fs/cifs/file.c

   1 /*
   2  *   fs/cifs/file.c
   3  *
   4  *   vfs operations that deal with files
   5  *
   6  *   Copyright (C) International Business Machines  Corp., 2002,2007
   7  *   Author(s): Steve French (sfrench@us.ibm.com)
   8  *              Jeremy Allison (jra@samba.org)
   9  *
  10  *   This library is free software; you can redistribute it and/or modify
  11  *   it under the terms of the GNU Lesser General Public License as published
  12  *   by the Free Software Foundation; either version 2.1 of the License, or
  13  *   (at your option) any later version.
  14  *
  15  *   This library is distributed in the hope that it will be useful,
  16  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
  18  *   the GNU Lesser General Public License for more details.
  19  *
  20  *   You should have received a copy of the GNU Lesser General Public License
  21  *   along with this library; if not, write to the Free Software
  22  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  23  */
  24 #include <linux/fs.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/stat.h>
  27 #include <linux/fcntl.h>
  28 #include <linux/pagemap.h>
  29 #include <linux/pagevec.h>
  30 #include <linux/writeback.h>
  31 #include <linux/task_io_accounting_ops.h>
  32 #include <linux/delay.h>
  33 #include <asm/div64.h>
  34 #include "cifsfs.h"
  35 #include "cifspdu.h"
  36 #include "cifsglob.h"
  37 #include "cifsproto.h"
  38 #include "cifs_unicode.h"
  39 #include "cifs_debug.h"
  40 #include "cifs_fs_sb.h"
  41
  42 static inline struct cifsFileInfo *cifs_init_private(
  43         struct cifsFileInfo *private_data, struct inode *inode,
  44         struct file *file, __u16 netfid)
  45 {
  46         memset(private_data, 0, sizeof(struct cifsFileInfo));
  47         private_data->netfid = netfid;
  48         private_data->pid = current->tgid;
  49         mutex_init(&private_data->fh_mutex);
  50         mutex_init(&private_data->lock_mutex);
  51         INIT_LIST_HEAD(&private_data->llist);
  52         private_data->pfile = file; /* needed for writepage */
  53         private_data->pInode = inode;
  54         private_data->invalidHandle = false;
  55         private_data->closePend = false;
  56         /* we have to track num writers to the inode, since writepages
  57         does not tell us which handle the write is for so there can
  58         be a close (overlapping with write) of the filehandle that
  59         cifs_writepages chose to use */
  60         atomic_set(&private_data->wrtPending, 0);
  61
  62         return private_data;
  63 }
  64
  65 static inline int cifs_convert_flags(unsigned int flags)
  66 {
  67         if ((flags & O_ACCMODE) == O_RDONLY)
  68                 return GENERIC_READ;
  69         else if ((flags & O_ACCMODE) == O_WRONLY)
  70                 return GENERIC_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                 return (GENERIC_READ | GENERIC_WRITE);
  76         }
  77
  78         return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
  79                 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
  80                 FILE_READ_DATA);
  81 }
  82
  83 static inline fmode_t cifs_posix_convert_flags(unsigned int flags)
  84 {
  85         fmode_t posix_flags = 0;
  86
  87         if ((flags & O_ACCMODE) == O_RDONLY)
  88                 posix_flags = FMODE_READ;
  89         else if ((flags & O_ACCMODE) == O_WRONLY)
  90                 posix_flags = FMODE_WRITE;
  91         else if ((flags & O_ACCMODE) == O_RDWR) {
  92                 /* GENERIC_ALL is too much permission to request
  93                    can cause unnecessary access denied on create */
  94                 /* return GENERIC_ALL; */
  95                 posix_flags = FMODE_READ | FMODE_WRITE;
  96         }
  97         /* can not map O_CREAT or O_EXCL or O_TRUNC flags when
  98            reopening a file.  They had their effect on the original open */
  99         if (flags & O_APPEND)
 100                 posix_flags |= (fmode_t)O_APPEND;
 101         if (flags & O_SYNC)
 102                 posix_flags |= (fmode_t)O_SYNC;
 103         if (flags & O_DIRECTORY)
 104                 posix_flags |= (fmode_t)O_DIRECTORY;
 105         if (flags & O_NOFOLLOW)
 106                 posix_flags |= (fmode_t)O_NOFOLLOW;
 107         if (flags & O_DIRECT)
 108                 posix_flags |= (fmode_t)O_DIRECT;
 109
 110         return posix_flags;
 111 }
 112
 113 static inline int cifs_get_disposition(unsigned int flags)
 114 {
 115         if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
 116                 return FILE_CREATE;
 117         else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
 118                 return FILE_OVERWRITE_IF;
 119         else if ((flags & O_CREAT) == O_CREAT)
 120                 return FILE_OPEN_IF;
 121         else if ((flags & O_TRUNC) == O_TRUNC)
 122                 return FILE_OVERWRITE;
 123         else
 124                 return FILE_OPEN;
 125 }
 126
 127 /* all arguments to this function must be checked for validity in caller */
 128 static inline int cifs_posix_open_inode_helper(struct inode *inode,
 129                         struct file *file, struct cifsInodeInfo *pCifsInode,
 130                         struct cifsFileInfo *pCifsFile, int oplock, u16 netfid)
 131 {
 132
 133         write_lock(&GlobalSMBSeslock);
 134
 135         pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
 136         if (pCifsInode == NULL) {
 137                 write_unlock(&GlobalSMBSeslock);
 138                 return -EINVAL;
 139         }
 140
 141         if (pCifsInode->clientCanCacheRead) {
 142                 /* we have the inode open somewhere else
 143                    no need to discard cache data */
 144                 goto psx_client_can_cache;
 145         }
 146
 147         /* BB FIXME need to fix this check to move it earlier into posix_open
 148            BB  fIX following section BB FIXME */
 149
 150         /* if not oplocked, invalidate inode pages if mtime or file
 151            size changed */
 152 /*      temp = cifs_NTtimeToUnix(le64_to_cpu(buf->LastWriteTime));
 153         if (timespec_equal(&file->f_path.dentry->d_inode->i_mtime, &temp) &&
 154                            (file->f_path.dentry->d_inode->i_size ==
 155                             (loff_t)le64_to_cpu(buf->EndOfFile))) {
 156                 cFYI(1, ("inode unchanged on server"));
 157         } else {
 158                 if (file->f_path.dentry->d_inode->i_mapping) {
 159                         rc = filemap_write_and_wait(file->f_path.dentry->d_inode->i_mapping);
 160                         if (rc != 0)
 161                                 CIFS_I(file->f_path.dentry->d_inode)->write_behind_rc = rc;
 162                 }
 163                 cFYI(1, ("invalidating remote inode since open detected it "
 164                          "changed"));
 165                 invalidate_remote_inode(file->f_path.dentry->d_inode);
 166         } */
 167
 168 psx_client_can_cache:
 169         if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
 170                 pCifsInode->clientCanCacheAll = true;
 171                 pCifsInode->clientCanCacheRead = true;
 172                 cFYI(1, ("Exclusive Oplock granted on inode %p",
 173                          file->f_path.dentry->d_inode));
 174         } else if ((oplock & 0xF) == OPLOCK_READ)
 175                 pCifsInode->clientCanCacheRead = true;
 176
 177         /* will have to change the unlock if we reenable the
 178            filemap_fdatawrite (which does not seem necessary */
 179         write_unlock(&GlobalSMBSeslock);
 180         return 0;
 181 }
 182
 183 static struct cifsFileInfo *
 184 cifs_fill_filedata(struct file *file)
 185 {
 186         struct list_head *tmp;
 187         struct cifsFileInfo *pCifsFile = NULL;
 188         struct cifsInodeInfo *pCifsInode = NULL;
 189
 190         /* search inode for this file and fill in file->private_data */
 191         pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
 192         read_lock(&GlobalSMBSeslock);
 193         list_for_each(tmp, &pCifsInode->openFileList) {
 194                 pCifsFile = list_entry(tmp, struct cifsFileInfo, flist);
 195                 if ((pCifsFile->pfile == NULL) &&
 196                     (pCifsFile->pid == current->tgid)) {
 197                         /* mode set in cifs_create */
 198
 199                         /* needed for writepage */
 200                         pCifsFile->pfile = file;
 201                         file->private_data = pCifsFile;
 202                         break;
 203                 }
 204         }
 205         read_unlock(&GlobalSMBSeslock);
 206
 207         if (file->private_data != NULL) {
 208                 return pCifsFile;
 209         } else if ((file->f_flags & O_CREAT) && (file->f_flags & O_EXCL))
 210                         cERROR(1, ("could not find file instance for "
 211                                    "new file %p", file));
 212         return NULL;
 213 }
 214
 215 /* all arguments to this function must be checked for validity in caller */
 216 static inline int cifs_open_inode_helper(struct inode *inode, struct file *file,
 217         struct cifsInodeInfo *pCifsInode, struct cifsFileInfo *pCifsFile,
 218         struct cifsTconInfo *pTcon, int *oplock, FILE_ALL_INFO *buf,
 219         char *full_path, int xid)
 220 {
 221         struct timespec temp;
 222         int rc;
 223
 224         /* want handles we can use to read with first
 225            in the list so we do not have to walk the
 226            list to search for one in write_begin */
 227         if ((file->f_flags & O_ACCMODE) == O_WRONLY) {
 228                 list_add_tail(&pCifsFile->flist,
 229                               &pCifsInode->openFileList);
 230         } else {
 231                 list_add(&pCifsFile->flist,
 232                          &pCifsInode->openFileList);
 233         }
 234         write_unlock(&GlobalSMBSeslock);
 235         if (pCifsInode->clientCanCacheRead) {
 236                 /* we have the inode open somewhere else
 237                    no need to discard cache data */
 238                 goto client_can_cache;
 239         }
 240
 241         /* BB need same check in cifs_create too? */
 242         /* if not oplocked, invalidate inode pages if mtime or file
 243            size changed */
 244         temp = cifs_NTtimeToUnix(buf->LastWriteTime);
 245         if (timespec_equal(&file->f_path.dentry->d_inode->i_mtime, &temp) &&
 246                            (file->f_path.dentry->d_inode->i_size ==
 247                             (loff_t)le64_to_cpu(buf->EndOfFile))) {
 248                 cFYI(1, ("inode unchanged on server"));
 249         } else {
 250                 if (file->f_path.dentry->d_inode->i_mapping) {
 251                 /* BB no need to lock inode until after invalidate
 252                    since namei code should already have it locked? */
 253                         rc = filemap_write_and_wait(file->f_path.dentry->d_inode->i_mapping);
 254                         if (rc != 0)
 255                                 CIFS_I(file->f_path.dentry->d_inode)->write_behind_rc = rc;
 256                 }
 257                 cFYI(1, ("invalidating remote inode since open detected it "
 258                          "changed"));
 259                 invalidate_remote_inode(file->f_path.dentry->d_inode);
 260         }
 261
 262 client_can_cache:
 263         if (pTcon->unix_ext)
 264                 rc = cifs_get_inode_info_unix(&file->f_path.dentry->d_inode,
 265                         full_path, inode->i_sb, xid);
 266         else
 267                 rc = cifs_get_inode_info(&file->f_path.dentry->d_inode,
 268                         full_path, buf, inode->i_sb, xid, NULL);
 269
 270         if ((*oplock & 0xF) == OPLOCK_EXCLUSIVE) {
 271                 pCifsInode->clientCanCacheAll = true;
 272                 pCifsInode->clientCanCacheRead = true;
 273                 cFYI(1, ("Exclusive Oplock granted on inode %p",
 274                          file->f_path.dentry->d_inode));
 275         } else if ((*oplock & 0xF) == OPLOCK_READ)
 276                 pCifsInode->clientCanCacheRead = true;
 277
 278         return rc;
 279 }
 280
 281 int cifs_open(struct inode *inode, struct file *file)
 282 {
 283         int rc = -EACCES;
 284         int xid, oplock;
 285         struct cifs_sb_info *cifs_sb;
 286         struct cifsTconInfo *tcon;
 287         struct cifsFileInfo *pCifsFile;
 288         struct cifsInodeInfo *pCifsInode;
 289         char *full_path = NULL;
 290         int desiredAccess;
 291         int disposition;
 292         __u16 netfid;
 293         FILE_ALL_INFO *buf = NULL;
 294
 295         xid = GetXid();
 296
 297         cifs_sb = CIFS_SB(inode->i_sb);
 298         tcon = cifs_sb->tcon;
 299
 300         pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
 301         pCifsFile = cifs_fill_filedata(file);
 302         if (pCifsFile) {
 303                 rc = 0;
 304                 FreeXid(xid);
 305                 return rc;
 306         }
 307
 308         full_path = build_path_from_dentry(file->f_path.dentry);
 309         if (full_path == NULL) {
 310                 rc = -ENOMEM;
 311                 FreeXid(xid);
 312                 return rc;
 313         }
 314
 315         cFYI(1, ("inode = 0x%p file flags are 0x%x for %s",
 316                  inode, file->f_flags, full_path));
 317
 318         if (oplockEnabled)
 319                 oplock = REQ_OPLOCK;
 320         else
 321                 oplock = 0;
 322
 323         if (!tcon->broken_posix_open && tcon->unix_ext &&
 324             (tcon->ses->capabilities & CAP_UNIX) &&
 325             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 326                         le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 327                 int oflags = (int) cifs_posix_convert_flags(file->f_flags);
 328                 /* can not refresh inode info since size could be stale */
 329                 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
 330                                      cifs_sb->mnt_file_mode /* ignored */,
 331                                      oflags, &oplock, &netfid, xid);
 332                 if (rc == 0) {
 333                         cFYI(1, ("posix open succeeded"));
 334                         /* no need for special case handling of setting mode
 335                            on read only files needed here */
 336
 337                         pCifsFile = cifs_fill_filedata(file);
 338                         cifs_posix_open_inode_helper(inode, file, pCifsInode,
 339                                                      pCifsFile, oplock, netfid);
 340                         goto out;
 341                 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
 342                         if (tcon->ses->serverNOS)
 343                                 cERROR(1, ("server %s of type %s returned"
 344                                            " unexpected error on SMB posix open"
 345                                            ", disabling posix open support."
 346                                            " Check if server update available.",
 347                                            tcon->ses->serverName,
 348                                            tcon->ses->serverNOS));
 349                         tcon->broken_posix_open = true;
 350                 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
 351                          (rc != -EOPNOTSUPP)) /* path not found or net err */
 352                         goto out;
 353                 /* else fallthrough to retry open the old way on network i/o
 354                    or DFS errors */
 355         }
 356
 357         desiredAccess = cifs_convert_flags(file->f_flags);
 358
 359 /*********************************************************************
 360  *  open flag mapping table:
 361  *
 362  *      POSIX Flag            CIFS Disposition
 363  *      ----------            ----------------
 364  *      O_CREAT               FILE_OPEN_IF
 365  *      O_CREAT | O_EXCL      FILE_CREATE
 366  *      O_CREAT | O_TRUNC     FILE_OVERWRITE_IF
 367  *      O_TRUNC               FILE_OVERWRITE
 368  *      none of the above     FILE_OPEN
 369  *
 370  *      Note that there is not a direct match between disposition
 371  *      FILE_SUPERSEDE (ie create whether or not file exists although
 372  *      O_CREAT | O_TRUNC is similar but truncates the existing
 373  *      file rather than creating a new file as FILE_SUPERSEDE does
 374  *      (which uses the attributes / metadata passed in on open call)
 375  *?
 376  *?  O_SYNC is a reasonable match to CIFS writethrough flag
 377  *?  and the read write flags match reasonably.  O_LARGEFILE
 378  *?  is irrelevant because largefile support is always used
 379  *?  by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
 380  *       O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
 381  *********************************************************************/
 382
 383         disposition = cifs_get_disposition(file->f_flags);
 384
 385         /* BB pass O_SYNC flag through on file attributes .. BB */
 386
 387         /* Also refresh inode by passing in file_info buf returned by SMBOpen
 388            and calling get_inode_info with returned buf (at least helps
 389            non-Unix server case) */
 390
 391         /* BB we can not do this if this is the second open of a file
 392            and the first handle has writebehind data, we might be
 393            able to simply do a filemap_fdatawrite/filemap_fdatawait first */
 394         buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
 395         if (!buf) {
 396                 rc = -ENOMEM;
 397                 goto out;
 398         }
 399
 400         if (cifs_sb->tcon->ses->capabilities & CAP_NT_SMBS)
 401                 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
 402                          desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
 403                          cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
 404                                  & CIFS_MOUNT_MAP_SPECIAL_CHR);
 405         else
 406                 rc = -EIO; /* no NT SMB support fall into legacy open below */
 407
 408         if (rc == -EIO) {
 409                 /* Old server, try legacy style OpenX */
 410                 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
 411                         desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
 412                         cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
 413                                 & CIFS_MOUNT_MAP_SPECIAL_CHR);
 414         }
 415         if (rc) {
 416                 cFYI(1, ("cifs_open returned 0x%x", rc));
 417                 goto out;
 418         }
 419         file->private_data =
 420                 kmalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
 421         if (file->private_data == NULL) {
 422                 rc = -ENOMEM;
 423                 goto out;
 424         }
 425         pCifsFile = cifs_init_private(file->private_data, inode, file, netfid);
 426         write_lock(&GlobalSMBSeslock);
 427         list_add(&pCifsFile->tlist, &tcon->openFileList);
 428
 429         pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
 430         if (pCifsInode) {
 431                 rc = cifs_open_inode_helper(inode, file, pCifsInode,
 432                                             pCifsFile, tcon,
 433                                             &oplock, buf, full_path, xid);
 434         } else {
 435                 write_unlock(&GlobalSMBSeslock);
 436         }
 437
 438         if (oplock & CIFS_CREATE_ACTION) {
 439                 /* time to set mode which we can not set earlier due to
 440                    problems creating new read-only files */
 441                 if (tcon->unix_ext) {
 442                         struct cifs_unix_set_info_args args = {
 443                                 .mode   = inode->i_mode,
 444                                 .uid    = NO_CHANGE_64,
 445                                 .gid    = NO_CHANGE_64,
 446                                 .ctime  = NO_CHANGE_64,
 447                                 .atime  = NO_CHANGE_64,
 448                                 .mtime  = NO_CHANGE_64,
 449                                 .device = 0,
 450                         };
 451                         CIFSSMBUnixSetPathInfo(xid, tcon, full_path, &args,
 452                                                cifs_sb->local_nls,
 453                                                cifs_sb->mnt_cifs_flags &
 454                                                 CIFS_MOUNT_MAP_SPECIAL_CHR);
 455                 }
 456         }
 457
 458 out:
 459         kfree(buf);
 460         kfree(full_path);
 461         FreeXid(xid);
 462         return rc;
 463 }
 464
 465 /* Try to reacquire byte range locks that were released when session */
 466 /* to server was lost */
 467 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
 468 {
 469         int rc = 0;
 470
 471 /* BB list all locks open on this file and relock */
 472
 473         return rc;
 474 }
 475
 476 static int cifs_reopen_file(struct file *file, bool can_flush)
 477 {
 478         int rc = -EACCES;
 479         int xid, oplock;
 480         struct cifs_sb_info *cifs_sb;
 481         struct cifsTconInfo *tcon;
 482         struct cifsFileInfo *pCifsFile;
 483         struct cifsInodeInfo *pCifsInode;
 484         struct inode *inode;
 485         char *full_path = NULL;
 486         int desiredAccess;
 487         int disposition = FILE_OPEN;
 488         __u16 netfid;
 489
 490         if (file->private_data)
 491                 pCifsFile = (struct cifsFileInfo *)file->private_data;
 492         else
 493                 return -EBADF;
 494
 495         xid = GetXid();
 496         mutex_lock(&pCifsFile->fh_mutex);
 497         if (!pCifsFile->invalidHandle) {
 498                 mutex_unlock(&pCifsFile->fh_mutex);
 499                 rc = 0;
 500                 FreeXid(xid);
 501                 return rc;
 502         }
 503
 504         if (file->f_path.dentry == NULL) {
 505                 cERROR(1, ("no valid name if dentry freed"));
 506                 dump_stack();
 507                 rc = -EBADF;
 508                 goto reopen_error_exit;
 509         }
 510
 511         inode = file->f_path.dentry->d_inode;
 512         if (inode == NULL) {
 513                 cERROR(1, ("inode not valid"));
 514                 dump_stack();
 515                 rc = -EBADF;
 516                 goto reopen_error_exit;
 517         }
 518
 519         cifs_sb = CIFS_SB(inode->i_sb);
 520         tcon = cifs_sb->tcon;
 521
 522 /* can not grab rename sem here because various ops, including
 523    those that already have the rename sem can end up causing writepage
 524    to get called and if the server was down that means we end up here,
 525    and we can never tell if the caller already has the rename_sem */
 526         full_path = build_path_from_dentry(file->f_path.dentry);
 527         if (full_path == NULL) {
 528                 rc = -ENOMEM;
 529 reopen_error_exit:
 530                 mutex_unlock(&pCifsFile->fh_mutex);
 531                 FreeXid(xid);
 532                 return rc;
 533         }
 534
 535         cFYI(1, ("inode = 0x%p file flags 0x%x for %s",
 536                  inode, file->f_flags, full_path));
 537
 538         if (oplockEnabled)
 539                 oplock = REQ_OPLOCK;
 540         else
 541                 oplock = 0;
 542
 543         if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
 544             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 545                         le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 546                 int oflags = (int) cifs_posix_convert_flags(file->f_flags);
 547                 /* can not refresh inode info since size could be stale */
 548                 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
 549                                      cifs_sb->mnt_file_mode /* ignored */,
 550                                      oflags, &oplock, &netfid, xid);
 551                 if (rc == 0) {
 552                         cFYI(1, ("posix reopen succeeded"));
 553                         goto reopen_success;
 554                 }
 555                 /* fallthrough to retry open the old way on errors, especially
 556                    in the reconnect path it is important to retry hard */
 557         }
 558
 559         desiredAccess = cifs_convert_flags(file->f_flags);
 560
 561         /* Can not refresh inode by passing in file_info buf to be returned
 562            by SMBOpen and then calling get_inode_info with returned buf
 563            since file might have write behind data that needs to be flushed
 564            and server version of file size can be stale. If we knew for sure
 565            that inode was not dirty locally we could do this */
 566
 567         rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
 568                          CREATE_NOT_DIR, &netfid, &oplock, NULL,
 569                          cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
 570                                 CIFS_MOUNT_MAP_SPECIAL_CHR);
 571         if (rc) {
 572                 mutex_unlock(&pCifsFile->fh_mutex);
 573                 cFYI(1, ("cifs_open returned 0x%x", rc));
 574                 cFYI(1, ("oplock: %d", oplock));
 575         } else {
 576 reopen_success:
 577                 pCifsFile->netfid = netfid;
 578                 pCifsFile->invalidHandle = false;
 579                 mutex_unlock(&pCifsFile->fh_mutex);
 580                 pCifsInode = CIFS_I(inode);
 581                 if (pCifsInode) {
 582                         if (can_flush) {
 583                                 rc = filemap_write_and_wait(inode->i_mapping);
 584                                 if (rc != 0)
 585                                         CIFS_I(inode)->write_behind_rc = rc;
 586                         /* temporarily disable caching while we
 587                            go to server to get inode info */
 588                                 pCifsInode->clientCanCacheAll = false;
 589                                 pCifsInode->clientCanCacheRead = false;
 590                                 if (tcon->unix_ext)
 591                                         rc = cifs_get_inode_info_unix(&inode,
 592                                                 full_path, inode->i_sb, xid);
 593                                 else
 594                                         rc = cifs_get_inode_info(&inode,
 595                                                 full_path, NULL, inode->i_sb,
 596                                                 xid, NULL);
 597                         } /* else we are writing out data to server already
 598                              and could deadlock if we tried to flush data, and
 599                              since we do not know if we have data that would
 600                              invalidate the current end of file on the server
 601                              we can not go to the server to get the new inod
 602                              info */
 603                         if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
 604                                 pCifsInode->clientCanCacheAll = true;
 605                                 pCifsInode->clientCanCacheRead = true;
 606                                 cFYI(1, ("Exclusive Oplock granted on inode %p",
 607                                          file->f_path.dentry->d_inode));
 608                         } else if ((oplock & 0xF) == OPLOCK_READ) {
 609                                 pCifsInode->clientCanCacheRead = true;
 610                                 pCifsInode->clientCanCacheAll = false;
 611                         } else {
 612                                 pCifsInode->clientCanCacheRead = false;
 613                                 pCifsInode->clientCanCacheAll = false;
 614                         }
 615                         cifs_relock_file(pCifsFile);
 616                 }
 617         }
 618         kfree(full_path);
 619         FreeXid(xid);
 620         return rc;
 621 }
 622
 623 int cifs_close(struct inode *inode, struct file *file)
 624 {
 625         int rc = 0;
 626         int xid, timeout;
 627         struct cifs_sb_info *cifs_sb;
 628         struct cifsTconInfo *pTcon;
 629         struct cifsFileInfo *pSMBFile =
 630                 (struct cifsFileInfo *)file->private_data;
 631
 632         xid = GetXid();
 633
 634         cifs_sb = CIFS_SB(inode->i_sb);
 635         pTcon = cifs_sb->tcon;
 636         if (pSMBFile) {
 637                 struct cifsLockInfo *li, *tmp;
 638                 write_lock(&GlobalSMBSeslock);
 639                 pSMBFile->closePend = true;
 640                 if (pTcon) {
 641                         /* no sense reconnecting to close a file that is
 642                            already closed */
 643                         if (!pTcon->need_reconnect) {
 644                                 write_unlock(&GlobalSMBSeslock);
 645                                 timeout = 2;
 646                                 while ((atomic_read(&pSMBFile->wrtPending) != 0)
 647                                         && (timeout <= 2048)) {
 648                                         /* Give write a better chance to get to
 649                                         server ahead of the close.  We do not
 650                                         want to add a wait_q here as it would
 651                                         increase the memory utilization as
 652                                         the struct would be in each open file,
 653                                         but this should give enough time to
 654                                         clear the socket */
 655                                         cFYI(DBG2,
 656                                                 ("close delay, write pending"));
 657                                         msleep(timeout);
 658                                         timeout *= 4;
 659                                 }
 660                                 if (atomic_read(&pSMBFile->wrtPending))
 661                                         cERROR(1, ("close with pending write"));
 662                                 if (!pTcon->need_reconnect &&
 663                                     !pSMBFile->invalidHandle)
 664                                         rc = CIFSSMBClose(xid, pTcon,
 665                                                   pSMBFile->netfid);
 666                         } else
 667                                 write_unlock(&GlobalSMBSeslock);
 668                 } else
 669                         write_unlock(&GlobalSMBSeslock);
 670
 671                 /* Delete any outstanding lock records.
 672                    We'll lose them when the file is closed anyway. */
 673                 mutex_lock(&pSMBFile->lock_mutex);
 674                 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
 675                         list_del(&li->llist);
 676                         kfree(li);
 677                 }
 678                 mutex_unlock(&pSMBFile->lock_mutex);
 679
 680                 write_lock(&GlobalSMBSeslock);
 681                 list_del(&pSMBFile->flist);
 682                 list_del(&pSMBFile->tlist);
 683                 write_unlock(&GlobalSMBSeslock);
 684                 timeout = 10;
 685                 /* We waited above to give the SMBWrite a chance to issue
 686                    on the wire (so we do not get SMBWrite returning EBADF
 687                    if writepages is racing with close.  Note that writepages
 688                    does not specify a file handle, so it is possible for a file
 689                    to be opened twice, and the application close the "wrong"
 690                    file handle - in these cases we delay long enough to allow
 691                    the SMBWrite to get on the wire before the SMB Close.
 692                    We allow total wait here over 45 seconds, more than
 693                    oplock break time, and more than enough to allow any write
 694                    to complete on the server, or to time out on the client */
 695                 while ((atomic_read(&pSMBFile->wrtPending) != 0)
 696                                 && (timeout <= 50000)) {
 697                         cERROR(1, ("writes pending, delay free of handle"));
 698                         msleep(timeout);
 699                         timeout *= 8;
 700                 }
 701                 kfree(file->private_data);
 702                 file->private_data = NULL;
 703         } else
 704                 rc = -EBADF;
 705
 706         read_lock(&GlobalSMBSeslock);
 707         if (list_empty(&(CIFS_I(inode)->openFileList))) {
 708                 cFYI(1, ("closing last open instance for inode %p", inode));
 709                 /* if the file is not open we do not know if we can cache info
 710                    on this inode, much less write behind and read ahead */
 711                 CIFS_I(inode)->clientCanCacheRead = false;
 712                 CIFS_I(inode)->clientCanCacheAll  = false;
 713         }
 714         read_unlock(&GlobalSMBSeslock);
 715         if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
 716                 rc = CIFS_I(inode)->write_behind_rc;
 717         FreeXid(xid);
 718         return rc;
 719 }
 720
 721 int cifs_closedir(struct inode *inode, struct file *file)
 722 {
 723         int rc = 0;
 724         int xid;
 725         struct cifsFileInfo *pCFileStruct =
 726             (struct cifsFileInfo *)file->private_data;
 727         char *ptmp;
 728
 729         cFYI(1, ("Closedir inode = 0x%p", inode));
 730
 731         xid = GetXid();
 732
 733         if (pCFileStruct) {
 734                 struct cifsTconInfo *pTcon;
 735                 struct cifs_sb_info *cifs_sb =
 736                         CIFS_SB(file->f_path.dentry->d_sb);
 737
 738                 pTcon = cifs_sb->tcon;
 739
 740                 cFYI(1, ("Freeing private data in close dir"));
 741                 write_lock(&GlobalSMBSeslock);
 742                 if (!pCFileStruct->srch_inf.endOfSearch &&
 743                     !pCFileStruct->invalidHandle) {
 744                         pCFileStruct->invalidHandle = true;
 745                         write_unlock(&GlobalSMBSeslock);
 746                         rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
 747                         cFYI(1, ("Closing uncompleted readdir with rc %d",
 748                                  rc));
 749                         /* not much we can do if it fails anyway, ignore rc */
 750                         rc = 0;
 751                 } else
 752                         write_unlock(&GlobalSMBSeslock);
 753                 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
 754                 if (ptmp) {
 755                         cFYI(1, ("closedir free smb buf in srch struct"));
 756                         pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
 757                         if (pCFileStruct->srch_inf.smallBuf)
 758                                 cifs_small_buf_release(ptmp);
 759                         else
 760                                 cifs_buf_release(ptmp);
 761                 }
 762                 kfree(file->private_data);
 763                 file->private_data = NULL;
 764         }
 765         /* BB can we lock the filestruct while this is going on? */
 766         FreeXid(xid);
 767         return rc;
 768 }
 769
 770 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
 771                                 __u64 offset, __u8 lockType)
 772 {
 773         struct cifsLockInfo *li =
 774                 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
 775         if (li == NULL)
 776                 return -ENOMEM;
 777         li->offset = offset;
 778         li->length = len;
 779         li->type = lockType;
 780         mutex_lock(&fid->lock_mutex);
 781         list_add(&li->llist, &fid->llist);
 782         mutex_unlock(&fid->lock_mutex);
 783         return 0;
 784 }
 785
 786 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
 787 {
 788         int rc, xid;
 789         __u32 numLock = 0;
 790         __u32 numUnlock = 0;
 791         __u64 length;
 792         bool wait_flag = false;
 793         struct cifs_sb_info *cifs_sb;
 794         struct cifsTconInfo *tcon;
 795         __u16 netfid;
 796         __u8 lockType = LOCKING_ANDX_LARGE_FILES;
 797         bool posix_locking = 0;
 798
 799         length = 1 + pfLock->fl_end - pfLock->fl_start;
 800         rc = -EACCES;
 801         xid = GetXid();
 802
 803         cFYI(1, ("Lock parm: 0x%x flockflags: "
 804                  "0x%x flocktype: 0x%x start: %lld end: %lld",
 805                 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
 806                 pfLock->fl_end));
 807
 808         if (pfLock->fl_flags & FL_POSIX)
 809                 cFYI(1, ("Posix"));
 810         if (pfLock->fl_flags & FL_FLOCK)
 811                 cFYI(1, ("Flock"));
 812         if (pfLock->fl_flags & FL_SLEEP) {
 813                 cFYI(1, ("Blocking lock"));
 814                 wait_flag = true;
 815         }
 816         if (pfLock->fl_flags & FL_ACCESS)
 817                 cFYI(1, ("Process suspended by mandatory locking - "
 818                          "not implemented yet"));
 819         if (pfLock->fl_flags & FL_LEASE)
 820                 cFYI(1, ("Lease on file - not implemented yet"));
 821         if (pfLock->fl_flags &
 822             (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
 823                 cFYI(1, ("Unknown lock flags 0x%x", pfLock->fl_flags));
 824
 825         if (pfLock->fl_type == F_WRLCK) {
 826                 cFYI(1, ("F_WRLCK "));
 827                 numLock = 1;
 828         } else if (pfLock->fl_type == F_UNLCK) {
 829                 cFYI(1, ("F_UNLCK"));
 830                 numUnlock = 1;
 831                 /* Check if unlock includes more than
 832                 one lock range */
 833         } else if (pfLock->fl_type == F_RDLCK) {
 834                 cFYI(1, ("F_RDLCK"));
 835                 lockType |= LOCKING_ANDX_SHARED_LOCK;
 836                 numLock = 1;
 837         } else if (pfLock->fl_type == F_EXLCK) {
 838                 cFYI(1, ("F_EXLCK"));
 839                 numLock = 1;
 840         } else if (pfLock->fl_type == F_SHLCK) {
 841                 cFYI(1, ("F_SHLCK"));
 842                 lockType |= LOCKING_ANDX_SHARED_LOCK;
 843                 numLock = 1;
 844         } else
 845                 cFYI(1, ("Unknown type of lock"));
 846
 847         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
 848         tcon = cifs_sb->tcon;
 849
 850         if (file->private_data == NULL) {
 851                 rc = -EBADF;
 852                 FreeXid(xid);
 853                 return rc;
 854         }
 855         netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
 856
 857         if ((tcon->ses->capabilities & CAP_UNIX) &&
 858             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
 859             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
 860                 posix_locking = 1;
 861         /* BB add code here to normalize offset and length to
 862         account for negative length which we can not accept over the
 863         wire */
 864         if (IS_GETLK(cmd)) {
 865                 if (posix_locking) {
 866                         int posix_lock_type;
 867                         if (lockType & LOCKING_ANDX_SHARED_LOCK)
 868                                 posix_lock_type = CIFS_RDLCK;
 869                         else
 870                                 posix_lock_type = CIFS_WRLCK;
 871                         rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
 872                                         length, pfLock,
 873                                         posix_lock_type, wait_flag);
 874                         FreeXid(xid);
 875                         return rc;
 876                 }
 877
 878                 /* BB we could chain these into one lock request BB */
 879                 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
 880                                  0, 1, lockType, 0 /* wait flag */ );
 881                 if (rc == 0) {
 882                         rc = CIFSSMBLock(xid, tcon, netfid, length,
 883                                          pfLock->fl_start, 1 /* numUnlock */ ,
 884                                          0 /* numLock */ , lockType,
 885                                          0 /* wait flag */ );
 886                         pfLock->fl_type = F_UNLCK;
 887                         if (rc != 0)
 888                                 cERROR(1, ("Error unlocking previously locked "
 889                                            "range %d during test of lock", rc));
 890                         rc = 0;
 891
 892                 } else {
 893                         /* if rc == ERR_SHARING_VIOLATION ? */
 894                         rc = 0; /* do not change lock type to unlock
 895                                    since range in use */
 896                 }
 897
 898                 FreeXid(xid);
 899                 return rc;
 900         }
 901
 902         if (!numLock && !numUnlock) {
 903                 /* if no lock or unlock then nothing
 904                 to do since we do not know what it is */
 905                 FreeXid(xid);
 906                 return -EOPNOTSUPP;
 907         }
 908
 909         if (posix_locking) {
 910                 int posix_lock_type;
 911                 if (lockType & LOCKING_ANDX_SHARED_LOCK)
 912                         posix_lock_type = CIFS_RDLCK;
 913                 else
 914                         posix_lock_type = CIFS_WRLCK;
 915
 916                 if (numUnlock == 1)
 917                         posix_lock_type = CIFS_UNLCK;
 918
 919                 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
 920                                       length, pfLock,
 921                                       posix_lock_type, wait_flag);
 922         } else {
 923                 struct cifsFileInfo *fid =
 924                         (struct cifsFileInfo *)file->private_data;
 925
 926                 if (numLock) {
 927                         rc = CIFSSMBLock(xid, tcon, netfid, length,
 928                                         pfLock->fl_start,
 929                                         0, numLock, lockType, wait_flag);
 930
 931                         if (rc == 0) {
 932                                 /* For Windows locks we must store them. */
 933                                 rc = store_file_lock(fid, length,
 934                                                 pfLock->fl_start, lockType);
 935                         }
 936                 } else if (numUnlock) {
 937                         /* For each stored lock that this unlock overlaps
 938                            completely, unlock it. */
 939                         int stored_rc = 0;
 940                         struct cifsLockInfo *li, *tmp;
 941
 942                         rc = 0;
 943                         mutex_lock(&fid->lock_mutex);
 944                         list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
 945                                 if (pfLock->fl_start <= li->offset &&
 946                                                 (pfLock->fl_start + length) >=
 947                                                 (li->offset + li->length)) {
 948                                         stored_rc = CIFSSMBLock(xid, tcon,
 949                                                         netfid,
 950                                                         li->length, li->offset,
 951                                                         1, 0, li->type, false);
 952                                         if (stored_rc)
 953                                                 rc = stored_rc;
 954
 955                                         list_del(&li->llist);
 956                                         kfree(li);
 957                                 }
 958                         }
 959                         mutex_unlock(&fid->lock_mutex);
 960                 }
 961         }
 962
 963         if (pfLock->fl_flags & FL_POSIX)
 964                 posix_lock_file_wait(file, pfLock);
 965         FreeXid(xid);
 966         return rc;
 967 }
 968
 969 /*
 970  * Set the timeout on write requests past EOF. For some servers (Windows)
 971  * these calls can be very long.
 972  *
 973  * If we're writing >10M past the EOF we give a 180s timeout. Anything less
 974  * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
 975  * The 10M cutoff is totally arbitrary. A better scheme for this would be
 976  * welcome if someone wants to suggest one.
 977  *
 978  * We may be able to do a better job with this if there were some way to
 979  * declare that a file should be sparse.
 980  */
 981 static int
 982 cifs_write_timeout(struct cifsInodeInfo *cifsi, loff_t offset)
 983 {
 984         if (offset <= cifsi->server_eof)
 985                 return CIFS_STD_OP;
 986         else if (offset > (cifsi->server_eof + (10 * 1024 * 1024)))
 987                 return CIFS_VLONG_OP;
 988         else
 989                 return CIFS_LONG_OP;
 990 }
 991
 992 /* update the file size (if needed) after a write */
 993 static void
 994 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
 995                       unsigned int bytes_written)
 996 {
 997         loff_t end_of_write = offset + bytes_written;
 998
 999         if (end_of_write > cifsi->server_eof)
1000                 cifsi->server_eof = end_of_write;
1001 }
1002
1003 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
1004         size_t write_size, loff_t *poffset)
1005 {
1006         int rc = 0;
1007         unsigned int bytes_written = 0;
1008         unsigned int total_written;
1009         struct cifs_sb_info *cifs_sb;
1010         struct cifsTconInfo *pTcon;
1011         int xid, long_op;
1012         struct cifsFileInfo *open_file;
1013         struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
1014
1015         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1016
1017         pTcon = cifs_sb->tcon;
1018
1019         /* cFYI(1,
1020            (" write %d bytes to offset %lld of %s", write_size,
1021            *poffset, file->f_path.dentry->d_name.name)); */
1022
1023         if (file->private_data == NULL)
1024                 return -EBADF;
1025         open_file = (struct cifsFileInfo *) file->private_data;
1026
1027         rc = generic_write_checks(file, poffset, &write_size, 0);
1028         if (rc)
1029                 return rc;
1030
1031         xid = GetXid();
1032
1033         long_op = cifs_write_timeout(cifsi, *poffset);
1034         for (total_written = 0; write_size > total_written;
1035              total_written += bytes_written) {
1036                 rc = -EAGAIN;
1037                 while (rc == -EAGAIN) {
1038                         if (file->private_data == NULL) {
1039                                 /* file has been closed on us */
1040                                 FreeXid(xid);
1041                         /* if we have gotten here we have written some data
1042                            and blocked, and the file has been freed on us while
1043                            we blocked so return what we managed to write */
1044                                 return total_written;
1045                         }
1046                         if (open_file->closePend) {
1047                                 FreeXid(xid);
1048                                 if (total_written)
1049                                         return total_written;
1050                                 else
1051                                         return -EBADF;
1052                         }
1053                         if (open_file->invalidHandle) {
1054                                 /* we could deadlock if we called
1055                                    filemap_fdatawait from here so tell
1056                                    reopen_file not to flush data to server
1057                                    now */
1058                                 rc = cifs_reopen_file(file, false);
1059                                 if (rc != 0)
1060                                         break;
1061                         }
1062
1063                         rc = CIFSSMBWrite(xid, pTcon,
1064                                 open_file->netfid,
1065                                 min_t(const int, cifs_sb->wsize,
1066                                       write_size - total_written),
1067                                 *poffset, &bytes_written,
1068                                 NULL, write_data + total_written, long_op);
1069                 }
1070                 if (rc || (bytes_written == 0)) {
1071                         if (total_written)
1072                                 break;
1073                         else {
1074                                 FreeXid(xid);
1075                                 return rc;
1076                         }
1077                 } else {
1078                         cifs_update_eof(cifsi, *poffset, bytes_written);
1079                         *poffset += bytes_written;
1080                 }
1081                 long_op = CIFS_STD_OP; /* subsequent writes fast -
1082                                     15 seconds is plenty */
1083         }
1084
1085         cifs_stats_bytes_written(pTcon, total_written);
1086
1087         /* since the write may have blocked check these pointers again */
1088         if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1089                 struct inode *inode = file->f_path.dentry->d_inode;
1090 /* Do not update local mtime - server will set its actual value on write
1091  *              inode->i_ctime = inode->i_mtime =
1092  *                      current_fs_time(inode->i_sb);*/
1093                 if (total_written > 0) {
1094                         spin_lock(&inode->i_lock);
1095                         if (*poffset > file->f_path.dentry->d_inode->i_size)
1096                                 i_size_write(file->f_path.dentry->d_inode,
1097                                         *poffset);
1098                         spin_unlock(&inode->i_lock);
1099                 }
1100                 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1101         }
1102         FreeXid(xid);
1103         return total_written;
1104 }
1105
1106 static ssize_t cifs_write(struct file *file, const char *write_data,
1107                           size_t write_size, loff_t *poffset)
1108 {
1109         int rc = 0;
1110         unsigned int bytes_written = 0;
1111         unsigned int total_written;
1112         struct cifs_sb_info *cifs_sb;
1113         struct cifsTconInfo *pTcon;
1114         int xid, long_op;
1115         struct cifsFileInfo *open_file;
1116         struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
1117
1118         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1119
1120         pTcon = cifs_sb->tcon;
1121
1122         cFYI(1, ("write %zd bytes to offset %lld of %s", write_size,
1123            *poffset, file->f_path.dentry->d_name.name));
1124
1125         if (file->private_data == NULL)
1126                 return -EBADF;
1127         open_file = (struct cifsFileInfo *)file->private_data;
1128
1129         xid = GetXid();
1130
1131         long_op = cifs_write_timeout(cifsi, *poffset);
1132         for (total_written = 0; write_size > total_written;
1133              total_written += bytes_written) {
1134                 rc = -EAGAIN;
1135                 while (rc == -EAGAIN) {
1136                         if (file->private_data == NULL) {
1137                                 /* file has been closed on us */
1138                                 FreeXid(xid);
1139                         /* if we have gotten here we have written some data
1140                            and blocked, and the file has been freed on us
1141                            while we blocked so return what we managed to
1142                            write */
1143                                 return total_written;
1144                         }
1145                         if (open_file->closePend) {
1146                                 FreeXid(xid);
1147                                 if (total_written)
1148                                         return total_written;
1149                                 else
1150                                         return -EBADF;
1151                         }
1152                         if (open_file->invalidHandle) {
1153                                 /* we could deadlock if we called
1154                                    filemap_fdatawait from here so tell
1155                                    reopen_file not to flush data to
1156                                    server now */
1157                                 rc = cifs_reopen_file(file, false);
1158                                 if (rc != 0)
1159                                         break;
1160                         }
1161                         if (experimEnabled || (pTcon->ses->server &&
1162                                 ((pTcon->ses->server->secMode &
1163                                 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1164                                 == 0))) {
1165                                 struct kvec iov[2];
1166                                 unsigned int len;
1167
1168                                 len = min((size_t)cifs_sb->wsize,
1169                                           write_size - total_written);
1170                                 /* iov[0] is reserved for smb header */
1171                                 iov[1].iov_base = (char *)write_data +
1172                                                   total_written;
1173                                 iov[1].iov_len = len;
1174                                 rc = CIFSSMBWrite2(xid, pTcon,
1175                                                 open_file->netfid, len,
1176                                                 *poffset, &bytes_written,
1177                                                 iov, 1, long_op);
1178                         } else
1179                                 rc = CIFSSMBWrite(xid, pTcon,
1180                                          open_file->netfid,
1181                                          min_t(const int, cifs_sb->wsize,
1182                                                write_size - total_written),
1183                                          *poffset, &bytes_written,
1184                                          write_data + total_written,
1185                                          NULL, long_op);
1186                 }
1187                 if (rc || (bytes_written == 0)) {
1188                         if (total_written)
1189                                 break;
1190                         else {
1191                                 FreeXid(xid);
1192                                 return rc;
1193                         }
1194                 } else {
1195                         cifs_update_eof(cifsi, *poffset, bytes_written);
1196                         *poffset += bytes_written;
1197                 }
1198                 long_op = CIFS_STD_OP; /* subsequent writes fast -
1199                                     15 seconds is plenty */
1200         }
1201
1202         cifs_stats_bytes_written(pTcon, total_written);
1203
1204         /* since the write may have blocked check these pointers again */
1205         if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1206 /*BB We could make this contingent on superblock ATIME flag too */
1207 /*              file->f_path.dentry->d_inode->i_ctime =
1208                 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
1209                 if (total_written > 0) {
1210                         spin_lock(&file->f_path.dentry->d_inode->i_lock);
1211                         if (*poffset > file->f_path.dentry->d_inode->i_size)
1212                                 i_size_write(file->f_path.dentry->d_inode,
1213                                              *poffset);
1214                         spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1215                 }
1216                 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1217         }
1218         FreeXid(xid);
1219         return total_written;
1220 }
1221
1222 #ifdef CONFIG_CIFS_EXPERIMENTAL
1223 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode)
1224 {
1225         struct cifsFileInfo *open_file = NULL;
1226
1227         read_lock(&GlobalSMBSeslock);
1228         /* we could simply get the first_list_entry since write-only entries
1229            are always at the end of the list but since the first entry might
1230            have a close pending, we go through the whole list */
1231         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1232                 if (open_file->closePend)
1233                         continue;
1234                 if (open_file->pfile && ((open_file->pfile->f_flags & O_RDWR) ||
1235                     (open_file->pfile->f_flags & O_RDONLY))) {
1236                         if (!open_file->invalidHandle) {
1237                                 /* found a good file */
1238                                 /* lock it so it will not be closed on us */
1239                                 atomic_inc(&open_file->wrtPending);
1240                                 read_unlock(&GlobalSMBSeslock);
1241                                 return open_file;
1242                         } /* else might as well continue, and look for
1243                              another, or simply have the caller reopen it
1244                              again rather than trying to fix this handle */
1245                 } else /* write only file */
1246                         break; /* write only files are last so must be done */
1247         }
1248         read_unlock(&GlobalSMBSeslock);
1249         return NULL;
1250 }
1251 #endif
1252
1253 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode)
1254 {
1255         struct cifsFileInfo *open_file;
1256         bool any_available = false;
1257         int rc;
1258
1259         /* Having a null inode here (because mapping->host was set to zero by
1260         the VFS or MM) should not happen but we had reports of on oops (due to
1261         it being zero) during stress testcases so we need to check for it */
1262
1263         if (cifs_inode == NULL) {
1264                 cERROR(1, ("Null inode passed to cifs_writeable_file"));
1265                 dump_stack();
1266                 return NULL;
1267         }
1268
1269         read_lock(&GlobalSMBSeslock);
1270 refind_writable:
1271         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1272                 if (open_file->closePend ||
1273                     (!any_available && open_file->pid != current->tgid))
1274                         continue;
1275
1276                 if (open_file->pfile &&
1277                     ((open_file->pfile->f_flags & O_RDWR) ||
1278                      (open_file->pfile->f_flags & O_WRONLY))) {
1279                         atomic_inc(&open_file->wrtPending);
1280
1281                         if (!open_file->invalidHandle) {
1282                                 /* found a good writable file */
1283                                 read_unlock(&GlobalSMBSeslock);
1284                                 return open_file;
1285                         }
1286
1287                         read_unlock(&GlobalSMBSeslock);
1288                         /* Had to unlock since following call can block */
1289                         rc = cifs_reopen_file(open_file->pfile, false);
1290                         if (!rc) {
1291                                 if (!open_file->closePend)
1292                                         return open_file;
1293                                 else { /* start over in case this was deleted */
1294                                        /* since the list could be modified */
1295                                         read_lock(&GlobalSMBSeslock);
1296                                         atomic_dec(&open_file->wrtPending);
1297                                         goto refind_writable;
1298                                 }
1299                         }
1300
1301                         /* if it fails, try another handle if possible -
1302                         (we can not do this if closePending since
1303                         loop could be modified - in which case we
1304                         have to start at the beginning of the list
1305                         again. Note that it would be bad
1306                         to hold up writepages here (rather than
1307                         in caller) with continuous retries */
1308                         cFYI(1, ("wp failed on reopen file"));
1309                         read_lock(&GlobalSMBSeslock);
1310                         /* can not use this handle, no write
1311                            pending on this one after all */
1312                         atomic_dec(&open_file->wrtPending);
1313
1314                         if (open_file->closePend) /* list could have changed */
1315                                 goto refind_writable;
1316                         /* else we simply continue to the next entry. Thus
1317                            we do not loop on reopen errors.  If we
1318                            can not reopen the file, for example if we
1319                            reconnected to a server with another client
1320                            racing to delete or lock the file we would not
1321                            make progress if we restarted before the beginning
1322                            of the loop here. */
1323                 }
1324         }
1325         /* couldn't find useable FH with same pid, try any available */
1326         if (!any_available) {
1327                 any_available = true;
1328                 goto refind_writable;
1329         }
1330         read_unlock(&GlobalSMBSeslock);
1331         return NULL;
1332 }
1333
1334 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1335 {
1336         struct address_space *mapping = page->mapping;
1337         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1338         char *write_data;
1339         int rc = -EFAULT;
1340         int bytes_written = 0;
1341         struct cifs_sb_info *cifs_sb;
1342         struct cifsTconInfo *pTcon;
1343         struct inode *inode;
1344         struct cifsFileInfo *open_file;
1345
1346         if (!mapping || !mapping->host)
1347                 return -EFAULT;
1348
1349         inode = page->mapping->host;
1350         cifs_sb = CIFS_SB(inode->i_sb);
1351         pTcon = cifs_sb->tcon;
1352
1353         offset += (loff_t)from;
1354         write_data = kmap(page);
1355         write_data += from;
1356
1357         if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1358                 kunmap(page);
1359                 return -EIO;
1360         }
1361
1362         /* racing with truncate? */
1363         if (offset > mapping->host->i_size) {
1364                 kunmap(page);
1365                 return 0; /* don't care */
1366         }
1367
1368         /* check to make sure that we are not extending the file */
1369         if (mapping->host->i_size - offset < (loff_t)to)
1370                 to = (unsigned)(mapping->host->i_size - offset);
1371
1372         open_file = find_writable_file(CIFS_I(mapping->host));
1373         if (open_file) {
1374                 bytes_written = cifs_write(open_file->pfile, write_data,
1375                                            to-from, &offset);
1376                 atomic_dec(&open_file->wrtPending);
1377                 /* Does mm or vfs already set times? */
1378                 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1379                 if ((bytes_written > 0) && (offset))
1380                         rc = 0;
1381                 else if (bytes_written < 0)
1382                         rc = bytes_written;
1383         } else {
1384                 cFYI(1, ("No writeable filehandles for inode"));
1385                 rc = -EIO;
1386         }
1387
1388         kunmap(page);
1389         return rc;
1390 }
1391
1392 static int cifs_writepages(struct address_space *mapping,
1393                            struct writeback_control *wbc)
1394 {
1395         struct backing_dev_info *bdi = mapping->backing_dev_info;
1396         unsigned int bytes_to_write;
1397         unsigned int bytes_written;
1398         struct cifs_sb_info *cifs_sb;
1399         int done = 0;
1400         pgoff_t end;
1401         pgoff_t index;
1402         int range_whole = 0;
1403         struct kvec *iov;
1404         int len;
1405         int n_iov = 0;
1406         pgoff_t next;
1407         int nr_pages;
1408         __u64 offset = 0;
1409         struct cifsFileInfo *open_file;
1410         struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1411         struct page *page;
1412         struct pagevec pvec;
1413         int rc = 0;
1414         int scanned = 0;
1415         int xid, long_op;
1416
1417         cifs_sb = CIFS_SB(mapping->host->i_sb);
1418
1419         /*
1420          * If wsize is smaller that the page cache size, default to writing
1421          * one page at a time via cifs_writepage
1422          */
1423         if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1424                 return generic_writepages(mapping, wbc);
1425
1426         if ((cifs_sb->tcon->ses) && (cifs_sb->tcon->ses->server))
1427                 if (cifs_sb->tcon->ses->server->secMode &
1428                                 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1429                         if (!experimEnabled)
1430                                 return generic_writepages(mapping, wbc);
1431
1432         iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1433         if (iov == NULL)
1434                 return generic_writepages(mapping, wbc);
1435
1436
1437         /*
1438          * BB: Is this meaningful for a non-block-device file system?
1439          * If it is, we should test it again after we do I/O
1440          */
1441         if (wbc->nonblocking && bdi_write_congested(bdi)) {
1442                 wbc->encountered_congestion = 1;
1443                 kfree(iov);
1444                 return 0;
1445         }
1446
1447         xid = GetXid();
1448
1449         pagevec_init(&pvec, 0);
1450         if (wbc->range_cyclic) {
1451                 index = mapping->writeback_index; /* Start from prev offset */
1452                 end = -1;
1453         } else {
1454                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1455                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1456                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1457                         range_whole = 1;
1458                 scanned = 1;
1459         }
1460 retry:
1461         while (!done && (index <= end) &&
1462                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1463                         PAGECACHE_TAG_DIRTY,
1464                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1465                 int first;
1466                 unsigned int i;
1467
1468                 first = -1;
1469                 next = 0;
1470                 n_iov = 0;
1471                 bytes_to_write = 0;
1472
1473                 for (i = 0; i < nr_pages; i++) {
1474                         page = pvec.pages[i];
1475                         /*
1476                          * At this point we hold neither mapping->tree_lock nor
1477                          * lock on the page itself: the page may be truncated or
1478                          * invalidated (changing page->mapping to NULL), or even
1479                          * swizzled back from swapper_space to tmpfs file
1480                          * mapping
1481                          */
1482
1483                         if (first < 0)
1484                                 lock_page(page);
1485                         else if (!trylock_page(page))
1486                                 break;
1487
1488                         if (unlikely(page->mapping != mapping)) {
1489                                 unlock_page(page);
1490                                 break;
1491                         }
1492
1493                         if (!wbc->range_cyclic && page->index > end) {
1494                                 done = 1;
1495                                 unlock_page(page);
1496                                 break;
1497                         }
1498
1499                         if (next && (page->index != next)) {
1500                                 /* Not next consecutive page */
1501                                 unlock_page(page);
1502                                 break;
1503                         }
1504
1505                         if (wbc->sync_mode != WB_SYNC_NONE)
1506                                 wait_on_page_writeback(page);
1507
1508                         if (PageWriteback(page) ||
1509                                         !clear_page_dirty_for_io(page)) {
1510                                 unlock_page(page);
1511                                 break;
1512                         }
1513
1514                         /*
1515                          * This actually clears the dirty bit in the radix tree.
1516                          * See cifs_writepage() for more commentary.
1517                          */
1518                         set_page_writeback(page);
1519
1520                         if (page_offset(page) >= mapping->host->i_size) {
1521                                 done = 1;
1522                                 unlock_page(page);
1523                                 end_page_writeback(page);
1524                                 break;
1525                         }
1526
1527                         /*
1528                          * BB can we get rid of this?  pages are held by pvec
1529                          */
1530                         page_cache_get(page);
1531
1532                         len = min(mapping->host->i_size - page_offset(page),
1533                                   (loff_t)PAGE_CACHE_SIZE);
1534
1535                         /* reserve iov[0] for the smb header */
1536                         n_iov++;
1537                         iov[n_iov].iov_base = kmap(page);
1538                         iov[n_iov].iov_len = len;
1539                         bytes_to_write += len;
1540
1541                         if (first < 0) {
1542                                 first = i;
1543                                 offset = page_offset(page);
1544                         }
1545                         next = page->index + 1;
1546                         if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1547                                 break;
1548                 }
1549                 if (n_iov) {
1550                         /* Search for a writable handle every time we call
1551                          * CIFSSMBWrite2.  We can't rely on the last handle
1552                          * we used to still be valid
1553                          */
1554                         open_file = find_writable_file(CIFS_I(mapping->host));
1555                         if (!open_file) {
1556                                 cERROR(1, ("No writable handles for inode"));
1557                                 rc = -EBADF;
1558                         } else {
1559                                 long_op = cifs_write_timeout(cifsi, offset);
1560                                 rc = CIFSSMBWrite2(xid, cifs_sb->tcon,
1561                                                    open_file->netfid,
1562                                                    bytes_to_write, offset,
1563                                                    &bytes_written, iov, n_iov,
1564                                                    long_op);
1565                                 atomic_dec(&open_file->wrtPending);
1566                                 cifs_update_eof(cifsi, offset, bytes_written);
1567
1568                                 if (rc || bytes_written < bytes_to_write) {
1569                                         cERROR(1, ("Write2 ret %d, wrote %d",
1570                                                   rc, bytes_written));
1571                                         /* BB what if continued retry is
1572                                            requested via mount flags? */
1573                                         if (rc == -ENOSPC)
1574                                                 set_bit(AS_ENOSPC, &mapping->flags);
1575                                         else
1576                                                 set_bit(AS_EIO, &mapping->flags);
1577                                 } else {
1578                                         cifs_stats_bytes_written(cifs_sb->tcon,
1579                                                                  bytes_written);
1580                                 }
1581                         }
1582                         for (i = 0; i < n_iov; i++) {
1583                                 page = pvec.pages[first + i];
1584                                 /* Should we also set page error on
1585                                 success rc but too little data written? */
1586                                 /* BB investigate retry logic on temporary
1587                                 server crash cases and how recovery works
1588                                 when page marked as error */
1589                                 if (rc)
1590                                         SetPageError(page);
1591                                 kunmap(page);
1592                                 unlock_page(page);
1593                                 end_page_writeback(page);
1594                                 page_cache_release(page);
1595                         }
1596                         if ((wbc->nr_to_write -= n_iov) <= 0)
1597                                 done = 1;
1598                         index = next;
1599                 } else
1600                         /* Need to re-find the pages we skipped */
1601                         index = pvec.pages[0]->index + 1;
1602
1603                 pagevec_release(&pvec);
1604         }
1605         if (!scanned && !done) {
1606                 /*
1607                  * We hit the last page and there is more work to be done: wrap
1608                  * back to the start of the file
1609                  */
1610                 scanned = 1;
1611                 index = 0;
1612                 goto retry;
1613         }
1614         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1615                 mapping->writeback_index = index;
1616
1617         FreeXid(xid);
1618         kfree(iov);
1619         return rc;
1620 }
1621
1622 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1623 {
1624         int rc = -EFAULT;
1625         int xid;
1626
1627         xid = GetXid();
1628 /* BB add check for wbc flags */
1629         page_cache_get(page);
1630         if (!PageUptodate(page))
1631                 cFYI(1, ("ppw - page not up to date"));
1632
1633         /*
1634          * Set the "writeback" flag, and clear "dirty" in the radix tree.
1635          *
1636          * A writepage() implementation always needs to do either this,
1637          * or re-dirty the page with "redirty_page_for_writepage()" in
1638          * the case of a failure.
1639          *
1640          * Just unlocking the page will cause the radix tree tag-bits
1641          * to fail to update with the state of the page correctly.
1642          */
1643         set_page_writeback(page);
1644         rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1645         SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1646         unlock_page(page);
1647         end_page_writeback(page);
1648         page_cache_release(page);
1649         FreeXid(xid);
1650         return rc;
1651 }
1652
1653 static int cifs_write_end(struct file *file, struct address_space *mapping,
1654                         loff_t pos, unsigned len, unsigned copied,
1655                         struct page *page, void *fsdata)
1656 {
1657         int rc;
1658         struct inode *inode = mapping->host;
1659
1660         cFYI(1, ("write_end for page %p from pos %lld with %d bytes",
1661                  page, pos, copied));
1662
1663         if (PageChecked(page)) {
1664                 if (copied == len)
1665                         SetPageUptodate(page);
1666                 ClearPageChecked(page);
1667         } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1668                 SetPageUptodate(page);
1669
1670         if (!PageUptodate(page)) {
1671                 char *page_data;
1672                 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1673                 int xid;
1674
1675                 xid = GetXid();
1676                 /* this is probably better than directly calling
1677                    partialpage_write since in this function the file handle is
1678                    known which we might as well leverage */
1679                 /* BB check if anything else missing out of ppw
1680                    such as updating last write time */
1681                 page_data = kmap(page);
1682                 rc = cifs_write(file, page_data + offset, copied, &pos);
1683                 /* if (rc < 0) should we set writebehind rc? */
1684                 kunmap(page);
1685
1686                 FreeXid(xid);
1687         } else {
1688                 rc = copied;
1689                 pos += copied;
1690                 set_page_dirty(page);
1691         }
1692
1693         if (rc > 0) {
1694                 spin_lock(&inode->i_lock);
1695                 if (pos > inode->i_size)
1696                         i_size_write(inode, pos);
1697                 spin_unlock(&inode->i_lock);
1698         }
1699
1700         unlock_page(page);
1701         page_cache_release(page);
1702
1703         return rc;
1704 }
1705
1706 int cifs_fsync(struct file *file, struct dentry *dentry, int datasync)
1707 {
1708         int xid;
1709         int rc = 0;
1710         struct cifsTconInfo *tcon;
1711         struct cifsFileInfo *smbfile =
1712                 (struct cifsFileInfo *)file->private_data;
1713         struct inode *inode = file->f_path.dentry->d_inode;
1714
1715         xid = GetXid();
1716
1717         cFYI(1, ("Sync file - name: %s datasync: 0x%x",
1718                 dentry->d_name.name, datasync));
1719
1720         rc = filemap_write_and_wait(inode->i_mapping);
1721         if (rc == 0) {
1722                 rc = CIFS_I(inode)->write_behind_rc;
1723                 CIFS_I(inode)->write_behind_rc = 0;
1724                 tcon = CIFS_SB(inode->i_sb)->tcon;
1725                 if (!rc && tcon && smbfile &&
1726                    !(CIFS_SB(inode->i_sb)->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1727                         rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1728         }
1729
1730         FreeXid(xid);
1731         return rc;
1732 }
1733
1734 /* static void cifs_sync_page(struct page *page)
1735 {
1736         struct address_space *mapping;
1737         struct inode *inode;
1738         unsigned long index = page->index;
1739         unsigned int rpages = 0;
1740         int rc = 0;
1741
1742         cFYI(1, ("sync page %p",page));
1743         mapping = page->mapping;
1744         if (!mapping)
1745                 return 0;
1746         inode = mapping->host;
1747         if (!inode)
1748                 return; */
1749
1750 /*      fill in rpages then
1751         result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1752
1753 /*      cFYI(1, ("rpages is %d for sync page of Index %ld", rpages, index));
1754
1755 #if 0
1756         if (rc < 0)
1757                 return rc;
1758         return 0;
1759 #endif
1760 } */
1761
1762 /*
1763  * As file closes, flush all cached write data for this inode checking
1764  * for write behind errors.
1765  */
1766 int cifs_flush(struct file *file, fl_owner_t id)
1767 {
1768         struct inode *inode = file->f_path.dentry->d_inode;
1769         int rc = 0;
1770
1771         /* Rather than do the steps manually:
1772            lock the inode for writing
1773            loop through pages looking for write behind data (dirty pages)
1774            coalesce into contiguous 16K (or smaller) chunks to write to server
1775            send to server (prefer in parallel)
1776            deal with writebehind errors
1777            unlock inode for writing
1778            filemapfdatawrite appears easier for the time being */
1779
1780         rc = filemap_fdatawrite(inode->i_mapping);
1781         /* reset wb rc if we were able to write out dirty pages */
1782         if (!rc) {
1783                 rc = CIFS_I(inode)->write_behind_rc;
1784                 CIFS_I(inode)->write_behind_rc = 0;
1785         }
1786
1787         cFYI(1, ("Flush inode %p file %p rc %d", inode, file, rc));
1788
1789         return rc;
1790 }
1791
1792 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1793         size_t read_size, loff_t *poffset)
1794 {
1795         int rc = -EACCES;
1796         unsigned int bytes_read = 0;
1797         unsigned int total_read = 0;
1798         unsigned int current_read_size;
1799         struct cifs_sb_info *cifs_sb;
1800         struct cifsTconInfo *pTcon;
1801         int xid;
1802         struct cifsFileInfo *open_file;
1803         char *smb_read_data;
1804         char __user *current_offset;
1805         struct smb_com_read_rsp *pSMBr;
1806
1807         xid = GetXid();
1808         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1809         pTcon = cifs_sb->tcon;
1810
1811         if (file->private_data == NULL) {
1812                 rc = -EBADF;
1813                 FreeXid(xid);
1814                 return rc;
1815         }
1816         open_file = (struct cifsFileInfo *)file->private_data;
1817
1818         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1819                 cFYI(1, ("attempting read on write only file instance"));
1820
1821         for (total_read = 0, current_offset = read_data;
1822              read_size > total_read;
1823              total_read += bytes_read, current_offset += bytes_read) {
1824                 current_read_size = min_t(const int, read_size - total_read,
1825                                           cifs_sb->rsize);
1826                 rc = -EAGAIN;
1827                 smb_read_data = NULL;
1828                 while (rc == -EAGAIN) {
1829                         int buf_type = CIFS_NO_BUFFER;
1830                         if ((open_file->invalidHandle) &&
1831                             (!open_file->closePend)) {
1832                                 rc = cifs_reopen_file(file, true);
1833                                 if (rc != 0)
1834                                         break;
1835                         }
1836                         rc = CIFSSMBRead(xid, pTcon,
1837                                          open_file->netfid,
1838                                          current_read_size, *poffset,
1839                                          &bytes_read, &smb_read_data,
1840                                          &buf_type);
1841                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1842                         if (smb_read_data) {
1843                                 if (copy_to_user(current_offset,
1844                                                 smb_read_data +
1845                                                 4 /* RFC1001 length field */ +
1846                                                 le16_to_cpu(pSMBr->DataOffset),
1847                                                 bytes_read))
1848                                         rc = -EFAULT;
1849
1850                                 if (buf_type == CIFS_SMALL_BUFFER)
1851                                         cifs_small_buf_release(smb_read_data);
1852                                 else if (buf_type == CIFS_LARGE_BUFFER)
1853                                         cifs_buf_release(smb_read_data);
1854                                 smb_read_data = NULL;
1855                         }
1856                 }
1857                 if (rc || (bytes_read == 0)) {
1858                         if (total_read) {
1859                                 break;
1860                         } else {
1861                                 FreeXid(xid);
1862                                 return rc;
1863                         }
1864                 } else {
1865                         cifs_stats_bytes_read(pTcon, bytes_read);
1866                         *poffset += bytes_read;
1867                 }
1868         }
1869         FreeXid(xid);
1870         return total_read;
1871 }
1872
1873
1874 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1875         loff_t *poffset)
1876 {
1877         int rc = -EACCES;
1878         unsigned int bytes_read = 0;
1879         unsigned int total_read;
1880         unsigned int current_read_size;
1881         struct cifs_sb_info *cifs_sb;
1882         struct cifsTconInfo *pTcon;
1883         int xid;
1884         char *current_offset;
1885         struct cifsFileInfo *open_file;
1886         int buf_type = CIFS_NO_BUFFER;
1887
1888         xid = GetXid();
1889         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1890         pTcon = cifs_sb->tcon;
1891
1892         if (file->private_data == NULL) {
1893                 rc = -EBADF;
1894                 FreeXid(xid);
1895                 return rc;
1896         }
1897         open_file = (struct cifsFileInfo *)file->private_data;
1898
1899         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1900                 cFYI(1, ("attempting read on write only file instance"));
1901
1902         for (total_read = 0, current_offset = read_data;
1903              read_size > total_read;
1904              total_read += bytes_read, current_offset += bytes_read) {
1905                 current_read_size = min_t(const int, read_size - total_read,
1906                                           cifs_sb->rsize);
1907                 /* For windows me and 9x we do not want to request more
1908                 than it negotiated since it will refuse the read then */
1909                 if ((pTcon->ses) &&
1910                         !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1911                         current_read_size = min_t(const int, current_read_size,
1912                                         pTcon->ses->server->maxBuf - 128);
1913                 }
1914                 rc = -EAGAIN;
1915                 while (rc == -EAGAIN) {
1916                         if ((open_file->invalidHandle) &&
1917                             (!open_file->closePend)) {
1918                                 rc = cifs_reopen_file(file, true);
1919                                 if (rc != 0)
1920                                         break;
1921                         }
1922                         rc = CIFSSMBRead(xid, pTcon,
1923                                          open_file->netfid,
1924                                          current_read_size, *poffset,
1925                                          &bytes_read, &current_offset,
1926                                          &buf_type);
1927                 }
1928                 if (rc || (bytes_read == 0)) {
1929                         if (total_read) {
1930                                 break;
1931                         } else {
1932                                 FreeXid(xid);
1933                                 return rc;
1934                         }
1935                 } else {
1936                         cifs_stats_bytes_read(pTcon, total_read);
1937                         *poffset += bytes_read;
1938                 }
1939         }
1940         FreeXid(xid);
1941         return total_read;
1942 }
1943
1944 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1945 {
1946         struct dentry *dentry = file->f_path.dentry;
1947         int rc, xid;
1948
1949         xid = GetXid();
1950         rc = cifs_revalidate(dentry);
1951         if (rc) {
1952                 cFYI(1, ("Validation prior to mmap failed, error=%d", rc));
1953                 FreeXid(xid);
1954                 return rc;
1955         }
1956         rc = generic_file_mmap(file, vma);
1957         FreeXid(xid);
1958         return rc;
1959 }
1960
1961
1962 static void cifs_copy_cache_pages(struct address_space *mapping,
1963         struct list_head *pages, int bytes_read, char *data,
1964         struct pagevec *plru_pvec)
1965 {
1966         struct page *page;
1967         char *target;
1968
1969         while (bytes_read > 0) {
1970                 if (list_empty(pages))
1971                         break;
1972
1973                 page = list_entry(pages->prev, struct page, lru);
1974                 list_del(&page->lru);
1975
1976                 if (add_to_page_cache(page, mapping, page->index,
1977                                       GFP_KERNEL)) {
1978                         page_cache_release(page);
1979                         cFYI(1, ("Add page cache failed"));
1980                         data += PAGE_CACHE_SIZE;
1981                         bytes_read -= PAGE_CACHE_SIZE;
1982                         continue;
1983                 }
1984
1985                 target = kmap_atomic(page, KM_USER0);
1986
1987                 if (PAGE_CACHE_SIZE > bytes_read) {
1988                         memcpy(target, data, bytes_read);
1989                         /* zero the tail end of this partial page */
1990                         memset(target + bytes_read, 0,
1991                                PAGE_CACHE_SIZE - bytes_read);
1992                         bytes_read = 0;
1993                 } else {
1994                         memcpy(target, data, PAGE_CACHE_SIZE);
1995                         bytes_read -= PAGE_CACHE_SIZE;
1996                 }
1997                 kunmap_atomic(target, KM_USER0);
1998
1999                 flush_dcache_page(page);
2000                 SetPageUptodate(page);
2001                 unlock_page(page);
2002                 if (!pagevec_add(plru_pvec, page))
2003                         __pagevec_lru_add_file(plru_pvec);
2004                 data += PAGE_CACHE_SIZE;
2005         }
2006         return;
2007 }
2008
2009 static int cifs_readpages(struct file *file, struct address_space *mapping,
2010         struct list_head *page_list, unsigned num_pages)
2011 {
2012         int rc = -EACCES;
2013         int xid;
2014         loff_t offset;
2015         struct page *page;
2016         struct cifs_sb_info *cifs_sb;
2017         struct cifsTconInfo *pTcon;
2018         unsigned int bytes_read = 0;
2019         unsigned int read_size, i;
2020         char *smb_read_data = NULL;
2021         struct smb_com_read_rsp *pSMBr;
2022         struct pagevec lru_pvec;
2023         struct cifsFileInfo *open_file;
2024         int buf_type = CIFS_NO_BUFFER;
2025
2026         xid = GetXid();
2027         if (file->private_data == NULL) {
2028                 rc = -EBADF;
2029                 FreeXid(xid);
2030                 return rc;
2031         }
2032         open_file = (struct cifsFileInfo *)file->private_data;
2033         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2034         pTcon = cifs_sb->tcon;
2035
2036         pagevec_init(&lru_pvec, 0);
2037         cFYI(DBG2, ("rpages: num pages %d", num_pages));
2038         for (i = 0; i < num_pages; ) {
2039                 unsigned contig_pages;
2040                 struct page *tmp_page;
2041                 unsigned long expected_index;
2042
2043                 if (list_empty(page_list))
2044                         break;
2045
2046                 page = list_entry(page_list->prev, struct page, lru);
2047                 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2048
2049                 /* count adjacent pages that we will read into */
2050                 contig_pages = 0;
2051                 expected_index =
2052                         list_entry(page_list->prev, struct page, lru)->index;
2053                 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2054                         if (tmp_page->index == expected_index) {
2055                                 contig_pages++;
2056                                 expected_index++;
2057                         } else
2058                                 break;
2059                 }
2060                 if (contig_pages + i >  num_pages)
2061                         contig_pages = num_pages - i;
2062
2063                 /* for reads over a certain size could initiate async
2064                    read ahead */
2065
2066                 read_size = contig_pages * PAGE_CACHE_SIZE;
2067                 /* Read size needs to be in multiples of one page */
2068                 read_size = min_t(const unsigned int, read_size,
2069                                   cifs_sb->rsize & PAGE_CACHE_MASK);
2070                 cFYI(DBG2, ("rpages: read size 0x%x  contiguous pages %d",
2071                                 read_size, contig_pages));
2072                 rc = -EAGAIN;
2073                 while (rc == -EAGAIN) {
2074                         if ((open_file->invalidHandle) &&
2075                             (!open_file->closePend)) {
2076                                 rc = cifs_reopen_file(file, true);
2077                                 if (rc != 0)
2078                                         break;
2079                         }
2080
2081                         rc = CIFSSMBRead(xid, pTcon,
2082                                          open_file->netfid,
2083                                          read_size, offset,
2084                                          &bytes_read, &smb_read_data,
2085                                          &buf_type);
2086                         /* BB more RC checks ? */
2087                         if (rc == -EAGAIN) {
2088                                 if (smb_read_data) {
2089                                         if (buf_type == CIFS_SMALL_BUFFER)
2090                                                 cifs_small_buf_release(smb_read_data);
2091                                         else if (buf_type == CIFS_LARGE_BUFFER)
2092                                                 cifs_buf_release(smb_read_data);
2093                                         smb_read_data = NULL;
2094                                 }
2095                         }
2096                 }
2097                 if ((rc < 0) || (smb_read_data == NULL)) {
2098                         cFYI(1, ("Read error in readpages: %d", rc));
2099                         break;
2100                 } else if (bytes_read > 0) {
2101                         task_io_account_read(bytes_read);
2102                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2103                         cifs_copy_cache_pages(mapping, page_list, bytes_read,
2104                                 smb_read_data + 4 /* RFC1001 hdr */ +
2105                                 le16_to_cpu(pSMBr->DataOffset), &lru_pvec);
2106
2107                         i +=  bytes_read >> PAGE_CACHE_SHIFT;
2108                         cifs_stats_bytes_read(pTcon, bytes_read);
2109                         if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2110                                 i++; /* account for partial page */
2111
2112                                 /* server copy of file can have smaller size
2113                                    than client */
2114                                 /* BB do we need to verify this common case ?
2115                                    this case is ok - if we are at server EOF
2116                                    we will hit it on next read */
2117
2118                                 /* break; */
2119                         }
2120                 } else {
2121                         cFYI(1, ("No bytes read (%d) at offset %lld . "
2122                                  "Cleaning remaining pages from readahead list",
2123                                  bytes_read, offset));
2124                         /* BB turn off caching and do new lookup on
2125                            file size at server? */
2126                         break;
2127                 }
2128                 if (smb_read_data) {
2129                         if (buf_type == CIFS_SMALL_BUFFER)
2130                                 cifs_small_buf_release(smb_read_data);
2131                         else if (buf_type == CIFS_LARGE_BUFFER)
2132                                 cifs_buf_release(smb_read_data);
2133                         smb_read_data = NULL;
2134                 }
2135                 bytes_read = 0;
2136         }
2137
2138         pagevec_lru_add_file(&lru_pvec);
2139
2140 /* need to free smb_read_data buf before exit */
2141         if (smb_read_data) {
2142                 if (buf_type == CIFS_SMALL_BUFFER)
2143                         cifs_small_buf_release(smb_read_data);
2144                 else if (buf_type == CIFS_LARGE_BUFFER)
2145                         cifs_buf_release(smb_read_data);
2146                 smb_read_data = NULL;
2147         }
2148
2149         FreeXid(xid);
2150         return rc;
2151 }
2152
2153 static int cifs_readpage_worker(struct file *file, struct page *page,
2154         loff_t *poffset)
2155 {
2156         char *read_data;
2157         int rc;
2158
2159         page_cache_get(page);
2160         read_data = kmap(page);
2161         /* for reads over a certain size could initiate async read ahead */
2162
2163         rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2164
2165         if (rc < 0)
2166                 goto io_error;
2167         else
2168                 cFYI(1, ("Bytes read %d", rc));
2169
2170         file->f_path.dentry->d_inode->i_atime =
2171                 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2172
2173         if (PAGE_CACHE_SIZE > rc)
2174                 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2175
2176         flush_dcache_page(page);
2177         SetPageUptodate(page);
2178         rc = 0;
2179
2180 io_error:
2181         kunmap(page);
2182         page_cache_release(page);
2183         return rc;
2184 }
2185
2186 static int cifs_readpage(struct file *file, struct page *page)
2187 {
2188         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2189         int rc = -EACCES;
2190         int xid;
2191
2192         xid = GetXid();
2193
2194         if (file->private_data == NULL) {
2195                 rc = -EBADF;
2196                 FreeXid(xid);
2197                 return rc;
2198         }
2199
2200         cFYI(1, ("readpage %p at offset %d 0x%x\n",
2201                  page, (int)offset, (int)offset));
2202
2203         rc = cifs_readpage_worker(file, page, &offset);
2204
2205         unlock_page(page);
2206
2207         FreeXid(xid);
2208         return rc;
2209 }
2210
2211 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2212 {
2213         struct cifsFileInfo *open_file;
2214
2215         read_lock(&GlobalSMBSeslock);
2216         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2217                 if (open_file->closePend)
2218                         continue;
2219                 if (open_file->pfile &&
2220                     ((open_file->pfile->f_flags & O_RDWR) ||
2221                      (open_file->pfile->f_flags & O_WRONLY))) {
2222                         read_unlock(&GlobalSMBSeslock);
2223                         return 1;
2224                 }
2225         }
2226         read_unlock(&GlobalSMBSeslock);
2227         return 0;
2228 }
2229
2230 /* We do not want to update the file size from server for inodes
2231    open for write - to avoid races with writepage extending
2232    the file - in the future we could consider allowing
2233    refreshing the inode only on increases in the file size
2234    but this is tricky to do without racing with writebehind
2235    page caching in the current Linux kernel design */
2236 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2237 {
2238         if (!cifsInode)
2239                 return true;
2240
2241         if (is_inode_writable(cifsInode)) {
2242                 /* This inode is open for write at least once */
2243                 struct cifs_sb_info *cifs_sb;
2244
2245                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2246                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2247                         /* since no page cache to corrupt on directio
2248                         we can change size safely */
2249                         return true;
2250                 }
2251
2252                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2253                         return true;
2254
2255                 return false;
2256         } else
2257                 return true;
2258 }
2259
2260 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2261                         loff_t pos, unsigned len, unsigned flags,
2262                         struct page **pagep, void **fsdata)
2263 {
2264         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2265         loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2266         loff_t page_start = pos & PAGE_MASK;
2267         loff_t i_size;
2268         struct page *page;
2269         int rc = 0;
2270
2271         cFYI(1, ("write_begin from %lld len %d", (long long)pos, len));
2272
2273         page = grab_cache_page_write_begin(mapping, index, flags);
2274         if (!page) {
2275                 rc = -ENOMEM;
2276                 goto out;
2277         }
2278
2279         if (PageUptodate(page))
2280                 goto out;
2281
2282         /*
2283          * If we write a full page it will be up to date, no need to read from
2284          * the server. If the write is short, we'll end up doing a sync write
2285          * instead.
2286          */
2287         if (len == PAGE_CACHE_SIZE)
2288                 goto out;
2289
2290         /*
2291          * optimize away the read when we have an oplock, and we're not
2292          * expecting to use any of the data we'd be reading in. That
2293          * is, when the page lies beyond the EOF, or straddles the EOF
2294          * and the write will cover all of the existing data.
2295          */
2296         if (CIFS_I(mapping->host)->clientCanCacheRead) {
2297                 i_size = i_size_read(mapping->host);
2298                 if (page_start >= i_size ||
2299                     (offset == 0 && (pos + len) >= i_size)) {
2300                         zero_user_segments(page, 0, offset,
2301                                            offset + len,
2302                                            PAGE_CACHE_SIZE);
2303                         /*
2304                          * PageChecked means that the parts of the page
2305                          * to which we're not writing are considered up
2306                          * to date. Once the data is copied to the
2307                          * page, it can be set uptodate.
2308                          */
2309                         SetPageChecked(page);
2310                         goto out;
2311                 }
2312         }
2313
2314         if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2315                 /*
2316                  * might as well read a page, it is fast enough. If we get
2317                  * an error, we don't need to return it. cifs_write_end will
2318                  * do a sync write instead since PG_uptodate isn't set.
2319                  */
2320                 cifs_readpage_worker(file, page, &page_start);
2321         } else {
2322                 /* we could try using another file handle if there is one -
2323                    but how would we lock it to prevent close of that handle
2324                    racing with this read? In any case
2325                    this will be written out by write_end so is fine */
2326         }
2327 out:
2328         *pagep = page;
2329         return rc;
2330 }
2331
2332 const struct address_space_operations cifs_addr_ops = {
2333         .readpage = cifs_readpage,
2334         .readpages = cifs_readpages,
2335         .writepage = cifs_writepage,
2336         .writepages = cifs_writepages,
2337         .write_begin = cifs_write_begin,
2338         .write_end = cifs_write_end,
2339         .set_page_dirty = __set_page_dirty_nobuffers,
2340         /* .sync_page = cifs_sync_page, */
2341         /* .direct_IO = */
2342 };
2343
2344 /*
2345  * cifs_readpages requires the server to support a buffer large enough to
2346  * contain the header plus one complete page of data.  Otherwise, we need
2347  * to leave cifs_readpages out of the address space operations.
2348  */
2349 const struct address_space_operations cifs_addr_ops_smallbuf = {
2350         .readpage = cifs_readpage,
2351         .writepage = cifs_writepage,
2352         .writepages = cifs_writepages,
2353         .write_begin = cifs_write_begin,
2354         .write_end = cifs_write_end,
2355         .set_page_dirty = __set_page_dirty_nobuffers,
2356         /* .sync_page = cifs_sync_page, */
2357         /* .direct_IO = */
2358 };