                         if (!pTcon->need_reconnect) {
 587                                 write_unlock(&GlobalSMBSeslock);
 588                                 timeout = 2;
 589                                 while ((atomic_read(&pSMBFile->count) != 1)
 590                                         && (timeout <= 2048)) {
 591                                         /* Give write a better chance to get to
 592                                         server ahead of the close.  We do not
 593                                         want to add a wait_q here as it would
 594                                         increase the memory utilization as
 595                                         the struct would be in each open file,
 596                                         but this should give enough time to
 597                                         clear the socket */
 598                                         cFYI(DBG2, "close delay, write pending");
 599                                         msleep(timeout);
 600                                         timeout *= 4;
 601                                 }
 602                                 if (!pTcon->need_reconnect &&
 603                                     !pSMBFile->invalidHandle)
 604                                         rc = CIFSSMBClose(xid, pTcon,
 605                                                   pSMBFile->netfid);
 606                         } else
 607                                 write_unlock(&GlobalSMBSeslock);
 608                 } else
 609                         write_unlock(&GlobalSMBSeslock);
 610
 611                 /* Delete any outstanding lock records.
 612                    We'll lose them when the file is closed anyway. */
 613                 mutex_lock(&pSMBFile->lock_mutex);
 614                 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
 615                         list_del(&li->llist);
 616                         kfree(li);
 617                 }
 618                 mutex_unlock(&pSMBFile->lock_mutex);
 619
 620                 write_lock(&GlobalSMBSeslock);
 621                 list_del(&pSMBFile->flist);
 622                 list_del(&pSMBFile->tlist);
 623                 write_unlock(&GlobalSMBSeslock);
 624                 cifsFileInfo_put(file->private_data);
 625                 file->private_data = NULL;
 626         } else
 627                 rc = -EBADF;
 628
 629         read_lock(&GlobalSMBSeslock);
 630         if (list_empty(&(CIFS_I(inode)->openFileList))) {
 631                 cFYI(1, "closing last open instance for inode %p", inode);
 632                 /* if the file is not open we do not know if we can cache info
 633                    on this inode, much less write behind and read ahead */
 634                 CIFS_I(inode)->clientCanCacheRead = false;
 635                 CIFS_I(inode)->clientCanCacheAll  = false;
 636         }
 637         read_unlock(&GlobalSMBSeslock);
 638         if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
 639                 rc = CIFS_I(inode)->write_behind_rc;
 640         FreeXid(xid);
 641         return rc;
 642 }
 643
 644 int cifs_closedir(struct inode *inode, struct file *file)
 645 {
 646         int rc = 0;
 647         int xid;
 648         struct cifsFileInfo *pCFileStruct = file->private_data;
 649         char *ptmp;
 650
 651         cFYI(1, "Closedir inode = 0x%p", inode);
 652
 653         xid = GetXid();
 654
 655         if (pCFileStruct) {
 656                 struct cifsTconInfo *pTcon = pCFileStruct->tcon;
 657
 658                 cFYI(1, "Freeing private data in close dir");
 659                 write_lock(&GlobalSMBSeslock);
 660                 if (!pCFileStruct->srch_inf.endOfSearch &&
 661                     !pCFileStruct->invalidHandle) {
 662                         pCFileStruct->invalidHandle = true;
 663                         write_unlock(&GlobalSMBSeslock);
 664                         rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
 665                         cFYI(1, "Closing uncompleted readdir with rc %d",
 666                                  rc);
 667                         /* not much we can do if it fails anyway, ignore rc */
 668                         rc = 0;
 669                 } else
 670                         write_unlock(&GlobalSMBSeslock);
 671                 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
 672                 if (ptmp) {
 673                         cFYI(1, "closedir free smb buf in srch struct");
 674                         pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
 675                         if (pCFileStruct->srch_inf.smallBuf)
 676                                 cifs_small_buf_release(ptmp);
 677                         else
 678                                 cifs_buf_release(ptmp);
 679                 }
 680                 kfree(file->private_data);
 681                 file->private_data = NULL;
 682         }
 683         /* BB can we lock the filestruct while this is going on? */
 684         FreeXid(xid);
 685         return rc;
 686 }
 687
 688 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
 689                                 __u64 offset, __u8 lockType)
 690 {
 691         struct cifsLockInfo *li =
 692                 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
 693         if (li == NULL)
 694                 return -ENOMEM;
 695         li->offset = offset;
 696         li->length = len;
 697         li->type = lockType;
 698         mutex_lock(&fid->lock_mutex);
 699         list_add(&li->llist, &fid->llist);
 700         mutex_unlock(&fid->lock_mutex);
 701         return 0;
 702 }
 703
 704 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
 705 {
 706         int rc, xid;
 707         __u32 numLock = 0;
 708         __u32 numUnlock = 0;
 709         __u64 length;
 710         bool wait_flag = false;
 711         struct cifs_sb_info *cifs_sb;
 712         struct cifsTconInfo *tcon;
 713         __u16 netfid;
 714         __u8 lockType = LOCKING_ANDX_LARGE_FILES;
 715         bool posix_locking = 0;
 716
 717         length = 1 + pfLock->fl_end - pfLock->fl_start;
 718         rc = -EACCES;
 719         xid = GetXid();
 720
 721         cFYI(1, "Lock parm: 0x%x flockflags: "
 722                  "0x%x flocktype: 0x%x start: %lld end: %lld",
 723                 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
 724                 pfLock->fl_end);
 725
 726         if (pfLock->fl_flags & FL_POSIX)
 727                 cFYI(1, "Posix");
 728         if (pfLock->fl_flags & FL_FLOCK)
 729                 cFYI(1, "Flock");
 730         if (pfLock->fl_flags & FL_SLEEP) {
 731                 cFYI(1, "Blocking lock");
 732                 wait_flag = true;
 733         }
 734         if (pfLock->fl_flags & FL_ACCESS)
 735                 cFYI(1, "Process suspended by mandatory locking - "
 736                          "not implemented yet");
 737         if (pfLock->fl_flags & FL_LEASE)
 738                 cFYI(1, "Lease on file - not implemented yet");
 739         if (pfLock->fl_flags &
 740             (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
 741                 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
 742
 743         if (pfLock->fl_type == F_WRLCK) {
 744                 cFYI(1, "F_WRLCK ");
 745                 numLock = 1;
 746         } else if (pfLock->fl_type == F_UNLCK) {
 747                 cFYI(1, "F_UNLCK");
 748                 numUnlock = 1;
 749                 /* Check if unlock includes more than
 750                 one lock range */
 751         } else if (pfLock->fl_type == F_RDLCK) {
 752                 cFYI(1, "F_RDLCK");
 753                 lockType |= LOCKING_ANDX_SHARED_LOCK;
 754                 numLock = 1;
 755         } else if (pfLock->fl_type == F_EXLCK) {
 756                 cFYI(1, "F_EXLCK");
 757                 numLock = 1;
 758         } else if (pfLock->fl_type == F_SHLCK) {
 759                 cFYI(1, "F_SHLCK");
 760                 lockType |= LOCKING_ANDX_SHARED_LOCK;
 761                 numLock = 1;
 762         } else
 763                 cFYI(1, "Unknown type of lock");
 764
 765         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
 766         tcon = ((struct cifsFileInfo *)file->private_data)->tcon;
 767
 768         if (file->private_data == NULL) {
 769                 rc = -EBADF;
 770                 FreeXid(xid);
 771                 return rc;
 772         }
 773         netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
 774
 775         if ((tcon->ses->capabilities & CAP_UNIX) &&
 776             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
 777             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
 778                 posix_locking = 1;
 779         /* BB add code here to normalize offset and length to
 780         account for negative length which we can not accept over the
 781         wire */
 782         if (IS_GETLK(cmd)) {
 783                 if (posix_locking) {
 784                         int posix_lock_type;
 785                         if (lockType & LOCKING_ANDX_SHARED_LOCK)
 786                                 posix_lock_type = CIFS_RDLCK;
 787                         else
 788                                 posix_lock_type = CIFS_WRLCK;
 789                         rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
 790                                         length, pfLock,
 791                                         posix_lock_type, wait_flag);
 792                         FreeXid(xid);
 793                         return rc;
 794                 }
 795
 796                 /* BB we could chain these into one lock request BB */
 797                 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
 798                                  0, 1, lockType, 0 /* wait flag */ );
 799                 if (rc == 0) {
 800                         rc = CIFSSMBLock(xid, tcon, netfid, length,
 801                                          pfLock->fl_start, 1 /* numUnlock */ ,
 802                                          0 /* numLock */ , lockType,
 803                                          0 /* wait flag */ );
 804                         pfLock->fl_type = F_UNLCK;
 805                         if (rc != 0)
 806                                 cERROR(1, "Error unlocking previously locked "
 807                                            "range %d during test of lock", rc);
 808                         rc = 0;
 809
 810                 } else {
 811                         /* if rc == ERR_SHARING_VIOLATION ? */
 812                         rc = 0;
 813
 814                         if (lockType & LOCKING_ANDX_SHARED_LOCK) {
 815                                 pfLock->fl_type = F_WRLCK;
 816                         } else {
 817                                 rc = CIFSSMBLock(xid, tcon, netfid, length,
 818                                         pfLock->fl_start, 0, 1,
 819                                         lockType | LOCKING_ANDX_SHARED_LOCK,
 820                                         0 /* wait flag */);
 821                                 if (rc == 0) {
 822                                         rc = CIFSSMBLock(xid, tcon, netfid,
 823                                                 length, pfLock->fl_start, 1, 0,
 824                                                 lockType |
 825                                                 LOCKING_ANDX_SHARED_LOCK,
 826                                                 0 /* wait flag */);
 827                                         pfLock->fl_type = F_RDLCK;
 828                                         if (rc != 0)
 829                                                 cERROR(1, "Error unlocking "
 830                                                 "previously locked range %d "
 831                                                 "during test of lock", rc);
 832                                         rc = 0;
 833                                 } else {
 834                                         pfLock->fl_type = F_WRLCK;
 835                                         rc = 0;
 836                                 }
 837                         }
 838                 }
 839
 840                 FreeXid(xid);
 841                 return rc;
 842         }
 843
 844         if (!numLock && !numUnlock) {
 845                 /* if no lock or unlock then nothing
 846                 to do since we do not know what it is */
 847                 FreeXid(xid);
 848                 return -EOPNOTSUPP;
 849         }
 850
 851         if (posix_locking) {
 852                 int posix_lock_type;
 853                 if (lockType & LOCKING_ANDX_SHARED_LOCK)
 854                         posix_lock_type = CIFS_RDLCK;
 855                 else
 856                         posix_lock_type = CIFS_WRLCK;
 857
 858                 if (numUnlock == 1)
 859                         posix_lock_type = CIFS_UNLCK;
 860
 861                 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
 862                                       length, pfLock,
 863                                       posix_lock_type, wait_flag);
 864         } else {
 865                 struct cifsFileInfo *fid = file->private_data;
 866
 867                 if (numLock) {
 868                         rc = CIFSSMBLock(xid, tcon, netfid, length,
 869                                         pfLock->fl_start,
 870                                         0, numLock, lockType, wait_flag);
 871
 872                         if (rc == 0) {
 873                                 /* For Windows locks we must store them. */
 874                                 rc = store_file_lock(fid, length,
 875                                                 pfLock->fl_start, lockType);
 876                         }
 877                 } else if (numUnlock) {
 878                         /* For each stored lock that this unlock overlaps
 879                            completely, unlock it. */
 880                         int stored_rc = 0;
 881                         struct cifsLockInfo *li, *tmp;
 882
 883                         rc = 0;
 884                         mutex_lock(&fid->lock_mutex);
 885                         list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
 886                                 if (pfLock->fl_start <= li->offset &&
 887                                                 (pfLock->fl_start + length) >=
 888                                                 (li->offset + li->length)) {
 889                                         stored_rc = CIFSSMBLock(xid, tcon,
 890                                                         netfid,
 891                                                         li->length, li->offset,
 892                                                         1, 0, li->type, false);
 893                                         if (stored_rc)
 894                                                 rc = stored_rc;
 895                                         else {
 896                                                 list_del(&li->llist);
 897                                                 kfree(li);
 898                                         }
 899                                 }
 900                         }
 901                         mutex_unlock(&fid->lock_mutex);
 902                 }
 903         }
 904
 905         if (pfLock->fl_flags & FL_POSIX)
 906                 posix_lock_file_wait(file, pfLock);
 907         FreeXid(xid);
 908         return rc;
 909 }
 910
 911 /*
 912  * Set the timeout on write requests past EOF. For some servers (Windows)
 913  * these calls can be very long.
 914  *
 915  * If we're writing >10M past the EOF we give a 180s timeout. Anything less
 916  * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
 917  * The 10M cutoff is totally arbitrary. A better scheme for this would be
 918  * welcome if someone wants to suggest one.
 919  *
 920  * We may be able to do a better job with this if there were some way to
 921  * declare that a file should be sparse.
 922  */
 923 static int
 924 cifs_write_timeout(struct cifsInodeInfo *cifsi, loff_t offset)
 925 {
 926         if (offset <= cifsi->server_eof)
 927                 return CIFS_STD_OP;
 928         else if (offset > (cifsi->server_eof + (10 * 1024 * 1024)))
 929                 return CIFS_VLONG_OP;
 930         else
 931                 return CIFS_LONG_OP;
 932 }
 933
 934 /* update the file size (if needed) after a write */
 935 static void
 936 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
 937                       unsigned int bytes_written)
 938 {
 939         loff_t end_of_write = offset + bytes_written;
 940
 941         if (end_of_write > cifsi->server_eof)
 942                 cifsi->server_eof = end_of_write;
 943 }
 944
 945 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
 946         size_t write_size, loff_t *poffset)
 947 {
 948         int rc = 0;
 949         unsigned int bytes_written = 0;
 950         unsigned int total_written;
 951         struct cifs_sb_info *cifs_sb;
 952         struct cifsTconInfo *pTcon;
 953         int xid, long_op;
 954         struct cifsFileInfo *open_file;
 955         struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
 956
 957         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
 958
 959         /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
 960            *poffset, file->f_path.dentry->d_name.name); */
 961
 962         if (file->private_data == NULL)
 963                 return -EBADF;
 964
 965         open_file = file->private_data;
 966         pTcon = open_file->tcon;
 967
 968         rc = generic_write_checks(file, poffset, &write_size, 0);
 969         if (rc)
 970                 return rc;
 971
 972         xid = GetXid();
 973
 974         long_op = cifs_write_timeout(cifsi, *poffset);
 975         for (total_written = 0; write_size > total_written;
 976              total_written += bytes_written) {
 977                 rc = -EAGAIN;
 978                 while (rc == -EAGAIN) {
 979                         if (file->private_data == NULL) {
 980                                 /* file has been closed on us */
 981                                 FreeXid(xid);
 982                         /* if we have gotten here we have written some data
 983                            and blocked, and the file has been freed on us while
 984                            we blocked so return what we managed to write */
 985                                 return total_written;
 986                         }
 987                         if (open_file->closePend) {
 988                                 FreeXid(xid);
 989                                 if (total_written)
 990                                         return total_written;
 991                                 else
 992                                         return -EBADF;
 993                         }
 994                         if (open_file->invalidHandle) {
 995                                 /* we could deadlock if we called
 996                                    filemap_fdatawait from here so tell
 997                                    reopen_file not to flush data to server
 998                                    now */
 999                                 rc = cifs_reopen_file(file, false);
1000                                 if (rc != 0)
1001                                         break;
1002                         }
1003
1004                         rc = CIFSSMBWrite(xid, pTcon,
1005                                 open_file->netfid,
1006                                 min_t(const int, cifs_sb->wsize,
1007                                       write_size - total_written),
1008                                 *poffset, &bytes_written,
1009                                 NULL, write_data + total_written, long_op);
1010                 }
1011                 if (rc || (bytes_written == 0)) {
1012                         if (total_written)
1013                                 break;
1014                         else {
1015                                 FreeXid(xid);
1016                                 return rc;
1017                         }
1018                 } else {
1019                         cifs_update_eof(cifsi, *poffset, bytes_written);
1020                         *poffset += bytes_written;
1021                 }
1022                 long_op = CIFS_STD_OP; /* subsequent writes fast -
1023                                     15 seconds is plenty */
1024         }
1025
1026         cifs_stats_bytes_written(pTcon, total_written);
1027
1028         /* since the write may have blocked check these pointers again */
1029         if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1030                 struct inode *inode = file->f_path.dentry->d_inode;
1031 /* Do not update local mtime - server will set its actual value on write
1032  *              inode->i_ctime = inode->i_mtime =
1033  *                      current_fs_time(inode->i_sb);*/
1034                 if (total_written > 0) {
1035                         spin_lock(&inode->i_lock);
1036                         if (*poffset > file->f_path.dentry->d_inode->i_size)
1037                                 i_size_write(file->f_path.dentry->d_inode,
1038                                         *poffset);
1039                         spin_unlock(&inode->i_lock);
1040                 }
1041                 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1042         }
1043         FreeXid(xid);
1044         return total_written;
1045 }
1046
1047 static ssize_t cifs_write(struct file *file, const char *write_data,
1048                           size_t write_size, loff_t *poffset)
1049 {
1050         int rc = 0;
1051         unsigned int bytes_written = 0;
1052         unsigned int total_written;
1053         struct cifs_sb_info *cifs_sb;
1054         struct cifsTconInfo *pTcon;
1055         int xid, long_op;
1056         struct cifsFileInfo *open_file;
1057         struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
1058
1059         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1060
1061         cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
1062            *poffset, file->f_path.dentry->d_name.name);
1063
1064         if (file->private_data == NULL)
1065                 return -EBADF;
1066         open_file = file->private_data;
1067         pTcon = open_file->tcon;
1068
1069         xid = GetXid();
1070
1071         long_op = cifs_write_timeout(cifsi, *poffset);
1072         for (total_written = 0; write_size > total_written;
1073              total_written += bytes_written) {
1074                 rc = -EAGAIN;
1075                 while (rc == -EAGAIN) {
1076                         if (file->private_data == NULL) {
1077                                 /* file has been closed on us */
1078                                 FreeXid(xid);
1079                         /* if we have gotten here we have written some data
1080                            and blocked, and the file has been freed on us
1081                            while we blocked so return what we managed to
1082                            write */
1083                                 return total_written;
1084                         }
1085                         if (open_file->closePend) {
1086                                 FreeXid(xid);
1087                                 if (total_written)
1088                                         return total_written;
1089                                 else
1090                                         return -EBADF;
1091                         }
1092                         if (open_file->invalidHandle) {
1093                                 /* we could deadlock if we called
1094                                    filemap_fdatawait from here so tell
1095                                    reopen_file not to flush data to
1096                                    server now */
1097                                 rc = cifs_reopen_file(file, false);
1098                                 if (rc != 0)
1099                                         break;
1100                         }
1101                         if (experimEnabled || (pTcon->ses->server &&
1102                                 ((pTcon->ses->server->secMode &
1103                                 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1104                                 == 0))) {
1105                                 struct kvec iov[2];
1106                                 unsigned int len;
1107
1108                                 len = min((size_t)cifs_sb->wsize,
1109                                           write_size - total_written);
1110                                 /* iov[0] is reserved for smb header */
1111                                 iov[1].iov_base = (char *)write_data +
1112                                                   total_written;
1113                                 iov[1].iov_len = len;
1114                                 rc = CIFSSMBWrite2(xid, pTcon,
1115                                                 open_file->netfid, len,
1116                                                 *poffset, &bytes_written,
1117                                                 iov, 1, long_op);
1118                         } else
1119                                 rc = CIFSSMBWrite(xid, pTcon,
1120                                          open_file->netfid,
1121                                          min_t(const int, cifs_sb->wsize,
1122                                                write_size - total_written),
1123                                          *poffset, &bytes_written,
1124                                          write_data + total_written,
1125                                          NULL, long_op);
1126                 }
1127                 if (rc || (bytes_written == 0)) {
1128                         if (total_written)
1129                                 break;
1130                         else {
1131                                 FreeXid(xid);
1132                                 return rc;
1133                         }
1134                 } else {
1135                         cifs_update_eof(cifsi, *poffset, bytes_written);
1136                         *poffset += bytes_written;
1137                 }
1138                 long_op = CIFS_STD_OP; /* subsequent writes fast -
1139                                     15 seconds is plenty */
1140         }
1141
1142         cifs_stats_bytes_written(pTcon, total_written);
1143
1144         /* since the write may have blocked check these pointers again */
1145         if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1146 /*BB We could make this contingent on superblock ATIME flag too */
1147 /*              file->f_path.dentry->d_inode->i_ctime =
1148                 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
1149                 if (total_written > 0) {
1150                         spin_lock(&file->f_path.dentry->d_inode->i_lock);
1151                         if (*poffset > file->f_path.dentry->d_inode->i_size)
1152                                 i_size_write(file->f_path.dentry->d_inode,
1153                                              *poffset);
1154                         spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1155                 }
1156                 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1157         }
1158         FreeXid(xid);
1159         return total_written;
1160 }
1161
1162 #ifdef CONFIG_CIFS_EXPERIMENTAL
1163 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode)
1164 {
1165         struct cifsFileInfo *open_file = NULL;
1166
1167         read_lock(&GlobalSMBSeslock);
1168         /* we could simply get the first_list_entry since write-only entries
1169            are always at the end of the list but since the first entry might
1170            have a close pending, we go through the whole list */
1171         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1172                 if (open_file->closePend)
1173                         continue;
1174                 if (open_file->pfile && ((open_file->pfile->f_flags & O_RDWR) ||
1175                     (open_file->pfile->f_flags & O_RDONLY))) {
1176                         if (!open_file->invalidHandle) {
1177                                 /* found a good file */
1178                                 /* lock it so it will not be closed on us */
1179                                 cifsFileInfo_get(open_file);
1180                                 read_unlock(&GlobalSMBSeslock);
1181                                 return open_file;
1182                         } /* else might as well continue, and look for
1183                              another, or simply have the caller reopen it
1184                              again rather than trying to fix this handle */
1185                 } else /* write only file */
1186                         break; /* write only files are last so must be done */
1187         }
1188         read_unlock(&GlobalSMBSeslock);
1189         return NULL;
1190 }
1191 #endif
1192
1193 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1194 {
1195         struct cifsFileInfo *open_file;
1196         bool any_available = false;
1197         int rc;
1198
1199         /* Having a null inode here (because mapping->host was set to zero by
1200         the VFS or MM) should not happen but we had reports of on oops (due to
1201         it being zero) during stress testcases so we need to check for it */
1202
1203         if (cifs_inode == NULL) {
1204                 cERROR(1, "Null inode passed to cifs_writeable_file");
1205                 dump_stack();
1206                 return NULL;
1207         }
1208
1209         read_lock(&GlobalSMBSeslock);
1210 refind_writable:
1211         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1212                 if (open_file->closePend ||
1213                     (!any_available && open_file->pid != current->tgid))
1214                         continue;
1215
1216                 if (open_file->pfile &&
1217                     ((open_file->pfile->f_flags & O_RDWR) ||
1218                      (open_file->pfile->f_flags & O_WRONLY))) {
1219                         cifsFileInfo_get(open_file);
1220
1221                         if (!open_file->invalidHandle) {
1222                                 /* found a good writable file */
1223                                 read_unlock(&GlobalSMBSeslock);
1224                                 return open_file;
1225                         }
1226
1227                         read_unlock(&GlobalSMBSeslock);
1228                         /* Had to unlock since following call can block */
1229                         rc = cifs_reopen_file(open_file->pfile, false);
1230                         if (!rc) {
1231                                 if (!open_file->closePend)
1232                                         return open_file;
1233                                 else { /* start over in case this was deleted */
1234                                        /* since the list could be modified */
1235                                         read_lock(&GlobalSMBSeslock);
1236                                         cifsFileInfo_put(open_file);
1237                                         goto refind_writable;
1238                                 }
1239                         }
1240
1241                         /* if it fails, try another handle if possible -
1242                         (we can not do this if closePending since
1243                         loop could be modified - in which case we
1244                         have to start at the beginning of the list
1245                         again. Note that it would be bad
1246                         to hold up writepages here (rather than
1247                         in caller) with continuous retries */
1248                         cFYI(1, "wp failed on reopen file");
1249                         read_lock(&GlobalSMBSeslock);
1250                         /* can not use this handle, no write
1251                            pending on this one after all */
1252                         cifsFileInfo_put(open_file);
1253
1254                         if (open_file->closePend) /* list could have changed */
1255                                 goto refind_writable;
1256                         /* else we simply continue to the next entry. Thus
1257                            we do not loop on reopen errors.  If we
1258                            can not reopen the file, for example if we
1259                            reconnected to a server with another client
1260                            racing to delete or lock the file we would not
1261                            make progress if we restarted before the beginning
1262                            of the loop here. */
1263                 }
1264         }
1265         /* couldn't find useable FH with same pid, try any available */
1266         if (!any_available) {
1267                 any_available = true;
1268                 goto refind_writable;
1269         }
1270         read_unlock(&GlobalSMBSeslock);
1271         return NULL;
1272 }
1273
1274 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1275 {
1276         struct address_space *mapping = page->mapping;
1277         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1278         char *write_data;
1279         int rc = -EFAULT;
1280         int bytes_written = 0;
1281         struct cifs_sb_info *cifs_sb;
1282         struct inode *inode;
1283         struct cifsFileInfo *open_file;
1284
1285         if (!mapping || !mapping->host)
1286                 return -EFAULT;
1287
1288         inode = page->mapping->host;
1289         cifs_sb = CIFS_SB(inode->i_sb);
1290
1291         offset += (loff_t)from;
1292         write_data = kmap(page);
1293         write_data += from;
1294
1295         if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1296                 kunmap(page);
1297                 return -EIO;
1298         }
1299
1300         /* racing with truncate? */
1301         if (offset > mapping->host->i_size) {
1302                 kunmap(page);
1303                 return 0; /* don't care */
1304         }
1305
1306         /* check to make sure that we are not extending the file */
1307         if (mapping->host->i_size - offset < (loff_t)to)
1308                 to = (unsigned)(mapping->host->i_size - offset);
1309
1310         open_file = find_writable_file(CIFS_I(mapping->host));
1311         if (open_file) {
1312                 bytes_written = cifs_write(open_file->pfile, write_data,
1313                                            to-from, &offset);
1314                 cifsFileInfo_put(open_file);
1315                 /* Does mm or vfs already set times? */
1316                 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1317                 if ((bytes_written > 0) && (offset))
1318                         rc = 0;
1319                 else if (bytes_written < 0)
1320                         rc = bytes_written;
1321         } else {
1322                 cFYI(1, "No writeable filehandles for inode");
1323                 rc = -EIO;
1324         }
1325
1326         kunmap(page);
1327         return rc;
1328 }
1329
1330 static int cifs_writepages(struct address_space *mapping,
1331                            struct writeback_control *wbc)
1332 {
1333         struct backing_dev_info *bdi = mapping->backing_dev_info;
1334         unsigned int bytes_to_write;
1335         unsigned int bytes_written;
1336         struct cifs_sb_info *cifs_sb;
1337         int done = 0;
1338         pgoff_t end;
1339         pgoff_t index;
1340         int range_whole = 0;
1341         struct kvec *iov;
1342         int len;
1343         int n_iov = 0;
1344         pgoff_t next;
1345         int nr_pages;
1346         __u64 offset = 0;
1347         struct cifsFileInfo *open_file;
1348         struct cifsTconInfo *tcon;
1349         struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1350         struct page *page;
1351         struct pagevec pvec;
1352         int rc = 0;
1353         int scanned = 0;
1354         int xid, long_op;
1355
1356         cifs_sb = CIFS_SB(mapping->host->i_sb);
1357
1358         /*
1359          * If wsize is smaller that the page cache size, default to writing
1360          * one page at a time via cifs_writepage
1361          */
1362         if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1363                 return generic_writepages(mapping, wbc);
1364
1365         if ((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1366                 if (cifs_sb->tcon->ses->server->secMode &
1367                                 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1368                         if (!experimEnabled)
1369                                 return generic_writepages(mapping, wbc);
1370
1371         iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1372         if (iov == NULL)
1373                 return generic_writepages(mapping, wbc);
1374
1375
1376         /*
1377          * BB: Is this meaningful for a non-block-device file system?
1378          * If it is, we should test it again after we do I/O
1379          */
1380         if (wbc->nonblocking && bdi_write_congested(bdi)) {
1381                 wbc->encountered_congestion = 1;
1382                 kfree(iov);
1383                 return 0;
1384         }
1385
1386         xid = GetXid();
1387
1388         pagevec_init(&pvec, 0);
1389         if (wbc->range_cyclic) {
1390                 index = mapping->writeback_index; /* Start from prev offset */
1391                 end = -1;
1392         } else {
1393                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1394                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1395                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1396                         range_whole = 1;
1397                 scanned = 1;
1398         }
1399 retry:
1400         while (!done && (index <= end) &&
1401                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1402                         PAGECACHE_TAG_DIRTY,
1403                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1404                 int first;
1405                 unsigned int i;
1406
1407                 first = -1;
1408                 next = 0;
1409                 n_iov = 0;
1410                 bytes_to_write = 0;
1411
1412                 for (i = 0; i < nr_pages; i++) {
1413                         page = pvec.pages[i];
1414                         /*
1415                          * At this point we hold neither mapping->tree_lock nor
1416                          * lock on the page itself: the page may be truncated or
1417                          * invalidated (changing page->mapping to NULL), or even
1418                          * swizzled back from swapper_space to tmpfs file
1419                          * mapping
1420                          */
1421
1422                         if (first < 0)
1423                                 lock_page(page);
1424                         else if (!trylock_page(page))
1425                                 break;
1426
1427                         if (unlikely(page->mapping != mapping)) {
1428                                 unlock_page(page);
1429                                 break;
1430                         }
1431
1432                         if (!wbc->range_cyclic && page->index > end) {
1433                                 done = 1;
1434                                 unlock_page(page);
1435                                 break;
1436                         }
1437
1438                         if (next && (page->index != next)) {
1439                                 /* Not next consecutive page */
1440                                 unlock_page(page);
1441                                 break;
1442                         }
1443
1444                         if (wbc->sync_mode != WB_SYNC_NONE)
1445                                 wait_on_page_writeback(page);
1446
1447                         if (PageWriteback(page) ||
1448                                         !clear_page_dirty_for_io(page)) {
1449                                 unlock_page(page);
1450                                 break;
1451                         }
1452
1453                         /*
1454                          * This actually clears the dirty bit in the radix tree.
1455                          * See cifs_writepage() for more commentary.
1456                          */
1457                         set_page_writeback(page);
1458
1459                         if (page_offset(page) >= mapping->host->i_size) {
1460                                 done = 1;
1461                                 unlock_page(page);
1462                                 end_page_writeback(page);
1463                                 break;
1464                         }
1465
1466                         /*
1467                          * BB can we get rid of this?  pages are held by pvec
1468                          */
1469                         page_cache_get(page);
1470
1471                         len = min(mapping->host->i_size - page_offset(page),
1472                                   (loff_t)PAGE_CACHE_SIZE);
1473
1474                         /* reserve iov[0] for the smb header */
1475                         n_iov++;
1476                         iov[n_iov].iov_base = kmap(page);
1477                         iov[n_iov].iov_len = len;
1478                         bytes_to_write += len;
1479
1480                         if (first < 0) {
1481                                 first = i;
1482                                 offset = page_offset(page);
1483                         }
1484                         next = page->index + 1;
1485                         if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1486                                 break;
1487                 }
1488                 if (n_iov) {
1489                         /* Search for a writable handle every time we call
1490                          * CIFSSMBWrite2.  We can't rely on the last handle
1491                          * we used to still be valid
1492                          */
1493                         open_file = find_writable_file(CIFS_I(mapping->host));
1494                         if (!open_file) {
1495                                 cERROR(1, "No writable handles for inode");
1496                                 rc = -EBADF;
1497                         } else {
1498                                 tcon = open_file->tcon;
1499                                 long_op = cifs_write_timeout(cifsi, offset);
1500                                 rc = CIFSSMBWrite2(xid, tcon,
1501                                                    open_file->netfid,
1502                                                    bytes_to_write, offset,
1503                                                    &bytes_written, iov, n_iov,
1504                                                    long_op);
1505                                 cifsFileInfo_put(open_file);
1506                                 cifs_update_eof(cifsi, offset, bytes_written);
1507
1508                                 if (rc || bytes_written < bytes_to_write) {
1509                                         cERROR(1, "Write2 ret %d, wrote %d",
1510                                                   rc, bytes_written);
1511                                         /* BB what if continued retry is
1512                                            requested via mount flags? */
1513                                         if (rc == -ENOSPC)
1514                                                 set_bit(AS_ENOSPC, &mapping->flags);
1515                                         else
1516                                                 set_bit(AS_EIO, &mapping->flags);
1517                                 } else {
1518                                         cifs_stats_bytes_written(tcon, bytes_written);
1519                                 }
1520                         }
1521                         for (i = 0; i < n_iov; i++) {
1522                                 page = pvec.pages[first + i];
1523                                 /* Should we also set page error on
1524                                 success rc but too little data written? */
1525                                 /* BB investigate retry logic on temporary
1526                                 server crash cases and how recovery works
1527                                 when page marked as error */
1528                                 if (rc)
1529                                         SetPageError(page);
1530                                 kunmap(page);
1531                                 unlock_page(page);
1532                                 end_page_writeback(page);
1533                                 page_cache_release(page);
1534                         }
1535                         if ((wbc->nr_to_write -= n_iov) <= 0)
1536                                 done = 1;
1537                         index = next;
1538                 } else
1539                         /* Need to re-find the pages we skipped */
1540                         index = pvec.pages[0]->index + 1;
1541
1542                 pagevec_release(&pvec);
1543         }
1544         if (!scanned && !done) {
1545                 /*
1546                  * We hit the last page and there is more work to be done: wrap
1547                  * back to the start of the file
1548                  */
1549                 scanned = 1;
1550                 index = 0;
1551                 goto retry;
1552         }
1553         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1554                 mapping->writeback_index = index;
1555
1556         FreeXid(xid);
1557         kfree(iov);
1558         return rc;
1559 }
1560
1561 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1562 {
1563         int rc = -EFAULT;
1564         int xid;
1565
1566         xid = GetXid();
1567 /* BB add check for wbc flags */
1568         page_cache_get(page);
1569         if (!PageUptodate(page))
1570                 cFYI(1, "ppw - page not up to date");
1571
1572         /*
1573          * Set the "writeback" flag, and clear "dirty" in the radix tree.
1574          *
1575          * A writepage() implementation always needs to do either this,
1576          * or re-dirty the page with "redirty_page_for_writepage()" in
1577          * the case of a failure.
1578          *
1579          * Just unlocking the page will cause the radix tree tag-bits
1580          * to fail to update with the state of the page correctly.
1581          */
1582         set_page_writeback(page);
1583         rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1584         SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1585         unlock_page(page);
1586         end_page_writeback(page);
1587         page_cache_release(page);
1588         FreeXid(xid);
1589         return rc;
1590 }
1591
1592 static int cifs_write_end(struct file *file, struct address_space *mapping,
1593                         loff_t pos, unsigned len, unsigned copied,
1594                         struct page *page, void *fsdata)
1595 {
1596         int rc;
1597         struct inode *inode = mapping->host;
1598
1599         cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1600                  page, pos, copied);
1601
1602         if (PageChecked(page)) {
1603                 if (copied == len)
1604                         SetPageUptodate(page);
1605                 ClearPageChecked(page);
1606         } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1607                 SetPageUptodate(page);
1608
1609         if (!PageUptodate(page)) {
1610                 char *page_data;
1611                 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1612                 int xid;
1613
1614                 xid = GetXid();
1615                 /* this is probably better than directly calling
1616                    partialpage_write since in this function the file handle is
1617                    known which we might as well leverage */
1618                 /* BB check if anything else missing out of ppw
1619                    such as updating last write time */
1620                 page_data = kmap(page);
1621                 rc = cifs_write(file, page_data + offset, copied, &pos);
1622                 /* if (rc < 0) should we set writebehind rc? */
1623                 kunmap(page);
1624
1625                 FreeXid(xid);
1626         } else {
1627                 rc = copied;
1628                 pos += copied;
1629                 set_page_dirty(page);
1630         }
1631
1632         if (rc > 0) {
1633                 spin_lock(&inode->i_lock);
1634                 if (pos > inode->i_size)
1635                         i_size_write(inode, pos);
1636                 spin_unlock(&inode->i_lock);
1637         }
1638
1639         unlock_page(page);
1640         page_cache_release(page);
1641
1642         return rc;
1643 }
1644
1645 int cifs_fsync(struct file *file, int datasync)
1646 {
1647         int xid;
1648         int rc = 0;
1649         struct cifsTconInfo *tcon;
1650         struct cifsFileInfo *smbfile = file->private_data;
1651         struct inode *inode = file->f_path.dentry->d_inode;
1652
1653         xid = GetXid();
1654
1655         cFYI(1, "Sync file - name: %s datasync: 0x%x",
1656                 file->f_path.dentry->d_name.name, datasync);
1657
1658         rc = filemap_write_and_wait(inode->i_mapping);
1659         if (rc == 0) {
1660                 rc = CIFS_I(inode)->write_behind_rc;
1661                 CIFS_I(inode)->write_behind_rc = 0;
1662                 tcon = smbfile->tcon;
1663                 if (!rc && tcon && smbfile &&
1664                    !(CIFS_SB(inode->i_sb)->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1665                         rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1666         }
1667
1668         FreeXid(xid);
1669         return rc;
1670 }
1671
1672 /* static void cifs_sync_page(struct page *page)
1673 {
1674         struct address_space *mapping;
1675         struct inode *inode;
1676         unsigned long index = page->index;
1677         unsigned int rpages = 0;
1678         int rc = 0;
1679
1680         cFYI(1, "sync page %p", page);
1681         mapping = page->mapping;
1682         if (!mapping)
1683                 return 0;
1684         inode = mapping->host;
1685         if (!inode)
1686                 return; */
1687
1688 /*      fill in rpages then
1689         result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1690
1691 /*      cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1692
1693 #if 0
1694         if (rc < 0)
1695                 return rc;
1696         return 0;
1697 #endif
1698 } */
1699
1700 /*
1701  * As file closes, flush all cached write data for this inode checking
1702  * for write behind errors.
1703  */
1704 int cifs_flush(struct file *file, fl_owner_t id)
1705 {
1706         struct inode *inode = file->f_path.dentry->d_inode;
1707         int rc = 0;
1708
1709         /* Rather than do the steps manually:
1710            lock the inode for writing
1711            loop through pages looking for write behind data (dirty pages)
1712            coalesce into contiguous 16K (or smaller) chunks to write to server
1713            send to server (prefer in parallel)
1714            deal with writebehind errors
1715            unlock inode for writing
1716            filemapfdatawrite appears easier for the time being */
1717
1718         rc = filemap_fdatawrite(inode->i_mapping);
1719         /* reset wb rc if we were able to write out dirty pages */
1720         if (!rc) {
1721                 rc = CIFS_I(inode)->write_behind_rc;
1722                 CIFS_I(inode)->write_behind_rc = 0;
1723         }
1724
1725         cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1726
1727         return rc;
1728 }
1729
1730 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1731         size_t read_size, loff_t *poffset)
1732 {
1733         int rc = -EACCES;
1734         unsigned int bytes_read = 0;
1735         unsigned int total_read = 0;
1736         unsigned int current_read_size;
1737         struct cifs_sb_info *cifs_sb;
1738         struct cifsTconInfo *pTcon;
1739         int xid;
1740         struct cifsFileInfo *open_file;
1741         char *smb_read_data;
1742         char __user *current_offset;
1743         struct smb_com_read_rsp *pSMBr;
1744
1745         xid = GetXid();
1746         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1747
1748         if (file->private_data == NULL) {
1749                 rc = -EBADF;
1750                 FreeXid(xid);
1751                 return rc;
1752         }
1753         open_file = file->private_data;
1754         pTcon = open_file->tcon;
1755
1756         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1757                 cFYI(1, "attempting read on write only file instance");
1758
1759         for (total_read = 0, current_offset = read_data;
1760              read_size > total_read;
1761              total_read += bytes_read, current_offset += bytes_read) {
1762                 current_read_size = min_t(const int, read_size - total_read,
1763                                           cifs_sb->rsize);
1764                 rc = -EAGAIN;
1765                 smb_read_data = NULL;
1766                 while (rc == -EAGAIN) {
1767                         int buf_type = CIFS_NO_BUFFER;
1768                         if ((open_file->invalidHandle) &&
1769                             (!open_file->closePend)) {
1770                                 rc = cifs_reopen_file(file, true);
1771                                 if (rc != 0)
1772                                         break;
1773                         }
1774                         rc = CIFSSMBRead(xid, pTcon,
1775                                          open_file->netfid,
1776                                          current_read_size, *poffset,
1777                                          &bytes_read, &smb_read_data,
1778                                          &buf_type);
1779                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1780                         if (smb_read_data) {
1781                                 if (copy_to_user(current_offset,
1782                                                 smb_read_data +
1783                                                 4 /* RFC1001 length field */ +
1784                                                 le16_to_cpu(pSMBr->DataOffset),
1785                                                 bytes_read))
1786                                         rc = -EFAULT;
1787
1788                                 if (buf_type == CIFS_SMALL_BUFFER)
1789                                         cifs_small_buf_release(smb_read_data);
1790                                 else if (buf_type == CIFS_LARGE_BUFFER)
1791                                         cifs_buf_release(smb_read_data);
1792                                 smb_read_data = NULL;
1793                         }
1794                 }
1795                 if (rc || (bytes_read == 0)) {
1796                         if (total_read) {
1797                                 break;
1798                         } else {
1799                                 FreeXid(xid);
1800                                 return rc;
1801                         }
1802                 } else {
1803                         cifs_stats_bytes_read(pTcon, bytes_read);
1804                         *poffset += bytes_read;
1805                 }
1806         }
1807         FreeXid(xid);
1808         return total_read;
1809 }
1810
1811
1812 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1813         loff_t *poffset)
1814 {
1815         int rc = -EACCES;
1816         unsigned int bytes_read = 0;
1817         unsigned int total_read;
1818         unsigned int current_read_size;
1819         struct cifs_sb_info *cifs_sb;
1820         struct cifsTconInfo *pTcon;
1821         int xid;
1822         char *current_offset;
1823         struct cifsFileInfo *open_file;
1824         int buf_type = CIFS_NO_BUFFER;
1825
1826         xid = GetXid();
1827         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1828
1829         if (file->private_data == NULL) {
1830                 rc = -EBADF;
1831                 FreeXid(xid);
1832                 return rc;
1833         }
1834         open_file = file->private_data;
1835         pTcon = open_file->tcon;
1836
1837         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1838                 cFYI(1, "attempting read on write only file instance");
1839
1840         for (total_read = 0, current_offset = read_data;
1841              read_size > total_read;
1842              total_read += bytes_read, current_offset += bytes_read) {
1843                 current_read_size = min_t(const int, read_size - total_read,
1844                                           cifs_sb->rsize);
1845                 /* For windows me and 9x we do not want to request more
1846                 than it negotiated since it will refuse the read then */
1847                 if ((pTcon->ses) &&
1848                         !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1849                         current_read_size = min_t(const int, current_read_size,
1850                                         pTcon->ses->server->maxBuf - 128);
1851                 }
1852                 rc = -EAGAIN;
1853                 while (rc == -EAGAIN) {
1854                         if ((open_file->invalidHandle) &&
1855                             (!open_file->closePend)) {
1856                                 rc = cifs_reopen_file(file, true);
1857                                 if (rc != 0)
1858                                         break;
1859                         }
1860                         rc = CIFSSMBRead(xid, pTcon,
1861                                          open_file->netfid,
1862                                          current_read_size, *poffset,
1863                                          &bytes_read, &current_offset,
1864                                          &buf_type);
1865                 }
1866                 if (rc || (bytes_read == 0)) {
1867                         if (total_read) {
1868                                 break;
1869                         } else {
1870                                 FreeXid(xid);
1871                                 return rc;
1872                         }
1873                 } else {
1874                         cifs_stats_bytes_read(pTcon, total_read);
1875                         *poffset += bytes_read;
1876                 }
1877         }
1878         FreeXid(xid);
1879         return total_read;
1880 }
1881
1882 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1883 {
1884         int rc, xid;
1885
1886         xid = GetXid();
1887         rc = cifs_revalidate_file(file);
1888         if (rc) {
1889                 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1890                 FreeXid(xid);
1891                 return rc;
1892         }
1893         rc = generic_file_mmap(file, vma);
1894         FreeXid(xid);
1895         return rc;
1896 }
1897
1898
1899 static void cifs_copy_cache_pages(struct address_space *mapping,
1900         struct list_head *pages, int bytes_read, char *data)
1901 {
1902         struct page *page;
1903         char *target;
1904
1905         while (bytes_read > 0) {
1906                 if (list_empty(pages))
1907                         break;
1908
1909                 page = list_entry(pages->prev, struct page, lru);
1910                 list_del(&page->lru);
1911
1912                 if (add_to_page_cache_lru(page, mapping, page->index,
1913                                       GFP_KERNEL)) {
1914                         page_cache_release(page);
1915                         cFYI(1, "Add page cache failed");
1916                         data += PAGE_CACHE_SIZE;
1917                         bytes_read -= PAGE_CACHE_SIZE;
1918                         continue;
1919                 }
1920                 page_cache_release(page);
1921
1922                 target = kmap_atomic(page, KM_USER0);
1923
1924                 if (PAGE_CACHE_SIZE > bytes_read) {
1925                         memcpy(target, data, bytes_read);
1926                         /* zero the tail end of this partial page */
1927                         memset(target + bytes_read, 0,
1928                                PAGE_CACHE_SIZE - bytes_read);
1929                         bytes_read = 0;
1930                 } else {
1931                         memcpy(target, data, PAGE_CACHE_SIZE);
1932                         bytes_read -= PAGE_CACHE_SIZE;
1933                 }
1934                 kunmap_atomic(target, KM_USER0);
1935
1936                 flush_dcache_page(page);
1937                 SetPageUptodate(page);
1938                 unlock_page(page);
1939                 data += PAGE_CACHE_SIZE;
1940
1941                 /* add page to FS-Cache */
1942                 cifs_readpage_to_fscache(mapping->host, page);
1943         }
1944         return;
1945 }
1946
1947 static int cifs_readpages(struct file *file, struct address_space *mapping,
1948         struct list_head *page_list, unsigned num_pages)
1949 {
1950         int rc = -EACCES;
1951         int xid;
1952         loff_t offset;
1953         struct page *page;
1954         struct cifs_sb_info *cifs_sb;
1955         struct cifsTconInfo *pTcon;
1956         unsigned int bytes_read = 0;
1957         unsigned int read_size, i;
1958         char *smb_read_data = NULL;
1959         struct smb_com_read_rsp *pSMBr;
1960         struct cifsFileInfo *open_file;
1961         int buf_type = CIFS_NO_BUFFER;
1962
1963         xid = GetXid();
1964         if (file->private_data == NULL) {
1965                 rc = -EBADF;
1966                 FreeXid(xid);
1967                 return rc;
1968         }
1969         open_file = file->private_data;
1970         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1971         pTcon = open_file->tcon;
1972
1973         /*
1974          * Reads as many pages as possible from fscache. Returns -ENOBUFS
1975          * immediately if the cookie is negative
1976          */
1977         rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
1978                                          &num_pages);
1979         if (rc == 0)
1980                 goto read_complete;
1981
1982         cFYI(DBG2, "rpages: num pages %d", num_pages);
1983         for (i = 0; i < num_pages; ) {
1984                 unsigned contig_pages;
1985                 struct page *tmp_page;
1986                 unsigned long expected_index;
1987
1988                 if (list_empty(page_list))
1989                         break;
1990
1991                 page = list_entry(page_list->prev, struct page, lru);
1992                 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1993
1994                 /* count adjacent pages that we will read into */
1995                 contig_pages = 0;
1996                 expected_index =
1997                         list_entry(page_list->prev, struct page, lru)->index;
1998                 list_for_each_entry_reverse(tmp_page, page_list, lru) {
1999                         if (tmp_page->index == expected_index) {
2000                                 contig_pages++;
2001                                 expected_index++;
2002                         } else
2003                                 break;
2004                 }
2005                 if (contig_pages + i >  num_pages)
2006                         contig_pages = num_pages - i;
2007
2008                 /* for reads over a certain size could initiate async
2009                    read ahead */
2010
2011                 read_size = contig_pages * PAGE_CACHE_SIZE;
2012                 /* Read size needs to be in multiples of one page */
2013                 read_size = min_t(const unsigned int, read_size,
2014                                   cifs_sb->rsize & PAGE_CACHE_MASK);
2015                 cFYI(DBG2, "rpages: read size 0x%x  contiguous pages %d",
2016                                 read_size, contig_pages);
2017                 rc = -EAGAIN;
2018                 while (rc == -EAGAIN) {
2019                         if ((open_file->invalidHandle) &&
2020                             (!open_file->closePend)) {
2021                                 rc = cifs_reopen_file(file, true);
2022                                 if (rc != 0)
2023                                         break;
2024                         }
2025
2026                         rc = CIFSSMBRead(xid, pTcon,
2027                                          open_file->netfid,
2028                                          read_size, offset,
2029                                          &bytes_read, &smb_read_data,
2030                                          &buf_type);
2031                         /* BB more RC checks ? */
2032                         if (rc == -EAGAIN) {
2033                                 if (smb_read_data) {
2034                                         if (buf_type == CIFS_SMALL_BUFFER)
2035                                                 cifs_small_buf_release(smb_read_data);
2036                                         else if (buf_type == CIFS_LARGE_BUFFER)
2037                                                 cifs_buf_release(smb_read_data);
2038                                         smb_read_data = NULL;
2039                                 }
2040                         }
2041                 }
2042                 if ((rc < 0) || (smb_read_data == NULL)) {
2043                         cFYI(1, "Read error in readpages: %d", rc);
2044                         break;
2045                 } else if (bytes_read > 0) {
2046                         task_io_account_read(bytes_read);
2047                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2048                         cifs_copy_cache_pages(mapping, page_list, bytes_read,
2049                                 smb_read_data + 4 /* RFC1001 hdr */ +
2050                                 le16_to_cpu(pSMBr->DataOffset));
2051
2052                         i +=  bytes_read >> PAGE_CACHE_SHIFT;
2053                         cifs_stats_bytes_read(pTcon, bytes_read);
2054                         if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2055                                 i++; /* account for partial page */
2056
2057                                 /* server copy of file can have smaller size
2058                                    than client */
2059                                 /* BB do we need to verify this common case ?
2060                                    this case is ok - if we are at server EOF
2061                                    we will hit it on next read */
2062
2063                                 /* break; */
2064                         }
2065                 } else {
2066                         cFYI(1, "No bytes read (%d) at offset %lld . "
2067                                 "Cleaning remaining pages from readahead list",
2068                                 bytes_read, offset);
2069                         /* BB turn off caching and do new lookup on
2070                            file size at server? */
2071                         break;
2072                 }
2073                 if (smb_read_data) {
2074                         if (buf_type == CIFS_SMALL_BUFFER)
2075                                 cifs_small_buf_release(smb_read_data);
2076                         else if (buf_type == CIFS_LARGE_BUFFER)
2077                                 cifs_buf_release(smb_read_data);
2078                         smb_read_data = NULL;
2079                 }
2080                 bytes_read = 0;
2081         }
2082
2083 /* need to free smb_read_data buf before exit */
2084         if (smb_read_data) {
2085                 if (buf_type == CIFS_SMALL_BUFFER)
2086                         cifs_small_buf_release(smb_read_data);
2087                 else if (buf_type == CIFS_LARGE_BUFFER)
2088                         cifs_buf_release(smb_read_data);
2089                 smb_read_data = NULL;
2090         }
2091
2092 read_complete:
2093         FreeXid(xid);
2094         return rc;
2095 }
2096
2097 static int cifs_readpage_worker(struct file *file, struct page *page,
2098         loff_t *poffset)
2099 {
2100         char *read_data;
2101         int rc;
2102
2103         /* Is the page cached? */
2104         rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2105         if (rc == 0)
2106                 goto read_complete;
2107
2108         page_cache_get(page);
2109         read_data = kmap(page);
2110         /* for reads over a certain size could initiate async read ahead */
2111
2112         rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2113
2114         if (rc < 0)
2115                 goto io_error;
2116         else
2117                 cFYI(1, "Bytes read %d", rc);
2118
2119         file->f_path.dentry->d_inode->i_atime =
2120                 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2121
2122         if (PAGE_CACHE_SIZE > rc)
2123                 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2124
2125         flush_dcache_page(page);
2126         SetPageUptodate(page);
2127
2128         /* send this page to the cache */
2129         cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2130
2131         rc = 0;
2132
2133 io_error:
2134         kunmap(page);
2135         page_cache_release(page);
2136
2137 read_complete:
2138         return rc;
2139 }
2140
2141 static int cifs_readpage(struct file *file, struct page *page)
2142 {
2143         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2144         int rc = -EACCES;
2145         int xid;
2146
2147         xid = GetXid();
2148
2149         if (file->private_data == NULL) {
2150                 rc = -EBADF;
2151                 FreeXid(xid);
2152                 return rc;
2153         }
2154
2155         cFYI(1, "readpage %p at offset %d 0x%x\n",
2156                  page, (int)offset, (int)offset);
2157
2158         rc = cifs_readpage_worker(file, page, &offset);
2159
2160         unlock_page(page);
2161
2162         FreeXid(xid);
2163         return rc;
2164 }
2165
2166 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2167 {
2168         struct cifsFileInfo *open_file;
2169
2170         read_lock(&GlobalSMBSeslock);
2171         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2172                 if (open_file->closePend)
2173                         continue;
2174                 if (open_file->pfile &&
2175                     ((open_file->pfile->f_flags & O_RDWR) ||
2176                      (open_file->pfile->f_flags & O_WRONLY))) {
2177                         read_unlock(&GlobalSMBSeslock);
2178                         return 1;
2179                 }
2180         }
2181         read_unlock(&GlobalSMBSeslock);
2182         return 0;
2183 }
2184
2185 /* We do not want to update the file size from server for inodes
2186    open for write - to avoid races with writepage extending
2187    the file - in the future we could consider allowing
2188    refreshing the inode only on increases in the file size
2189    but this is tricky to do without racing with writebehind
2190    page caching in the current Linux kernel design */
2191 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2192 {
2193         if (!cifsInode)
2194                 return true;
2195
2196         if (is_inode_writable(cifsInode)) {
2197                 /* This inode is open for write at least once */
2198                 struct cifs_sb_info *cifs_sb;
2199
2200                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2201                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2202                         /* since no page cache to corrupt on directio
2203                         we can change size safely */
2204                         return true;
2205                 }
2206
2207                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2208                         return true;
2209
2210                 return false;
2211         } else
2212                 return true;
2213 }
2214
2215 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2216                         loff_t pos, unsigned len, unsigned flags,
2217                         struct page **pagep, void **fsdata)
2218 {
2219         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2220         loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2221         loff_t page_start = pos & PAGE_MASK;
2222         loff_t i_size;
2223         struct page *page;
2224         int rc = 0;
2225
2226         cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2227
2228         page = grab_cache_page_write_begin(mapping, index, flags);
2229         if (!page) {
2230                 rc = -ENOMEM;
2231                 goto out;
2232         }
2233
2234         if (PageUptodate(page))
2235                 goto out;
2236
2237         /*
2238          * If we write a full page it will be up to date, no need to read from
2239          * the server. If the write is short, we'll end up doing a sync write
2240          * instead.
2241          */
2242         if (len == PAGE_CACHE_SIZE)
2243                 goto out;
2244
2245         /*
2246          * optimize away the read when we have an oplock, and we're not
2247          * expecting to use any of the data we'd be reading in. That
2248          * is, when the page lies beyond the EOF, or straddles the EOF
2249          * and the write will cover all of the existing data.
2250          */
2251         if (CIFS_I(mapping->host)->clientCanCacheRead) {
2252                 i_size = i_size_read(mapping->host);
2253                 if (page_start >= i_size ||
2254                     (offset == 0 && (pos + len) >= i_size)) {
2255                         zero_user_segments(page, 0, offset,
2256                                            offset + len,
2257                                            PAGE_CACHE_SIZE);
2258                         /*
2259                          * PageChecked means that the parts of the page
2260                          * to which we're not writing are considered up
2261                          * to date. Once the data is copied to the
2262                          * page, it can be set uptodate.
2263                          */
2264                         SetPageChecked(page);
2265                         goto out;
2266                 }
2267         }
2268
2269         if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2270                 /*
2271                  * might as well read a page, it is fast enough. If we get
2272                  * an error, we don't need to return it. cifs_write_end will
2273                  * do a sync write instead since PG_uptodate isn't set.
2274                  */
2275                 cifs_readpage_worker(file, page, &page_start);
2276         } else {
2277                 /* we could try using another file handle if there is one -
2278                    but how would we lock it to prevent close of that handle
2279                    racing with this read? In any case
2280                    this will be written out by write_end so is fine */
2281         }
2282 out:
2283         *pagep = page;
2284         return rc;
2285 }
2286
2287 static int cifs_release_page(struct page *page, gfp_t gfp)
2288 {
2289         if (PagePrivate(page))
2290                 return 0;
2291
2292         return cifs_fscache_release_page(page, gfp);
2293 }
2294
2295 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2296 {
2297         struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2298
2299         if (offset == 0)
2300                 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2301 }
2302
2303 void cifs_oplock_break(struct work_struct *work)
2304 {
2305         struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2306                                                   oplock_break);
2307         struct inode *inode = cfile->pInode;
2308         struct cifsInodeInfo *cinode = CIFS_I(inode);
2309         int rc, waitrc = 0;
2310
2311         if (inode && S_ISREG(inode->i_mode)) {
2312                 if (cinode->clientCanCacheRead)
2313                         break_lease(inode, O_RDONLY);
2314                 else
2315                         break_lease(inode, O_WRONLY);
2316                 rc = filemap_fdatawrite(inode->i_mapping);
2317                 if (cinode->clientCanCacheRead == 0) {
2318                         waitrc = filemap_fdatawait(inode->i_mapping);
2319                         invalidate_remote_inode(inode);
2320                 }
2321                 if (!rc)
2322                         rc = waitrc;
2323                 if (rc)
2324                         cinode->write_behind_rc = rc;
2325                 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2326         }
2327
2328         /*
2329          * releasing stale oplock after recent reconnect of smb session using
2330          * a now incorrect file handle is not a data integrity issue but do
2331          * not bother sending an oplock release if session to server still is
2332          * disconnected since oplock already released by the server
2333          */
2334         if (!cfile->closePend && !cfile->oplock_break_cancelled) {
2335                 rc = CIFSSMBLock(0, cfile->tcon, cfile->netfid, 0, 0, 0, 0,
2336                                  LOCKING_ANDX_OPLOCK_RELEASE, false);
2337                 cFYI(1, "Oplock release rc = %d", rc);
2338         }
2339
2340         /*
2341          * We might have kicked in before is_valid_oplock_break()
2342          * finished grabbing reference for us.  Make sure it's done by
2343          * waiting for GlobalSMSSeslock.
2344          */
2345         write_lock(&GlobalSMBSeslock);
2346         write_unlock(&GlobalSMBSeslock);
2347
2348         cifs_oplock_break_put(cfile);
2349 }
2350
2351 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2352 {
2353         mntget(cfile->mnt);
2354         cifsFileInfo_get(cfile);
2355 }
2356
2357 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2358 {
2359         mntput(cfile->mnt);
2360         cifsFileInfo_put(cfile);
2361 }
2362
2363 const struct address_space_operations cifs_addr_ops = {
2364         .readpage = cifs_readpage,
2365         .readpages = cifs_readpages,
2366         .writepage = cifs_writepage,
2367         .writepages = cifs_writepages,
2368         .write_begin = cifs_write_begin,
2369         .write_end = cifs_write_end,
2370         .set_page_dirty = __set_page_dirty_nobuffers,
2371         .releasepage = cifs_release_page,
2372         .invalidatepage = cifs_invalidate_page,
2373         /* .sync_page = cifs_sync_page, */
2374         /* .direct_IO = */
2375 };
2376
2377 /*
2378  * cifs_readpages requires the server to support a buffer large enough to
2379  * contain the header plus one complete page of data.  Otherwise, we need
2380  * to leave cifs_readpages out of the address space operations.
2381  */
2382 const struct address_space_operations cifs_addr_ops_smallbuf = {
2383         .readpage = cifs_readpage,
2384         .writepage = cifs_writepage,
2385         .writepages = cifs_writepages,
2386         .write_begin = cifs_write_begin,
2387         .write_end = cifs_write_end,
2388         .set_page_dirty = __set_page_dirty_nobuffers,
2389         .releasepage = cifs_release_page,
2390         .invalidatepage = cifs_invalidate_page,
2391         /* .sync_page = cifs_sync_page, */
2392         /* .direct_IO = */
2393 };