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 <linux/swap.h>
  36 #include <asm/div64.h>
  37 #include "cifsfs.h"
  38 #include "cifspdu.h"
  39 #include "cifsglob.h"
  40 #include "cifsproto.h"
  41 #include "cifs_unicode.h"
  42 #include "cifs_debug.h"
  43 #include "cifs_fs_sb.h"
  44 #include "fscache.h"
  45
  46
  47 static inline int cifs_convert_flags(unsigned int flags)
  48 {
  49         if ((flags & O_ACCMODE) == O_RDONLY)
  50                 return GENERIC_READ;
  51         else if ((flags & O_ACCMODE) == O_WRONLY)
  52                 return GENERIC_WRITE;
  53         else if ((flags & O_ACCMODE) == O_RDWR) {
  54                 /* GENERIC_ALL is too much permission to request
  55                    can cause unnecessary access denied on create */
  56                 /* return GENERIC_ALL; */
  57                 return (GENERIC_READ | GENERIC_WRITE);
  58         }
  59
  60         return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
  61                 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
  62                 FILE_READ_DATA);
  63 }
  64
  65 static u32 cifs_posix_convert_flags(unsigned int flags)
  66 {
  67         u32 posix_flags = 0;
  68
  69         if ((flags & O_ACCMODE) == O_RDONLY)
  70                 posix_flags = SMB_O_RDONLY;
  71         else if ((flags & O_ACCMODE) == O_WRONLY)
  72                 posix_flags = SMB_O_WRONLY;
  73         else if ((flags & O_ACCMODE) == O_RDWR)
  74                 posix_flags = SMB_O_RDWR;
  75
  76         if (flags & O_CREAT) {
  77                 posix_flags |= SMB_O_CREAT;
  78                 if (flags & O_EXCL)
  79                         posix_flags |= SMB_O_EXCL;
  80         } else if (flags & O_EXCL)
  81                 cifs_dbg(FYI, "Application %s pid %d has incorrectly set O_EXCL flag but not O_CREAT on file open. Ignoring O_EXCL\n",
  82                          current->comm, current->tgid);
  83
  84         if (flags & O_TRUNC)
  85                 posix_flags |= SMB_O_TRUNC;
  86         /* be safe and imply O_SYNC for O_DSYNC */
  87         if (flags & O_DSYNC)
  88                 posix_flags |= SMB_O_SYNC;
  89         if (flags & O_DIRECTORY)
  90                 posix_flags |= SMB_O_DIRECTORY;
  91         if (flags & O_NOFOLLOW)
  92                 posix_flags |= SMB_O_NOFOLLOW;
  93         if (flags & O_DIRECT)
  94                 posix_flags |= SMB_O_DIRECT;
  95
  96         return posix_flags;
  97 }
  98
  99 static inline int cifs_get_disposition(unsigned int flags)
 100 {
 101         if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
 102                 return FILE_CREATE;
 103         else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
 104                 return FILE_OVERWRITE_IF;
 105         else if ((flags & O_CREAT) == O_CREAT)
 106                 return FILE_OPEN_IF;
 107         else if ((flags & O_TRUNC) == O_TRUNC)
 108                 return FILE_OVERWRITE;
 109         else
 110                 return FILE_OPEN;
 111 }
 112
 113 int cifs_posix_open(char *full_path, struct inode **pinode,
 114                         struct super_block *sb, int mode, unsigned int f_flags,
 115                         __u32 *poplock, __u16 *pnetfid, unsigned int xid)
 116 {
 117         int rc;
 118         FILE_UNIX_BASIC_INFO *presp_data;
 119         __u32 posix_flags = 0;
 120         struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
 121         struct cifs_fattr fattr;
 122         struct tcon_link *tlink;
 123         struct cifs_tcon *tcon;
 124
 125         cifs_dbg(FYI, "posix open %s\n", full_path);
 126
 127         presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
 128         if (presp_data == NULL)
 129                 return -ENOMEM;
 130
 131         tlink = cifs_sb_tlink(cifs_sb);
 132         if (IS_ERR(tlink)) {
 133                 rc = PTR_ERR(tlink);
 134                 goto posix_open_ret;
 135         }
 136
 137         tcon = tlink_tcon(tlink);
 138         mode &= ~current_umask();
 139
 140         posix_flags = cifs_posix_convert_flags(f_flags);
 141         rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
 142                              poplock, full_path, cifs_sb->local_nls,
 143                              cifs_sb->mnt_cifs_flags &
 144                                         CIFS_MOUNT_MAP_SPECIAL_CHR);
 145         cifs_put_tlink(tlink);
 146
 147         if (rc)
 148                 goto posix_open_ret;
 149
 150         if (presp_data->Type == cpu_to_le32(-1))
 151                 goto posix_open_ret; /* open ok, caller does qpathinfo */
 152
 153         if (!pinode)
 154                 goto posix_open_ret; /* caller does not need info */
 155
 156         cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
 157
 158         /* get new inode and set it up */
 159         if (*pinode == NULL) {
 160                 cifs_fill_uniqueid(sb, &fattr);
 161                 *pinode = cifs_iget(sb, &fattr);
 162                 if (!*pinode) {
 163                         rc = -ENOMEM;
 164                         goto posix_open_ret;
 165                 }
 166         } else {
 167                 cifs_fattr_to_inode(*pinode, &fattr);
 168         }
 169
 170 posix_open_ret:
 171         kfree(presp_data);
 172         return rc;
 173 }
 174
 175 static int
 176 cifs_nt_open(char *full_path, struct inode *inode, struct cifs_sb_info *cifs_sb,
 177              struct cifs_tcon *tcon, unsigned int f_flags, __u32 *oplock,
 178              struct cifs_fid *fid, unsigned int xid)
 179 {
 180         int rc;
 181         int desired_access;
 182         int disposition;
 183         int create_options = CREATE_NOT_DIR;
 184         FILE_ALL_INFO *buf;
 185         struct TCP_Server_Info *server = tcon->ses->server;
 186         struct cifs_open_parms oparms;
 187
 188         if (!server->ops->open)
 189                 return -ENOSYS;
 190
 191         desired_access = cifs_convert_flags(f_flags);
 192
 193 /*********************************************************************
 194  *  open flag mapping table:
 195  *
 196  *      POSIX Flag            CIFS Disposition
 197  *      ----------            ----------------
 198  *      O_CREAT               FILE_OPEN_IF
 199  *      O_CREAT | O_EXCL      FILE_CREATE
 200  *      O_CREAT | O_TRUNC     FILE_OVERWRITE_IF
 201  *      O_TRUNC               FILE_OVERWRITE
 202  *      none of the above     FILE_OPEN
 203  *
 204  *      Note that there is not a direct match between disposition
 205  *      FILE_SUPERSEDE (ie create whether or not file exists although
 206  *      O_CREAT | O_TRUNC is similar but truncates the existing
 207  *      file rather than creating a new file as FILE_SUPERSEDE does
 208  *      (which uses the attributes / metadata passed in on open call)
 209  *?
 210  *?  O_SYNC is a reasonable match to CIFS writethrough flag
 211  *?  and the read write flags match reasonably.  O_LARGEFILE
 212  *?  is irrelevant because largefile support is always used
 213  *?  by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
 214  *       O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
 215  *********************************************************************/
 216
 217         disposition = cifs_get_disposition(f_flags);
 218
 219         /* BB pass O_SYNC flag through on file attributes .. BB */
 220
 221         buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
 222         if (!buf)
 223                 return -ENOMEM;
 224
 225         if (backup_cred(cifs_sb))
 226                 create_options |= CREATE_OPEN_BACKUP_INTENT;
 227
 228         oparms.tcon = tcon;
 229         oparms.cifs_sb = cifs_sb;
 230         oparms.desired_access = desired_access;
 231         oparms.create_options = create_options;
 232         oparms.disposition = disposition;
 233         oparms.path = full_path;
 234         oparms.fid = fid;
 235         oparms.reconnect = false;
 236
 237         rc = server->ops->open(xid, &oparms, oplock, buf);
 238
 239         if (rc)
 240                 goto out;
 241
 242         if (tcon->unix_ext)
 243                 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
 244                                               xid);
 245         else
 246                 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
 247                                          xid, fid);
 248
 249 out:
 250         kfree(buf);
 251         return rc;
 252 }
 253
 254 static bool
 255 cifs_has_mand_locks(struct cifsInodeInfo *cinode)
 256 {
 257         struct cifs_fid_locks *cur;
 258         bool has_locks = false;
 259
 260         down_read(&cinode->lock_sem);
 261         list_for_each_entry(cur, &cinode->llist, llist) {
 262                 if (!list_empty(&cur->locks)) {
 263                         has_locks = true;
 264                         break;
 265                 }
 266         }
 267         up_read(&cinode->lock_sem);
 268         return has_locks;
 269 }
 270
 271 struct cifsFileInfo *
 272 cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
 273                   struct tcon_link *tlink, __u32 oplock)
 274 {
 275         struct dentry *dentry = file->f_path.dentry;
 276         struct inode *inode = dentry->d_inode;
 277         struct cifsInodeInfo *cinode = CIFS_I(inode);
 278         struct cifsFileInfo *cfile;
 279         struct cifs_fid_locks *fdlocks;
 280         struct cifs_tcon *tcon = tlink_tcon(tlink);
 281         struct TCP_Server_Info *server = tcon->ses->server;
 282
 283         cfile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
 284         if (cfile == NULL)
 285                 return cfile;
 286
 287         fdlocks = kzalloc(sizeof(struct cifs_fid_locks), GFP_KERNEL);
 288         if (!fdlocks) {
 289                 kfree(cfile);
 290                 return NULL;
 291         }
 292
 293         INIT_LIST_HEAD(&fdlocks->locks);
 294         fdlocks->cfile = cfile;
 295         cfile->llist = fdlocks;
 296         down_write(&cinode->lock_sem);
 297         list_add(&fdlocks->llist, &cinode->llist);
 298         up_write(&cinode->lock_sem);
 299
 300         cfile->count = 1;
 301         cfile->pid = current->tgid;
 302         cfile->uid = current_fsuid();
 303         cfile->dentry = dget(dentry);
 304         cfile->f_flags = file->f_flags;
 305         cfile->invalidHandle = false;
 306         cfile->tlink = cifs_get_tlink(tlink);
 307         INIT_WORK(&cfile->oplock_break, cifs_oplock_break);
 308         mutex_init(&cfile->fh_mutex);
 309
 310         cifs_sb_active(inode->i_sb);
 311
 312         /*
 313          * If the server returned a read oplock and we have mandatory brlocks,
 314          * set oplock level to None.
 315          */
 316         if (server->ops->is_read_op(oplock) && cifs_has_mand_locks(cinode)) {
 317                 cifs_dbg(FYI, "Reset oplock val from read to None due to mand locks\n");
 318                 oplock = 0;
 319         }
 320
 321         spin_lock(&cifs_file_list_lock);
 322         if (fid->pending_open->oplock != CIFS_OPLOCK_NO_CHANGE && oplock)
 323                 oplock = fid->pending_open->oplock;
 324         list_del(&fid->pending_open->olist);
 325
 326         fid->purge_cache = false;
 327         server->ops->set_fid(cfile, fid, oplock);
 328
 329         list_add(&cfile->tlist, &tcon->openFileList);
 330         /* if readable file instance put first in list*/
 331         if (file->f_mode & FMODE_READ)
 332                 list_add(&cfile->flist, &cinode->openFileList);
 333         else
 334                 list_add_tail(&cfile->flist, &cinode->openFileList);
 335         spin_unlock(&cifs_file_list_lock);
 336
 337         if (fid->purge_cache)
 338                 cifs_invalidate_mapping(inode);
 339
 340         file->private_data = cfile;
 341         return cfile;
 342 }
 343
 344 struct cifsFileInfo *
 345 cifsFileInfo_get(struct cifsFileInfo *cifs_file)
 346 {
 347         spin_lock(&cifs_file_list_lock);
 348         cifsFileInfo_get_locked(cifs_file);
 349         spin_unlock(&cifs_file_list_lock);
 350         return cifs_file;
 351 }
 352
 353 /*
 354  * Release a reference on the file private data. This may involve closing
 355  * the filehandle out on the server. Must be called without holding
 356  * cifs_file_list_lock.
 357  */
 358 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
 359 {
 360         struct inode *inode = cifs_file->dentry->d_inode;
 361         struct cifs_tcon *tcon = tlink_tcon(cifs_file->tlink);
 362         struct TCP_Server_Info *server = tcon->ses->server;
 363         struct cifsInodeInfo *cifsi = CIFS_I(inode);
 364         struct super_block *sb = inode->i_sb;
 365         struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
 366         struct cifsLockInfo *li, *tmp;
 367         struct cifs_fid fid;
 368         struct cifs_pending_open open;
 369
 370         spin_lock(&cifs_file_list_lock);
 371         if (--cifs_file->count > 0) {
 372                 spin_unlock(&cifs_file_list_lock);
 373                 return;
 374         }
 375
 376         if (server->ops->get_lease_key)
 377                 server->ops->get_lease_key(inode, &fid);
 378
 379         /* store open in pending opens to make sure we don't miss lease break */
 380         cifs_add_pending_open_locked(&fid, cifs_file->tlink, &open);
 381
 382         /* remove it from the lists */
 383         list_del(&cifs_file->flist);
 384         list_del(&cifs_file->tlist);
 385
 386         if (list_empty(&cifsi->openFileList)) {
 387                 cifs_dbg(FYI, "closing last open instance for inode %p\n",
 388                          cifs_file->dentry->d_inode);
 389                 /*
 390                  * In strict cache mode we need invalidate mapping on the last
 391                  * close  because it may cause a error when we open this file
 392                  * again and get at least level II oplock.
 393                  */
 394                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
 395                         CIFS_I(inode)->invalid_mapping = true;
 396                 cifs_set_oplock_level(cifsi, 0);
 397         }
 398         spin_unlock(&cifs_file_list_lock);
 399
 400         cancel_work_sync(&cifs_file->oplock_break);
 401
 402         if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
 403                 struct TCP_Server_Info *server = tcon->ses->server;
 404                 unsigned int xid;
 405
 406                 xid = get_xid();
 407                 if (server->ops->close)
 408                         server->ops->close(xid, tcon, &cifs_file->fid);
 409                 _free_xid(xid);
 410         }
 411
 412         cifs_del_pending_open(&open);
 413
 414         /*
 415          * Delete any outstanding lock records. We'll lose them when the file
 416          * is closed anyway.
 417          */
 418         down_write(&cifsi->lock_sem);
 419         list_for_each_entry_safe(li, tmp, &cifs_file->llist->locks, llist) {
 420                 list_del(&li->llist);
 421                 cifs_del_lock_waiters(li);
 422                 kfree(li);
 423         }
 424         list_del(&cifs_file->llist->llist);
 425         kfree(cifs_file->llist);
 426         up_write(&cifsi->lock_sem);
 427
 428         cifs_put_tlink(cifs_file->tlink);
 429         dput(cifs_file->dentry);
 430         cifs_sb_deactive(sb);
 431         kfree(cifs_file);
 432 }
 433
 434 int cifs_open(struct inode *inode, struct file *file)
 435
 436 {
 437         int rc = -EACCES;
 438         unsigned int xid;
 439         __u32 oplock;
 440         struct cifs_sb_info *cifs_sb;
 441         struct TCP_Server_Info *server;
 442         struct cifs_tcon *tcon;
 443         struct tcon_link *tlink;
 444         struct cifsFileInfo *cfile = NULL;
 445         char *full_path = NULL;
 446         bool posix_open_ok = false;
 447         struct cifs_fid fid;
 448         struct cifs_pending_open open;
 449
 450         xid = get_xid();
 451
 452         cifs_sb = CIFS_SB(inode->i_sb);
 453         tlink = cifs_sb_tlink(cifs_sb);
 454         if (IS_ERR(tlink)) {
 455                 free_xid(xid);
 456                 return PTR_ERR(tlink);
 457         }
 458         tcon = tlink_tcon(tlink);
 459         server = tcon->ses->server;
 460
 461         full_path = build_path_from_dentry(file->f_path.dentry);
 462         if (full_path == NULL) {
 463                 rc = -ENOMEM;
 464                 goto out;
 465         }
 466
 467         cifs_dbg(FYI, "inode = 0x%p file flags are 0x%x for %s\n",
 468                  inode, file->f_flags, full_path);
 469
 470         if (server->oplocks)
 471                 oplock = REQ_OPLOCK;
 472         else
 473                 oplock = 0;
 474
 475         if (!tcon->broken_posix_open && tcon->unix_ext &&
 476             cap_unix(tcon->ses) && (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 477                                 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 478                 /* can not refresh inode info since size could be stale */
 479                 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
 480                                 cifs_sb->mnt_file_mode /* ignored */,
 481                                 file->f_flags, &oplock, &fid.netfid, xid);
 482                 if (rc == 0) {
 483                         cifs_dbg(FYI, "posix open succeeded\n");
 484                         posix_open_ok = true;
 485                 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
 486                         if (tcon->ses->serverNOS)
 487                                 cifs_dbg(VFS, "server %s of type %s returned unexpected error on SMB posix open, disabling posix open support. Check if server update available.\n",
 488                                          tcon->ses->serverName,
 489                                          tcon->ses->serverNOS);
 490                         tcon->broken_posix_open = true;
 491                 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
 492                          (rc != -EOPNOTSUPP)) /* path not found or net err */
 493                         goto out;
 494                 /*
 495                  * Else fallthrough to retry open the old way on network i/o
 496                  * or DFS errors.
 497                  */
 498         }
 499
 500         if (server->ops->get_lease_key)
 501                 server->ops->get_lease_key(inode, &fid);
 502
 503         cifs_add_pending_open(&fid, tlink, &open);
 504
 505         if (!posix_open_ok) {
 506                 if (server->ops->get_lease_key)
 507                         server->ops->get_lease_key(inode, &fid);
 508
 509                 rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
 510                                   file->f_flags, &oplock, &fid, xid);
 511                 if (rc) {
 512                         cifs_del_pending_open(&open);
 513                         goto out;
 514                 }
 515         }
 516
 517         cfile = cifs_new_fileinfo(&fid, file, tlink, oplock);
 518         if (cfile == NULL) {
 519                 if (server->ops->close)
 520                         server->ops->close(xid, tcon, &fid);
 521                 cifs_del_pending_open(&open);
 522                 rc = -ENOMEM;
 523                 goto out;
 524         }
 525
 526         cifs_fscache_set_inode_cookie(inode, file);
 527
 528         if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
 529                 /*
 530                  * Time to set mode which we can not set earlier due to
 531                  * problems creating new read-only files.
 532                  */
 533                 struct cifs_unix_set_info_args args = {
 534                         .mode   = inode->i_mode,
 535                         .uid    = INVALID_UID, /* no change */
 536                         .gid    = INVALID_GID, /* no change */
 537                         .ctime  = NO_CHANGE_64,
 538                         .atime  = NO_CHANGE_64,
 539                         .mtime  = NO_CHANGE_64,
 540                         .device = 0,
 541                 };
 542                 CIFSSMBUnixSetFileInfo(xid, tcon, &args, fid.netfid,
 543                                        cfile->pid);
 544         }
 545
 546 out:
 547         kfree(full_path);
 548         free_xid(xid);
 549         cifs_put_tlink(tlink);
 550         return rc;
 551 }
 552
 553 static int cifs_push_posix_locks(struct cifsFileInfo *cfile);
 554
 555 /*
 556  * Try to reacquire byte range locks that were released when session
 557  * to server was lost.
 558  */
 559 static int
 560 cifs_relock_file(struct cifsFileInfo *cfile)
 561 {
 562         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
 563         struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
 564         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
 565         int rc = 0;
 566
 567         down_read(&cinode->lock_sem);
 568         if (cinode->can_cache_brlcks) {
 569                 /* can cache locks - no need to relock */
 570                 up_read(&cinode->lock_sem);
 571                 return rc;
 572         }
 573
 574         if (cap_unix(tcon->ses) &&
 575             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
 576             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
 577                 rc = cifs_push_posix_locks(cfile);
 578         else
 579                 rc = tcon->ses->server->ops->push_mand_locks(cfile);
 580
 581         up_read(&cinode->lock_sem);
 582         return rc;
 583 }
 584
 585 static int
 586 cifs_reopen_file(struct cifsFileInfo *cfile, bool can_flush)
 587 {
 588         int rc = -EACCES;
 589         unsigned int xid;
 590         __u32 oplock;
 591         struct cifs_sb_info *cifs_sb;
 592         struct cifs_tcon *tcon;
 593         struct TCP_Server_Info *server;
 594         struct cifsInodeInfo *cinode;
 595         struct inode *inode;
 596         char *full_path = NULL;
 597         int desired_access;
 598         int disposition = FILE_OPEN;
 599         int create_options = CREATE_NOT_DIR;
 600         struct cifs_open_parms oparms;
 601
 602         xid = get_xid();
 603         mutex_lock(&cfile->fh_mutex);
 604         if (!cfile->invalidHandle) {
 605                 mutex_unlock(&cfile->fh_mutex);
 606                 rc = 0;
 607                 free_xid(xid);
 608                 return rc;
 609         }
 610
 611         inode = cfile->dentry->d_inode;
 612         cifs_sb = CIFS_SB(inode->i_sb);
 613         tcon = tlink_tcon(cfile->tlink);
 614         server = tcon->ses->server;
 615
 616         /*
 617          * Can not grab rename sem here because various ops, including those
 618          * that already have the rename sem can end up causing writepage to get
 619          * called and if the server was down that means we end up here, and we
 620          * can never tell if the caller already has the rename_sem.
 621          */
 622         full_path = build_path_from_dentry(cfile->dentry);
 623         if (full_path == NULL) {
 624                 rc = -ENOMEM;
 625                 mutex_unlock(&cfile->fh_mutex);
 626                 free_xid(xid);
 627                 return rc;
 628         }
 629
 630         cifs_dbg(FYI, "inode = 0x%p file flags 0x%x for %s\n",
 631                  inode, cfile->f_flags, full_path);
 632
 633         if (tcon->ses->server->oplocks)
 634                 oplock = REQ_OPLOCK;
 635         else
 636                 oplock = 0;
 637
 638         if (tcon->unix_ext && cap_unix(tcon->ses) &&
 639             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 640                                 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 641                 /*
 642                  * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
 643                  * original open. Must mask them off for a reopen.
 644                  */
 645                 unsigned int oflags = cfile->f_flags &
 646                                                 ~(O_CREAT | O_EXCL | O_TRUNC);
 647
 648                 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
 649                                      cifs_sb->mnt_file_mode /* ignored */,
 650                                      oflags, &oplock, &cfile->fid.netfid, xid);
 651                 if (rc == 0) {
 652                         cifs_dbg(FYI, "posix reopen succeeded\n");
 653                         oparms.reconnect = true;
 654                         goto reopen_success;
 655                 }
 656                 /*
 657                  * fallthrough to retry open the old way on errors, especially
 658                  * in the reconnect path it is important to retry hard
 659                  */
 660         }
 661
 662         desired_access = cifs_convert_flags(cfile->f_flags);
 663
 664         if (backup_cred(cifs_sb))
 665                 create_options |= CREATE_OPEN_BACKUP_INTENT;
 666
 667         if (server->ops->get_lease_key)
 668                 server->ops->get_lease_key(inode, &cfile->fid);
 669
 670         oparms.tcon = tcon;
 671         oparms.cifs_sb = cifs_sb;
 672         oparms.desired_access = desired_access;
 673         oparms.create_options = create_options;
 674         oparms.disposition = disposition;
 675         oparms.path = full_path;
 676         oparms.fid = &cfile->fid;
 677         oparms.reconnect = true;
 678
 679         /*
 680          * Can not refresh inode by passing in file_info buf to be returned by
 681          * ops->open and then calling get_inode_info with returned buf since
 682          * file might have write behind data that needs to be flushed and server
 683          * version of file size can be stale. If we knew for sure that inode was
 684          * not dirty locally we could do this.
 685          */
 686         rc = server->ops->open(xid, &oparms, &oplock, NULL);
 687         if (rc == -ENOENT && oparms.reconnect == false) {
 688                 /* durable handle timeout is expired - open the file again */
 689                 rc = server->ops->open(xid, &oparms, &oplock, NULL);
 690                 /* indicate that we need to relock the file */
 691                 oparms.reconnect = true;
 692         }
 693
 694         if (rc) {
 695                 mutex_unlock(&cfile->fh_mutex);
 696                 cifs_dbg(FYI, "cifs_reopen returned 0x%x\n", rc);
 697                 cifs_dbg(FYI, "oplock: %d\n", oplock);
 698                 goto reopen_error_exit;
 699         }
 700
 701 reopen_success:
 702         cfile->invalidHandle = false;
 703         mutex_unlock(&cfile->fh_mutex);
 704         cinode = CIFS_I(inode);
 705
 706         if (can_flush) {
 707                 rc = filemap_write_and_wait(inode->i_mapping);
 708                 mapping_set_error(inode->i_mapping, rc);
 709
 710                 if (tcon->unix_ext)
 711                         rc = cifs_get_inode_info_unix(&inode, full_path,
 712                                                       inode->i_sb, xid);
 713                 else
 714                         rc = cifs_get_inode_info(&inode, full_path, NULL,
 715                                                  inode->i_sb, xid, NULL);
 716         }
 717         /*
 718          * Else we are writing out data to server already and could deadlock if
 719          * we tried to flush data, and since we do not know if we have data that
 720          * would invalidate the current end of file on the server we can not go
 721          * to the server to get the new inode info.
 722          */
 723
 724         server->ops->set_fid(cfile, &cfile->fid, oplock);
 725         if (oparms.reconnect)
 726                 cifs_relock_file(cfile);
 727
 728 reopen_error_exit:
 729         kfree(full_path);
 730         free_xid(xid);
 731         return rc;
 732 }
 733
 734 int cifs_close(struct inode *inode, struct file *file)
 735 {
 736         if (file->private_data != NULL) {
 737                 cifsFileInfo_put(file->private_data);
 738                 file->private_data = NULL;
 739         }
 740
 741         /* return code from the ->release op is always ignored */
 742         return 0;
 743 }
 744
 745 int cifs_closedir(struct inode *inode, struct file *file)
 746 {
 747         int rc = 0;
 748         unsigned int xid;
 749         struct cifsFileInfo *cfile = file->private_data;
 750         struct cifs_tcon *tcon;
 751         struct TCP_Server_Info *server;
 752         char *buf;
 753
 754         cifs_dbg(FYI, "Closedir inode = 0x%p\n", inode);
 755
 756         if (cfile == NULL)
 757                 return rc;
 758
 759         xid = get_xid();
 760         tcon = tlink_tcon(cfile->tlink);
 761         server = tcon->ses->server;
 762
 763         cifs_dbg(FYI, "Freeing private data in close dir\n");
 764         spin_lock(&cifs_file_list_lock);
 765         if (!cfile->srch_inf.endOfSearch && !cfile->invalidHandle) {
 766                 cfile->invalidHandle = true;
 767                 spin_unlock(&cifs_file_list_lock);
 768                 if (server->ops->close_dir)
 769                         rc = server->ops->close_dir(xid, tcon, &cfile->fid);
 770                 else
 771                         rc = -ENOSYS;
 772                 cifs_dbg(FYI, "Closing uncompleted readdir with rc %d\n", rc);
 773                 /* not much we can do if it fails anyway, ignore rc */
 774                 rc = 0;
 775         } else
 776                 spin_unlock(&cifs_file_list_lock);
 777
 778         buf = cfile->srch_inf.ntwrk_buf_start;
 779         if (buf) {
 780                 cifs_dbg(FYI, "closedir free smb buf in srch struct\n");
 781                 cfile->srch_inf.ntwrk_buf_start = NULL;
 782                 if (cfile->srch_inf.smallBuf)
 783                         cifs_small_buf_release(buf);
 784                 else
 785                         cifs_buf_release(buf);
 786         }
 787
 788         cifs_put_tlink(cfile->tlink);
 789         kfree(file->private_data);
 790         file->private_data = NULL;
 791         /* BB can we lock the filestruct while this is going on? */
 792         free_xid(xid);
 793         return rc;
 794 }
 795
 796 static struct cifsLockInfo *
 797 cifs_lock_init(__u64 offset, __u64 length, __u8 type)
 798 {
 799         struct cifsLockInfo *lock =
 800                 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
 801         if (!lock)
 802                 return lock;
 803         lock->offset = offset;
 804         lock->length = length;
 805         lock->type = type;
 806         lock->pid = current->tgid;
 807         INIT_LIST_HEAD(&lock->blist);
 808         init_waitqueue_head(&lock->block_q);
 809         return lock;
 810 }
 811
 812 void
 813 cifs_del_lock_waiters(struct cifsLockInfo *lock)
 814 {
 815         struct cifsLockInfo *li, *tmp;
 816         list_for_each_entry_safe(li, tmp, &lock->blist, blist) {
 817                 list_del_init(&li->blist);
 818                 wake_up(&li->block_q);
 819         }
 820 }
 821
 822 #define CIFS_LOCK_OP    0
 823 #define CIFS_READ_OP    1
 824 #define CIFS_WRITE_OP   2
 825
 826 /* @rw_check : 0 - no op, 1 - read, 2 - write */
 827 static bool
 828 cifs_find_fid_lock_conflict(struct cifs_fid_locks *fdlocks, __u64 offset,
 829                             __u64 length, __u8 type, struct cifsFileInfo *cfile,
 830                             struct cifsLockInfo **conf_lock, int rw_check)
 831 {
 832         struct cifsLockInfo *li;
 833         struct cifsFileInfo *cur_cfile = fdlocks->cfile;
 834         struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
 835
 836         list_for_each_entry(li, &fdlocks->locks, llist) {
 837                 if (offset + length <= li->offset ||
 838                     offset >= li->offset + li->length)
 839                         continue;
 840                 if (rw_check != CIFS_LOCK_OP && current->tgid == li->pid &&
 841                     server->ops->compare_fids(cfile, cur_cfile)) {
 842                         /* shared lock prevents write op through the same fid */
 843                         if (!(li->type & server->vals->shared_lock_type) ||
 844                             rw_check != CIFS_WRITE_OP)
 845                                 continue;
 846                 }
 847                 if ((type & server->vals->shared_lock_type) &&
 848                     ((server->ops->compare_fids(cfile, cur_cfile) &&
 849                      current->tgid == li->pid) || type == li->type))
 850                         continue;
 851                 if (conf_lock)
 852                         *conf_lock = li;
 853                 return true;
 854         }
 855         return false;
 856 }
 857
 858 bool
 859 cifs_find_lock_conflict(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
 860                         __u8 type, struct cifsLockInfo **conf_lock,
 861                         int rw_check)
 862 {
 863         bool rc = false;
 864         struct cifs_fid_locks *cur;
 865         struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
 866
 867         list_for_each_entry(cur, &cinode->llist, llist) {
 868                 rc = cifs_find_fid_lock_conflict(cur, offset, length, type,
 869                                                  cfile, conf_lock, rw_check);
 870                 if (rc)
 871                         break;
 872         }
 873
 874         return rc;
 875 }
 876
 877 /*
 878  * Check if there is another lock that prevents us to set the lock (mandatory
 879  * style). If such a lock exists, update the flock structure with its
 880  * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
 881  * or leave it the same if we can't. Returns 0 if we don't need to request to
 882  * the server or 1 otherwise.
 883  */
 884 static int
 885 cifs_lock_test(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
 886                __u8 type, struct file_lock *flock)
 887 {
 888         int rc = 0;
 889         struct cifsLockInfo *conf_lock;
 890         struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
 891         struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
 892         bool exist;
 893
 894         down_read(&cinode->lock_sem);
 895
 896         exist = cifs_find_lock_conflict(cfile, offset, length, type,
 897                                         &conf_lock, CIFS_LOCK_OP);
 898         if (exist) {
 899                 flock->fl_start = conf_lock->offset;
 900                 flock->fl_end = conf_lock->offset + conf_lock->length - 1;
 901                 flock->fl_pid = conf_lock->pid;
 902                 if (conf_lock->type & server->vals->shared_lock_type)
 903                         flock->fl_type = F_RDLCK;
 904                 else
 905                         flock->fl_type = F_WRLCK;
 906         } else if (!cinode->can_cache_brlcks)
 907                 rc = 1;
 908         else
 909                 flock->fl_type = F_UNLCK;
 910
 911         up_read(&cinode->lock_sem);
 912         return rc;
 913 }
 914
 915 static void
 916 cifs_lock_add(struct cifsFileInfo *cfile, struct cifsLockInfo *lock)
 917 {
 918         struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
 919         down_write(&cinode->lock_sem);
 920         list_add_tail(&lock->llist, &cfile->llist->locks);
 921         up_write(&cinode->lock_sem);
 922 }
 923
 924 /*
 925  * Set the byte-range lock (mandatory style). Returns:
 926  * 1) 0, if we set the lock and don't need to request to the server;
 927  * 2) 1, if no locks prevent us but we need to request to the server;
 928  * 3) -EACCESS, if there is a lock that prevents us and wait is false.
 929  */
 930 static int
 931 cifs_lock_add_if(struct cifsFileInfo *cfile, struct cifsLockInfo *lock,
 932                  bool wait)
 933 {
 934         struct cifsLockInfo *conf_lock;
 935         struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
 936         bool exist;
 937         int rc = 0;
 938
 939 try_again:
 940         exist = false;
 941         down_write(&cinode->lock_sem);
 942
 943         exist = cifs_find_lock_conflict(cfile, lock->offset, lock->length,
 944                                         lock->type, &conf_lock, CIFS_LOCK_OP);
 945         if (!exist && cinode->can_cache_brlcks) {
 946                 list_add_tail(&lock->llist, &cfile->llist->locks);
 947                 up_write(&cinode->lock_sem);
 948                 return rc;
 949         }
 950
 951         if (!exist)
 952                 rc = 1;
 953         else if (!wait)
 954                 rc = -EACCES;
 955         else {
 956                 list_add_tail(&lock->blist, &conf_lock->blist);
 957                 up_write(&cinode->lock_sem);
 958                 rc = wait_event_interruptible(lock->block_q,
 959                                         (lock->blist.prev == &lock->blist) &&
 960                                         (lock->blist.next == &lock->blist));
 961                 if (!rc)
 962                         goto try_again;
 963                 down_write(&cinode->lock_sem);
 964                 list_del_init(&lock->blist);
 965         }
 966
 967         up_write(&cinode->lock_sem);
 968         return rc;
 969 }
 970
 971 /*
 972  * Check if there is another lock that prevents us to set the lock (posix
 973  * style). If such a lock exists, update the flock structure with its
 974  * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
 975  * or leave it the same if we can't. Returns 0 if we don't need to request to
 976  * the server or 1 otherwise.
 977  */
 978 static int
 979 cifs_posix_lock_test(struct file *file, struct file_lock *flock)
 980 {
 981         int rc = 0;
 982         struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
 983         unsigned char saved_type = flock->fl_type;
 984
 985         if ((flock->fl_flags & FL_POSIX) == 0)
 986                 return 1;
 987
 988         down_read(&cinode->lock_sem);
 989         posix_test_lock(file, flock);
 990
 991         if (flock->fl_type == F_UNLCK && !cinode->can_cache_brlcks) {
 992                 flock->fl_type = saved_type;
 993                 rc = 1;
 994         }
 995
 996         up_read(&cinode->lock_sem);
 997         return rc;
 998 }
 999
1000 /*
1001  * Set the byte-range lock (posix style). Returns:
1002  * 1) 0, if we set the lock and don't need to request to the server;
1003  * 2) 1, if we need to request to the server;
1004  * 3) <0, if the error occurs while setting the lock.
1005  */
1006 static int
1007 cifs_posix_lock_set(struct file *file, struct file_lock *flock)
1008 {
1009         struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
1010         int rc = 1;
1011
1012         if ((flock->fl_flags & FL_POSIX) == 0)
1013                 return rc;
1014
1015 try_again:
1016         down_write(&cinode->lock_sem);
1017         if (!cinode->can_cache_brlcks) {
1018                 up_write(&cinode->lock_sem);
1019                 return rc;
1020         }
1021
1022         rc = posix_lock_file(file, flock, NULL);
1023         up_write(&cinode->lock_sem);
1024         if (rc == FILE_LOCK_DEFERRED) {
1025                 rc = wait_event_interruptible(flock->fl_wait, !flock->fl_next);
1026                 if (!rc)
1027                         goto try_again;
1028                 posix_unblock_lock(flock);
1029         }
1030         return rc;
1031 }
1032
1033 int
1034 cifs_push_mandatory_locks(struct cifsFileInfo *cfile)
1035 {
1036         unsigned int xid;
1037         int rc = 0, stored_rc;
1038         struct cifsLockInfo *li, *tmp;
1039         struct cifs_tcon *tcon;
1040         unsigned int num, max_num, max_buf;
1041         LOCKING_ANDX_RANGE *buf, *cur;
1042         int types[] = {LOCKING_ANDX_LARGE_FILES,
1043                        LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES};
1044         int i;
1045
1046         xid = get_xid();
1047         tcon = tlink_tcon(cfile->tlink);
1048
1049         /*
1050          * Accessing maxBuf is racy with cifs_reconnect - need to store value
1051          * and check it for zero before using.
1052          */
1053         max_buf = tcon->ses->server->maxBuf;
1054         if (!max_buf) {
1055                 free_xid(xid);
1056                 return -EINVAL;
1057         }
1058
1059         max_num = (max_buf - sizeof(struct smb_hdr)) /
1060                                                 sizeof(LOCKING_ANDX_RANGE);
1061         buf = kzalloc(max_num * sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1062         if (!buf) {
1063                 free_xid(xid);
1064                 return -ENOMEM;
1065         }
1066
1067         for (i = 0; i < 2; i++) {
1068                 cur = buf;
1069                 num = 0;
1070                 list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1071                         if (li->type != types[i])
1072                                 continue;
1073                         cur->Pid = cpu_to_le16(li->pid);
1074                         cur->LengthLow = cpu_to_le32((u32)li->length);
1075                         cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1076                         cur->OffsetLow = cpu_to_le32((u32)li->offset);
1077                         cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1078                         if (++num == max_num) {
1079                                 stored_rc = cifs_lockv(xid, tcon,
1080                                                        cfile->fid.netfid,
1081                                                        (__u8)li->type, 0, num,
1082                                                        buf);
1083                                 if (stored_rc)
1084                                         rc = stored_rc;
1085                                 cur = buf;
1086                                 num = 0;
1087                         } else
1088                                 cur++;
1089                 }
1090
1091                 if (num) {
1092                         stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1093                                                (__u8)types[i], 0, num, buf);
1094                         if (stored_rc)
1095                                 rc = stored_rc;
1096                 }
1097         }
1098
1099         kfree(buf);
1100         free_xid(xid);
1101         return rc;
1102 }
1103
1104 /* copied from fs/locks.c with a name change */
1105 #define cifs_for_each_lock(inode, lockp) \
1106         for (lockp = &inode->i_flock; *lockp != NULL; \
1107              lockp = &(*lockp)->fl_next)
1108
1109 struct lock_to_push {
1110         struct list_head llist;
1111         __u64 offset;
1112         __u64 length;
1113         __u32 pid;
1114         __u16 netfid;
1115         __u8 type;
1116 };
1117
1118 static int
1119 cifs_push_posix_locks(struct cifsFileInfo *cfile)
1120 {
1121         struct inode *inode = cfile->dentry->d_inode;
1122         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1123         struct file_lock *flock, **before;
1124         unsigned int count = 0, i = 0;
1125         int rc = 0, xid, type;
1126         struct list_head locks_to_send, *el;
1127         struct lock_to_push *lck, *tmp;
1128         __u64 length;
1129
1130         xid = get_xid();
1131
1132         spin_lock(&inode->i_lock);
1133         cifs_for_each_lock(inode, before) {
1134                 if ((*before)->fl_flags & FL_POSIX)
1135                         count++;
1136         }
1137         spin_unlock(&inode->i_lock);
1138
1139         INIT_LIST_HEAD(&locks_to_send);
1140
1141         /*
1142          * Allocating count locks is enough because no FL_POSIX locks can be
1143          * added to the list while we are holding cinode->lock_sem that
1144          * protects locking operations of this inode.
1145          */
1146         for (; i < count; i++) {
1147                 lck = kmalloc(sizeof(struct lock_to_push), GFP_KERNEL);
1148                 if (!lck) {
1149                         rc = -ENOMEM;
1150                         goto err_out;
1151                 }
1152                 list_add_tail(&lck->llist, &locks_to_send);
1153         }
1154
1155         el = locks_to_send.next;
1156         spin_lock(&inode->i_lock);
1157         cifs_for_each_lock(inode, before) {
1158                 flock = *before;
1159                 if ((flock->fl_flags & FL_POSIX) == 0)
1160                         continue;
1161                 if (el == &locks_to_send) {
1162                         /*
1163                          * The list ended. We don't have enough allocated
1164                          * structures - something is really wrong.
1165                          */
1166                         cifs_dbg(VFS, "Can't push all brlocks!\n");
1167                         break;
1168                 }
1169                 length = 1 + flock->fl_end - flock->fl_start;
1170                 if (flock->fl_type == F_RDLCK || flock->fl_type == F_SHLCK)
1171                         type = CIFS_RDLCK;
1172                 else
1173                         type = CIFS_WRLCK;
1174                 lck = list_entry(el, struct lock_to_push, llist);
1175                 lck->pid = flock->fl_pid;
1176                 lck->netfid = cfile->fid.netfid;
1177                 lck->length = length;
1178                 lck->type = type;
1179                 lck->offset = flock->fl_start;
1180                 el = el->next;
1181         }
1182         spin_unlock(&inode->i_lock);
1183
1184         list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1185                 int stored_rc;
1186
1187                 stored_rc = CIFSSMBPosixLock(xid, tcon, lck->netfid, lck->pid,
1188                                              lck->offset, lck->length, NULL,
1189                                              lck->type, 0);
1190                 if (stored_rc)
1191                         rc = stored_rc;
1192                 list_del(&lck->llist);
1193                 kfree(lck);
1194         }
1195
1196 out:
1197         free_xid(xid);
1198         return rc;
1199 err_out:
1200         list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1201                 list_del(&lck->llist);
1202                 kfree(lck);
1203         }
1204         goto out;
1205 }
1206
1207 static int
1208 cifs_push_locks(struct cifsFileInfo *cfile)
1209 {
1210         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
1211         struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
1212         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1213         int rc = 0;
1214
1215         /* we are going to update can_cache_brlcks here - need a write access */
1216         down_write(&cinode->lock_sem);
1217         if (!cinode->can_cache_brlcks) {
1218                 up_write(&cinode->lock_sem);
1219                 return rc;
1220         }
1221
1222         if (cap_unix(tcon->ses) &&
1223             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1224             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1225                 rc = cifs_push_posix_locks(cfile);
1226         else
1227                 rc = tcon->ses->server->ops->push_mand_locks(cfile);
1228
1229         cinode->can_cache_brlcks = false;
1230         up_write(&cinode->lock_sem);
1231         return rc;
1232 }
1233
1234 static void
1235 cifs_read_flock(struct file_lock *flock, __u32 *type, int *lock, int *unlock,
1236                 bool *wait_flag, struct TCP_Server_Info *server)
1237 {
1238         if (flock->fl_flags & FL_POSIX)
1239                 cifs_dbg(FYI, "Posix\n");
1240         if (flock->fl_flags & FL_FLOCK)
1241                 cifs_dbg(FYI, "Flock\n");
1242         if (flock->fl_flags & FL_SLEEP) {
1243                 cifs_dbg(FYI, "Blocking lock\n");
1244                 *wait_flag = true;
1245         }
1246         if (flock->fl_flags & FL_ACCESS)
1247                 cifs_dbg(FYI, "Process suspended by mandatory locking - not implemented yet\n");
1248         if (flock->fl_flags & FL_LEASE)
1249                 cifs_dbg(FYI, "Lease on file - not implemented yet\n");
1250         if (flock->fl_flags &
1251             (~(FL_POSIX | FL_FLOCK | FL_SLEEP |
1252                FL_ACCESS | FL_LEASE | FL_CLOSE)))
1253                 cifs_dbg(FYI, "Unknown lock flags 0x%x\n", flock->fl_flags);
1254
1255         *type = server->vals->large_lock_type;
1256         if (flock->fl_type == F_WRLCK) {
1257                 cifs_dbg(FYI, "F_WRLCK\n");
1258                 *type |= server->vals->exclusive_lock_type;
1259                 *lock = 1;
1260         } else if (flock->fl_type == F_UNLCK) {
1261                 cifs_dbg(FYI, "F_UNLCK\n");
1262                 *type |= server->vals->unlock_lock_type;
1263                 *unlock = 1;
1264                 /* Check if unlock includes more than one lock range */
1265         } else if (flock->fl_type == F_RDLCK) {
1266                 cifs_dbg(FYI, "F_RDLCK\n");
1267                 *type |= server->vals->shared_lock_type;
1268                 *lock = 1;
1269         } else if (flock->fl_type == F_EXLCK) {
1270                 cifs_dbg(FYI, "F_EXLCK\n");
1271                 *type |= server->vals->exclusive_lock_type;
1272                 *lock = 1;
1273         } else if (flock->fl_type == F_SHLCK) {
1274                 cifs_dbg(FYI, "F_SHLCK\n");
1275                 *type |= server->vals->shared_lock_type;
1276                 *lock = 1;
1277         } else
1278                 cifs_dbg(FYI, "Unknown type of lock\n");
1279 }
1280
1281 static int
1282 cifs_getlk(struct file *file, struct file_lock *flock, __u32 type,
1283            bool wait_flag, bool posix_lck, unsigned int xid)
1284 {
1285         int rc = 0;
1286         __u64 length = 1 + flock->fl_end - flock->fl_start;
1287         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1288         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1289         struct TCP_Server_Info *server = tcon->ses->server;
1290         __u16 netfid = cfile->fid.netfid;
1291
1292         if (posix_lck) {
1293                 int posix_lock_type;
1294
1295                 rc = cifs_posix_lock_test(file, flock);
1296                 if (!rc)
1297                         return rc;
1298
1299                 if (type & server->vals->shared_lock_type)
1300                         posix_lock_type = CIFS_RDLCK;
1301                 else
1302                         posix_lock_type = CIFS_WRLCK;
1303                 rc = CIFSSMBPosixLock(xid, tcon, netfid, current->tgid,
1304                                       flock->fl_start, length, flock,
1305                                       posix_lock_type, wait_flag);
1306                 return rc;
1307         }
1308
1309         rc = cifs_lock_test(cfile, flock->fl_start, length, type, flock);
1310         if (!rc)
1311                 return rc;
1312
1313         /* BB we could chain these into one lock request BB */
1314         rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length, type,
1315                                     1, 0, false);
1316         if (rc == 0) {
1317                 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1318                                             type, 0, 1, false);
1319                 flock->fl_type = F_UNLCK;
1320                 if (rc != 0)
1321                         cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1322                                  rc);
1323                 return 0;
1324         }
1325
1326         if (type & server->vals->shared_lock_type) {
1327                 flock->fl_type = F_WRLCK;
1328                 return 0;
1329         }
1330
1331         type &= ~server->vals->exclusive_lock_type;
1332
1333         rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1334                                     type | server->vals->shared_lock_type,
1335                                     1, 0, false);
1336         if (rc == 0) {
1337                 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1338                         type | server->vals->shared_lock_type, 0, 1, false);
1339                 flock->fl_type = F_RDLCK;
1340                 if (rc != 0)
1341                         cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1342                                  rc);
1343         } else
1344                 flock->fl_type = F_WRLCK;
1345
1346         return 0;
1347 }
1348
1349 void
1350 cifs_move_llist(struct list_head *source, struct list_head *dest)
1351 {
1352         struct list_head *li, *tmp;
1353         list_for_each_safe(li, tmp, source)
1354                 list_move(li, dest);
1355 }
1356
1357 void
1358 cifs_free_llist(struct list_head *llist)
1359 {
1360         struct cifsLockInfo *li, *tmp;
1361         list_for_each_entry_safe(li, tmp, llist, llist) {
1362                 cifs_del_lock_waiters(li);
1363                 list_del(&li->llist);
1364                 kfree(li);
1365         }
1366 }
1367
1368 int
1369 cifs_unlock_range(struct cifsFileInfo *cfile, struct file_lock *flock,
1370                   unsigned int xid)
1371 {
1372         int rc = 0, stored_rc;
1373         int types[] = {LOCKING_ANDX_LARGE_FILES,
1374                        LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES};
1375         unsigned int i;
1376         unsigned int max_num, num, max_buf;
1377         LOCKING_ANDX_RANGE *buf, *cur;
1378         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1379         struct cifsInodeInfo *cinode = CIFS_I(cfile->dentry->d_inode);
1380         struct cifsLockInfo *li, *tmp;
1381         __u64 length = 1 + flock->fl_end - flock->fl_start;
1382         struct list_head tmp_llist;
1383
1384         INIT_LIST_HEAD(&tmp_llist);
1385
1386         /*
1387          * Accessing maxBuf is racy with cifs_reconnect - need to store value
1388          * and check it for zero before using.
1389          */
1390         max_buf = tcon->ses->server->maxBuf;
1391         if (!max_buf)
1392                 return -EINVAL;
1393
1394         max_num = (max_buf - sizeof(struct smb_hdr)) /
1395                                                 sizeof(LOCKING_ANDX_RANGE);
1396         buf = kzalloc(max_num * sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1397         if (!buf)
1398                 return -ENOMEM;
1399
1400         down_write(&cinode->lock_sem);
1401         for (i = 0; i < 2; i++) {
1402                 cur = buf;
1403                 num = 0;
1404                 list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1405                         if (flock->fl_start > li->offset ||
1406                             (flock->fl_start + length) <
1407                             (li->offset + li->length))
1408                                 continue;
1409                         if (current->tgid != li->pid)
1410                                 continue;
1411                         if (types[i] != li->type)
1412                                 continue;
1413                         if (cinode->can_cache_brlcks) {
1414                                 /*
1415                                  * We can cache brlock requests - simply remove
1416                                  * a lock from the file's list.
1417                                  */
1418                                 list_del(&li->llist);
1419                                 cifs_del_lock_waiters(li);
1420                                 kfree(li);
1421                                 continue;
1422                         }
1423                         cur->Pid = cpu_to_le16(li->pid);
1424                         cur->LengthLow = cpu_to_le32((u32)li->length);
1425                         cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1426                         cur->OffsetLow = cpu_to_le32((u32)li->offset);
1427                         cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1428                         /*
1429                          * We need to save a lock here to let us add it again to
1430                          * the file's list if the unlock range request fails on
1431                          * the server.
1432                          */
1433                         list_move(&li->llist, &tmp_llist);
1434                         if (++num == max_num) {
1435                                 stored_rc = cifs_lockv(xid, tcon,
1436                                                        cfile->fid.netfid,
1437                                                        li->type, num, 0, buf);
1438                                 if (stored_rc) {
1439                                         /*
1440                                          * We failed on the unlock range
1441                                          * request - add all locks from the tmp
1442                                          * list to the head of the file's list.
1443                                          */
1444                                         cifs_move_llist(&tmp_llist,
1445                                                         &cfile->llist->locks);
1446                                         rc = stored_rc;
1447                                 } else
1448                                         /*
1449                                          * The unlock range request succeed -
1450                                          * free the tmp list.
1451                                          */
1452                                         cifs_free_llist(&tmp_llist);
1453                                 cur = buf;
1454                                 num = 0;
1455                         } else
1456                                 cur++;
1457                 }
1458                 if (num) {
1459                         stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1460                                                types[i], num, 0, buf);
1461                         if (stored_rc) {
1462                                 cifs_move_llist(&tmp_llist,
1463                                                 &cfile->llist->locks);
1464                                 rc = stored_rc;
1465                         } else
1466                                 cifs_free_llist(&tmp_llist);
1467                 }
1468         }
1469
1470         up_write(&cinode->lock_sem);
1471         kfree(buf);
1472         return rc;
1473 }
1474
1475 static int
1476 cifs_setlk(struct file *file, struct file_lock *flock, __u32 type,
1477            bool wait_flag, bool posix_lck, int lock, int unlock,
1478            unsigned int xid)
1479 {
1480         int rc = 0;
1481         __u64 length = 1 + flock->fl_end - flock->fl_start;
1482         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1483         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1484         struct TCP_Server_Info *server = tcon->ses->server;
1485         struct inode *inode = cfile->dentry->d_inode;
1486
1487         if (posix_lck) {
1488                 int posix_lock_type;
1489
1490                 rc = cifs_posix_lock_set(file, flock);
1491                 if (!rc || rc < 0)
1492                         return rc;
1493
1494                 if (type & server->vals->shared_lock_type)
1495                         posix_lock_type = CIFS_RDLCK;
1496                 else
1497                         posix_lock_type = CIFS_WRLCK;
1498
1499                 if (unlock == 1)
1500                         posix_lock_type = CIFS_UNLCK;
1501
1502                 rc = CIFSSMBPosixLock(xid, tcon, cfile->fid.netfid,
1503                                       current->tgid, flock->fl_start, length,
1504                                       NULL, posix_lock_type, wait_flag);
1505                 goto out;
1506         }
1507
1508         if (lock) {
1509                 struct cifsLockInfo *lock;
1510
1511                 lock = cifs_lock_init(flock->fl_start, length, type);
1512                 if (!lock)
1513                         return -ENOMEM;
1514
1515                 rc = cifs_lock_add_if(cfile, lock, wait_flag);
1516                 if (rc < 0) {
1517                         kfree(lock);
1518                         return rc;
1519                 }
1520                 if (!rc)
1521                         goto out;
1522
1523                 /*
1524                  * Windows 7 server can delay breaking lease from read to None
1525                  * if we set a byte-range lock on a file - break it explicitly
1526                  * before sending the lock to the server to be sure the next
1527                  * read won't conflict with non-overlapted locks due to
1528                  * pagereading.
1529                  */
1530                 if (!CIFS_CACHE_WRITE(CIFS_I(inode)) &&
1531                                         CIFS_CACHE_READ(CIFS_I(inode))) {
1532                         cifs_invalidate_mapping(inode);
1533                         cifs_dbg(FYI, "Set no oplock for inode=%p due to mand locks\n",
1534                                  inode);
1535                         CIFS_I(inode)->oplock = 0;
1536                 }
1537
1538                 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1539                                             type, 1, 0, wait_flag);
1540                 if (rc) {
1541                         kfree(lock);
1542                         return rc;
1543                 }
1544
1545                 cifs_lock_add(cfile, lock);
1546         } else if (unlock)
1547                 rc = server->ops->mand_unlock_range(cfile, flock, xid);
1548
1549 out:
1550         if (flock->fl_flags & FL_POSIX)
1551                 posix_lock_file_wait(file, flock);
1552         return rc;
1553 }
1554
1555 int cifs_lock(struct file *file, int cmd, struct file_lock *flock)
1556 {
1557         int rc, xid;
1558         int lock = 0, unlock = 0;
1559         bool wait_flag = false;
1560         bool posix_lck = false;
1561         struct cifs_sb_info *cifs_sb;
1562         struct cifs_tcon *tcon;
1563         struct cifsInodeInfo *cinode;
1564         struct cifsFileInfo *cfile;
1565         __u16 netfid;
1566         __u32 type;
1567
1568         rc = -EACCES;
1569         xid = get_xid();
1570
1571         cifs_dbg(FYI, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld end: %lld\n",
1572                  cmd, flock->fl_flags, flock->fl_type,
1573                  flock->fl_start, flock->fl_end);
1574
1575         cfile = (struct cifsFileInfo *)file->private_data;
1576         tcon = tlink_tcon(cfile->tlink);
1577
1578         cifs_read_flock(flock, &type, &lock, &unlock, &wait_flag,
1579                         tcon->ses->server);
1580
1581         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
1582         netfid = cfile->fid.netfid;
1583         cinode = CIFS_I(file_inode(file));
1584
1585         if (cap_unix(tcon->ses) &&
1586             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1587             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1588                 posix_lck = true;
1589         /*
1590          * BB add code here to normalize offset and length to account for
1591          * negative length which we can not accept over the wire.
1592          */
1593         if (IS_GETLK(cmd)) {
1594                 rc = cifs_getlk(file, flock, type, wait_flag, posix_lck, xid);
1595                 free_xid(xid);
1596                 return rc;
1597         }
1598
1599         if (!lock && !unlock) {
1600                 /*
1601                  * if no lock or unlock then nothing to do since we do not
1602                  * know what it is
1603                  */
1604                 free_xid(xid);
1605                 return -EOPNOTSUPP;
1606         }
1607
1608         rc = cifs_setlk(file, flock, type, wait_flag, posix_lck, lock, unlock,
1609                         xid);
1610         free_xid(xid);
1611         return rc;
1612 }
1613
1614 /*
1615  * update the file size (if needed) after a write. Should be called with
1616  * the inode->i_lock held
1617  */
1618 void
1619 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
1620                       unsigned int bytes_written)
1621 {
1622         loff_t end_of_write = offset + bytes_written;
1623
1624         if (end_of_write > cifsi->server_eof)
1625                 cifsi->server_eof = end_of_write;
1626 }
1627
1628 static ssize_t
1629 cifs_write(struct cifsFileInfo *open_file, __u32 pid, const char *write_data,
1630            size_t write_size, loff_t *offset)
1631 {
1632         int rc = 0;
1633         unsigned int bytes_written = 0;
1634         unsigned int total_written;
1635         struct cifs_sb_info *cifs_sb;
1636         struct cifs_tcon *tcon;
1637         struct TCP_Server_Info *server;
1638         unsigned int xid;
1639         struct dentry *dentry = open_file->dentry;
1640         struct cifsInodeInfo *cifsi = CIFS_I(dentry->d_inode);
1641         struct cifs_io_parms io_parms;
1642
1643         cifs_sb = CIFS_SB(dentry->d_sb);
1644
1645         cifs_dbg(FYI, "write %zd bytes to offset %lld of %s\n",
1646                  write_size, *offset, dentry->d_name.name);
1647
1648         tcon = tlink_tcon(open_file->tlink);
1649         server = tcon->ses->server;
1650
1651         if (!server->ops->sync_write)
1652                 return -ENOSYS;
1653
1654         xid = get_xid();
1655
1656         for (total_written = 0; write_size > total_written;
1657              total_written += bytes_written) {
1658                 rc = -EAGAIN;
1659                 while (rc == -EAGAIN) {
1660                         struct kvec iov[2];
1661                         unsigned int len;
1662
1663                         if (open_file->invalidHandle) {
1664                                 /* we could deadlock if we called
1665                                    filemap_fdatawait from here so tell
1666                                    reopen_file not to flush data to
1667                                    server now */
1668                                 rc = cifs_reopen_file(open_file, false);
1669                                 if (rc != 0)
1670                                         break;
1671                         }
1672
1673                         len = min((size_t)cifs_sb->wsize,
1674                                   write_size - total_written);
1675                         /* iov[0] is reserved for smb header */
1676                         iov[1].iov_base = (char *)write_data + total_written;
1677                         iov[1].iov_len = len;
1678                         io_parms.pid = pid;
1679                         io_parms.tcon = tcon;
1680                         io_parms.offset = *offset;
1681                         io_parms.length = len;
1682                         rc = server->ops->sync_write(xid, open_file, &io_parms,
1683                                                      &bytes_written, iov, 1);
1684                 }
1685                 if (rc || (bytes_written == 0)) {
1686                         if (total_written)
1687                                 break;
1688                         else {
1689                                 free_xid(xid);
1690                                 return rc;
1691                         }
1692                 } else {
1693                         spin_lock(&dentry->d_inode->i_lock);
1694                         cifs_update_eof(cifsi, *offset, bytes_written);
1695                         spin_unlock(&dentry->d_inode->i_lock);
1696                         *offset += bytes_written;
1697                 }
1698         }
1699
1700         cifs_stats_bytes_written(tcon, total_written);
1701
1702         if (total_written > 0) {
1703                 spin_lock(&dentry->d_inode->i_lock);
1704                 if (*offset > dentry->d_inode->i_size)
1705                         i_size_write(dentry->d_inode, *offset);
1706                 spin_unlock(&dentry->d_inode->i_lock);
1707         }
1708         mark_inode_dirty_sync(dentry->d_inode);
1709         free_xid(xid);
1710         return total_written;
1711 }
1712
1713 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1714                                         bool fsuid_only)
1715 {
1716         struct cifsFileInfo *open_file = NULL;
1717         struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1718
1719         /* only filter by fsuid on multiuser mounts */
1720         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1721                 fsuid_only = false;
1722
1723         spin_lock(&cifs_file_list_lock);
1724         /* we could simply get the first_list_entry since write-only entries
1725            are always at the end of the list but since the first entry might
1726            have a close pending, we go through the whole list */
1727         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1728                 if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
1729                         continue;
1730                 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1731                         if (!open_file->invalidHandle) {
1732                                 /* found a good file */
1733                                 /* lock it so it will not be closed on us */
1734                                 cifsFileInfo_get_locked(open_file);
1735                                 spin_unlock(&cifs_file_list_lock);
1736                                 return open_file;
1737                         } /* else might as well continue, and look for
1738                              another, or simply have the caller reopen it
1739                              again rather than trying to fix this handle */
1740                 } else /* write only file */
1741                         break; /* write only files are last so must be done */
1742         }
1743         spin_unlock(&cifs_file_list_lock);
1744         return NULL;
1745 }
1746
1747 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1748                                         bool fsuid_only)
1749 {
1750         struct cifsFileInfo *open_file, *inv_file = NULL;
1751         struct cifs_sb_info *cifs_sb;
1752         bool any_available = false;
1753         int rc;
1754         unsigned int refind = 0;
1755
1756         /* Having a null inode here (because mapping->host was set to zero by
1757         the VFS or MM) should not happen but we had reports of on oops (due to
1758         it being zero) during stress testcases so we need to check for it */
1759
1760         if (cifs_inode == NULL) {
1761                 cifs_dbg(VFS, "Null inode passed to cifs_writeable_file\n");
1762                 dump_stack();
1763                 return NULL;
1764         }
1765
1766         cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1767
1768         /* only filter by fsuid on multiuser mounts */
1769         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1770                 fsuid_only = false;
1771
1772         spin_lock(&cifs_file_list_lock);
1773 refind_writable:
1774         if (refind > MAX_REOPEN_ATT) {
1775                 spin_unlock(&cifs_file_list_lock);
1776                 return NULL;
1777         }
1778         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1779                 if (!any_available && open_file->pid != current->tgid)
1780                         continue;
1781                 if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
1782                         continue;
1783                 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1784                         if (!open_file->invalidHandle) {
1785                                 /* found a good writable file */
1786                                 cifsFileInfo_get_locked(open_file);
1787                                 spin_unlock(&cifs_file_list_lock);
1788                                 return open_file;
1789                         } else {
1790                                 if (!inv_file)
1791                                         inv_file = open_file;
1792                         }
1793                 }
1794         }
1795         /* couldn't find useable FH with same pid, try any available */
1796         if (!any_available) {
1797                 any_available = true;
1798                 goto refind_writable;
1799         }
1800
1801         if (inv_file) {
1802                 any_available = false;
1803                 cifsFileInfo_get_locked(inv_file);
1804         }
1805
1806         spin_unlock(&cifs_file_list_lock);
1807
1808         if (inv_file) {
1809                 rc = cifs_reopen_file(inv_file, false);
1810                 if (!rc)
1811                         return inv_file;
1812                 else {
1813                         spin_lock(&cifs_file_list_lock);
1814                         list_move_tail(&inv_file->flist,
1815                                         &cifs_inode->openFileList);
1816                         spin_unlock(&cifs_file_list_lock);
1817                         cifsFileInfo_put(inv_file);
1818                         spin_lock(&cifs_file_list_lock);
1819                         ++refind;
1820                         goto refind_writable;
1821                 }
1822         }
1823
1824         return NULL;
1825 }
1826
1827 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1828 {
1829         struct address_space *mapping = page->mapping;
1830         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
1831         char *write_data;
1832         int rc = -EFAULT;
1833         int bytes_written = 0;
1834         struct inode *inode;
1835         struct cifsFileInfo *open_file;
1836
1837         if (!mapping || !mapping->host)
1838                 return -EFAULT;
1839
1840         inode = page->mapping->host;
1841
1842         offset += (loff_t)from;
1843         write_data = kmap(page);
1844         write_data += from;
1845
1846         if ((to > PAGE_CACHE_SIZE) || (from > to)) {
1847                 kunmap(page);
1848                 return -EIO;
1849         }
1850
1851         /* racing with truncate? */
1852         if (offset > mapping->host->i_size) {
1853                 kunmap(page);
1854                 return 0; /* don't care */
1855         }
1856
1857         /* check to make sure that we are not extending the file */
1858         if (mapping->host->i_size - offset < (loff_t)to)
1859                 to = (unsigned)(mapping->host->i_size - offset);
1860
1861         open_file = find_writable_file(CIFS_I(mapping->host), false);
1862         if (open_file) {
1863                 bytes_written = cifs_write(open_file, open_file->pid,
1864                                            write_data, to - from, &offset);
1865                 cifsFileInfo_put(open_file);
1866                 /* Does mm or vfs already set times? */
1867                 inode->i_atime = inode->i_mtime = current_fs_time(inode->i_sb);
1868                 if ((bytes_written > 0) && (offset))
1869                         rc = 0;
1870                 else if (bytes_written < 0)
1871                         rc = bytes_written;
1872         } else {
1873                 cifs_dbg(FYI, "No writeable filehandles for inode\n");
1874                 rc = -EIO;
1875         }
1876
1877         kunmap(page);
1878         return rc;
1879 }
1880
1881 static int cifs_writepages(struct address_space *mapping,
1882                            struct writeback_control *wbc)
1883 {
1884         struct cifs_sb_info *cifs_sb = CIFS_SB(mapping->host->i_sb);
1885         bool done = false, scanned = false, range_whole = false;
1886         pgoff_t end, index;
1887         struct cifs_writedata *wdata;
1888         struct TCP_Server_Info *server;
1889         struct page *page;
1890         int rc = 0;
1891
1892         /*
1893          * If wsize is smaller than the page cache size, default to writing
1894          * one page at a time via cifs_writepage
1895          */
1896         if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1897                 return generic_writepages(mapping, wbc);
1898
1899         if (wbc->range_cyclic) {
1900                 index = mapping->writeback_index; /* Start from prev offset */
1901                 end = -1;
1902         } else {
1903                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1904                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1905                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1906                         range_whole = true;
1907                 scanned = true;
1908         }
1909 retry:
1910         while (!done && index <= end) {
1911                 unsigned int i, nr_pages, found_pages;
1912                 pgoff_t next = 0, tofind;
1913                 struct page **pages;
1914
1915                 tofind = min((cifs_sb->wsize / PAGE_CACHE_SIZE) - 1,
1916                                 end - index) + 1;
1917
1918                 wdata = cifs_writedata_alloc((unsigned int)tofind,
1919                                              cifs_writev_complete);
1920                 if (!wdata) {
1921                         rc = -ENOMEM;
1922                         break;
1923                 }
1924
1925                 /*
1926                  * find_get_pages_tag seems to return a max of 256 on each
1927                  * iteration, so we must call it several times in order to
1928                  * fill the array or the wsize is effectively limited to
1929                  * 256 * PAGE_CACHE_SIZE.
1930                  */
1931                 found_pages = 0;
1932                 pages = wdata->pages;
1933                 do {
1934                         nr_pages = find_get_pages_tag(mapping, &index,
1935                                                         PAGECACHE_TAG_DIRTY,
1936                                                         tofind, pages);
1937                         found_pages += nr_pages;
1938                         tofind -= nr_pages;
1939                         pages += nr_pages;
1940                 } while (nr_pages && tofind && index <= end);
1941
1942                 if (found_pages == 0) {
1943                         kref_put(&wdata->refcount, cifs_writedata_release);
1944                         break;
1945                 }
1946
1947                 nr_pages = 0;
1948                 for (i = 0; i < found_pages; i++) {
1949                         page = wdata->pages[i];
1950                         /*
1951                          * At this point we hold neither mapping->tree_lock nor
1952                          * lock on the page itself: the page may be truncated or
1953                          * invalidated (changing page->mapping to NULL), or even
1954                          * swizzled back from swapper_space to tmpfs file
1955                          * mapping
1956                          */
1957
1958                         if (nr_pages == 0)
1959                                 lock_page(page);
1960                         else if (!trylock_page(page))
1961                                 break;
1962
1963                         if (unlikely(page->mapping != mapping)) {
1964                                 unlock_page(page);
1965                                 break;
1966                         }
1967
1968                         if (!wbc->range_cyclic && page->index > end) {
1969                                 done = true;
1970                                 unlock_page(page);
1971                                 break;
1972                         }
1973
1974                         if (next && (page->index != next)) {
1975                                 /* Not next consecutive page */
1976                                 unlock_page(page);
1977                                 break;
1978                         }
1979
1980                         if (wbc->sync_mode != WB_SYNC_NONE)
1981                                 wait_on_page_writeback(page);
1982
1983                         if (PageWriteback(page) ||
1984                                         !clear_page_dirty_for_io(page)) {
1985                                 unlock_page(page);
1986                                 break;
1987                         }
1988
1989                         /*
1990                          * This actually clears the dirty bit in the radix tree.
1991                          * See cifs_writepage() for more commentary.
1992                          */
1993                         set_page_writeback(page);
1994
1995                         if (page_offset(page) >= i_size_read(mapping->host)) {
1996                                 done = true;
1997                                 unlock_page(page);
1998                                 end_page_writeback(page);
1999                                 break;
2000                         }
2001
2002                         wdata->pages[i] = page;
2003                         next = page->index + 1;
2004                         ++nr_pages;
2005                 }
2006
2007                 /* reset index to refind any pages skipped */
2008                 if (nr_pages == 0)
2009                         index = wdata->pages[0]->index + 1;
2010
2011                 /* put any pages we aren't going to use */
2012                 for (i = nr_pages; i < found_pages; i++) {
2013                         page_cache_release(wdata->pages[i]);
2014                         wdata->pages[i] = NULL;
2015                 }
2016
2017                 /* nothing to write? */
2018                 if (nr_pages == 0) {
2019                         kref_put(&wdata->refcount, cifs_writedata_release);
2020                         continue;
2021                 }
2022
2023                 wdata->sync_mode = wbc->sync_mode;
2024                 wdata->nr_pages = nr_pages;
2025                 wdata->offset = page_offset(wdata->pages[0]);
2026                 wdata->pagesz = PAGE_CACHE_SIZE;
2027                 wdata->tailsz =
2028                         min(i_size_read(mapping->host) -
2029                             page_offset(wdata->pages[nr_pages - 1]),
2030                             (loff_t)PAGE_CACHE_SIZE);
2031                 wdata->bytes = ((nr_pages - 1) * PAGE_CACHE_SIZE) +
2032                                         wdata->tailsz;
2033
2034                 do {
2035                         if (wdata->cfile != NULL)
2036                                 cifsFileInfo_put(wdata->cfile);
2037                         wdata->cfile = find_writable_file(CIFS_I(mapping->host),
2038                                                           false);
2039                         if (!wdata->cfile) {
2040                                 cifs_dbg(VFS, "No writable handles for inode\n");
2041                                 rc = -EBADF;
2042                                 break;
2043                         }
2044                         wdata->pid = wdata->cfile->pid;
2045                         server = tlink_tcon(wdata->cfile->tlink)->ses->server;
2046                         rc = server->ops->async_writev(wdata,
2047                                                         cifs_writedata_release);
2048                 } while (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN);
2049
2050                 for (i = 0; i < nr_pages; ++i)
2051                         unlock_page(wdata->pages[i]);
2052
2053                 /* send failure -- clean up the mess */
2054                 if (rc != 0) {
2055                         for (i = 0; i < nr_pages; ++i) {
2056                                 if (rc == -EAGAIN)
2057                                         redirty_page_for_writepage(wbc,
2058                                                            wdata->pages[i]);
2059                                 else
2060                                         SetPageError(wdata->pages[i]);
2061                                 end_page_writeback(wdata->pages[i]);
2062                                 page_cache_release(wdata->pages[i]);
2063                         }
2064                         if (rc != -EAGAIN)
2065                                 mapping_set_error(mapping, rc);
2066                 }
2067                 kref_put(&wdata->refcount, cifs_writedata_release);
2068
2069                 wbc->nr_to_write -= nr_pages;
2070                 if (wbc->nr_to_write <= 0)
2071                         done = true;
2072
2073                 index = next;
2074         }
2075
2076         if (!scanned && !done) {
2077                 /*
2078                  * We hit the last page and there is more work to be done: wrap
2079                  * back to the start of the file
2080                  */
2081                 scanned = true;
2082                 index = 0;
2083                 goto retry;
2084         }
2085
2086         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2087                 mapping->writeback_index = index;
2088
2089         return rc;
2090 }
2091
2092 static int
2093 cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
2094 {
2095         int rc;
2096         unsigned int xid;
2097
2098         xid = get_xid();
2099 /* BB add check for wbc flags */
2100         page_cache_get(page);
2101         if (!PageUptodate(page))
2102                 cifs_dbg(FYI, "ppw - page not up to date\n");
2103
2104         /*
2105          * Set the "writeback" flag, and clear "dirty" in the radix tree.
2106          *
2107          * A writepage() implementation always needs to do either this,
2108          * or re-dirty the page with "redirty_page_for_writepage()" in
2109          * the case of a failure.
2110          *
2111          * Just unlocking the page will cause the radix tree tag-bits
2112          * to fail to update with the state of the page correctly.
2113          */
2114         set_page_writeback(page);
2115 retry_write:
2116         rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
2117         if (rc == -EAGAIN && wbc->sync_mode == WB_SYNC_ALL)
2118                 goto retry_write;
2119         else if (rc == -EAGAIN)
2120                 redirty_page_for_writepage(wbc, page);
2121         else if (rc != 0)
2122                 SetPageError(page);
2123         else
2124                 SetPageUptodate(page);
2125         end_page_writeback(page);
2126         page_cache_release(page);
2127         free_xid(xid);
2128         return rc;
2129 }
2130
2131 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
2132 {
2133         int rc = cifs_writepage_locked(page, wbc);
2134         unlock_page(page);
2135         return rc;
2136 }
2137
2138 static int cifs_write_end(struct file *file, struct address_space *mapping,
2139                         loff_t pos, unsigned len, unsigned copied,
2140                         struct page *page, void *fsdata)
2141 {
2142         int rc;
2143         struct inode *inode = mapping->host;
2144         struct cifsFileInfo *cfile = file->private_data;
2145         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
2146         __u32 pid;
2147
2148         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2149                 pid = cfile->pid;
2150         else
2151                 pid = current->tgid;
2152
2153         cifs_dbg(FYI, "write_end for page %p from pos %lld with %d bytes\n",
2154                  page, pos, copied);
2155
2156         if (PageChecked(page)) {
2157                 if (copied == len)
2158                         SetPageUptodate(page);
2159                 ClearPageChecked(page);
2160         } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
2161                 SetPageUptodate(page);
2162
2163         if (!PageUptodate(page)) {
2164                 char *page_data;
2165                 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
2166                 unsigned int xid;
2167
2168                 xid = get_xid();
2169                 /* this is probably better than directly calling
2170                    partialpage_write since in this function the file handle is
2171                    known which we might as well leverage */
2172                 /* BB check if anything else missing out of ppw
2173                    such as updating last write time */
2174                 page_data = kmap(page);
2175                 rc = cifs_write(cfile, pid, page_data + offset, copied, &pos);
2176                 /* if (rc < 0) should we set writebehind rc? */
2177                 kunmap(page);
2178
2179                 free_xid(xid);
2180         } else {
2181                 rc = copied;
2182                 pos += copied;
2183                 set_page_dirty(page);
2184         }
2185
2186         if (rc > 0) {
2187                 spin_lock(&inode->i_lock);
2188                 if (pos > inode->i_size)
2189                         i_size_write(inode, pos);
2190                 spin_unlock(&inode->i_lock);
2191         }
2192
2193         unlock_page(page);
2194         page_cache_release(page);
2195
2196         return rc;
2197 }
2198
2199 int cifs_strict_fsync(struct file *file, loff_t start, loff_t end,
2200                       int datasync)
2201 {
2202         unsigned int xid;
2203         int rc = 0;
2204         struct cifs_tcon *tcon;
2205         struct TCP_Server_Info *server;
2206         struct cifsFileInfo *smbfile = file->private_data;
2207         struct inode *inode = file_inode(file);
2208         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2209
2210         rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2211         if (rc)
2212                 return rc;
2213         mutex_lock(&inode->i_mutex);
2214
2215         xid = get_xid();
2216
2217         cifs_dbg(FYI, "Sync file - name: %s datasync: 0x%x\n",
2218                  file->f_path.dentry->d_name.name, datasync);
2219
2220         if (!CIFS_CACHE_READ(CIFS_I(inode))) {
2221                 rc = cifs_invalidate_mapping(inode);
2222                 if (rc) {
2223                         cifs_dbg(FYI, "rc: %d during invalidate phase\n", rc);
2224                         rc = 0; /* don't care about it in fsync */
2225                 }
2226         }
2227
2228         tcon = tlink_tcon(smbfile->tlink);
2229         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2230                 server = tcon->ses->server;
2231                 if (server->ops->flush)
2232                         rc = server->ops->flush(xid, tcon, &smbfile->fid);
2233                 else
2234                         rc = -ENOSYS;
2235         }
2236
2237         free_xid(xid);
2238         mutex_unlock(&inode->i_mutex);
2239         return rc;
2240 }
2241
2242 int cifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2243 {
2244         unsigned int xid;
2245         int rc = 0;
2246         struct cifs_tcon *tcon;
2247         struct TCP_Server_Info *server;
2248         struct cifsFileInfo *smbfile = file->private_data;
2249         struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2250         struct inode *inode = file->f_mapping->host;
2251
2252         rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2253         if (rc)
2254                 return rc;
2255         mutex_lock(&inode->i_mutex);
2256
2257         xid = get_xid();
2258
2259         cifs_dbg(FYI, "Sync file - name: %s datasync: 0x%x\n",
2260                  file->f_path.dentry->d_name.name, datasync);
2261
2262         tcon = tlink_tcon(smbfile->tlink);
2263         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2264                 server = tcon->ses->server;
2265                 if (server->ops->flush)
2266                         rc = server->ops->flush(xid, tcon, &smbfile->fid);
2267                 else
2268                         rc = -ENOSYS;
2269         }
2270
2271         free_xid(xid);
2272         mutex_unlock(&inode->i_mutex);
2273         return rc;
2274 }
2275
2276 /*
2277  * As file closes, flush all cached write data for this inode checking
2278  * for write behind errors.
2279  */
2280 int cifs_flush(struct file *file, fl_owner_t id)
2281 {
2282         struct inode *inode = file_inode(file);
2283         int rc = 0;
2284
2285         if (file->f_mode & FMODE_WRITE)
2286                 rc = filemap_write_and_wait(inode->i_mapping);
2287
2288         cifs_dbg(FYI, "Flush inode %p file %p rc %d\n", inode, file, rc);
2289
2290         return rc;
2291 }
2292
2293 static int
2294 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
2295 {
2296         int rc = 0;
2297         unsigned long i;
2298
2299         for (i = 0; i < num_pages; i++) {
2300                 pages[i] = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2301                 if (!pages[i]) {
2302                         /*
2303                          * save number of pages we have already allocated and
2304                          * return with ENOMEM error
2305                          */
2306                         num_pages = i;
2307                         rc = -ENOMEM;
2308                         break;
2309                 }
2310         }
2311
2312         if (rc) {
2313                 for (i = 0; i < num_pages; i++)
2314                         put_page(pages[i]);
2315         }
2316         return rc;
2317 }
2318
2319 static inline
2320 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
2321 {
2322         size_t num_pages;
2323         size_t clen;
2324
2325         clen = min_t(const size_t, len, wsize);
2326         num_pages = DIV_ROUND_UP(clen, PAGE_SIZE);
2327
2328         if (cur_len)
2329                 *cur_len = clen;
2330
2331         return num_pages;
2332 }
2333
2334 static void
2335 cifs_uncached_writedata_release(struct kref *refcount)
2336 {
2337         int i;
2338         struct cifs_writedata *wdata = container_of(refcount,
2339                                         struct cifs_writedata, refcount);
2340
2341         for (i = 0; i < wdata->nr_pages; i++)
2342                 put_page(wdata->pages[i]);
2343         cifs_writedata_release(refcount);
2344 }
2345
2346 static void
2347 cifs_uncached_writev_complete(struct work_struct *work)
2348 {
2349         struct cifs_writedata *wdata = container_of(work,
2350                                         struct cifs_writedata, work);
2351         struct inode *inode = wdata->cfile->dentry->d_inode;
2352         struct cifsInodeInfo *cifsi = CIFS_I(inode);
2353
2354         spin_lock(&inode->i_lock);
2355         cifs_update_eof(cifsi, wdata->offset, wdata->bytes);
2356         if (cifsi->server_eof > inode->i_size)
2357                 i_size_write(inode, cifsi->server_eof);
2358         spin_unlock(&inode->i_lock);
2359
2360         complete(&wdata->done);
2361
2362         kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2363 }
2364
2365 /* attempt to send write to server, retry on any -EAGAIN errors */
2366 static int
2367 cifs_uncached_retry_writev(struct cifs_writedata *wdata)
2368 {
2369         int rc;
2370         struct TCP_Server_Info *server;
2371
2372         server = tlink_tcon(wdata->cfile->tlink)->ses->server;
2373
2374         do {
2375                 if (wdata->cfile->invalidHandle) {
2376                         rc = cifs_reopen_file(wdata->cfile, false);
2377                         if (rc != 0)
2378                                 continue;
2379                 }
2380                 rc = server->ops->async_writev(wdata,
2381                                                cifs_uncached_writedata_release);
2382         } while (rc == -EAGAIN);
2383
2384         return rc;
2385 }
2386
2387 static ssize_t
2388 cifs_iovec_write(struct file *file, const struct iovec *iov,
2389                  unsigned long nr_segs, loff_t *poffset)
2390 {
2391         unsigned long nr_pages, i;
2392         size_t bytes, copied, len, cur_len;
2393         ssize_t total_written = 0;
2394         loff_t offset;
2395         struct iov_iter it;
2396         struct cifsFileInfo *open_file;
2397         struct cifs_tcon *tcon;
2398         struct cifs_sb_info *cifs_sb;
2399         struct cifs_writedata *wdata, *tmp;
2400         struct list_head wdata_list;
2401         int rc;
2402         pid_t pid;
2403
2404         len = iov_length(iov, nr_segs);
2405         if (!len)
2406                 return 0;
2407
2408         rc = generic_write_checks(file, poffset, &len, 0);
2409         if (rc)
2410                 return rc;
2411
2412         INIT_LIST_HEAD(&wdata_list);
2413         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2414         open_file = file->private_data;
2415         tcon = tlink_tcon(open_file->tlink);
2416
2417         if (!tcon->ses->server->ops->async_writev)
2418                 return -ENOSYS;
2419
2420         offset = *poffset;
2421
2422         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2423                 pid = open_file->pid;
2424         else
2425                 pid = current->tgid;
2426
2427         iov_iter_init(&it, iov, nr_segs, len, 0);
2428         do {
2429                 size_t save_len;
2430
2431                 nr_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
2432                 wdata = cifs_writedata_alloc(nr_pages,
2433                                              cifs_uncached_writev_complete);
2434                 if (!wdata) {
2435                         rc = -ENOMEM;
2436                         break;
2437                 }
2438
2439                 rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
2440                 if (rc) {
2441                         kfree(wdata);
2442                         break;
2443                 }
2444
2445                 save_len = cur_len;
2446                 for (i = 0; i < nr_pages; i++) {
2447                         bytes = min_t(const size_t, cur_len, PAGE_SIZE);
2448                         copied = iov_iter_copy_from_user(wdata->pages[i], &it,
2449                                                          0, bytes);
2450                         cur_len -= copied;
2451                         iov_iter_advance(&it, copied);
2452                         /*
2453                          * If we didn't copy as much as we expected, then that
2454                          * may mean we trod into an unmapped area. Stop copying
2455                          * at that point. On the next pass through the big
2456                          * loop, we'll likely end up getting a zero-length
2457                          * write and bailing out of it.
2458                          */
2459                         if (copied < bytes)
2460                                 break;
2461                 }
2462                 cur_len = save_len - cur_len;
2463
2464                 /*
2465                  * If we have no data to send, then that probably means that
2466                  * the copy above failed altogether. That's most likely because
2467                  * the address in the iovec was bogus. Set the rc to -EFAULT,
2468                  * free anything we allocated and bail out.
2469                  */
2470                 if (!cur_len) {
2471                         for (i = 0; i < nr_pages; i++)
2472                                 put_page(wdata->pages[i]);
2473                         kfree(wdata);
2474                         rc = -EFAULT;
2475                         break;
2476                 }
2477
2478                 /*
2479                  * i + 1 now represents the number of pages we actually used in
2480                  * the copy phase above. Bring nr_pages down to that, and free
2481                  * any pages that we didn't use.
2482                  */
2483                 for ( ; nr_pages > i + 1; nr_pages--)
2484                         put_page(wdata->pages[nr_pages - 1]);
2485
2486                 wdata->sync_mode = WB_SYNC_ALL;
2487                 wdata->nr_pages = nr_pages;
2488                 wdata->offset = (__u64)offset;
2489                 wdata->cfile = cifsFileInfo_get(open_file);
2490                 wdata->pid = pid;
2491                 wdata->bytes = cur_len;
2492                 wdata->pagesz = PAGE_SIZE;
2493                 wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
2494                 rc = cifs_uncached_retry_writev(wdata);
2495                 if (rc) {
2496                         kref_put(&wdata->refcount,
2497                                  cifs_uncached_writedata_release);
2498                         break;
2499                 }
2500
2501                 list_add_tail(&wdata->list, &wdata_list);
2502                 offset += cur_len;
2503                 len -= cur_len;
2504         } while (len > 0);
2505
2506         /*
2507          * If at least one write was successfully sent, then discard any rc
2508          * value from the later writes. If the other write succeeds, then
2509          * we'll end up returning whatever was written. If it fails, then
2510          * we'll get a new rc value from that.
2511          */
2512         if (!list_empty(&wdata_list))
2513                 rc = 0;
2514
2515         /*
2516          * Wait for and collect replies for any successful sends in order of
2517          * increasing offset. Once an error is hit or we get a fatal signal
2518          * while waiting, then return without waiting for any more replies.
2519          */
2520 restart_loop:
2521         list_for_each_entry_safe(wdata, tmp, &wdata_list, list) {
2522                 if (!rc) {
2523                         /* FIXME: freezable too? */
2524                         rc = wait_for_completion_killable(&wdata->done);
2525                         if (rc)
2526                                 rc = -EINTR;
2527                         else if (wdata->result)
2528                                 rc = wdata->result;
2529                         else
2530                                 total_written += wdata->bytes;
2531
2532                         /* resend call if it's a retryable error */
2533                         if (rc == -EAGAIN) {
2534                                 rc = cifs_uncached_retry_writev(wdata);
2535                                 goto restart_loop;
2536                         }
2537                 }
2538                 list_del_init(&wdata->list);
2539                 kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2540         }
2541
2542         if (total_written > 0)
2543                 *poffset += total_written;
2544
2545         cifs_stats_bytes_written(tcon, total_written);
2546         return total_written ? total_written : (ssize_t)rc;
2547 }
2548
2549 ssize_t cifs_user_writev(struct kiocb *iocb, const struct iovec *iov,
2550                                 unsigned long nr_segs, loff_t pos)
2551 {
2552         ssize_t written;
2553         struct inode *inode;
2554
2555         inode = file_inode(iocb->ki_filp);
2556
2557         /*
2558          * BB - optimize the way when signing is disabled. We can drop this
2559          * extra memory-to-memory copying and use iovec buffers for constructing
2560          * write request.
2561          */
2562
2563         written = cifs_iovec_write(iocb->ki_filp, iov, nr_segs, &pos);
2564         if (written > 0) {
2565                 CIFS_I(inode)->invalid_mapping = true;
2566                 iocb->ki_pos = pos;
2567         }
2568
2569         return written;
2570 }
2571
2572 static ssize_t
2573 cifs_writev(struct kiocb *iocb, const struct iovec *iov,
2574             unsigned long nr_segs, loff_t pos)
2575 {
2576         struct file *file = iocb->ki_filp;
2577         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
2578         struct inode *inode = file->f_mapping->host;
2579         struct cifsInodeInfo *cinode = CIFS_I(inode);
2580         struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
2581         ssize_t rc = -EACCES;
2582         loff_t lock_pos = pos;
2583
2584         if (file->f_flags & O_APPEND)
2585                 lock_pos = i_size_read(inode);
2586         /*
2587          * We need to hold the sem to be sure nobody modifies lock list
2588          * with a brlock that prevents writing.
2589          */
2590         down_read(&cinode->lock_sem);
2591         if (!cifs_find_lock_conflict(cfile, lock_pos, iov_length(iov, nr_segs),
2592                                      server->vals->exclusive_lock_type, NULL,
2593                                      CIFS_WRITE_OP))
2594                 rc = generic_file_aio_write(iocb, iov, nr_segs, pos);
2595         up_read(&cinode->lock_sem);
2596         return rc;
2597 }
2598
2599 ssize_t
2600 cifs_strict_writev(struct kiocb *iocb, const struct iovec *iov,
2601                    unsigned long nr_segs, loff_t pos)
2602 {
2603         struct inode *inode = file_inode(iocb->ki_filp);
2604         struct cifsInodeInfo *cinode = CIFS_I(inode);
2605         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2606         struct cifsFileInfo *cfile = (struct cifsFileInfo *)
2607                                                 iocb->ki_filp->private_data;
2608         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
2609         ssize_t written;
2610
2611         written = cifs_get_writer(cinode);
2612         if (written)
2613                 return written;
2614
2615         if (CIFS_CACHE_WRITE(cinode)) {
2616                 if (cap_unix(tcon->ses) &&
2617                 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability))
2618                   && ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0)) {
2619                         written = generic_file_aio_write(
2620                                         iocb, iov, nr_segs, pos);
2621                         goto out;
2622                 }
2623                 written = cifs_writev(iocb, iov, nr_segs, pos);
2624                 goto out;
2625         }
2626         /*
2627          * For non-oplocked files in strict cache mode we need to write the data
2628          * to the server exactly from the pos to pos+len-1 rather than flush all
2629          * affected pages because it may cause a error with mandatory locks on
2630          * these pages but not on the region from pos to ppos+len-1.
2631          */
2632         written = cifs_user_writev(iocb, iov, nr_segs, pos);
2633         if (written > 0 && CIFS_CACHE_READ(cinode)) {
2634                 /*
2635                  * Windows 7 server can delay breaking level2 oplock if a write
2636                  * request comes - break it on the client to prevent reading
2637                  * an old data.
2638                  */
2639                 cifs_invalidate_mapping(inode);
2640                 cifs_dbg(FYI, "Set no oplock for inode=%p after a write operation\n",
2641                          inode);
2642                 cinode->oplock = 0;
2643         }
2644 out:
2645         cifs_put_writer(cinode);
2646         return written;
2647 }
2648
2649 static struct cifs_readdata *
2650 cifs_readdata_alloc(unsigned int nr_pages, work_func_t complete)
2651 {
2652         struct cifs_readdata *rdata;
2653
2654         rdata = kzalloc(sizeof(*rdata) + (sizeof(struct page *) * nr_pages),
2655                         GFP_KERNEL);
2656         if (rdata != NULL) {
2657                 kref_init(&rdata->refcount);
2658                 INIT_LIST_HEAD(&rdata->list);
2659                 init_completion(&rdata->done);
2660                 INIT_WORK(&rdata->work, complete);
2661         }
2662
2663         return rdata;
2664 }
2665
2666 void
2667 cifs_readdata_release(struct kref *refcount)
2668 {
2669         struct cifs_readdata *rdata = container_of(refcount,
2670                                         struct cifs_readdata, refcount);
2671
2672         if (rdata->cfile)
2673                 cifsFileInfo_put(rdata->cfile);
2674
2675         kfree(rdata);
2676 }
2677
2678 static int
2679 cifs_read_allocate_pages(struct cifs_readdata *rdata, unsigned int nr_pages)
2680 {
2681         int rc = 0;
2682         struct page *page;
2683         unsigned int i;
2684
2685         for (i = 0; i < nr_pages; i++) {
2686                 page = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2687                 if (!page) {
2688                         rc = -ENOMEM;
2689                         break;
2690                 }
2691                 rdata->pages[i] = page;
2692         }
2693
2694         if (rc) {
2695                 for (i = 0; i < nr_pages; i++) {
2696                         put_page(rdata->pages[i]);
2697                         rdata->pages[i] = NULL;
2698                 }
2699         }
2700         return rc;
2701 }
2702
2703 static void
2704 cifs_uncached_readdata_release(struct kref *refcount)
2705 {
2706         struct cifs_readdata *rdata = container_of(refcount,
2707                                         struct cifs_readdata, refcount);
2708         unsigned int i;
2709
2710         for (i = 0; i < rdata->nr_pages; i++) {
2711                 put_page(rdata->pages[i]);
2712                 rdata->pages[i] = NULL;
2713         }
2714         cifs_readdata_release(refcount);
2715 }
2716
2717 static int
2718 cifs_retry_async_readv(struct cifs_readdata *rdata)
2719 {
2720         int rc;
2721         struct TCP_Server_Info *server;
2722
2723         server = tlink_tcon(rdata->cfile->tlink)->ses->server;
2724
2725         do {
2726                 if (rdata->cfile->invalidHandle) {
2727                         rc = cifs_reopen_file(rdata->cfile, true);
2728                         if (rc != 0)
2729                                 continue;
2730                 }
2731                 rc = server->ops->async_readv(rdata);
2732         } while (rc == -EAGAIN);
2733
2734         return rc;
2735 }
2736
2737 /**
2738  * cifs_readdata_to_iov - copy data from pages in response to an iovec
2739  * @rdata:      the readdata response with list of pages holding data
2740  * @iov:        vector in which we should copy the data
2741  * @nr_segs:    number of segments in vector
2742  * @offset:     offset into file of the first iovec
2743  * @copied:     used to return the amount of data copied to the iov
2744  *
2745  * This function copies data from a list of pages in a readdata response into
2746  * an array of iovecs. It will first calculate where the data should go
2747  * based on the info in the readdata and then copy the data into that spot.
2748  */
2749 static ssize_t
2750 cifs_readdata_to_iov(struct cifs_readdata *rdata, const struct iovec *iov,
2751                         unsigned long nr_segs, loff_t offset, ssize_t *copied)
2752 {
2753         int rc = 0;
2754         struct iov_iter ii;
2755         size_t pos = rdata->offset - offset;
2756         ssize_t remaining = rdata->bytes;
2757         unsigned char *pdata;
2758         unsigned int i;
2759
2760         /* set up iov_iter and advance to the correct offset */
2761         iov_iter_init(&ii, iov, nr_segs, iov_length(iov, nr_segs), 0);
2762         iov_iter_advance(&ii, pos);
2763
2764         *copied = 0;
2765         for (i = 0; i < rdata->nr_pages; i++) {
2766                 ssize_t copy;
2767                 struct page *page = rdata->pages[i];
2768
2769                 /* copy a whole page or whatever's left */
2770                 copy = min_t(ssize_t, remaining, PAGE_SIZE);
2771
2772                 /* ...but limit it to whatever space is left in the iov */
2773                 copy = min_t(ssize_t, copy, iov_iter_count(&ii));
2774
2775                 /* go while there's data to be copied and no errors */
2776                 if (copy && !rc) {
2777                         pdata = kmap(page);
2778                         rc = memcpy_toiovecend(ii.iov, pdata, ii.iov_offset,
2779                                                 (int)copy);
2780                         kunmap(page);
2781                         if (!rc) {
2782                                 *copied += copy;
2783                                 remaining -= copy;
2784                                 iov_iter_advance(&ii, copy);
2785                         }
2786                 }
2787         }
2788
2789         return rc;
2790 }
2791
2792 static void
2793 cifs_uncached_readv_complete(struct work_struct *work)
2794 {
2795         struct cifs_readdata *rdata = container_of(work,
2796                                                 struct cifs_readdata, work);
2797
2798         complete(&rdata->done);
2799         kref_put(&rdata->refcount, cifs_uncached_readdata_release);
2800 }
2801
2802 static int
2803 cifs_uncached_read_into_pages(struct TCP_Server_Info *server,
2804                         struct cifs_readdata *rdata, unsigned int len)
2805 {
2806         int total_read = 0, result = 0;
2807         unsigned int i;
2808         unsigned int nr_pages = rdata->nr_pages;
2809         struct kvec iov;
2810
2811         rdata->tailsz = PAGE_SIZE;
2812         for (i = 0; i < nr_pages; i++) {
2813                 struct page *page = rdata->pages[i];
2814
2815                 if (len >= PAGE_SIZE) {
2816                         /* enough data to fill the page */
2817                         iov.iov_base = kmap(page);
2818                         iov.iov_len = PAGE_SIZE;
2819                         cifs_dbg(FYI, "%u: iov_base=%p iov_len=%zu\n",
2820                                  i, iov.iov_base, iov.iov_len);
2821                         len -= PAGE_SIZE;
2822                 } else if (len > 0) {
2823                         /* enough for partial page, fill and zero the rest */
2824                         iov.iov_base = kmap(page);
2825                         iov.iov_len = len;
2826                         cifs_dbg(FYI, "%u: iov_base=%p iov_len=%zu\n",
2827                                  i, iov.iov_base, iov.iov_len);
2828                         memset(iov.iov_base + len, '\0', PAGE_SIZE - len);
2829                         rdata->tailsz = len;
2830                         len = 0;
2831                 } else {
2832                         /* no need to hold page hostage */
2833                         rdata->pages[i] = NULL;
2834                         rdata->nr_pages--;
2835                         put_page(page);
2836                         continue;
2837                 }
2838
2839                 result = cifs_readv_from_socket(server, &iov, 1, iov.iov_len);
2840                 kunmap(page);
2841                 if (result < 0)
2842                         break;
2843
2844                 total_read += result;
2845         }
2846
2847         return total_read > 0 && result != -EAGAIN ? total_read : result;
2848 }
2849
2850 static ssize_t
2851 cifs_iovec_read(struct file *file, const struct iovec *iov,
2852                  unsigned long nr_segs, loff_t *poffset)
2853 {
2854         ssize_t rc;
2855         size_t len, cur_len;
2856         ssize_t total_read = 0;
2857         loff_t offset = *poffset;
2858         unsigned int npages;
2859         struct cifs_sb_info *cifs_sb;
2860         struct cifs_tcon *tcon;
2861         struct cifsFileInfo *open_file;
2862         struct cifs_readdata *rdata, *tmp;
2863         struct list_head rdata_list;
2864         pid_t pid;
2865
2866         if (!nr_segs)
2867                 return 0;
2868
2869         len = iov_length(iov, nr_segs);
2870         if (!len)
2871                 return 0;
2872
2873         INIT_LIST_HEAD(&rdata_list);
2874         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2875         open_file = file->private_data;
2876         tcon = tlink_tcon(open_file->tlink);
2877
2878         if (!tcon->ses->server->ops->async_readv)
2879                 return -ENOSYS;
2880
2881         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2882                 pid = open_file->pid;
2883         else
2884                 pid = current->tgid;
2885
2886         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
2887                 cifs_dbg(FYI, "attempting read on write only file instance\n");
2888
2889         do {
2890                 cur_len = min_t(const size_t, len - total_read, cifs_sb->rsize);
2891                 npages = DIV_ROUND_UP(cur_len, PAGE_SIZE);
2892
2893                 /* allocate a readdata struct */
2894                 rdata = cifs_readdata_alloc(npages,
2895                                             cifs_uncached_readv_complete);
2896                 if (!rdata) {
2897                         rc = -ENOMEM;
2898                         goto error;
2899                 }
2900
2901                 rc = cifs_read_allocate_pages(rdata, npages);
2902                 if (rc)
2903                         goto error;
2904
2905                 rdata->cfile = cifsFileInfo_get(open_file);
2906                 rdata->nr_pages = npages;
2907                 rdata->offset = offset;
2908                 rdata->bytes = cur_len;
2909                 rdata->pid = pid;
2910                 rdata->pagesz = PAGE_SIZE;
2911                 rdata->read_into_pages = cifs_uncached_read_into_pages;
2912
2913                 rc = cifs_retry_async_readv(rdata);
2914 error:
2915                 if (rc) {
2916                         kref_put(&rdata->refcount,
2917                                  cifs_uncached_readdata_release);
2918                         break;
2919                 }
2920
2921                 list_add_tail(&rdata->list, &rdata_list);
2922                 offset += cur_len;
2923                 len -= cur_len;
2924         } while (len > 0);
2925
2926         /* if at least one read request send succeeded, then reset rc */
2927         if (!list_empty(&rdata_list))
2928                 rc = 0;
2929
2930         /* the loop below should proceed in the order of increasing offsets */
2931 restart_loop:
2932         list_for_each_entry_safe(rdata, tmp, &rdata_list, list) {
2933                 if (!rc) {
2934                         ssize_t copied;
2935
2936                         /* FIXME: freezable sleep too? */
2937                         rc = wait_for_completion_killable(&rdata->done);
2938                         if (rc)
2939                                 rc = -EINTR;
2940                         else if (rdata->result)
2941                                 rc = rdata->result;
2942                         else {
2943                                 rc = cifs_readdata_to_iov(rdata, iov,
2944                                                         nr_segs, *poffset,
2945                                                         &copied);
2946                                 total_read += copied;
2947                         }
2948
2949                         /* resend call if it's a retryable error */
2950                         if (rc == -EAGAIN) {
2951                                 rc = cifs_retry_async_readv(rdata);
2952                                 goto restart_loop;
2953                         }
2954                 }
2955                 list_del_init(&rdata->list);
2956                 kref_put(&rdata->refcount, cifs_uncached_readdata_release);
2957         }
2958
2959         cifs_stats_bytes_read(tcon, total_read);
2960         *poffset += total_read;
2961
2962         /* mask nodata case */
2963         if (rc == -ENODATA)
2964                 rc = 0;
2965
2966         return total_read ? total_read : rc;
2967 }
2968
2969 ssize_t cifs_user_readv(struct kiocb *iocb, const struct iovec *iov,
2970                                unsigned long nr_segs, loff_t pos)
2971 {
2972         ssize_t read;
2973
2974         read = cifs_iovec_read(iocb->ki_filp, iov, nr_segs, &pos);
2975         if (read > 0)
2976                 iocb->ki_pos = pos;
2977
2978         return read;
2979 }
2980
2981 ssize_t
2982 cifs_strict_readv(struct kiocb *iocb, const struct iovec *iov,
2983                   unsigned long nr_segs, loff_t pos)
2984 {
2985         struct inode *inode = file_inode(iocb->ki_filp);
2986         struct cifsInodeInfo *cinode = CIFS_I(inode);
2987         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2988         struct cifsFileInfo *cfile = (struct cifsFileInfo *)
2989                                                 iocb->ki_filp->private_data;
2990         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
2991         int rc = -EACCES;
2992
2993         /*
2994          * In strict cache mode we need to read from the server all the time
2995          * if we don't have level II oplock because the server can delay mtime
2996          * change - so we can't make a decision about inode invalidating.
2997          * And we can also fail with pagereading if there are mandatory locks
2998          * on pages affected by this read but not on the region from pos to
2999          * pos+len-1.
3000          */
3001         if (!CIFS_CACHE_READ(cinode))
3002                 return cifs_user_readv(iocb, iov, nr_segs, pos);
3003
3004         if (cap_unix(tcon->ses) &&
3005             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
3006             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
3007                 return generic_file_aio_read(iocb, iov, nr_segs, pos);
3008
3009         /*
3010          * We need to hold the sem to be sure nobody modifies lock list
3011          * with a brlock that prevents reading.
3012          */
3013         down_read(&cinode->lock_sem);
3014         if (!cifs_find_lock_conflict(cfile, pos, iov_length(iov, nr_segs),
3015                                      tcon->ses->server->vals->shared_lock_type,
3016                                      NULL, CIFS_READ_OP))
3017                 rc = generic_file_aio_read(iocb, iov, nr_segs, pos);
3018         up_read(&cinode->lock_sem);
3019         return rc;
3020 }
3021
3022 static ssize_t
3023 cifs_read(struct file *file, char *read_data, size_t read_size, loff_t *offset)
3024 {
3025         int rc = -EACCES;
3026         unsigned int bytes_read = 0;
3027         unsigned int total_read;
3028         unsigned int current_read_size;
3029         unsigned int rsize;
3030         struct cifs_sb_info *cifs_sb;
3031         struct cifs_tcon *tcon;
3032         struct TCP_Server_Info *server;
3033         unsigned int xid;
3034         char *cur_offset;
3035         struct cifsFileInfo *open_file;
3036         struct cifs_io_parms io_parms;
3037         int buf_type = CIFS_NO_BUFFER;
3038         __u32 pid;
3039
3040         xid = get_xid();
3041         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
3042
3043         /* FIXME: set up handlers for larger reads and/or convert to async */
3044         rsize = min_t(unsigned int, cifs_sb->rsize, CIFSMaxBufSize);
3045
3046         if (file->private_data == NULL) {
3047                 rc = -EBADF;
3048                 free_xid(xid);
3049                 return rc;
3050         }
3051         open_file = file->private_data;
3052         tcon = tlink_tcon(open_file->tlink);
3053         server = tcon->ses->server;
3054
3055         if (!server->ops->sync_read) {
3056                 free_xid(xid);
3057                 return -ENOSYS;
3058         }
3059
3060         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3061                 pid = open_file->pid;
3062         else
3063                 pid = current->tgid;
3064
3065         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3066                 cifs_dbg(FYI, "attempting read on write only file instance\n");
3067
3068         for (total_read = 0, cur_offset = read_data; read_size > total_read;
3069              total_read += bytes_read, cur_offset += bytes_read) {
3070                 current_read_size = min_t(uint, read_size - total_read, rsize);
3071                 /*
3072                  * For windows me and 9x we do not want to request more than it
3073                  * negotiated since it will refuse the read then.
3074                  */
3075                 if ((tcon->ses) && !(tcon->ses->capabilities &
3076                                 tcon->ses->server->vals->cap_large_files)) {
3077                         current_read_size = min_t(uint, current_read_size,
3078                                         CIFSMaxBufSize);
3079                 }
3080                 rc = -EAGAIN;
3081                 while (rc == -EAGAIN) {
3082                         if (open_file->invalidHandle) {
3083                                 rc = cifs_reopen_file(open_file, true);
3084                                 if (rc != 0)
3085                                         break;
3086                         }
3087                         io_parms.pid = pid;
3088                         io_parms.tcon = tcon;
3089                         io_parms.offset = *offset;
3090                         io_parms.length = current_read_size;
3091                         rc = server->ops->sync_read(xid, open_file, &io_parms,
3092                                                     &bytes_read, &cur_offset,
3093                                                     &buf_type);
3094                 }
3095                 if (rc || (bytes_read == 0)) {
3096                         if (total_read) {
3097                                 break;
3098                         } else {
3099                                 free_xid(xid);
3100                                 return rc;
3101                         }
3102                 } else {
3103                         cifs_stats_bytes_read(tcon, total_read);
3104                         *offset += bytes_read;
3105                 }
3106         }
3107         free_xid(xid);
3108         return total_read;
3109 }
3110
3111 /*
3112  * If the page is mmap'ed into a process' page tables, then we need to make
3113  * sure that it doesn't change while being written back.
3114  */
3115 static int
3116 cifs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
3117 {
3118         struct page *page = vmf->page;
3119
3120         lock_page(page);
3121         return VM_FAULT_LOCKED;
3122 }
3123
3124 static struct vm_operations_struct cifs_file_vm_ops = {
3125         .fault = filemap_fault,
3126         .page_mkwrite = cifs_page_mkwrite,
3127         .remap_pages = generic_file_remap_pages,
3128 };
3129
3130 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
3131 {
3132         int rc, xid;
3133         struct inode *inode = file_inode(file);
3134
3135         xid = get_xid();
3136
3137         if (!CIFS_CACHE_READ(CIFS_I(inode))) {
3138                 rc = cifs_invalidate_mapping(inode);
3139                 if (rc)
3140                         return rc;
3141         }
3142
3143         rc = generic_file_mmap(file, vma);
3144         if (rc == 0)
3145                 vma->vm_ops = &cifs_file_vm_ops;
3146         free_xid(xid);
3147         return rc;
3148 }
3149
3150 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
3151 {
3152         int rc, xid;
3153
3154         xid = get_xid();
3155         rc = cifs_revalidate_file(file);
3156         if (rc) {
3157                 cifs_dbg(FYI, "Validation prior to mmap failed, error=%d\n",
3158                          rc);
3159                 free_xid(xid);
3160                 return rc;
3161         }
3162         rc = generic_file_mmap(file, vma);
3163         if (rc == 0)
3164                 vma->vm_ops = &cifs_file_vm_ops;
3165         free_xid(xid);
3166         return rc;
3167 }
3168
3169 static void
3170 cifs_readv_complete(struct work_struct *work)
3171 {
3172         unsigned int i;
3173         struct cifs_readdata *rdata = container_of(work,
3174                                                 struct cifs_readdata, work);
3175
3176         for (i = 0; i < rdata->nr_pages; i++) {
3177                 struct page *page = rdata->pages[i];
3178
3179                 lru_cache_add_file(page);
3180
3181                 if (rdata->result == 0) {
3182                         flush_dcache_page(page);
3183                         SetPageUptodate(page);
3184                 }
3185
3186                 unlock_page(page);
3187
3188                 if (rdata->result == 0)
3189                         cifs_readpage_to_fscache(rdata->mapping->host, page);
3190
3191                 page_cache_release(page);
3192                 rdata->pages[i] = NULL;
3193         }
3194         kref_put(&rdata->refcount, cifs_readdata_release);
3195 }
3196
3197 static int
3198 cifs_readpages_read_into_pages(struct TCP_Server_Info *server,
3199                         struct cifs_readdata *rdata, unsigned int len)
3200 {
3201         int total_read = 0, result = 0;
3202         unsigned int i;
3203         u64 eof;
3204         pgoff_t eof_index;
3205         unsigned int nr_pages = rdata->nr_pages;
3206         struct kvec iov;
3207
3208         /* determine the eof that the server (probably) has */
3209         eof = CIFS_I(rdata->mapping->host)->server_eof;
3210         eof_index = eof ? (eof - 1) >> PAGE_CACHE_SHIFT : 0;
3211         cifs_dbg(FYI, "eof=%llu eof_index=%lu\n", eof, eof_index);
3212
3213         rdata->tailsz = PAGE_CACHE_SIZE;
3214         for (i = 0; i < nr_pages; i++) {
3215                 struct page *page = rdata->pages[i];
3216
3217                 if (len >= PAGE_CACHE_SIZE) {
3218                         /* enough data to fill the page */
3219                         iov.iov_base = kmap(page);
3220                         iov.iov_len = PAGE_CACHE_SIZE;
3221                         cifs_dbg(FYI, "%u: idx=%lu iov_base=%p iov_len=%zu\n",
3222                                  i, page->index, iov.iov_base, iov.iov_len);
3223                         len -= PAGE_CACHE_SIZE;
3224                 } else if (len > 0) {
3225                         /* enough for partial page, fill and zero the rest */
3226                         iov.iov_base = kmap(page);
3227                         iov.iov_len = len;
3228                         cifs_dbg(FYI, "%u: idx=%lu iov_base=%p iov_len=%zu\n",
3229                                  i, page->index, iov.iov_base, iov.iov_len);
3230                         memset(iov.iov_base + len,
3231                                 '\0', PAGE_CACHE_SIZE - len);
3232                         rdata->tailsz = len;
3233                         len = 0;
3234                 } else if (page->index > eof_index) {
3235                         /*
3236                          * The VFS will not try to do readahead past the
3237                          * i_size, but it's possible that we have outstanding
3238                          * writes with gaps in the middle and the i_size hasn't
3239                          * caught up yet. Populate those with zeroed out pages
3240                          * to prevent the VFS from repeatedly attempting to
3241                          * fill them until the writes are flushed.
3242                          */
3243                         zero_user(page, 0, PAGE_CACHE_SIZE);
3244                         lru_cache_add_file(page);
3245                         flush_dcache_page(page);
3246                         SetPageUptodate(page);
3247                         unlock_page(page);
3248                         page_cache_release(page);
3249                         rdata->pages[i] = NULL;
3250                         rdata->nr_pages--;
3251                         continue;
3252                 } else {
3253                         /* no need to hold page hostage */
3254                         lru_cache_add_file(page);
3255                         unlock_page(page);
3256                         page_cache_release(page);
3257                         rdata->pages[i] = NULL;
3258                         rdata->nr_pages--;
3259                         continue;
3260                 }
3261
3262                 result = cifs_readv_from_socket(server, &iov, 1, iov.iov_len);
3263                 kunmap(page);
3264                 if (result < 0)
3265                         break;
3266
3267                 total_read += result;
3268         }
3269
3270         return total_read > 0 && result != -EAGAIN ? total_read : result;
3271 }
3272
3273 static int cifs_readpages(struct file *file, struct address_space *mapping,
3274         struct list_head *page_list, unsigned num_pages)
3275 {
3276         int rc;
3277         struct list_head tmplist;
3278         struct cifsFileInfo *open_file = file->private_data;
3279         struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
3280         unsigned int rsize = cifs_sb->rsize;
3281         pid_t pid;
3282
3283         /*
3284          * Give up immediately if rsize is too small to read an entire page.
3285          * The VFS will fall back to readpage. We should never reach this
3286          * point however since we set ra_pages to 0 when the rsize is smaller
3287          * than a cache page.
3288          */
3289         if (unlikely(rsize < PAGE_CACHE_SIZE))
3290                 return 0;
3291
3292         /*
3293          * Reads as many pages as possible from fscache. Returns -ENOBUFS
3294          * immediately if the cookie is negative
3295          *
3296          * After this point, every page in the list might have PG_fscache set,
3297          * so we will need to clean that up off of every page we don't use.
3298          */
3299         rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
3300                                          &num_pages);
3301         if (rc == 0)
3302                 return rc;
3303
3304         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3305                 pid = open_file->pid;
3306         else
3307                 pid = current->tgid;
3308
3309         rc = 0;
3310         INIT_LIST_HEAD(&tmplist);
3311
3312         cifs_dbg(FYI, "%s: file=%p mapping=%p num_pages=%u\n",
3313                  __func__, file, mapping, num_pages);
3314
3315         /*
3316          * Start with the page at end of list and move it to private
3317          * list. Do the same with any following pages until we hit
3318          * the rsize limit, hit an index discontinuity, or run out of
3319          * pages. Issue the async read and then start the loop again
3320          * until the list is empty.
3321          *
3322          * Note that list order is important. The page_list is in
3323          * the order of declining indexes. When we put the pages in
3324          * the rdata->pages, then we want them in increasing order.
3325          */
3326         while (!list_empty(page_list)) {
3327                 unsigned int i;
3328                 unsigned int bytes = PAGE_CACHE_SIZE;
3329                 unsigned int expected_index;
3330                 unsigned int nr_pages = 1;
3331                 loff_t offset;
3332                 struct page *page, *tpage;
3333                 struct cifs_readdata *rdata;
3334
3335                 page = list_entry(page_list->prev, struct page, lru);
3336
3337                 /*
3338                  * Lock the page and put it in the cache. Since no one else
3339                  * should have access to this page, we're safe to simply set
3340                  * PG_locked without checking it first.
3341                  */
3342                 __set_page_locked(page);
3343                 rc = add_to_page_cache_locked(page, mapping,
3344                                               page->index, GFP_KERNEL);
3345
3346                 /* give up if we can't stick it in the cache */
3347                 if (rc) {
3348                         __clear_page_locked(page);
3349                         break;
3350                 }
3351
3352                 /* move first page to the tmplist */
3353                 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
3354                 list_move_tail(&page->lru, &tmplist);
3355
3356                 /* now try and add more pages onto the request */
3357                 expected_index = page->index + 1;
3358                 list_for_each_entry_safe_reverse(page, tpage, page_list, lru) {
3359                         /* discontinuity ? */
3360                         if (page->index != expected_index)
3361                                 break;
3362
3363                         /* would this page push the read over the rsize? */
3364                         if (bytes + PAGE_CACHE_SIZE > rsize)
3365                                 break;
3366
3367                         __set_page_locked(page);
3368                         if (add_to_page_cache_locked(page, mapping,
3369                                                 page->index, GFP_KERNEL)) {
3370                                 __clear_page_locked(page);
3371                                 break;
3372                         }
3373                         list_move_tail(&page->lru, &tmplist);
3374                         bytes += PAGE_CACHE_SIZE;
3375                         expected_index++;
3376                         nr_pages++;
3377                 }
3378
3379                 rdata = cifs_readdata_alloc(nr_pages, cifs_readv_complete);
3380                 if (!rdata) {
3381                         /* best to give up if we're out of mem */
3382                         list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3383                                 list_del(&page->lru);
3384                                 lru_cache_add_file(page);
3385                                 unlock_page(page);
3386                                 page_cache_release(page);
3387                         }
3388                         rc = -ENOMEM;
3389                         break;
3390                 }
3391
3392                 rdata->cfile = cifsFileInfo_get(open_file);
3393                 rdata->mapping = mapping;
3394                 rdata->offset = offset;
3395                 rdata->bytes = bytes;
3396                 rdata->pid = pid;
3397                 rdata->pagesz = PAGE_CACHE_SIZE;
3398                 rdata->read_into_pages = cifs_readpages_read_into_pages;
3399
3400                 list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3401                         list_del(&page->lru);
3402                         rdata->pages[rdata->nr_pages++] = page;
3403                 }
3404
3405                 rc = cifs_retry_async_readv(rdata);
3406                 if (rc != 0) {
3407                         for (i = 0; i < rdata->nr_pages; i++) {
3408                                 page = rdata->pages[i];
3409                                 lru_cache_add_file(page);
3410                                 unlock_page(page);
3411                                 page_cache_release(page);
3412                         }
3413                         kref_put(&rdata->refcount, cifs_readdata_release);
3414                         break;
3415                 }
3416
3417                 kref_put(&rdata->refcount, cifs_readdata_release);
3418         }
3419
3420         /* Any pages that have been shown to fscache but didn't get added to
3421          * the pagecache must be uncached before they get returned to the
3422          * allocator.
3423          */
3424         cifs_fscache_readpages_cancel(mapping->host, page_list);
3425         return rc;
3426 }
3427
3428 /*
3429  * cifs_readpage_worker must be called with the page pinned
3430  */
3431 static int cifs_readpage_worker(struct file *file, struct page *page,
3432         loff_t *poffset)
3433 {
3434         char *read_data;
3435         int rc;
3436
3437         /* Is the page cached? */
3438         rc = cifs_readpage_from_fscache(file_inode(file), page);
3439         if (rc == 0)
3440                 goto read_complete;
3441
3442         read_data = kmap(page);
3443         /* for reads over a certain size could initiate async read ahead */
3444
3445         rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
3446
3447         if (rc < 0)
3448                 goto io_error;
3449         else
3450                 cifs_dbg(FYI, "Bytes read %d\n", rc);
3451
3452         file_inode(file)->i_atime =
3453                 current_fs_time(file_inode(file)->i_sb);
3454
3455         if (PAGE_CACHE_SIZE > rc)
3456                 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
3457
3458         flush_dcache_page(page);
3459         SetPageUptodate(page);
3460
3461         /* send this page to the cache */
3462         cifs_readpage_to_fscache(file_inode(file), page);
3463
3464         rc = 0;
3465
3466 io_error:
3467         kunmap(page);
3468         unlock_page(page);
3469
3470 read_complete:
3471         return rc;
3472 }
3473
3474 static int cifs_readpage(struct file *file, struct page *page)
3475 {
3476         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
3477         int rc = -EACCES;
3478         unsigned int xid;
3479
3480         xid = get_xid();
3481
3482         if (file->private_data == NULL) {
3483                 rc = -EBADF;
3484                 free_xid(xid);
3485                 return rc;
3486         }
3487
3488         cifs_dbg(FYI, "readpage %p at offset %d 0x%x\n",
3489                  page, (int)offset, (int)offset);
3490
3491         rc = cifs_readpage_worker(file, page, &offset);
3492
3493         free_xid(xid);
3494         return rc;
3495 }
3496
3497 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
3498 {
3499         struct cifsFileInfo *open_file;
3500
3501         spin_lock(&cifs_file_list_lock);
3502         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
3503                 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
3504                         spin_unlock(&cifs_file_list_lock);
3505                         return 1;
3506                 }
3507         }
3508         spin_unlock(&cifs_file_list_lock);
3509         return 0;
3510 }
3511
3512 /* We do not want to update the file size from server for inodes
3513    open for write - to avoid races with writepage extending
3514    the file - in the future we could consider allowing
3515    refreshing the inode only on increases in the file size
3516    but this is tricky to do without racing with writebehind
3517    page caching in the current Linux kernel design */
3518 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
3519 {
3520         if (!cifsInode)
3521                 return true;
3522
3523         if (is_inode_writable(cifsInode)) {
3524                 /* This inode is open for write at least once */
3525                 struct cifs_sb_info *cifs_sb;
3526
3527                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
3528                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
3529                         /* since no page cache to corrupt on directio
3530                         we can change size safely */
3531                         return true;
3532                 }
3533
3534                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
3535                         return true;
3536
3537                 return false;
3538         } else
3539                 return true;
3540 }
3541
3542 static int cifs_write_begin(struct file *file, struct address_space *mapping,
3543                         loff_t pos, unsigned len, unsigned flags,
3544                         struct page **pagep, void **fsdata)
3545 {
3546         int oncethru = 0;
3547         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
3548         loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
3549         loff_t page_start = pos & PAGE_MASK;
3550         loff_t i_size;
3551         struct page *page;
3552         int rc = 0;
3553
3554         cifs_dbg(FYI, "write_begin from %lld len %d\n", (long long)pos, len);
3555
3556 start:
3557         page = grab_cache_page_write_begin(mapping, index, flags);
3558         if (!page) {
3559                 rc = -ENOMEM;
3560                 goto out;
3561         }
3562
3563         if (PageUptodate(page))
3564                 goto out;
3565
3566         /*
3567          * If we write a full page it will be up to date, no need to read from
3568          * the server. If the write is short, we'll end up doing a sync write
3569          * instead.
3570          */
3571         if (len == PAGE_CACHE_SIZE)
3572                 goto out;
3573
3574         /*
3575          * optimize away the read when we have an oplock, and we're not
3576          * expecting to use any of the data we'd be reading in. That
3577          * is, when the page lies beyond the EOF, or straddles the EOF
3578          * and the write will cover all of the existing data.
3579          */
3580         if (CIFS_CACHE_READ(CIFS_I(mapping->host))) {
3581                 i_size = i_size_read(mapping->host);
3582                 if (page_start >= i_size ||
3583                     (offset == 0 && (pos + len) >= i_size)) {
3584                         zero_user_segments(page, 0, offset,
3585                                            offset + len,
3586                                            PAGE_CACHE_SIZE);
3587                         /*
3588                          * PageChecked means that the parts of the page
3589                          * to which we're not writing are considered up
3590                          * to date. Once the data is copied to the
3591                          * page, it can be set uptodate.
3592                          */
3593                         SetPageChecked(page);
3594                         goto out;
3595                 }
3596         }
3597
3598         if ((file->f_flags & O_ACCMODE) != O_WRONLY && !oncethru) {
3599                 /*
3600                  * might as well read a page, it is fast enough. If we get
3601                  * an error, we don't need to return it. cifs_write_end will
3602                  * do a sync write instead since PG_uptodate isn't set.
3603                  */
3604                 cifs_readpage_worker(file, page, &page_start);
3605                 page_cache_release(page);
3606                 oncethru = 1;
3607                 goto start;
3608         } else {
3609                 /* we could try using another file handle if there is one -
3610                    but how would we lock it to prevent close of that handle
3611                    racing with this read? In any case
3612                    this will be written out by write_end so is fine */
3613         }
3614 out:
3615         *pagep = page;
3616         return rc;
3617 }
3618
3619 static int cifs_release_page(struct page *page, gfp_t gfp)
3620 {
3621         if (PagePrivate(page))
3622                 return 0;
3623
3624         return cifs_fscache_release_page(page, gfp);
3625 }
3626
3627 static void cifs_invalidate_page(struct page *page, unsigned int offset,
3628                                  unsigned int length)
3629 {
3630         struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
3631
3632         if (offset == 0 && length == PAGE_CACHE_SIZE)
3633                 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
3634 }
3635
3636 static int cifs_launder_page(struct page *page)
3637 {
3638         int rc = 0;
3639         loff_t range_start = page_offset(page);
3640         loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
3641         struct writeback_control wbc = {
3642                 .sync_mode = WB_SYNC_ALL,
3643                 .nr_to_write = 0,
3644                 .range_start = range_start,
3645                 .range_end = range_end,
3646         };
3647
3648         cifs_dbg(FYI, "Launder page: %p\n", page);
3649
3650         if (clear_page_dirty_for_io(page))
3651                 rc = cifs_writepage_locked(page, &wbc);
3652
3653         cifs_fscache_invalidate_page(page, page->mapping->host);
3654         return rc;
3655 }
3656
3657 static int
3658 cifs_pending_writers_wait(void *unused)
3659 {
3660         schedule();
3661         return 0;
3662 }
3663
3664 void cifs_oplock_break(struct work_struct *work)
3665 {
3666         struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
3667                                                   oplock_break);
3668         struct inode *inode = cfile->dentry->d_inode;
3669         struct cifsInodeInfo *cinode = CIFS_I(inode);
3670         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
3671         struct TCP_Server_Info *server = tcon->ses->server;
3672         int rc = 0;
3673
3674         wait_on_bit(&cinode->flags, CIFS_INODE_PENDING_WRITERS,
3675                         cifs_pending_writers_wait, TASK_UNINTERRUPTIBLE);
3676
3677         server->ops->downgrade_oplock(server, cinode,
3678                 test_bit(CIFS_INODE_DOWNGRADE_OPLOCK_TO_L2, &cinode->flags));
3679
3680         if (!CIFS_CACHE_WRITE(cinode) && CIFS_CACHE_READ(cinode) &&
3681                                                 cifs_has_mand_locks(cinode)) {
3682                 cifs_dbg(FYI, "Reset oplock to None for inode=%p due to mand locks\n",
3683                          inode);
3684                 cinode->oplock = 0;
3685         }
3686
3687         if (inode && S_ISREG(inode->i_mode)) {
3688                 if (CIFS_CACHE_READ(cinode))
3689                         break_lease(inode, O_RDONLY);
3690                 else
3691                         break_lease(inode, O_WRONLY);
3692                 rc = filemap_fdatawrite(inode->i_mapping);
3693                 if (!CIFS_CACHE_READ(cinode)) {
3694                         rc = filemap_fdatawait(inode->i_mapping);
3695                         mapping_set_error(inode->i_mapping, rc);
3696                         cifs_invalidate_mapping(inode);
3697                 }
3698                 cifs_dbg(FYI, "Oplock flush inode %p rc %d\n", inode, rc);
3699         }
3700
3701         rc = cifs_push_locks(cfile);
3702         if (rc)
3703                 cifs_dbg(VFS, "Push locks rc = %d\n", rc);
3704
3705         /*
3706          * releasing stale oplock after recent reconnect of smb session using
3707          * a now incorrect file handle is not a data integrity issue but do
3708          * not bother sending an oplock release if session to server still is
3709          * disconnected since oplock already released by the server
3710          */
3711         if (!cfile->oplock_break_cancelled) {
3712                 rc = tcon->ses->server->ops->oplock_response(tcon, &cfile->fid,
3713                                                              cinode);
3714                 cifs_dbg(FYI, "Oplock release rc = %d\n", rc);
3715         }
3716         cifs_done_oplock_break(cinode);
3717 }
3718
3719 /*
3720  * The presence of cifs_direct_io() in the address space ops vector
3721  * allowes open() O_DIRECT flags which would have failed otherwise.
3722  *
3723  * In the non-cached mode (mount with cache=none), we shunt off direct read and write requests
3724  * so this method should never be called.
3725  *
3726  * Direct IO is not yet supported in the cached mode.
3727  */
3728 static ssize_t
3729 cifs_direct_io(int rw, struct kiocb *iocb, const struct iovec *iov,
3730                loff_t pos, unsigned long nr_segs)
3731 {
3732         /*
3733          * FIXME
3734          * Eventually need to support direct IO for non forcedirectio mounts
3735          */
3736         return -EINVAL;
3737 }
3738
3739
3740 const struct address_space_operations cifs_addr_ops = {
3741         .readpage = cifs_readpage,
3742         .readpages = cifs_readpages,
3743         .writepage = cifs_writepage,
3744         .writepages = cifs_writepages,
3745         .write_begin = cifs_write_begin,
3746         .write_end = cifs_write_end,
3747         .set_page_dirty = __set_page_dirty_nobuffers,
3748         .releasepage = cifs_release_page,
3749         .direct_IO = cifs_direct_io,
3750         .invalidatepage = cifs_invalidate_page,
3751         .launder_page = cifs_launder_page,
3752 };
3753
3754 /*
3755  * cifs_readpages requires the server to support a buffer large enough to
3756  * contain the header plus one complete page of data.  Otherwise, we need
3757  * to leave cifs_readpages out of the address space operations.
3758  */
3759 const struct address_space_operations cifs_addr_ops_smallbuf = {
3760         .readpage = cifs_readpage,
3761         .writepage = cifs_writepage,
3762         .writepages = cifs_writepages,
3763         .write_begin = cifs_write_begin,
3764         .write_end = cifs_write_end,
3765         .set_page_dirty = __set_page_dirty_nobuffers,
3766         .releasepage = cifs_release_page,
3767         .invalidatepage = cifs_invalidate_page,
3768         .launder_page = cifs_launder_page,
3769 };