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 u32 cifs_posix_convert_flags(unsigned int flags)
  64 {
  65         u32 posix_flags = 0;
  66
  67         if ((flags & O_ACCMODE) == O_RDONLY)
  68                 posix_flags = SMB_O_RDONLY;
  69         else if ((flags & O_ACCMODE) == O_WRONLY)
  70                 posix_flags = SMB_O_WRONLY;
  71         else if ((flags & O_ACCMODE) == O_RDWR)
  72                 posix_flags = SMB_O_RDWR;
  73
  74         if (flags & O_CREAT)
  75                 posix_flags |= SMB_O_CREAT;
  76         if (flags & O_EXCL)
  77                 posix_flags |= SMB_O_EXCL;
  78         if (flags & O_TRUNC)
  79                 posix_flags |= SMB_O_TRUNC;
  80         /* be safe and imply O_SYNC for O_DSYNC */
  81         if (flags & O_DSYNC)
  82                 posix_flags |= SMB_O_SYNC;
  83         if (flags & O_DIRECTORY)
  84                 posix_flags |= SMB_O_DIRECTORY;
  85         if (flags & O_NOFOLLOW)
  86                 posix_flags |= SMB_O_NOFOLLOW;
  87         if (flags & O_DIRECT)
  88                 posix_flags |= SMB_O_DIRECT;
  89
  90         return posix_flags;
  91 }
  92
  93 static inline int cifs_get_disposition(unsigned int flags)
  94 {
  95         if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
  96                 return FILE_CREATE;
  97         else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
  98                 return FILE_OVERWRITE_IF;
  99         else if ((flags & O_CREAT) == O_CREAT)
 100                 return FILE_OPEN_IF;
 101         else if ((flags & O_TRUNC) == O_TRUNC)
 102                 return FILE_OVERWRITE;
 103         else
 104                 return FILE_OPEN;
 105 }
 106
 107 int cifs_posix_open(char *full_path, struct inode **pinode,
 108                         struct super_block *sb, int mode, unsigned int f_flags,
 109                         __u32 *poplock, __u16 *pnetfid, int xid)
 110 {
 111         int rc;
 112         FILE_UNIX_BASIC_INFO *presp_data;
 113         __u32 posix_flags = 0;
 114         struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
 115         struct cifs_fattr fattr;
 116         struct tcon_link *tlink;
 117         struct cifs_tcon *tcon;
 118
 119         cFYI(1, "posix open %s", full_path);
 120
 121         presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
 122         if (presp_data == NULL)
 123                 return -ENOMEM;
 124
 125         tlink = cifs_sb_tlink(cifs_sb);
 126         if (IS_ERR(tlink)) {
 127                 rc = PTR_ERR(tlink);
 128                 goto posix_open_ret;
 129         }
 130
 131         tcon = tlink_tcon(tlink);
 132         mode &= ~current_umask();
 133
 134         posix_flags = cifs_posix_convert_flags(f_flags);
 135         rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
 136                              poplock, full_path, cifs_sb->local_nls,
 137                              cifs_sb->mnt_cifs_flags &
 138                                         CIFS_MOUNT_MAP_SPECIAL_CHR);
 139         cifs_put_tlink(tlink);
 140
 141         if (rc)
 142                 goto posix_open_ret;
 143
 144         if (presp_data->Type == cpu_to_le32(-1))
 145                 goto posix_open_ret; /* open ok, caller does qpathinfo */
 146
 147         if (!pinode)
 148                 goto posix_open_ret; /* caller does not need info */
 149
 150         cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
 151
 152         /* get new inode and set it up */
 153         if (*pinode == NULL) {
 154                 cifs_fill_uniqueid(sb, &fattr);
 155                 *pinode = cifs_iget(sb, &fattr);
 156                 if (!*pinode) {
 157                         rc = -ENOMEM;
 158                         goto posix_open_ret;
 159                 }
 160         } else {
 161                 cifs_fattr_to_inode(*pinode, &fattr);
 162         }
 163
 164 posix_open_ret:
 165         kfree(presp_data);
 166         return rc;
 167 }
 168
 169 static int
 170 cifs_nt_open(char *full_path, struct inode *inode, struct cifs_sb_info *cifs_sb,
 171              struct cifs_tcon *tcon, unsigned int f_flags, __u32 *poplock,
 172              __u16 *pnetfid, int xid)
 173 {
 174         int rc;
 175         int desiredAccess;
 176         int disposition;
 177         int create_options = CREATE_NOT_DIR;
 178         FILE_ALL_INFO *buf;
 179
 180         desiredAccess = cifs_convert_flags(f_flags);
 181
 182 /*********************************************************************
 183  *  open flag mapping table:
 184  *
 185  *      POSIX Flag            CIFS Disposition
 186  *      ----------            ----------------
 187  *      O_CREAT               FILE_OPEN_IF
 188  *      O_CREAT | O_EXCL      FILE_CREATE
 189  *      O_CREAT | O_TRUNC     FILE_OVERWRITE_IF
 190  *      O_TRUNC               FILE_OVERWRITE
 191  *      none of the above     FILE_OPEN
 192  *
 193  *      Note that there is not a direct match between disposition
 194  *      FILE_SUPERSEDE (ie create whether or not file exists although
 195  *      O_CREAT | O_TRUNC is similar but truncates the existing
 196  *      file rather than creating a new file as FILE_SUPERSEDE does
 197  *      (which uses the attributes / metadata passed in on open call)
 198  *?
 199  *?  O_SYNC is a reasonable match to CIFS writethrough flag
 200  *?  and the read write flags match reasonably.  O_LARGEFILE
 201  *?  is irrelevant because largefile support is always used
 202  *?  by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
 203  *       O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
 204  *********************************************************************/
 205
 206         disposition = cifs_get_disposition(f_flags);
 207
 208         /* BB pass O_SYNC flag through on file attributes .. BB */
 209
 210         buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
 211         if (!buf)
 212                 return -ENOMEM;
 213
 214         if (backup_cred(cifs_sb))
 215                 create_options |= CREATE_OPEN_BACKUP_INTENT;
 216
 217         if (tcon->ses->capabilities & CAP_NT_SMBS)
 218                 rc = CIFSSMBOpen(xid, tcon, full_path, disposition,
 219                          desiredAccess, create_options, pnetfid, poplock, buf,
 220                          cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
 221                                  & CIFS_MOUNT_MAP_SPECIAL_CHR);
 222         else
 223                 rc = SMBLegacyOpen(xid, tcon, full_path, disposition,
 224                         desiredAccess, CREATE_NOT_DIR, pnetfid, poplock, buf,
 225                         cifs_sb->local_nls, cifs_sb->mnt_cifs_flags
 226                                 & CIFS_MOUNT_MAP_SPECIAL_CHR);
 227
 228         if (rc)
 229                 goto out;
 230
 231         if (tcon->unix_ext)
 232                 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
 233                                               xid);
 234         else
 235                 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
 236                                          xid, pnetfid);
 237
 238 out:
 239         kfree(buf);
 240         return rc;
 241 }
 242
 243 struct cifsFileInfo *
 244 cifs_new_fileinfo(__u16 fileHandle, struct file *file,
 245                   struct tcon_link *tlink, __u32 oplock)
 246 {
 247         struct dentry *dentry = file->f_path.dentry;
 248         struct inode *inode = dentry->d_inode;
 249         struct cifsInodeInfo *pCifsInode = CIFS_I(inode);
 250         struct cifsFileInfo *pCifsFile;
 251
 252         pCifsFile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
 253         if (pCifsFile == NULL)
 254                 return pCifsFile;
 255
 256         pCifsFile->count = 1;
 257         pCifsFile->netfid = fileHandle;
 258         pCifsFile->pid = current->tgid;
 259         pCifsFile->uid = current_fsuid();
 260         pCifsFile->dentry = dget(dentry);
 261         pCifsFile->f_flags = file->f_flags;
 262         pCifsFile->invalidHandle = false;
 263         pCifsFile->tlink = cifs_get_tlink(tlink);
 264         mutex_init(&pCifsFile->fh_mutex);
 265         INIT_WORK(&pCifsFile->oplock_break, cifs_oplock_break);
 266
 267         spin_lock(&cifs_file_list_lock);
 268         list_add(&pCifsFile->tlist, &(tlink_tcon(tlink)->openFileList));
 269         /* if readable file instance put first in list*/
 270         if (file->f_mode & FMODE_READ)
 271                 list_add(&pCifsFile->flist, &pCifsInode->openFileList);
 272         else
 273                 list_add_tail(&pCifsFile->flist, &pCifsInode->openFileList);
 274         spin_unlock(&cifs_file_list_lock);
 275
 276         cifs_set_oplock_level(pCifsInode, oplock);
 277
 278         file->private_data = pCifsFile;
 279         return pCifsFile;
 280 }
 281
 282 /*
 283  * Release a reference on the file private data. This may involve closing
 284  * the filehandle out on the server. Must be called without holding
 285  * cifs_file_list_lock.
 286  */
 287 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
 288 {
 289         struct inode *inode = cifs_file->dentry->d_inode;
 290         struct cifs_tcon *tcon = tlink_tcon(cifs_file->tlink);
 291         struct cifsInodeInfo *cifsi = CIFS_I(inode);
 292         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
 293         struct cifsLockInfo *li, *tmp;
 294
 295         spin_lock(&cifs_file_list_lock);
 296         if (--cifs_file->count > 0) {
 297                 spin_unlock(&cifs_file_list_lock);
 298                 return;
 299         }
 300
 301         /* remove it from the lists */
 302         list_del(&cifs_file->flist);
 303         list_del(&cifs_file->tlist);
 304
 305         if (list_empty(&cifsi->openFileList)) {
 306                 cFYI(1, "closing last open instance for inode %p",
 307                         cifs_file->dentry->d_inode);
 308
 309                 /* in strict cache mode we need invalidate mapping on the last
 310                    close  because it may cause a error when we open this file
 311                    again and get at least level II oplock */
 312                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
 313                         CIFS_I(inode)->invalid_mapping = true;
 314
 315                 cifs_set_oplock_level(cifsi, 0);
 316         }
 317         spin_unlock(&cifs_file_list_lock);
 318
 319         cancel_work_sync(&cifs_file->oplock_break);
 320
 321         if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
 322                 int xid, rc;
 323
 324                 xid = GetXid();
 325                 rc = CIFSSMBClose(xid, tcon, cifs_file->netfid);
 326                 FreeXid(xid);
 327         }
 328
 329         /* Delete any outstanding lock records. We'll lose them when the file
 330          * is closed anyway.
 331          */
 332         mutex_lock(&cifsi->lock_mutex);
 333         list_for_each_entry_safe(li, tmp, &cifsi->llist, llist) {
 334                 if (li->netfid != cifs_file->netfid)
 335                         continue;
 336                 list_del(&li->llist);
 337                 kfree(li);
 338         }
 339         mutex_unlock(&cifsi->lock_mutex);
 340
 341         cifs_put_tlink(cifs_file->tlink);
 342         dput(cifs_file->dentry);
 343         kfree(cifs_file);
 344 }
 345
 346 int cifs_open(struct inode *inode, struct file *file)
 347 {
 348         int rc = -EACCES;
 349         int xid;
 350         __u32 oplock;
 351         struct cifs_sb_info *cifs_sb;
 352         struct cifs_tcon *tcon;
 353         struct tcon_link *tlink;
 354         struct cifsFileInfo *pCifsFile = NULL;
 355         char *full_path = NULL;
 356         bool posix_open_ok = false;
 357         __u16 netfid;
 358
 359         xid = GetXid();
 360
 361         cifs_sb = CIFS_SB(inode->i_sb);
 362         tlink = cifs_sb_tlink(cifs_sb);
 363         if (IS_ERR(tlink)) {
 364                 FreeXid(xid);
 365                 return PTR_ERR(tlink);
 366         }
 367         tcon = tlink_tcon(tlink);
 368
 369         full_path = build_path_from_dentry(file->f_path.dentry);
 370         if (full_path == NULL) {
 371                 rc = -ENOMEM;
 372                 goto out;
 373         }
 374
 375         cFYI(1, "inode = 0x%p file flags are 0x%x for %s",
 376                  inode, file->f_flags, full_path);
 377
 378         if (enable_oplocks)
 379                 oplock = REQ_OPLOCK;
 380         else
 381                 oplock = 0;
 382
 383         if (!tcon->broken_posix_open && tcon->unix_ext &&
 384             (tcon->ses->capabilities & CAP_UNIX) &&
 385             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 386                         le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 387                 /* can not refresh inode info since size could be stale */
 388                 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
 389                                 cifs_sb->mnt_file_mode /* ignored */,
 390                                 file->f_flags, &oplock, &netfid, xid);
 391                 if (rc == 0) {
 392                         cFYI(1, "posix open succeeded");
 393                         posix_open_ok = true;
 394                 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
 395                         if (tcon->ses->serverNOS)
 396                                 cERROR(1, "server %s of type %s returned"
 397                                            " unexpected error on SMB posix open"
 398                                            ", disabling posix open support."
 399                                            " Check if server update available.",
 400                                            tcon->ses->serverName,
 401                                            tcon->ses->serverNOS);
 402                         tcon->broken_posix_open = true;
 403                 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
 404                          (rc != -EOPNOTSUPP)) /* path not found or net err */
 405                         goto out;
 406                 /* else fallthrough to retry open the old way on network i/o
 407                    or DFS errors */
 408         }
 409
 410         if (!posix_open_ok) {
 411                 rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
 412                                   file->f_flags, &oplock, &netfid, xid);
 413                 if (rc)
 414                         goto out;
 415         }
 416
 417         pCifsFile = cifs_new_fileinfo(netfid, file, tlink, oplock);
 418         if (pCifsFile == NULL) {
 419                 CIFSSMBClose(xid, tcon, netfid);
 420                 rc = -ENOMEM;
 421                 goto out;
 422         }
 423
 424         cifs_fscache_set_inode_cookie(inode, file);
 425
 426         if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
 427                 /* time to set mode which we can not set earlier due to
 428                    problems creating new read-only files */
 429                 struct cifs_unix_set_info_args args = {
 430                         .mode   = inode->i_mode,
 431                         .uid    = NO_CHANGE_64,
 432                         .gid    = NO_CHANGE_64,
 433                         .ctime  = NO_CHANGE_64,
 434                         .atime  = NO_CHANGE_64,
 435                         .mtime  = NO_CHANGE_64,
 436                         .device = 0,
 437                 };
 438                 CIFSSMBUnixSetFileInfo(xid, tcon, &args, netfid,
 439                                         pCifsFile->pid);
 440         }
 441
 442 out:
 443         kfree(full_path);
 444         FreeXid(xid);
 445         cifs_put_tlink(tlink);
 446         return rc;
 447 }
 448
 449 /* Try to reacquire byte range locks that were released when session */
 450 /* to server was lost */
 451 static int cifs_relock_file(struct cifsFileInfo *cifsFile)
 452 {
 453         int rc = 0;
 454
 455 /* BB list all locks open on this file and relock */
 456
 457         return rc;
 458 }
 459
 460 static int cifs_reopen_file(struct cifsFileInfo *pCifsFile, bool can_flush)
 461 {
 462         int rc = -EACCES;
 463         int xid;
 464         __u32 oplock;
 465         struct cifs_sb_info *cifs_sb;
 466         struct cifs_tcon *tcon;
 467         struct cifsInodeInfo *pCifsInode;
 468         struct inode *inode;
 469         char *full_path = NULL;
 470         int desiredAccess;
 471         int disposition = FILE_OPEN;
 472         int create_options = CREATE_NOT_DIR;
 473         __u16 netfid;
 474
 475         xid = GetXid();
 476         mutex_lock(&pCifsFile->fh_mutex);
 477         if (!pCifsFile->invalidHandle) {
 478                 mutex_unlock(&pCifsFile->fh_mutex);
 479                 rc = 0;
 480                 FreeXid(xid);
 481                 return rc;
 482         }
 483
 484         inode = pCifsFile->dentry->d_inode;
 485         cifs_sb = CIFS_SB(inode->i_sb);
 486         tcon = tlink_tcon(pCifsFile->tlink);
 487
 488 /* can not grab rename sem here because various ops, including
 489    those that already have the rename sem can end up causing writepage
 490    to get called and if the server was down that means we end up here,
 491    and we can never tell if the caller already has the rename_sem */
 492         full_path = build_path_from_dentry(pCifsFile->dentry);
 493         if (full_path == NULL) {
 494                 rc = -ENOMEM;
 495                 mutex_unlock(&pCifsFile->fh_mutex);
 496                 FreeXid(xid);
 497                 return rc;
 498         }
 499
 500         cFYI(1, "inode = 0x%p file flags 0x%x for %s",
 501                  inode, pCifsFile->f_flags, full_path);
 502
 503         if (enable_oplocks)
 504                 oplock = REQ_OPLOCK;
 505         else
 506                 oplock = 0;
 507
 508         if (tcon->unix_ext && (tcon->ses->capabilities & CAP_UNIX) &&
 509             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 510                         le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 511
 512                 /*
 513                  * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
 514                  * original open. Must mask them off for a reopen.
 515                  */
 516                 unsigned int oflags = pCifsFile->f_flags &
 517                                                 ~(O_CREAT | O_EXCL | O_TRUNC);
 518
 519                 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
 520                                 cifs_sb->mnt_file_mode /* ignored */,
 521                                 oflags, &oplock, &netfid, xid);
 522                 if (rc == 0) {
 523                         cFYI(1, "posix reopen succeeded");
 524                         goto reopen_success;
 525                 }
 526                 /* fallthrough to retry open the old way on errors, especially
 527                    in the reconnect path it is important to retry hard */
 528         }
 529
 530         desiredAccess = cifs_convert_flags(pCifsFile->f_flags);
 531
 532         if (backup_cred(cifs_sb))
 533                 create_options |= CREATE_OPEN_BACKUP_INTENT;
 534
 535         /* Can not refresh inode by passing in file_info buf to be returned
 536            by SMBOpen and then calling get_inode_info with returned buf
 537            since file might have write behind data that needs to be flushed
 538            and server version of file size can be stale. If we knew for sure
 539            that inode was not dirty locally we could do this */
 540
 541         rc = CIFSSMBOpen(xid, tcon, full_path, disposition, desiredAccess,
 542                          create_options, &netfid, &oplock, NULL,
 543                          cifs_sb->local_nls, cifs_sb->mnt_cifs_flags &
 544                                 CIFS_MOUNT_MAP_SPECIAL_CHR);
 545         if (rc) {
 546                 mutex_unlock(&pCifsFile->fh_mutex);
 547                 cFYI(1, "cifs_open returned 0x%x", rc);
 548                 cFYI(1, "oplock: %d", oplock);
 549                 goto reopen_error_exit;
 550         }
 551
 552 reopen_success:
 553         pCifsFile->netfid = netfid;
 554         pCifsFile->invalidHandle = false;
 555         mutex_unlock(&pCifsFile->fh_mutex);
 556         pCifsInode = CIFS_I(inode);
 557
 558         if (can_flush) {
 559                 rc = filemap_write_and_wait(inode->i_mapping);
 560                 mapping_set_error(inode->i_mapping, rc);
 561
 562                 if (tcon->unix_ext)
 563                         rc = cifs_get_inode_info_unix(&inode,
 564                                 full_path, inode->i_sb, xid);
 565                 else
 566                         rc = cifs_get_inode_info(&inode,
 567                                 full_path, NULL, inode->i_sb,
 568                                 xid, NULL);
 569         } /* else we are writing out data to server already
 570              and could deadlock if we tried to flush data, and
 571              since we do not know if we have data that would
 572              invalidate the current end of file on the server
 573              we can not go to the server to get the new inod
 574              info */
 575
 576         cifs_set_oplock_level(pCifsInode, oplock);
 577
 578         cifs_relock_file(pCifsFile);
 579
 580 reopen_error_exit:
 581         kfree(full_path);
 582         FreeXid(xid);
 583         return rc;
 584 }
 585
 586 int cifs_close(struct inode *inode, struct file *file)
 587 {
 588         if (file->private_data != NULL) {
 589                 cifsFileInfo_put(file->private_data);
 590                 file->private_data = NULL;
 591         }
 592
 593         /* return code from the ->release op is always ignored */
 594         return 0;
 595 }
 596
 597 int cifs_closedir(struct inode *inode, struct file *file)
 598 {
 599         int rc = 0;
 600         int xid;
 601         struct cifsFileInfo *pCFileStruct = file->private_data;
 602         char *ptmp;
 603
 604         cFYI(1, "Closedir inode = 0x%p", inode);
 605
 606         xid = GetXid();
 607
 608         if (pCFileStruct) {
 609                 struct cifs_tcon *pTcon = tlink_tcon(pCFileStruct->tlink);
 610
 611                 cFYI(1, "Freeing private data in close dir");
 612                 spin_lock(&cifs_file_list_lock);
 613                 if (!pCFileStruct->srch_inf.endOfSearch &&
 614                     !pCFileStruct->invalidHandle) {
 615                         pCFileStruct->invalidHandle = true;
 616                         spin_unlock(&cifs_file_list_lock);
 617                         rc = CIFSFindClose(xid, pTcon, pCFileStruct->netfid);
 618                         cFYI(1, "Closing uncompleted readdir with rc %d",
 619                                  rc);
 620                         /* not much we can do if it fails anyway, ignore rc */
 621                         rc = 0;
 622                 } else
 623                         spin_unlock(&cifs_file_list_lock);
 624                 ptmp = pCFileStruct->srch_inf.ntwrk_buf_start;
 625                 if (ptmp) {
 626                         cFYI(1, "closedir free smb buf in srch struct");
 627                         pCFileStruct->srch_inf.ntwrk_buf_start = NULL;
 628                         if (pCFileStruct->srch_inf.smallBuf)
 629                                 cifs_small_buf_release(ptmp);
 630                         else
 631                                 cifs_buf_release(ptmp);
 632                 }
 633                 cifs_put_tlink(pCFileStruct->tlink);
 634                 kfree(file->private_data);
 635                 file->private_data = NULL;
 636         }
 637         /* BB can we lock the filestruct while this is going on? */
 638         FreeXid(xid);
 639         return rc;
 640 }
 641
 642 static int store_file_lock(struct cifsInodeInfo *cinode, __u64 len,
 643                            __u64 offset, __u8 type, __u16 netfid)
 644 {
 645         struct cifsLockInfo *li =
 646                 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
 647         if (li == NULL)
 648                 return -ENOMEM;
 649         li->netfid = netfid;
 650         li->offset = offset;
 651         li->length = len;
 652         li->type = type;
 653         li->pid = current->tgid;
 654         mutex_lock(&cinode->lock_mutex);
 655         list_add_tail(&li->llist, &cinode->llist);
 656         mutex_unlock(&cinode->lock_mutex);
 657         return 0;
 658 }
 659
 660 static void
 661 cifs_read_flock(struct file_lock *flock, __u8 *type, int *lock, int *unlock,
 662                 bool *wait_flag)
 663 {
 664         if (flock->fl_flags & FL_POSIX)
 665                 cFYI(1, "Posix");
 666         if (flock->fl_flags & FL_FLOCK)
 667                 cFYI(1, "Flock");
 668         if (flock->fl_flags & FL_SLEEP) {
 669                 cFYI(1, "Blocking lock");
 670                 *wait_flag = true;
 671         }
 672         if (flock->fl_flags & FL_ACCESS)
 673                 cFYI(1, "Process suspended by mandatory locking - "
 674                         "not implemented yet");
 675         if (flock->fl_flags & FL_LEASE)
 676                 cFYI(1, "Lease on file - not implemented yet");
 677         if (flock->fl_flags &
 678             (~(FL_POSIX | FL_FLOCK | FL_SLEEP | FL_ACCESS | FL_LEASE)))
 679                 cFYI(1, "Unknown lock flags 0x%x", flock->fl_flags);
 680
 681         *type = LOCKING_ANDX_LARGE_FILES;
 682         if (flock->fl_type == F_WRLCK) {
 683                 cFYI(1, "F_WRLCK ");
 684                 *lock = 1;
 685         } else if (flock->fl_type == F_UNLCK) {
 686                 cFYI(1, "F_UNLCK");
 687                 *unlock = 1;
 688                 /* Check if unlock includes more than one lock range */
 689         } else if (flock->fl_type == F_RDLCK) {
 690                 cFYI(1, "F_RDLCK");
 691                 *type |= LOCKING_ANDX_SHARED_LOCK;
 692                 *lock = 1;
 693         } else if (flock->fl_type == F_EXLCK) {
 694                 cFYI(1, "F_EXLCK");
 695                 *lock = 1;
 696         } else if (flock->fl_type == F_SHLCK) {
 697                 cFYI(1, "F_SHLCK");
 698                 *type |= LOCKING_ANDX_SHARED_LOCK;
 699                 *lock = 1;
 700         } else
 701                 cFYI(1, "Unknown type of lock");
 702 }
 703
 704 static int
 705 cifs_getlk(struct cifsFileInfo *cfile, struct file_lock *flock, __u8 type,
 706            bool wait_flag, bool posix_lck, int xid)
 707 {
 708         int rc = 0;
 709         __u64 length = 1 + flock->fl_end - flock->fl_start;
 710         __u16 netfid = cfile->netfid;
 711         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
 712
 713         if (posix_lck) {
 714                 int posix_lock_type;
 715                 if (type & LOCKING_ANDX_SHARED_LOCK)
 716                         posix_lock_type = CIFS_RDLCK;
 717                 else
 718                         posix_lock_type = CIFS_WRLCK;
 719                 rc = CIFSSMBPosixLock(xid, tcon, netfid, 1 /* get */,
 720                                       length, flock, posix_lock_type,
 721                                       wait_flag);
 722                 return rc;
 723         }
 724
 725         /* BB we could chain these into one lock request BB */
 726         rc = CIFSSMBLock(xid, tcon, netfid, current->tgid, length,
 727                          flock->fl_start, 0, 1, type, 0, 0);
 728         if (rc == 0) {
 729                 rc = CIFSSMBLock(xid, tcon, netfid, current->tgid,
 730                                  length, flock->fl_start, 1, 0,
 731                                  type, 0, 0);
 732                 flock->fl_type = F_UNLCK;
 733                 if (rc != 0)
 734                         cERROR(1, "Error unlocking previously locked "
 735                                    "range %d during test of lock", rc);
 736                 rc = 0;
 737                 return rc;
 738         }
 739
 740         if (type & LOCKING_ANDX_SHARED_LOCK) {
 741                 flock->fl_type = F_WRLCK;
 742                 rc = 0;
 743                 return rc;
 744         }
 745
 746         rc = CIFSSMBLock(xid, tcon, netfid, current->tgid, length,
 747                          flock->fl_start, 0, 1,
 748                          type | LOCKING_ANDX_SHARED_LOCK, 0, 0);
 749         if (rc == 0) {
 750                 rc = CIFSSMBLock(xid, tcon, netfid, current->tgid,
 751                                  length, flock->fl_start, 1, 0,
 752                                  type | LOCKING_ANDX_SHARED_LOCK,
 753                                  0, 0);
 754                 flock->fl_type = F_RDLCK;
 755                 if (rc != 0)
 756                         cERROR(1, "Error unlocking previously locked "
 757                                   "range %d during test of lock", rc);
 758         } else
 759                 flock->fl_type = F_WRLCK;
 760
 761         rc = 0;
 762         return rc;
 763 }
 764
 765 static int
 766 cifs_setlk(struct file *file,  struct file_lock *flock, __u8 type,
 767            bool wait_flag, bool posix_lck, int lock, int unlock, int xid)
 768 {
 769         int rc = 0;
 770         __u64 length = 1 + flock->fl_end - flock->fl_start;
 771         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
 772         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
 773         struct cifsInodeInfo *cinode = CIFS_I(file->f_path.dentry->d_inode);
 774         __u16 netfid = cfile->netfid;
 775
 776         if (posix_lck) {
 777                 int posix_lock_type;
 778                 if (type & LOCKING_ANDX_SHARED_LOCK)
 779                         posix_lock_type = CIFS_RDLCK;
 780                 else
 781                         posix_lock_type = CIFS_WRLCK;
 782
 783                 if (unlock == 1)
 784                         posix_lock_type = CIFS_UNLCK;
 785
 786                 rc = CIFSSMBPosixLock(xid, tcon, netfid, 0 /* set */, length,
 787                                       flock, posix_lock_type, wait_flag);
 788                 goto out;
 789         }
 790
 791         if (lock) {
 792                 rc = CIFSSMBLock(xid, tcon, netfid, current->tgid, length,
 793                                  flock->fl_start, 0, lock, type, wait_flag, 0);
 794                 if (rc == 0) {
 795                         /* For Windows locks we must store them. */
 796                         rc = store_file_lock(cinode, length, flock->fl_start,
 797                                              type, netfid);
 798                 }
 799         } else if (unlock) {
 800                 /*
 801                  * For each stored lock that this unlock overlaps completely,
 802                  * unlock it.
 803                  */
 804                 int stored_rc = 0;
 805                 struct cifsLockInfo *li, *tmp;
 806
 807                 mutex_lock(&cinode->lock_mutex);
 808                 list_for_each_entry_safe(li, tmp, &cinode->llist, llist) {
 809                         if (flock->fl_start > li->offset ||
 810                             (flock->fl_start + length) <
 811                             (li->offset + li->length))
 812                                 continue;
 813                         if (current->tgid != li->pid)
 814                                 continue;
 815                         if (cfile->netfid != li->netfid)
 816                                 continue;
 817
 818                         stored_rc = CIFSSMBLock(xid, tcon, netfid,
 819                                                 current->tgid, li->length,
 820                                                 li->offset, 1, 0, li->type,
 821                                                 0, 0);
 822                         if (stored_rc)
 823                                 rc = stored_rc;
 824                         else {
 825                                 list_del(&li->llist);
 826                                 kfree(li);
 827                         }
 828                 }
 829                 mutex_unlock(&cinode->lock_mutex);
 830         }
 831 out:
 832         if (flock->fl_flags & FL_POSIX)
 833                 posix_lock_file_wait(file, flock);
 834         return rc;
 835 }
 836
 837 int cifs_lock(struct file *file, int cmd, struct file_lock *flock)
 838 {
 839         int rc, xid;
 840         int lock = 0, unlock = 0;
 841         bool wait_flag = false;
 842         bool posix_lck = false;
 843         struct cifs_sb_info *cifs_sb;
 844         struct cifs_tcon *tcon;
 845         struct cifsInodeInfo *cinode;
 846         struct cifsFileInfo *cfile;
 847         __u16 netfid;
 848         __u8 type;
 849
 850         rc = -EACCES;
 851         xid = GetXid();
 852
 853         cFYI(1, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld "
 854                 "end: %lld", cmd, flock->fl_flags, flock->fl_type,
 855                 flock->fl_start, flock->fl_end);
 856
 857         cifs_read_flock(flock, &type, &lock, &unlock, &wait_flag);
 858
 859         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
 860         cfile = (struct cifsFileInfo *)file->private_data;
 861         tcon = tlink_tcon(cfile->tlink);
 862         netfid = cfile->netfid;
 863         cinode = CIFS_I(file->f_path.dentry->d_inode);
 864
 865         if ((tcon->ses->capabilities & CAP_UNIX) &&
 866             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
 867             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
 868                 posix_lck = true;
 869         /*
 870          * BB add code here to normalize offset and length to account for
 871          * negative length which we can not accept over the wire.
 872          */
 873         if (IS_GETLK(cmd)) {
 874                 rc = cifs_getlk(cfile, flock, type, wait_flag, posix_lck, xid);
 875                 FreeXid(xid);
 876                 return rc;
 877         }
 878
 879         if (!lock && !unlock) {
 880                 /*
 881                  * if no lock or unlock then nothing to do since we do not
 882                  * know what it is
 883                  */
 884                 FreeXid(xid);
 885                 return -EOPNOTSUPP;
 886         }
 887
 888         rc = cifs_setlk(file, flock, type, wait_flag, posix_lck, lock, unlock,
 889                         xid);
 890         FreeXid(xid);
 891         return rc;
 892 }
 893
 894 /* update the file size (if needed) after a write */
 895 void
 896 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
 897                       unsigned int bytes_written)
 898 {
 899         loff_t end_of_write = offset + bytes_written;
 900
 901         if (end_of_write > cifsi->server_eof)
 902                 cifsi->server_eof = end_of_write;
 903 }
 904
 905 static ssize_t cifs_write(struct cifsFileInfo *open_file, __u32 pid,
 906                           const char *write_data, size_t write_size,
 907                           loff_t *poffset)
 908 {
 909         int rc = 0;
 910         unsigned int bytes_written = 0;
 911         unsigned int total_written;
 912         struct cifs_sb_info *cifs_sb;
 913         struct cifs_tcon *pTcon;
 914         int xid;
 915         struct dentry *dentry = open_file->dentry;
 916         struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
 917         struct cifs_io_parms io_parms;
 918
 919         cifs_sb = CIFS_SB(dentry->d_sb);
 920
 921         cFYI(1, "write %zd bytes to offset %lld of %s", write_size,
 922            *poffset, dentry->d_name.name);
 923
 924         pTcon = tlink_tcon(open_file->tlink);
 925
 926         xid = GetXid();
 927
 928         for (total_written = 0; write_size > total_written;
 929              total_written += bytes_written) {
 930                 rc = -EAGAIN;
 931                 while (rc == -EAGAIN) {
 932                         struct kvec iov[2];
 933                         unsigned int len;
 934
 935                         if (open_file->invalidHandle) {
 936                                 /* we could deadlock if we called
 937                                    filemap_fdatawait from here so tell
 938                                    reopen_file not to flush data to
 939                                    server now */
 940                                 rc = cifs_reopen_file(open_file, false);
 941                                 if (rc != 0)
 942                                         break;
 943                         }
 944
 945                         len = min((size_t)cifs_sb->wsize,
 946                                   write_size - total_written);
 947                         /* iov[0] is reserved for smb header */
 948                         iov[1].iov_base = (char *)write_data + total_written;
 949                         iov[1].iov_len = len;
 950                         io_parms.netfid = open_file->netfid;
 951                         io_parms.pid = pid;
 952                         io_parms.tcon = pTcon;
 953                         io_parms.offset = *poffset;
 954                         io_parms.length = len;
 955                         rc = CIFSSMBWrite2(xid, &io_parms, &bytes_written, iov,
 956                                            1, 0);
 957                 }
 958                 if (rc || (bytes_written == 0)) {
 959                         if (total_written)
 960                                 break;
 961                         else {
 962                                 FreeXid(xid);
 963                                 return rc;
 964                         }
 965                 } else {
 966                         cifs_update_eof(cifsi, *poffset, bytes_written);
 967                         *poffset += bytes_written;
 968                 }
 969         }
 970
 971         cifs_stats_bytes_written(pTcon, total_written);
 972
 973         if (total_written > 0) {
 974                 spin_lock(&dentry->d_inode->i_lock);
 975                 if (*poffset > dentry->d_inode->i_size)
 976                         i_size_write(dentry->d_inode, *poffset);
 977                 spin_unlock(&dentry->d_inode->i_lock);
 978         }
 979         mark_inode_dirty_sync(dentry->d_inode);
 980         FreeXid(xid);
 981         return total_written;
 982 }
 983
 984 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
 985                                         bool fsuid_only)
 986 {
 987         struct cifsFileInfo *open_file = NULL;
 988         struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
 989
 990         /* only filter by fsuid on multiuser mounts */
 991         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
 992                 fsuid_only = false;
 993
 994         spin_lock(&cifs_file_list_lock);
 995         /* we could simply get the first_list_entry since write-only entries
 996            are always at the end of the list but since the first entry might
 997            have a close pending, we go through the whole list */
 998         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
 999                 if (fsuid_only && open_file->uid != current_fsuid())
1000                         continue;
1001                 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1002                         if (!open_file->invalidHandle) {
1003                                 /* found a good file */
1004                                 /* lock it so it will not be closed on us */
1005                                 cifsFileInfo_get(open_file);
1006                                 spin_unlock(&cifs_file_list_lock);
1007                                 return open_file;
1008                         } /* else might as well continue, and look for
1009                              another, or simply have the caller reopen it
1010                              again rather than trying to fix this handle */
1011                 } else /* write only file */
1012                         break; /* write only files are last so must be done */
1013         }
1014         spin_unlock(&cifs_file_list_lock);
1015         return NULL;
1016 }
1017
1018 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1019                                         bool fsuid_only)
1020 {
1021         struct cifsFileInfo *open_file;
1022         struct cifs_sb_info *cifs_sb;
1023         bool any_available = false;
1024         int rc;
1025
1026         /* Having a null inode here (because mapping->host was set to zero by
1027         the VFS or MM) should not happen but we had reports of on oops (due to
1028         it being zero) during stress testcases so we need to check for it */
1029
1030         if (cifs_inode == NULL) {
1031                 cERROR(1, "Null inode passed to cifs_writeable_file");
1032                 dump_stack();
1033                 return NULL;
1034         }
1035
1036         cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1037
1038         /* only filter by fsuid on multiuser mounts */
1039         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1040                 fsuid_only = false;
1041
1042         spin_lock(&cifs_file_list_lock);
1043 refind_writable:
1044         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1045                 if (!any_available && open_file->pid != current->tgid)
1046                         continue;
1047                 if (fsuid_only && open_file->uid != current_fsuid())
1048                         continue;
1049                 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1050                         cifsFileInfo_get(open_file);
1051
1052                         if (!open_file->invalidHandle) {
1053                                 /* found a good writable file */
1054                                 spin_unlock(&cifs_file_list_lock);
1055                                 return open_file;
1056                         }
1057
1058                         spin_unlock(&cifs_file_list_lock);
1059
1060                         /* Had to unlock since following call can block */
1061                         rc = cifs_reopen_file(open_file, false);
1062                         if (!rc)
1063                                 return open_file;
1064
1065                         /* if it fails, try another handle if possible */
1066                         cFYI(1, "wp failed on reopen file");
1067                         cifsFileInfo_put(open_file);
1068
1069                         spin_lock(&cifs_file_list_lock);
1070
1071                         /* else we simply continue to the next entry. Thus
1072                            we do not loop on reopen errors.  If we
1073                            can not reopen the file, for example if we
1074                            reconnected to a server with another client
1075                            racing to delete or lock the file we would not
1076                            make progress if we restarted before the beginning
1077                            of the loop here. */
1078                 }
1079         }
1080         /* couldn't find useable FH with same pid, try any available */
1081         if (!any_available) {
1082                 any_available = true;
1083                 goto refind_writable;
1084         }
1085         spin_unlock(&cifs_file_list_lock);
1086         return NULL;
1087 }
1088
1089 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1090 {
1091         struct address_space *mapping = page->mapping;
1092         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1093         char *write_data;
1094         int rc = -EFAULT;
1095         int bytes_written = 0;
1096         struct inode *inode;
1097         struct cifsFileInfo *open_file;
1098
1099         if (!mapping || !mapping->host)
1100                 return -EFAULT;
1101
1102         inode = page->mapping->host;
1103
1104         offset += (loff_t)from;
1105         write_data = kmap(page);
1106         write_data += from;
1107
1108         if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1109                 kunmap(page);
1110                 return -EIO;
1111         }
1112
1113         /* racing with truncate? */
1114         if (offset > mapping->host->i_size) {
1115                 kunmap(page);
1116                 return 0; /* don't care */
1117         }
1118
1119         /* check to make sure that we are not extending the file */
1120         if (mapping->host->i_size - offset < (loff_t)to)
1121                 to = (unsigned)(mapping->host->i_size - offset);
1122
1123         open_file = find_writable_file(CIFS_I(mapping->host), false);
1124         if (open_file) {
1125                 bytes_written = cifs_write(open_file, open_file->pid,
1126                                            write_data, to - from, &offset);
1127                 cifsFileInfo_put(open_file);
1128                 /* Does mm or vfs already set times? */
1129                 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1130                 if ((bytes_written > 0) && (offset))
1131                         rc = 0;
1132                 else if (bytes_written < 0)
1133                         rc = bytes_written;
1134         } else {
1135                 cFYI(1, "No writeable filehandles for inode");
1136                 rc = -EIO;
1137         }
1138
1139         kunmap(page);
1140         return rc;
1141 }
1142
1143 static int cifs_writepages(struct address_space *mapping,
1144                            struct writeback_control *wbc)
1145 {
1146         struct cifs_sb_info *cifs_sb = CIFS_SB(mapping->host->i_sb);
1147         bool done = false, scanned = false, range_whole = false;
1148         pgoff_t end, index;
1149         struct cifs_writedata *wdata;
1150         struct page *page;
1151         int rc = 0;
1152
1153         /*
1154          * If wsize is smaller than the page cache size, default to writing
1155          * one page at a time via cifs_writepage
1156          */
1157         if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1158                 return generic_writepages(mapping, wbc);
1159
1160         if (wbc->range_cyclic) {
1161                 index = mapping->writeback_index; /* Start from prev offset */
1162                 end = -1;
1163         } else {
1164                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1165                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1166                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1167                         range_whole = true;
1168                 scanned = true;
1169         }
1170 retry:
1171         while (!done && index <= end) {
1172                 unsigned int i, nr_pages, found_pages;
1173                 pgoff_t next = 0, tofind;
1174                 struct page **pages;
1175
1176                 tofind = min((cifs_sb->wsize / PAGE_CACHE_SIZE) - 1,
1177                                 end - index) + 1;
1178
1179                 wdata = cifs_writedata_alloc((unsigned int)tofind);
1180                 if (!wdata) {
1181                         rc = -ENOMEM;
1182                         break;
1183                 }
1184
1185                 /*
1186                  * find_get_pages_tag seems to return a max of 256 on each
1187                  * iteration, so we must call it several times in order to
1188                  * fill the array or the wsize is effectively limited to
1189                  * 256 * PAGE_CACHE_SIZE.
1190                  */
1191                 found_pages = 0;
1192                 pages = wdata->pages;
1193                 do {
1194                         nr_pages = find_get_pages_tag(mapping, &index,
1195                                                         PAGECACHE_TAG_DIRTY,
1196                                                         tofind, pages);
1197                         found_pages += nr_pages;
1198                         tofind -= nr_pages;
1199                         pages += nr_pages;
1200                 } while (nr_pages && tofind && index <= end);
1201
1202                 if (found_pages == 0) {
1203                         kref_put(&wdata->refcount, cifs_writedata_release);
1204                         break;
1205                 }
1206
1207                 nr_pages = 0;
1208                 for (i = 0; i < found_pages; i++) {
1209                         page = wdata->pages[i];
1210                         /*
1211                          * At this point we hold neither mapping->tree_lock nor
1212                          * lock on the page itself: the page may be truncated or
1213                          * invalidated (changing page->mapping to NULL), or even
1214                          * swizzled back from swapper_space to tmpfs file
1215                          * mapping
1216                          */
1217
1218                         if (nr_pages == 0)
1219                                 lock_page(page);
1220                         else if (!trylock_page(page))
1221                                 break;
1222
1223                         if (unlikely(page->mapping != mapping)) {
1224                                 unlock_page(page);
1225                                 break;
1226                         }
1227
1228                         if (!wbc->range_cyclic && page->index > end) {
1229                                 done = true;
1230                                 unlock_page(page);
1231                                 break;
1232                         }
1233
1234                         if (next && (page->index != next)) {
1235                                 /* Not next consecutive page */
1236                                 unlock_page(page);
1237                                 break;
1238                         }
1239
1240                         if (wbc->sync_mode != WB_SYNC_NONE)
1241                                 wait_on_page_writeback(page);
1242
1243                         if (PageWriteback(page) ||
1244                                         !clear_page_dirty_for_io(page)) {
1245                                 unlock_page(page);
1246                                 break;
1247                         }
1248
1249                         /*
1250                          * This actually clears the dirty bit in the radix tree.
1251                          * See cifs_writepage() for more commentary.
1252                          */
1253                         set_page_writeback(page);
1254
1255                         if (page_offset(page) >= mapping->host->i_size) {
1256                                 done = true;
1257                                 unlock_page(page);
1258                                 end_page_writeback(page);
1259                                 break;
1260                         }
1261
1262                         wdata->pages[i] = page;
1263                         next = page->index + 1;
1264                         ++nr_pages;
1265                 }
1266
1267                 /* reset index to refind any pages skipped */
1268                 if (nr_pages == 0)
1269                         index = wdata->pages[0]->index + 1;
1270
1271                 /* put any pages we aren't going to use */
1272                 for (i = nr_pages; i < found_pages; i++) {
1273                         page_cache_release(wdata->pages[i]);
1274                         wdata->pages[i] = NULL;
1275                 }
1276
1277                 /* nothing to write? */
1278                 if (nr_pages == 0) {
1279                         kref_put(&wdata->refcount, cifs_writedata_release);
1280                         continue;
1281                 }
1282
1283                 wdata->sync_mode = wbc->sync_mode;
1284                 wdata->nr_pages = nr_pages;
1285                 wdata->offset = page_offset(wdata->pages[0]);
1286
1287                 do {
1288                         if (wdata->cfile != NULL)
1289                                 cifsFileInfo_put(wdata->cfile);
1290                         wdata->cfile = find_writable_file(CIFS_I(mapping->host),
1291                                                           false);
1292                         if (!wdata->cfile) {
1293                                 cERROR(1, "No writable handles for inode");
1294                                 rc = -EBADF;
1295                                 break;
1296                         }
1297                         rc = cifs_async_writev(wdata);
1298                 } while (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN);
1299
1300                 for (i = 0; i < nr_pages; ++i)
1301                         unlock_page(wdata->pages[i]);
1302
1303                 /* send failure -- clean up the mess */
1304                 if (rc != 0) {
1305                         for (i = 0; i < nr_pages; ++i) {
1306                                 if (rc == -EAGAIN)
1307                                         redirty_page_for_writepage(wbc,
1308                                                            wdata->pages[i]);
1309                                 else
1310                                         SetPageError(wdata->pages[i]);
1311                                 end_page_writeback(wdata->pages[i]);
1312                                 page_cache_release(wdata->pages[i]);
1313                         }
1314                         if (rc != -EAGAIN)
1315                                 mapping_set_error(mapping, rc);
1316                 }
1317                 kref_put(&wdata->refcount, cifs_writedata_release);
1318
1319                 wbc->nr_to_write -= nr_pages;
1320                 if (wbc->nr_to_write <= 0)
1321                         done = true;
1322
1323                 index = next;
1324         }
1325
1326         if (!scanned && !done) {
1327                 /*
1328                  * We hit the last page and there is more work to be done: wrap
1329                  * back to the start of the file
1330                  */
1331                 scanned = true;
1332                 index = 0;
1333                 goto retry;
1334         }
1335
1336         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1337                 mapping->writeback_index = index;
1338
1339         return rc;
1340 }
1341
1342 static int
1343 cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
1344 {
1345         int rc;
1346         int xid;
1347
1348         xid = GetXid();
1349 /* BB add check for wbc flags */
1350         page_cache_get(page);
1351         if (!PageUptodate(page))
1352                 cFYI(1, "ppw - page not up to date");
1353
1354         /*
1355          * Set the "writeback" flag, and clear "dirty" in the radix tree.
1356          *
1357          * A writepage() implementation always needs to do either this,
1358          * or re-dirty the page with "redirty_page_for_writepage()" in
1359          * the case of a failure.
1360          *
1361          * Just unlocking the page will cause the radix tree tag-bits
1362          * to fail to update with the state of the page correctly.
1363          */
1364         set_page_writeback(page);
1365 retry_write:
1366         rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
1367         if (rc == -EAGAIN && wbc->sync_mode == WB_SYNC_ALL)
1368                 goto retry_write;
1369         else if (rc == -EAGAIN)
1370                 redirty_page_for_writepage(wbc, page);
1371         else if (rc != 0)
1372                 SetPageError(page);
1373         else
1374                 SetPageUptodate(page);
1375         end_page_writeback(page);
1376         page_cache_release(page);
1377         FreeXid(xid);
1378         return rc;
1379 }
1380
1381 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
1382 {
1383         int rc = cifs_writepage_locked(page, wbc);
1384         unlock_page(page);
1385         return rc;
1386 }
1387
1388 static int cifs_write_end(struct file *file, struct address_space *mapping,
1389                         loff_t pos, unsigned len, unsigned copied,
1390                         struct page *page, void *fsdata)
1391 {
1392         int rc;
1393         struct inode *inode = mapping->host;
1394         struct cifsFileInfo *cfile = file->private_data;
1395         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
1396         __u32 pid;
1397
1398         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
1399                 pid = cfile->pid;
1400         else
1401                 pid = current->tgid;
1402
1403         cFYI(1, "write_end for page %p from pos %lld with %d bytes",
1404                  page, pos, copied);
1405
1406         if (PageChecked(page)) {
1407                 if (copied == len)
1408                         SetPageUptodate(page);
1409                 ClearPageChecked(page);
1410         } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
1411                 SetPageUptodate(page);
1412
1413         if (!PageUptodate(page)) {
1414                 char *page_data;
1415                 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
1416                 int xid;
1417
1418                 xid = GetXid();
1419                 /* this is probably better than directly calling
1420                    partialpage_write since in this function the file handle is
1421                    known which we might as well leverage */
1422                 /* BB check if anything else missing out of ppw
1423                    such as updating last write time */
1424                 page_data = kmap(page);
1425                 rc = cifs_write(cfile, pid, page_data + offset, copied, &pos);
1426                 /* if (rc < 0) should we set writebehind rc? */
1427                 kunmap(page);
1428
1429                 FreeXid(xid);
1430         } else {
1431                 rc = copied;
1432                 pos += copied;
1433                 set_page_dirty(page);
1434         }
1435
1436         if (rc > 0) {
1437                 spin_lock(&inode->i_lock);
1438                 if (pos > inode->i_size)
1439                         i_size_write(inode, pos);
1440                 spin_unlock(&inode->i_lock);
1441         }
1442
1443         unlock_page(page);
1444         page_cache_release(page);
1445
1446         return rc;
1447 }
1448
1449 int cifs_strict_fsync(struct file *file, loff_t start, loff_t end,
1450                       int datasync)
1451 {
1452         int xid;
1453         int rc = 0;
1454         struct cifs_tcon *tcon;
1455         struct cifsFileInfo *smbfile = file->private_data;
1456         struct inode *inode = file->f_path.dentry->d_inode;
1457         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
1458
1459         rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
1460         if (rc)
1461                 return rc;
1462         mutex_lock(&inode->i_mutex);
1463
1464         xid = GetXid();
1465
1466         cFYI(1, "Sync file - name: %s datasync: 0x%x",
1467                 file->f_path.dentry->d_name.name, datasync);
1468
1469         if (!CIFS_I(inode)->clientCanCacheRead) {
1470                 rc = cifs_invalidate_mapping(inode);
1471                 if (rc) {
1472                         cFYI(1, "rc: %d during invalidate phase", rc);
1473                         rc = 0; /* don't care about it in fsync */
1474                 }
1475         }
1476
1477         tcon = tlink_tcon(smbfile->tlink);
1478         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1479                 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1480
1481         FreeXid(xid);
1482         mutex_unlock(&inode->i_mutex);
1483         return rc;
1484 }
1485
1486 int cifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1487 {
1488         int xid;
1489         int rc = 0;
1490         struct cifs_tcon *tcon;
1491         struct cifsFileInfo *smbfile = file->private_data;
1492         struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1493         struct inode *inode = file->f_mapping->host;
1494
1495         rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
1496         if (rc)
1497                 return rc;
1498         mutex_lock(&inode->i_mutex);
1499
1500         xid = GetXid();
1501
1502         cFYI(1, "Sync file - name: %s datasync: 0x%x",
1503                 file->f_path.dentry->d_name.name, datasync);
1504
1505         tcon = tlink_tcon(smbfile->tlink);
1506         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC))
1507                 rc = CIFSSMBFlush(xid, tcon, smbfile->netfid);
1508
1509         FreeXid(xid);
1510         mutex_unlock(&inode->i_mutex);
1511         return rc;
1512 }
1513
1514 /*
1515  * As file closes, flush all cached write data for this inode checking
1516  * for write behind errors.
1517  */
1518 int cifs_flush(struct file *file, fl_owner_t id)
1519 {
1520         struct inode *inode = file->f_path.dentry->d_inode;
1521         int rc = 0;
1522
1523         if (file->f_mode & FMODE_WRITE)
1524                 rc = filemap_write_and_wait(inode->i_mapping);
1525
1526         cFYI(1, "Flush inode %p file %p rc %d", inode, file, rc);
1527
1528         return rc;
1529 }
1530
1531 static int
1532 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
1533 {
1534         int rc = 0;
1535         unsigned long i;
1536
1537         for (i = 0; i < num_pages; i++) {
1538                 pages[i] = alloc_page(__GFP_HIGHMEM);
1539                 if (!pages[i]) {
1540                         /*
1541                          * save number of pages we have already allocated and
1542                          * return with ENOMEM error
1543                          */
1544                         num_pages = i;
1545                         rc = -ENOMEM;
1546                         goto error;
1547                 }
1548         }
1549
1550         return rc;
1551
1552 error:
1553         for (i = 0; i < num_pages; i++)
1554                 put_page(pages[i]);
1555         return rc;
1556 }
1557
1558 static inline
1559 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
1560 {
1561         size_t num_pages;
1562         size_t clen;
1563
1564         clen = min_t(const size_t, len, wsize);
1565         num_pages = clen / PAGE_CACHE_SIZE;
1566         if (clen % PAGE_CACHE_SIZE)
1567                 num_pages++;
1568
1569         if (cur_len)
1570                 *cur_len = clen;
1571
1572         return num_pages;
1573 }
1574
1575 static ssize_t
1576 cifs_iovec_write(struct file *file, const struct iovec *iov,
1577                  unsigned long nr_segs, loff_t *poffset)
1578 {
1579         unsigned int written;
1580         unsigned long num_pages, npages, i;
1581         size_t copied, len, cur_len;
1582         ssize_t total_written = 0;
1583         struct kvec *to_send;
1584         struct page **pages;
1585         struct iov_iter it;
1586         struct inode *inode;
1587         struct cifsFileInfo *open_file;
1588         struct cifs_tcon *pTcon;
1589         struct cifs_sb_info *cifs_sb;
1590         struct cifs_io_parms io_parms;
1591         int xid, rc;
1592         __u32 pid;
1593
1594         len = iov_length(iov, nr_segs);
1595         if (!len)
1596                 return 0;
1597
1598         rc = generic_write_checks(file, poffset, &len, 0);
1599         if (rc)
1600                 return rc;
1601
1602         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1603         num_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
1604
1605         pages = kmalloc(sizeof(struct pages *)*num_pages, GFP_KERNEL);
1606         if (!pages)
1607                 return -ENOMEM;
1608
1609         to_send = kmalloc(sizeof(struct kvec)*(num_pages + 1), GFP_KERNEL);
1610         if (!to_send) {
1611                 kfree(pages);
1612                 return -ENOMEM;
1613         }
1614
1615         rc = cifs_write_allocate_pages(pages, num_pages);
1616         if (rc) {
1617                 kfree(pages);
1618                 kfree(to_send);
1619                 return rc;
1620         }
1621
1622         xid = GetXid();
1623         open_file = file->private_data;
1624
1625         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
1626                 pid = open_file->pid;
1627         else
1628                 pid = current->tgid;
1629
1630         pTcon = tlink_tcon(open_file->tlink);
1631         inode = file->f_path.dentry->d_inode;
1632
1633         iov_iter_init(&it, iov, nr_segs, len, 0);
1634         npages = num_pages;
1635
1636         do {
1637                 size_t save_len = cur_len;
1638                 for (i = 0; i < npages; i++) {
1639                         copied = min_t(const size_t, cur_len, PAGE_CACHE_SIZE);
1640                         copied = iov_iter_copy_from_user(pages[i], &it, 0,
1641                                                          copied);
1642                         cur_len -= copied;
1643                         iov_iter_advance(&it, copied);
1644                         to_send[i+1].iov_base = kmap(pages[i]);
1645                         to_send[i+1].iov_len = copied;
1646                 }
1647
1648                 cur_len = save_len - cur_len;
1649
1650                 do {
1651                         if (open_file->invalidHandle) {
1652                                 rc = cifs_reopen_file(open_file, false);
1653                                 if (rc != 0)
1654                                         break;
1655                         }
1656                         io_parms.netfid = open_file->netfid;
1657                         io_parms.pid = pid;
1658                         io_parms.tcon = pTcon;
1659                         io_parms.offset = *poffset;
1660                         io_parms.length = cur_len;
1661                         rc = CIFSSMBWrite2(xid, &io_parms, &written, to_send,
1662                                            npages, 0);
1663                 } while (rc == -EAGAIN);
1664
1665                 for (i = 0; i < npages; i++)
1666                         kunmap(pages[i]);
1667
1668                 if (written) {
1669                         len -= written;
1670                         total_written += written;
1671                         cifs_update_eof(CIFS_I(inode), *poffset, written);
1672                         *poffset += written;
1673                 } else if (rc < 0) {
1674                         if (!total_written)
1675                                 total_written = rc;
1676                         break;
1677                 }
1678
1679                 /* get length and number of kvecs of the next write */
1680                 npages = get_numpages(cifs_sb->wsize, len, &cur_len);
1681         } while (len > 0);
1682
1683         if (total_written > 0) {
1684                 spin_lock(&inode->i_lock);
1685                 if (*poffset > inode->i_size)
1686                         i_size_write(inode, *poffset);
1687                 spin_unlock(&inode->i_lock);
1688         }
1689
1690         cifs_stats_bytes_written(pTcon, total_written);
1691         mark_inode_dirty_sync(inode);
1692
1693         for (i = 0; i < num_pages; i++)
1694                 put_page(pages[i]);
1695         kfree(to_send);
1696         kfree(pages);
1697         FreeXid(xid);
1698         return total_written;
1699 }
1700
1701 ssize_t cifs_user_writev(struct kiocb *iocb, const struct iovec *iov,
1702                                 unsigned long nr_segs, loff_t pos)
1703 {
1704         ssize_t written;
1705         struct inode *inode;
1706
1707         inode = iocb->ki_filp->f_path.dentry->d_inode;
1708
1709         /*
1710          * BB - optimize the way when signing is disabled. We can drop this
1711          * extra memory-to-memory copying and use iovec buffers for constructing
1712          * write request.
1713          */
1714
1715         written = cifs_iovec_write(iocb->ki_filp, iov, nr_segs, &pos);
1716         if (written > 0) {
1717                 CIFS_I(inode)->invalid_mapping = true;
1718                 iocb->ki_pos = pos;
1719         }
1720
1721         return written;
1722 }
1723
1724 ssize_t cifs_strict_writev(struct kiocb *iocb, const struct iovec *iov,
1725                            unsigned long nr_segs, loff_t pos)
1726 {
1727         struct inode *inode;
1728
1729         inode = iocb->ki_filp->f_path.dentry->d_inode;
1730
1731         if (CIFS_I(inode)->clientCanCacheAll)
1732                 return generic_file_aio_write(iocb, iov, nr_segs, pos);
1733
1734         /*
1735          * In strict cache mode we need to write the data to the server exactly
1736          * from the pos to pos+len-1 rather than flush all affected pages
1737          * because it may cause a error with mandatory locks on these pages but
1738          * not on the region from pos to ppos+len-1.
1739          */
1740
1741         return cifs_user_writev(iocb, iov, nr_segs, pos);
1742 }
1743
1744 static ssize_t
1745 cifs_iovec_read(struct file *file, const struct iovec *iov,
1746                  unsigned long nr_segs, loff_t *poffset)
1747 {
1748         int rc;
1749         int xid;
1750         ssize_t total_read;
1751         unsigned int bytes_read = 0;
1752         size_t len, cur_len;
1753         int iov_offset = 0;
1754         struct cifs_sb_info *cifs_sb;
1755         struct cifs_tcon *pTcon;
1756         struct cifsFileInfo *open_file;
1757         struct smb_com_read_rsp *pSMBr;
1758         struct cifs_io_parms io_parms;
1759         char *read_data;
1760         __u32 pid;
1761
1762         if (!nr_segs)
1763                 return 0;
1764
1765         len = iov_length(iov, nr_segs);
1766         if (!len)
1767                 return 0;
1768
1769         xid = GetXid();
1770         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1771
1772         open_file = file->private_data;
1773         pTcon = tlink_tcon(open_file->tlink);
1774
1775         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
1776                 pid = open_file->pid;
1777         else
1778                 pid = current->tgid;
1779
1780         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1781                 cFYI(1, "attempting read on write only file instance");
1782
1783         for (total_read = 0; total_read < len; total_read += bytes_read) {
1784                 cur_len = min_t(const size_t, len - total_read, cifs_sb->rsize);
1785                 rc = -EAGAIN;
1786                 read_data = NULL;
1787
1788                 while (rc == -EAGAIN) {
1789                         int buf_type = CIFS_NO_BUFFER;
1790                         if (open_file->invalidHandle) {
1791                                 rc = cifs_reopen_file(open_file, true);
1792                                 if (rc != 0)
1793                                         break;
1794                         }
1795                         io_parms.netfid = open_file->netfid;
1796                         io_parms.pid = pid;
1797                         io_parms.tcon = pTcon;
1798                         io_parms.offset = *poffset;
1799                         io_parms.length = cur_len;
1800                         rc = CIFSSMBRead(xid, &io_parms, &bytes_read,
1801                                          &read_data, &buf_type);
1802                         pSMBr = (struct smb_com_read_rsp *)read_data;
1803                         if (read_data) {
1804                                 char *data_offset = read_data + 4 +
1805                                                 le16_to_cpu(pSMBr->DataOffset);
1806                                 if (memcpy_toiovecend(iov, data_offset,
1807                                                       iov_offset, bytes_read))
1808                                         rc = -EFAULT;
1809                                 if (buf_type == CIFS_SMALL_BUFFER)
1810                                         cifs_small_buf_release(read_data);
1811                                 else if (buf_type == CIFS_LARGE_BUFFER)
1812                                         cifs_buf_release(read_data);
1813                                 read_data = NULL;
1814                                 iov_offset += bytes_read;
1815                         }
1816                 }
1817
1818                 if (rc || (bytes_read == 0)) {
1819                         if (total_read) {
1820                                 break;
1821                         } else {
1822                                 FreeXid(xid);
1823                                 return rc;
1824                         }
1825                 } else {
1826                         cifs_stats_bytes_read(pTcon, bytes_read);
1827                         *poffset += bytes_read;
1828                 }
1829         }
1830
1831         FreeXid(xid);
1832         return total_read;
1833 }
1834
1835 ssize_t cifs_user_readv(struct kiocb *iocb, const struct iovec *iov,
1836                                unsigned long nr_segs, loff_t pos)
1837 {
1838         ssize_t read;
1839
1840         read = cifs_iovec_read(iocb->ki_filp, iov, nr_segs, &pos);
1841         if (read > 0)
1842                 iocb->ki_pos = pos;
1843
1844         return read;
1845 }
1846
1847 ssize_t cifs_strict_readv(struct kiocb *iocb, const struct iovec *iov,
1848                           unsigned long nr_segs, loff_t pos)
1849 {
1850         struct inode *inode;
1851
1852         inode = iocb->ki_filp->f_path.dentry->d_inode;
1853
1854         if (CIFS_I(inode)->clientCanCacheRead)
1855                 return generic_file_aio_read(iocb, iov, nr_segs, pos);
1856
1857         /*
1858          * In strict cache mode we need to read from the server all the time
1859          * if we don't have level II oplock because the server can delay mtime
1860          * change - so we can't make a decision about inode invalidating.
1861          * And we can also fail with pagereading if there are mandatory locks
1862          * on pages affected by this read but not on the region from pos to
1863          * pos+len-1.
1864          */
1865
1866         return cifs_user_readv(iocb, iov, nr_segs, pos);
1867 }
1868
1869 static ssize_t cifs_read(struct file *file, char *read_data, size_t read_size,
1870                          loff_t *poffset)
1871 {
1872         int rc = -EACCES;
1873         unsigned int bytes_read = 0;
1874         unsigned int total_read;
1875         unsigned int current_read_size;
1876         struct cifs_sb_info *cifs_sb;
1877         struct cifs_tcon *pTcon;
1878         int xid;
1879         char *current_offset;
1880         struct cifsFileInfo *open_file;
1881         struct cifs_io_parms io_parms;
1882         int buf_type = CIFS_NO_BUFFER;
1883         __u32 pid;
1884
1885         xid = GetXid();
1886         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1887
1888         if (file->private_data == NULL) {
1889                 rc = -EBADF;
1890                 FreeXid(xid);
1891                 return rc;
1892         }
1893         open_file = file->private_data;
1894         pTcon = tlink_tcon(open_file->tlink);
1895
1896         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
1897                 pid = open_file->pid;
1898         else
1899                 pid = current->tgid;
1900
1901         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
1902                 cFYI(1, "attempting read on write only file instance");
1903
1904         for (total_read = 0, current_offset = read_data;
1905              read_size > total_read;
1906              total_read += bytes_read, current_offset += bytes_read) {
1907                 current_read_size = min_t(const int, read_size - total_read,
1908                                           cifs_sb->rsize);
1909                 /* For windows me and 9x we do not want to request more
1910                 than it negotiated since it will refuse the read then */
1911                 if ((pTcon->ses) &&
1912                         !(pTcon->ses->capabilities & CAP_LARGE_FILES)) {
1913                         current_read_size = min_t(const int, current_read_size,
1914                                         CIFSMaxBufSize);
1915                 }
1916                 rc = -EAGAIN;
1917                 while (rc == -EAGAIN) {
1918                         if (open_file->invalidHandle) {
1919                                 rc = cifs_reopen_file(open_file, true);
1920                                 if (rc != 0)
1921                                         break;
1922                         }
1923                         io_parms.netfid = open_file->netfid;
1924                         io_parms.pid = pid;
1925                         io_parms.tcon = pTcon;
1926                         io_parms.offset = *poffset;
1927                         io_parms.length = current_read_size;
1928                         rc = CIFSSMBRead(xid, &io_parms, &bytes_read,
1929                                          &current_offset, &buf_type);
1930                 }
1931                 if (rc || (bytes_read == 0)) {
1932                         if (total_read) {
1933                                 break;
1934                         } else {
1935                                 FreeXid(xid);
1936                                 return rc;
1937                         }
1938                 } else {
1939                         cifs_stats_bytes_read(pTcon, total_read);
1940                         *poffset += bytes_read;
1941                 }
1942         }
1943         FreeXid(xid);
1944         return total_read;
1945 }
1946
1947 /*
1948  * If the page is mmap'ed into a process' page tables, then we need to make
1949  * sure that it doesn't change while being written back.
1950  */
1951 static int
1952 cifs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1953 {
1954         struct page *page = vmf->page;
1955
1956         lock_page(page);
1957         return VM_FAULT_LOCKED;
1958 }
1959
1960 static struct vm_operations_struct cifs_file_vm_ops = {
1961         .fault = filemap_fault,
1962         .page_mkwrite = cifs_page_mkwrite,
1963 };
1964
1965 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
1966 {
1967         int rc, xid;
1968         struct inode *inode = file->f_path.dentry->d_inode;
1969
1970         xid = GetXid();
1971
1972         if (!CIFS_I(inode)->clientCanCacheRead) {
1973                 rc = cifs_invalidate_mapping(inode);
1974                 if (rc)
1975                         return rc;
1976         }
1977
1978         rc = generic_file_mmap(file, vma);
1979         if (rc == 0)
1980                 vma->vm_ops = &cifs_file_vm_ops;
1981         FreeXid(xid);
1982         return rc;
1983 }
1984
1985 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1986 {
1987         int rc, xid;
1988
1989         xid = GetXid();
1990         rc = cifs_revalidate_file(file);
1991         if (rc) {
1992                 cFYI(1, "Validation prior to mmap failed, error=%d", rc);
1993                 FreeXid(xid);
1994                 return rc;
1995         }
1996         rc = generic_file_mmap(file, vma);
1997         if (rc == 0)
1998                 vma->vm_ops = &cifs_file_vm_ops;
1999         FreeXid(xid);
2000         return rc;
2001 }
2002
2003
2004 static void cifs_copy_cache_pages(struct address_space *mapping,
2005         struct list_head *pages, int bytes_read, char *data)
2006 {
2007         struct page *page;
2008         char *target;
2009
2010         while (bytes_read > 0) {
2011                 if (list_empty(pages))
2012                         break;
2013
2014                 page = list_entry(pages->prev, struct page, lru);
2015                 list_del(&page->lru);
2016
2017                 if (add_to_page_cache_lru(page, mapping, page->index,
2018                                       GFP_KERNEL)) {
2019                         page_cache_release(page);
2020                         cFYI(1, "Add page cache failed");
2021                         data += PAGE_CACHE_SIZE;
2022                         bytes_read -= PAGE_CACHE_SIZE;
2023                         continue;
2024                 }
2025                 page_cache_release(page);
2026
2027                 target = kmap_atomic(page, KM_USER0);
2028
2029                 if (PAGE_CACHE_SIZE > bytes_read) {
2030                         memcpy(target, data, bytes_read);
2031                         /* zero the tail end of this partial page */
2032                         memset(target + bytes_read, 0,
2033                                PAGE_CACHE_SIZE - bytes_read);
2034                         bytes_read = 0;
2035                 } else {
2036                         memcpy(target, data, PAGE_CACHE_SIZE);
2037                         bytes_read -= PAGE_CACHE_SIZE;
2038                 }
2039                 kunmap_atomic(target, KM_USER0);
2040
2041                 flush_dcache_page(page);
2042                 SetPageUptodate(page);
2043                 unlock_page(page);
2044                 data += PAGE_CACHE_SIZE;
2045
2046                 /* add page to FS-Cache */
2047                 cifs_readpage_to_fscache(mapping->host, page);
2048         }
2049         return;
2050 }
2051
2052 static int cifs_readpages(struct file *file, struct address_space *mapping,
2053         struct list_head *page_list, unsigned num_pages)
2054 {
2055         int rc = -EACCES;
2056         int xid;
2057         loff_t offset;
2058         struct page *page;
2059         struct cifs_sb_info *cifs_sb;
2060         struct cifs_tcon *pTcon;
2061         unsigned int bytes_read = 0;
2062         unsigned int read_size, i;
2063         char *smb_read_data = NULL;
2064         struct smb_com_read_rsp *pSMBr;
2065         struct cifsFileInfo *open_file;
2066         struct cifs_io_parms io_parms;
2067         int buf_type = CIFS_NO_BUFFER;
2068         __u32 pid;
2069
2070         xid = GetXid();
2071         if (file->private_data == NULL) {
2072                 rc = -EBADF;
2073                 FreeXid(xid);
2074                 return rc;
2075         }
2076         open_file = file->private_data;
2077         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2078         pTcon = tlink_tcon(open_file->tlink);
2079
2080         /*
2081          * Reads as many pages as possible from fscache. Returns -ENOBUFS
2082          * immediately if the cookie is negative
2083          */
2084         rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
2085                                          &num_pages);
2086         if (rc == 0)
2087                 goto read_complete;
2088
2089         cFYI(DBG2, "rpages: num pages %d", num_pages);
2090         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2091                 pid = open_file->pid;
2092         else
2093                 pid = current->tgid;
2094
2095         for (i = 0; i < num_pages; ) {
2096                 unsigned contig_pages;
2097                 struct page *tmp_page;
2098                 unsigned long expected_index;
2099
2100                 if (list_empty(page_list))
2101                         break;
2102
2103                 page = list_entry(page_list->prev, struct page, lru);
2104                 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2105
2106                 /* count adjacent pages that we will read into */
2107                 contig_pages = 0;
2108                 expected_index =
2109                         list_entry(page_list->prev, struct page, lru)->index;
2110                 list_for_each_entry_reverse(tmp_page, page_list, lru) {
2111                         if (tmp_page->index == expected_index) {
2112                                 contig_pages++;
2113                                 expected_index++;
2114                         } else
2115                                 break;
2116                 }
2117                 if (contig_pages + i >  num_pages)
2118                         contig_pages = num_pages - i;
2119
2120                 /* for reads over a certain size could initiate async
2121                    read ahead */
2122
2123                 read_size = contig_pages * PAGE_CACHE_SIZE;
2124                 /* Read size needs to be in multiples of one page */
2125                 read_size = min_t(const unsigned int, read_size,
2126                                   cifs_sb->rsize & PAGE_CACHE_MASK);
2127                 cFYI(DBG2, "rpages: read size 0x%x  contiguous pages %d",
2128                                 read_size, contig_pages);
2129                 rc = -EAGAIN;
2130                 while (rc == -EAGAIN) {
2131                         if (open_file->invalidHandle) {
2132                                 rc = cifs_reopen_file(open_file, true);
2133                                 if (rc != 0)
2134                                         break;
2135                         }
2136                         io_parms.netfid = open_file->netfid;
2137                         io_parms.pid = pid;
2138                         io_parms.tcon = pTcon;
2139                         io_parms.offset = offset;
2140                         io_parms.length = read_size;
2141                         rc = CIFSSMBRead(xid, &io_parms, &bytes_read,
2142                                          &smb_read_data, &buf_type);
2143                         /* BB more RC checks ? */
2144                         if (rc == -EAGAIN) {
2145                                 if (smb_read_data) {
2146                                         if (buf_type == CIFS_SMALL_BUFFER)
2147                                                 cifs_small_buf_release(smb_read_data);
2148                                         else if (buf_type == CIFS_LARGE_BUFFER)
2149                                                 cifs_buf_release(smb_read_data);
2150                                         smb_read_data = NULL;
2151                                 }
2152                         }
2153                 }
2154                 if ((rc < 0) || (smb_read_data == NULL)) {
2155                         cFYI(1, "Read error in readpages: %d", rc);
2156                         break;
2157                 } else if (bytes_read > 0) {
2158                         task_io_account_read(bytes_read);
2159                         pSMBr = (struct smb_com_read_rsp *)smb_read_data;
2160                         cifs_copy_cache_pages(mapping, page_list, bytes_read,
2161                                 smb_read_data + 4 /* RFC1001 hdr */ +
2162                                 le16_to_cpu(pSMBr->DataOffset));
2163
2164                         i +=  bytes_read >> PAGE_CACHE_SHIFT;
2165                         cifs_stats_bytes_read(pTcon, bytes_read);
2166                         if ((bytes_read & PAGE_CACHE_MASK) != bytes_read) {
2167                                 i++; /* account for partial page */
2168
2169                                 /* server copy of file can have smaller size
2170                                    than client */
2171                                 /* BB do we need to verify this common case ?
2172                                    this case is ok - if we are at server EOF
2173                                    we will hit it on next read */
2174
2175                                 /* break; */
2176                         }
2177                 } else {
2178                         cFYI(1, "No bytes read (%d) at offset %lld . "
2179                                 "Cleaning remaining pages from readahead list",
2180                                 bytes_read, offset);
2181                         /* BB turn off caching and do new lookup on
2182                            file size at server? */
2183                         break;
2184                 }
2185                 if (smb_read_data) {
2186                         if (buf_type == CIFS_SMALL_BUFFER)
2187                                 cifs_small_buf_release(smb_read_data);
2188                         else if (buf_type == CIFS_LARGE_BUFFER)
2189                                 cifs_buf_release(smb_read_data);
2190                         smb_read_data = NULL;
2191                 }
2192                 bytes_read = 0;
2193         }
2194
2195 /* need to free smb_read_data buf before exit */
2196         if (smb_read_data) {
2197                 if (buf_type == CIFS_SMALL_BUFFER)
2198                         cifs_small_buf_release(smb_read_data);
2199                 else if (buf_type == CIFS_LARGE_BUFFER)
2200                         cifs_buf_release(smb_read_data);
2201                 smb_read_data = NULL;
2202         }
2203
2204 read_complete:
2205         FreeXid(xid);
2206         return rc;
2207 }
2208
2209 static int cifs_readpage_worker(struct file *file, struct page *page,
2210         loff_t *poffset)
2211 {
2212         char *read_data;
2213         int rc;
2214
2215         /* Is the page cached? */
2216         rc = cifs_readpage_from_fscache(file->f_path.dentry->d_inode, page);
2217         if (rc == 0)
2218                 goto read_complete;
2219
2220         page_cache_get(page);
2221         read_data = kmap(page);
2222         /* for reads over a certain size could initiate async read ahead */
2223
2224         rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
2225
2226         if (rc < 0)
2227                 goto io_error;
2228         else
2229                 cFYI(1, "Bytes read %d", rc);
2230
2231         file->f_path.dentry->d_inode->i_atime =
2232                 current_fs_time(file->f_path.dentry->d_inode->i_sb);
2233
2234         if (PAGE_CACHE_SIZE > rc)
2235                 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
2236
2237         flush_dcache_page(page);
2238         SetPageUptodate(page);
2239
2240         /* send this page to the cache */
2241         cifs_readpage_to_fscache(file->f_path.dentry->d_inode, page);
2242
2243         rc = 0;
2244
2245 io_error:
2246         kunmap(page);
2247         page_cache_release(page);
2248
2249 read_complete:
2250         return rc;
2251 }
2252
2253 static int cifs_readpage(struct file *file, struct page *page)
2254 {
2255         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
2256         int rc = -EACCES;
2257         int xid;
2258
2259         xid = GetXid();
2260
2261         if (file->private_data == NULL) {
2262                 rc = -EBADF;
2263                 FreeXid(xid);
2264                 return rc;
2265         }
2266
2267         cFYI(1, "readpage %p at offset %d 0x%x\n",
2268                  page, (int)offset, (int)offset);
2269
2270         rc = cifs_readpage_worker(file, page, &offset);
2271
2272         unlock_page(page);
2273
2274         FreeXid(xid);
2275         return rc;
2276 }
2277
2278 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
2279 {
2280         struct cifsFileInfo *open_file;
2281
2282         spin_lock(&cifs_file_list_lock);
2283         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
2284                 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
2285                         spin_unlock(&cifs_file_list_lock);
2286                         return 1;
2287                 }
2288         }
2289         spin_unlock(&cifs_file_list_lock);
2290         return 0;
2291 }
2292
2293 /* We do not want to update the file size from server for inodes
2294    open for write - to avoid races with writepage extending
2295    the file - in the future we could consider allowing
2296    refreshing the inode only on increases in the file size
2297    but this is tricky to do without racing with writebehind
2298    page caching in the current Linux kernel design */
2299 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
2300 {
2301         if (!cifsInode)
2302                 return true;
2303
2304         if (is_inode_writable(cifsInode)) {
2305                 /* This inode is open for write at least once */
2306                 struct cifs_sb_info *cifs_sb;
2307
2308                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
2309                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
2310                         /* since no page cache to corrupt on directio
2311                         we can change size safely */
2312                         return true;
2313                 }
2314
2315                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
2316                         return true;
2317
2318                 return false;
2319         } else
2320                 return true;
2321 }
2322
2323 static int cifs_write_begin(struct file *file, struct address_space *mapping,
2324                         loff_t pos, unsigned len, unsigned flags,
2325                         struct page **pagep, void **fsdata)
2326 {
2327         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2328         loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
2329         loff_t page_start = pos & PAGE_MASK;
2330         loff_t i_size;
2331         struct page *page;
2332         int rc = 0;
2333
2334         cFYI(1, "write_begin from %lld len %d", (long long)pos, len);
2335
2336         page = grab_cache_page_write_begin(mapping, index, flags);
2337         if (!page) {
2338                 rc = -ENOMEM;
2339                 goto out;
2340         }
2341
2342         if (PageUptodate(page))
2343                 goto out;
2344
2345         /*
2346          * If we write a full page it will be up to date, no need to read from
2347          * the server. If the write is short, we'll end up doing a sync write
2348          * instead.
2349          */
2350         if (len == PAGE_CACHE_SIZE)
2351                 goto out;
2352
2353         /*
2354          * optimize away the read when we have an oplock, and we're not
2355          * expecting to use any of the data we'd be reading in. That
2356          * is, when the page lies beyond the EOF, or straddles the EOF
2357          * and the write will cover all of the existing data.
2358          */
2359         if (CIFS_I(mapping->host)->clientCanCacheRead) {
2360                 i_size = i_size_read(mapping->host);
2361                 if (page_start >= i_size ||
2362                     (offset == 0 && (pos + len) >= i_size)) {
2363                         zero_user_segments(page, 0, offset,
2364                                            offset + len,
2365                                            PAGE_CACHE_SIZE);
2366                         /*
2367                          * PageChecked means that the parts of the page
2368                          * to which we're not writing are considered up
2369                          * to date. Once the data is copied to the
2370                          * page, it can be set uptodate.
2371                          */
2372                         SetPageChecked(page);
2373                         goto out;
2374                 }
2375         }
2376
2377         if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
2378                 /*
2379                  * might as well read a page, it is fast enough. If we get
2380                  * an error, we don't need to return it. cifs_write_end will
2381                  * do a sync write instead since PG_uptodate isn't set.
2382                  */
2383                 cifs_readpage_worker(file, page, &page_start);
2384         } else {
2385                 /* we could try using another file handle if there is one -
2386                    but how would we lock it to prevent close of that handle
2387                    racing with this read? In any case
2388                    this will be written out by write_end so is fine */
2389         }
2390 out:
2391         *pagep = page;
2392         return rc;
2393 }
2394
2395 static int cifs_release_page(struct page *page, gfp_t gfp)
2396 {
2397         if (PagePrivate(page))
2398                 return 0;
2399
2400         return cifs_fscache_release_page(page, gfp);
2401 }
2402
2403 static void cifs_invalidate_page(struct page *page, unsigned long offset)
2404 {
2405         struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
2406
2407         if (offset == 0)
2408                 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
2409 }
2410
2411 static int cifs_launder_page(struct page *page)
2412 {
2413         int rc = 0;
2414         loff_t range_start = page_offset(page);
2415         loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
2416         struct writeback_control wbc = {
2417                 .sync_mode = WB_SYNC_ALL,
2418                 .nr_to_write = 0,
2419                 .range_start = range_start,
2420                 .range_end = range_end,
2421         };
2422
2423         cFYI(1, "Launder page: %p", page);
2424
2425         if (clear_page_dirty_for_io(page))
2426                 rc = cifs_writepage_locked(page, &wbc);
2427
2428         cifs_fscache_invalidate_page(page, page->mapping->host);
2429         return rc;
2430 }
2431
2432 void cifs_oplock_break(struct work_struct *work)
2433 {
2434         struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
2435                                                   oplock_break);
2436         struct inode *inode = cfile->dentry->d_inode;
2437         struct cifsInodeInfo *cinode = CIFS_I(inode);
2438         int rc = 0;
2439
2440         if (inode && S_ISREG(inode->i_mode)) {
2441                 if (cinode->clientCanCacheRead)
2442                         break_lease(inode, O_RDONLY);
2443                 else
2444                         break_lease(inode, O_WRONLY);
2445                 rc = filemap_fdatawrite(inode->i_mapping);
2446                 if (cinode->clientCanCacheRead == 0) {
2447                         rc = filemap_fdatawait(inode->i_mapping);
2448                         mapping_set_error(inode->i_mapping, rc);
2449                         invalidate_remote_inode(inode);
2450                 }
2451                 cFYI(1, "Oplock flush inode %p rc %d", inode, rc);
2452         }
2453
2454         /*
2455          * releasing stale oplock after recent reconnect of smb session using
2456          * a now incorrect file handle is not a data integrity issue but do
2457          * not bother sending an oplock release if session to server still is
2458          * disconnected since oplock already released by the server
2459          */
2460         if (!cfile->oplock_break_cancelled) {
2461                 rc = CIFSSMBLock(0, tlink_tcon(cfile->tlink), cfile->netfid,
2462                                  current->tgid, 0, 0, 0, 0,
2463                                  LOCKING_ANDX_OPLOCK_RELEASE, false,
2464                                  cinode->clientCanCacheRead ? 1 : 0);
2465                 cFYI(1, "Oplock release rc = %d", rc);
2466         }
2467 }
2468
2469 const struct address_space_operations cifs_addr_ops = {
2470         .readpage = cifs_readpage,
2471         .readpages = cifs_readpages,
2472         .writepage = cifs_writepage,
2473         .writepages = cifs_writepages,
2474         .write_begin = cifs_write_begin,
2475         .write_end = cifs_write_end,
2476         .set_page_dirty = __set_page_dirty_nobuffers,
2477         .releasepage = cifs_release_page,
2478         .invalidatepage = cifs_invalidate_page,
2479         .launder_page = cifs_launder_page,
2480 };
2481
2482 /*
2483  * cifs_readpages requires the server to support a buffer large enough to
2484  * contain the header plus one complete page of data.  Otherwise, we need
2485  * to leave cifs_readpages out of the address space operations.
2486  */
2487 const struct address_space_operations cifs_addr_ops_smallbuf = {
2488         .readpage = cifs_readpage,
2489         .writepage = cifs_writepage,
2490         .writepages = cifs_writepages,
2491         .write_begin = cifs_write_begin,
2492         .write_end = cifs_write_end,
2493         .set_page_dirty = __set_page_dirty_nobuffers,
2494         .releasepage = cifs_release_page,
2495         .invalidatepage = cifs_invalidate_page,
2496         .launder_page = cifs_launder_page,
2497 };