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_zap_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                         set_bit(CIFS_INO_INVALID_MAPPING, &cifsi->flags);
 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_zap_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 unsigned int
1882 wdata_prepare_pages(struct cifs_writedata *wdata, unsigned int found_pages,
1883                     struct address_space *mapping,
1884                     struct writeback_control *wbc,
1885                     pgoff_t end, pgoff_t *index, pgoff_t *next, bool *done)
1886 {
1887         unsigned int nr_pages = 0, i;
1888         struct page *page;
1889
1890         for (i = 0; i < found_pages; i++) {
1891                 page = wdata->pages[i];
1892                 /*
1893                  * At this point we hold neither mapping->tree_lock nor
1894                  * lock on the page itself: the page may be truncated or
1895                  * invalidated (changing page->mapping to NULL), or even
1896                  * swizzled back from swapper_space to tmpfs file
1897                  * mapping
1898                  */
1899
1900                 if (nr_pages == 0)
1901                         lock_page(page);
1902                 else if (!trylock_page(page))
1903                         break;
1904
1905                 if (unlikely(page->mapping != mapping)) {
1906                         unlock_page(page);
1907                         break;
1908                 }
1909
1910                 if (!wbc->range_cyclic && page->index > end) {
1911                         *done = true;
1912                         unlock_page(page);
1913                         break;
1914                 }
1915
1916                 if (*next && (page->index != *next)) {
1917                         /* Not next consecutive page */
1918                         unlock_page(page);
1919                         break;
1920                 }
1921
1922                 if (wbc->sync_mode != WB_SYNC_NONE)
1923                         wait_on_page_writeback(page);
1924
1925                 if (PageWriteback(page) ||
1926                                 !clear_page_dirty_for_io(page)) {
1927                         unlock_page(page);
1928                         break;
1929                 }
1930
1931                 /*
1932                  * This actually clears the dirty bit in the radix tree.
1933                  * See cifs_writepage() for more commentary.
1934                  */
1935                 set_page_writeback(page);
1936                 if (page_offset(page) >= i_size_read(mapping->host)) {
1937                         *done = true;
1938                         unlock_page(page);
1939                         end_page_writeback(page);
1940                         break;
1941                 }
1942
1943                 wdata->pages[i] = page;
1944                 *next = page->index + 1;
1945                 ++nr_pages;
1946         }
1947
1948         /* reset index to refind any pages skipped */
1949         if (nr_pages == 0)
1950                 *index = wdata->pages[0]->index + 1;
1951
1952         /* put any pages we aren't going to use */
1953         for (i = nr_pages; i < found_pages; i++) {
1954                 page_cache_release(wdata->pages[i]);
1955                 wdata->pages[i] = NULL;
1956         }
1957
1958         return nr_pages;
1959 }
1960
1961 static int cifs_writepages(struct address_space *mapping,
1962                            struct writeback_control *wbc)
1963 {
1964         struct cifs_sb_info *cifs_sb = CIFS_SB(mapping->host->i_sb);
1965         bool done = false, scanned = false, range_whole = false;
1966         pgoff_t end, index;
1967         struct cifs_writedata *wdata;
1968         struct TCP_Server_Info *server;
1969         int rc = 0;
1970
1971         /*
1972          * If wsize is smaller than the page cache size, default to writing
1973          * one page at a time via cifs_writepage
1974          */
1975         if (cifs_sb->wsize < PAGE_CACHE_SIZE)
1976                 return generic_writepages(mapping, wbc);
1977
1978         if (wbc->range_cyclic) {
1979                 index = mapping->writeback_index; /* Start from prev offset */
1980                 end = -1;
1981         } else {
1982                 index = wbc->range_start >> PAGE_CACHE_SHIFT;
1983                 end = wbc->range_end >> PAGE_CACHE_SHIFT;
1984                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1985                         range_whole = true;
1986                 scanned = true;
1987         }
1988 retry:
1989         while (!done && index <= end) {
1990                 unsigned int i, nr_pages, found_pages;
1991                 pgoff_t next = 0, tofind;
1992                 struct page **pages;
1993
1994                 tofind = min((cifs_sb->wsize / PAGE_CACHE_SIZE) - 1,
1995                                 end - index) + 1;
1996
1997                 wdata = cifs_writedata_alloc((unsigned int)tofind,
1998                                              cifs_writev_complete);
1999                 if (!wdata) {
2000                         rc = -ENOMEM;
2001                         break;
2002                 }
2003
2004                 /*
2005                  * find_get_pages_tag seems to return a max of 256 on each
2006                  * iteration, so we must call it several times in order to
2007                  * fill the array or the wsize is effectively limited to
2008                  * 256 * PAGE_CACHE_SIZE.
2009                  */
2010                 found_pages = 0;
2011                 pages = wdata->pages;
2012                 do {
2013                         nr_pages = find_get_pages_tag(mapping, &index,
2014                                                         PAGECACHE_TAG_DIRTY,
2015                                                         tofind, pages);
2016                         found_pages += nr_pages;
2017                         tofind -= nr_pages;
2018                         pages += nr_pages;
2019                 } while (nr_pages && tofind && index <= end);
2020
2021                 if (found_pages == 0) {
2022                         kref_put(&wdata->refcount, cifs_writedata_release);
2023                         break;
2024                 }
2025
2026                 nr_pages = wdata_prepare_pages(wdata, found_pages, mapping, wbc,
2027                                                end, &index, &next, &done);
2028
2029                 /* nothing to write? */
2030                 if (nr_pages == 0) {
2031                         kref_put(&wdata->refcount, cifs_writedata_release);
2032                         continue;
2033                 }
2034
2035                 wdata->sync_mode = wbc->sync_mode;
2036                 wdata->nr_pages = nr_pages;
2037                 wdata->offset = page_offset(wdata->pages[0]);
2038                 wdata->pagesz = PAGE_CACHE_SIZE;
2039                 wdata->tailsz =
2040                         min(i_size_read(mapping->host) -
2041                             page_offset(wdata->pages[nr_pages - 1]),
2042                             (loff_t)PAGE_CACHE_SIZE);
2043                 wdata->bytes = ((nr_pages - 1) * PAGE_CACHE_SIZE) +
2044                                         wdata->tailsz;
2045
2046                 do {
2047                         if (wdata->cfile != NULL)
2048                                 cifsFileInfo_put(wdata->cfile);
2049                         wdata->cfile = find_writable_file(CIFS_I(mapping->host),
2050                                                           false);
2051                         if (!wdata->cfile) {
2052                                 cifs_dbg(VFS, "No writable handles for inode\n");
2053                                 rc = -EBADF;
2054                                 break;
2055                         }
2056                         wdata->pid = wdata->cfile->pid;
2057                         server = tlink_tcon(wdata->cfile->tlink)->ses->server;
2058                         rc = server->ops->async_writev(wdata,
2059                                                         cifs_writedata_release);
2060                 } while (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN);
2061
2062                 for (i = 0; i < nr_pages; ++i)
2063                         unlock_page(wdata->pages[i]);
2064
2065                 /* send failure -- clean up the mess */
2066                 if (rc != 0) {
2067                         for (i = 0; i < nr_pages; ++i) {
2068                                 if (rc == -EAGAIN)
2069                                         redirty_page_for_writepage(wbc,
2070                                                            wdata->pages[i]);
2071                                 else
2072                                         SetPageError(wdata->pages[i]);
2073                                 end_page_writeback(wdata->pages[i]);
2074                                 page_cache_release(wdata->pages[i]);
2075                         }
2076                         if (rc != -EAGAIN)
2077                                 mapping_set_error(mapping, rc);
2078                 }
2079                 kref_put(&wdata->refcount, cifs_writedata_release);
2080
2081                 wbc->nr_to_write -= nr_pages;
2082                 if (wbc->nr_to_write <= 0)
2083                         done = true;
2084
2085                 index = next;
2086         }
2087
2088         if (!scanned && !done) {
2089                 /*
2090                  * We hit the last page and there is more work to be done: wrap
2091                  * back to the start of the file
2092                  */
2093                 scanned = true;
2094                 index = 0;
2095                 goto retry;
2096         }
2097
2098         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2099                 mapping->writeback_index = index;
2100
2101         return rc;
2102 }
2103
2104 static int
2105 cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
2106 {
2107         int rc;
2108         unsigned int xid;
2109
2110         xid = get_xid();
2111 /* BB add check for wbc flags */
2112         page_cache_get(page);
2113         if (!PageUptodate(page))
2114                 cifs_dbg(FYI, "ppw - page not up to date\n");
2115
2116         /*
2117          * Set the "writeback" flag, and clear "dirty" in the radix tree.
2118          *
2119          * A writepage() implementation always needs to do either this,
2120          * or re-dirty the page with "redirty_page_for_writepage()" in
2121          * the case of a failure.
2122          *
2123          * Just unlocking the page will cause the radix tree tag-bits
2124          * to fail to update with the state of the page correctly.
2125          */
2126         set_page_writeback(page);
2127 retry_write:
2128         rc = cifs_partialpagewrite(page, 0, PAGE_CACHE_SIZE);
2129         if (rc == -EAGAIN && wbc->sync_mode == WB_SYNC_ALL)
2130                 goto retry_write;
2131         else if (rc == -EAGAIN)
2132                 redirty_page_for_writepage(wbc, page);
2133         else if (rc != 0)
2134                 SetPageError(page);
2135         else
2136                 SetPageUptodate(page);
2137         end_page_writeback(page);
2138         page_cache_release(page);
2139         free_xid(xid);
2140         return rc;
2141 }
2142
2143 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
2144 {
2145         int rc = cifs_writepage_locked(page, wbc);
2146         unlock_page(page);
2147         return rc;
2148 }
2149
2150 static int cifs_write_end(struct file *file, struct address_space *mapping,
2151                         loff_t pos, unsigned len, unsigned copied,
2152                         struct page *page, void *fsdata)
2153 {
2154         int rc;
2155         struct inode *inode = mapping->host;
2156         struct cifsFileInfo *cfile = file->private_data;
2157         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
2158         __u32 pid;
2159
2160         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2161                 pid = cfile->pid;
2162         else
2163                 pid = current->tgid;
2164
2165         cifs_dbg(FYI, "write_end for page %p from pos %lld with %d bytes\n",
2166                  page, pos, copied);
2167
2168         if (PageChecked(page)) {
2169                 if (copied == len)
2170                         SetPageUptodate(page);
2171                 ClearPageChecked(page);
2172         } else if (!PageUptodate(page) && copied == PAGE_CACHE_SIZE)
2173                 SetPageUptodate(page);
2174
2175         if (!PageUptodate(page)) {
2176                 char *page_data;
2177                 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
2178                 unsigned int xid;
2179
2180                 xid = get_xid();
2181                 /* this is probably better than directly calling
2182                    partialpage_write since in this function the file handle is
2183                    known which we might as well leverage */
2184                 /* BB check if anything else missing out of ppw
2185                    such as updating last write time */
2186                 page_data = kmap(page);
2187                 rc = cifs_write(cfile, pid, page_data + offset, copied, &pos);
2188                 /* if (rc < 0) should we set writebehind rc? */
2189                 kunmap(page);
2190
2191                 free_xid(xid);
2192         } else {
2193                 rc = copied;
2194                 pos += copied;
2195                 set_page_dirty(page);
2196         }
2197
2198         if (rc > 0) {
2199                 spin_lock(&inode->i_lock);
2200                 if (pos > inode->i_size)
2201                         i_size_write(inode, pos);
2202                 spin_unlock(&inode->i_lock);
2203         }
2204
2205         unlock_page(page);
2206         page_cache_release(page);
2207
2208         return rc;
2209 }
2210
2211 int cifs_strict_fsync(struct file *file, loff_t start, loff_t end,
2212                       int datasync)
2213 {
2214         unsigned int xid;
2215         int rc = 0;
2216         struct cifs_tcon *tcon;
2217         struct TCP_Server_Info *server;
2218         struct cifsFileInfo *smbfile = file->private_data;
2219         struct inode *inode = file_inode(file);
2220         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2221
2222         rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2223         if (rc)
2224                 return rc;
2225         mutex_lock(&inode->i_mutex);
2226
2227         xid = get_xid();
2228
2229         cifs_dbg(FYI, "Sync file - name: %s datasync: 0x%x\n",
2230                  file->f_path.dentry->d_name.name, datasync);
2231
2232         if (!CIFS_CACHE_READ(CIFS_I(inode))) {
2233                 rc = cifs_zap_mapping(inode);
2234                 if (rc) {
2235                         cifs_dbg(FYI, "rc: %d during invalidate phase\n", rc);
2236                         rc = 0; /* don't care about it in fsync */
2237                 }
2238         }
2239
2240         tcon = tlink_tcon(smbfile->tlink);
2241         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2242                 server = tcon->ses->server;
2243                 if (server->ops->flush)
2244                         rc = server->ops->flush(xid, tcon, &smbfile->fid);
2245                 else
2246                         rc = -ENOSYS;
2247         }
2248
2249         free_xid(xid);
2250         mutex_unlock(&inode->i_mutex);
2251         return rc;
2252 }
2253
2254 int cifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2255 {
2256         unsigned int xid;
2257         int rc = 0;
2258         struct cifs_tcon *tcon;
2259         struct TCP_Server_Info *server;
2260         struct cifsFileInfo *smbfile = file->private_data;
2261         struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2262         struct inode *inode = file->f_mapping->host;
2263
2264         rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2265         if (rc)
2266                 return rc;
2267         mutex_lock(&inode->i_mutex);
2268
2269         xid = get_xid();
2270
2271         cifs_dbg(FYI, "Sync file - name: %s datasync: 0x%x\n",
2272                  file->f_path.dentry->d_name.name, datasync);
2273
2274         tcon = tlink_tcon(smbfile->tlink);
2275         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2276                 server = tcon->ses->server;
2277                 if (server->ops->flush)
2278                         rc = server->ops->flush(xid, tcon, &smbfile->fid);
2279                 else
2280                         rc = -ENOSYS;
2281         }
2282
2283         free_xid(xid);
2284         mutex_unlock(&inode->i_mutex);
2285         return rc;
2286 }
2287
2288 /*
2289  * As file closes, flush all cached write data for this inode checking
2290  * for write behind errors.
2291  */
2292 int cifs_flush(struct file *file, fl_owner_t id)
2293 {
2294         struct inode *inode = file_inode(file);
2295         int rc = 0;
2296
2297         if (file->f_mode & FMODE_WRITE)
2298                 rc = filemap_write_and_wait(inode->i_mapping);
2299
2300         cifs_dbg(FYI, "Flush inode %p file %p rc %d\n", inode, file, rc);
2301
2302         return rc;
2303 }
2304
2305 static int
2306 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
2307 {
2308         int rc = 0;
2309         unsigned long i;
2310
2311         for (i = 0; i < num_pages; i++) {
2312                 pages[i] = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2313                 if (!pages[i]) {
2314                         /*
2315                          * save number of pages we have already allocated and
2316                          * return with ENOMEM error
2317                          */
2318                         num_pages = i;
2319                         rc = -ENOMEM;
2320                         break;
2321                 }
2322         }
2323
2324         if (rc) {
2325                 for (i = 0; i < num_pages; i++)
2326                         put_page(pages[i]);
2327         }
2328         return rc;
2329 }
2330
2331 static inline
2332 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
2333 {
2334         size_t num_pages;
2335         size_t clen;
2336
2337         clen = min_t(const size_t, len, wsize);
2338         num_pages = DIV_ROUND_UP(clen, PAGE_SIZE);
2339
2340         if (cur_len)
2341                 *cur_len = clen;
2342
2343         return num_pages;
2344 }
2345
2346 static void
2347 cifs_uncached_writedata_release(struct kref *refcount)
2348 {
2349         int i;
2350         struct cifs_writedata *wdata = container_of(refcount,
2351                                         struct cifs_writedata, refcount);
2352
2353         for (i = 0; i < wdata->nr_pages; i++)
2354                 put_page(wdata->pages[i]);
2355         cifs_writedata_release(refcount);
2356 }
2357
2358 static void
2359 cifs_uncached_writev_complete(struct work_struct *work)
2360 {
2361         struct cifs_writedata *wdata = container_of(work,
2362                                         struct cifs_writedata, work);
2363         struct inode *inode = wdata->cfile->dentry->d_inode;
2364         struct cifsInodeInfo *cifsi = CIFS_I(inode);
2365
2366         spin_lock(&inode->i_lock);
2367         cifs_update_eof(cifsi, wdata->offset, wdata->bytes);
2368         if (cifsi->server_eof > inode->i_size)
2369                 i_size_write(inode, cifsi->server_eof);
2370         spin_unlock(&inode->i_lock);
2371
2372         complete(&wdata->done);
2373
2374         kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2375 }
2376
2377 /* attempt to send write to server, retry on any -EAGAIN errors */
2378 static int
2379 cifs_uncached_retry_writev(struct cifs_writedata *wdata)
2380 {
2381         int rc;
2382         struct TCP_Server_Info *server;
2383
2384         server = tlink_tcon(wdata->cfile->tlink)->ses->server;
2385
2386         do {
2387                 if (wdata->cfile->invalidHandle) {
2388                         rc = cifs_reopen_file(wdata->cfile, false);
2389                         if (rc != 0)
2390                                 continue;
2391                 }
2392                 rc = server->ops->async_writev(wdata,
2393                                                cifs_uncached_writedata_release);
2394         } while (rc == -EAGAIN);
2395
2396         return rc;
2397 }
2398
2399 static ssize_t
2400 cifs_iovec_write(struct file *file, struct iov_iter *from, loff_t *poffset)
2401 {
2402         unsigned long nr_pages, i;
2403         size_t bytes, copied, len, cur_len;
2404         ssize_t total_written = 0;
2405         loff_t offset;
2406         struct cifsFileInfo *open_file;
2407         struct cifs_tcon *tcon;
2408         struct cifs_sb_info *cifs_sb;
2409         struct cifs_writedata *wdata, *tmp;
2410         struct list_head wdata_list;
2411         int rc;
2412         pid_t pid;
2413
2414         len = iov_iter_count(from);
2415         rc = generic_write_checks(file, poffset, &len, 0);
2416         if (rc)
2417                 return rc;
2418
2419         if (!len)
2420                 return 0;
2421
2422         iov_iter_truncate(from, len);
2423
2424         INIT_LIST_HEAD(&wdata_list);
2425         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2426         open_file = file->private_data;
2427         tcon = tlink_tcon(open_file->tlink);
2428
2429         if (!tcon->ses->server->ops->async_writev)
2430                 return -ENOSYS;
2431
2432         offset = *poffset;
2433
2434         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2435                 pid = open_file->pid;
2436         else
2437                 pid = current->tgid;
2438
2439         do {
2440                 size_t save_len;
2441
2442                 nr_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
2443                 wdata = cifs_writedata_alloc(nr_pages,
2444                                              cifs_uncached_writev_complete);
2445                 if (!wdata) {
2446                         rc = -ENOMEM;
2447                         break;
2448                 }
2449
2450                 rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
2451                 if (rc) {
2452                         kfree(wdata);
2453                         break;
2454                 }
2455
2456                 save_len = cur_len;
2457                 for (i = 0; i < nr_pages; i++) {
2458                         bytes = min_t(size_t, cur_len, PAGE_SIZE);
2459                         copied = copy_page_from_iter(wdata->pages[i], 0, bytes,
2460                                                      from);
2461                         cur_len -= copied;
2462                         /*
2463                          * If we didn't copy as much as we expected, then that
2464                          * may mean we trod into an unmapped area. Stop copying
2465                          * at that point. On the next pass through the big
2466                          * loop, we'll likely end up getting a zero-length
2467                          * write and bailing out of it.
2468                          */
2469                         if (copied < bytes)
2470                                 break;
2471                 }
2472                 cur_len = save_len - cur_len;
2473
2474                 /*
2475                  * If we have no data to send, then that probably means that
2476                  * the copy above failed altogether. That's most likely because
2477                  * the address in the iovec was bogus. Set the rc to -EFAULT,
2478                  * free anything we allocated and bail out.
2479                  */
2480                 if (!cur_len) {
2481                         for (i = 0; i < nr_pages; i++)
2482                                 put_page(wdata->pages[i]);
2483                         kfree(wdata);
2484                         rc = -EFAULT;
2485                         break;
2486                 }
2487
2488                 /*
2489                  * i + 1 now represents the number of pages we actually used in
2490                  * the copy phase above. Bring nr_pages down to that, and free
2491                  * any pages that we didn't use.
2492                  */
2493                 for ( ; nr_pages > i + 1; nr_pages--)
2494                         put_page(wdata->pages[nr_pages - 1]);
2495
2496                 wdata->sync_mode = WB_SYNC_ALL;
2497                 wdata->nr_pages = nr_pages;
2498                 wdata->offset = (__u64)offset;
2499                 wdata->cfile = cifsFileInfo_get(open_file);
2500                 wdata->pid = pid;
2501                 wdata->bytes = cur_len;
2502                 wdata->pagesz = PAGE_SIZE;
2503                 wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
2504                 rc = cifs_uncached_retry_writev(wdata);
2505                 if (rc) {
2506                         kref_put(&wdata->refcount,
2507                                  cifs_uncached_writedata_release);
2508                         break;
2509                 }
2510
2511                 list_add_tail(&wdata->list, &wdata_list);
2512                 offset += cur_len;
2513                 len -= cur_len;
2514         } while (len > 0);
2515
2516         /*
2517          * If at least one write was successfully sent, then discard any rc
2518          * value from the later writes. If the other write succeeds, then
2519          * we'll end up returning whatever was written. If it fails, then
2520          * we'll get a new rc value from that.
2521          */
2522         if (!list_empty(&wdata_list))
2523                 rc = 0;
2524
2525         /*
2526          * Wait for and collect replies for any successful sends in order of
2527          * increasing offset. Once an error is hit or we get a fatal signal
2528          * while waiting, then return without waiting for any more replies.
2529          */
2530 restart_loop:
2531         list_for_each_entry_safe(wdata, tmp, &wdata_list, list) {
2532                 if (!rc) {
2533                         /* FIXME: freezable too? */
2534                         rc = wait_for_completion_killable(&wdata->done);
2535                         if (rc)
2536                                 rc = -EINTR;
2537                         else if (wdata->result)
2538                                 rc = wdata->result;
2539                         else
2540                                 total_written += wdata->bytes;
2541
2542                         /* resend call if it's a retryable error */
2543                         if (rc == -EAGAIN) {
2544                                 rc = cifs_uncached_retry_writev(wdata);
2545                                 goto restart_loop;
2546                         }
2547                 }
2548                 list_del_init(&wdata->list);
2549                 kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2550         }
2551
2552         if (total_written > 0)
2553                 *poffset += total_written;
2554
2555         cifs_stats_bytes_written(tcon, total_written);
2556         return total_written ? total_written : (ssize_t)rc;
2557 }
2558
2559 ssize_t cifs_user_writev(struct kiocb *iocb, struct iov_iter *from)
2560 {
2561         ssize_t written;
2562         struct inode *inode;
2563         loff_t pos = iocb->ki_pos;
2564
2565         inode = file_inode(iocb->ki_filp);
2566
2567         /*
2568          * BB - optimize the way when signing is disabled. We can drop this
2569          * extra memory-to-memory copying and use iovec buffers for constructing
2570          * write request.
2571          */
2572
2573         written = cifs_iovec_write(iocb->ki_filp, from, &pos);
2574         if (written > 0) {
2575                 set_bit(CIFS_INO_INVALID_MAPPING, &CIFS_I(inode)->flags);
2576                 iocb->ki_pos = pos;
2577         }
2578
2579         return written;
2580 }
2581
2582 static ssize_t
2583 cifs_writev(struct kiocb *iocb, struct iov_iter *from)
2584 {
2585         struct file *file = iocb->ki_filp;
2586         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
2587         struct inode *inode = file->f_mapping->host;
2588         struct cifsInodeInfo *cinode = CIFS_I(inode);
2589         struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
2590         ssize_t rc = -EACCES;
2591         loff_t lock_pos = iocb->ki_pos;
2592
2593         /*
2594          * We need to hold the sem to be sure nobody modifies lock list
2595          * with a brlock that prevents writing.
2596          */
2597         down_read(&cinode->lock_sem);
2598         mutex_lock(&inode->i_mutex);
2599         if (file->f_flags & O_APPEND)
2600                 lock_pos = i_size_read(inode);
2601         if (!cifs_find_lock_conflict(cfile, lock_pos, iov_iter_count(from),
2602                                      server->vals->exclusive_lock_type, NULL,
2603                                      CIFS_WRITE_OP)) {
2604                 rc = __generic_file_write_iter(iocb, from);
2605                 mutex_unlock(&inode->i_mutex);
2606
2607                 if (rc > 0) {
2608                         ssize_t err;
2609
2610                         err = generic_write_sync(file, iocb->ki_pos - rc, rc);
2611                         if (err < 0)
2612                                 rc = err;
2613                 }
2614         } else {
2615                 mutex_unlock(&inode->i_mutex);
2616         }
2617         up_read(&cinode->lock_sem);
2618         return rc;
2619 }
2620
2621 ssize_t
2622 cifs_strict_writev(struct kiocb *iocb, struct iov_iter *from)
2623 {
2624         struct inode *inode = file_inode(iocb->ki_filp);
2625         struct cifsInodeInfo *cinode = CIFS_I(inode);
2626         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2627         struct cifsFileInfo *cfile = (struct cifsFileInfo *)
2628                                                 iocb->ki_filp->private_data;
2629         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
2630         ssize_t written;
2631
2632         written = cifs_get_writer(cinode);
2633         if (written)
2634                 return written;
2635
2636         if (CIFS_CACHE_WRITE(cinode)) {
2637                 if (cap_unix(tcon->ses) &&
2638                 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability))
2639                   && ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0)) {
2640                         written = generic_file_write_iter(iocb, from);
2641                         goto out;
2642                 }
2643                 written = cifs_writev(iocb, from);
2644                 goto out;
2645         }
2646         /*
2647          * For non-oplocked files in strict cache mode we need to write the data
2648          * to the server exactly from the pos to pos+len-1 rather than flush all
2649          * affected pages because it may cause a error with mandatory locks on
2650          * these pages but not on the region from pos to ppos+len-1.
2651          */
2652         written = cifs_user_writev(iocb, from);
2653         if (written > 0 && CIFS_CACHE_READ(cinode)) {
2654                 /*
2655                  * Windows 7 server can delay breaking level2 oplock if a write
2656                  * request comes - break it on the client to prevent reading
2657                  * an old data.
2658                  */
2659                 cifs_zap_mapping(inode);
2660                 cifs_dbg(FYI, "Set no oplock for inode=%p after a write operation\n",
2661                          inode);
2662                 cinode->oplock = 0;
2663         }
2664 out:
2665         cifs_put_writer(cinode);
2666         return written;
2667 }
2668
2669 static struct cifs_readdata *
2670 cifs_readdata_alloc(unsigned int nr_pages, work_func_t complete)
2671 {
2672         struct cifs_readdata *rdata;
2673
2674         rdata = kzalloc(sizeof(*rdata) + (sizeof(struct page *) * nr_pages),
2675                         GFP_KERNEL);
2676         if (rdata != NULL) {
2677                 kref_init(&rdata->refcount);
2678                 INIT_LIST_HEAD(&rdata->list);
2679                 init_completion(&rdata->done);
2680                 INIT_WORK(&rdata->work, complete);
2681         }
2682
2683         return rdata;
2684 }
2685
2686 void
2687 cifs_readdata_release(struct kref *refcount)
2688 {
2689         struct cifs_readdata *rdata = container_of(refcount,
2690                                         struct cifs_readdata, refcount);
2691
2692         if (rdata->cfile)
2693                 cifsFileInfo_put(rdata->cfile);
2694
2695         kfree(rdata);
2696 }
2697
2698 static int
2699 cifs_read_allocate_pages(struct cifs_readdata *rdata, unsigned int nr_pages)
2700 {
2701         int rc = 0;
2702         struct page *page;
2703         unsigned int i;
2704
2705         for (i = 0; i < nr_pages; i++) {
2706                 page = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2707                 if (!page) {
2708                         rc = -ENOMEM;
2709                         break;
2710                 }
2711                 rdata->pages[i] = page;
2712         }
2713
2714         if (rc) {
2715                 for (i = 0; i < nr_pages; i++) {
2716                         put_page(rdata->pages[i]);
2717                         rdata->pages[i] = NULL;
2718                 }
2719         }
2720         return rc;
2721 }
2722
2723 static void
2724 cifs_uncached_readdata_release(struct kref *refcount)
2725 {
2726         struct cifs_readdata *rdata = container_of(refcount,
2727                                         struct cifs_readdata, refcount);
2728         unsigned int i;
2729
2730         for (i = 0; i < rdata->nr_pages; i++) {
2731                 put_page(rdata->pages[i]);
2732                 rdata->pages[i] = NULL;
2733         }
2734         cifs_readdata_release(refcount);
2735 }
2736
2737 static int
2738 cifs_retry_async_readv(struct cifs_readdata *rdata)
2739 {
2740         int rc;
2741         struct TCP_Server_Info *server;
2742
2743         server = tlink_tcon(rdata->cfile->tlink)->ses->server;
2744
2745         do {
2746                 if (rdata->cfile->invalidHandle) {
2747                         rc = cifs_reopen_file(rdata->cfile, true);
2748                         if (rc != 0)
2749                                 continue;
2750                 }
2751                 rc = server->ops->async_readv(rdata);
2752         } while (rc == -EAGAIN);
2753
2754         return rc;
2755 }
2756
2757 /**
2758  * cifs_readdata_to_iov - copy data from pages in response to an iovec
2759  * @rdata:      the readdata response with list of pages holding data
2760  * @iter:       destination for our data
2761  *
2762  * This function copies data from a list of pages in a readdata response into
2763  * an array of iovecs. It will first calculate where the data should go
2764  * based on the info in the readdata and then copy the data into that spot.
2765  */
2766 static int
2767 cifs_readdata_to_iov(struct cifs_readdata *rdata, struct iov_iter *iter)
2768 {
2769         size_t remaining = rdata->bytes;
2770         unsigned int i;
2771
2772         for (i = 0; i < rdata->nr_pages; i++) {
2773                 struct page *page = rdata->pages[i];
2774                 size_t copy = min_t(size_t, remaining, PAGE_SIZE);
2775                 size_t written = copy_page_to_iter(page, 0, copy, iter);
2776                 remaining -= written;
2777                 if (written < copy && iov_iter_count(iter) > 0)
2778                         break;
2779         }
2780         return remaining ? -EFAULT : 0;
2781 }
2782
2783 static void
2784 cifs_uncached_readv_complete(struct work_struct *work)
2785 {
2786         struct cifs_readdata *rdata = container_of(work,
2787                                                 struct cifs_readdata, work);
2788
2789         complete(&rdata->done);
2790         kref_put(&rdata->refcount, cifs_uncached_readdata_release);
2791 }
2792
2793 static int
2794 cifs_uncached_read_into_pages(struct TCP_Server_Info *server,
2795                         struct cifs_readdata *rdata, unsigned int len)
2796 {
2797         int total_read = 0, result = 0;
2798         unsigned int i;
2799         unsigned int nr_pages = rdata->nr_pages;
2800         struct kvec iov;
2801
2802         rdata->tailsz = PAGE_SIZE;
2803         for (i = 0; i < nr_pages; i++) {
2804                 struct page *page = rdata->pages[i];
2805
2806                 if (len >= PAGE_SIZE) {
2807                         /* enough data to fill the page */
2808                         iov.iov_base = kmap(page);
2809                         iov.iov_len = PAGE_SIZE;
2810                         cifs_dbg(FYI, "%u: iov_base=%p iov_len=%zu\n",
2811                                  i, iov.iov_base, iov.iov_len);
2812                         len -= PAGE_SIZE;
2813                 } else if (len > 0) {
2814                         /* enough for partial page, fill and zero the rest */
2815                         iov.iov_base = kmap(page);
2816                         iov.iov_len = len;
2817                         cifs_dbg(FYI, "%u: iov_base=%p iov_len=%zu\n",
2818                                  i, iov.iov_base, iov.iov_len);
2819                         memset(iov.iov_base + len, '\0', PAGE_SIZE - len);
2820                         rdata->tailsz = len;
2821                         len = 0;
2822                 } else {
2823                         /* no need to hold page hostage */
2824                         rdata->pages[i] = NULL;
2825                         rdata->nr_pages--;
2826                         put_page(page);
2827                         continue;
2828                 }
2829
2830                 result = cifs_readv_from_socket(server, &iov, 1, iov.iov_len);
2831                 kunmap(page);
2832                 if (result < 0)
2833                         break;
2834
2835                 total_read += result;
2836         }
2837
2838         return total_read > 0 && result != -EAGAIN ? total_read : result;
2839 }
2840
2841 ssize_t cifs_user_readv(struct kiocb *iocb, struct iov_iter *to)
2842 {
2843         struct file *file = iocb->ki_filp;
2844         ssize_t rc;
2845         size_t len, cur_len;
2846         ssize_t total_read = 0;
2847         loff_t offset = iocb->ki_pos;
2848         unsigned int npages;
2849         struct cifs_sb_info *cifs_sb;
2850         struct cifs_tcon *tcon;
2851         struct cifsFileInfo *open_file;
2852         struct cifs_readdata *rdata, *tmp;
2853         struct list_head rdata_list;
2854         pid_t pid;
2855
2856         len = iov_iter_count(to);
2857         if (!len)
2858                 return 0;
2859
2860         INIT_LIST_HEAD(&rdata_list);
2861         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
2862         open_file = file->private_data;
2863         tcon = tlink_tcon(open_file->tlink);
2864
2865         if (!tcon->ses->server->ops->async_readv)
2866                 return -ENOSYS;
2867
2868         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2869                 pid = open_file->pid;
2870         else
2871                 pid = current->tgid;
2872
2873         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
2874                 cifs_dbg(FYI, "attempting read on write only file instance\n");
2875
2876         do {
2877                 cur_len = min_t(const size_t, len - total_read, cifs_sb->rsize);
2878                 npages = DIV_ROUND_UP(cur_len, PAGE_SIZE);
2879
2880                 /* allocate a readdata struct */
2881                 rdata = cifs_readdata_alloc(npages,
2882                                             cifs_uncached_readv_complete);
2883                 if (!rdata) {
2884                         rc = -ENOMEM;
2885                         break;
2886                 }
2887
2888                 rc = cifs_read_allocate_pages(rdata, npages);
2889                 if (rc)
2890                         goto error;
2891
2892                 rdata->cfile = cifsFileInfo_get(open_file);
2893                 rdata->nr_pages = npages;
2894                 rdata->offset = offset;
2895                 rdata->bytes = cur_len;
2896                 rdata->pid = pid;
2897                 rdata->pagesz = PAGE_SIZE;
2898                 rdata->read_into_pages = cifs_uncached_read_into_pages;
2899
2900                 rc = cifs_retry_async_readv(rdata);
2901 error:
2902                 if (rc) {
2903                         kref_put(&rdata->refcount,
2904                                  cifs_uncached_readdata_release);
2905                         break;
2906                 }
2907
2908                 list_add_tail(&rdata->list, &rdata_list);
2909                 offset += cur_len;
2910                 len -= cur_len;
2911         } while (len > 0);
2912
2913         /* if at least one read request send succeeded, then reset rc */
2914         if (!list_empty(&rdata_list))
2915                 rc = 0;
2916
2917         len = iov_iter_count(to);
2918         /* the loop below should proceed in the order of increasing offsets */
2919         list_for_each_entry_safe(rdata, tmp, &rdata_list, list) {
2920         again:
2921                 if (!rc) {
2922                         /* FIXME: freezable sleep too? */
2923                         rc = wait_for_completion_killable(&rdata->done);
2924                         if (rc)
2925                                 rc = -EINTR;
2926                         else if (rdata->result) {
2927                                 rc = rdata->result;
2928                                 /* resend call if it's a retryable error */
2929                                 if (rc == -EAGAIN) {
2930                                         rc = cifs_retry_async_readv(rdata);
2931                                         goto again;
2932                                 }
2933                         } else {
2934                                 rc = cifs_readdata_to_iov(rdata, to);
2935                         }
2936
2937                 }
2938                 list_del_init(&rdata->list);
2939                 kref_put(&rdata->refcount, cifs_uncached_readdata_release);
2940         }
2941
2942         total_read = len - iov_iter_count(to);
2943
2944         cifs_stats_bytes_read(tcon, total_read);
2945
2946         /* mask nodata case */
2947         if (rc == -ENODATA)
2948                 rc = 0;
2949
2950         if (total_read) {
2951                 iocb->ki_pos += total_read;
2952                 return total_read;
2953         }
2954         return rc;
2955 }
2956
2957 ssize_t
2958 cifs_strict_readv(struct kiocb *iocb, struct iov_iter *to)
2959 {
2960         struct inode *inode = file_inode(iocb->ki_filp);
2961         struct cifsInodeInfo *cinode = CIFS_I(inode);
2962         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2963         struct cifsFileInfo *cfile = (struct cifsFileInfo *)
2964                                                 iocb->ki_filp->private_data;
2965         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
2966         int rc = -EACCES;
2967
2968         /*
2969          * In strict cache mode we need to read from the server all the time
2970          * if we don't have level II oplock because the server can delay mtime
2971          * change - so we can't make a decision about inode invalidating.
2972          * And we can also fail with pagereading if there are mandatory locks
2973          * on pages affected by this read but not on the region from pos to
2974          * pos+len-1.
2975          */
2976         if (!CIFS_CACHE_READ(cinode))
2977                 return cifs_user_readv(iocb, to);
2978
2979         if (cap_unix(tcon->ses) &&
2980             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
2981             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
2982                 return generic_file_read_iter(iocb, to);
2983
2984         /*
2985          * We need to hold the sem to be sure nobody modifies lock list
2986          * with a brlock that prevents reading.
2987          */
2988         down_read(&cinode->lock_sem);
2989         if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(to),
2990                                      tcon->ses->server->vals->shared_lock_type,
2991                                      NULL, CIFS_READ_OP))
2992                 rc = generic_file_read_iter(iocb, to);
2993         up_read(&cinode->lock_sem);
2994         return rc;
2995 }
2996
2997 static ssize_t
2998 cifs_read(struct file *file, char *read_data, size_t read_size, loff_t *offset)
2999 {
3000         int rc = -EACCES;
3001         unsigned int bytes_read = 0;
3002         unsigned int total_read;
3003         unsigned int current_read_size;
3004         unsigned int rsize;
3005         struct cifs_sb_info *cifs_sb;
3006         struct cifs_tcon *tcon;
3007         struct TCP_Server_Info *server;
3008         unsigned int xid;
3009         char *cur_offset;
3010         struct cifsFileInfo *open_file;
3011         struct cifs_io_parms io_parms;
3012         int buf_type = CIFS_NO_BUFFER;
3013         __u32 pid;
3014
3015         xid = get_xid();
3016         cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
3017
3018         /* FIXME: set up handlers for larger reads and/or convert to async */
3019         rsize = min_t(unsigned int, cifs_sb->rsize, CIFSMaxBufSize);
3020
3021         if (file->private_data == NULL) {
3022                 rc = -EBADF;
3023                 free_xid(xid);
3024                 return rc;
3025         }
3026         open_file = file->private_data;
3027         tcon = tlink_tcon(open_file->tlink);
3028         server = tcon->ses->server;
3029
3030         if (!server->ops->sync_read) {
3031                 free_xid(xid);
3032                 return -ENOSYS;
3033         }
3034
3035         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3036                 pid = open_file->pid;
3037         else
3038                 pid = current->tgid;
3039
3040         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3041                 cifs_dbg(FYI, "attempting read on write only file instance\n");
3042
3043         for (total_read = 0, cur_offset = read_data; read_size > total_read;
3044              total_read += bytes_read, cur_offset += bytes_read) {
3045                 current_read_size = min_t(uint, read_size - total_read, rsize);
3046                 /*
3047                  * For windows me and 9x we do not want to request more than it
3048                  * negotiated since it will refuse the read then.
3049                  */
3050                 if ((tcon->ses) && !(tcon->ses->capabilities &
3051                                 tcon->ses->server->vals->cap_large_files)) {
3052                         current_read_size = min_t(uint, current_read_size,
3053                                         CIFSMaxBufSize);
3054                 }
3055                 rc = -EAGAIN;
3056                 while (rc == -EAGAIN) {
3057                         if (open_file->invalidHandle) {
3058                                 rc = cifs_reopen_file(open_file, true);
3059                                 if (rc != 0)
3060                                         break;
3061                         }
3062                         io_parms.pid = pid;
3063                         io_parms.tcon = tcon;
3064                         io_parms.offset = *offset;
3065                         io_parms.length = current_read_size;
3066                         rc = server->ops->sync_read(xid, open_file, &io_parms,
3067                                                     &bytes_read, &cur_offset,
3068                                                     &buf_type);
3069                 }
3070                 if (rc || (bytes_read == 0)) {
3071                         if (total_read) {
3072                                 break;
3073                         } else {
3074                                 free_xid(xid);
3075                                 return rc;
3076                         }
3077                 } else {
3078                         cifs_stats_bytes_read(tcon, total_read);
3079                         *offset += bytes_read;
3080                 }
3081         }
3082         free_xid(xid);
3083         return total_read;
3084 }
3085
3086 /*
3087  * If the page is mmap'ed into a process' page tables, then we need to make
3088  * sure that it doesn't change while being written back.
3089  */
3090 static int
3091 cifs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
3092 {
3093         struct page *page = vmf->page;
3094
3095         lock_page(page);
3096         return VM_FAULT_LOCKED;
3097 }
3098
3099 static struct vm_operations_struct cifs_file_vm_ops = {
3100         .fault = filemap_fault,
3101         .map_pages = filemap_map_pages,
3102         .page_mkwrite = cifs_page_mkwrite,
3103         .remap_pages = generic_file_remap_pages,
3104 };
3105
3106 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
3107 {
3108         int rc, xid;
3109         struct inode *inode = file_inode(file);
3110
3111         xid = get_xid();
3112
3113         if (!CIFS_CACHE_READ(CIFS_I(inode))) {
3114                 rc = cifs_zap_mapping(inode);
3115                 if (rc)
3116                         return rc;
3117         }
3118
3119         rc = generic_file_mmap(file, vma);
3120         if (rc == 0)
3121                 vma->vm_ops = &cifs_file_vm_ops;
3122         free_xid(xid);
3123         return rc;
3124 }
3125
3126 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
3127 {
3128         int rc, xid;
3129
3130         xid = get_xid();
3131         rc = cifs_revalidate_file(file);
3132         if (rc) {
3133                 cifs_dbg(FYI, "Validation prior to mmap failed, error=%d\n",
3134                          rc);
3135                 free_xid(xid);
3136                 return rc;
3137         }
3138         rc = generic_file_mmap(file, vma);
3139         if (rc == 0)
3140                 vma->vm_ops = &cifs_file_vm_ops;
3141         free_xid(xid);
3142         return rc;
3143 }
3144
3145 static void
3146 cifs_readv_complete(struct work_struct *work)
3147 {
3148         unsigned int i;
3149         struct cifs_readdata *rdata = container_of(work,
3150                                                 struct cifs_readdata, work);
3151
3152         for (i = 0; i < rdata->nr_pages; i++) {
3153                 struct page *page = rdata->pages[i];
3154
3155                 lru_cache_add_file(page);
3156
3157                 if (rdata->result == 0) {
3158                         flush_dcache_page(page);
3159                         SetPageUptodate(page);
3160                 }
3161
3162                 unlock_page(page);
3163
3164                 if (rdata->result == 0)
3165                         cifs_readpage_to_fscache(rdata->mapping->host, page);
3166
3167                 page_cache_release(page);
3168                 rdata->pages[i] = NULL;
3169         }
3170         kref_put(&rdata->refcount, cifs_readdata_release);
3171 }
3172
3173 static int
3174 cifs_readpages_read_into_pages(struct TCP_Server_Info *server,
3175                         struct cifs_readdata *rdata, unsigned int len)
3176 {
3177         int total_read = 0, result = 0;
3178         unsigned int i;
3179         u64 eof;
3180         pgoff_t eof_index;
3181         unsigned int nr_pages = rdata->nr_pages;
3182         struct kvec iov;
3183
3184         /* determine the eof that the server (probably) has */
3185         eof = CIFS_I(rdata->mapping->host)->server_eof;
3186         eof_index = eof ? (eof - 1) >> PAGE_CACHE_SHIFT : 0;
3187         cifs_dbg(FYI, "eof=%llu eof_index=%lu\n", eof, eof_index);
3188
3189         rdata->tailsz = PAGE_CACHE_SIZE;
3190         for (i = 0; i < nr_pages; i++) {
3191                 struct page *page = rdata->pages[i];
3192
3193                 if (len >= PAGE_CACHE_SIZE) {
3194                         /* enough data to fill the page */
3195                         iov.iov_base = kmap(page);
3196                         iov.iov_len = PAGE_CACHE_SIZE;
3197                         cifs_dbg(FYI, "%u: idx=%lu iov_base=%p iov_len=%zu\n",
3198                                  i, page->index, iov.iov_base, iov.iov_len);
3199                         len -= PAGE_CACHE_SIZE;
3200                 } else if (len > 0) {
3201                         /* enough for partial page, fill and zero the rest */
3202                         iov.iov_base = kmap(page);
3203                         iov.iov_len = len;
3204                         cifs_dbg(FYI, "%u: idx=%lu iov_base=%p iov_len=%zu\n",
3205                                  i, page->index, iov.iov_base, iov.iov_len);
3206                         memset(iov.iov_base + len,
3207                                 '\0', PAGE_CACHE_SIZE - len);
3208                         rdata->tailsz = len;
3209                         len = 0;
3210                 } else if (page->index > eof_index) {
3211                         /*
3212                          * The VFS will not try to do readahead past the
3213                          * i_size, but it's possible that we have outstanding
3214                          * writes with gaps in the middle and the i_size hasn't
3215                          * caught up yet. Populate those with zeroed out pages
3216                          * to prevent the VFS from repeatedly attempting to
3217                          * fill them until the writes are flushed.
3218                          */
3219                         zero_user(page, 0, PAGE_CACHE_SIZE);
3220                         lru_cache_add_file(page);
3221                         flush_dcache_page(page);
3222                         SetPageUptodate(page);
3223                         unlock_page(page);
3224                         page_cache_release(page);
3225                         rdata->pages[i] = NULL;
3226                         rdata->nr_pages--;
3227                         continue;
3228                 } else {
3229                         /* no need to hold page hostage */
3230                         lru_cache_add_file(page);
3231                         unlock_page(page);
3232                         page_cache_release(page);
3233                         rdata->pages[i] = NULL;
3234                         rdata->nr_pages--;
3235                         continue;
3236                 }
3237
3238                 result = cifs_readv_from_socket(server, &iov, 1, iov.iov_len);
3239                 kunmap(page);
3240                 if (result < 0)
3241                         break;
3242
3243                 total_read += result;
3244         }
3245
3246         return total_read > 0 && result != -EAGAIN ? total_read : result;
3247 }
3248
3249 static int cifs_readpages(struct file *file, struct address_space *mapping,
3250         struct list_head *page_list, unsigned num_pages)
3251 {
3252         int rc;
3253         struct list_head tmplist;
3254         struct cifsFileInfo *open_file = file->private_data;
3255         struct cifs_sb_info *cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
3256         unsigned int rsize = cifs_sb->rsize;
3257         pid_t pid;
3258
3259         /*
3260          * Give up immediately if rsize is too small to read an entire page.
3261          * The VFS will fall back to readpage. We should never reach this
3262          * point however since we set ra_pages to 0 when the rsize is smaller
3263          * than a cache page.
3264          */
3265         if (unlikely(rsize < PAGE_CACHE_SIZE))
3266                 return 0;
3267
3268         /*
3269          * Reads as many pages as possible from fscache. Returns -ENOBUFS
3270          * immediately if the cookie is negative
3271          *
3272          * After this point, every page in the list might have PG_fscache set,
3273          * so we will need to clean that up off of every page we don't use.
3274          */
3275         rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
3276                                          &num_pages);
3277         if (rc == 0)
3278                 return rc;
3279
3280         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3281                 pid = open_file->pid;
3282         else
3283                 pid = current->tgid;
3284
3285         rc = 0;
3286         INIT_LIST_HEAD(&tmplist);
3287
3288         cifs_dbg(FYI, "%s: file=%p mapping=%p num_pages=%u\n",
3289                  __func__, file, mapping, num_pages);
3290
3291         /*
3292          * Start with the page at end of list and move it to private
3293          * list. Do the same with any following pages until we hit
3294          * the rsize limit, hit an index discontinuity, or run out of
3295          * pages. Issue the async read and then start the loop again
3296          * until the list is empty.
3297          *
3298          * Note that list order is important. The page_list is in
3299          * the order of declining indexes. When we put the pages in
3300          * the rdata->pages, then we want them in increasing order.
3301          */
3302         while (!list_empty(page_list)) {
3303                 unsigned int i;
3304                 unsigned int bytes = PAGE_CACHE_SIZE;
3305                 unsigned int expected_index;
3306                 unsigned int nr_pages = 1;
3307                 loff_t offset;
3308                 struct page *page, *tpage;
3309                 struct cifs_readdata *rdata;
3310
3311                 page = list_entry(page_list->prev, struct page, lru);
3312
3313                 /*
3314                  * Lock the page and put it in the cache. Since no one else
3315                  * should have access to this page, we're safe to simply set
3316                  * PG_locked without checking it first.
3317                  */
3318                 __set_page_locked(page);
3319                 rc = add_to_page_cache_locked(page, mapping,
3320                                               page->index, GFP_KERNEL);
3321
3322                 /* give up if we can't stick it in the cache */
3323                 if (rc) {
3324                         __clear_page_locked(page);
3325                         break;
3326                 }
3327
3328                 /* move first page to the tmplist */
3329                 offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
3330                 list_move_tail(&page->lru, &tmplist);
3331
3332                 /* now try and add more pages onto the request */
3333                 expected_index = page->index + 1;
3334                 list_for_each_entry_safe_reverse(page, tpage, page_list, lru) {
3335                         /* discontinuity ? */
3336                         if (page->index != expected_index)
3337                                 break;
3338
3339                         /* would this page push the read over the rsize? */
3340                         if (bytes + PAGE_CACHE_SIZE > rsize)
3341                                 break;
3342
3343                         __set_page_locked(page);
3344                         if (add_to_page_cache_locked(page, mapping,
3345                                                 page->index, GFP_KERNEL)) {
3346                                 __clear_page_locked(page);
3347                                 break;
3348                         }
3349                         list_move_tail(&page->lru, &tmplist);
3350                         bytes += PAGE_CACHE_SIZE;
3351                         expected_index++;
3352                         nr_pages++;
3353                 }
3354
3355                 rdata = cifs_readdata_alloc(nr_pages, cifs_readv_complete);
3356                 if (!rdata) {
3357                         /* best to give up if we're out of mem */
3358                         list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3359                                 list_del(&page->lru);
3360                                 lru_cache_add_file(page);
3361                                 unlock_page(page);
3362                                 page_cache_release(page);
3363                         }
3364                         rc = -ENOMEM;
3365                         break;
3366                 }
3367
3368                 rdata->cfile = cifsFileInfo_get(open_file);
3369                 rdata->mapping = mapping;
3370                 rdata->offset = offset;
3371                 rdata->bytes = bytes;
3372                 rdata->pid = pid;
3373                 rdata->pagesz = PAGE_CACHE_SIZE;
3374                 rdata->read_into_pages = cifs_readpages_read_into_pages;
3375
3376                 list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3377                         list_del(&page->lru);
3378                         rdata->pages[rdata->nr_pages++] = page;
3379                 }
3380
3381                 rc = cifs_retry_async_readv(rdata);
3382                 if (rc != 0) {
3383                         for (i = 0; i < rdata->nr_pages; i++) {
3384                                 page = rdata->pages[i];
3385                                 lru_cache_add_file(page);
3386                                 unlock_page(page);
3387                                 page_cache_release(page);
3388                         }
3389                         kref_put(&rdata->refcount, cifs_readdata_release);
3390                         break;
3391                 }
3392
3393                 kref_put(&rdata->refcount, cifs_readdata_release);
3394         }
3395
3396         /* Any pages that have been shown to fscache but didn't get added to
3397          * the pagecache must be uncached before they get returned to the
3398          * allocator.
3399          */
3400         cifs_fscache_readpages_cancel(mapping->host, page_list);
3401         return rc;
3402 }
3403
3404 /*
3405  * cifs_readpage_worker must be called with the page pinned
3406  */
3407 static int cifs_readpage_worker(struct file *file, struct page *page,
3408         loff_t *poffset)
3409 {
3410         char *read_data;
3411         int rc;
3412
3413         /* Is the page cached? */
3414         rc = cifs_readpage_from_fscache(file_inode(file), page);
3415         if (rc == 0)
3416                 goto read_complete;
3417
3418         read_data = kmap(page);
3419         /* for reads over a certain size could initiate async read ahead */
3420
3421         rc = cifs_read(file, read_data, PAGE_CACHE_SIZE, poffset);
3422
3423         if (rc < 0)
3424                 goto io_error;
3425         else
3426                 cifs_dbg(FYI, "Bytes read %d\n", rc);
3427
3428         file_inode(file)->i_atime =
3429                 current_fs_time(file_inode(file)->i_sb);
3430
3431         if (PAGE_CACHE_SIZE > rc)
3432                 memset(read_data + rc, 0, PAGE_CACHE_SIZE - rc);
3433
3434         flush_dcache_page(page);
3435         SetPageUptodate(page);
3436
3437         /* send this page to the cache */
3438         cifs_readpage_to_fscache(file_inode(file), page);
3439
3440         rc = 0;
3441
3442 io_error:
3443         kunmap(page);
3444         unlock_page(page);
3445
3446 read_complete:
3447         return rc;
3448 }
3449
3450 static int cifs_readpage(struct file *file, struct page *page)
3451 {
3452         loff_t offset = (loff_t)page->index << PAGE_CACHE_SHIFT;
3453         int rc = -EACCES;
3454         unsigned int xid;
3455
3456         xid = get_xid();
3457
3458         if (file->private_data == NULL) {
3459                 rc = -EBADF;
3460                 free_xid(xid);
3461                 return rc;
3462         }
3463
3464         cifs_dbg(FYI, "readpage %p at offset %d 0x%x\n",
3465                  page, (int)offset, (int)offset);
3466
3467         rc = cifs_readpage_worker(file, page, &offset);
3468
3469         free_xid(xid);
3470         return rc;
3471 }
3472
3473 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
3474 {
3475         struct cifsFileInfo *open_file;
3476
3477         spin_lock(&cifs_file_list_lock);
3478         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
3479                 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
3480                         spin_unlock(&cifs_file_list_lock);
3481                         return 1;
3482                 }
3483         }
3484         spin_unlock(&cifs_file_list_lock);
3485         return 0;
3486 }
3487
3488 /* We do not want to update the file size from server for inodes
3489    open for write - to avoid races with writepage extending
3490    the file - in the future we could consider allowing
3491    refreshing the inode only on increases in the file size
3492    but this is tricky to do without racing with writebehind
3493    page caching in the current Linux kernel design */
3494 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
3495 {
3496         if (!cifsInode)
3497                 return true;
3498
3499         if (is_inode_writable(cifsInode)) {
3500                 /* This inode is open for write at least once */
3501                 struct cifs_sb_info *cifs_sb;
3502
3503                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
3504                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
3505                         /* since no page cache to corrupt on directio
3506                         we can change size safely */
3507                         return true;
3508                 }
3509
3510                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
3511                         return true;
3512
3513                 return false;
3514         } else
3515                 return true;
3516 }
3517
3518 static int cifs_write_begin(struct file *file, struct address_space *mapping,
3519                         loff_t pos, unsigned len, unsigned flags,
3520                         struct page **pagep, void **fsdata)
3521 {
3522         int oncethru = 0;
3523         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
3524         loff_t offset = pos & (PAGE_CACHE_SIZE - 1);
3525         loff_t page_start = pos & PAGE_MASK;
3526         loff_t i_size;
3527         struct page *page;
3528         int rc = 0;
3529
3530         cifs_dbg(FYI, "write_begin from %lld len %d\n", (long long)pos, len);
3531
3532 start:
3533         page = grab_cache_page_write_begin(mapping, index, flags);
3534         if (!page) {
3535                 rc = -ENOMEM;
3536                 goto out;
3537         }
3538
3539         if (PageUptodate(page))
3540                 goto out;
3541
3542         /*
3543          * If we write a full page it will be up to date, no need to read from
3544          * the server. If the write is short, we'll end up doing a sync write
3545          * instead.
3546          */
3547         if (len == PAGE_CACHE_SIZE)
3548                 goto out;
3549
3550         /*
3551          * optimize away the read when we have an oplock, and we're not
3552          * expecting to use any of the data we'd be reading in. That
3553          * is, when the page lies beyond the EOF, or straddles the EOF
3554          * and the write will cover all of the existing data.
3555          */
3556         if (CIFS_CACHE_READ(CIFS_I(mapping->host))) {
3557                 i_size = i_size_read(mapping->host);
3558                 if (page_start >= i_size ||
3559                     (offset == 0 && (pos + len) >= i_size)) {
3560                         zero_user_segments(page, 0, offset,
3561                                            offset + len,
3562                                            PAGE_CACHE_SIZE);
3563                         /*
3564                          * PageChecked means that the parts of the page
3565                          * to which we're not writing are considered up
3566                          * to date. Once the data is copied to the
3567                          * page, it can be set uptodate.
3568                          */
3569                         SetPageChecked(page);
3570                         goto out;
3571                 }
3572         }
3573
3574         if ((file->f_flags & O_ACCMODE) != O_WRONLY && !oncethru) {
3575                 /*
3576                  * might as well read a page, it is fast enough. If we get
3577                  * an error, we don't need to return it. cifs_write_end will
3578                  * do a sync write instead since PG_uptodate isn't set.
3579                  */
3580                 cifs_readpage_worker(file, page, &page_start);
3581                 page_cache_release(page);
3582                 oncethru = 1;
3583                 goto start;
3584         } else {
3585                 /* we could try using another file handle if there is one -
3586                    but how would we lock it to prevent close of that handle
3587                    racing with this read? In any case
3588                    this will be written out by write_end so is fine */
3589         }
3590 out:
3591         *pagep = page;
3592         return rc;
3593 }
3594
3595 static int cifs_release_page(struct page *page, gfp_t gfp)
3596 {
3597         if (PagePrivate(page))
3598                 return 0;
3599
3600         return cifs_fscache_release_page(page, gfp);
3601 }
3602
3603 static void cifs_invalidate_page(struct page *page, unsigned int offset,
3604                                  unsigned int length)
3605 {
3606         struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
3607
3608         if (offset == 0 && length == PAGE_CACHE_SIZE)
3609                 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
3610 }
3611
3612 static int cifs_launder_page(struct page *page)
3613 {
3614         int rc = 0;
3615         loff_t range_start = page_offset(page);
3616         loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
3617         struct writeback_control wbc = {
3618                 .sync_mode = WB_SYNC_ALL,
3619                 .nr_to_write = 0,
3620                 .range_start = range_start,
3621                 .range_end = range_end,
3622         };
3623
3624         cifs_dbg(FYI, "Launder page: %p\n", page);
3625
3626         if (clear_page_dirty_for_io(page))
3627                 rc = cifs_writepage_locked(page, &wbc);
3628
3629         cifs_fscache_invalidate_page(page, page->mapping->host);
3630         return rc;
3631 }
3632
3633 static int
3634 cifs_pending_writers_wait(void *unused)
3635 {
3636         schedule();
3637         return 0;
3638 }
3639
3640 void cifs_oplock_break(struct work_struct *work)
3641 {
3642         struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
3643                                                   oplock_break);
3644         struct inode *inode = cfile->dentry->d_inode;
3645         struct cifsInodeInfo *cinode = CIFS_I(inode);
3646         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
3647         struct TCP_Server_Info *server = tcon->ses->server;
3648         int rc = 0;
3649
3650         wait_on_bit(&cinode->flags, CIFS_INODE_PENDING_WRITERS,
3651                         cifs_pending_writers_wait, TASK_UNINTERRUPTIBLE);
3652
3653         server->ops->downgrade_oplock(server, cinode,
3654                 test_bit(CIFS_INODE_DOWNGRADE_OPLOCK_TO_L2, &cinode->flags));
3655
3656         if (!CIFS_CACHE_WRITE(cinode) && CIFS_CACHE_READ(cinode) &&
3657                                                 cifs_has_mand_locks(cinode)) {
3658                 cifs_dbg(FYI, "Reset oplock to None for inode=%p due to mand locks\n",
3659                          inode);
3660                 cinode->oplock = 0;
3661         }
3662
3663         if (inode && S_ISREG(inode->i_mode)) {
3664                 if (CIFS_CACHE_READ(cinode))
3665                         break_lease(inode, O_RDONLY);
3666                 else
3667                         break_lease(inode, O_WRONLY);
3668                 rc = filemap_fdatawrite(inode->i_mapping);
3669                 if (!CIFS_CACHE_READ(cinode)) {
3670                         rc = filemap_fdatawait(inode->i_mapping);
3671                         mapping_set_error(inode->i_mapping, rc);
3672                         cifs_zap_mapping(inode);
3673                 }
3674                 cifs_dbg(FYI, "Oplock flush inode %p rc %d\n", inode, rc);
3675         }
3676
3677         rc = cifs_push_locks(cfile);
3678         if (rc)
3679                 cifs_dbg(VFS, "Push locks rc = %d\n", rc);
3680
3681         /*
3682          * releasing stale oplock after recent reconnect of smb session using
3683          * a now incorrect file handle is not a data integrity issue but do
3684          * not bother sending an oplock release if session to server still is
3685          * disconnected since oplock already released by the server
3686          */
3687         if (!cfile->oplock_break_cancelled) {
3688                 rc = tcon->ses->server->ops->oplock_response(tcon, &cfile->fid,
3689                                                              cinode);
3690                 cifs_dbg(FYI, "Oplock release rc = %d\n", rc);
3691         }
3692         cifs_done_oplock_break(cinode);
3693 }
3694
3695 /*
3696  * The presence of cifs_direct_io() in the address space ops vector
3697  * allowes open() O_DIRECT flags which would have failed otherwise.
3698  *
3699  * In the non-cached mode (mount with cache=none), we shunt off direct read and write requests
3700  * so this method should never be called.
3701  *
3702  * Direct IO is not yet supported in the cached mode.
3703  */
3704 static ssize_t
3705 cifs_direct_io(int rw, struct kiocb *iocb, struct iov_iter *iter,
3706                loff_t pos)
3707 {
3708         /*
3709          * FIXME
3710          * Eventually need to support direct IO for non forcedirectio mounts
3711          */
3712         return -EINVAL;
3713 }
3714
3715
3716 const struct address_space_operations cifs_addr_ops = {
3717         .readpage = cifs_readpage,
3718         .readpages = cifs_readpages,
3719         .writepage = cifs_writepage,
3720         .writepages = cifs_writepages,
3721         .write_begin = cifs_write_begin,
3722         .write_end = cifs_write_end,
3723         .set_page_dirty = __set_page_dirty_nobuffers,
3724         .releasepage = cifs_release_page,
3725         .direct_IO = cifs_direct_io,
3726         .invalidatepage = cifs_invalidate_page,
3727         .launder_page = cifs_launder_page,
3728 };
3729
3730 /*
3731  * cifs_readpages requires the server to support a buffer large enough to
3732  * contain the header plus one complete page of data.  Otherwise, we need
3733  * to leave cifs_readpages out of the address space operations.
3734  */
3735 const struct address_space_operations cifs_addr_ops_smallbuf = {
3736         .readpage = cifs_readpage,
3737         .writepage = cifs_writepage,
3738         .writepages = cifs_writepages,
3739         .write_begin = cifs_write_begin,
3740         .write_end = cifs_write_end,
3741         .set_page_dirty = __set_page_dirty_nobuffers,
3742         .releasepage = cifs_release_page,
3743         .invalidatepage = cifs_invalidate_page,
3744         .launder_page = cifs_launder_page,
3745 };