fs/cifs/file.c

   1 /*
   2  *   fs/cifs/file.c
   3  *
   4  *   vfs operations that deal with files
   5  *
   6  *   Copyright (C) International Business Machines  Corp., 2002,2010
   7  *   Author(s): Steve French (sfrench@us.ibm.com)
   8  *              Jeremy Allison (jra@samba.org)
   9  *
  10  *   This library is free software; you can redistribute it and/or modify
  11  *   it under the terms of the GNU Lesser General Public License as published
  12  *   by the Free Software Foundation; either version 2.1 of the License, or
  13  *   (at your option) any later version.
  14  *
  15  *   This library is distributed in the hope that it will be useful,
  16  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
  18  *   the GNU Lesser General Public License for more details.
  19  *
  20  *   You should have received a copy of the GNU Lesser General Public License
  21  *   along with this library; if not, write to the Free Software
  22  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  23  */
  24 #include <linux/fs.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/stat.h>
  27 #include <linux/fcntl.h>
  28 #include <linux/pagemap.h>
  29 #include <linux/pagevec.h>
  30 #include <linux/writeback.h>
  31 #include <linux/task_io_accounting_ops.h>
  32 #include <linux/delay.h>
  33 #include <linux/mount.h>
  34 #include <linux/slab.h>
  35 #include <asm/div64.h>
  36 #include "cifsfs.h"
  37 #include "cifspdu.h"
  38 #include "cifsglob.h"
  39 #include "cifsproto.h"
  40 #include "cifs_unicode.h"
  41 #include "cifs_debug.h"
  42 #include "cifs_fs_sb.h"
  43 #include "fscache.h"
  44
  45 static inline int cifs_convert_flags(unsigned int flags)
  46 {
  47         if ((flags & O_ACCMODE) == O_RDONLY)
  48                 return GENERIC_READ;
  49         else if ((flags & O_ACCMODE) == O_WRONLY)
  50                 return GENERIC_WRITE;
  51         else if ((flags & O_ACCMODE) == O_RDWR) {
  52                 /* GENERIC_ALL is too much permission to request
  53                    can cause unnecessary access denied on create */
  54                 /* return GENERIC_ALL; */
  55                 return (GENERIC_READ | GENERIC_WRITE);
  56         }
  57
  58         return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
  59                 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
  60                 FILE_READ_DATA);
  61 }
  62
  63 static inline fmode_t cifs_posix_convert_flags(unsigned int flags)
  64 {
  65         fmode_t posix_flags = 0;
  66
  67         if ((flags & O_ACCMODE) == O_RDONLY)
  68                 posix_flags = FMODE_READ;
  69         else if ((flags & O_ACCMODE) == O_WRONLY)
  70                 posix_flags = FMODE_WRITE;
  71         else if ((flags & O_ACCMODE) == O_RDWR) {
  72                 /* GENERIC_ALL is too much permission to request
  73                    can cause unnecessary access denied on create */
  74                 /* return GENERIC_ALL; */
  75                 posix_flags = FMODE_READ | FMODE_WRITE;
  76         }
  77         /* can not map O_CREAT or O_EXCL or O_TRUNC flags when
  78            reopening a file.  They had their effect on the original open */
  79         if (flags & O_APPEND)
  80                 posix_flags |= (fmode_t)O_APPEND;
  81         if (flags & O_DSYNC)
  82                 posix_flags |= (fmode_t)O_DSYNC;
  83         if (flags & __O_SYNC)
  84                 posix_flags |= (fmode_t)__O_SYNC;
  85         if (flags & O_DIRECTORY)
  86                 posix_flags |= (fmode_t)O_DIRECTORY;
  87         if (flags & O_NOFOLLOW)
  88                 posix_flags |= (fmode_t)O_NOFOLLOW;
  89         if (flags & O_DIRECT)
  90                 posix_flags |= (fmode_t)O_DIRECT;
  91
  92         return posix_flags;
  93 }
  94
  95 static inline int cifs_get_disposition(unsigned int flags)
  96 {
  97         if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
  98                 return FILE_CREATE;
  99         else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
 100                 return FILE_OVERWRITE_IF;
 101         else if ((flags & O_CREAT) == O_CREAT)
 102                 return FILE_OPEN_IF;
 103         else if ((flags & O_TRUNC) == O_TRUNC)
 104                 return FILE_OVERWRITE;
 105         else
 106                 return FILE_OPEN;
 107 }
 108
 109 /* all arguments to this function must be checked for validity in caller */
 110 static inline int
 111 cifs_posix_open_inode_helper(struct inode *inode, struct file *file,
 112                              struct cifsInodeInfo *pCifsInode, __u32 oplock,
 113                              u16 netfid)
 114 {
 115
 116         write_lock(&GlobalSMBSeslock);
 117
 118         pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
 119         if (pCifsInode == NULL) {
 120                 write_unlock(&GlobalSMBSeslock);
 121                 return -EINVAL;
 122         }
 123
 124         if (pCifsInode->clientCanCacheRead) {
 125                 /* we have the inode open somewhere else
 126                    no need to discard cache data */
 127                 goto psx_client_can_cache;
 128         }
 129
 130         /* BB FIXME need to fix this check to move it earlier into posix_open
 131            BB  fIX following section BB FIXME */
 132
 133         /* if not oplocked, invalidate inode pages if mtime or file
 134            size changed */
 135 /*      temp = cifs_NTtimeToUnix(le64_to_cpu(buf->LastWriteTime));
 136         if (timespec_equal(&file->f_path.dentry->d_inode->i_mtime, &temp) &&
 137                            (file->f_path.dentry->d_inode->i_size ==
 138                             (loff_t)le64_to_cpu(buf->EndOfFile))) {
 139                 cFYI(1, "inode unchanged on server");
 140         } else {
 141                 if (file->f_path.dentry->d_inode->i_mapping) {
 142                         rc = filemap_write_and_wait(file->f_path.dentry->d_inode->i_mapping);
 143                         if (rc != 0)
 144                                 CIFS_I(file->f_path.dentry->d_inode)->write_behind_rc = rc;
 145                 }
 146                 cFYI(1, "invalidating remote inode since open detected it "
 147                          "changed");
 148                 invalidate_remote_inode(file->f_path.dentry->d_inode);
 149         } */
 150
 151 psx_client_can_cache:
 152         if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
 153                 pCifsInode->clientCanCacheAll = true;
 154                 pCifsInode->clientCanCacheRead = true;
 155                 cFYI(1, "Exclusive Oplock granted on inode %p",
 156                          file->f_path.dentry->d_inode);
 157         } else if ((oplock & 0xF) == OPLOCK_READ)
 158                 pCifsInode->clientCanCacheRead = true;
 159
 160         /* will have to change the unlock if we reenable the
 161            filemap_fdatawrite (which does not seem necessary */
 162         write_unlock(&GlobalSMBSeslock);
 163         return 0;
 164 }
 165
 166 /* all arguments to this function must be checked for validity in caller */
 167 static inline int cifs_open_inode_helper(struct inode *inode,
 168         struct cifsTconInfo *pTcon, __u32 oplock, FILE_ALL_INFO *buf,
 169         char *full_path, int xid)
 170 {
 171         struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
 172         struct timespec temp;
 173         int rc;
 174
 175         if (pCifsInode->clientCanCacheRead) {
 176                 /* we have the inode open somewhere else
 177                    no need to discard cache data */
 178                 goto client_can_cache;
 179         }
 180
 181         /* BB need same check in cifs_create too? */
 182         /* if not oplocked, invalidate inode pages if mtime or file
 183            size changed */
 184         temp = cifs_NTtimeToUnix(buf->LastWriteTime);
 185         if (timespec_equal(&inode->i_mtime, &temp) &&
 186                            (inode->i_size ==
 187                             (loff_t)le64_to_cpu(buf->EndOfFile))) {
 188                 cFYI(1, "inode unchanged on server");
 189         } else {
 190                 if (inode->i_mapping) {
 191                         /* BB no need to lock inode until after invalidate
 192                         since namei code should already have it locked? */
 193                         rc = filemap_write_and_wait(inode->i_mapping);
 194                         if (rc != 0)
 195                                 pCifsInode->write_behind_rc = rc;
 196                 }
 197                 cFYI(1, "invalidating remote inode since open detected it "
 198                          "changed");
 199                 invalidate_remote_inode(inode);
 200         }
 201
 202 client_can_cache:
 203         if (pTcon->unix_ext)
 204                 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
 205                                               xid);
 206         else
 207                 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
 208                                          xid, NULL);
 209
 210         if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
 211                 pCifsInode->clientCanCacheAll = true;
 212                 pCifsInode->clientCanCacheRead = true;
 213                 cFYI(1, "Exclusive Oplock granted on inode %p", inode);
 214         } else if ((oplock & 0xF) == OPLOCK_READ)
 215                 pCifsInode->clientCanCacheRead = true;
 216
 217         return rc;
 218 }
 219
 220 int cifs_open(struct inode *inode, struct file *file)
 221 {
 222         int rc = -EACCES;
 223         int xid;
 224         __u32 oplock;
 225         struct cifs_sb_info *cifs_sb;
 226         struct cifsTconInfo *tcon;
 227         struct tcon_link *tlink;
 228         struct cifsFileInfo *pCifsFile = NULL;
 229         struct cifsInodeInfo *pCifsInode;
 230         char *full_path = NULL;
 231         int desiredAccess;
 232         int disposition;
 233         __u16 netfid;
 234         FILE_ALL_INFO *buf = NULL;
 235
 236         xid = GetXid();
 237
 238         cifs_sb = CIFS_SB(inode->i_sb);
 239         tlink = cifs_sb_tlink(cifs_sb);
 240         if (IS_ERR(tlink)) {
 241                 FreeXid(xid);
 242                 return PTR_ERR(tlink);
 243         }
 244         tcon = tlink_tcon(tlink);
 245
 246         pCifsInode = CIFS_I(file->f_path.dentry->d_inode);
 247
 248         full_path = build_path_from_dentry(file->f_path.dentry);
 249         if (full_path == NULL) {
 250                 rc = -ENOMEM;
 251                 goto out;
 252         }
 253
 254         cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
 255                  inode, file->f_flags, full_path);
 256
 257         if (oplockEnabled)
 258                 oplock = REQ_OPLOCK;
 259         else
 260                 oplock = 0;
 261
 262         if (!tcon->broken_posix_open && tcon->unix_ext &&
 263             (tcon->ses->capabilities & CAP_UNIX) &&
 264             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 265                         le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 266                 int oflags = (int) cifs_posix_convert_flags(file->f_flags);
 267                 oflags |= SMB_O_CREAT;
 268                 /* can not refresh inode info since size could be stale */
 269                 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
 270                                 cifs_sb->mnt_file_mode /* ignored */,
 271                                 oflags, &oplock, &netfid, xid);
 272                 if (rc == 0) {
 273                         cFYI(1, "posix open succeeded");
 274                         /* no need for special case handling of setting mode
 275                            on read only files needed here */
 276
 277                         rc = cifs_posix_open_inode_helper(inode, file,
 278                                         pCifsInode, oplock, netfid);
 279                         if (rc != 0) {
 280                                 CIFSSMBClose(xid, tcon, netfid);
 281                                 goto out;
 282                         }
 283
 284                         pCifsFile = cifs_new_fileinfo(inode, netfid, file,
 285                                                         file->f_path.mnt,
 286                                                         tlink, oflags, oplock);
 287                         if (pCifsFile == NULL) {
 288                                 CIFSSMBClose(xid, tcon, netfid);
 289                                 rc = -ENOMEM;
 290                         }
 291
 292                         cifs_fscache_set_inode_cookie(inode, file);
 293
 294                         goto out;
 295                 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
 296                         if (tcon->ses->serverNOS)
 297                                 cERROR(1, "server %s of type %s returned"
 298                                            " unexpected error on SMB posix open"
 299                                            ", disabling posix open support."
 300                                            " Check if server update available.",
 301                                            tcon->ses->serverName,
 302                                            tcon->ses->serverNOS);
 303                         tcon->broken_posix_open = true;
 304                 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
 305                          (rc != -EOPNOTSUPP)) /* path not found or net err */
 306                         goto out;
 307                 /* else fallthrough to retry open the old way on network i/o
 308                    or DFS errors */
 309         }
 310
 311         desiredAccess = cifs_convert_flags(file->f_flags);
 312
 313 /*********************************************************************
 314  *  open flag mapping table:
 315  *
 316  *      POSIX Flag            CIFS Disposition
 317  *      ----------            ----------------
 318  *      O_CREAT               FILE_OPEN_IF
 319  *      O_CREAT | O_EXCL      FILE_CREATE
 320  *      O_CREAT | O_TRUNC     FILE_OVERWRITE_IF
 321  *      O_TRUNC               FILE_OVERWRITE
 322  *      none of the above     FILE_OPEN
 323  *
 324  *      Note that there is not a direct match between disposition
 325  *      FILE_SUPERSEDE (ie create whether or not file exists although
 326  *      O_CREAT | O_TRUNC is similar but truncates the existing
 327  *      file rather than creating a new file as FILE_SUPERSEDE does
 328  *      (which uses the attributes / metadata passed in on open call)
 329  *?
 330  *?  O_SYNC is a reasonable match to CIFS writethrough flag
 331  *?  and the read write flags match reasonably.  O_LARGEFILE
 332  *?  is irrelevant because largefile support is always used
 333  *?  by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
 334  *       O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
 335  *********************************************************************/
 336
 337         disposition = cifs_get_disposition(file->f_flags);
 338
 339         /* BB pass O_SYNC flag through on file attributes .. BB */
 340
 341         /* Also refresh inode by passing in file_info buf returned by SMBOpen
 342            and calling get_inode_info with returned buf (at least helps
 343            non-Unix server case) */
 344
 345         /* BB we can not do this if this is the second open of a file
 346            and the first handle has writebehind data, we might be
 347            able to simply do a filemap_fdatawrite/filemap_fdatawait first */
 348         buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
 349         if (!buf) {
 350                 rc = -ENOMEM;
 351                 goto out;
 352         }
 353
 354         if (tcon->ses->capabilities & CAP_NT_SMBS)
 355                 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
 356                          desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
 357                          cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
 358                                  & CIFS_MOUNT_MAP_SPECIAL_CHR);
 359         else
 360                 rc = -EIO; /* no NT SMB support fall into legacy open below */
 361
 362         if (rc == -EIO) {
 363                 /* Old server, try legacy style OpenX */
 364                 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
 365                         desiredAccess, CREATE_NOT_DIR, &netfid, &oplock, buf,
 366                         cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
 367                                 & CIFS_MOUNT_MAP_SPECIAL_CHR);
 368         }
 369         if (rc) {
 370                 cFYI(1, "cifs_open returned 0x%x", rc);
 371                 goto out;
 372         }
 373
 374         rc = cifs_open_inode_helper(inode, tcon, oplock, buf, full_path, xid);
 375         if (rc != 0)
 376                 goto out;
 377
 378         pCifsFile = cifs_new_fileinfo(inode, netfid, file, file->f_path.mnt,
 379                                         tlink, file->f_flags, oplock);
 380         if (pCifsFile == NULL) {
 381                 rc = -ENOMEM;
 382                 goto out;
 383         }
 384
 385         cifs_fscache_set_inode_cookie(inode, file);
 386
 387         if (oplock & CIFS_CREATE_ACTION) {
 388                 /* time to set mode which we can not set earlier due to
 389                    problems creating new read-only files */
 390                 if (tcon->unix_ext) {
 391                         struct cifs_unix_set_info_args args = {
 392                                 .mode   = inode->i_mode,
 393                                 .uid    = NO_CHANGE_64,
 394                                 .gid    = NO_CHANGE_64,
 395                                 .ctime  = NO_CHANGE_64,
 396                                 .atime  = NO_CHANGE_64,
 397                                 .mtime  = NO_CHANGE_64,
 398                                 .device = 0,
 399                         };
 400                         CIFSSMBUnixSetPathInfo(xid, tcon, full_path, &args,
 401                                                cifs_sb->local_nls,
 402                                                cifs_sb->mnt_cifs_flags &
 403                                                 CIFS_MOUNT_MAP_SPECIAL_CHR);
 404                 }
 405         }
 406
 407 out:
 408         kfree(buf);
 409         kfree(full_path);
 410         FreeXid(xid);
 411         cifs_put_tlink(tlink);
 412         return rc;
 413 }
 414
 415 /* Try to reacquire byte range locks that were released when session */
 416 /* to server was lost */
 417 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
 418 {
 419         int rc = 0;
 420
 421 /* BB list all locks open on this file and relock */
 422
 423         return rc;
 424 }
 425
 426 static int cifs_reopen_file(struct file *file, bool can_flush)
 427 {
 428         int rc = -EACCES;
 429         int xid;
 430         __u32 oplock;
 431         struct cifs_sb_info *cifs_sb;
 432         struct cifsTconInfo *tcon;
 433         struct cifsFileInfo *pCifsFile;
 434         struct cifsInodeInfo *pCifsInode;
 435         struct inode *inode;
 436         char *full_path = NULL;
 437         int desiredAccess;
 438         int disposition = FILE_OPEN;
 439         __u16 netfid;
 440
 441         if (file->private_data)
 442                 pCifsFile = file->private_data;
 443         else
 444                 return -EBADF;
 445
 446         xid = GetXid();
 447         mutex_lock(&pCifsFile->fh_mutex);
 448         if (!pCifsFile->invalidHandle) {
 449                 mutex_unlock(&pCifsFile->fh_mutex);
 450                 rc = 0;
 451                 FreeXid(xid);
 452                 return rc;
 453         }
 454
 455         if (file->f_path.dentry == NULL) {
 456                 cERROR(1, "no valid name if dentry freed");
 457                 dump_stack();
 458                 rc = -EBADF;
 459                 goto reopen_error_exit;
 460         }
 461
 462         inode = file->f_path.dentry->d_inode;
 463         if (inode == NULL) {
 464                 cERROR(1, "inode not valid");
 465                 dump_stack();
 466                 rc = -EBADF;
 467                 goto reopen_error_exit;
 468         }
 469
 470         cifs_sb = CIFS_SB(inode->i_sb);
 471         tcon = tlink_tcon(pCifsFile->tlink);
 472
 473 /* can not grab rename sem here because various ops, including
 474    those that already have the rename sem can end up causing writepage
 475    to get called and if the server was down that means we end up here,
 476    and we can never tell if the caller already has the rename_sem */
 477         full_path = build_path_from_dentry(file->f_path.dentry);
 478         if (full_path == NULL) {
 479                 rc = -ENOMEM;
 480 reopen_error_exit:
 481                 mutex_unlock(&pCifsFile->fh_mutex);
 482                 FreeXid(xid);
 483                 return rc;
 484         }
 485
 486         cFYI(1, "inode = 0x%p file flags 0x%x for %s",
 487                  inode, file->f_flags, full_path);
 488
 489         if (oplockEnabled)
 490                 oplock = REQ_OPLOCK;
 491         else
 492                 oplock = 0;
 493
 494         if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
 495             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 496                         le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 497                 int oflags = (int) cifs_posix_convert_flags(file->f_flags);
 498                 /* can not refresh inode info since size could be stale */
 499                 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
 500                                 cifs_sb->mnt_file_mode /* ignored */,
 501                                 oflags, &oplock, &netfid, xid);
 502                 if (rc == 0) {
 503                         cFYI(1, "posix reopen succeeded");
 504                         goto reopen_success;
 505                 }
 506                 /* fallthrough to retry open the old way on errors, especially
 507                    in the reconnect path it is important to retry hard */
 508         }
 509
 510         desiredAccess = cifs_convert_flags(file->f_flags);
 511
 512         /* Can not refresh inode by passing in file_info buf to be returned
 513            by SMBOpen and then calling get_inode_info with returned buf
 514            since file might have write behind data that needs to be flushed
 515            and server version of file size can be stale. If we knew for sure
 516            that inode was not dirty locally we could do this */
 517
 518         rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
 519                          CREATE_NOT_DIR, &netfid, &oplock, NULL,
 520                          cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
 521                                 CIFS_MOUNT_MAP_SPECIAL_CHR);
 522         if (rc) {
 523                 mutex_unlock(&pCifsFile->fh_mutex);
 524                 cFYI(1, "cifs_open returned 0x%x", rc);
 525                 cFYI(1, "oplock: %d", oplock);
 526         } else {
 527 reopen_success:
 528                 pCifsFile->netfid = netfid;
 529                 pCifsFile->invalidHandle = false;
 530                 mutex_unlock(&pCifsFile->fh_mutex);
 531                 pCifsInode = CIFS_I(inode);
 532                 if (pCifsInode) {
 533                         if (can_flush) {
 534                                 rc = filemap_write_and_wait(inode->i_mapping);
 535                                 if (rc != 0)
 536                                         CIFS_I(inode)->write_behind_rc = rc;
 537                         /* temporarily disable caching while we
 538                            go to server to get inode info */
 539                                 pCifsInode->clientCanCacheAll = false;
 540                                 pCifsInode->clientCanCacheRead = false;
 541                                 if (tcon->unix_ext)
 542                                         rc = cifs_get_inode_info_unix(&inode,
 543                                                 full_path, inode->i_sb, xid);
 544                                 else
 545                                         rc = cifs_get_inode_info(&inode,
 546                                                 full_path, NULL, inode->i_sb,
 547                                                 xid, NULL);
 548                         } /* else we are writing out data to server already
 549                              and could deadlock if we tried to flush data, and
 550                              since we do not know if we have data that would
 551                              invalidate the current end of file on the server
 552                              we can not go to the server to get the new inod
 553                              info */
 554                         if ((oplock & 0xF) == OPLOCK_EXCLUSIVE) {
 555                                 pCifsInode->clientCanCacheAll = true;
 556                                 pCifsInode->clientCanCacheRead = true;
 557                                 cFYI(1, "Exclusive Oplock granted on inode %p",
 558                                          file->f_path.dentry->d_inode);
 559                         } else if ((oplock & 0xF) == OPLOCK_READ) {
 560                                 pCifsInode->clientCanCacheRead = true;
 561                                 pCifsInode->clientCanCacheAll = false;
 562                         } else {
 563                                 pCifsInode->clientCanCacheRead = false;
 564                                 pCifsInode->clientCanCacheAll = false;
 565                         }
 566                         cifs_relock_file(pCifsFile);
 567                 }
 568         }
 569         kfree(full_path);
 570         FreeXid(xid);
 571         return rc;
 572 }
 573
 574 int cifs_close(struct inode *inode, struct file *file)
 575 {
 576         int rc = 0;
 577         int xid, timeout;
 578         struct cifs_sb_info *cifs_sb;
 579         struct cifsTconInfo *pTcon;
 580         struct cifsFileInfo *pSMBFile = file->private_data;
 581
 582         xid = GetXid();
 583
 584         cifs_sb = CIFS_SB(inode->i_sb);
 585         pTcon = tlink_tcon(pSMBFile->tlink);
 586         if (pSMBFile) {
 587                 struct cifsLockInfo *li, *tmp;
 588                 write_lock(&GlobalSMBSeslock);
 589                 pSMBFile->closePend = true;
 590                 if (pTcon) {
 591                         /* no sense reconnecting to close a file that is
 592                            already closed */
 593                         if (!pTcon->need_reconnect) {
 594                                 write_unlock(&GlobalSMBSeslock);
 595                                 timeout = 2;
 596                                 while ((atomic_read(&pSMBFile->count) != 1)
 597                                         && (timeout <= 2048)) {
 598                                         /* Give write a better chance to get to
 599                                         server ahead of the close.  We do not
 600                                         want to add a wait_q here as it would
 601                                         increase the memory utilization as
 602                                         the struct would be in each open file,
 603                                         but this should give enough time to
 604                                         clear the socket */
 605                                         cFYI(DBG2, "close delay, write pending");
 606                                         msleep(timeout);
 607                                         timeout *= 4;
 608                                 }
 609                                 if (!pTcon->need_reconnect &&
 610                                     !pSMBFile->invalidHandle)
 611                                         rc = CIFSSMBClose(xid, pTcon,
 612                                                   pSMBFile->netfid);
 613                         } else
 614                                 write_unlock(&GlobalSMBSeslock);
 615                 } else
 616                         write_unlock(&GlobalSMBSeslock);
 617
 618                 /* Delete any outstanding lock records.
 619                    We'll lose them when the file is closed anyway. */
 620                 mutex_lock(&pSMBFile->lock_mutex);
 621                 list_for_each_entry_safe(li, tmp, &pSMBFile->llist, llist) {
 622                         list_del(&li->llist);
 623                         kfree(li);
 624                 }
 625                 mutex_unlock(&pSMBFile->lock_mutex);
 626
 627                 write_lock(&GlobalSMBSeslock);
 628                 list_del(&pSMBFile->flist);
 629                 list_del(&pSMBFile->tlist);
 630                 write_unlock(&GlobalSMBSeslock);
 631                 cifsFileInfo_put(file->private_data);
 632                 file->private_data = NULL;
 633         } else
 634                 rc = -EBADF;
 635
 636         read_lock(&GlobalSMBSeslock);
 637         if (list_empty(&(CIFS_I(inode)->openFileList))) {
 638                 cFYI(1, "closing last open instance for inode %p", inode);
 639                 /* if the file is not open we do not know if we can cache info
 640                    on this inode, much less write behind and read ahead */
 641                 CIFS_I(inode)->clientCanCacheRead = false;
 642                 CIFS_I(inode)->clientCanCacheAll  = false;
 643         }
 644         read_unlock(&GlobalSMBSeslock);
 645         if ((rc == 0) && CIFS_I(inode)->write_behind_rc)
 646                 rc = CIFS_I(inode)->write_behind_rc;
 647         FreeXid(xid);
 648         return rc;
 649 }
 650
 651 int cifs_closedir(struct inode *inode, struct file *file)
 652 {
 653         int rc = 0;
 654         int xid;
 655         struct cifsFileInfo *pCFileStruct = file->private_data;
 656         char *ptmp;
 657
 658         cFYI(1, "Closedir inode = 0x%p", inode);
 659
 660         xid = GetXid();
 661
 662         if (pCFileStruct) {
 663                 struct cifsTconInfo *pTcon = tlink_tcon(pCFileStruct->tlink);
 664
 665                 cFYI(1, "Freeing private data in close dir");
 666                 write_lock(&GlobalSMBSeslock);
 667                 if (!pCFileStruct->srch_inf.endOfSearch &&
 668                     !pCFileStruct->invalidHandle) {
 669                         pCFileStruct->invalidHandle = true;
 670                         write_unlock(&GlobalSMBSeslock);
 671                         rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
 672                         cFYI(1, "Closing uncompleted readdir with rc %d",
 673                                  rc);
 674                         /* not much we can do if it fails anyway, ignore rc */
 675                         rc = 0;
 676                 } else
 677                         write_unlock(&GlobalSMBSeslock);
 678                 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
 679                 if (ptmp) {
 680                         cFYI(1, "closedir free smb buf in srch struct");
 681                         pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
 682                         if (pCFileStruct->srch_inf.smallBuf)
 683                                 cifs_small_buf_release(ptmp);
 684                         else
 685                                 cifs_buf_release(ptmp);
 686                 }
 687                 cifs_put_tlink(pCFileStruct->tlink);
 688                 kfree(file->private_data);
 689                 file->private_data = NULL;
 690         }
 691         /* BB can we lock the filestruct while this is going on? */
 692         FreeXid(xid);
 693         return rc;
 694 }
 695
 696 static int store_file_lock(struct cifsFileInfo *fid, __u64 len,
 697                                 __u64 offset, __u8 lockType)
 698 {
 699         struct cifsLockInfo *li =
 700                 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
 701         if (li == NULL)
 702                 return -ENOMEM;
 703         li->offset = offset;
 704         li->length = len;
 705         li->type = lockType;
 706         mutex_lock(&fid->lock_mutex);
 707         list_add(&li->llist, &fid->llist);
 708         mutex_unlock(&fid->lock_mutex);
 709         return 0;
 710 }
 711
 712 int cifs_lock(struct file *file, int cmd, struct file_lock *pfLock)
 713 {
 714         int rc, xid;
 715         __u32 numLock = 0;
 716         __u32 numUnlock = 0;
 717         __u64 length;
 718         bool wait_flag = false;
 719         struct cifs_sb_info *cifs_sb;
 720         struct cifsTconInfo *tcon;
 721         __u16 netfid;
 722         __u8 lockType = LOCKING_ANDX_LARGE_FILES;
 723         bool posix_locking = 0;
 724
 725         length = 1 + pfLock->fl_end - pfLock->fl_start;
 726         rc = -EACCES;
 727         xid = GetXid();
 728
 729         cFYI(1, "Lock parm: 0x%x flockflags: "
 730                  "0x%x flocktype: 0x%x start: %lld end: %lld",
 731                 cmd, pfLock->fl_flags, pfLock->fl_type, pfLock->fl_start,
 732                 pfLock->fl_end);
 733
 734         if (pfLock->fl_flags & FL_POSIX)
 735                 cFYI(1, "Posix");
 736         if (pfLock->fl_flags & FL_FLOCK)
 737                 cFYI(1, "Flock");
 738         if (pfLock->fl_flags & FL_SLEEP) {
 739                 cFYI(1, "Blocking lock");
 740                 wait_flag = true;
 741         }
 742         if (pfLock->fl_flags & FL_ACCESS)
 743                 cFYI(1, "Process suspended by mandatory locking - "
 744                          "not implemented yet");
 745         if (pfLock->fl_flags & FL_LEASE)
 746                 cFYI(1, "Lease on file - not implemented yet");
 747         if (pfLock->fl_flags &
 748             (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
 749                 cFYI(1, "Unknown lock flags 0x%x", pfLock->fl_flags);
 750
 751         if (pfLock->fl_type == F_WRLCK) {
 752                 cFYI(1, "F_WRLCK ");
 753                 numLock = 1;
 754         } else if (pfLock->fl_type == F_UNLCK) {
 755                 cFYI(1, "F_UNLCK");
 756                 numUnlock = 1;
 757                 /* Check if unlock includes more than
 758                 one lock range */
 759         } else if (pfLock->fl_type == F_RDLCK) {
 760                 cFYI(1, "F_RDLCK");
 761                 lockType |= LOCKING_ANDX_SHARED_LOCK;
 762                 numLock = 1;
 763         } else if (pfLock->fl_type == F_EXLCK) {
 764                 cFYI(1, "F_EXLCK");
 765                 numLock = 1;
 766         } else if (pfLock->fl_type == F_SHLCK) {
 767                 cFYI(1, "F_SHLCK");
 768                 lockType |= LOCKING_ANDX_SHARED_LOCK;
 769                 numLock = 1;
 770         } else
 771                 cFYI(1, "Unknown type of lock");
 772
 773         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
 774         tcon = tlink_tcon(((struct cifsFileInfo *)file->private_data)->tlink);
 775
 776         if (file->private_data == NULL) {
 777                 rc = -EBADF;
 778                 FreeXid(xid);
 779                 return rc;
 780         }
 781         netfid = ((struct cifsFileInfo *)file->private_data)->netfid;
 782
 783         if ((tcon->ses->capabilities & CAP_UNIX) &&
 784             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
 785             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
 786                 posix_locking = 1;
 787         /* BB add code here to normalize offset and length to
 788         account for negative length which we can not accept over the
 789         wire */
 790         if (IS_GETLK(cmd)) {
 791                 if (posix_locking) {
 792                         int posix_lock_type;
 793                         if (lockType & LOCKING_ANDX_SHARED_LOCK)
 794                                 posix_lock_type = CIFS_RDLCK;
 795                         else
 796                                 posix_lock_type = CIFS_WRLCK;
 797                         rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
 798                                         length, pfLock,
 799                                         posix_lock_type, wait_flag);
 800                         FreeXid(xid);
 801                         return rc;
 802                 }
 803
 804                 /* BB we could chain these into one lock request BB */
 805                 rc = CIFSSMBLock(xid, tcon, netfid, length, pfLock->fl_start,
 806                                  0, 1, lockType, 0 /* wait flag */ );
 807                 if (rc == 0) {
 808                         rc = CIFSSMBLock(xid, tcon, netfid, length,
 809                                          pfLock->fl_start, 1 /* numUnlock */ ,
 810                                          0 /* numLock */ , lockType,
 811                                          0 /* wait flag */ );
 812                         pfLock->fl_type = F_UNLCK;
 813                         if (rc != 0)
 814                                 cERROR(1, "Error unlocking previously locked "
 815                                            "range %d during test of lock", rc);
 816                         rc = 0;
 817
 818                 } else {
 819                         /* if rc == ERR_SHARING_VIOLATION ? */
 820                         rc = 0;
 821
 822                         if (lockType & LOCKING_ANDX_SHARED_LOCK) {
 823                                 pfLock->fl_type = F_WRLCK;
 824                         } else {
 825                                 rc = CIFSSMBLock(xid, tcon, netfid, length,
 826                                         pfLock->fl_start, 0, 1,
 827                                         lockType | LOCKING_ANDX_SHARED_LOCK,
 828                                         0 /* wait flag */);
 829                                 if (rc == 0) {
 830                                         rc = CIFSSMBLock(xid, tcon, netfid,
 831                                                 length, pfLock->fl_start, 1, 0,
 832                                                 lockType |
 833                                                 LOCKING_ANDX_SHARED_LOCK,
 834                                                 0 /* wait flag */);
 835                                         pfLock->fl_type = F_RDLCK;
 836                                         if (rc != 0)
 837                                                 cERROR(1, "Error unlocking "
 838                                                 "previously locked range %d "
 839                                                 "during test of lock", rc);
 840                                         rc = 0;
 841                                 } else {
 842                                         pfLock->fl_type = F_WRLCK;
 843                                         rc = 0;
 844                                 }
 845                         }
 846                 }
 847
 848                 FreeXid(xid);
 849                 return rc;
 850         }
 851
 852         if (!numLock && !numUnlock) {
 853                 /* if no lock or unlock then nothing
 854                 to do since we do not know what it is */
 855                 FreeXid(xid);
 856                 return -EOPNOTSUPP;
 857         }
 858
 859         if (posix_locking) {
 860                 int posix_lock_type;
 861                 if (lockType & LOCKING_ANDX_SHARED_LOCK)
 862                         posix_lock_type = CIFS_RDLCK;
 863                 else
 864                         posix_lock_type = CIFS_WRLCK;
 865
 866                 if (numUnlock == 1)
 867                         posix_lock_type = CIFS_UNLCK;
 868
 869                 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */,
 870                                       length, pfLock,
 871                                       posix_lock_type, wait_flag);
 872         } else {
 873                 struct cifsFileInfo *fid = file->private_data;
 874
 875                 if (numLock) {
 876                         rc = CIFSSMBLock(xid, tcon, netfid, length,
 877                                         pfLock->fl_start,
 878                                         0, numLock, lockType, wait_flag);
 879
 880                         if (rc == 0) {
 881                                 /* For Windows locks we must store them. */
 882                                 rc = store_file_lock(fid, length,
 883                                                 pfLock->fl_start, lockType);
 884                         }
 885                 } else if (numUnlock) {
 886                         /* For each stored lock that this unlock overlaps
 887                            completely, unlock it. */
 888                         int stored_rc = 0;
 889                         struct cifsLockInfo *li, *tmp;
 890
 891                         rc = 0;
 892                         mutex_lock(&fid->lock_mutex);
 893                         list_for_each_entry_safe(li, tmp, &fid->llist, llist) {
 894                                 if (pfLock->fl_start <= li->offset &&
 895                                                 (pfLock->fl_start + length) >=
 896                                                 (li->offset + li->length)) {
 897                                         stored_rc = CIFSSMBLock(xid, tcon,
 898                                                         netfid,
 899                                                         li->length, li->offset,
 900                                                         1, 0, li->type, false);
 901                                         if (stored_rc)
 902                                                 rc = stored_rc;
 903                                         else {
 904                                                 list_del(&li->llist);
 905                                                 kfree(li);
 906                                         }
 907                                 }
 908                         }
 909                         mutex_unlock(&fid->lock_mutex);
 910                 }
 911         }
 912
 913         if (pfLock->fl_flags & FL_POSIX)
 914                 posix_lock_file_wait(file, pfLock);
 915         FreeXid(xid);
 916         return rc;
 917 }
 918
 919 /*
 920  * Set the timeout on write requests past EOF. For some servers (Windows)
 921  * these calls can be very long.
 922  *
 923  * If we're writing >10M past the EOF we give a 180s timeout. Anything less
 924  * than that gets a 45s timeout. Writes not past EOF get 15s timeouts.
 925  * The 10M cutoff is totally arbitrary. A better scheme for this would be
 926  * welcome if someone wants to suggest one.
 927  *
 928  * We may be able to do a better job with this if there were some way to
 929  * declare that a file should be sparse.
 930  */
 931 static int
 932 cifs_write_timeout(struct cifsInodeInfo *cifsi, loff_t offset)
 933 {
 934         if (offset <= cifsi->server_eof)
 935                 return CIFS_STD_OP;
 936         else if (offset > (cifsi->server_eof + (10 * 1024 * 1024)))
 937                 return CIFS_VLONG_OP;
 938         else
 939                 return CIFS_LONG_OP;
 940 }
 941
 942 /* update the file size (if needed) after a write */
 943 static void
 944 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
 945                       unsigned int bytes_written)
 946 {
 947         loff_t end_of_write = offset + bytes_written;
 948
 949         if (end_of_write > cifsi->server_eof)
 950                 cifsi->server_eof = end_of_write;
 951 }
 952
 953 ssize_t cifs_user_write(struct file *file, const char __user *write_data,
 954         size_t write_size, loff_t *poffset)
 955 {
 956         int rc = 0;
 957         unsigned int bytes_written = 0;
 958         unsigned int total_written;
 959         struct cifs_sb_info *cifs_sb;
 960         struct cifsTconInfo *pTcon;
 961         int xid, long_op;
 962         struct cifsFileInfo *open_file;
 963         struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
 964
 965         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
 966
 967         /* cFYI(1, " write %d bytes to offset %lld of %s", write_size,
 968            *poffset, file->f_path.dentry->d_name.name); */
 969
 970         if (file->private_data == NULL)
 971                 return -EBADF;
 972
 973         open_file = file->private_data;
 974         pTcon = tlink_tcon(open_file->tlink);
 975
 976         rc = generic_write_checks(file, poffset, &write_size, 0);
 977         if (rc)
 978                 return rc;
 979
 980         xid = GetXid();
 981
 982         long_op = cifs_write_timeout(cifsi, *poffset);
 983         for (total_written = 0; write_size > total_written;
 984              total_written += bytes_written) {
 985                 rc = -EAGAIN;
 986                 while (rc == -EAGAIN) {
 987                         if (file->private_data == NULL) {
 988                                 /* file has been closed on us */
 989                                 FreeXid(xid);
 990                         /* if we have gotten here we have written some data
 991                            and blocked, and the file has been freed on us while
 992                            we blocked so return what we managed to write */
 993                                 return total_written;
 994                         }
 995                         if (open_file->closePend) {
 996                                 FreeXid(xid);
 997                                 if (total_written)
 998                                         return total_written;
 999                                 else
1000                                         return -EBADF;
1001                         }
1002                         if (open_file->invalidHandle) {
1003                                 /* we could deadlock if we called
1004                                    filemap_fdatawait from here so tell
1005                                    reopen_file not to flush data to server
1006                                    now */
1007                                 rc = cifs_reopen_file(file, false);
1008                                 if (rc != 0)
1009                                         break;
1010                         }
1011
1012                         rc = CIFSSMBWrite(xid, pTcon,
1013                                 open_file->netfid,
1014                                 min_t(const int, cifs_sb->wsize,
1015                                       write_size - total_written),
1016                                 *poffset, &bytes_written,
1017                                 NULL, write_data + total_written, long_op);
1018                 }
1019                 if (rc || (bytes_written == 0)) {
1020                         if (total_written)
1021                                 break;
1022                         else {
1023                                 FreeXid(xid);
1024                                 return rc;
1025                         }
1026                 } else {
1027                         cifs_update_eof(cifsi, *poffset, bytes_written);
1028                         *poffset += bytes_written;
1029                 }
1030                 long_op = CIFS_STD_OP; /* subsequent writes fast -
1031                                     15 seconds is plenty */
1032         }
1033
1034         cifs_stats_bytes_written(pTcon, total_written);
1035
1036         /* since the write may have blocked check these pointers again */
1037         if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1038                 struct inode *inode = file->f_path.dentry->d_inode;
1039 /* Do not update local mtime - server will set its actual value on write
1040  *              inode->i_ctime = inode->i_mtime =
1041  *                      current_fs_time(inode->i_sb);*/
1042                 if (total_written > 0) {
1043                         spin_lock(&inode->i_lock);
1044                         if (*poffset > file->f_path.dentry->d_inode->i_size)
1045                                 i_size_write(file->f_path.dentry->d_inode,
1046                                         *poffset);
1047                         spin_unlock(&inode->i_lock);
1048                 }
1049                 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1050         }
1051         FreeXid(xid);
1052         return total_written;
1053 }
1054
1055 static ssize_t cifs_write(struct file *file, const char *write_data,
1056                           size_t write_size, loff_t *poffset)
1057 {
1058         int rc = 0;
1059         unsigned int bytes_written = 0;
1060         unsigned int total_written;
1061         struct cifs_sb_info *cifs_sb;
1062         struct cifsTconInfo *pTcon;
1063         int xid, long_op;
1064         struct cifsFileInfo *open_file;
1065         struct cifsInodeInfo *cifsi = CIFS_I(file->f_path.dentry->d_inode);
1066
1067         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1068
1069         cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
1070            *poffset, file->f_path.dentry->d_name.name);
1071
1072         if (file->private_data == NULL)
1073                 return -EBADF;
1074         open_file = file->private_data;
1075         pTcon = tlink_tcon(open_file->tlink);
1076
1077         xid = GetXid();
1078
1079         long_op = cifs_write_timeout(cifsi, *poffset);
1080         for (total_written = 0; write_size > total_written;
1081              total_written += bytes_written) {
1082                 rc = -EAGAIN;
1083                 while (rc == -EAGAIN) {
1084                         if (file->private_data == NULL) {
1085                                 /* file has been closed on us */
1086                                 FreeXid(xid);
1087                         /* if we have gotten here we have written some data
1088                            and blocked, and the file has been freed on us
1089                            while we blocked so return what we managed to
1090                            write */
1091                                 return total_written;
1092                         }
1093                         if (open_file->closePend) {
1094                                 FreeXid(xid);
1095                                 if (total_written)
1096                                         return total_written;
1097                                 else
1098                                         return -EBADF;
1099                         }
1100                         if (open_file->invalidHandle) {
1101                                 /* we could deadlock if we called
1102                                    filemap_fdatawait from here so tell
1103                                    reopen_file not to flush data to
1104                                    server now */
1105                                 rc = cifs_reopen_file(file, false);
1106                                 if (rc != 0)
1107                                         break;
1108                         }
1109                         if (experimEnabled || (pTcon->ses->server &&
1110                                 ((pTcon->ses->server->secMode &
1111                                 (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED))
1112                                 == 0))) {
1113                                 struct kvec iov[2];
1114                                 unsigned int len;
1115
1116                                 len = min((size_t)cifs_sb->wsize,
1117                                           write_size - total_written);
1118                                 /* iov[0] is reserved for smb header */
1119                                 iov[1].iov_base = (char *)write_data +
1120                                                   total_written;
1121                                 iov[1].iov_len = len;
1122                                 rc = CIFSSMBWrite2(xid, pTcon,
1123                                                 open_file->netfid, len,
1124                                                 *poffset, &bytes_written,
1125                                                 iov, 1, long_op);
1126                         } else
1127                                 rc = CIFSSMBWrite(xid, pTcon,
1128                                          open_file->netfid,
1129                                          min_t(const int, cifs_sb->wsize,
1130                                                write_size - total_written),
1131                                          *poffset, &bytes_written,
1132                                          write_data + total_written,
1133                                          NULL, long_op);
1134                 }
1135                 if (rc || (bytes_written == 0)) {
1136                         if (total_written)
1137                                 break;
1138                         else {
1139                                 FreeXid(xid);
1140                                 return rc;
1141                         }
1142                 } else {
1143                         cifs_update_eof(cifsi, *poffset, bytes_written);
1144                         *poffset += bytes_written;
1145                 }
1146                 long_op = CIFS_STD_OP; /* subsequent writes fast -
1147                                     15 seconds is plenty */
1148         }
1149
1150         cifs_stats_bytes_written(pTcon, total_written);
1151
1152         /* since the write may have blocked check these pointers again */
1153         if ((file->f_path.dentry) && (file->f_path.dentry->d_inode)) {
1154 /*BB We could make this contingent on superblock ATIME flag too */
1155 /*              file->f_path.dentry->d_inode->i_ctime =
1156                 file->f_path.dentry->d_inode->i_mtime = CURRENT_TIME;*/
1157                 if (total_written > 0) {
1158                         spin_lock(&file->f_path.dentry->d_inode->i_lock);
1159                         if (*poffset > file->f_path.dentry->d_inode->i_size)
1160                                 i_size_write(file->f_path.dentry->d_inode,
1161                                              *poffset);
1162                         spin_unlock(&file->f_path.dentry->d_inode->i_lock);
1163                 }
1164                 mark_inode_dirty_sync(file->f_path.dentry->d_inode);
1165         }
1166         FreeXid(xid);
1167         return total_written;
1168 }
1169
1170 #ifdef CONFIG_CIFS_EXPERIMENTAL
1171 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1172                                         bool fsuid_only)
1173 {
1174         struct cifsFileInfo *open_file = NULL;
1175         struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1176
1177         /* only filter by fsuid on multiuser mounts */
1178         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1179                 fsuid_only = false;
1180
1181         read_lock(&GlobalSMBSeslock);
1182         /* we could simply get the first_list_entry since write-only entries
1183            are always at the end of the list but since the first entry might
1184            have a close pending, we go through the whole list */
1185         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1186                 if (open_file->closePend)
1187                         continue;
1188                 if (fsuid_only && open_file->uid != current_fsuid())
1189                         continue;
1190                 if (open_file->pfile && ((open_file->pfile->f_flags & O_RDWR) ||
1191                     (open_file->pfile->f_flags & O_RDONLY))) {
1192                         if (!open_file->invalidHandle) {
1193                                 /* found a good file */
1194                                 /* lock it so it will not be closed on us */
1195                                 cifsFileInfo_get(open_file);
1196                                 read_unlock(&GlobalSMBSeslock);
1197                                 return open_file;
1198                         } /* else might as well continue, and look for
1199                              another, or simply have the caller reopen it
1200                              again rather than trying to fix this handle */
1201                 } else /* write only file */
1202                         break; /* write only files are last so must be done */
1203         }
1204         read_unlock(&GlobalSMBSeslock);
1205         return NULL;
1206 }
1207 #endif
1208
1209 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1210                                         bool fsuid_only)
1211 {
1212         struct cifsFileInfo *open_file;
1213         struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1214         bool any_available = false;
1215         int rc;
1216
1217         /* Having a null inode here (because mapping->host was set to zero by
1218         the VFS or MM) should not happen but we had reports of on oops (due to
1219         it being zero) during stress testcases so we need to check for it */
1220
1221         if (cifs_inode == NULL) {
1222                 cERROR(1, "Null inode passed to cifs_writeable_file");
1223                 dump_stack();
1224                 return NULL;
1225         }
1226
1227         /* only filter by fsuid on multiuser mounts */
1228         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1229                 fsuid_only = false;
1230
1231         read_lock(&GlobalSMBSeslock);
1232 refind_writable:
1233         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1234                 if (open_file->closePend)
1235                         continue;
1236                 if (!any_available && open_file->pid != current->tgid)
1237                         continue;
1238                 if (fsuid_only && open_file->uid != current_fsuid())
1239                         continue;
1240                 if (open_file->pfile &&
1241                     ((open_file->pfile->f_flags & O_RDWR) ||
1242                      (open_file->pfile->f_flags & O_WRONLY))) {
1243                         cifsFileInfo_get(open_file);
1244
1245                         if (!open_file->invalidHandle) {
1246                                 /* found a good writable file */
1247                                 read_unlock(&GlobalSMBSeslock);
1248                                 return open_file;
1249                         }
1250
1251                         read_unlock(&GlobalSMBSeslock);
1252                         /* Had to unlock since following call can block */
1253                         rc = cifs_reopen_file(open_file->pfile, false);
1254                         if (!rc) {
1255                                 if (!open_file->closePend)
1256                                         return open_file;
1257                                 else { /* start over in case this was deleted */
1258                                        /* since the list could be modified */
1259                                         read_lock(&GlobalSMBSeslock);
1260                                         cifsFileInfo_put(open_file);
1261                                         goto refind_writable;
1262                                 }
1263                         }
1264
1265                         /* if it fails, try another handle if possible -
1266                         (we can not do this if closePending since
1267                         loop could be modified - in which case we
1268                         have to start at the beginning of the list
1269                         again. Note that it would be bad
1270                         to hold up writepages here (rather than
1271                         in caller) with continuous retries */
1272                         cFYI(1, "wp failed on reopen file");
1273                         read_lock(&GlobalSMBSeslock);
1274                         /* can not use this handle, no write
1275                            pending on this one after all */
1276                         cifsFileInfo_put(open_file);
1277
1278                         if (open_file->closePend) /* list could have changed */
1279                                 goto refind_writable;
1280                         /* else we simply continue to the next entry. Thus
1281                            we do not loop on reopen errors.  If we
1282                            can not reopen the file, for example if we
1283                            reconnected to a server with another client
1284                            racing to delete or lock the file we would not
1285                            make progress if we restarted before the beginning
1286                            of the loop here. */
1287                 }
1288         }
1289         /* couldn't find useable FH with same pid, try any available */
1290         if (!any_available) {
1291                 any_available = true;
1292                 goto refind_writable;
1293         }
1294         read_unlock(&GlobalSMBSeslock);
1295         return NULL;
1296 }
1297
1298 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1299 {
1300         struct address_space *mapping = page->mapping;
1301         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1302         char *write_data;
1303         int rc = -EFAULT;
1304         int bytes_written = 0;
1305         struct cifs_sb_info *cifs_sb;
1306         struct inode *inode;
1307         struct cifsFileInfo *open_file;
1308
1309         if (!mapping || !mapping->host)
1310                 return -EFAULT;
1311
1312         inode = page->mapping->host;
1313         cifs_sb = CIFS_SB(inode->i_sb);
1314
1315         offset += (loff_t)from;
1316         write_data = kmap(page);
1317         write_data += from;
1318
1319         if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1320                 kunmap(page);
1321                 return -EIO;
1322         }
1323
1324         /* racing with truncate? */
1325         if (offset > mapping->host->i_size) {
1326                 kunmap(page);
1327                 return 0; /* don't care */
1328         }
1329
1330         /* check to make sure that we are not extending the file */
1331         if (mapping->host->i_size - offset < (loff_t)to)
1332                 to = (unsigned)(mapping->host->i_size - offset);
1333
1334         open_file = find_writable_file(CIFS_I(mapping->host), false);
1335         if (open_file) {
1336                 bytes_written = cifs_write(open_file->pfile, write_data,
1337                                            to-from, &offset);
1338                 cifsFileInfo_put(open_file);
1339                 /* Does mm or vfs already set times? */
1340                 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1341                 if ((bytes_written > 0) && (offset))
1342                         rc = 0;
1343                 else if (bytes_written < 0)
1344                         rc = bytes_written;
1345         } else {
1346                 cFYI(1, "No writeable filehandles for inode");
1347                 rc = -EIO;
1348         }
1349
1350         kunmap(page);
1351         return rc;
1352 }
1353
1354 static int cifs_writepages(struct address_space *mapping,
1355                            struct writeback_control *wbc)
1356 {
1357         struct backing_dev_info *bdi = mapping->backing_dev_info;
1358         unsigned int bytes_to_write;
1359         unsigned int bytes_written;
1360         struct cifs_sb_info *cifs_sb;
1361         int done = 0;
1362         pgoff_t end;
1363         pgoff_t index;
1364         int range_whole = 0;
1365         struct kvec *iov;
1366         int len;
1367         int n_iov = 0;
1368         pgoff_t next;
1369         int nr_pages;
1370         __u64 offset = 0;
1371         struct cifsFileInfo *open_file;
1372         struct cifsTconInfo *tcon;
1373         struct cifsInodeInfo *cifsi = CIFS_I(mapping->host);
1374         struct page *page;
1375         struct pagevec pvec;
1376         int rc = 0;
1377         int scanned = 0;
1378         int xid, long_op;
1379
1380         /*
1381          * BB: Is this meaningful for a non-block-device file system?
1382          * If it is, we should test it again after we do I/O
1383          */
1384         if (wbc->nonblocking && bdi_write_congested(bdi)) {
1385                 wbc->encountered_congestion = 1;
1386                 return 0;
1387         }
1388
1389         cifs_sb = CIFS_SB(mapping->host->i_sb);
1390
1391         /*
1392          * If wsize is smaller that the page cache size, default to writing
1393          * one page at a time via cifs_writepage
1394          */
1395         if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1396                 return generic_writepages(mapping, wbc);
1397
1398         iov = kmalloc(32 * sizeof(struct kvec), GFP_KERNEL);
1399         if (iov == NULL)
1400                 return generic_writepages(mapping, wbc);
1401
1402         /*
1403          * if there's no open file, then this is likely to fail too,
1404          * but it'll at least handle the return. Maybe it should be
1405          * a BUG() instead?
1406          */
1407         open_file = find_writable_file(CIFS_I(mapping->host), false);
1408         if (!open_file) {
1409                 kfree(iov);
1410                 return generic_writepages(mapping, wbc);
1411         }
1412
1413         tcon = tlink_tcon(open_file->tlink);
1414         if (!experimEnabled && tcon->ses->server->secMode &
1415                         (SECMODE_SIGN_REQUIRED | SECMODE_SIGN_ENABLED)) {
1416                 cifsFileInfo_put(open_file);
1417                 return generic_writepages(mapping, wbc);
1418         }
1419         cifsFileInfo_put(open_file);
1420
1421         xid = GetXid();
1422
1423         pagevec_init(&pvec, 0);
1424         if (wbc->range_cyclic) {
1425                 index = mapping->writeback_index; /* Start from prev offset */
1426                 end = -1;
1427         } else {
1428                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1429                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1430                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1431                         range_whole = 1;
1432                 scanned = 1;
1433         }
1434 retry:
1435         while (!done && (index <= end) &&
1436                (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1437                         PAGECACHE_TAG_DIRTY,
1438                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1))) {
1439                 int first;
1440                 unsigned int i;
1441
1442                 first = -1;
1443                 next = 0;
1444                 n_iov = 0;
1445                 bytes_to_write = 0;
1446
1447                 for (i = 0; i < nr_pages; i++) {
1448                         page = pvec.pages[i];
1449                         /*
1450                          * At this point we hold neither mapping->tree_lock nor
1451                          * lock on the page itself: the page may be truncated or
1452                          * invalidated (changing page->mapping to NULL), or even
1453                          * swizzled back from swapper_space to tmpfs file
1454                          * mapping
1455                          */
1456
1457                         if (first < 0)
1458                                 lock_page(page);
1459                         else if (!trylock_page(page))
1460                                 break;
1461
1462                         if (unlikely(page->mapping != mapping)) {
1463                                 unlock_page(page);
1464                                 break;
1465                         }
1466
1467                         if (!wbc->range_cyclic && page->index > end) {
1468                                 done = 1;
1469                                 unlock_page(page);
1470                                 break;
1471                         }
1472
1473                         if (next && (page->index != next)) {
1474                                 /* Not next consecutive page */
1475                                 unlock_page(page);
1476                                 break;
1477                         }
1478
1479                         if (wbc->sync_mode != WB_SYNC_NONE)
1480                                 wait_on_page_writeback(page);
1481
1482                         if (PageWriteback(page) ||
1483                                         !clear_page_dirty_for_io(page)) {
1484                                 unlock_page(page);
1485                                 break;
1486                         }
1487
1488                         /*
1489                          * This actually clears the dirty bit in the radix tree.
1490                          * See cifs_writepage() for more commentary.
1491                          */
1492                         set_page_writeback(page);
1493
1494                         if (page_offset(page) >= mapping->host->i_size) {
1495                                 done = 1;
1496                                 unlock_page(page);
1497                                 end_page_writeback(page);
1498                                 break;
1499                         }
1500
1501                         /*
1502                          * BB can we get rid of this?  pages are held by pvec
1503                          */
1504                         page_cache_get(page);
1505
1506                         len = min(mapping->host->i_size - page_offset(page),
1507                                   (loff_t)PAGE_CACHE_SIZE);
1508
1509                         /* reserve iov[0] for the smb header */
1510                         n_iov++;
1511                         iov[n_iov].iov_base = kmap(page);
1512                         iov[n_iov].iov_len = len;
1513                         bytes_to_write += len;
1514
1515                         if (first < 0) {
1516                                 first = i;
1517                                 offset = page_offset(page);
1518                         }
1519                         next = page->index + 1;
1520                         if (bytes_to_write + PAGE_CACHE_SIZE > cifs_sb->wsize)
1521                                 break;
1522                 }
1523                 if (n_iov) {
1524                         open_file = find_writable_file(CIFS_I(mapping->host),
1525                                                         false);
1526                         if (!open_file) {
1527                                 cERROR(1, "No writable handles for inode");
1528                                 rc = -EBADF;
1529                         } else {
1530                                 long_op = cifs_write_timeout(cifsi, offset);
1531                                 rc = CIFSSMBWrite2(xid, tcon, open_file->netfid,
1532                                                    bytes_to_write, offset,
1533                                                    &bytes_written, iov, n_iov,
1534                                                    long_op);
1535                                 cifsFileInfo_put(open_file);
1536                                 cifs_update_eof(cifsi, offset, bytes_written);
1537                         }
1538
1539                         if (rc || bytes_written < bytes_to_write) {
1540                                 cERROR(1, "Write2 ret %d, wrote %d",
1541                                           rc, bytes_written);
1542                                 /* BB what if continued retry is
1543                                    requested via mount flags? */
1544                                 if (rc == -ENOSPC)
1545                                         set_bit(AS_ENOSPC, &mapping->flags);
1546                                 else
1547                                         set_bit(AS_EIO, &mapping->flags);
1548                         } else {
1549                                 cifs_stats_bytes_written(tcon, bytes_written);
1550                         }
1551
1552                         for (i = 0; i < n_iov; i++) {
1553                                 page = pvec.pages[first + i];
1554                                 /* Should we also set page error on
1555                                 success rc but too little data written? */
1556                                 /* BB investigate retry logic on temporary
1557                                 server crash cases and how recovery works
1558                                 when page marked as error */
1559                                 if (rc)
1560                                         SetPageError(page);
1561                                 kunmap(page);
1562                                 unlock_page(page);
1563                                 end_page_writeback(page);
1564                                 page_cache_release(page);
1565                         }
1566                         if ((wbc->nr_to_write -= n_iov) <= 0)
1567                                 done = 1;
1568                         index = next;
1569                 } else
1570                         /* Need to re-find the pages we skipped */
1571                         index = pvec.pages[0]->index + 1;
1572
1573                 pagevec_release(&pvec);
1574         }
1575         if (!scanned && !done) {
1576                 /*
1577                  * We hit the last page and there is more work to be done: wrap
1578                  * back to the start of the file
1579                  */
1580                 scanned = 1;
1581                 index = 0;
1582                 goto retry;
1583         }
1584         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1585                 mapping->writeback_index = index;
1586
1587         FreeXid(xid);
1588         kfree(iov);
1589         return rc;
1590 }
1591
1592 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1593 {
1594         int rc = -EFAULT;
1595         int xid;
1596
1597         xid = GetXid();
1598 /* BB add check for wbc flags */
1599         page_cache_get(page);
1600         if (!PageUptodate(page))
1601                 cFYI(1, "ppw - page not up to date");
1602
1603         /*
1604          * Set the "writeback" flag, and clear "dirty" in the radix tree.
1605          *
1606          * A writepage() implementation always needs to do either this,
1607          * or re-dirty the page with "redirty_page_for_writepage()" in
1608          * the case of a failure.
1609          *
1610          * Just unlocking the page will cause the radix tree tag-bits
1611          * to fail to update with the state of the page correctly.
1612          */
1613         set_page_writeback(page);
1614         rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1615         SetPageUptodate(page); /* BB add check for error and Clearuptodate? */
1616         unlock_page(page);
1617         end_page_writeback(page);
1618         page_cache_release(page);
1619         FreeXid(xid);
1620         return rc;
1621 }
1622
1623 static int cifs_write_end(struct file *file, struct address_space *mapping,
1624                         loff_t pos, unsigned len, unsigned copied,
1625                         struct page *page, void *fsdata)
1626 {
1627         int rc;
1628         struct inode *inode = mapping->host;
1629
1630         cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1631                  page, pos, copied);
1632
1633         if (PageChecked(page)) {
1634                 if (copied == len)
1635                         SetPageUptodate(page);
1636                 ClearPageChecked(page);
1637         } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1638                 SetPageUptodate(page);
1639
1640         if (!PageUptodate(page)) {
1641                 char *page_data;
1642                 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1643                 int xid;
1644
1645                 xid = GetXid();
1646                 /* this is probably better than directly calling
1647                    partialpage_write since in this function the file handle is
1648                    known which we might as well leverage */
1649                 /* BB check if anything else missing out of ppw
1650                    such as updating last write time */
1651                 page_data = kmap(page);
1652                 rc = cifs_write(file, page_data + offset, copied, &pos);
1653                 /* if (rc < 0) should we set writebehind rc? */
1654                 kunmap(page);
1655
1656                 FreeXid(xid);
1657         } else {
1658                 rc = copied;
1659                 pos += copied;
1660                 set_page_dirty(page);
1661         }
1662
1663         if (rc > 0) {
1664                 spin_lock(&inode->i_lock);
1665                 if (pos > inode->i_size)
1666                         i_size_write(inode, pos);
1667                 spin_unlock(&inode->i_lock);
1668         }
1669
1670         unlock_page(page);
1671         page_cache_release(page);
1672
1673         return rc;
1674 }
1675
1676 int cifs_fsync(struct file *file, int datasync)
1677 {
1678         int xid;
1679         int rc = 0;
1680         struct cifsTconInfo *tcon;
1681         struct cifsFileInfo *smbfile = file->private_data;
1682         struct inode *inode = file->f_path.dentry->d_inode;
1683
1684         xid = GetXid();
1685
1686         cFYI(1, "Sync file - name: %s datasync: 0x%x",
1687                 file->f_path.dentry->d_name.name, datasync);
1688
1689         rc = filemap_write_and_wait(inode->i_mapping);
1690         if (rc == 0) {
1691                 rc = CIFS_I(inode)->write_behind_rc;
1692                 CIFS_I(inode)->write_behind_rc = 0;
1693                 tcon = tlink_tcon(smbfile->tlink);
1694                 if (!rc && tcon && smbfile &&
1695                    !(CIFS_SB(inode->i_sb)->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1696                         rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1697         }
1698
1699         FreeXid(xid);
1700         return rc;
1701 }
1702
1703 /* static void cifs_sync_page(struct page *page)
1704 {
1705         struct address_space *mapping;
1706         struct inode *inode;
1707         unsigned long index = page->index;
1708         unsigned int rpages = 0;
1709         int rc = 0;
1710
1711         cFYI(1, "sync page %p", page);
1712         mapping = page->mapping;
1713         if (!mapping)
1714                 return 0;
1715         inode = mapping->host;
1716         if (!inode)
1717                 return; */
1718
1719 /*      fill in rpages then
1720         result = cifs_pagein_inode(inode, index, rpages); */ /* BB finish */
1721
1722 /*      cFYI(1, "rpages is %d for sync page of Index %ld", rpages, index);
1723
1724 #if 0
1725         if (rc < 0)
1726                 return rc;
1727         return 0;
1728 #endif
1729 } */
1730
1731 /*
1732  * As file closes, flush all cached write data for this inode checking
1733  * for write behind errors.
1734  */
1735 int cifs_flush(struct file *file, fl_owner_t id)
1736 {
1737         struct inode *inode = file->f_path.dentry->d_inode;
1738         int rc = 0;
1739
1740         /* Rather than do the steps manually:
1741            lock the inode for writing
1742            loop through pages looking for write behind data (dirty pages)
1743            coalesce into contiguous 16K (or smaller) chunks to write to server
1744            send to server (prefer in parallel)
1745            deal with writebehind errors
1746            unlock inode for writing
1747            filemapfdatawrite appears easier for the time being */
1748
1749         rc = filemap_fdatawrite(inode->i_mapping);
1750         /* reset wb rc if we were able to write out dirty pages */
1751         if (!rc) {
1752                 rc = CIFS_I(inode)->write_behind_rc;
1753                 CIFS_I(inode)->write_behind_rc = 0;
1754         }
1755
1756         cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1757
1758         return rc;
1759 }
1760
1761 ssize_t cifs_user_read(struct file *file, char __user *read_data,
1762         size_t read_size, loff_t *poffset)
1763 {
1764         int rc = -EACCES;
1765         unsigned int bytes_read = 0;
1766         unsigned int total_read = 0;
1767         unsigned int current_read_size;
1768         struct cifs_sb_info *cifs_sb;
1769         struct cifsTconInfo *pTcon;
1770         int xid;
1771         struct cifsFileInfo *open_file;
1772         char *smb_read_data;
1773         char __user *current_offset;
1774         struct smb_com_read_rsp *pSMBr;
1775
1776         xid = GetXid();
1777         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1778
1779         if (file->private_data == NULL) {
1780                 rc = -EBADF;
1781                 FreeXid(xid);
1782                 return rc;
1783         }
1784         open_file = file->private_data;
1785         pTcon = tlink_tcon(open_file->tlink);
1786
1787         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1788                 cFYI(1, "attempting read on write only file instance");
1789
1790         for (total_read = 0, current_offset = read_data;
1791              read_size > total_read;
1792              total_read += bytes_read, current_offset += bytes_read) {
1793                 current_read_size = min_t(const int, read_size - total_read,
1794                                           cifs_sb->rsize);
1795                 rc = -EAGAIN;
1796                 smb_read_data = NULL;
1797                 while (rc == -EAGAIN) {
1798                         int buf_type = CIFS_NO_BUFFER;
1799                         if ((open_file->invalidHandle) &&
1800                             (!open_file->closePend)) {
1801                                 rc = cifs_reopen_file(file, true);
1802                                 if (rc != 0)
1803                                         break;
1804                         }
1805                         rc = CIFSSMBRead(xid, pTcon,
1806                                          open_file->netfid,
1807                                          current_read_size, *poffset,
1808                                          &bytes_read, &smb_read_data,
1809                                          &buf_type);
1810                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
1811                         if (smb_read_data) {
1812                                 if (copy_to_user(current_offset,
1813                                                 smb_read_data +
1814                                                 4 /* RFC1001 length field */ +
1815                                                 le16_to_cpu(pSMBr->DataOffset),
1816                                                 bytes_read))
1817                                         rc = -EFAULT;
1818
1819                                 if (buf_type == CIFS_SMALL_BUFFER)
1820                                         cifs_small_buf_release(smb_read_data);
1821                                 else if (buf_type == CIFS_LARGE_BUFFER)
1822                                         cifs_buf_release(smb_read_data);
1823                                 smb_read_data = NULL;
1824                         }
1825                 }
1826                 if (rc || (bytes_read == 0)) {
1827                         if (total_read) {
1828                                 break;
1829                         } else {
1830                                 FreeXid(xid);
1831                                 return rc;
1832                         }
1833                 } else {
1834                         cifs_stats_bytes_read(pTcon, bytes_read);
1835                         *poffset += bytes_read;
1836                 }
1837         }
1838         FreeXid(xid);
1839         return total_read;
1840 }
1841
1842
1843 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1844         loff_t *poffset)
1845 {
1846         int rc = -EACCES;
1847         unsigned int bytes_read = 0;
1848         unsigned int total_read;
1849         unsigned int current_read_size;
1850         struct cifs_sb_info *cifs_sb;
1851         struct cifsTconInfo *pTcon;
1852         int xid;
1853         char *current_offset;
1854         struct cifsFileInfo *open_file;
1855         int buf_type = CIFS_NO_BUFFER;
1856
1857         xid = GetXid();
1858         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1859
1860         if (file->private_data == NULL) {
1861                 rc = -EBADF;
1862                 FreeXid(xid);
1863                 return rc;
1864         }
1865         open_file = file->private_data;
1866         pTcon = tlink_tcon(open_file->tlink);
1867
1868         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1869                 cFYI(1, "attempting read on write only file instance");
1870
1871         for (total_read = 0, current_offset = read_data;
1872              read_size > total_read;
1873              total_read += bytes_read, current_offset += bytes_read) {
1874                 current_read_size = min_t(const int, read_size - total_read,
1875                                           cifs_sb->rsize);
1876                 /* For windows me and 9x we do not want to request more
1877                 than it negotiated since it will refuse the read then */
1878                 if ((pTcon->ses) &&
1879                         !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1880                         current_read_size = min_t(const int, current_read_size,
1881                                         pTcon->ses->server->maxBuf - 128);
1882                 }
1883                 rc = -EAGAIN;
1884                 while (rc == -EAGAIN) {
1885                         if ((open_file->invalidHandle) &&
1886                             (!open_file->closePend)) {
1887                                 rc = cifs_reopen_file(file, true);
1888                                 if (rc != 0)
1889                                         break;
1890                         }
1891                         rc = CIFSSMBRead(xid, pTcon,
1892                                          open_file->netfid,
1893                                          current_read_size, *poffset,
1894                                          &bytes_read, &current_offset,
1895                                          &buf_type);
1896                 }
1897                 if (rc || (bytes_read == 0)) {
1898                         if (total_read) {
1899                                 break;
1900                         } else {
1901                                 FreeXid(xid);
1902                                 return rc;
1903                         }
1904                 } else {
1905                         cifs_stats_bytes_read(pTcon, total_read);
1906                         *poffset += bytes_read;
1907                 }
1908         }
1909         FreeXid(xid);
1910         return total_read;
1911 }
1912
1913 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1914 {
1915         int rc, xid;
1916
1917         xid = GetXid();
1918         rc = cifs_revalidate_file(file);
1919         if (rc) {
1920                 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1921                 FreeXid(xid);
1922                 return rc;
1923         }
1924         rc = generic_file_mmap(file, vma);
1925         FreeXid(xid);
1926         return rc;
1927 }
1928
1929
1930 static void cifs_copy_cache_pages(struct address_space *mapping,
1931         struct list_head *pages, int bytes_read, char *data)
1932 {
1933         struct page *page;
1934         char *target;
1935
1936         while (bytes_read > 0) {
1937                 if (list_empty(pages))
1938                         break;
1939
1940                 page = list_entry(pages->prev, struct page, lru);
1941                 list_del(&page->lru);
1942
1943                 if (add_to_page_cache_lru(page, mapping, page->index,
1944                                       GFP_KERNEL)) {
1945                         page_cache_release(page);
1946                         cFYI(1, "Add page cache failed");
1947                         data += PAGE_CACHE_SIZE;
1948                         bytes_read -= PAGE_CACHE_SIZE;
1949                         continue;
1950                 }
1951                 page_cache_release(page);
1952
1953                 target = kmap_atomic(page, KM_USER0);
1954
1955                 if (PAGE_CACHE_SIZE > bytes_read) {
1956                         memcpy(target, data, bytes_read);
1957                         /* zero the tail end of this partial page */
1958                         memset(target + bytes_read, 0,
1959                                PAGE_CACHE_SIZE - bytes_read);
1960                         bytes_read = 0;
1961                 } else {
1962                         memcpy(target, data, PAGE_CACHE_SIZE);
1963                         bytes_read -= PAGE_CACHE_SIZE;
1964                 }
1965                 kunmap_atomic(target, KM_USER0);
1966
1967                 flush_dcache_page(page);
1968                 SetPageUptodate(page);
1969                 unlock_page(page);
1970                 data += PAGE_CACHE_SIZE;
1971
1972                 /* add page to FS-Cache */
1973                 cifs_readpage_to_fscache(mapping->host, page);
1974         }
1975         return;
1976 }
1977
1978 static int cifs_readpages(struct file *file, struct address_space *mapping,
1979         struct list_head *page_list, unsigned num_pages)
1980 {
1981         int rc = -EACCES;
1982         int xid;
1983         loff_t offset;
1984         struct page *page;
1985         struct cifs_sb_info *cifs_sb;
1986         struct cifsTconInfo *pTcon;
1987         unsigned int bytes_read = 0;
1988         unsigned int read_size, i;
1989         char *smb_read_data = NULL;
1990         struct smb_com_read_rsp *pSMBr;
1991         struct cifsFileInfo *open_file;
1992         int buf_type = CIFS_NO_BUFFER;
1993
1994         xid = GetXid();
1995         if (file->private_data == NULL) {
1996                 rc = -EBADF;
1997                 FreeXid(xid);
1998                 return rc;
1999         }
2000         open_file = file->private_data;
2001         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2002         pTcon = tlink_tcon(open_file->tlink);
2003
2004         /*
2005          * Reads as many pages as possible from fscache. Returns -ENOBUFS
2006          * immediately if the cookie is negative
2007          */
2008         rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
2009                                          &num_pages);
2010         if (rc == 0)
2011                 goto read_complete;
2012
2013         cFYI(DBG2, "rpages: num pages %d", num_pages);
2014         for (i = 0; i < num_pages; ) {
2015                 unsigned contig_pages;
2016                 struct page *tmp_page;
2017                 unsigned long expected_index;
2018
2019                 if (list_empty(page_list))
2020                         break;
2021
2022                 page = list_entry(page_list->prev, struct page, lru);
2023                 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2024
2025                 /* count adjacent pages that we will read into */
2026                 contig_pages = 0;
2027                 expected_index =
2028                         list_entry(page_list->prev, struct page, lru)->index;
2029                 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2030                         if (tmp_page->index == expected_index) {
2031                                 contig_pages++;
2032                                 expected_index++;
2033                         } else
2034                                 break;
2035                 }
2036                 if (contig_pages + i >  num_pages)
2037                         contig_pages = num_pages - i;
2038
2039                 /* for reads over a certain size could initiate async
2040                    read ahead */
2041
2042                 read_size = contig_pages * PAGE_CACHE_SIZE;
2043                 /* Read size needs to be in multiples of one page */
2044                 read_size = min_t(const unsigned int, read_size,
2045                                   cifs_sb->rsize & PAGE_CACHE_MASK);
2046                 cFYI(DBG2, "rpages: read size 0x%x  contiguous pages %d",
2047                                 read_size, contig_pages);
2048                 rc = -EAGAIN;
2049                 while (rc == -EAGAIN) {
2050                         if ((open_file->invalidHandle) &&
2051                             (!open_file->closePend)) {
2052                                 rc = cifs_reopen_file(file, true);
2053                                 if (rc != 0)
2054                                         break;
2055                         }
2056
2057                         rc = CIFSSMBRead(xid, pTcon,
2058                                          open_file->netfid,
2059                                          read_size, offset,
2060                                          &bytes_read, &smb_read_data,
2061                                          &buf_type);
2062                         /* BB more RC checks ? */
2063                         if (rc == -EAGAIN) {
2064                                 if (smb_read_data) {
2065                                         if (buf_type == CIFS_SMALL_BUFFER)
2066                                                 cifs_small_buf_release(smb_read_data);
2067                                         else if (buf_type == CIFS_LARGE_BUFFER)
2068                                                 cifs_buf_release(smb_read_data);
2069                                         smb_read_data = NULL;
2070                                 }
2071                         }
2072                 }
2073                 if ((rc < 0) || (smb_read_data == NULL)) {
2074                         cFYI(1, "Read error in readpages: %d", rc);
2075                         break;
2076                 } else if (bytes_read > 0) {
2077                         task_io_account_read(bytes_read);
2078                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2079                         cifs_copy_cache_pages(mapping, page_list, bytes_read,
2080                                 smb_read_data + 4 /* RFC1001 hdr */ +
2081                                 le16_to_cpu(pSMBr->DataOffset));
2082
2083                         i +=  bytes_read >> PAGE_CACHE_SHIFT;
2084                         cifs_stats_bytes_read(pTcon, bytes_read);
2085                         if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2086                                 i++; /* account for partial page */
2087
2088                                 /* server copy of file can have smaller size
2089                                    than client */
2090                                 /* BB do we need to verify this common case ?
2091                                    this case is ok - if we are at server EOF
2092                                    we will hit it on next read */
2093
2094                                 /* break; */
2095                         }
2096                 } else {
2097                         cFYI(1, "No bytes read (%d) at offset %lld . "
2098                                 "Cleaning remaining pages from readahead list",
2099                                 bytes_read, offset);
2100                         /* BB turn off caching and do new lookup on
2101                            file size at server? */
2102                         break;
2103                 }
2104                 if (smb_read_data) {
2105                         if (buf_type == CIFS_SMALL_BUFFER)
2106                                 cifs_small_buf_release(smb_read_data);
2107                         else if (buf_type == CIFS_LARGE_BUFFER)
2108                                 cifs_buf_release(smb_read_data);
2109                         smb_read_data = NULL;
2110                 }
2111                 bytes_read = 0;
2112         }
2113
2114 /* need to free smb_read_data buf before exit */
2115         if (smb_read_data) {
2116                 if (buf_type == CIFS_SMALL_BUFFER)
2117                         cifs_small_buf_release(smb_read_data);
2118                 else if (buf_type == CIFS_LARGE_BUFFER)
2119                         cifs_buf_release(smb_read_data);
2120                 smb_read_data = NULL;
2121         }
2122
2123 read_complete:
2124         FreeXid(xid);
2125         return rc;
2126 }
2127
2128 static int cifs_readpage_worker(struct file *file, struct page *page,
2129         loff_t *poffset)
2130 {
2131         char *read_data;
2132         int rc;
2133
2134         /* Is the page cached? */
2135         rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2136         if (rc == 0)
2137                 goto read_complete;
2138
2139         page_cache_get(page);
2140         read_data = kmap(page);
2141         /* for reads over a certain size could initiate async read ahead */
2142
2143         rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2144
2145         if (rc < 0)
2146                 goto io_error;
2147         else
2148                 cFYI(1, "Bytes read %d", rc);
2149
2150         file->f_path.dentry->d_inode->i_atime =
2151                 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2152
2153         if (PAGE_CACHE_SIZE > rc)
2154                 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2155
2156         flush_dcache_page(page);
2157         SetPageUptodate(page);
2158
2159         /* send this page to the cache */
2160         cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2161
2162         rc = 0;
2163
2164 io_error:
2165         kunmap(page);
2166         page_cache_release(page);
2167
2168 read_complete:
2169         return rc;
2170 }
2171
2172 static int cifs_readpage(struct file *file, struct page *page)
2173 {
2174         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2175         int rc = -EACCES;
2176         int xid;
2177
2178         xid = GetXid();
2179
2180         if (file->private_data == NULL) {
2181                 rc = -EBADF;
2182                 FreeXid(xid);
2183                 return rc;
2184         }
2185
2186         cFYI(1, "readpage %p at offset %d 0x%x\n",
2187                  page, (int)offset, (int)offset);
2188
2189         rc = cifs_readpage_worker(file, page, &offset);
2190
2191         unlock_page(page);
2192
2193         FreeXid(xid);
2194         return rc;
2195 }
2196
2197 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2198 {
2199         struct cifsFileInfo *open_file;
2200
2201         read_lock(&GlobalSMBSeslock);
2202         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2203                 if (open_file->closePend)
2204                         continue;
2205                 if (open_file->pfile &&
2206                     ((open_file->pfile->f_flags & O_RDWR) ||
2207                      (open_file->pfile->f_flags & O_WRONLY))) {
2208                         read_unlock(&GlobalSMBSeslock);
2209                         return 1;
2210                 }
2211         }
2212         read_unlock(&GlobalSMBSeslock);
2213         return 0;
2214 }
2215
2216 /* We do not want to update the file size from server for inodes
2217    open for write - to avoid races with writepage extending
2218    the file - in the future we could consider allowing
2219    refreshing the inode only on increases in the file size
2220    but this is tricky to do without racing with writebehind
2221    page caching in the current Linux kernel design */
2222 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2223 {
2224         if (!cifsInode)
2225                 return true;
2226
2227         if (is_inode_writable(cifsInode)) {
2228                 /* This inode is open for write at least once */
2229                 struct cifs_sb_info *cifs_sb;
2230
2231                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2232                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2233                         /* since no page cache to corrupt on directio
2234                         we can change size safely */
2235                         return true;
2236                 }
2237
2238                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2239                         return true;
2240
2241                 return false;
2242         } else
2243                 return true;
2244 }
2245
2246 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2247                         loff_t pos, unsigned len, unsigned flags,
2248                         struct page **pagep, void **fsdata)
2249 {
2250         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2251         loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2252         loff_t page_start = pos & PAGE_MASK;
2253         loff_t i_size;
2254         struct page *page;
2255         int rc = 0;
2256
2257         cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2258
2259         page = grab_cache_page_write_begin(mapping, index, flags);
2260         if (!page) {
2261                 rc = -ENOMEM;
2262                 goto out;
2263         }
2264
2265         if (PageUptodate(page))
2266                 goto out;
2267
2268         /*
2269          * If we write a full page it will be up to date, no need to read from
2270          * the server. If the write is short, we'll end up doing a sync write
2271          * instead.
2272          */
2273         if (len == PAGE_CACHE_SIZE)
2274                 goto out;
2275
2276         /*
2277          * optimize away the read when we have an oplock, and we're not
2278          * expecting to use any of the data we'd be reading in. That
2279          * is, when the page lies beyond the EOF, or straddles the EOF
2280          * and the write will cover all of the existing data.
2281          */
2282         if (CIFS_I(mapping->host)->clientCanCacheRead) {
2283                 i_size = i_size_read(mapping->host);
2284                 if (page_start >= i_size ||
2285                     (offset == 0 && (pos + len) >= i_size)) {
2286                         zero_user_segments(page, 0, offset,
2287                                            offset + len,
2288                                            PAGE_CACHE_SIZE);
2289                         /*
2290                          * PageChecked means that the parts of the page
2291                          * to which we're not writing are considered up
2292                          * to date. Once the data is copied to the
2293                          * page, it can be set uptodate.
2294                          */
2295                         SetPageChecked(page);
2296                         goto out;
2297                 }
2298         }
2299
2300         if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2301                 /*
2302                  * might as well read a page, it is fast enough. If we get
2303                  * an error, we don't need to return it. cifs_write_end will
2304                  * do a sync write instead since PG_uptodate isn't set.
2305                  */
2306                 cifs_readpage_worker(file, page, &page_start);
2307         } else {
2308                 /* we could try using another file handle if there is one -
2309                    but how would we lock it to prevent close of that handle
2310                    racing with this read? In any case
2311                    this will be written out by write_end so is fine */
2312         }
2313 out:
2314         *pagep = page;
2315         return rc;
2316 }
2317
2318 static int cifs_release_page(struct page *page, gfp_t gfp)
2319 {
2320         if (PagePrivate(page))
2321                 return 0;
2322
2323         return cifs_fscache_release_page(page, gfp);
2324 }
2325
2326 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2327 {
2328         struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2329
2330         if (offset == 0)
2331                 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2332 }
2333
2334 void cifs_oplock_break(struct work_struct *work)
2335 {
2336         struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2337                                                   oplock_break);
2338         struct inode *inode = cfile->pInode;
2339         struct cifsInodeInfo *cinode = CIFS_I(inode);
2340         int rc, waitrc = 0;
2341
2342         if (inode && S_ISREG(inode->i_mode)) {
2343                 if (cinode->clientCanCacheRead)
2344                         break_lease(inode, O_RDONLY);
2345                 else
2346                         break_lease(inode, O_WRONLY);
2347                 rc = filemap_fdatawrite(inode->i_mapping);
2348                 if (cinode->clientCanCacheRead == 0) {
2349                         waitrc = filemap_fdatawait(inode->i_mapping);
2350                         invalidate_remote_inode(inode);
2351                 }
2352                 if (!rc)
2353                         rc = waitrc;
2354                 if (rc)
2355                         cinode->write_behind_rc = rc;
2356                 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2357         }
2358
2359         /*
2360          * releasing stale oplock after recent reconnect of smb session using
2361          * a now incorrect file handle is not a data integrity issue but do
2362          * not bother sending an oplock release if session to server still is
2363          * disconnected since oplock already released by the server
2364          */
2365         if (!cfile->closePend && !cfile->oplock_break_cancelled) {
2366                 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid, 0,
2367                                  0, 0, 0, LOCKING_ANDX_OPLOCK_RELEASE, false);
2368                 cFYI(1, "Oplock release rc = %d", rc);
2369         }
2370
2371         /*
2372          * We might have kicked in before is_valid_oplock_break()
2373          * finished grabbing reference for us.  Make sure it's done by
2374          * waiting for GlobalSMSSeslock.
2375          */
2376         write_lock(&GlobalSMBSeslock);
2377         write_unlock(&GlobalSMBSeslock);
2378
2379         cifs_oplock_break_put(cfile);
2380 }
2381
2382 void cifs_oplock_break_get(struct cifsFileInfo *cfile)
2383 {
2384         mntget(cfile->mnt);
2385         cifsFileInfo_get(cfile);
2386 }
2387
2388 void cifs_oplock_break_put(struct cifsFileInfo *cfile)
2389 {
2390         mntput(cfile->mnt);
2391         cifsFileInfo_put(cfile);
2392 }
2393
2394 const struct address_space_operations cifs_addr_ops = {
2395         .readpage = cifs_readpage,
2396         .readpages = cifs_readpages,
2397         .writepage = cifs_writepage,
2398         .writepages = cifs_writepages,
2399         .write_begin = cifs_write_begin,
2400         .write_end = cifs_write_end,
2401         .set_page_dirty = __set_page_dirty_nobuffers,
2402         .releasepage = cifs_release_page,
2403         .invalidatepage = cifs_invalidate_page,
2404         /* .sync_page = cifs_sync_page, */
2405         /* .direct_IO = */
2406 };
2407
2408 /*
2409  * cifs_readpages requires the server to support a buffer large enough to
2410  * contain the header plus one complete page of data.  Otherwise, we need
2411  * to leave cifs_readpages out of the address space operations.
2412  */
2413 const struct address_space_operations cifs_addr_ops_smallbuf = {
2414         .readpage = cifs_readpage,
2415         .writepage = cifs_writepage,
2416         .writepages = cifs_writepages,
2417         .write_begin = cifs_write_begin,
2418         .write_end = cifs_write_end,
2419         .set_page_dirty = __set_page_dirty_nobuffers,
2420         .releasepage = cifs_release_page,
2421         .invalidatepage = cifs_invalidate_page,
2422         /* .sync_page = cifs_sync_page, */
2423         /* .direct_IO = */
2424 };