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 #include "smbdirect.h"
  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_remap(cifs_sb));
 144         cifs_put_tlink(tlink);
 145
 146         if (rc)
 147                 goto posix_open_ret;
 148
 149         if (presp_data->Type == cpu_to_le32(-1))
 150                 goto posix_open_ret; /* open ok, caller does qpathinfo */
 151
 152         if (!pinode)
 153                 goto posix_open_ret; /* caller does not need info */
 154
 155         cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
 156
 157         /* get new inode and set it up */
 158         if (*pinode == NULL) {
 159                 cifs_fill_uniqueid(sb, &fattr);
 160                 *pinode = cifs_iget(sb, &fattr);
 161                 if (!*pinode) {
 162                         rc = -ENOMEM;
 163                         goto posix_open_ret;
 164                 }
 165         } else {
 166                 cifs_fattr_to_inode(*pinode, &fattr);
 167         }
 168
 169 posix_open_ret:
 170         kfree(presp_data);
 171         return rc;
 172 }
 173
 174 static int
 175 cifs_nt_open(char *full_path, struct inode *inode, struct cifs_sb_info *cifs_sb,
 176              struct cifs_tcon *tcon, unsigned int f_flags, __u32 *oplock,
 177              struct cifs_fid *fid, unsigned int xid)
 178 {
 179         int rc;
 180         int desired_access;
 181         int disposition;
 182         int create_options = CREATE_NOT_DIR;
 183         FILE_ALL_INFO *buf;
 184         struct TCP_Server_Info *server = tcon->ses->server;
 185         struct cifs_open_parms oparms;
 186
 187         if (!server->ops->open)
 188                 return -ENOSYS;
 189
 190         desired_access = cifs_convert_flags(f_flags);
 191
 192 /*********************************************************************
 193  *  open flag mapping table:
 194  *
 195  *      POSIX Flag            CIFS Disposition
 196  *      ----------            ----------------
 197  *      O_CREAT               FILE_OPEN_IF
 198  *      O_CREAT | O_EXCL      FILE_CREATE
 199  *      O_CREAT | O_TRUNC     FILE_OVERWRITE_IF
 200  *      O_TRUNC               FILE_OVERWRITE
 201  *      none of the above     FILE_OPEN
 202  *
 203  *      Note that there is not a direct match between disposition
 204  *      FILE_SUPERSEDE (ie create whether or not file exists although
 205  *      O_CREAT | O_TRUNC is similar but truncates the existing
 206  *      file rather than creating a new file as FILE_SUPERSEDE does
 207  *      (which uses the attributes / metadata passed in on open call)
 208  *?
 209  *?  O_SYNC is a reasonable match to CIFS writethrough flag
 210  *?  and the read write flags match reasonably.  O_LARGEFILE
 211  *?  is irrelevant because largefile support is always used
 212  *?  by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
 213  *       O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
 214  *********************************************************************/
 215
 216         disposition = cifs_get_disposition(f_flags);
 217
 218         /* BB pass O_SYNC flag through on file attributes .. BB */
 219
 220         buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
 221         if (!buf)
 222                 return -ENOMEM;
 223
 224         if (backup_cred(cifs_sb))
 225                 create_options |= CREATE_OPEN_BACKUP_INTENT;
 226
 227         /* O_SYNC also has bit for O_DSYNC so following check picks up either */
 228         if (f_flags & O_SYNC)
 229                 create_options |= CREATE_WRITE_THROUGH;
 230
 231         if (f_flags & O_DIRECT)
 232                 create_options |= CREATE_NO_BUFFER;
 233
 234         oparms.tcon = tcon;
 235         oparms.cifs_sb = cifs_sb;
 236         oparms.desired_access = desired_access;
 237         oparms.create_options = create_options;
 238         oparms.disposition = disposition;
 239         oparms.path = full_path;
 240         oparms.fid = fid;
 241         oparms.reconnect = false;
 242
 243         rc = server->ops->open(xid, &oparms, oplock, buf);
 244
 245         if (rc)
 246                 goto out;
 247
 248         if (tcon->unix_ext)
 249                 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
 250                                               xid);
 251         else
 252                 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
 253                                          xid, fid);
 254
 255 out:
 256         kfree(buf);
 257         return rc;
 258 }
 259
 260 static bool
 261 cifs_has_mand_locks(struct cifsInodeInfo *cinode)
 262 {
 263         struct cifs_fid_locks *cur;
 264         bool has_locks = false;
 265
 266         down_read(&cinode->lock_sem);
 267         list_for_each_entry(cur, &cinode->llist, llist) {
 268                 if (!list_empty(&cur->locks)) {
 269                         has_locks = true;
 270                         break;
 271                 }
 272         }
 273         up_read(&cinode->lock_sem);
 274         return has_locks;
 275 }
 276
 277 struct cifsFileInfo *
 278 cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
 279                   struct tcon_link *tlink, __u32 oplock)
 280 {
 281         struct dentry *dentry = file_dentry(file);
 282         struct inode *inode = d_inode(dentry);
 283         struct cifsInodeInfo *cinode = CIFS_I(inode);
 284         struct cifsFileInfo *cfile;
 285         struct cifs_fid_locks *fdlocks;
 286         struct cifs_tcon *tcon = tlink_tcon(tlink);
 287         struct TCP_Server_Info *server = tcon->ses->server;
 288
 289         cfile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
 290         if (cfile == NULL)
 291                 return cfile;
 292
 293         fdlocks = kzalloc(sizeof(struct cifs_fid_locks), GFP_KERNEL);
 294         if (!fdlocks) {
 295                 kfree(cfile);
 296                 return NULL;
 297         }
 298
 299         INIT_LIST_HEAD(&fdlocks->locks);
 300         fdlocks->cfile = cfile;
 301         cfile->llist = fdlocks;
 302         down_write(&cinode->lock_sem);
 303         list_add(&fdlocks->llist, &cinode->llist);
 304         up_write(&cinode->lock_sem);
 305
 306         cfile->count = 1;
 307         cfile->pid = current->tgid;
 308         cfile->uid = current_fsuid();
 309         cfile->dentry = dget(dentry);
 310         cfile->f_flags = file->f_flags;
 311         cfile->invalidHandle = false;
 312         cfile->tlink = cifs_get_tlink(tlink);
 313         INIT_WORK(&cfile->oplock_break, cifs_oplock_break);
 314         mutex_init(&cfile->fh_mutex);
 315         spin_lock_init(&cfile->file_info_lock);
 316
 317         cifs_sb_active(inode->i_sb);
 318
 319         /*
 320          * If the server returned a read oplock and we have mandatory brlocks,
 321          * set oplock level to None.
 322          */
 323         if (server->ops->is_read_op(oplock) && cifs_has_mand_locks(cinode)) {
 324                 cifs_dbg(FYI, "Reset oplock val from read to None due to mand locks\n");
 325                 oplock = 0;
 326         }
 327
 328         spin_lock(&tcon->open_file_lock);
 329         if (fid->pending_open->oplock != CIFS_OPLOCK_NO_CHANGE && oplock)
 330                 oplock = fid->pending_open->oplock;
 331         list_del(&fid->pending_open->olist);
 332
 333         fid->purge_cache = false;
 334         server->ops->set_fid(cfile, fid, oplock);
 335
 336         list_add(&cfile->tlist, &tcon->openFileList);
 337
 338         /* if readable file instance put first in list*/
 339         if (file->f_mode & FMODE_READ)
 340                 list_add(&cfile->flist, &cinode->openFileList);
 341         else
 342                 list_add_tail(&cfile->flist, &cinode->openFileList);
 343         spin_unlock(&tcon->open_file_lock);
 344
 345         if (fid->purge_cache)
 346                 cifs_zap_mapping(inode);
 347
 348         file->private_data = cfile;
 349         return cfile;
 350 }
 351
 352 struct cifsFileInfo *
 353 cifsFileInfo_get(struct cifsFileInfo *cifs_file)
 354 {
 355         spin_lock(&cifs_file->file_info_lock);
 356         cifsFileInfo_get_locked(cifs_file);
 357         spin_unlock(&cifs_file->file_info_lock);
 358         return cifs_file;
 359 }
 360
 361 /*
 362  * Release a reference on the file private data. This may involve closing
 363  * the filehandle out on the server. Must be called without holding
 364  * tcon->open_file_lock and cifs_file->file_info_lock.
 365  */
 366 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
 367 {
 368         struct inode *inode = d_inode(cifs_file->dentry);
 369         struct cifs_tcon *tcon = tlink_tcon(cifs_file->tlink);
 370         struct TCP_Server_Info *server = tcon->ses->server;
 371         struct cifsInodeInfo *cifsi = CIFS_I(inode);
 372         struct super_block *sb = inode->i_sb;
 373         struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
 374         struct cifsLockInfo *li, *tmp;
 375         struct cifs_fid fid;
 376         struct cifs_pending_open open;
 377         bool oplock_break_cancelled;
 378
 379         spin_lock(&tcon->open_file_lock);
 380
 381         spin_lock(&cifs_file->file_info_lock);
 382         if (--cifs_file->count > 0) {
 383                 spin_unlock(&cifs_file->file_info_lock);
 384                 spin_unlock(&tcon->open_file_lock);
 385                 return;
 386         }
 387         spin_unlock(&cifs_file->file_info_lock);
 388
 389         if (server->ops->get_lease_key)
 390                 server->ops->get_lease_key(inode, &fid);
 391
 392         /* store open in pending opens to make sure we don't miss lease break */
 393         cifs_add_pending_open_locked(&fid, cifs_file->tlink, &open);
 394
 395         /* remove it from the lists */
 396         list_del(&cifs_file->flist);
 397         list_del(&cifs_file->tlist);
 398
 399         if (list_empty(&cifsi->openFileList)) {
 400                 cifs_dbg(FYI, "closing last open instance for inode %p\n",
 401                          d_inode(cifs_file->dentry));
 402                 /*
 403                  * In strict cache mode we need invalidate mapping on the last
 404                  * close  because it may cause a error when we open this file
 405                  * again and get at least level II oplock.
 406                  */
 407                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
 408                         set_bit(CIFS_INO_INVALID_MAPPING, &cifsi->flags);
 409                 cifs_set_oplock_level(cifsi, 0);
 410         }
 411
 412         spin_unlock(&tcon->open_file_lock);
 413
 414         oplock_break_cancelled = cancel_work_sync(&cifs_file->oplock_break);
 415
 416         if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
 417                 struct TCP_Server_Info *server = tcon->ses->server;
 418                 unsigned int xid;
 419
 420                 xid = get_xid();
 421                 if (server->ops->close)
 422                         server->ops->close(xid, tcon, &cifs_file->fid);
 423                 _free_xid(xid);
 424         }
 425
 426         if (oplock_break_cancelled)
 427                 cifs_done_oplock_break(cifsi);
 428
 429         cifs_del_pending_open(&open);
 430
 431         /*
 432          * Delete any outstanding lock records. We'll lose them when the file
 433          * is closed anyway.
 434          */
 435         down_write(&cifsi->lock_sem);
 436         list_for_each_entry_safe(li, tmp, &cifs_file->llist->locks, llist) {
 437                 list_del(&li->llist);
 438                 cifs_del_lock_waiters(li);
 439                 kfree(li);
 440         }
 441         list_del(&cifs_file->llist->llist);
 442         kfree(cifs_file->llist);
 443         up_write(&cifsi->lock_sem);
 444
 445         cifs_put_tlink(cifs_file->tlink);
 446         dput(cifs_file->dentry);
 447         cifs_sb_deactive(sb);
 448         kfree(cifs_file);
 449 }
 450
 451 int cifs_open(struct inode *inode, struct file *file)
 452
 453 {
 454         int rc = -EACCES;
 455         unsigned int xid;
 456         __u32 oplock;
 457         struct cifs_sb_info *cifs_sb;
 458         struct TCP_Server_Info *server;
 459         struct cifs_tcon *tcon;
 460         struct tcon_link *tlink;
 461         struct cifsFileInfo *cfile = NULL;
 462         char *full_path = NULL;
 463         bool posix_open_ok = false;
 464         struct cifs_fid fid;
 465         struct cifs_pending_open open;
 466
 467         xid = get_xid();
 468
 469         cifs_sb = CIFS_SB(inode->i_sb);
 470         tlink = cifs_sb_tlink(cifs_sb);
 471         if (IS_ERR(tlink)) {
 472                 free_xid(xid);
 473                 return PTR_ERR(tlink);
 474         }
 475         tcon = tlink_tcon(tlink);
 476         server = tcon->ses->server;
 477
 478         full_path = build_path_from_dentry(file_dentry(file));
 479         if (full_path == NULL) {
 480                 rc = -ENOMEM;
 481                 goto out;
 482         }
 483
 484         cifs_dbg(FYI, "inode = 0x%p file flags are 0x%x for %s\n",
 485                  inode, file->f_flags, full_path);
 486
 487         if (file->f_flags & O_DIRECT &&
 488             cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO) {
 489                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_BRL)
 490                         file->f_op = &cifs_file_direct_nobrl_ops;
 491                 else
 492                         file->f_op = &cifs_file_direct_ops;
 493         }
 494
 495         if (server->oplocks)
 496                 oplock = REQ_OPLOCK;
 497         else
 498                 oplock = 0;
 499
 500         if (!tcon->broken_posix_open && tcon->unix_ext &&
 501             cap_unix(tcon->ses) && (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 502                                 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 503                 /* can not refresh inode info since size could be stale */
 504                 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
 505                                 cifs_sb->mnt_file_mode /* ignored */,
 506                                 file->f_flags, &oplock, &fid.netfid, xid);
 507                 if (rc == 0) {
 508                         cifs_dbg(FYI, "posix open succeeded\n");
 509                         posix_open_ok = true;
 510                 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
 511                         if (tcon->ses->serverNOS)
 512                                 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",
 513                                          tcon->ses->serverName,
 514                                          tcon->ses->serverNOS);
 515                         tcon->broken_posix_open = true;
 516                 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
 517                          (rc != -EOPNOTSUPP)) /* path not found or net err */
 518                         goto out;
 519                 /*
 520                  * Else fallthrough to retry open the old way on network i/o
 521                  * or DFS errors.
 522                  */
 523         }
 524
 525         if (server->ops->get_lease_key)
 526                 server->ops->get_lease_key(inode, &fid);
 527
 528         cifs_add_pending_open(&fid, tlink, &open);
 529
 530         if (!posix_open_ok) {
 531                 if (server->ops->get_lease_key)
 532                         server->ops->get_lease_key(inode, &fid);
 533
 534                 rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
 535                                   file->f_flags, &oplock, &fid, xid);
 536                 if (rc) {
 537                         cifs_del_pending_open(&open);
 538                         goto out;
 539                 }
 540         }
 541
 542         cfile = cifs_new_fileinfo(&fid, file, tlink, oplock);
 543         if (cfile == NULL) {
 544                 if (server->ops->close)
 545                         server->ops->close(xid, tcon, &fid);
 546                 cifs_del_pending_open(&open);
 547                 rc = -ENOMEM;
 548                 goto out;
 549         }
 550
 551         cifs_fscache_set_inode_cookie(inode, file);
 552
 553         if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
 554                 /*
 555                  * Time to set mode which we can not set earlier due to
 556                  * problems creating new read-only files.
 557                  */
 558                 struct cifs_unix_set_info_args args = {
 559                         .mode   = inode->i_mode,
 560                         .uid    = INVALID_UID, /* no change */
 561                         .gid    = INVALID_GID, /* no change */
 562                         .ctime  = NO_CHANGE_64,
 563                         .atime  = NO_CHANGE_64,
 564                         .mtime  = NO_CHANGE_64,
 565                         .device = 0,
 566                 };
 567                 CIFSSMBUnixSetFileInfo(xid, tcon, &args, fid.netfid,
 568                                        cfile->pid);
 569         }
 570
 571 out:
 572         kfree(full_path);
 573         free_xid(xid);
 574         cifs_put_tlink(tlink);
 575         return rc;
 576 }
 577
 578 static int cifs_push_posix_locks(struct cifsFileInfo *cfile);
 579
 580 /*
 581  * Try to reacquire byte range locks that were released when session
 582  * to server was lost.
 583  */
 584 static int
 585 cifs_relock_file(struct cifsFileInfo *cfile)
 586 {
 587         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
 588         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 589         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
 590         int rc = 0;
 591
 592         down_read_nested(&cinode->lock_sem, SINGLE_DEPTH_NESTING);
 593         if (cinode->can_cache_brlcks) {
 594                 /* can cache locks - no need to relock */
 595                 up_read(&cinode->lock_sem);
 596                 return rc;
 597         }
 598
 599         if (cap_unix(tcon->ses) &&
 600             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
 601             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
 602                 rc = cifs_push_posix_locks(cfile);
 603         else
 604                 rc = tcon->ses->server->ops->push_mand_locks(cfile);
 605
 606         up_read(&cinode->lock_sem);
 607         return rc;
 608 }
 609
 610 static int
 611 cifs_reopen_file(struct cifsFileInfo *cfile, bool can_flush)
 612 {
 613         int rc = -EACCES;
 614         unsigned int xid;
 615         __u32 oplock;
 616         struct cifs_sb_info *cifs_sb;
 617         struct cifs_tcon *tcon;
 618         struct TCP_Server_Info *server;
 619         struct cifsInodeInfo *cinode;
 620         struct inode *inode;
 621         char *full_path = NULL;
 622         int desired_access;
 623         int disposition = FILE_OPEN;
 624         int create_options = CREATE_NOT_DIR;
 625         struct cifs_open_parms oparms;
 626
 627         xid = get_xid();
 628         mutex_lock(&cfile->fh_mutex);
 629         if (!cfile->invalidHandle) {
 630                 mutex_unlock(&cfile->fh_mutex);
 631                 rc = 0;
 632                 free_xid(xid);
 633                 return rc;
 634         }
 635
 636         inode = d_inode(cfile->dentry);
 637         cifs_sb = CIFS_SB(inode->i_sb);
 638         tcon = tlink_tcon(cfile->tlink);
 639         server = tcon->ses->server;
 640
 641         /*
 642          * Can not grab rename sem here because various ops, including those
 643          * that already have the rename sem can end up causing writepage to get
 644          * called and if the server was down that means we end up here, and we
 645          * can never tell if the caller already has the rename_sem.
 646          */
 647         full_path = build_path_from_dentry(cfile->dentry);
 648         if (full_path == NULL) {
 649                 rc = -ENOMEM;
 650                 mutex_unlock(&cfile->fh_mutex);
 651                 free_xid(xid);
 652                 return rc;
 653         }
 654
 655         cifs_dbg(FYI, "inode = 0x%p file flags 0x%x for %s\n",
 656                  inode, cfile->f_flags, full_path);
 657
 658         if (tcon->ses->server->oplocks)
 659                 oplock = REQ_OPLOCK;
 660         else
 661                 oplock = 0;
 662
 663         if (tcon->unix_ext && cap_unix(tcon->ses) &&
 664             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 665                                 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 666                 /*
 667                  * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
 668                  * original open. Must mask them off for a reopen.
 669                  */
 670                 unsigned int oflags = cfile->f_flags &
 671                                                 ~(O_CREAT | O_EXCL | O_TRUNC);
 672
 673                 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
 674                                      cifs_sb->mnt_file_mode /* ignored */,
 675                                      oflags, &oplock, &cfile->fid.netfid, xid);
 676                 if (rc == 0) {
 677                         cifs_dbg(FYI, "posix reopen succeeded\n");
 678                         oparms.reconnect = true;
 679                         goto reopen_success;
 680                 }
 681                 /*
 682                  * fallthrough to retry open the old way on errors, especially
 683                  * in the reconnect path it is important to retry hard
 684                  */
 685         }
 686
 687         desired_access = cifs_convert_flags(cfile->f_flags);
 688
 689         if (backup_cred(cifs_sb))
 690                 create_options |= CREATE_OPEN_BACKUP_INTENT;
 691
 692         if (server->ops->get_lease_key)
 693                 server->ops->get_lease_key(inode, &cfile->fid);
 694
 695         oparms.tcon = tcon;
 696         oparms.cifs_sb = cifs_sb;
 697         oparms.desired_access = desired_access;
 698         oparms.create_options = create_options;
 699         oparms.disposition = disposition;
 700         oparms.path = full_path;
 701         oparms.fid = &cfile->fid;
 702         oparms.reconnect = true;
 703
 704         /*
 705          * Can not refresh inode by passing in file_info buf to be returned by
 706          * ops->open and then calling get_inode_info with returned buf since
 707          * file might have write behind data that needs to be flushed and server
 708          * version of file size can be stale. If we knew for sure that inode was
 709          * not dirty locally we could do this.
 710          */
 711         rc = server->ops->open(xid, &oparms, &oplock, NULL);
 712         if (rc == -ENOENT && oparms.reconnect == false) {
 713                 /* durable handle timeout is expired - open the file again */
 714                 rc = server->ops->open(xid, &oparms, &oplock, NULL);
 715                 /* indicate that we need to relock the file */
 716                 oparms.reconnect = true;
 717         }
 718
 719         if (rc) {
 720                 mutex_unlock(&cfile->fh_mutex);
 721                 cifs_dbg(FYI, "cifs_reopen returned 0x%x\n", rc);
 722                 cifs_dbg(FYI, "oplock: %d\n", oplock);
 723                 goto reopen_error_exit;
 724         }
 725
 726 reopen_success:
 727         cfile->invalidHandle = false;
 728         mutex_unlock(&cfile->fh_mutex);
 729         cinode = CIFS_I(inode);
 730
 731         if (can_flush) {
 732                 rc = filemap_write_and_wait(inode->i_mapping);
 733                 mapping_set_error(inode->i_mapping, rc);
 734
 735                 if (tcon->unix_ext)
 736                         rc = cifs_get_inode_info_unix(&inode, full_path,
 737                                                       inode->i_sb, xid);
 738                 else
 739                         rc = cifs_get_inode_info(&inode, full_path, NULL,
 740                                                  inode->i_sb, xid, NULL);
 741         }
 742         /*
 743          * Else we are writing out data to server already and could deadlock if
 744          * we tried to flush data, and since we do not know if we have data that
 745          * would invalidate the current end of file on the server we can not go
 746          * to the server to get the new inode info.
 747          */
 748
 749         /*
 750          * If the server returned a read oplock and we have mandatory brlocks,
 751          * set oplock level to None.
 752          */
 753         if (server->ops->is_read_op(oplock) && cifs_has_mand_locks(cinode)) {
 754                 cifs_dbg(FYI, "Reset oplock val from read to None due to mand locks\n");
 755                 oplock = 0;
 756         }
 757
 758         server->ops->set_fid(cfile, &cfile->fid, oplock);
 759         if (oparms.reconnect)
 760                 cifs_relock_file(cfile);
 761
 762 reopen_error_exit:
 763         kfree(full_path);
 764         free_xid(xid);
 765         return rc;
 766 }
 767
 768 int cifs_close(struct inode *inode, struct file *file)
 769 {
 770         if (file->private_data != NULL) {
 771                 cifsFileInfo_put(file->private_data);
 772                 file->private_data = NULL;
 773         }
 774
 775         /* return code from the ->release op is always ignored */
 776         return 0;
 777 }
 778
 779 void
 780 cifs_reopen_persistent_handles(struct cifs_tcon *tcon)
 781 {
 782         struct cifsFileInfo *open_file;
 783         struct list_head *tmp;
 784         struct list_head *tmp1;
 785         struct list_head tmp_list;
 786
 787         if (!tcon->use_persistent || !tcon->need_reopen_files)
 788                 return;
 789
 790         tcon->need_reopen_files = false;
 791
 792         cifs_dbg(FYI, "Reopen persistent handles");
 793         INIT_LIST_HEAD(&tmp_list);
 794
 795         /* list all files open on tree connection, reopen resilient handles  */
 796         spin_lock(&tcon->open_file_lock);
 797         list_for_each(tmp, &tcon->openFileList) {
 798                 open_file = list_entry(tmp, struct cifsFileInfo, tlist);
 799                 if (!open_file->invalidHandle)
 800                         continue;
 801                 cifsFileInfo_get(open_file);
 802                 list_add_tail(&open_file->rlist, &tmp_list);
 803         }
 804         spin_unlock(&tcon->open_file_lock);
 805
 806         list_for_each_safe(tmp, tmp1, &tmp_list) {
 807                 open_file = list_entry(tmp, struct cifsFileInfo, rlist);
 808                 if (cifs_reopen_file(open_file, false /* do not flush */))
 809                         tcon->need_reopen_files = true;
 810                 list_del_init(&open_file->rlist);
 811                 cifsFileInfo_put(open_file);
 812         }
 813 }
 814
 815 int cifs_closedir(struct inode *inode, struct file *file)
 816 {
 817         int rc = 0;
 818         unsigned int xid;
 819         struct cifsFileInfo *cfile = file->private_data;
 820         struct cifs_tcon *tcon;
 821         struct TCP_Server_Info *server;
 822         char *buf;
 823
 824         cifs_dbg(FYI, "Closedir inode = 0x%p\n", inode);
 825
 826         if (cfile == NULL)
 827                 return rc;
 828
 829         xid = get_xid();
 830         tcon = tlink_tcon(cfile->tlink);
 831         server = tcon->ses->server;
 832
 833         cifs_dbg(FYI, "Freeing private data in close dir\n");
 834         spin_lock(&cfile->file_info_lock);
 835         if (server->ops->dir_needs_close(cfile)) {
 836                 cfile->invalidHandle = true;
 837                 spin_unlock(&cfile->file_info_lock);
 838                 if (server->ops->close_dir)
 839                         rc = server->ops->close_dir(xid, tcon, &cfile->fid);
 840                 else
 841                         rc = -ENOSYS;
 842                 cifs_dbg(FYI, "Closing uncompleted readdir with rc %d\n", rc);
 843                 /* not much we can do if it fails anyway, ignore rc */
 844                 rc = 0;
 845         } else
 846                 spin_unlock(&cfile->file_info_lock);
 847
 848         buf = cfile->srch_inf.ntwrk_buf_start;
 849         if (buf) {
 850                 cifs_dbg(FYI, "closedir free smb buf in srch struct\n");
 851                 cfile->srch_inf.ntwrk_buf_start = NULL;
 852                 if (cfile->srch_inf.smallBuf)
 853                         cifs_small_buf_release(buf);
 854                 else
 855                         cifs_buf_release(buf);
 856         }
 857
 858         cifs_put_tlink(cfile->tlink);
 859         kfree(file->private_data);
 860         file->private_data = NULL;
 861         /* BB can we lock the filestruct while this is going on? */
 862         free_xid(xid);
 863         return rc;
 864 }
 865
 866 static struct cifsLockInfo *
 867 cifs_lock_init(__u64 offset, __u64 length, __u8 type)
 868 {
 869         struct cifsLockInfo *lock =
 870                 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
 871         if (!lock)
 872                 return lock;
 873         lock->offset = offset;
 874         lock->length = length;
 875         lock->type = type;
 876         lock->pid = current->tgid;
 877         INIT_LIST_HEAD(&lock->blist);
 878         init_waitqueue_head(&lock->block_q);
 879         return lock;
 880 }
 881
 882 void
 883 cifs_del_lock_waiters(struct cifsLockInfo *lock)
 884 {
 885         struct cifsLockInfo *li, *tmp;
 886         list_for_each_entry_safe(li, tmp, &lock->blist, blist) {
 887                 list_del_init(&li->blist);
 888                 wake_up(&li->block_q);
 889         }
 890 }
 891
 892 #define CIFS_LOCK_OP    0
 893 #define CIFS_READ_OP    1
 894 #define CIFS_WRITE_OP   2
 895
 896 /* @rw_check : 0 - no op, 1 - read, 2 - write */
 897 static bool
 898 cifs_find_fid_lock_conflict(struct cifs_fid_locks *fdlocks, __u64 offset,
 899                             __u64 length, __u8 type, struct cifsFileInfo *cfile,
 900                             struct cifsLockInfo **conf_lock, int rw_check)
 901 {
 902         struct cifsLockInfo *li;
 903         struct cifsFileInfo *cur_cfile = fdlocks->cfile;
 904         struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
 905
 906         list_for_each_entry(li, &fdlocks->locks, llist) {
 907                 if (offset + length <= li->offset ||
 908                     offset >= li->offset + li->length)
 909                         continue;
 910                 if (rw_check != CIFS_LOCK_OP && current->tgid == li->pid &&
 911                     server->ops->compare_fids(cfile, cur_cfile)) {
 912                         /* shared lock prevents write op through the same fid */
 913                         if (!(li->type & server->vals->shared_lock_type) ||
 914                             rw_check != CIFS_WRITE_OP)
 915                                 continue;
 916                 }
 917                 if ((type & server->vals->shared_lock_type) &&
 918                     ((server->ops->compare_fids(cfile, cur_cfile) &&
 919                      current->tgid == li->pid) || type == li->type))
 920                         continue;
 921                 if (conf_lock)
 922                         *conf_lock = li;
 923                 return true;
 924         }
 925         return false;
 926 }
 927
 928 bool
 929 cifs_find_lock_conflict(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
 930                         __u8 type, struct cifsLockInfo **conf_lock,
 931                         int rw_check)
 932 {
 933         bool rc = false;
 934         struct cifs_fid_locks *cur;
 935         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 936
 937         list_for_each_entry(cur, &cinode->llist, llist) {
 938                 rc = cifs_find_fid_lock_conflict(cur, offset, length, type,
 939                                                  cfile, conf_lock, rw_check);
 940                 if (rc)
 941                         break;
 942         }
 943
 944         return rc;
 945 }
 946
 947 /*
 948  * Check if there is another lock that prevents us to set the lock (mandatory
 949  * style). If such a lock exists, update the flock structure with its
 950  * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
 951  * or leave it the same if we can't. Returns 0 if we don't need to request to
 952  * the server or 1 otherwise.
 953  */
 954 static int
 955 cifs_lock_test(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
 956                __u8 type, struct file_lock *flock)
 957 {
 958         int rc = 0;
 959         struct cifsLockInfo *conf_lock;
 960         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 961         struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
 962         bool exist;
 963
 964         down_read(&cinode->lock_sem);
 965
 966         exist = cifs_find_lock_conflict(cfile, offset, length, type,
 967                                         &conf_lock, CIFS_LOCK_OP);
 968         if (exist) {
 969                 flock->fl_start = conf_lock->offset;
 970                 flock->fl_end = conf_lock->offset + conf_lock->length - 1;
 971                 flock->fl_pid = conf_lock->pid;
 972                 if (conf_lock->type & server->vals->shared_lock_type)
 973                         flock->fl_type = F_RDLCK;
 974                 else
 975                         flock->fl_type = F_WRLCK;
 976         } else if (!cinode->can_cache_brlcks)
 977                 rc = 1;
 978         else
 979                 flock->fl_type = F_UNLCK;
 980
 981         up_read(&cinode->lock_sem);
 982         return rc;
 983 }
 984
 985 static void
 986 cifs_lock_add(struct cifsFileInfo *cfile, struct cifsLockInfo *lock)
 987 {
 988         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 989         down_write(&cinode->lock_sem);
 990         list_add_tail(&lock->llist, &cfile->llist->locks);
 991         up_write(&cinode->lock_sem);
 992 }
 993
 994 /*
 995  * Set the byte-range lock (mandatory style). Returns:
 996  * 1) 0, if we set the lock and don't need to request to the server;
 997  * 2) 1, if no locks prevent us but we need to request to the server;
 998  * 3) -EACCESS, if there is a lock that prevents us and wait is false.
 999  */
1000 static int
1001 cifs_lock_add_if(struct cifsFileInfo *cfile, struct cifsLockInfo *lock,
1002                  bool wait)
1003 {
1004         struct cifsLockInfo *conf_lock;
1005         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1006         bool exist;
1007         int rc = 0;
1008
1009 try_again:
1010         exist = false;
1011         down_write(&cinode->lock_sem);
1012
1013         exist = cifs_find_lock_conflict(cfile, lock->offset, lock->length,
1014                                         lock->type, &conf_lock, CIFS_LOCK_OP);
1015         if (!exist && cinode->can_cache_brlcks) {
1016                 list_add_tail(&lock->llist, &cfile->llist->locks);
1017                 up_write(&cinode->lock_sem);
1018                 return rc;
1019         }
1020
1021         if (!exist)
1022                 rc = 1;
1023         else if (!wait)
1024                 rc = -EACCES;
1025         else {
1026                 list_add_tail(&lock->blist, &conf_lock->blist);
1027                 up_write(&cinode->lock_sem);
1028                 rc = wait_event_interruptible(lock->block_q,
1029                                         (lock->blist.prev == &lock->blist) &&
1030                                         (lock->blist.next == &lock->blist));
1031                 if (!rc)
1032                         goto try_again;
1033                 down_write(&cinode->lock_sem);
1034                 list_del_init(&lock->blist);
1035         }
1036
1037         up_write(&cinode->lock_sem);
1038         return rc;
1039 }
1040
1041 /*
1042  * Check if there is another lock that prevents us to set the lock (posix
1043  * style). If such a lock exists, update the flock structure with its
1044  * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
1045  * or leave it the same if we can't. Returns 0 if we don't need to request to
1046  * the server or 1 otherwise.
1047  */
1048 static int
1049 cifs_posix_lock_test(struct file *file, struct file_lock *flock)
1050 {
1051         int rc = 0;
1052         struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
1053         unsigned char saved_type = flock->fl_type;
1054
1055         if ((flock->fl_flags & FL_POSIX) == 0)
1056                 return 1;
1057
1058         down_read(&cinode->lock_sem);
1059         posix_test_lock(file, flock);
1060
1061         if (flock->fl_type == F_UNLCK && !cinode->can_cache_brlcks) {
1062                 flock->fl_type = saved_type;
1063                 rc = 1;
1064         }
1065
1066         up_read(&cinode->lock_sem);
1067         return rc;
1068 }
1069
1070 /*
1071  * Set the byte-range lock (posix style). Returns:
1072  * 1) 0, if we set the lock and don't need to request to the server;
1073  * 2) 1, if we need to request to the server;
1074  * 3) <0, if the error occurs while setting the lock.
1075  */
1076 static int
1077 cifs_posix_lock_set(struct file *file, struct file_lock *flock)
1078 {
1079         struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
1080         int rc = 1;
1081
1082         if ((flock->fl_flags & FL_POSIX) == 0)
1083                 return rc;
1084
1085 try_again:
1086         down_write(&cinode->lock_sem);
1087         if (!cinode->can_cache_brlcks) {
1088                 up_write(&cinode->lock_sem);
1089                 return rc;
1090         }
1091
1092         rc = posix_lock_file(file, flock, NULL);
1093         up_write(&cinode->lock_sem);
1094         if (rc == FILE_LOCK_DEFERRED) {
1095                 rc = wait_event_interruptible(flock->fl_wait, !flock->fl_next);
1096                 if (!rc)
1097                         goto try_again;
1098                 posix_unblock_lock(flock);
1099         }
1100         return rc;
1101 }
1102
1103 int
1104 cifs_push_mandatory_locks(struct cifsFileInfo *cfile)
1105 {
1106         unsigned int xid;
1107         int rc = 0, stored_rc;
1108         struct cifsLockInfo *li, *tmp;
1109         struct cifs_tcon *tcon;
1110         unsigned int num, max_num, max_buf;
1111         LOCKING_ANDX_RANGE *buf, *cur;
1112         static const int types[] = {
1113                 LOCKING_ANDX_LARGE_FILES,
1114                 LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES
1115         };
1116         int i;
1117
1118         xid = get_xid();
1119         tcon = tlink_tcon(cfile->tlink);
1120
1121         /*
1122          * Accessing maxBuf is racy with cifs_reconnect - need to store value
1123          * and check it for zero before using.
1124          */
1125         max_buf = tcon->ses->server->maxBuf;
1126         if (!max_buf) {
1127                 free_xid(xid);
1128                 return -EINVAL;
1129         }
1130
1131         max_num = (max_buf - sizeof(struct smb_hdr)) /
1132                                                 sizeof(LOCKING_ANDX_RANGE);
1133         buf = kcalloc(max_num, sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1134         if (!buf) {
1135                 free_xid(xid);
1136                 return -ENOMEM;
1137         }
1138
1139         for (i = 0; i < 2; i++) {
1140                 cur = buf;
1141                 num = 0;
1142                 list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1143                         if (li->type != types[i])
1144                                 continue;
1145                         cur->Pid = cpu_to_le16(li->pid);
1146                         cur->LengthLow = cpu_to_le32((u32)li->length);
1147                         cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1148                         cur->OffsetLow = cpu_to_le32((u32)li->offset);
1149                         cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1150                         if (++num == max_num) {
1151                                 stored_rc = cifs_lockv(xid, tcon,
1152                                                        cfile->fid.netfid,
1153                                                        (__u8)li->type, 0, num,
1154                                                        buf);
1155                                 if (stored_rc)
1156                                         rc = stored_rc;
1157                                 cur = buf;
1158                                 num = 0;
1159                         } else
1160                                 cur++;
1161                 }
1162
1163                 if (num) {
1164                         stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1165                                                (__u8)types[i], 0, num, buf);
1166                         if (stored_rc)
1167                                 rc = stored_rc;
1168                 }
1169         }
1170
1171         kfree(buf);
1172         free_xid(xid);
1173         return rc;
1174 }
1175
1176 static __u32
1177 hash_lockowner(fl_owner_t owner)
1178 {
1179         return cifs_lock_secret ^ hash32_ptr((const void *)owner);
1180 }
1181
1182 struct lock_to_push {
1183         struct list_head llist;
1184         __u64 offset;
1185         __u64 length;
1186         __u32 pid;
1187         __u16 netfid;
1188         __u8 type;
1189 };
1190
1191 static int
1192 cifs_push_posix_locks(struct cifsFileInfo *cfile)
1193 {
1194         struct inode *inode = d_inode(cfile->dentry);
1195         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1196         struct file_lock *flock;
1197         struct file_lock_context *flctx = inode->i_flctx;
1198         unsigned int count = 0, i;
1199         int rc = 0, xid, type;
1200         struct list_head locks_to_send, *el;
1201         struct lock_to_push *lck, *tmp;
1202         __u64 length;
1203
1204         xid = get_xid();
1205
1206         if (!flctx)
1207                 goto out;
1208
1209         spin_lock(&flctx->flc_lock);
1210         list_for_each(el, &flctx->flc_posix) {
1211                 count++;
1212         }
1213         spin_unlock(&flctx->flc_lock);
1214
1215         INIT_LIST_HEAD(&locks_to_send);
1216
1217         /*
1218          * Allocating count locks is enough because no FL_POSIX locks can be
1219          * added to the list while we are holding cinode->lock_sem that
1220          * protects locking operations of this inode.
1221          */
1222         for (i = 0; i < count; i++) {
1223                 lck = kmalloc(sizeof(struct lock_to_push), GFP_KERNEL);
1224                 if (!lck) {
1225                         rc = -ENOMEM;
1226                         goto err_out;
1227                 }
1228                 list_add_tail(&lck->llist, &locks_to_send);
1229         }
1230
1231         el = locks_to_send.next;
1232         spin_lock(&flctx->flc_lock);
1233         list_for_each_entry(flock, &flctx->flc_posix, fl_list) {
1234                 if (el == &locks_to_send) {
1235                         /*
1236                          * The list ended. We don't have enough allocated
1237                          * structures - something is really wrong.
1238                          */
1239                         cifs_dbg(VFS, "Can't push all brlocks!\n");
1240                         break;
1241                 }
1242                 length = 1 + flock->fl_end - flock->fl_start;
1243                 if (flock->fl_type == F_RDLCK || flock->fl_type == F_SHLCK)
1244                         type = CIFS_RDLCK;
1245                 else
1246                         type = CIFS_WRLCK;
1247                 lck = list_entry(el, struct lock_to_push, llist);
1248                 lck->pid = hash_lockowner(flock->fl_owner);
1249                 lck->netfid = cfile->fid.netfid;
1250                 lck->length = length;
1251                 lck->type = type;
1252                 lck->offset = flock->fl_start;
1253         }
1254         spin_unlock(&flctx->flc_lock);
1255
1256         list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1257                 int stored_rc;
1258
1259                 stored_rc = CIFSSMBPosixLock(xid, tcon, lck->netfid, lck->pid,
1260                                              lck->offset, lck->length, NULL,
1261                                              lck->type, 0);
1262                 if (stored_rc)
1263                         rc = stored_rc;
1264                 list_del(&lck->llist);
1265                 kfree(lck);
1266         }
1267
1268 out:
1269         free_xid(xid);
1270         return rc;
1271 err_out:
1272         list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1273                 list_del(&lck->llist);
1274                 kfree(lck);
1275         }
1276         goto out;
1277 }
1278
1279 static int
1280 cifs_push_locks(struct cifsFileInfo *cfile)
1281 {
1282         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
1283         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1284         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1285         int rc = 0;
1286
1287         /* we are going to update can_cache_brlcks here - need a write access */
1288         down_write(&cinode->lock_sem);
1289         if (!cinode->can_cache_brlcks) {
1290                 up_write(&cinode->lock_sem);
1291                 return rc;
1292         }
1293
1294         if (cap_unix(tcon->ses) &&
1295             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1296             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1297                 rc = cifs_push_posix_locks(cfile);
1298         else
1299                 rc = tcon->ses->server->ops->push_mand_locks(cfile);
1300
1301         cinode->can_cache_brlcks = false;
1302         up_write(&cinode->lock_sem);
1303         return rc;
1304 }
1305
1306 static void
1307 cifs_read_flock(struct file_lock *flock, __u32 *type, int *lock, int *unlock,
1308                 bool *wait_flag, struct TCP_Server_Info *server)
1309 {
1310         if (flock->fl_flags & FL_POSIX)
1311                 cifs_dbg(FYI, "Posix\n");
1312         if (flock->fl_flags & FL_FLOCK)
1313                 cifs_dbg(FYI, "Flock\n");
1314         if (flock->fl_flags & FL_SLEEP) {
1315                 cifs_dbg(FYI, "Blocking lock\n");
1316                 *wait_flag = true;
1317         }
1318         if (flock->fl_flags & FL_ACCESS)
1319                 cifs_dbg(FYI, "Process suspended by mandatory locking - not implemented yet\n");
1320         if (flock->fl_flags & FL_LEASE)
1321                 cifs_dbg(FYI, "Lease on file - not implemented yet\n");
1322         if (flock->fl_flags &
1323             (~(FL_POSIX | FL_FLOCK | FL_SLEEP |
1324                FL_ACCESS | FL_LEASE | FL_CLOSE)))
1325                 cifs_dbg(FYI, "Unknown lock flags 0x%x\n", flock->fl_flags);
1326
1327         *type = server->vals->large_lock_type;
1328         if (flock->fl_type == F_WRLCK) {
1329                 cifs_dbg(FYI, "F_WRLCK\n");
1330                 *type |= server->vals->exclusive_lock_type;
1331                 *lock = 1;
1332         } else if (flock->fl_type == F_UNLCK) {
1333                 cifs_dbg(FYI, "F_UNLCK\n");
1334                 *type |= server->vals->unlock_lock_type;
1335                 *unlock = 1;
1336                 /* Check if unlock includes more than one lock range */
1337         } else if (flock->fl_type == F_RDLCK) {
1338                 cifs_dbg(FYI, "F_RDLCK\n");
1339                 *type |= server->vals->shared_lock_type;
1340                 *lock = 1;
1341         } else if (flock->fl_type == F_EXLCK) {
1342                 cifs_dbg(FYI, "F_EXLCK\n");
1343                 *type |= server->vals->exclusive_lock_type;
1344                 *lock = 1;
1345         } else if (flock->fl_type == F_SHLCK) {
1346                 cifs_dbg(FYI, "F_SHLCK\n");
1347                 *type |= server->vals->shared_lock_type;
1348                 *lock = 1;
1349         } else
1350                 cifs_dbg(FYI, "Unknown type of lock\n");
1351 }
1352
1353 static int
1354 cifs_getlk(struct file *file, struct file_lock *flock, __u32 type,
1355            bool wait_flag, bool posix_lck, unsigned int xid)
1356 {
1357         int rc = 0;
1358         __u64 length = 1 + flock->fl_end - flock->fl_start;
1359         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1360         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1361         struct TCP_Server_Info *server = tcon->ses->server;
1362         __u16 netfid = cfile->fid.netfid;
1363
1364         if (posix_lck) {
1365                 int posix_lock_type;
1366
1367                 rc = cifs_posix_lock_test(file, flock);
1368                 if (!rc)
1369                         return rc;
1370
1371                 if (type & server->vals->shared_lock_type)
1372                         posix_lock_type = CIFS_RDLCK;
1373                 else
1374                         posix_lock_type = CIFS_WRLCK;
1375                 rc = CIFSSMBPosixLock(xid, tcon, netfid,
1376                                       hash_lockowner(flock->fl_owner),
1377                                       flock->fl_start, length, flock,
1378                                       posix_lock_type, wait_flag);
1379                 return rc;
1380         }
1381
1382         rc = cifs_lock_test(cfile, flock->fl_start, length, type, flock);
1383         if (!rc)
1384                 return rc;
1385
1386         /* BB we could chain these into one lock request BB */
1387         rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length, type,
1388                                     1, 0, false);
1389         if (rc == 0) {
1390                 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1391                                             type, 0, 1, false);
1392                 flock->fl_type = F_UNLCK;
1393                 if (rc != 0)
1394                         cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1395                                  rc);
1396                 return 0;
1397         }
1398
1399         if (type & server->vals->shared_lock_type) {
1400                 flock->fl_type = F_WRLCK;
1401                 return 0;
1402         }
1403
1404         type &= ~server->vals->exclusive_lock_type;
1405
1406         rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1407                                     type | server->vals->shared_lock_type,
1408                                     1, 0, false);
1409         if (rc == 0) {
1410                 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1411                         type | server->vals->shared_lock_type, 0, 1, false);
1412                 flock->fl_type = F_RDLCK;
1413                 if (rc != 0)
1414                         cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1415                                  rc);
1416         } else
1417                 flock->fl_type = F_WRLCK;
1418
1419         return 0;
1420 }
1421
1422 void
1423 cifs_move_llist(struct list_head *source, struct list_head *dest)
1424 {
1425         struct list_head *li, *tmp;
1426         list_for_each_safe(li, tmp, source)
1427                 list_move(li, dest);
1428 }
1429
1430 void
1431 cifs_free_llist(struct list_head *llist)
1432 {
1433         struct cifsLockInfo *li, *tmp;
1434         list_for_each_entry_safe(li, tmp, llist, llist) {
1435                 cifs_del_lock_waiters(li);
1436                 list_del(&li->llist);
1437                 kfree(li);
1438         }
1439 }
1440
1441 int
1442 cifs_unlock_range(struct cifsFileInfo *cfile, struct file_lock *flock,
1443                   unsigned int xid)
1444 {
1445         int rc = 0, stored_rc;
1446         static const int types[] = {
1447                 LOCKING_ANDX_LARGE_FILES,
1448                 LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES
1449         };
1450         unsigned int i;
1451         unsigned int max_num, num, max_buf;
1452         LOCKING_ANDX_RANGE *buf, *cur;
1453         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1454         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1455         struct cifsLockInfo *li, *tmp;
1456         __u64 length = 1 + flock->fl_end - flock->fl_start;
1457         struct list_head tmp_llist;
1458
1459         INIT_LIST_HEAD(&tmp_llist);
1460
1461         /*
1462          * Accessing maxBuf is racy with cifs_reconnect - need to store value
1463          * and check it for zero before using.
1464          */
1465         max_buf = tcon->ses->server->maxBuf;
1466         if (!max_buf)
1467                 return -EINVAL;
1468
1469         max_num = (max_buf - sizeof(struct smb_hdr)) /
1470                                                 sizeof(LOCKING_ANDX_RANGE);
1471         buf = kcalloc(max_num, sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1472         if (!buf)
1473                 return -ENOMEM;
1474
1475         down_write(&cinode->lock_sem);
1476         for (i = 0; i < 2; i++) {
1477                 cur = buf;
1478                 num = 0;
1479                 list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1480                         if (flock->fl_start > li->offset ||
1481                             (flock->fl_start + length) <
1482                             (li->offset + li->length))
1483                                 continue;
1484                         if (current->tgid != li->pid)
1485                                 continue;
1486                         if (types[i] != li->type)
1487                                 continue;
1488                         if (cinode->can_cache_brlcks) {
1489                                 /*
1490                                  * We can cache brlock requests - simply remove
1491                                  * a lock from the file's list.
1492                                  */
1493                                 list_del(&li->llist);
1494                                 cifs_del_lock_waiters(li);
1495                                 kfree(li);
1496                                 continue;
1497                         }
1498                         cur->Pid = cpu_to_le16(li->pid);
1499                         cur->LengthLow = cpu_to_le32((u32)li->length);
1500                         cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1501                         cur->OffsetLow = cpu_to_le32((u32)li->offset);
1502                         cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1503                         /*
1504                          * We need to save a lock here to let us add it again to
1505                          * the file's list if the unlock range request fails on
1506                          * the server.
1507                          */
1508                         list_move(&li->llist, &tmp_llist);
1509                         if (++num == max_num) {
1510                                 stored_rc = cifs_lockv(xid, tcon,
1511                                                        cfile->fid.netfid,
1512                                                        li->type, num, 0, buf);
1513                                 if (stored_rc) {
1514                                         /*
1515                                          * We failed on the unlock range
1516                                          * request - add all locks from the tmp
1517                                          * list to the head of the file's list.
1518                                          */
1519                                         cifs_move_llist(&tmp_llist,
1520                                                         &cfile->llist->locks);
1521                                         rc = stored_rc;
1522                                 } else
1523                                         /*
1524                                          * The unlock range request succeed -
1525                                          * free the tmp list.
1526                                          */
1527                                         cifs_free_llist(&tmp_llist);
1528                                 cur = buf;
1529                                 num = 0;
1530                         } else
1531                                 cur++;
1532                 }
1533                 if (num) {
1534                         stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1535                                                types[i], num, 0, buf);
1536                         if (stored_rc) {
1537                                 cifs_move_llist(&tmp_llist,
1538                                                 &cfile->llist->locks);
1539                                 rc = stored_rc;
1540                         } else
1541                                 cifs_free_llist(&tmp_llist);
1542                 }
1543         }
1544
1545         up_write(&cinode->lock_sem);
1546         kfree(buf);
1547         return rc;
1548 }
1549
1550 static int
1551 cifs_setlk(struct file *file, struct file_lock *flock, __u32 type,
1552            bool wait_flag, bool posix_lck, int lock, int unlock,
1553            unsigned int xid)
1554 {
1555         int rc = 0;
1556         __u64 length = 1 + flock->fl_end - flock->fl_start;
1557         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1558         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1559         struct TCP_Server_Info *server = tcon->ses->server;
1560         struct inode *inode = d_inode(cfile->dentry);
1561
1562         if (posix_lck) {
1563                 int posix_lock_type;
1564
1565                 rc = cifs_posix_lock_set(file, flock);
1566                 if (!rc || rc < 0)
1567                         return rc;
1568
1569                 if (type & server->vals->shared_lock_type)
1570                         posix_lock_type = CIFS_RDLCK;
1571                 else
1572                         posix_lock_type = CIFS_WRLCK;
1573
1574                 if (unlock == 1)
1575                         posix_lock_type = CIFS_UNLCK;
1576
1577                 rc = CIFSSMBPosixLock(xid, tcon, cfile->fid.netfid,
1578                                       hash_lockowner(flock->fl_owner),
1579                                       flock->fl_start, length,
1580                                       NULL, posix_lock_type, wait_flag);
1581                 goto out;
1582         }
1583
1584         if (lock) {
1585                 struct cifsLockInfo *lock;
1586
1587                 lock = cifs_lock_init(flock->fl_start, length, type);
1588                 if (!lock)
1589                         return -ENOMEM;
1590
1591                 rc = cifs_lock_add_if(cfile, lock, wait_flag);
1592                 if (rc < 0) {
1593                         kfree(lock);
1594                         return rc;
1595                 }
1596                 if (!rc)
1597                         goto out;
1598
1599                 /*
1600                  * Windows 7 server can delay breaking lease from read to None
1601                  * if we set a byte-range lock on a file - break it explicitly
1602                  * before sending the lock to the server to be sure the next
1603                  * read won't conflict with non-overlapted locks due to
1604                  * pagereading.
1605                  */
1606                 if (!CIFS_CACHE_WRITE(CIFS_I(inode)) &&
1607                                         CIFS_CACHE_READ(CIFS_I(inode))) {
1608                         cifs_zap_mapping(inode);
1609                         cifs_dbg(FYI, "Set no oplock for inode=%p due to mand locks\n",
1610                                  inode);
1611                         CIFS_I(inode)->oplock = 0;
1612                 }
1613
1614                 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1615                                             type, 1, 0, wait_flag);
1616                 if (rc) {
1617                         kfree(lock);
1618                         return rc;
1619                 }
1620
1621                 cifs_lock_add(cfile, lock);
1622         } else if (unlock)
1623                 rc = server->ops->mand_unlock_range(cfile, flock, xid);
1624
1625 out:
1626         if (flock->fl_flags & FL_POSIX && !rc)
1627                 rc = locks_lock_file_wait(file, flock);
1628         return rc;
1629 }
1630
1631 int cifs_lock(struct file *file, int cmd, struct file_lock *flock)
1632 {
1633         int rc, xid;
1634         int lock = 0, unlock = 0;
1635         bool wait_flag = false;
1636         bool posix_lck = false;
1637         struct cifs_sb_info *cifs_sb;
1638         struct cifs_tcon *tcon;
1639         struct cifsInodeInfo *cinode;
1640         struct cifsFileInfo *cfile;
1641         __u16 netfid;
1642         __u32 type;
1643
1644         rc = -EACCES;
1645         xid = get_xid();
1646
1647         cifs_dbg(FYI, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld end: %lld\n",
1648                  cmd, flock->fl_flags, flock->fl_type,
1649                  flock->fl_start, flock->fl_end);
1650
1651         cfile = (struct cifsFileInfo *)file->private_data;
1652         tcon = tlink_tcon(cfile->tlink);
1653
1654         cifs_read_flock(flock, &type, &lock, &unlock, &wait_flag,
1655                         tcon->ses->server);
1656
1657         cifs_sb = CIFS_FILE_SB(file);
1658         netfid = cfile->fid.netfid;
1659         cinode = CIFS_I(file_inode(file));
1660
1661         if (cap_unix(tcon->ses) &&
1662             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1663             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1664                 posix_lck = true;
1665         /*
1666          * BB add code here to normalize offset and length to account for
1667          * negative length which we can not accept over the wire.
1668          */
1669         if (IS_GETLK(cmd)) {
1670                 rc = cifs_getlk(file, flock, type, wait_flag, posix_lck, xid);
1671                 free_xid(xid);
1672                 return rc;
1673         }
1674
1675         if (!lock && !unlock) {
1676                 /*
1677                  * if no lock or unlock then nothing to do since we do not
1678                  * know what it is
1679                  */
1680                 free_xid(xid);
1681                 return -EOPNOTSUPP;
1682         }
1683
1684         rc = cifs_setlk(file, flock, type, wait_flag, posix_lck, lock, unlock,
1685                         xid);
1686         free_xid(xid);
1687         return rc;
1688 }
1689
1690 /*
1691  * update the file size (if needed) after a write. Should be called with
1692  * the inode->i_lock held
1693  */
1694 void
1695 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
1696                       unsigned int bytes_written)
1697 {
1698         loff_t end_of_write = offset + bytes_written;
1699
1700         if (end_of_write > cifsi->server_eof)
1701                 cifsi->server_eof = end_of_write;
1702 }
1703
1704 static ssize_t
1705 cifs_write(struct cifsFileInfo *open_file, __u32 pid, const char *write_data,
1706            size_t write_size, loff_t *offset)
1707 {
1708         int rc = 0;
1709         unsigned int bytes_written = 0;
1710         unsigned int total_written;
1711         struct cifs_sb_info *cifs_sb;
1712         struct cifs_tcon *tcon;
1713         struct TCP_Server_Info *server;
1714         unsigned int xid;
1715         struct dentry *dentry = open_file->dentry;
1716         struct cifsInodeInfo *cifsi = CIFS_I(d_inode(dentry));
1717         struct cifs_io_parms io_parms;
1718
1719         cifs_sb = CIFS_SB(dentry->d_sb);
1720
1721         cifs_dbg(FYI, "write %zd bytes to offset %lld of %pd\n",
1722                  write_size, *offset, dentry);
1723
1724         tcon = tlink_tcon(open_file->tlink);
1725         server = tcon->ses->server;
1726
1727         if (!server->ops->sync_write)
1728                 return -ENOSYS;
1729
1730         xid = get_xid();
1731
1732         for (total_written = 0; write_size > total_written;
1733              total_written += bytes_written) {
1734                 rc = -EAGAIN;
1735                 while (rc == -EAGAIN) {
1736                         struct kvec iov[2];
1737                         unsigned int len;
1738
1739                         if (open_file->invalidHandle) {
1740                                 /* we could deadlock if we called
1741                                    filemap_fdatawait from here so tell
1742                                    reopen_file not to flush data to
1743                                    server now */
1744                                 rc = cifs_reopen_file(open_file, false);
1745                                 if (rc != 0)
1746                                         break;
1747                         }
1748
1749                         len = min(server->ops->wp_retry_size(d_inode(dentry)),
1750                                   (unsigned int)write_size - total_written);
1751                         /* iov[0] is reserved for smb header */
1752                         iov[1].iov_base = (char *)write_data + total_written;
1753                         iov[1].iov_len = len;
1754                         io_parms.pid = pid;
1755                         io_parms.tcon = tcon;
1756                         io_parms.offset = *offset;
1757                         io_parms.length = len;
1758                         rc = server->ops->sync_write(xid, &open_file->fid,
1759                                         &io_parms, &bytes_written, iov, 1);
1760                 }
1761                 if (rc || (bytes_written == 0)) {
1762                         if (total_written)
1763                                 break;
1764                         else {
1765                                 free_xid(xid);
1766                                 return rc;
1767                         }
1768                 } else {
1769                         spin_lock(&d_inode(dentry)->i_lock);
1770                         cifs_update_eof(cifsi, *offset, bytes_written);
1771                         spin_unlock(&d_inode(dentry)->i_lock);
1772                         *offset += bytes_written;
1773                 }
1774         }
1775
1776         cifs_stats_bytes_written(tcon, total_written);
1777
1778         if (total_written > 0) {
1779                 spin_lock(&d_inode(dentry)->i_lock);
1780                 if (*offset > d_inode(dentry)->i_size)
1781                         i_size_write(d_inode(dentry), *offset);
1782                 spin_unlock(&d_inode(dentry)->i_lock);
1783         }
1784         mark_inode_dirty_sync(d_inode(dentry));
1785         free_xid(xid);
1786         return total_written;
1787 }
1788
1789 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1790                                         bool fsuid_only)
1791 {
1792         struct cifsFileInfo *open_file = NULL;
1793         struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1794         struct cifs_tcon *tcon = cifs_sb_master_tcon(cifs_sb);
1795
1796         /* only filter by fsuid on multiuser mounts */
1797         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1798                 fsuid_only = false;
1799
1800         spin_lock(&tcon->open_file_lock);
1801         /* we could simply get the first_list_entry since write-only entries
1802            are always at the end of the list but since the first entry might
1803            have a close pending, we go through the whole list */
1804         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1805                 if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
1806                         continue;
1807                 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1808                         if (!open_file->invalidHandle) {
1809                                 /* found a good file */
1810                                 /* lock it so it will not be closed on us */
1811                                 cifsFileInfo_get(open_file);
1812                                 spin_unlock(&tcon->open_file_lock);
1813                                 return open_file;
1814                         } /* else might as well continue, and look for
1815                              another, or simply have the caller reopen it
1816                              again rather than trying to fix this handle */
1817                 } else /* write only file */
1818                         break; /* write only files are last so must be done */
1819         }
1820         spin_unlock(&tcon->open_file_lock);
1821         return NULL;
1822 }
1823
1824 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1825                                         bool fsuid_only)
1826 {
1827         struct cifsFileInfo *open_file, *inv_file = NULL;
1828         struct cifs_sb_info *cifs_sb;
1829         struct cifs_tcon *tcon;
1830         bool any_available = false;
1831         int rc;
1832         unsigned int refind = 0;
1833
1834         /* Having a null inode here (because mapping->host was set to zero by
1835         the VFS or MM) should not happen but we had reports of on oops (due to
1836         it being zero) during stress testcases so we need to check for it */
1837
1838         if (cifs_inode == NULL) {
1839                 cifs_dbg(VFS, "Null inode passed to cifs_writeable_file\n");
1840                 dump_stack();
1841                 return NULL;
1842         }
1843
1844         cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1845         tcon = cifs_sb_master_tcon(cifs_sb);
1846
1847         /* only filter by fsuid on multiuser mounts */
1848         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1849                 fsuid_only = false;
1850
1851         spin_lock(&tcon->open_file_lock);
1852 refind_writable:
1853         if (refind > MAX_REOPEN_ATT) {
1854                 spin_unlock(&tcon->open_file_lock);
1855                 return NULL;
1856         }
1857         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1858                 if (!any_available && open_file->pid != current->tgid)
1859                         continue;
1860                 if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
1861                         continue;
1862                 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1863                         if (!open_file->invalidHandle) {
1864                                 /* found a good writable file */
1865                                 cifsFileInfo_get(open_file);
1866                                 spin_unlock(&tcon->open_file_lock);
1867                                 return open_file;
1868                         } else {
1869                                 if (!inv_file)
1870                                         inv_file = open_file;
1871                         }
1872                 }
1873         }
1874         /* couldn't find useable FH with same pid, try any available */
1875         if (!any_available) {
1876                 any_available = true;
1877                 goto refind_writable;
1878         }
1879
1880         if (inv_file) {
1881                 any_available = false;
1882                 cifsFileInfo_get(inv_file);
1883         }
1884
1885         spin_unlock(&tcon->open_file_lock);
1886
1887         if (inv_file) {
1888                 rc = cifs_reopen_file(inv_file, false);
1889                 if (!rc)
1890                         return inv_file;
1891                 else {
1892                         spin_lock(&tcon->open_file_lock);
1893                         list_move_tail(&inv_file->flist,
1894                                         &cifs_inode->openFileList);
1895                         spin_unlock(&tcon->open_file_lock);
1896                         cifsFileInfo_put(inv_file);
1897                         ++refind;
1898                         inv_file = NULL;
1899                         spin_lock(&tcon->open_file_lock);
1900                         goto refind_writable;
1901                 }
1902         }
1903
1904         return NULL;
1905 }
1906
1907 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1908 {
1909         struct address_space *mapping = page->mapping;
1910         loff_t offset = (loff_t)page->index << PAGE_SHIFT;
1911         char *write_data;
1912         int rc = -EFAULT;
1913         int bytes_written = 0;
1914         struct inode *inode;
1915         struct cifsFileInfo *open_file;
1916
1917         if (!mapping || !mapping->host)
1918                 return -EFAULT;
1919
1920         inode = page->mapping->host;
1921
1922         offset += (loff_t)from;
1923         write_data = kmap(page);
1924         write_data += from;
1925
1926         if ((to > PAGE_SIZE) || (from > to)) {
1927                 kunmap(page);
1928                 return -EIO;
1929         }
1930
1931         /* racing with truncate? */
1932         if (offset > mapping->host->i_size) {
1933                 kunmap(page);
1934                 return 0; /* don't care */
1935         }
1936
1937         /* check to make sure that we are not extending the file */
1938         if (mapping->host->i_size - offset < (loff_t)to)
1939                 to = (unsigned)(mapping->host->i_size - offset);
1940
1941         open_file = find_writable_file(CIFS_I(mapping->host), false);
1942         if (open_file) {
1943                 bytes_written = cifs_write(open_file, open_file->pid,
1944                                            write_data, to - from, &offset);
1945                 cifsFileInfo_put(open_file);
1946                 /* Does mm or vfs already set times? */
1947                 inode->i_atime = inode->i_mtime = current_time(inode);
1948                 if ((bytes_written > 0) && (offset))
1949                         rc = 0;
1950                 else if (bytes_written < 0)
1951                         rc = bytes_written;
1952         } else {
1953                 cifs_dbg(FYI, "No writeable filehandles for inode\n");
1954                 rc = -EIO;
1955         }
1956
1957         kunmap(page);
1958         return rc;
1959 }
1960
1961 static struct cifs_writedata *
1962 wdata_alloc_and_fillpages(pgoff_t tofind, struct address_space *mapping,
1963                           pgoff_t end, pgoff_t *index,
1964                           unsigned int *found_pages)
1965 {
1966         struct cifs_writedata *wdata;
1967
1968         wdata = cifs_writedata_alloc((unsigned int)tofind,
1969                                      cifs_writev_complete);
1970         if (!wdata)
1971                 return NULL;
1972
1973         *found_pages = find_get_pages_range_tag(mapping, index, end,
1974                                 PAGECACHE_TAG_DIRTY, tofind, wdata->pages);
1975         return wdata;
1976 }
1977
1978 static unsigned int
1979 wdata_prepare_pages(struct cifs_writedata *wdata, unsigned int found_pages,
1980                     struct address_space *mapping,
1981                     struct writeback_control *wbc,
1982                     pgoff_t end, pgoff_t *index, pgoff_t *next, bool *done)
1983 {
1984         unsigned int nr_pages = 0, i;
1985         struct page *page;
1986
1987         for (i = 0; i < found_pages; i++) {
1988                 page = wdata->pages[i];
1989                 /*
1990                  * At this point we hold neither mapping->tree_lock nor
1991                  * lock on the page itself: the page may be truncated or
1992                  * invalidated (changing page->mapping to NULL), or even
1993                  * swizzled back from swapper_space to tmpfs file
1994                  * mapping
1995                  */
1996
1997                 if (nr_pages == 0)
1998                         lock_page(page);
1999                 else if (!trylock_page(page))
2000                         break;
2001
2002                 if (unlikely(page->mapping != mapping)) {
2003                         unlock_page(page);
2004                         break;
2005                 }
2006
2007                 if (!wbc->range_cyclic && page->index > end) {
2008                         *done = true;
2009                         unlock_page(page);
2010                         break;
2011                 }
2012
2013                 if (*next && (page->index != *next)) {
2014                         /* Not next consecutive page */
2015                         unlock_page(page);
2016                         break;
2017                 }
2018
2019                 if (wbc->sync_mode != WB_SYNC_NONE)
2020                         wait_on_page_writeback(page);
2021
2022                 if (PageWriteback(page) ||
2023                                 !clear_page_dirty_for_io(page)) {
2024                         unlock_page(page);
2025                         break;
2026                 }
2027
2028                 /*
2029                  * This actually clears the dirty bit in the radix tree.
2030                  * See cifs_writepage() for more commentary.
2031                  */
2032                 set_page_writeback(page);
2033                 if (page_offset(page) >= i_size_read(mapping->host)) {
2034                         *done = true;
2035                         unlock_page(page);
2036                         end_page_writeback(page);
2037                         break;
2038                 }
2039
2040                 wdata->pages[i] = page;
2041                 *next = page->index + 1;
2042                 ++nr_pages;
2043         }
2044
2045         /* reset index to refind any pages skipped */
2046         if (nr_pages == 0)
2047                 *index = wdata->pages[0]->index + 1;
2048
2049         /* put any pages we aren't going to use */
2050         for (i = nr_pages; i < found_pages; i++) {
2051                 put_page(wdata->pages[i]);
2052                 wdata->pages[i] = NULL;
2053         }
2054
2055         return nr_pages;
2056 }
2057
2058 static int
2059 wdata_send_pages(struct cifs_writedata *wdata, unsigned int nr_pages,
2060                  struct address_space *mapping, struct writeback_control *wbc)
2061 {
2062         int rc = 0;
2063         struct TCP_Server_Info *server;
2064         unsigned int i;
2065
2066         wdata->sync_mode = wbc->sync_mode;
2067         wdata->nr_pages = nr_pages;
2068         wdata->offset = page_offset(wdata->pages[0]);
2069         wdata->pagesz = PAGE_SIZE;
2070         wdata->tailsz = min(i_size_read(mapping->host) -
2071                         page_offset(wdata->pages[nr_pages - 1]),
2072                         (loff_t)PAGE_SIZE);
2073         wdata->bytes = ((nr_pages - 1) * PAGE_SIZE) + wdata->tailsz;
2074
2075         if (wdata->cfile != NULL)
2076                 cifsFileInfo_put(wdata->cfile);
2077         wdata->cfile = find_writable_file(CIFS_I(mapping->host), false);
2078         if (!wdata->cfile) {
2079                 cifs_dbg(VFS, "No writable handles for inode\n");
2080                 rc = -EBADF;
2081         } else {
2082                 wdata->pid = wdata->cfile->pid;
2083                 server = tlink_tcon(wdata->cfile->tlink)->ses->server;
2084                 rc = server->ops->async_writev(wdata, cifs_writedata_release);
2085         }
2086
2087         for (i = 0; i < nr_pages; ++i)
2088                 unlock_page(wdata->pages[i]);
2089
2090         return rc;
2091 }
2092
2093 static int cifs_writepages(struct address_space *mapping,
2094                            struct writeback_control *wbc)
2095 {
2096         struct cifs_sb_info *cifs_sb = CIFS_SB(mapping->host->i_sb);
2097         struct TCP_Server_Info *server;
2098         bool done = false, scanned = false, range_whole = false;
2099         pgoff_t end, index;
2100         struct cifs_writedata *wdata;
2101         int rc = 0;
2102
2103         /*
2104          * If wsize is smaller than the page cache size, default to writing
2105          * one page at a time via cifs_writepage
2106          */
2107         if (cifs_sb->wsize < PAGE_SIZE)
2108                 return generic_writepages(mapping, wbc);
2109
2110         if (wbc->range_cyclic) {
2111                 index = mapping->writeback_index; /* Start from prev offset */
2112                 end = -1;
2113         } else {
2114                 index = wbc->range_start >> PAGE_SHIFT;
2115                 end = wbc->range_end >> PAGE_SHIFT;
2116                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2117                         range_whole = true;
2118                 scanned = true;
2119         }
2120         server = cifs_sb_master_tcon(cifs_sb)->ses->server;
2121 retry:
2122         while (!done && index <= end) {
2123                 unsigned int i, nr_pages, found_pages, wsize, credits;
2124                 pgoff_t next = 0, tofind, saved_index = index;
2125
2126                 rc = server->ops->wait_mtu_credits(server, cifs_sb->wsize,
2127                                                    &wsize, &credits);
2128                 if (rc)
2129                         break;
2130
2131                 tofind = min((wsize / PAGE_SIZE) - 1, end - index) + 1;
2132
2133                 wdata = wdata_alloc_and_fillpages(tofind, mapping, end, &index,
2134                                                   &found_pages);
2135                 if (!wdata) {
2136                         rc = -ENOMEM;
2137                         add_credits_and_wake_if(server, credits, 0);
2138                         break;
2139                 }
2140
2141                 if (found_pages == 0) {
2142                         kref_put(&wdata->refcount, cifs_writedata_release);
2143                         add_credits_and_wake_if(server, credits, 0);
2144                         break;
2145                 }
2146
2147                 nr_pages = wdata_prepare_pages(wdata, found_pages, mapping, wbc,
2148                                                end, &index, &next, &done);
2149
2150                 /* nothing to write? */
2151                 if (nr_pages == 0) {
2152                         kref_put(&wdata->refcount, cifs_writedata_release);
2153                         add_credits_and_wake_if(server, credits, 0);
2154                         continue;
2155                 }
2156
2157                 wdata->credits = credits;
2158
2159                 rc = wdata_send_pages(wdata, nr_pages, mapping, wbc);
2160
2161                 /* send failure -- clean up the mess */
2162                 if (rc != 0) {
2163                         add_credits_and_wake_if(server, wdata->credits, 0);
2164                         for (i = 0; i < nr_pages; ++i) {
2165                                 if (rc == -EAGAIN)
2166                                         redirty_page_for_writepage(wbc,
2167                                                            wdata->pages[i]);
2168                                 else
2169                                         SetPageError(wdata->pages[i]);
2170                                 end_page_writeback(wdata->pages[i]);
2171                                 put_page(wdata->pages[i]);
2172                         }
2173                         if (rc != -EAGAIN)
2174                                 mapping_set_error(mapping, rc);
2175                 }
2176                 kref_put(&wdata->refcount, cifs_writedata_release);
2177
2178                 if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN) {
2179                         index = saved_index;
2180                         continue;
2181                 }
2182
2183                 wbc->nr_to_write -= nr_pages;
2184                 if (wbc->nr_to_write <= 0)
2185                         done = true;
2186
2187                 index = next;
2188         }
2189
2190         if (!scanned && !done) {
2191                 /*
2192                  * We hit the last page and there is more work to be done: wrap
2193                  * back to the start of the file
2194                  */
2195                 scanned = true;
2196                 index = 0;
2197                 goto retry;
2198         }
2199
2200         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2201                 mapping->writeback_index = index;
2202
2203         return rc;
2204 }
2205
2206 static int
2207 cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
2208 {
2209         int rc;
2210         unsigned int xid;
2211
2212         xid = get_xid();
2213 /* BB add check for wbc flags */
2214         get_page(page);
2215         if (!PageUptodate(page))
2216                 cifs_dbg(FYI, "ppw - page not up to date\n");
2217
2218         /*
2219          * Set the "writeback" flag, and clear "dirty" in the radix tree.
2220          *
2221          * A writepage() implementation always needs to do either this,
2222          * or re-dirty the page with "redirty_page_for_writepage()" in
2223          * the case of a failure.
2224          *
2225          * Just unlocking the page will cause the radix tree tag-bits
2226          * to fail to update with the state of the page correctly.
2227          */
2228         set_page_writeback(page);
2229 retry_write:
2230         rc = cifs_partialpagewrite(page, 0, PAGE_SIZE);
2231         if (rc == -EAGAIN) {
2232                 if (wbc->sync_mode == WB_SYNC_ALL)
2233                         goto retry_write;
2234                 redirty_page_for_writepage(wbc, page);
2235         } else if (rc != 0) {
2236                 SetPageError(page);
2237                 mapping_set_error(page->mapping, rc);
2238         } else {
2239                 SetPageUptodate(page);
2240         }
2241         end_page_writeback(page);
2242         put_page(page);
2243         free_xid(xid);
2244         return rc;
2245 }
2246
2247 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
2248 {
2249         int rc = cifs_writepage_locked(page, wbc);
2250         unlock_page(page);
2251         return rc;
2252 }
2253
2254 static int cifs_write_end(struct file *file, struct address_space *mapping,
2255                         loff_t pos, unsigned len, unsigned copied,
2256                         struct page *page, void *fsdata)
2257 {
2258         int rc;
2259         struct inode *inode = mapping->host;
2260         struct cifsFileInfo *cfile = file->private_data;
2261         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
2262         __u32 pid;
2263
2264         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2265                 pid = cfile->pid;
2266         else
2267                 pid = current->tgid;
2268
2269         cifs_dbg(FYI, "write_end for page %p from pos %lld with %d bytes\n",
2270                  page, pos, copied);
2271
2272         if (PageChecked(page)) {
2273                 if (copied == len)
2274                         SetPageUptodate(page);
2275                 ClearPageChecked(page);
2276         } else if (!PageUptodate(page) && copied == PAGE_SIZE)
2277                 SetPageUptodate(page);
2278
2279         if (!PageUptodate(page)) {
2280                 char *page_data;
2281                 unsigned offset = pos & (PAGE_SIZE - 1);
2282                 unsigned int xid;
2283
2284                 xid = get_xid();
2285                 /* this is probably better than directly calling
2286                    partialpage_write since in this function the file handle is
2287                    known which we might as well leverage */
2288                 /* BB check if anything else missing out of ppw
2289                    such as updating last write time */
2290                 page_data = kmap(page);
2291                 rc = cifs_write(cfile, pid, page_data + offset, copied, &pos);
2292                 /* if (rc < 0) should we set writebehind rc? */
2293                 kunmap(page);
2294
2295                 free_xid(xid);
2296         } else {
2297                 rc = copied;
2298                 pos += copied;
2299                 set_page_dirty(page);
2300         }
2301
2302         if (rc > 0) {
2303                 spin_lock(&inode->i_lock);
2304                 if (pos > inode->i_size)
2305                         i_size_write(inode, pos);
2306                 spin_unlock(&inode->i_lock);
2307         }
2308
2309         unlock_page(page);
2310         put_page(page);
2311
2312         return rc;
2313 }
2314
2315 int cifs_strict_fsync(struct file *file, loff_t start, loff_t end,
2316                       int datasync)
2317 {
2318         unsigned int xid;
2319         int rc = 0;
2320         struct cifs_tcon *tcon;
2321         struct TCP_Server_Info *server;
2322         struct cifsFileInfo *smbfile = file->private_data;
2323         struct inode *inode = file_inode(file);
2324         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2325
2326         rc = file_write_and_wait_range(file, start, end);
2327         if (rc)
2328                 return rc;
2329         inode_lock(inode);
2330
2331         xid = get_xid();
2332
2333         cifs_dbg(FYI, "Sync file - name: %pD datasync: 0x%x\n",
2334                  file, datasync);
2335
2336         if (!CIFS_CACHE_READ(CIFS_I(inode))) {
2337                 rc = cifs_zap_mapping(inode);
2338                 if (rc) {
2339                         cifs_dbg(FYI, "rc: %d during invalidate phase\n", rc);
2340                         rc = 0; /* don't care about it in fsync */
2341                 }
2342         }
2343
2344         tcon = tlink_tcon(smbfile->tlink);
2345         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2346                 server = tcon->ses->server;
2347                 if (server->ops->flush)
2348                         rc = server->ops->flush(xid, tcon, &smbfile->fid);
2349                 else
2350                         rc = -ENOSYS;
2351         }
2352
2353         free_xid(xid);
2354         inode_unlock(inode);
2355         return rc;
2356 }
2357
2358 int cifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2359 {
2360         unsigned int xid;
2361         int rc = 0;
2362         struct cifs_tcon *tcon;
2363         struct TCP_Server_Info *server;
2364         struct cifsFileInfo *smbfile = file->private_data;
2365         struct cifs_sb_info *cifs_sb = CIFS_FILE_SB(file);
2366         struct inode *inode = file->f_mapping->host;
2367
2368         rc = file_write_and_wait_range(file, start, end);
2369         if (rc)
2370                 return rc;
2371         inode_lock(inode);
2372
2373         xid = get_xid();
2374
2375         cifs_dbg(FYI, "Sync file - name: %pD datasync: 0x%x\n",
2376                  file, datasync);
2377
2378         tcon = tlink_tcon(smbfile->tlink);
2379         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2380                 server = tcon->ses->server;
2381                 if (server->ops->flush)
2382                         rc = server->ops->flush(xid, tcon, &smbfile->fid);
2383                 else
2384                         rc = -ENOSYS;
2385         }
2386
2387         free_xid(xid);
2388         inode_unlock(inode);
2389         return rc;
2390 }
2391
2392 /*
2393  * As file closes, flush all cached write data for this inode checking
2394  * for write behind errors.
2395  */
2396 int cifs_flush(struct file *file, fl_owner_t id)
2397 {
2398         struct inode *inode = file_inode(file);
2399         int rc = 0;
2400
2401         if (file->f_mode & FMODE_WRITE)
2402                 rc = filemap_write_and_wait(inode->i_mapping);
2403
2404         cifs_dbg(FYI, "Flush inode %p file %p rc %d\n", inode, file, rc);
2405
2406         return rc;
2407 }
2408
2409 static int
2410 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
2411 {
2412         int rc = 0;
2413         unsigned long i;
2414
2415         for (i = 0; i < num_pages; i++) {
2416                 pages[i] = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2417                 if (!pages[i]) {
2418                         /*
2419                          * save number of pages we have already allocated and
2420                          * return with ENOMEM error
2421                          */
2422                         num_pages = i;
2423                         rc = -ENOMEM;
2424                         break;
2425                 }
2426         }
2427
2428         if (rc) {
2429                 for (i = 0; i < num_pages; i++)
2430                         put_page(pages[i]);
2431         }
2432         return rc;
2433 }
2434
2435 static inline
2436 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
2437 {
2438         size_t num_pages;
2439         size_t clen;
2440
2441         clen = min_t(const size_t, len, wsize);
2442         num_pages = DIV_ROUND_UP(clen, PAGE_SIZE);
2443
2444         if (cur_len)
2445                 *cur_len = clen;
2446
2447         return num_pages;
2448 }
2449
2450 static void
2451 cifs_uncached_writedata_release(struct kref *refcount)
2452 {
2453         int i;
2454         struct cifs_writedata *wdata = container_of(refcount,
2455                                         struct cifs_writedata, refcount);
2456
2457         kref_put(&wdata->ctx->refcount, cifs_aio_ctx_release);
2458         for (i = 0; i < wdata->nr_pages; i++)
2459                 put_page(wdata->pages[i]);
2460         cifs_writedata_release(refcount);
2461 }
2462
2463 static void collect_uncached_write_data(struct cifs_aio_ctx *ctx);
2464
2465 static void
2466 cifs_uncached_writev_complete(struct work_struct *work)
2467 {
2468         struct cifs_writedata *wdata = container_of(work,
2469                                         struct cifs_writedata, work);
2470         struct inode *inode = d_inode(wdata->cfile->dentry);
2471         struct cifsInodeInfo *cifsi = CIFS_I(inode);
2472
2473         spin_lock(&inode->i_lock);
2474         cifs_update_eof(cifsi, wdata->offset, wdata->bytes);
2475         if (cifsi->server_eof > inode->i_size)
2476                 i_size_write(inode, cifsi->server_eof);
2477         spin_unlock(&inode->i_lock);
2478
2479         complete(&wdata->done);
2480         collect_uncached_write_data(wdata->ctx);
2481         /* the below call can possibly free the last ref to aio ctx */
2482         kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2483 }
2484
2485 static int
2486 wdata_fill_from_iovec(struct cifs_writedata *wdata, struct iov_iter *from,
2487                       size_t *len, unsigned long *num_pages)
2488 {
2489         size_t save_len, copied, bytes, cur_len = *len;
2490         unsigned long i, nr_pages = *num_pages;
2491
2492         save_len = cur_len;
2493         for (i = 0; i < nr_pages; i++) {
2494                 bytes = min_t(const size_t, cur_len, PAGE_SIZE);
2495                 copied = copy_page_from_iter(wdata->pages[i], 0, bytes, from);
2496                 cur_len -= copied;
2497                 /*
2498                  * If we didn't copy as much as we expected, then that
2499                  * may mean we trod into an unmapped area. Stop copying
2500                  * at that point. On the next pass through the big
2501                  * loop, we'll likely end up getting a zero-length
2502                  * write and bailing out of it.
2503                  */
2504                 if (copied < bytes)
2505                         break;
2506         }
2507         cur_len = save_len - cur_len;
2508         *len = cur_len;
2509
2510         /*
2511          * If we have no data to send, then that probably means that
2512          * the copy above failed altogether. That's most likely because
2513          * the address in the iovec was bogus. Return -EFAULT and let
2514          * the caller free anything we allocated and bail out.
2515          */
2516         if (!cur_len)
2517                 return -EFAULT;
2518
2519         /*
2520          * i + 1 now represents the number of pages we actually used in
2521          * the copy phase above.
2522          */
2523         *num_pages = i + 1;
2524         return 0;
2525 }
2526
2527 static int
2528 cifs_write_from_iter(loff_t offset, size_t len, struct iov_iter *from,
2529                      struct cifsFileInfo *open_file,
2530                      struct cifs_sb_info *cifs_sb, struct list_head *wdata_list,
2531                      struct cifs_aio_ctx *ctx)
2532 {
2533         int rc = 0;
2534         size_t cur_len;
2535         unsigned long nr_pages, num_pages, i;
2536         struct cifs_writedata *wdata;
2537         struct iov_iter saved_from = *from;
2538         loff_t saved_offset = offset;
2539         pid_t pid;
2540         struct TCP_Server_Info *server;
2541
2542         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2543                 pid = open_file->pid;
2544         else
2545                 pid = current->tgid;
2546
2547         server = tlink_tcon(open_file->tlink)->ses->server;
2548
2549         do {
2550                 unsigned int wsize, credits;
2551
2552                 rc = server->ops->wait_mtu_credits(server, cifs_sb->wsize,
2553                                                    &wsize, &credits);
2554                 if (rc)
2555                         break;
2556
2557                 nr_pages = get_numpages(wsize, len, &cur_len);
2558                 wdata = cifs_writedata_alloc(nr_pages,
2559                                              cifs_uncached_writev_complete);
2560                 if (!wdata) {
2561                         rc = -ENOMEM;
2562                         add_credits_and_wake_if(server, credits, 0);
2563                         break;
2564                 }
2565
2566                 rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
2567                 if (rc) {
2568                         kfree(wdata);
2569                         add_credits_and_wake_if(server, credits, 0);
2570                         break;
2571                 }
2572
2573                 num_pages = nr_pages;
2574                 rc = wdata_fill_from_iovec(wdata, from, &cur_len, &num_pages);
2575                 if (rc) {
2576                         for (i = 0; i < nr_pages; i++)
2577                                 put_page(wdata->pages[i]);
2578                         kfree(wdata);
2579                         add_credits_and_wake_if(server, credits, 0);
2580                         break;
2581                 }
2582
2583                 /*
2584                  * Bring nr_pages down to the number of pages we actually used,
2585                  * and free any pages that we didn't use.
2586                  */
2587                 for ( ; nr_pages > num_pages; nr_pages--)
2588                         put_page(wdata->pages[nr_pages - 1]);
2589
2590                 wdata->sync_mode = WB_SYNC_ALL;
2591                 wdata->nr_pages = nr_pages;
2592                 wdata->offset = (__u64)offset;
2593                 wdata->cfile = cifsFileInfo_get(open_file);
2594                 wdata->pid = pid;
2595                 wdata->bytes = cur_len;
2596                 wdata->pagesz = PAGE_SIZE;
2597                 wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
2598                 wdata->credits = credits;
2599                 wdata->ctx = ctx;
2600                 kref_get(&ctx->refcount);
2601
2602                 if (!wdata->cfile->invalidHandle ||
2603                     !(rc = cifs_reopen_file(wdata->cfile, false)))
2604                         rc = server->ops->async_writev(wdata,
2605                                         cifs_uncached_writedata_release);
2606                 if (rc) {
2607                         add_credits_and_wake_if(server, wdata->credits, 0);
2608                         kref_put(&wdata->refcount,
2609                                  cifs_uncached_writedata_release);
2610                         if (rc == -EAGAIN) {
2611                                 *from = saved_from;
2612                                 iov_iter_advance(from, offset - saved_offset);
2613                                 continue;
2614                         }
2615                         break;
2616                 }
2617
2618                 list_add_tail(&wdata->list, wdata_list);
2619                 offset += cur_len;
2620                 len -= cur_len;
2621         } while (len > 0);
2622
2623         return rc;
2624 }
2625
2626 static void collect_uncached_write_data(struct cifs_aio_ctx *ctx)
2627 {
2628         struct cifs_writedata *wdata, *tmp;
2629         struct cifs_tcon *tcon;
2630         struct cifs_sb_info *cifs_sb;
2631         struct dentry *dentry = ctx->cfile->dentry;
2632         unsigned int i;
2633         int rc;
2634
2635         tcon = tlink_tcon(ctx->cfile->tlink);
2636         cifs_sb = CIFS_SB(dentry->d_sb);
2637
2638         mutex_lock(&ctx->aio_mutex);
2639
2640         if (list_empty(&ctx->list)) {
2641                 mutex_unlock(&ctx->aio_mutex);
2642                 return;
2643         }
2644
2645         rc = ctx->rc;
2646         /*
2647          * Wait for and collect replies for any successful sends in order of
2648          * increasing offset. Once an error is hit, then return without waiting
2649          * for any more replies.
2650          */
2651 restart_loop:
2652         list_for_each_entry_safe(wdata, tmp, &ctx->list, list) {
2653                 if (!rc) {
2654                         if (!try_wait_for_completion(&wdata->done)) {
2655                                 mutex_unlock(&ctx->aio_mutex);
2656                                 return;
2657                         }
2658
2659                         if (wdata->result)
2660                                 rc = wdata->result;
2661                         else
2662                                 ctx->total_len += wdata->bytes;
2663
2664                         /* resend call if it's a retryable error */
2665                         if (rc == -EAGAIN) {
2666                                 struct list_head tmp_list;
2667                                 struct iov_iter tmp_from = ctx->iter;
2668
2669                                 INIT_LIST_HEAD(&tmp_list);
2670                                 list_del_init(&wdata->list);
2671
2672                                 iov_iter_advance(&tmp_from,
2673                                                  wdata->offset - ctx->pos);
2674
2675                                 rc = cifs_write_from_iter(wdata->offset,
2676                                                 wdata->bytes, &tmp_from,
2677                                                 ctx->cfile, cifs_sb, &tmp_list,
2678                                                 ctx);
2679
2680                                 list_splice(&tmp_list, &ctx->list);
2681
2682                                 kref_put(&wdata->refcount,
2683                                          cifs_uncached_writedata_release);
2684                                 goto restart_loop;
2685                         }
2686                 }
2687                 list_del_init(&wdata->list);
2688                 kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2689         }
2690
2691         for (i = 0; i < ctx->npages; i++)
2692                 put_page(ctx->bv[i].bv_page);
2693
2694         cifs_stats_bytes_written(tcon, ctx->total_len);
2695         set_bit(CIFS_INO_INVALID_MAPPING, &CIFS_I(dentry->d_inode)->flags);
2696
2697         ctx->rc = (rc == 0) ? ctx->total_len : rc;
2698
2699         mutex_unlock(&ctx->aio_mutex);
2700
2701         if (ctx->iocb && ctx->iocb->ki_complete)
2702                 ctx->iocb->ki_complete(ctx->iocb, ctx->rc, 0);
2703         else
2704                 complete(&ctx->done);
2705 }
2706
2707 ssize_t cifs_user_writev(struct kiocb *iocb, struct iov_iter *from)
2708 {
2709         struct file *file = iocb->ki_filp;
2710         ssize_t total_written = 0;
2711         struct cifsFileInfo *cfile;
2712         struct cifs_tcon *tcon;
2713         struct cifs_sb_info *cifs_sb;
2714         struct cifs_aio_ctx *ctx;
2715         struct iov_iter saved_from = *from;
2716         int rc;
2717
2718         /*
2719          * BB - optimize the way when signing is disabled. We can drop this
2720          * extra memory-to-memory copying and use iovec buffers for constructing
2721          * write request.
2722          */
2723
2724         rc = generic_write_checks(iocb, from);
2725         if (rc <= 0)
2726                 return rc;
2727
2728         cifs_sb = CIFS_FILE_SB(file);
2729         cfile = file->private_data;
2730         tcon = tlink_tcon(cfile->tlink);
2731
2732         if (!tcon->ses->server->ops->async_writev)
2733                 return -ENOSYS;
2734
2735         ctx = cifs_aio_ctx_alloc();
2736         if (!ctx)
2737                 return -ENOMEM;
2738
2739         ctx->cfile = cifsFileInfo_get(cfile);
2740
2741         if (!is_sync_kiocb(iocb))
2742                 ctx->iocb = iocb;
2743
2744         ctx->pos = iocb->ki_pos;
2745
2746         rc = setup_aio_ctx_iter(ctx, from, WRITE);
2747         if (rc) {
2748                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
2749                 return rc;
2750         }
2751
2752         /* grab a lock here due to read response handlers can access ctx */
2753         mutex_lock(&ctx->aio_mutex);
2754
2755         rc = cifs_write_from_iter(iocb->ki_pos, ctx->len, &saved_from,
2756                                   cfile, cifs_sb, &ctx->list, ctx);
2757
2758         /*
2759          * If at least one write was successfully sent, then discard any rc
2760          * value from the later writes. If the other write succeeds, then
2761          * we'll end up returning whatever was written. If it fails, then
2762          * we'll get a new rc value from that.
2763          */
2764         if (!list_empty(&ctx->list))
2765                 rc = 0;
2766
2767         mutex_unlock(&ctx->aio_mutex);
2768
2769         if (rc) {
2770                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
2771                 return rc;
2772         }
2773
2774         if (!is_sync_kiocb(iocb)) {
2775                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
2776                 return -EIOCBQUEUED;
2777         }
2778
2779         rc = wait_for_completion_killable(&ctx->done);
2780         if (rc) {
2781                 mutex_lock(&ctx->aio_mutex);
2782                 ctx->rc = rc = -EINTR;
2783                 total_written = ctx->total_len;
2784                 mutex_unlock(&ctx->aio_mutex);
2785         } else {
2786                 rc = ctx->rc;
2787                 total_written = ctx->total_len;
2788         }
2789
2790         kref_put(&ctx->refcount, cifs_aio_ctx_release);
2791
2792         if (unlikely(!total_written))
2793                 return rc;
2794
2795         iocb->ki_pos += total_written;
2796         return total_written;
2797 }
2798
2799 static ssize_t
2800 cifs_writev(struct kiocb *iocb, struct iov_iter *from)
2801 {
2802         struct file *file = iocb->ki_filp;
2803         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
2804         struct inode *inode = file->f_mapping->host;
2805         struct cifsInodeInfo *cinode = CIFS_I(inode);
2806         struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
2807         ssize_t rc;
2808
2809         inode_lock(inode);
2810         /*
2811          * We need to hold the sem to be sure nobody modifies lock list
2812          * with a brlock that prevents writing.
2813          */
2814         down_read(&cinode->lock_sem);
2815
2816         rc = generic_write_checks(iocb, from);
2817         if (rc <= 0)
2818                 goto out;
2819
2820         if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(from),
2821                                      server->vals->exclusive_lock_type, NULL,
2822                                      CIFS_WRITE_OP))
2823                 rc = __generic_file_write_iter(iocb, from);
2824         else
2825                 rc = -EACCES;
2826 out:
2827         up_read(&cinode->lock_sem);
2828         inode_unlock(inode);
2829
2830         if (rc > 0)
2831                 rc = generic_write_sync(iocb, rc);
2832         return rc;
2833 }
2834
2835 ssize_t
2836 cifs_strict_writev(struct kiocb *iocb, struct iov_iter *from)
2837 {
2838         struct inode *inode = file_inode(iocb->ki_filp);
2839         struct cifsInodeInfo *cinode = CIFS_I(inode);
2840         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2841         struct cifsFileInfo *cfile = (struct cifsFileInfo *)
2842                                                 iocb->ki_filp->private_data;
2843         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
2844         ssize_t written;
2845
2846         written = cifs_get_writer(cinode);
2847         if (written)
2848                 return written;
2849
2850         if (CIFS_CACHE_WRITE(cinode)) {
2851                 if (cap_unix(tcon->ses) &&
2852                 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability))
2853                   && ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0)) {
2854                         written = generic_file_write_iter(iocb, from);
2855                         goto out;
2856                 }
2857                 written = cifs_writev(iocb, from);
2858                 goto out;
2859         }
2860         /*
2861          * For non-oplocked files in strict cache mode we need to write the data
2862          * to the server exactly from the pos to pos+len-1 rather than flush all
2863          * affected pages because it may cause a error with mandatory locks on
2864          * these pages but not on the region from pos to ppos+len-1.
2865          */
2866         written = cifs_user_writev(iocb, from);
2867         if (written > 0 && CIFS_CACHE_READ(cinode)) {
2868                 /*
2869                  * Windows 7 server can delay breaking level2 oplock if a write
2870                  * request comes - break it on the client to prevent reading
2871                  * an old data.
2872                  */
2873                 cifs_zap_mapping(inode);
2874                 cifs_dbg(FYI, "Set no oplock for inode=%p after a write operation\n",
2875                          inode);
2876                 cinode->oplock = 0;
2877         }
2878 out:
2879         cifs_put_writer(cinode);
2880         return written;
2881 }
2882
2883 static struct cifs_readdata *
2884 cifs_readdata_alloc(unsigned int nr_pages, work_func_t complete)
2885 {
2886         struct cifs_readdata *rdata;
2887
2888         rdata = kzalloc(sizeof(*rdata) + (sizeof(struct page *) * nr_pages),
2889                         GFP_KERNEL);
2890         if (rdata != NULL) {
2891                 kref_init(&rdata->refcount);
2892                 INIT_LIST_HEAD(&rdata->list);
2893                 init_completion(&rdata->done);
2894                 INIT_WORK(&rdata->work, complete);
2895         }
2896
2897         return rdata;
2898 }
2899
2900 void
2901 cifs_readdata_release(struct kref *refcount)
2902 {
2903         struct cifs_readdata *rdata = container_of(refcount,
2904                                         struct cifs_readdata, refcount);
2905 #ifdef CONFIG_CIFS_SMB_DIRECT
2906         if (rdata->mr) {
2907                 smbd_deregister_mr(rdata->mr);
2908                 rdata->mr = NULL;
2909         }
2910 #endif
2911         if (rdata->cfile)
2912                 cifsFileInfo_put(rdata->cfile);
2913
2914         kfree(rdata);
2915 }
2916
2917 static int
2918 cifs_read_allocate_pages(struct cifs_readdata *rdata, unsigned int nr_pages)
2919 {
2920         int rc = 0;
2921         struct page *page;
2922         unsigned int i;
2923
2924         for (i = 0; i < nr_pages; i++) {
2925                 page = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2926                 if (!page) {
2927                         rc = -ENOMEM;
2928                         break;
2929                 }
2930                 rdata->pages[i] = page;
2931         }
2932
2933         if (rc) {
2934                 for (i = 0; i < nr_pages; i++) {
2935                         put_page(rdata->pages[i]);
2936                         rdata->pages[i] = NULL;
2937                 }
2938         }
2939         return rc;
2940 }
2941
2942 static void
2943 cifs_uncached_readdata_release(struct kref *refcount)
2944 {
2945         struct cifs_readdata *rdata = container_of(refcount,
2946                                         struct cifs_readdata, refcount);
2947         unsigned int i;
2948
2949         kref_put(&rdata->ctx->refcount, cifs_aio_ctx_release);
2950         for (i = 0; i < rdata->nr_pages; i++) {
2951                 put_page(rdata->pages[i]);
2952                 rdata->pages[i] = NULL;
2953         }
2954         cifs_readdata_release(refcount);
2955 }
2956
2957 /**
2958  * cifs_readdata_to_iov - copy data from pages in response to an iovec
2959  * @rdata:      the readdata response with list of pages holding data
2960  * @iter:       destination for our data
2961  *
2962  * This function copies data from a list of pages in a readdata response into
2963  * an array of iovecs. It will first calculate where the data should go
2964  * based on the info in the readdata and then copy the data into that spot.
2965  */
2966 static int
2967 cifs_readdata_to_iov(struct cifs_readdata *rdata, struct iov_iter *iter)
2968 {
2969         size_t remaining = rdata->got_bytes;
2970         unsigned int i;
2971
2972         for (i = 0; i < rdata->nr_pages; i++) {
2973                 struct page *page = rdata->pages[i];
2974                 size_t copy = min_t(size_t, remaining, PAGE_SIZE);
2975                 size_t written;
2976
2977                 if (unlikely(iter->type & ITER_PIPE)) {
2978                         void *addr = kmap_atomic(page);
2979
2980                         written = copy_to_iter(addr, copy, iter);
2981                         kunmap_atomic(addr);
2982                 } else
2983                         written = copy_page_to_iter(page, 0, copy, iter);
2984                 remaining -= written;
2985                 if (written < copy && iov_iter_count(iter) > 0)
2986                         break;
2987         }
2988         return remaining ? -EFAULT : 0;
2989 }
2990
2991 static void collect_uncached_read_data(struct cifs_aio_ctx *ctx);
2992
2993 static void
2994 cifs_uncached_readv_complete(struct work_struct *work)
2995 {
2996         struct cifs_readdata *rdata = container_of(work,
2997                                                 struct cifs_readdata, work);
2998
2999         complete(&rdata->done);
3000         collect_uncached_read_data(rdata->ctx);
3001         /* the below call can possibly free the last ref to aio ctx */
3002         kref_put(&rdata->refcount, cifs_uncached_readdata_release);
3003 }
3004
3005 static int
3006 uncached_fill_pages(struct TCP_Server_Info *server,
3007                     struct cifs_readdata *rdata, struct iov_iter *iter,
3008                     unsigned int len)
3009 {
3010         int result = 0;
3011         unsigned int i;
3012         unsigned int nr_pages = rdata->nr_pages;
3013
3014         rdata->got_bytes = 0;
3015         rdata->tailsz = PAGE_SIZE;
3016         for (i = 0; i < nr_pages; i++) {
3017                 struct page *page = rdata->pages[i];
3018                 size_t n;
3019
3020                 if (len <= 0) {
3021                         /* no need to hold page hostage */
3022                         rdata->pages[i] = NULL;
3023                         rdata->nr_pages--;
3024                         put_page(page);
3025                         continue;
3026                 }
3027                 n = len;
3028                 if (len >= PAGE_SIZE) {
3029                         /* enough data to fill the page */
3030                         n = PAGE_SIZE;
3031                         len -= n;
3032                 } else {
3033                         zero_user(page, len, PAGE_SIZE - len);
3034                         rdata->tailsz = len;
3035                         len = 0;
3036                 }
3037                 if (iter)
3038                         result = copy_page_from_iter(page, 0, n, iter);
3039 #ifdef CONFIG_CIFS_SMB_DIRECT
3040                 else if (rdata->mr)
3041                         result = n;
3042 #endif
3043                 else
3044                         result = cifs_read_page_from_socket(server, page, n);
3045                 if (result < 0)
3046                         break;
3047
3048                 rdata->got_bytes += result;
3049         }
3050
3051         return rdata->got_bytes > 0 && result != -ECONNABORTED ?
3052                                                 rdata->got_bytes : result;
3053 }
3054
3055 static int
3056 cifs_uncached_read_into_pages(struct TCP_Server_Info *server,
3057                               struct cifs_readdata *rdata, unsigned int len)
3058 {
3059         return uncached_fill_pages(server, rdata, NULL, len);
3060 }
3061
3062 static int
3063 cifs_uncached_copy_into_pages(struct TCP_Server_Info *server,
3064                               struct cifs_readdata *rdata,
3065                               struct iov_iter *iter)
3066 {
3067         return uncached_fill_pages(server, rdata, iter, iter->count);
3068 }
3069
3070 static int
3071 cifs_send_async_read(loff_t offset, size_t len, struct cifsFileInfo *open_file,
3072                      struct cifs_sb_info *cifs_sb, struct list_head *rdata_list,
3073                      struct cifs_aio_ctx *ctx)
3074 {
3075         struct cifs_readdata *rdata;
3076         unsigned int npages, rsize, credits;
3077         size_t cur_len;
3078         int rc;
3079         pid_t pid;
3080         struct TCP_Server_Info *server;
3081
3082         server = tlink_tcon(open_file->tlink)->ses->server;
3083
3084         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3085                 pid = open_file->pid;
3086         else
3087                 pid = current->tgid;
3088
3089         do {
3090                 rc = server->ops->wait_mtu_credits(server, cifs_sb->rsize,
3091                                                    &rsize, &credits);
3092                 if (rc)
3093                         break;
3094
3095                 cur_len = min_t(const size_t, len, rsize);
3096                 npages = DIV_ROUND_UP(cur_len, PAGE_SIZE);
3097
3098                 /* allocate a readdata struct */
3099                 rdata = cifs_readdata_alloc(npages,
3100                                             cifs_uncached_readv_complete);
3101                 if (!rdata) {
3102                         add_credits_and_wake_if(server, credits, 0);
3103                         rc = -ENOMEM;
3104                         break;
3105                 }
3106
3107                 rc = cifs_read_allocate_pages(rdata, npages);
3108                 if (rc)
3109                         goto error;
3110
3111                 rdata->cfile = cifsFileInfo_get(open_file);
3112                 rdata->nr_pages = npages;
3113                 rdata->offset = offset;
3114                 rdata->bytes = cur_len;
3115                 rdata->pid = pid;
3116                 rdata->pagesz = PAGE_SIZE;
3117                 rdata->read_into_pages = cifs_uncached_read_into_pages;
3118                 rdata->copy_into_pages = cifs_uncached_copy_into_pages;
3119                 rdata->credits = credits;
3120                 rdata->ctx = ctx;
3121                 kref_get(&ctx->refcount);
3122
3123                 if (!rdata->cfile->invalidHandle ||
3124                     !(rc = cifs_reopen_file(rdata->cfile, true)))
3125                         rc = server->ops->async_readv(rdata);
3126 error:
3127                 if (rc) {
3128                         add_credits_and_wake_if(server, rdata->credits, 0);
3129                         kref_put(&rdata->refcount,
3130                                  cifs_uncached_readdata_release);
3131                         if (rc == -EAGAIN)
3132                                 continue;
3133                         break;
3134                 }
3135
3136                 list_add_tail(&rdata->list, rdata_list);
3137                 offset += cur_len;
3138                 len -= cur_len;
3139         } while (len > 0);
3140
3141         return rc;
3142 }
3143
3144 static void
3145 collect_uncached_read_data(struct cifs_aio_ctx *ctx)
3146 {
3147         struct cifs_readdata *rdata, *tmp;
3148         struct iov_iter *to = &ctx->iter;
3149         struct cifs_sb_info *cifs_sb;
3150         struct cifs_tcon *tcon;
3151         unsigned int i;
3152         int rc;
3153
3154         tcon = tlink_tcon(ctx->cfile->tlink);
3155         cifs_sb = CIFS_SB(ctx->cfile->dentry->d_sb);
3156
3157         mutex_lock(&ctx->aio_mutex);
3158
3159         if (list_empty(&ctx->list)) {
3160                 mutex_unlock(&ctx->aio_mutex);
3161                 return;
3162         }
3163
3164         rc = ctx->rc;
3165         /* the loop below should proceed in the order of increasing offsets */
3166 again:
3167         list_for_each_entry_safe(rdata, tmp, &ctx->list, list) {
3168                 if (!rc) {
3169                         if (!try_wait_for_completion(&rdata->done)) {
3170                                 mutex_unlock(&ctx->aio_mutex);
3171                                 return;
3172                         }
3173
3174                         if (rdata->result == -EAGAIN) {
3175                                 /* resend call if it's a retryable error */
3176                                 struct list_head tmp_list;
3177                                 unsigned int got_bytes = rdata->got_bytes;
3178
3179                                 list_del_init(&rdata->list);
3180                                 INIT_LIST_HEAD(&tmp_list);
3181
3182                                 /*
3183                                  * Got a part of data and then reconnect has
3184                                  * happened -- fill the buffer and continue
3185                                  * reading.
3186                                  */
3187                                 if (got_bytes && got_bytes < rdata->bytes) {
3188                                         rc = cifs_readdata_to_iov(rdata, to);
3189                                         if (rc) {
3190                                                 kref_put(&rdata->refcount,
3191                                                 cifs_uncached_readdata_release);
3192                                                 continue;
3193                                         }
3194                                 }
3195
3196                                 rc = cifs_send_async_read(
3197                                                 rdata->offset + got_bytes,
3198                                                 rdata->bytes - got_bytes,
3199                                                 rdata->cfile, cifs_sb,
3200                                                 &tmp_list, ctx);
3201
3202                                 list_splice(&tmp_list, &ctx->list);
3203
3204                                 kref_put(&rdata->refcount,
3205                                          cifs_uncached_readdata_release);
3206                                 goto again;
3207                         } else if (rdata->result)
3208                                 rc = rdata->result;
3209                         else
3210                                 rc = cifs_readdata_to_iov(rdata, to);
3211
3212                         /* if there was a short read -- discard anything left */
3213                         if (rdata->got_bytes && rdata->got_bytes < rdata->bytes)
3214                                 rc = -ENODATA;
3215                 }
3216                 list_del_init(&rdata->list);
3217                 kref_put(&rdata->refcount, cifs_uncached_readdata_release);
3218         }
3219
3220         for (i = 0; i < ctx->npages; i++) {
3221                 if (ctx->should_dirty)
3222                         set_page_dirty(ctx->bv[i].bv_page);
3223                 put_page(ctx->bv[i].bv_page);
3224         }
3225
3226         ctx->total_len = ctx->len - iov_iter_count(to);
3227
3228         cifs_stats_bytes_read(tcon, ctx->total_len);
3229
3230         /* mask nodata case */
3231         if (rc == -ENODATA)
3232                 rc = 0;
3233
3234         ctx->rc = (rc == 0) ? ctx->total_len : rc;
3235
3236         mutex_unlock(&ctx->aio_mutex);
3237
3238         if (ctx->iocb && ctx->iocb->ki_complete)
3239                 ctx->iocb->ki_complete(ctx->iocb, ctx->rc, 0);
3240         else
3241                 complete(&ctx->done);
3242 }
3243
3244 ssize_t cifs_user_readv(struct kiocb *iocb, struct iov_iter *to)
3245 {
3246         struct file *file = iocb->ki_filp;
3247         ssize_t rc;
3248         size_t len;
3249         ssize_t total_read = 0;
3250         loff_t offset = iocb->ki_pos;
3251         struct cifs_sb_info *cifs_sb;
3252         struct cifs_tcon *tcon;
3253         struct cifsFileInfo *cfile;
3254         struct cifs_aio_ctx *ctx;
3255
3256         len = iov_iter_count(to);
3257         if (!len)
3258                 return 0;
3259
3260         cifs_sb = CIFS_FILE_SB(file);
3261         cfile = file->private_data;
3262         tcon = tlink_tcon(cfile->tlink);
3263
3264         if (!tcon->ses->server->ops->async_readv)
3265                 return -ENOSYS;
3266
3267         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3268                 cifs_dbg(FYI, "attempting read on write only file instance\n");
3269
3270         ctx = cifs_aio_ctx_alloc();
3271         if (!ctx)
3272                 return -ENOMEM;
3273
3274         ctx->cfile = cifsFileInfo_get(cfile);
3275
3276         if (!is_sync_kiocb(iocb))
3277                 ctx->iocb = iocb;
3278
3279         if (to->type == ITER_IOVEC)
3280                 ctx->should_dirty = true;
3281
3282         rc = setup_aio_ctx_iter(ctx, to, READ);
3283         if (rc) {
3284                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
3285                 return rc;
3286         }
3287
3288         len = ctx->len;
3289
3290         /* grab a lock here due to read response handlers can access ctx */
3291         mutex_lock(&ctx->aio_mutex);
3292
3293         rc = cifs_send_async_read(offset, len, cfile, cifs_sb, &ctx->list, ctx);
3294
3295         /* if at least one read request send succeeded, then reset rc */
3296         if (!list_empty(&ctx->list))
3297                 rc = 0;
3298
3299         mutex_unlock(&ctx->aio_mutex);
3300
3301         if (rc) {
3302                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
3303                 return rc;
3304         }
3305
3306         if (!is_sync_kiocb(iocb)) {
3307                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
3308                 return -EIOCBQUEUED;
3309         }
3310
3311         rc = wait_for_completion_killable(&ctx->done);
3312         if (rc) {
3313                 mutex_lock(&ctx->aio_mutex);
3314                 ctx->rc = rc = -EINTR;
3315                 total_read = ctx->total_len;
3316                 mutex_unlock(&ctx->aio_mutex);
3317         } else {
3318                 rc = ctx->rc;
3319                 total_read = ctx->total_len;
3320         }
3321
3322         kref_put(&ctx->refcount, cifs_aio_ctx_release);
3323
3324         if (total_read) {
3325                 iocb->ki_pos += total_read;
3326                 return total_read;
3327         }
3328         return rc;
3329 }
3330
3331 ssize_t
3332 cifs_strict_readv(struct kiocb *iocb, struct iov_iter *to)
3333 {
3334         struct inode *inode = file_inode(iocb->ki_filp);
3335         struct cifsInodeInfo *cinode = CIFS_I(inode);
3336         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
3337         struct cifsFileInfo *cfile = (struct cifsFileInfo *)
3338                                                 iocb->ki_filp->private_data;
3339         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
3340         int rc = -EACCES;
3341
3342         /*
3343          * In strict cache mode we need to read from the server all the time
3344          * if we don't have level II oplock because the server can delay mtime
3345          * change - so we can't make a decision about inode invalidating.
3346          * And we can also fail with pagereading if there are mandatory locks
3347          * on pages affected by this read but not on the region from pos to
3348          * pos+len-1.
3349          */
3350         if (!CIFS_CACHE_READ(cinode))
3351                 return cifs_user_readv(iocb, to);
3352
3353         if (cap_unix(tcon->ses) &&
3354             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
3355             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
3356                 return generic_file_read_iter(iocb, to);
3357
3358         /*
3359          * We need to hold the sem to be sure nobody modifies lock list
3360          * with a brlock that prevents reading.
3361          */
3362         down_read(&cinode->lock_sem);
3363         if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(to),
3364                                      tcon->ses->server->vals->shared_lock_type,
3365                                      NULL, CIFS_READ_OP))
3366                 rc = generic_file_read_iter(iocb, to);
3367         up_read(&cinode->lock_sem);
3368         return rc;
3369 }
3370
3371 static ssize_t
3372 cifs_read(struct file *file, char *read_data, size_t read_size, loff_t *offset)
3373 {
3374         int rc = -EACCES;
3375         unsigned int bytes_read = 0;
3376         unsigned int total_read;
3377         unsigned int current_read_size;
3378         unsigned int rsize;
3379         struct cifs_sb_info *cifs_sb;
3380         struct cifs_tcon *tcon;
3381         struct TCP_Server_Info *server;
3382         unsigned int xid;
3383         char *cur_offset;
3384         struct cifsFileInfo *open_file;
3385         struct cifs_io_parms io_parms;
3386         int buf_type = CIFS_NO_BUFFER;
3387         __u32 pid;
3388
3389         xid = get_xid();
3390         cifs_sb = CIFS_FILE_SB(file);
3391
3392         /* FIXME: set up handlers for larger reads and/or convert to async */
3393         rsize = min_t(unsigned int, cifs_sb->rsize, CIFSMaxBufSize);
3394
3395         if (file->private_data == NULL) {
3396                 rc = -EBADF;
3397                 free_xid(xid);
3398                 return rc;
3399         }
3400         open_file = file->private_data;
3401         tcon = tlink_tcon(open_file->tlink);
3402         server = tcon->ses->server;
3403
3404         if (!server->ops->sync_read) {
3405                 free_xid(xid);
3406                 return -ENOSYS;
3407         }
3408
3409         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3410                 pid = open_file->pid;
3411         else
3412                 pid = current->tgid;
3413
3414         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3415                 cifs_dbg(FYI, "attempting read on write only file instance\n");
3416
3417         for (total_read = 0, cur_offset = read_data; read_size > total_read;
3418              total_read += bytes_read, cur_offset += bytes_read) {
3419                 do {
3420                         current_read_size = min_t(uint, read_size - total_read,
3421                                                   rsize);
3422                         /*
3423                          * For windows me and 9x we do not want to request more
3424                          * than it negotiated since it will refuse the read
3425                          * then.
3426                          */
3427                         if ((tcon->ses) && !(tcon->ses->capabilities &
3428                                 tcon->ses->server->vals->cap_large_files)) {
3429                                 current_read_size = min_t(uint,
3430                                         current_read_size, CIFSMaxBufSize);
3431                         }
3432                         if (open_file->invalidHandle) {
3433                                 rc = cifs_reopen_file(open_file, true);
3434                                 if (rc != 0)
3435                                         break;
3436                         }
3437                         io_parms.pid = pid;
3438                         io_parms.tcon = tcon;
3439                         io_parms.offset = *offset;
3440                         io_parms.length = current_read_size;
3441                         rc = server->ops->sync_read(xid, &open_file->fid, &io_parms,
3442                                                     &bytes_read, &cur_offset,
3443                                                     &buf_type);
3444                 } while (rc == -EAGAIN);
3445
3446                 if (rc || (bytes_read == 0)) {
3447                         if (total_read) {
3448                                 break;
3449                         } else {
3450                                 free_xid(xid);
3451                                 return rc;
3452                         }
3453                 } else {
3454                         cifs_stats_bytes_read(tcon, total_read);
3455                         *offset += bytes_read;
3456                 }
3457         }
3458         free_xid(xid);
3459         return total_read;
3460 }
3461
3462 /*
3463  * If the page is mmap'ed into a process' page tables, then we need to make
3464  * sure that it doesn't change while being written back.
3465  */
3466 static int
3467 cifs_page_mkwrite(struct vm_fault *vmf)
3468 {
3469         struct page *page = vmf->page;
3470
3471         lock_page(page);
3472         return VM_FAULT_LOCKED;
3473 }
3474
3475 static const struct vm_operations_struct cifs_file_vm_ops = {
3476         .fault = filemap_fault,
3477         .map_pages = filemap_map_pages,
3478         .page_mkwrite = cifs_page_mkwrite,
3479 };
3480
3481 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
3482 {
3483         int xid, rc = 0;
3484         struct inode *inode = file_inode(file);
3485
3486         xid = get_xid();
3487
3488         if (!CIFS_CACHE_READ(CIFS_I(inode)))
3489                 rc = cifs_zap_mapping(inode);
3490         if (!rc)
3491                 rc = generic_file_mmap(file, vma);
3492         if (!rc)
3493                 vma->vm_ops = &cifs_file_vm_ops;
3494
3495         free_xid(xid);
3496         return rc;
3497 }
3498
3499 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
3500 {
3501         int rc, xid;
3502
3503         xid = get_xid();
3504
3505         rc = cifs_revalidate_file(file);
3506         if (rc)
3507                 cifs_dbg(FYI, "Validation prior to mmap failed, error=%d\n",
3508                          rc);
3509         if (!rc)
3510                 rc = generic_file_mmap(file, vma);
3511         if (!rc)
3512                 vma->vm_ops = &cifs_file_vm_ops;
3513
3514         free_xid(xid);
3515         return rc;
3516 }
3517
3518 static void
3519 cifs_readv_complete(struct work_struct *work)
3520 {
3521         unsigned int i, got_bytes;
3522         struct cifs_readdata *rdata = container_of(work,
3523                                                 struct cifs_readdata, work);
3524
3525         got_bytes = rdata->got_bytes;
3526         for (i = 0; i < rdata->nr_pages; i++) {
3527                 struct page *page = rdata->pages[i];
3528
3529                 lru_cache_add_file(page);
3530
3531                 if (rdata->result == 0 ||
3532                     (rdata->result == -EAGAIN && got_bytes)) {
3533                         flush_dcache_page(page);
3534                         SetPageUptodate(page);
3535                 }
3536
3537                 unlock_page(page);
3538
3539                 if (rdata->result == 0 ||
3540                     (rdata->result == -EAGAIN && got_bytes))
3541                         cifs_readpage_to_fscache(rdata->mapping->host, page);
3542
3543                 got_bytes -= min_t(unsigned int, PAGE_SIZE, got_bytes);
3544
3545                 put_page(page);
3546                 rdata->pages[i] = NULL;
3547         }
3548         kref_put(&rdata->refcount, cifs_readdata_release);
3549 }
3550
3551 static int
3552 readpages_fill_pages(struct TCP_Server_Info *server,
3553                      struct cifs_readdata *rdata, struct iov_iter *iter,
3554                      unsigned int len)
3555 {
3556         int result = 0;
3557         unsigned int i;
3558         u64 eof;
3559         pgoff_t eof_index;
3560         unsigned int nr_pages = rdata->nr_pages;
3561
3562         /* determine the eof that the server (probably) has */
3563         eof = CIFS_I(rdata->mapping->host)->server_eof;
3564         eof_index = eof ? (eof - 1) >> PAGE_SHIFT : 0;
3565         cifs_dbg(FYI, "eof=%llu eof_index=%lu\n", eof, eof_index);
3566
3567         rdata->got_bytes = 0;
3568         rdata->tailsz = PAGE_SIZE;
3569         for (i = 0; i < nr_pages; i++) {
3570                 struct page *page = rdata->pages[i];
3571                 size_t n = PAGE_SIZE;
3572
3573                 if (len >= PAGE_SIZE) {
3574                         len -= PAGE_SIZE;
3575                 } else if (len > 0) {
3576                         /* enough for partial page, fill and zero the rest */
3577                         zero_user(page, len, PAGE_SIZE - len);
3578                         n = rdata->tailsz = len;
3579                         len = 0;
3580                 } else if (page->index > eof_index) {
3581                         /*
3582                          * The VFS will not try to do readahead past the
3583                          * i_size, but it's possible that we have outstanding
3584                          * writes with gaps in the middle and the i_size hasn't
3585                          * caught up yet. Populate those with zeroed out pages
3586                          * to prevent the VFS from repeatedly attempting to
3587                          * fill them until the writes are flushed.
3588                          */
3589                         zero_user(page, 0, PAGE_SIZE);
3590                         lru_cache_add_file(page);
3591                         flush_dcache_page(page);
3592                         SetPageUptodate(page);
3593                         unlock_page(page);
3594                         put_page(page);
3595                         rdata->pages[i] = NULL;
3596                         rdata->nr_pages--;
3597                         continue;
3598                 } else {
3599                         /* no need to hold page hostage */
3600                         lru_cache_add_file(page);
3601                         unlock_page(page);
3602                         put_page(page);
3603                         rdata->pages[i] = NULL;
3604                         rdata->nr_pages--;
3605                         continue;
3606                 }
3607
3608                 if (iter)
3609                         result = copy_page_from_iter(page, 0, n, iter);
3610 #ifdef CONFIG_CIFS_SMB_DIRECT
3611                 else if (rdata->mr)
3612                         result = n;
3613 #endif
3614                 else
3615                         result = cifs_read_page_from_socket(server, page, n);
3616                 if (result < 0)
3617                         break;
3618
3619                 rdata->got_bytes += result;
3620         }
3621
3622         return rdata->got_bytes > 0 && result != -ECONNABORTED ?
3623                                                 rdata->got_bytes : result;
3624 }
3625
3626 static int
3627 cifs_readpages_read_into_pages(struct TCP_Server_Info *server,
3628                                struct cifs_readdata *rdata, unsigned int len)
3629 {
3630         return readpages_fill_pages(server, rdata, NULL, len);
3631 }
3632
3633 static int
3634 cifs_readpages_copy_into_pages(struct TCP_Server_Info *server,
3635                                struct cifs_readdata *rdata,
3636                                struct iov_iter *iter)
3637 {
3638         return readpages_fill_pages(server, rdata, iter, iter->count);
3639 }
3640
3641 static int
3642 readpages_get_pages(struct address_space *mapping, struct list_head *page_list,
3643                     unsigned int rsize, struct list_head *tmplist,
3644                     unsigned int *nr_pages, loff_t *offset, unsigned int *bytes)
3645 {
3646         struct page *page, *tpage;
3647         unsigned int expected_index;
3648         int rc;
3649         gfp_t gfp = readahead_gfp_mask(mapping);
3650
3651         INIT_LIST_HEAD(tmplist);
3652
3653         page = list_entry(page_list->prev, struct page, lru);
3654
3655         /*
3656          * Lock the page and put it in the cache. Since no one else
3657          * should have access to this page, we're safe to simply set
3658          * PG_locked without checking it first.
3659          */
3660         __SetPageLocked(page);
3661         rc = add_to_page_cache_locked(page, mapping,
3662                                       page->index, gfp);
3663
3664         /* give up if we can't stick it in the cache */
3665         if (rc) {
3666                 __ClearPageLocked(page);
3667                 return rc;
3668         }
3669
3670         /* move first page to the tmplist */
3671         *offset = (loff_t)page->index << PAGE_SHIFT;
3672         *bytes = PAGE_SIZE;
3673         *nr_pages = 1;
3674         list_move_tail(&page->lru, tmplist);
3675
3676         /* now try and add more pages onto the request */
3677         expected_index = page->index + 1;
3678         list_for_each_entry_safe_reverse(page, tpage, page_list, lru) {
3679                 /* discontinuity ? */
3680                 if (page->index != expected_index)
3681                         break;
3682
3683                 /* would this page push the read over the rsize? */
3684                 if (*bytes + PAGE_SIZE > rsize)
3685                         break;
3686
3687                 __SetPageLocked(page);
3688                 if (add_to_page_cache_locked(page, mapping, page->index, gfp)) {
3689                         __ClearPageLocked(page);
3690                         break;
3691                 }
3692                 list_move_tail(&page->lru, tmplist);
3693                 (*bytes) += PAGE_SIZE;
3694                 expected_index++;
3695                 (*nr_pages)++;
3696         }
3697         return rc;
3698 }
3699
3700 static int cifs_readpages(struct file *file, struct address_space *mapping,
3701         struct list_head *page_list, unsigned num_pages)
3702 {
3703         int rc;
3704         struct list_head tmplist;
3705         struct cifsFileInfo *open_file = file->private_data;
3706         struct cifs_sb_info *cifs_sb = CIFS_FILE_SB(file);
3707         struct TCP_Server_Info *server;
3708         pid_t pid;
3709
3710         /*
3711          * Reads as many pages as possible from fscache. Returns -ENOBUFS
3712          * immediately if the cookie is negative
3713          *
3714          * After this point, every page in the list might have PG_fscache set,
3715          * so we will need to clean that up off of every page we don't use.
3716          */
3717         rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
3718                                          &num_pages);
3719         if (rc == 0)
3720                 return rc;
3721
3722         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3723                 pid = open_file->pid;
3724         else
3725                 pid = current->tgid;
3726
3727         rc = 0;
3728         server = tlink_tcon(open_file->tlink)->ses->server;
3729
3730         cifs_dbg(FYI, "%s: file=%p mapping=%p num_pages=%u\n",
3731                  __func__, file, mapping, num_pages);
3732
3733         /*
3734          * Start with the page at end of list and move it to private
3735          * list. Do the same with any following pages until we hit
3736          * the rsize limit, hit an index discontinuity, or run out of
3737          * pages. Issue the async read and then start the loop again
3738          * until the list is empty.
3739          *
3740          * Note that list order is important. The page_list is in
3741          * the order of declining indexes. When we put the pages in
3742          * the rdata->pages, then we want them in increasing order.
3743          */
3744         while (!list_empty(page_list)) {
3745                 unsigned int i, nr_pages, bytes, rsize;
3746                 loff_t offset;
3747                 struct page *page, *tpage;
3748                 struct cifs_readdata *rdata;
3749                 unsigned credits;
3750
3751                 rc = server->ops->wait_mtu_credits(server, cifs_sb->rsize,
3752                                                    &rsize, &credits);
3753                 if (rc)
3754                         break;
3755
3756                 /*
3757                  * Give up immediately if rsize is too small to read an entire
3758                  * page. The VFS will fall back to readpage. We should never
3759                  * reach this point however since we set ra_pages to 0 when the
3760                  * rsize is smaller than a cache page.
3761                  */
3762                 if (unlikely(rsize < PAGE_SIZE)) {
3763                         add_credits_and_wake_if(server, credits, 0);
3764                         return 0;
3765                 }
3766
3767                 rc = readpages_get_pages(mapping, page_list, rsize, &tmplist,
3768                                          &nr_pages, &offset, &bytes);
3769                 if (rc) {
3770                         add_credits_and_wake_if(server, credits, 0);
3771                         break;
3772                 }
3773
3774                 rdata = cifs_readdata_alloc(nr_pages, cifs_readv_complete);
3775                 if (!rdata) {
3776                         /* best to give up if we're out of mem */
3777                         list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3778                                 list_del(&page->lru);
3779                                 lru_cache_add_file(page);
3780                                 unlock_page(page);
3781                                 put_page(page);
3782                         }
3783                         rc = -ENOMEM;
3784                         add_credits_and_wake_if(server, credits, 0);
3785                         break;
3786                 }
3787
3788                 rdata->cfile = cifsFileInfo_get(open_file);
3789                 rdata->mapping = mapping;
3790                 rdata->offset = offset;
3791                 rdata->bytes = bytes;
3792                 rdata->pid = pid;
3793                 rdata->pagesz = PAGE_SIZE;
3794                 rdata->read_into_pages = cifs_readpages_read_into_pages;
3795                 rdata->copy_into_pages = cifs_readpages_copy_into_pages;
3796                 rdata->credits = credits;
3797
3798                 list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3799                         list_del(&page->lru);
3800                         rdata->pages[rdata->nr_pages++] = page;
3801                 }
3802
3803                 if (!rdata->cfile->invalidHandle ||
3804                     !(rc = cifs_reopen_file(rdata->cfile, true)))
3805                         rc = server->ops->async_readv(rdata);
3806                 if (rc) {
3807                         add_credits_and_wake_if(server, rdata->credits, 0);
3808                         for (i = 0; i < rdata->nr_pages; i++) {
3809                                 page = rdata->pages[i];
3810                                 lru_cache_add_file(page);
3811                                 unlock_page(page);
3812                                 put_page(page);
3813                         }
3814                         /* Fallback to the readpage in error/reconnect cases */
3815                         kref_put(&rdata->refcount, cifs_readdata_release);
3816                         break;
3817                 }
3818
3819                 kref_put(&rdata->refcount, cifs_readdata_release);
3820         }
3821
3822         /* Any pages that have been shown to fscache but didn't get added to
3823          * the pagecache must be uncached before they get returned to the
3824          * allocator.
3825          */
3826         cifs_fscache_readpages_cancel(mapping->host, page_list);
3827         return rc;
3828 }
3829
3830 /*
3831  * cifs_readpage_worker must be called with the page pinned
3832  */
3833 static int cifs_readpage_worker(struct file *file, struct page *page,
3834         loff_t *poffset)
3835 {
3836         char *read_data;
3837         int rc;
3838
3839         /* Is the page cached? */
3840         rc = cifs_readpage_from_fscache(file_inode(file), page);
3841         if (rc == 0)
3842                 goto read_complete;
3843
3844         read_data = kmap(page);
3845         /* for reads over a certain size could initiate async read ahead */
3846
3847         rc = cifs_read(file, read_data, PAGE_SIZE, poffset);
3848
3849         if (rc < 0)
3850                 goto io_error;
3851         else
3852                 cifs_dbg(FYI, "Bytes read %d\n", rc);
3853
3854         file_inode(file)->i_atime =
3855                 current_time(file_inode(file));
3856
3857         if (PAGE_SIZE > rc)
3858                 memset(read_data + rc, 0, PAGE_SIZE - rc);
3859
3860         flush_dcache_page(page);
3861         SetPageUptodate(page);
3862
3863         /* send this page to the cache */
3864         cifs_readpage_to_fscache(file_inode(file), page);
3865
3866         rc = 0;
3867
3868 io_error:
3869         kunmap(page);
3870         unlock_page(page);
3871
3872 read_complete:
3873         return rc;
3874 }
3875
3876 static int cifs_readpage(struct file *file, struct page *page)
3877 {
3878         loff_t offset = (loff_t)page->index << PAGE_SHIFT;
3879         int rc = -EACCES;
3880         unsigned int xid;
3881
3882         xid = get_xid();
3883
3884         if (file->private_data == NULL) {
3885                 rc = -EBADF;
3886                 free_xid(xid);
3887                 return rc;
3888         }
3889
3890         cifs_dbg(FYI, "readpage %p at offset %d 0x%x\n",
3891                  page, (int)offset, (int)offset);
3892
3893         rc = cifs_readpage_worker(file, page, &offset);
3894
3895         free_xid(xid);
3896         return rc;
3897 }
3898
3899 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
3900 {
3901         struct cifsFileInfo *open_file;
3902         struct cifs_tcon *tcon =
3903                 cifs_sb_master_tcon(CIFS_SB(cifs_inode->vfs_inode.i_sb));
3904
3905         spin_lock(&tcon->open_file_lock);
3906         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
3907                 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
3908                         spin_unlock(&tcon->open_file_lock);
3909                         return 1;
3910                 }
3911         }
3912         spin_unlock(&tcon->open_file_lock);
3913         return 0;
3914 }
3915
3916 /* We do not want to update the file size from server for inodes
3917    open for write - to avoid races with writepage extending
3918    the file - in the future we could consider allowing
3919    refreshing the inode only on increases in the file size
3920    but this is tricky to do without racing with writebehind
3921    page caching in the current Linux kernel design */
3922 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
3923 {
3924         if (!cifsInode)
3925                 return true;
3926
3927         if (is_inode_writable(cifsInode)) {
3928                 /* This inode is open for write at least once */
3929                 struct cifs_sb_info *cifs_sb;
3930
3931                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
3932                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
3933                         /* since no page cache to corrupt on directio
3934                         we can change size safely */
3935                         return true;
3936                 }
3937
3938                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
3939                         return true;
3940
3941                 return false;
3942         } else
3943                 return true;
3944 }
3945
3946 static int cifs_write_begin(struct file *file, struct address_space *mapping,
3947                         loff_t pos, unsigned len, unsigned flags,
3948                         struct page **pagep, void **fsdata)
3949 {
3950         int oncethru = 0;
3951         pgoff_t index = pos >> PAGE_SHIFT;
3952         loff_t offset = pos & (PAGE_SIZE - 1);
3953         loff_t page_start = pos & PAGE_MASK;
3954         loff_t i_size;
3955         struct page *page;
3956         int rc = 0;
3957
3958         cifs_dbg(FYI, "write_begin from %lld len %d\n", (long long)pos, len);
3959
3960 start:
3961         page = grab_cache_page_write_begin(mapping, index, flags);
3962         if (!page) {
3963                 rc = -ENOMEM;
3964                 goto out;
3965         }
3966
3967         if (PageUptodate(page))
3968                 goto out;
3969
3970         /*
3971          * If we write a full page it will be up to date, no need to read from
3972          * the server. If the write is short, we'll end up doing a sync write
3973          * instead.
3974          */
3975         if (len == PAGE_SIZE)
3976                 goto out;
3977
3978         /*
3979          * optimize away the read when we have an oplock, and we're not
3980          * expecting to use any of the data we'd be reading in. That
3981          * is, when the page lies beyond the EOF, or straddles the EOF
3982          * and the write will cover all of the existing data.
3983          */
3984         if (CIFS_CACHE_READ(CIFS_I(mapping->host))) {
3985                 i_size = i_size_read(mapping->host);
3986                 if (page_start >= i_size ||
3987                     (offset == 0 && (pos + len) >= i_size)) {
3988                         zero_user_segments(page, 0, offset,
3989                                            offset + len,
3990                                            PAGE_SIZE);
3991                         /*
3992                          * PageChecked means that the parts of the page
3993                          * to which we're not writing are considered up
3994                          * to date. Once the data is copied to the
3995                          * page, it can be set uptodate.
3996                          */
3997                         SetPageChecked(page);
3998                         goto out;
3999                 }
4000         }
4001
4002         if ((file->f_flags & O_ACCMODE) != O_WRONLY && !oncethru) {
4003                 /*
4004                  * might as well read a page, it is fast enough. If we get
4005                  * an error, we don't need to return it. cifs_write_end will
4006                  * do a sync write instead since PG_uptodate isn't set.
4007                  */
4008                 cifs_readpage_worker(file, page, &page_start);
4009                 put_page(page);
4010                 oncethru = 1;
4011                 goto start;
4012         } else {
4013                 /* we could try using another file handle if there is one -
4014                    but how would we lock it to prevent close of that handle
4015                    racing with this read? In any case
4016                    this will be written out by write_end so is fine */
4017         }
4018 out:
4019         *pagep = page;
4020         return rc;
4021 }
4022
4023 static int cifs_release_page(struct page *page, gfp_t gfp)
4024 {
4025         if (PagePrivate(page))
4026                 return 0;
4027
4028         return cifs_fscache_release_page(page, gfp);
4029 }
4030
4031 static void cifs_invalidate_page(struct page *page, unsigned int offset,
4032                                  unsigned int length)
4033 {
4034         struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
4035
4036         if (offset == 0 && length == PAGE_SIZE)
4037                 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
4038 }
4039
4040 static int cifs_launder_page(struct page *page)
4041 {
4042         int rc = 0;
4043         loff_t range_start = page_offset(page);
4044         loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1);
4045         struct writeback_control wbc = {
4046                 .sync_mode = WB_SYNC_ALL,
4047                 .nr_to_write = 0,
4048                 .range_start = range_start,
4049                 .range_end = range_end,
4050         };
4051
4052         cifs_dbg(FYI, "Launder page: %p\n", page);
4053
4054         if (clear_page_dirty_for_io(page))
4055                 rc = cifs_writepage_locked(page, &wbc);
4056
4057         cifs_fscache_invalidate_page(page, page->mapping->host);
4058         return rc;
4059 }
4060
4061 void cifs_oplock_break(struct work_struct *work)
4062 {
4063         struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
4064                                                   oplock_break);
4065         struct inode *inode = d_inode(cfile->dentry);
4066         struct cifsInodeInfo *cinode = CIFS_I(inode);
4067         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
4068         struct TCP_Server_Info *server = tcon->ses->server;
4069         int rc = 0;
4070
4071         wait_on_bit(&cinode->flags, CIFS_INODE_PENDING_WRITERS,
4072                         TASK_UNINTERRUPTIBLE);
4073
4074         server->ops->downgrade_oplock(server, cinode,
4075                 test_bit(CIFS_INODE_DOWNGRADE_OPLOCK_TO_L2, &cinode->flags));
4076
4077         if (!CIFS_CACHE_WRITE(cinode) && CIFS_CACHE_READ(cinode) &&
4078                                                 cifs_has_mand_locks(cinode)) {
4079                 cifs_dbg(FYI, "Reset oplock to None for inode=%p due to mand locks\n",
4080                          inode);
4081                 cinode->oplock = 0;
4082         }
4083
4084         if (inode && S_ISREG(inode->i_mode)) {
4085                 if (CIFS_CACHE_READ(cinode))
4086                         break_lease(inode, O_RDONLY);
4087                 else
4088                         break_lease(inode, O_WRONLY);
4089                 rc = filemap_fdatawrite(inode->i_mapping);
4090                 if (!CIFS_CACHE_READ(cinode)) {
4091                         rc = filemap_fdatawait(inode->i_mapping);
4092                         mapping_set_error(inode->i_mapping, rc);
4093                         cifs_zap_mapping(inode);
4094                 }
4095                 cifs_dbg(FYI, "Oplock flush inode %p rc %d\n", inode, rc);
4096         }
4097
4098         rc = cifs_push_locks(cfile);
4099         if (rc)
4100                 cifs_dbg(VFS, "Push locks rc = %d\n", rc);
4101
4102         /*
4103          * releasing stale oplock after recent reconnect of smb session using
4104          * a now incorrect file handle is not a data integrity issue but do
4105          * not bother sending an oplock release if session to server still is
4106          * disconnected since oplock already released by the server
4107          */
4108         if (!cfile->oplock_break_cancelled) {
4109                 rc = tcon->ses->server->ops->oplock_response(tcon, &cfile->fid,
4110                                                              cinode);
4111                 cifs_dbg(FYI, "Oplock release rc = %d\n", rc);
4112         }
4113         cifs_done_oplock_break(cinode);
4114 }
4115
4116 /*
4117  * The presence of cifs_direct_io() in the address space ops vector
4118  * allowes open() O_DIRECT flags which would have failed otherwise.
4119  *
4120  * In the non-cached mode (mount with cache=none), we shunt off direct read and write requests
4121  * so this method should never be called.
4122  *
4123  * Direct IO is not yet supported in the cached mode.
4124  */
4125 static ssize_t
4126 cifs_direct_io(struct kiocb *iocb, struct iov_iter *iter)
4127 {
4128         /*
4129          * FIXME
4130          * Eventually need to support direct IO for non forcedirectio mounts
4131          */
4132         return -EINVAL;
4133 }
4134
4135
4136 const struct address_space_operations cifs_addr_ops = {
4137         .readpage = cifs_readpage,
4138         .readpages = cifs_readpages,
4139         .writepage = cifs_writepage,
4140         .writepages = cifs_writepages,
4141         .write_begin = cifs_write_begin,
4142         .write_end = cifs_write_end,
4143         .set_page_dirty = __set_page_dirty_nobuffers,
4144         .releasepage = cifs_release_page,
4145         .direct_IO = cifs_direct_io,
4146         .invalidatepage = cifs_invalidate_page,
4147         .launder_page = cifs_launder_page,
4148 };
4149
4150 /*
4151  * cifs_readpages requires the server to support a buffer large enough to
4152  * contain the header plus one complete page of data.  Otherwise, we need
4153  * to leave cifs_readpages out of the address space operations.
4154  */
4155 const struct address_space_operations cifs_addr_ops_smallbuf = {
4156         .readpage = cifs_readpage,
4157         .writepage = cifs_writepage,
4158         .writepages = cifs_writepages,
4159         .write_begin = cifs_write_begin,
4160         .write_end = cifs_write_end,
4161         .set_page_dirty = __set_page_dirty_nobuffers,
4162         .releasepage = cifs_release_page,
4163         .invalidatepage = cifs_invalidate_page,
4164         .launder_page = cifs_launder_page,
4165 };