fs/cifs/file.c

   1 /*
   2  *   fs/cifs/file.c
   3  *
   4  *   vfs operations that deal with files
   5  *
   6  *   Copyright (C) International Business Machines  Corp., 2002,2010
   7  *   Author(s): Steve French (sfrench@us.ibm.com)
   8  *              Jeremy Allison (jra@samba.org)
   9  *
  10  *   This library is free software; you can redistribute it and/or modify
  11  *   it under the terms of the GNU Lesser General Public License as published
  12  *   by the Free Software Foundation; either version 2.1 of the License, or
  13  *   (at your option) any later version.
  14  *
  15  *   This library is distributed in the hope that it will be useful,
  16  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
  18  *   the GNU Lesser General Public License for more details.
  19  *
  20  *   You should have received a copy of the GNU Lesser General Public License
  21  *   along with this library; if not, write to the Free Software
  22  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  23  */
  24 #include <linux/fs.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/stat.h>
  27 #include <linux/fcntl.h>
  28 #include <linux/pagemap.h>
  29 #include <linux/pagevec.h>
  30 #include <linux/writeback.h>
  31 #include <linux/task_io_accounting_ops.h>
  32 #include <linux/delay.h>
  33 #include <linux/mount.h>
  34 #include <linux/slab.h>
  35 #include <linux/swap.h>
  36 #include <asm/div64.h>
  37 #include "cifsfs.h"
  38 #include "cifspdu.h"
  39 #include "cifsglob.h"
  40 #include "cifsproto.h"
  41 #include "cifs_unicode.h"
  42 #include "cifs_debug.h"
  43 #include "cifs_fs_sb.h"
  44 #include "fscache.h"
  45
  46
  47 static inline int cifs_convert_flags(unsigned int flags)
  48 {
  49         if ((flags & O_ACCMODE) == O_RDONLY)
  50                 return GENERIC_READ;
  51         else if ((flags & O_ACCMODE) == O_WRONLY)
  52                 return GENERIC_WRITE;
  53         else if ((flags & O_ACCMODE) == O_RDWR) {
  54                 /* GENERIC_ALL is too much permission to request
  55                    can cause unnecessary access denied on create */
  56                 /* return GENERIC_ALL; */
  57                 return (GENERIC_READ | GENERIC_WRITE);
  58         }
  59
  60         return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
  61                 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
  62                 FILE_READ_DATA);
  63 }
  64
  65 static u32 cifs_posix_convert_flags(unsigned int flags)
  66 {
  67         u32 posix_flags = 0;
  68
  69         if ((flags & O_ACCMODE) == O_RDONLY)
  70                 posix_flags = SMB_O_RDONLY;
  71         else if ((flags & O_ACCMODE) == O_WRONLY)
  72                 posix_flags = SMB_O_WRONLY;
  73         else if ((flags & O_ACCMODE) == O_RDWR)
  74                 posix_flags = SMB_O_RDWR;
  75
  76         if (flags & O_CREAT) {
  77                 posix_flags |= SMB_O_CREAT;
  78                 if (flags & O_EXCL)
  79                         posix_flags |= SMB_O_EXCL;
  80         } else if (flags & O_EXCL)
  81                 cifs_dbg(FYI, "Application %s pid %d has incorrectly set O_EXCL flag but not O_CREAT on file open. Ignoring O_EXCL\n",
  82                          current->comm, current->tgid);
  83
  84         if (flags & O_TRUNC)
  85                 posix_flags |= SMB_O_TRUNC;
  86         /* be safe and imply O_SYNC for O_DSYNC */
  87         if (flags & O_DSYNC)
  88                 posix_flags |= SMB_O_SYNC;
  89         if (flags & O_DIRECTORY)
  90                 posix_flags |= SMB_O_DIRECTORY;
  91         if (flags & O_NOFOLLOW)
  92                 posix_flags |= SMB_O_NOFOLLOW;
  93         if (flags & O_DIRECT)
  94                 posix_flags |= SMB_O_DIRECT;
  95
  96         return posix_flags;
  97 }
  98
  99 static inline int cifs_get_disposition(unsigned int flags)
 100 {
 101         if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
 102                 return FILE_CREATE;
 103         else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
 104                 return FILE_OVERWRITE_IF;
 105         else if ((flags & O_CREAT) == O_CREAT)
 106                 return FILE_OPEN_IF;
 107         else if ((flags & O_TRUNC) == O_TRUNC)
 108                 return FILE_OVERWRITE;
 109         else
 110                 return FILE_OPEN;
 111 }
 112
 113 int cifs_posix_open(char *full_path, struct inode **pinode,
 114                         struct super_block *sb, int mode, unsigned int f_flags,
 115                         __u32 *poplock, __u16 *pnetfid, unsigned int xid)
 116 {
 117         int rc;
 118         FILE_UNIX_BASIC_INFO *presp_data;
 119         __u32 posix_flags = 0;
 120         struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
 121         struct cifs_fattr fattr;
 122         struct tcon_link *tlink;
 123         struct cifs_tcon *tcon;
 124
 125         cifs_dbg(FYI, "posix open %s\n", full_path);
 126
 127         presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
 128         if (presp_data == NULL)
 129                 return -ENOMEM;
 130
 131         tlink = cifs_sb_tlink(cifs_sb);
 132         if (IS_ERR(tlink)) {
 133                 rc = PTR_ERR(tlink);
 134                 goto posix_open_ret;
 135         }
 136
 137         tcon = tlink_tcon(tlink);
 138         mode &= ~current_umask();
 139
 140         posix_flags = cifs_posix_convert_flags(f_flags);
 141         rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
 142                              poplock, full_path, cifs_sb->local_nls,
 143                              cifs_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         oparms.tcon = tcon;
 228         oparms.cifs_sb = cifs_sb;
 229         oparms.desired_access = desired_access;
 230         oparms.create_options = create_options;
 231         oparms.disposition = disposition;
 232         oparms.path = full_path;
 233         oparms.fid = fid;
 234         oparms.reconnect = false;
 235
 236         rc = server->ops->open(xid, &oparms, oplock, buf);
 237
 238         if (rc)
 239                 goto out;
 240
 241         if (tcon->unix_ext)
 242                 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
 243                                               xid);
 244         else
 245                 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
 246                                          xid, fid);
 247
 248 out:
 249         kfree(buf);
 250         return rc;
 251 }
 252
 253 static bool
 254 cifs_has_mand_locks(struct cifsInodeInfo *cinode)
 255 {
 256         struct cifs_fid_locks *cur;
 257         bool has_locks = false;
 258
 259         down_read(&cinode->lock_sem);
 260         list_for_each_entry(cur, &cinode->llist, llist) {
 261                 if (!list_empty(&cur->locks)) {
 262                         has_locks = true;
 263                         break;
 264                 }
 265         }
 266         up_read(&cinode->lock_sem);
 267         return has_locks;
 268 }
 269
 270 struct cifsFileInfo *
 271 cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
 272                   struct tcon_link *tlink, __u32 oplock)
 273 {
 274         struct dentry *dentry = file_dentry(file);
 275         struct inode *inode = d_inode(dentry);
 276         struct cifsInodeInfo *cinode = CIFS_I(inode);
 277         struct cifsFileInfo *cfile;
 278         struct cifs_fid_locks *fdlocks;
 279         struct cifs_tcon *tcon = tlink_tcon(tlink);
 280         struct TCP_Server_Info *server = tcon->ses->server;
 281
 282         cfile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
 283         if (cfile == NULL)
 284                 return cfile;
 285
 286         fdlocks = kzalloc(sizeof(struct cifs_fid_locks), GFP_KERNEL);
 287         if (!fdlocks) {
 288                 kfree(cfile);
 289                 return NULL;
 290         }
 291
 292         INIT_LIST_HEAD(&fdlocks->locks);
 293         fdlocks->cfile = cfile;
 294         cfile->llist = fdlocks;
 295         down_write(&cinode->lock_sem);
 296         list_add(&fdlocks->llist, &cinode->llist);
 297         up_write(&cinode->lock_sem);
 298
 299         cfile->count = 1;
 300         cfile->pid = current->tgid;
 301         cfile->uid = current_fsuid();
 302         cfile->dentry = dget(dentry);
 303         cfile->f_flags = file->f_flags;
 304         cfile->invalidHandle = false;
 305         cfile->tlink = cifs_get_tlink(tlink);
 306         INIT_WORK(&cfile->oplock_break, cifs_oplock_break);
 307         mutex_init(&cfile->fh_mutex);
 308         spin_lock_init(&cfile->file_info_lock);
 309
 310         cifs_sb_active(inode->i_sb);
 311
 312         /*
 313          * If the server returned a read oplock and we have mandatory brlocks,
 314          * set oplock level to None.
 315          */
 316         if (server->ops->is_read_op(oplock) && cifs_has_mand_locks(cinode)) {
 317                 cifs_dbg(FYI, "Reset oplock val from read to None due to mand locks\n");
 318                 oplock = 0;
 319         }
 320
 321         spin_lock(&tcon->open_file_lock);
 322         if (fid->pending_open->oplock != CIFS_OPLOCK_NO_CHANGE && oplock)
 323                 oplock = fid->pending_open->oplock;
 324         list_del(&fid->pending_open->olist);
 325
 326         fid->purge_cache = false;
 327         server->ops->set_fid(cfile, fid, oplock);
 328
 329         list_add(&cfile->tlist, &tcon->openFileList);
 330
 331         /* if readable file instance put first in list*/
 332         if (file->f_mode & FMODE_READ)
 333                 list_add(&cfile->flist, &cinode->openFileList);
 334         else
 335                 list_add_tail(&cfile->flist, &cinode->openFileList);
 336         spin_unlock(&tcon->open_file_lock);
 337
 338         if (fid->purge_cache)
 339                 cifs_zap_mapping(inode);
 340
 341         file->private_data = cfile;
 342         return cfile;
 343 }
 344
 345 struct cifsFileInfo *
 346 cifsFileInfo_get(struct cifsFileInfo *cifs_file)
 347 {
 348         spin_lock(&cifs_file->file_info_lock);
 349         cifsFileInfo_get_locked(cifs_file);
 350         spin_unlock(&cifs_file->file_info_lock);
 351         return cifs_file;
 352 }
 353
 354 /*
 355  * Release a reference on the file private data. This may involve closing
 356  * the filehandle out on the server. Must be called without holding
 357  * tcon->open_file_lock and cifs_file->file_info_lock.
 358  */
 359 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
 360 {
 361         struct inode *inode = d_inode(cifs_file->dentry);
 362         struct cifs_tcon *tcon = tlink_tcon(cifs_file->tlink);
 363         struct TCP_Server_Info *server = tcon->ses->server;
 364         struct cifsInodeInfo *cifsi = CIFS_I(inode);
 365         struct super_block *sb = inode->i_sb;
 366         struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
 367         struct cifsLockInfo *li, *tmp;
 368         struct cifs_fid fid;
 369         struct cifs_pending_open open;
 370         bool oplock_break_cancelled;
 371
 372         spin_lock(&tcon->open_file_lock);
 373
 374         spin_lock(&cifs_file->file_info_lock);
 375         if (--cifs_file->count > 0) {
 376                 spin_unlock(&cifs_file->file_info_lock);
 377                 spin_unlock(&tcon->open_file_lock);
 378                 return;
 379         }
 380         spin_unlock(&cifs_file->file_info_lock);
 381
 382         if (server->ops->get_lease_key)
 383                 server->ops->get_lease_key(inode, &fid);
 384
 385         /* store open in pending opens to make sure we don't miss lease break */
 386         cifs_add_pending_open_locked(&fid, cifs_file->tlink, &open);
 387
 388         /* remove it from the lists */
 389         list_del(&cifs_file->flist);
 390         list_del(&cifs_file->tlist);
 391
 392         if (list_empty(&cifsi->openFileList)) {
 393                 cifs_dbg(FYI, "closing last open instance for inode %p\n",
 394                          d_inode(cifs_file->dentry));
 395                 /*
 396                  * In strict cache mode we need invalidate mapping on the last
 397                  * close  because it may cause a error when we open this file
 398                  * again and get at least level II oplock.
 399                  */
 400                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
 401                         set_bit(CIFS_INO_INVALID_MAPPING, &cifsi->flags);
 402                 cifs_set_oplock_level(cifsi, 0);
 403         }
 404
 405         spin_unlock(&tcon->open_file_lock);
 406
 407         oplock_break_cancelled = cancel_work_sync(&cifs_file->oplock_break);
 408
 409         if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
 410                 struct TCP_Server_Info *server = tcon->ses->server;
 411                 unsigned int xid;
 412
 413                 xid = get_xid();
 414                 if (server->ops->close)
 415                         server->ops->close(xid, tcon, &cifs_file->fid);
 416                 _free_xid(xid);
 417         }
 418
 419         if (oplock_break_cancelled)
 420                 cifs_done_oplock_break(cifsi);
 421
 422         cifs_del_pending_open(&open);
 423
 424         /*
 425          * Delete any outstanding lock records. We'll lose them when the file
 426          * is closed anyway.
 427          */
 428         down_write(&cifsi->lock_sem);
 429         list_for_each_entry_safe(li, tmp, &cifs_file->llist->locks, llist) {
 430                 list_del(&li->llist);
 431                 cifs_del_lock_waiters(li);
 432                 kfree(li);
 433         }
 434         list_del(&cifs_file->llist->llist);
 435         kfree(cifs_file->llist);
 436         up_write(&cifsi->lock_sem);
 437
 438         cifs_put_tlink(cifs_file->tlink);
 439         dput(cifs_file->dentry);
 440         cifs_sb_deactive(sb);
 441         kfree(cifs_file);
 442 }
 443
 444 int cifs_open(struct inode *inode, struct file *file)
 445
 446 {
 447         int rc = -EACCES;
 448         unsigned int xid;
 449         __u32 oplock;
 450         struct cifs_sb_info *cifs_sb;
 451         struct TCP_Server_Info *server;
 452         struct cifs_tcon *tcon;
 453         struct tcon_link *tlink;
 454         struct cifsFileInfo *cfile = NULL;
 455         char *full_path = NULL;
 456         bool posix_open_ok = false;
 457         struct cifs_fid fid;
 458         struct cifs_pending_open open;
 459
 460         xid = get_xid();
 461
 462         cifs_sb = CIFS_SB(inode->i_sb);
 463         tlink = cifs_sb_tlink(cifs_sb);
 464         if (IS_ERR(tlink)) {
 465                 free_xid(xid);
 466                 return PTR_ERR(tlink);
 467         }
 468         tcon = tlink_tcon(tlink);
 469         server = tcon->ses->server;
 470
 471         full_path = build_path_from_dentry(file_dentry(file));
 472         if (full_path == NULL) {
 473                 rc = -ENOMEM;
 474                 goto out;
 475         }
 476
 477         cifs_dbg(FYI, "inode = 0x%p file flags are 0x%x for %s\n",
 478                  inode, file->f_flags, full_path);
 479
 480         if (file->f_flags & O_DIRECT &&
 481             cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO) {
 482                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_BRL)
 483                         file->f_op = &cifs_file_direct_nobrl_ops;
 484                 else
 485                         file->f_op = &cifs_file_direct_ops;
 486         }
 487
 488         if (server->oplocks)
 489                 oplock = REQ_OPLOCK;
 490         else
 491                 oplock = 0;
 492
 493         if (!tcon->broken_posix_open && tcon->unix_ext &&
 494             cap_unix(tcon->ses) && (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 495                                 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 496                 /* can not refresh inode info since size could be stale */
 497                 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
 498                                 cifs_sb->mnt_file_mode /* ignored */,
 499                                 file->f_flags, &oplock, &fid.netfid, xid);
 500                 if (rc == 0) {
 501                         cifs_dbg(FYI, "posix open succeeded\n");
 502                         posix_open_ok = true;
 503                 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
 504                         if (tcon->ses->serverNOS)
 505                                 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",
 506                                          tcon->ses->serverName,
 507                                          tcon->ses->serverNOS);
 508                         tcon->broken_posix_open = true;
 509                 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
 510                          (rc != -EOPNOTSUPP)) /* path not found or net err */
 511                         goto out;
 512                 /*
 513                  * Else fallthrough to retry open the old way on network i/o
 514                  * or DFS errors.
 515                  */
 516         }
 517
 518         if (server->ops->get_lease_key)
 519                 server->ops->get_lease_key(inode, &fid);
 520
 521         cifs_add_pending_open(&fid, tlink, &open);
 522
 523         if (!posix_open_ok) {
 524                 if (server->ops->get_lease_key)
 525                         server->ops->get_lease_key(inode, &fid);
 526
 527                 rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
 528                                   file->f_flags, &oplock, &fid, xid);
 529                 if (rc) {
 530                         cifs_del_pending_open(&open);
 531                         goto out;
 532                 }
 533         }
 534
 535         cfile = cifs_new_fileinfo(&fid, file, tlink, oplock);
 536         if (cfile == NULL) {
 537                 if (server->ops->close)
 538                         server->ops->close(xid, tcon, &fid);
 539                 cifs_del_pending_open(&open);
 540                 rc = -ENOMEM;
 541                 goto out;
 542         }
 543
 544         cifs_fscache_set_inode_cookie(inode, file);
 545
 546         if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
 547                 /*
 548                  * Time to set mode which we can not set earlier due to
 549                  * problems creating new read-only files.
 550                  */
 551                 struct cifs_unix_set_info_args args = {
 552                         .mode   = inode->i_mode,
 553                         .uid    = INVALID_UID, /* no change */
 554                         .gid    = INVALID_GID, /* no change */
 555                         .ctime  = NO_CHANGE_64,
 556                         .atime  = NO_CHANGE_64,
 557                         .mtime  = NO_CHANGE_64,
 558                         .device = 0,
 559                 };
 560                 CIFSSMBUnixSetFileInfo(xid, tcon, &args, fid.netfid,
 561                                        cfile->pid);
 562         }
 563
 564 out:
 565         kfree(full_path);
 566         free_xid(xid);
 567         cifs_put_tlink(tlink);
 568         return rc;
 569 }
 570
 571 static int cifs_push_posix_locks(struct cifsFileInfo *cfile);
 572
 573 /*
 574  * Try to reacquire byte range locks that were released when session
 575  * to server was lost.
 576  */
 577 static int
 578 cifs_relock_file(struct cifsFileInfo *cfile)
 579 {
 580         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
 581         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 582         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
 583         int rc = 0;
 584
 585         down_read_nested(&cinode->lock_sem, SINGLE_DEPTH_NESTING);
 586         if (cinode->can_cache_brlcks) {
 587                 /* can cache locks - no need to relock */
 588                 up_read(&cinode->lock_sem);
 589                 return rc;
 590         }
 591
 592         if (cap_unix(tcon->ses) &&
 593             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
 594             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
 595                 rc = cifs_push_posix_locks(cfile);
 596         else
 597                 rc = tcon->ses->server->ops->push_mand_locks(cfile);
 598
 599         up_read(&cinode->lock_sem);
 600         return rc;
 601 }
 602
 603 static int
 604 cifs_reopen_file(struct cifsFileInfo *cfile, bool can_flush)
 605 {
 606         int rc = -EACCES;
 607         unsigned int xid;
 608         __u32 oplock;
 609         struct cifs_sb_info *cifs_sb;
 610         struct cifs_tcon *tcon;
 611         struct TCP_Server_Info *server;
 612         struct cifsInodeInfo *cinode;
 613         struct inode *inode;
 614         char *full_path = NULL;
 615         int desired_access;
 616         int disposition = FILE_OPEN;
 617         int create_options = CREATE_NOT_DIR;
 618         struct cifs_open_parms oparms;
 619
 620         xid = get_xid();
 621         mutex_lock(&cfile->fh_mutex);
 622         if (!cfile->invalidHandle) {
 623                 mutex_unlock(&cfile->fh_mutex);
 624                 rc = 0;
 625                 free_xid(xid);
 626                 return rc;
 627         }
 628
 629         inode = d_inode(cfile->dentry);
 630         cifs_sb = CIFS_SB(inode->i_sb);
 631         tcon = tlink_tcon(cfile->tlink);
 632         server = tcon->ses->server;
 633
 634         /*
 635          * Can not grab rename sem here because various ops, including those
 636          * that already have the rename sem can end up causing writepage to get
 637          * called and if the server was down that means we end up here, and we
 638          * can never tell if the caller already has the rename_sem.
 639          */
 640         full_path = build_path_from_dentry(cfile->dentry);
 641         if (full_path == NULL) {
 642                 rc = -ENOMEM;
 643                 mutex_unlock(&cfile->fh_mutex);
 644                 free_xid(xid);
 645                 return rc;
 646         }
 647
 648         cifs_dbg(FYI, "inode = 0x%p file flags 0x%x for %s\n",
 649                  inode, cfile->f_flags, full_path);
 650
 651         if (tcon->ses->server->oplocks)
 652                 oplock = REQ_OPLOCK;
 653         else
 654                 oplock = 0;
 655
 656         if (tcon->unix_ext && cap_unix(tcon->ses) &&
 657             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 658                                 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 659                 /*
 660                  * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
 661                  * original open. Must mask them off for a reopen.
 662                  */
 663                 unsigned int oflags = cfile->f_flags &
 664                                                 ~(O_CREAT | O_EXCL | O_TRUNC);
 665
 666                 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
 667                                      cifs_sb->mnt_file_mode /* ignored */,
 668                                      oflags, &oplock, &cfile->fid.netfid, xid);
 669                 if (rc == 0) {
 670                         cifs_dbg(FYI, "posix reopen succeeded\n");
 671                         oparms.reconnect = true;
 672                         goto reopen_success;
 673                 }
 674                 /*
 675                  * fallthrough to retry open the old way on errors, especially
 676                  * in the reconnect path it is important to retry hard
 677                  */
 678         }
 679
 680         desired_access = cifs_convert_flags(cfile->f_flags);
 681
 682         if (backup_cred(cifs_sb))
 683                 create_options |= CREATE_OPEN_BACKUP_INTENT;
 684
 685         if (server->ops->get_lease_key)
 686                 server->ops->get_lease_key(inode, &cfile->fid);
 687
 688         oparms.tcon = tcon;
 689         oparms.cifs_sb = cifs_sb;
 690         oparms.desired_access = desired_access;
 691         oparms.create_options = create_options;
 692         oparms.disposition = disposition;
 693         oparms.path = full_path;
 694         oparms.fid = &cfile->fid;
 695         oparms.reconnect = true;
 696
 697         /*
 698          * Can not refresh inode by passing in file_info buf to be returned by
 699          * ops->open and then calling get_inode_info with returned buf since
 700          * file might have write behind data that needs to be flushed and server
 701          * version of file size can be stale. If we knew for sure that inode was
 702          * not dirty locally we could do this.
 703          */
 704         rc = server->ops->open(xid, &oparms, &oplock, NULL);
 705         if (rc == -ENOENT && oparms.reconnect == false) {
 706                 /* durable handle timeout is expired - open the file again */
 707                 rc = server->ops->open(xid, &oparms, &oplock, NULL);
 708                 /* indicate that we need to relock the file */
 709                 oparms.reconnect = true;
 710         }
 711
 712         if (rc) {
 713                 mutex_unlock(&cfile->fh_mutex);
 714                 cifs_dbg(FYI, "cifs_reopen returned 0x%x\n", rc);
 715                 cifs_dbg(FYI, "oplock: %d\n", oplock);
 716                 goto reopen_error_exit;
 717         }
 718
 719 reopen_success:
 720         cfile->invalidHandle = false;
 721         mutex_unlock(&cfile->fh_mutex);
 722         cinode = CIFS_I(inode);
 723
 724         if (can_flush) {
 725                 rc = filemap_write_and_wait(inode->i_mapping);
 726                 mapping_set_error(inode->i_mapping, rc);
 727
 728                 if (tcon->unix_ext)
 729                         rc = cifs_get_inode_info_unix(&inode, full_path,
 730                                                       inode->i_sb, xid);
 731                 else
 732                         rc = cifs_get_inode_info(&inode, full_path, NULL,
 733                                                  inode->i_sb, xid, NULL);
 734         }
 735         /*
 736          * Else we are writing out data to server already and could deadlock if
 737          * we tried to flush data, and since we do not know if we have data that
 738          * would invalidate the current end of file on the server we can not go
 739          * to the server to get the new inode info.
 740          */
 741
 742         /*
 743          * If the server returned a read oplock and we have mandatory brlocks,
 744          * set oplock level to None.
 745          */
 746         if (server->ops->is_read_op(oplock) && cifs_has_mand_locks(cinode)) {
 747                 cifs_dbg(FYI, "Reset oplock val from read to None due to mand locks\n");
 748                 oplock = 0;
 749         }
 750
 751         server->ops->set_fid(cfile, &cfile->fid, oplock);
 752         if (oparms.reconnect)
 753                 cifs_relock_file(cfile);
 754
 755 reopen_error_exit:
 756         kfree(full_path);
 757         free_xid(xid);
 758         return rc;
 759 }
 760
 761 int cifs_close(struct inode *inode, struct file *file)
 762 {
 763         if (file->private_data != NULL) {
 764                 cifsFileInfo_put(file->private_data);
 765                 file->private_data = NULL;
 766         }
 767
 768         /* return code from the ->release op is always ignored */
 769         return 0;
 770 }
 771
 772 void
 773 cifs_reopen_persistent_handles(struct cifs_tcon *tcon)
 774 {
 775         struct cifsFileInfo *open_file;
 776         struct list_head *tmp;
 777         struct list_head *tmp1;
 778         struct list_head tmp_list;
 779
 780         if (!tcon->use_persistent || !tcon->need_reopen_files)
 781                 return;
 782
 783         tcon->need_reopen_files = false;
 784
 785         cifs_dbg(FYI, "Reopen persistent handles");
 786         INIT_LIST_HEAD(&tmp_list);
 787
 788         /* list all files open on tree connection, reopen resilient handles  */
 789         spin_lock(&tcon->open_file_lock);
 790         list_for_each(tmp, &tcon->openFileList) {
 791                 open_file = list_entry(tmp, struct cifsFileInfo, tlist);
 792                 if (!open_file->invalidHandle)
 793                         continue;
 794                 cifsFileInfo_get(open_file);
 795                 list_add_tail(&open_file->rlist, &tmp_list);
 796         }
 797         spin_unlock(&tcon->open_file_lock);
 798
 799         list_for_each_safe(tmp, tmp1, &tmp_list) {
 800                 open_file = list_entry(tmp, struct cifsFileInfo, rlist);
 801                 if (cifs_reopen_file(open_file, false /* do not flush */))
 802                         tcon->need_reopen_files = true;
 803                 list_del_init(&open_file->rlist);
 804                 cifsFileInfo_put(open_file);
 805         }
 806 }
 807
 808 int cifs_closedir(struct inode *inode, struct file *file)
 809 {
 810         int rc = 0;
 811         unsigned int xid;
 812         struct cifsFileInfo *cfile = file->private_data;
 813         struct cifs_tcon *tcon;
 814         struct TCP_Server_Info *server;
 815         char *buf;
 816
 817         cifs_dbg(FYI, "Closedir inode = 0x%p\n", inode);
 818
 819         if (cfile == NULL)
 820                 return rc;
 821
 822         xid = get_xid();
 823         tcon = tlink_tcon(cfile->tlink);
 824         server = tcon->ses->server;
 825
 826         cifs_dbg(FYI, "Freeing private data in close dir\n");
 827         spin_lock(&cfile->file_info_lock);
 828         if (server->ops->dir_needs_close(cfile)) {
 829                 cfile->invalidHandle = true;
 830                 spin_unlock(&cfile->file_info_lock);
 831                 if (server->ops->close_dir)
 832                         rc = server->ops->close_dir(xid, tcon, &cfile->fid);
 833                 else
 834                         rc = -ENOSYS;
 835                 cifs_dbg(FYI, "Closing uncompleted readdir with rc %d\n", rc);
 836                 /* not much we can do if it fails anyway, ignore rc */
 837                 rc = 0;
 838         } else
 839                 spin_unlock(&cfile->file_info_lock);
 840
 841         buf = cfile->srch_inf.ntwrk_buf_start;
 842         if (buf) {
 843                 cifs_dbg(FYI, "closedir free smb buf in srch struct\n");
 844                 cfile->srch_inf.ntwrk_buf_start = NULL;
 845                 if (cfile->srch_inf.smallBuf)
 846                         cifs_small_buf_release(buf);
 847                 else
 848                         cifs_buf_release(buf);
 849         }
 850
 851         cifs_put_tlink(cfile->tlink);
 852         kfree(file->private_data);
 853         file->private_data = NULL;
 854         /* BB can we lock the filestruct while this is going on? */
 855         free_xid(xid);
 856         return rc;
 857 }
 858
 859 static struct cifsLockInfo *
 860 cifs_lock_init(__u64 offset, __u64 length, __u8 type)
 861 {
 862         struct cifsLockInfo *lock =
 863                 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
 864         if (!lock)
 865                 return lock;
 866         lock->offset = offset;
 867         lock->length = length;
 868         lock->type = type;
 869         lock->pid = current->tgid;
 870         INIT_LIST_HEAD(&lock->blist);
 871         init_waitqueue_head(&lock->block_q);
 872         return lock;
 873 }
 874
 875 void
 876 cifs_del_lock_waiters(struct cifsLockInfo *lock)
 877 {
 878         struct cifsLockInfo *li, *tmp;
 879         list_for_each_entry_safe(li, tmp, &lock->blist, blist) {
 880                 list_del_init(&li->blist);
 881                 wake_up(&li->block_q);
 882         }
 883 }
 884
 885 #define CIFS_LOCK_OP    0
 886 #define CIFS_READ_OP    1
 887 #define CIFS_WRITE_OP   2
 888
 889 /* @rw_check : 0 - no op, 1 - read, 2 - write */
 890 static bool
 891 cifs_find_fid_lock_conflict(struct cifs_fid_locks *fdlocks, __u64 offset,
 892                             __u64 length, __u8 type, struct cifsFileInfo *cfile,
 893                             struct cifsLockInfo **conf_lock, int rw_check)
 894 {
 895         struct cifsLockInfo *li;
 896         struct cifsFileInfo *cur_cfile = fdlocks->cfile;
 897         struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
 898
 899         list_for_each_entry(li, &fdlocks->locks, llist) {
 900                 if (offset + length <= li->offset ||
 901                     offset >= li->offset + li->length)
 902                         continue;
 903                 if (rw_check != CIFS_LOCK_OP && current->tgid == li->pid &&
 904                     server->ops->compare_fids(cfile, cur_cfile)) {
 905                         /* shared lock prevents write op through the same fid */
 906                         if (!(li->type & server->vals->shared_lock_type) ||
 907                             rw_check != CIFS_WRITE_OP)
 908                                 continue;
 909                 }
 910                 if ((type & server->vals->shared_lock_type) &&
 911                     ((server->ops->compare_fids(cfile, cur_cfile) &&
 912                      current->tgid == li->pid) || type == li->type))
 913                         continue;
 914                 if (conf_lock)
 915                         *conf_lock = li;
 916                 return true;
 917         }
 918         return false;
 919 }
 920
 921 bool
 922 cifs_find_lock_conflict(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
 923                         __u8 type, struct cifsLockInfo **conf_lock,
 924                         int rw_check)
 925 {
 926         bool rc = false;
 927         struct cifs_fid_locks *cur;
 928         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 929
 930         list_for_each_entry(cur, &cinode->llist, llist) {
 931                 rc = cifs_find_fid_lock_conflict(cur, offset, length, type,
 932                                                  cfile, conf_lock, rw_check);
 933                 if (rc)
 934                         break;
 935         }
 936
 937         return rc;
 938 }
 939
 940 /*
 941  * Check if there is another lock that prevents us to set the lock (mandatory
 942  * style). If such a lock exists, update the flock structure with its
 943  * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
 944  * or leave it the same if we can't. Returns 0 if we don't need to request to
 945  * the server or 1 otherwise.
 946  */
 947 static int
 948 cifs_lock_test(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
 949                __u8 type, struct file_lock *flock)
 950 {
 951         int rc = 0;
 952         struct cifsLockInfo *conf_lock;
 953         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 954         struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
 955         bool exist;
 956
 957         down_read(&cinode->lock_sem);
 958
 959         exist = cifs_find_lock_conflict(cfile, offset, length, type,
 960                                         &conf_lock, CIFS_LOCK_OP);
 961         if (exist) {
 962                 flock->fl_start = conf_lock->offset;
 963                 flock->fl_end = conf_lock->offset + conf_lock->length - 1;
 964                 flock->fl_pid = conf_lock->pid;
 965                 if (conf_lock->type & server->vals->shared_lock_type)
 966                         flock->fl_type = F_RDLCK;
 967                 else
 968                         flock->fl_type = F_WRLCK;
 969         } else if (!cinode->can_cache_brlcks)
 970                 rc = 1;
 971         else
 972                 flock->fl_type = F_UNLCK;
 973
 974         up_read(&cinode->lock_sem);
 975         return rc;
 976 }
 977
 978 static void
 979 cifs_lock_add(struct cifsFileInfo *cfile, struct cifsLockInfo *lock)
 980 {
 981         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 982         down_write(&cinode->lock_sem);
 983         list_add_tail(&lock->llist, &cfile->llist->locks);
 984         up_write(&cinode->lock_sem);
 985 }
 986
 987 /*
 988  * Set the byte-range lock (mandatory style). Returns:
 989  * 1) 0, if we set the lock and don't need to request to the server;
 990  * 2) 1, if no locks prevent us but we need to request to the server;
 991  * 3) -EACCESS, if there is a lock that prevents us and wait is false.
 992  */
 993 static int
 994 cifs_lock_add_if(struct cifsFileInfo *cfile, struct cifsLockInfo *lock,
 995                  bool wait)
 996 {
 997         struct cifsLockInfo *conf_lock;
 998         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 999         bool exist;
1000         int rc = 0;
1001
1002 try_again:
1003         exist = false;
1004         down_write(&cinode->lock_sem);
1005
1006         exist = cifs_find_lock_conflict(cfile, lock->offset, lock->length,
1007                                         lock->type, &conf_lock, CIFS_LOCK_OP);
1008         if (!exist && cinode->can_cache_brlcks) {
1009                 list_add_tail(&lock->llist, &cfile->llist->locks);
1010                 up_write(&cinode->lock_sem);
1011                 return rc;
1012         }
1013
1014         if (!exist)
1015                 rc = 1;
1016         else if (!wait)
1017                 rc = -EACCES;
1018         else {
1019                 list_add_tail(&lock->blist, &conf_lock->blist);
1020                 up_write(&cinode->lock_sem);
1021                 rc = wait_event_interruptible(lock->block_q,
1022                                         (lock->blist.prev == &lock->blist) &&
1023                                         (lock->blist.next == &lock->blist));
1024                 if (!rc)
1025                         goto try_again;
1026                 down_write(&cinode->lock_sem);
1027                 list_del_init(&lock->blist);
1028         }
1029
1030         up_write(&cinode->lock_sem);
1031         return rc;
1032 }
1033
1034 /*
1035  * Check if there is another lock that prevents us to set the lock (posix
1036  * style). If such a lock exists, update the flock structure with its
1037  * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
1038  * or leave it the same if we can't. Returns 0 if we don't need to request to
1039  * the server or 1 otherwise.
1040  */
1041 static int
1042 cifs_posix_lock_test(struct file *file, struct file_lock *flock)
1043 {
1044         int rc = 0;
1045         struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
1046         unsigned char saved_type = flock->fl_type;
1047
1048         if ((flock->fl_flags & FL_POSIX) == 0)
1049                 return 1;
1050
1051         down_read(&cinode->lock_sem);
1052         posix_test_lock(file, flock);
1053
1054         if (flock->fl_type == F_UNLCK && !cinode->can_cache_brlcks) {
1055                 flock->fl_type = saved_type;
1056                 rc = 1;
1057         }
1058
1059         up_read(&cinode->lock_sem);
1060         return rc;
1061 }
1062
1063 /*
1064  * Set the byte-range lock (posix style). Returns:
1065  * 1) 0, if we set the lock and don't need to request to the server;
1066  * 2) 1, if we need to request to the server;
1067  * 3) <0, if the error occurs while setting the lock.
1068  */
1069 static int
1070 cifs_posix_lock_set(struct file *file, struct file_lock *flock)
1071 {
1072         struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
1073         int rc = 1;
1074
1075         if ((flock->fl_flags & FL_POSIX) == 0)
1076                 return rc;
1077
1078 try_again:
1079         down_write(&cinode->lock_sem);
1080         if (!cinode->can_cache_brlcks) {
1081                 up_write(&cinode->lock_sem);
1082                 return rc;
1083         }
1084
1085         rc = posix_lock_file(file, flock, NULL);
1086         up_write(&cinode->lock_sem);
1087         if (rc == FILE_LOCK_DEFERRED) {
1088                 rc = wait_event_interruptible(flock->fl_wait, !flock->fl_next);
1089                 if (!rc)
1090                         goto try_again;
1091                 posix_unblock_lock(flock);
1092         }
1093         return rc;
1094 }
1095
1096 int
1097 cifs_push_mandatory_locks(struct cifsFileInfo *cfile)
1098 {
1099         unsigned int xid;
1100         int rc = 0, stored_rc;
1101         struct cifsLockInfo *li, *tmp;
1102         struct cifs_tcon *tcon;
1103         unsigned int num, max_num, max_buf;
1104         LOCKING_ANDX_RANGE *buf, *cur;
1105         int types[] = {LOCKING_ANDX_LARGE_FILES,
1106                        LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES};
1107         int i;
1108
1109         xid = get_xid();
1110         tcon = tlink_tcon(cfile->tlink);
1111
1112         /*
1113          * Accessing maxBuf is racy with cifs_reconnect - need to store value
1114          * and check it for zero before using.
1115          */
1116         max_buf = tcon->ses->server->maxBuf;
1117         if (!max_buf) {
1118                 free_xid(xid);
1119                 return -EINVAL;
1120         }
1121
1122         max_num = (max_buf - sizeof(struct smb_hdr)) /
1123                                                 sizeof(LOCKING_ANDX_RANGE);
1124         buf = kcalloc(max_num, sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1125         if (!buf) {
1126                 free_xid(xid);
1127                 return -ENOMEM;
1128         }
1129
1130         for (i = 0; i < 2; i++) {
1131                 cur = buf;
1132                 num = 0;
1133                 list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1134                         if (li->type != types[i])
1135                                 continue;
1136                         cur->Pid = cpu_to_le16(li->pid);
1137                         cur->LengthLow = cpu_to_le32((u32)li->length);
1138                         cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1139                         cur->OffsetLow = cpu_to_le32((u32)li->offset);
1140                         cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1141                         if (++num == max_num) {
1142                                 stored_rc = cifs_lockv(xid, tcon,
1143                                                        cfile->fid.netfid,
1144                                                        (__u8)li->type, 0, num,
1145                                                        buf);
1146                                 if (stored_rc)
1147                                         rc = stored_rc;
1148                                 cur = buf;
1149                                 num = 0;
1150                         } else
1151                                 cur++;
1152                 }
1153
1154                 if (num) {
1155                         stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1156                                                (__u8)types[i], 0, num, buf);
1157                         if (stored_rc)
1158                                 rc = stored_rc;
1159                 }
1160         }
1161
1162         kfree(buf);
1163         free_xid(xid);
1164         return rc;
1165 }
1166
1167 static __u32
1168 hash_lockowner(fl_owner_t owner)
1169 {
1170         return cifs_lock_secret ^ hash32_ptr((const void *)owner);
1171 }
1172
1173 struct lock_to_push {
1174         struct list_head llist;
1175         __u64 offset;
1176         __u64 length;
1177         __u32 pid;
1178         __u16 netfid;
1179         __u8 type;
1180 };
1181
1182 static int
1183 cifs_push_posix_locks(struct cifsFileInfo *cfile)
1184 {
1185         struct inode *inode = d_inode(cfile->dentry);
1186         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1187         struct file_lock *flock;
1188         struct file_lock_context *flctx = inode->i_flctx;
1189         unsigned int count = 0, i;
1190         int rc = 0, xid, type;
1191         struct list_head locks_to_send, *el;
1192         struct lock_to_push *lck, *tmp;
1193         __u64 length;
1194
1195         xid = get_xid();
1196
1197         if (!flctx)
1198                 goto out;
1199
1200         spin_lock(&flctx->flc_lock);
1201         list_for_each(el, &flctx->flc_posix) {
1202                 count++;
1203         }
1204         spin_unlock(&flctx->flc_lock);
1205
1206         INIT_LIST_HEAD(&locks_to_send);
1207
1208         /*
1209          * Allocating count locks is enough because no FL_POSIX locks can be
1210          * added to the list while we are holding cinode->lock_sem that
1211          * protects locking operations of this inode.
1212          */
1213         for (i = 0; i < count; i++) {
1214                 lck = kmalloc(sizeof(struct lock_to_push), GFP_KERNEL);
1215                 if (!lck) {
1216                         rc = -ENOMEM;
1217                         goto err_out;
1218                 }
1219                 list_add_tail(&lck->llist, &locks_to_send);
1220         }
1221
1222         el = locks_to_send.next;
1223         spin_lock(&flctx->flc_lock);
1224         list_for_each_entry(flock, &flctx->flc_posix, fl_list) {
1225                 if (el == &locks_to_send) {
1226                         /*
1227                          * The list ended. We don't have enough allocated
1228                          * structures - something is really wrong.
1229                          */
1230                         cifs_dbg(VFS, "Can't push all brlocks!\n");
1231                         break;
1232                 }
1233                 length = 1 + flock->fl_end - flock->fl_start;
1234                 if (flock->fl_type == F_RDLCK || flock->fl_type == F_SHLCK)
1235                         type = CIFS_RDLCK;
1236                 else
1237                         type = CIFS_WRLCK;
1238                 lck = list_entry(el, struct lock_to_push, llist);
1239                 lck->pid = hash_lockowner(flock->fl_owner);
1240                 lck->netfid = cfile->fid.netfid;
1241                 lck->length = length;
1242                 lck->type = type;
1243                 lck->offset = flock->fl_start;
1244         }
1245         spin_unlock(&flctx->flc_lock);
1246
1247         list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1248                 int stored_rc;
1249
1250                 stored_rc = CIFSSMBPosixLock(xid, tcon, lck->netfid, lck->pid,
1251                                              lck->offset, lck->length, NULL,
1252                                              lck->type, 0);
1253                 if (stored_rc)
1254                         rc = stored_rc;
1255                 list_del(&lck->llist);
1256                 kfree(lck);
1257         }
1258
1259 out:
1260         free_xid(xid);
1261         return rc;
1262 err_out:
1263         list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1264                 list_del(&lck->llist);
1265                 kfree(lck);
1266         }
1267         goto out;
1268 }
1269
1270 static int
1271 cifs_push_locks(struct cifsFileInfo *cfile)
1272 {
1273         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
1274         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1275         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1276         int rc = 0;
1277
1278         /* we are going to update can_cache_brlcks here - need a write access */
1279         down_write(&cinode->lock_sem);
1280         if (!cinode->can_cache_brlcks) {
1281                 up_write(&cinode->lock_sem);
1282                 return rc;
1283         }
1284
1285         if (cap_unix(tcon->ses) &&
1286             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1287             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1288                 rc = cifs_push_posix_locks(cfile);
1289         else
1290                 rc = tcon->ses->server->ops->push_mand_locks(cfile);
1291
1292         cinode->can_cache_brlcks = false;
1293         up_write(&cinode->lock_sem);
1294         return rc;
1295 }
1296
1297 static void
1298 cifs_read_flock(struct file_lock *flock, __u32 *type, int *lock, int *unlock,
1299                 bool *wait_flag, struct TCP_Server_Info *server)
1300 {
1301         if (flock->fl_flags & FL_POSIX)
1302                 cifs_dbg(FYI, "Posix\n");
1303         if (flock->fl_flags & FL_FLOCK)
1304                 cifs_dbg(FYI, "Flock\n");
1305         if (flock->fl_flags & FL_SLEEP) {
1306                 cifs_dbg(FYI, "Blocking lock\n");
1307                 *wait_flag = true;
1308         }
1309         if (flock->fl_flags & FL_ACCESS)
1310                 cifs_dbg(FYI, "Process suspended by mandatory locking - not implemented yet\n");
1311         if (flock->fl_flags & FL_LEASE)
1312                 cifs_dbg(FYI, "Lease on file - not implemented yet\n");
1313         if (flock->fl_flags &
1314             (~(FL_POSIX | FL_FLOCK | FL_SLEEP |
1315                FL_ACCESS | FL_LEASE | FL_CLOSE)))
1316                 cifs_dbg(FYI, "Unknown lock flags 0x%x\n", flock->fl_flags);
1317
1318         *type = server->vals->large_lock_type;
1319         if (flock->fl_type == F_WRLCK) {
1320                 cifs_dbg(FYI, "F_WRLCK\n");
1321                 *type |= server->vals->exclusive_lock_type;
1322                 *lock = 1;
1323         } else if (flock->fl_type == F_UNLCK) {
1324                 cifs_dbg(FYI, "F_UNLCK\n");
1325                 *type |= server->vals->unlock_lock_type;
1326                 *unlock = 1;
1327                 /* Check if unlock includes more than one lock range */
1328         } else if (flock->fl_type == F_RDLCK) {
1329                 cifs_dbg(FYI, "F_RDLCK\n");
1330                 *type |= server->vals->shared_lock_type;
1331                 *lock = 1;
1332         } else if (flock->fl_type == F_EXLCK) {
1333                 cifs_dbg(FYI, "F_EXLCK\n");
1334                 *type |= server->vals->exclusive_lock_type;
1335                 *lock = 1;
1336         } else if (flock->fl_type == F_SHLCK) {
1337                 cifs_dbg(FYI, "F_SHLCK\n");
1338                 *type |= server->vals->shared_lock_type;
1339                 *lock = 1;
1340         } else
1341                 cifs_dbg(FYI, "Unknown type of lock\n");
1342 }
1343
1344 static int
1345 cifs_getlk(struct file *file, struct file_lock *flock, __u32 type,
1346            bool wait_flag, bool posix_lck, unsigned int xid)
1347 {
1348         int rc = 0;
1349         __u64 length = 1 + flock->fl_end - flock->fl_start;
1350         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1351         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1352         struct TCP_Server_Info *server = tcon->ses->server;
1353         __u16 netfid = cfile->fid.netfid;
1354
1355         if (posix_lck) {
1356                 int posix_lock_type;
1357
1358                 rc = cifs_posix_lock_test(file, flock);
1359                 if (!rc)
1360                         return rc;
1361
1362                 if (type & server->vals->shared_lock_type)
1363                         posix_lock_type = CIFS_RDLCK;
1364                 else
1365                         posix_lock_type = CIFS_WRLCK;
1366                 rc = CIFSSMBPosixLock(xid, tcon, netfid,
1367                                       hash_lockowner(flock->fl_owner),
1368                                       flock->fl_start, length, flock,
1369                                       posix_lock_type, wait_flag);
1370                 return rc;
1371         }
1372
1373         rc = cifs_lock_test(cfile, flock->fl_start, length, type, flock);
1374         if (!rc)
1375                 return rc;
1376
1377         /* BB we could chain these into one lock request BB */
1378         rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length, type,
1379                                     1, 0, false);
1380         if (rc == 0) {
1381                 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1382                                             type, 0, 1, false);
1383                 flock->fl_type = F_UNLCK;
1384                 if (rc != 0)
1385                         cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1386                                  rc);
1387                 return 0;
1388         }
1389
1390         if (type & server->vals->shared_lock_type) {
1391                 flock->fl_type = F_WRLCK;
1392                 return 0;
1393         }
1394
1395         type &= ~server->vals->exclusive_lock_type;
1396
1397         rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1398                                     type | server->vals->shared_lock_type,
1399                                     1, 0, false);
1400         if (rc == 0) {
1401                 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1402                         type | server->vals->shared_lock_type, 0, 1, false);
1403                 flock->fl_type = F_RDLCK;
1404                 if (rc != 0)
1405                         cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1406                                  rc);
1407         } else
1408                 flock->fl_type = F_WRLCK;
1409
1410         return 0;
1411 }
1412
1413 void
1414 cifs_move_llist(struct list_head *source, struct list_head *dest)
1415 {
1416         struct list_head *li, *tmp;
1417         list_for_each_safe(li, tmp, source)
1418                 list_move(li, dest);
1419 }
1420
1421 void
1422 cifs_free_llist(struct list_head *llist)
1423 {
1424         struct cifsLockInfo *li, *tmp;
1425         list_for_each_entry_safe(li, tmp, llist, llist) {
1426                 cifs_del_lock_waiters(li);
1427                 list_del(&li->llist);
1428                 kfree(li);
1429         }
1430 }
1431
1432 int
1433 cifs_unlock_range(struct cifsFileInfo *cfile, struct file_lock *flock,
1434                   unsigned int xid)
1435 {
1436         int rc = 0, stored_rc;
1437         int types[] = {LOCKING_ANDX_LARGE_FILES,
1438                        LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES};
1439         unsigned int i;
1440         unsigned int max_num, num, max_buf;
1441         LOCKING_ANDX_RANGE *buf, *cur;
1442         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1443         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1444         struct cifsLockInfo *li, *tmp;
1445         __u64 length = 1 + flock->fl_end - flock->fl_start;
1446         struct list_head tmp_llist;
1447
1448         INIT_LIST_HEAD(&tmp_llist);
1449
1450         /*
1451          * Accessing maxBuf is racy with cifs_reconnect - need to store value
1452          * and check it for zero before using.
1453          */
1454         max_buf = tcon->ses->server->maxBuf;
1455         if (!max_buf)
1456                 return -EINVAL;
1457
1458         max_num = (max_buf - sizeof(struct smb_hdr)) /
1459                                                 sizeof(LOCKING_ANDX_RANGE);
1460         buf = kcalloc(max_num, sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1461         if (!buf)
1462                 return -ENOMEM;
1463
1464         down_write(&cinode->lock_sem);
1465         for (i = 0; i < 2; i++) {
1466                 cur = buf;
1467                 num = 0;
1468                 list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1469                         if (flock->fl_start > li->offset ||
1470                             (flock->fl_start + length) <
1471                             (li->offset + li->length))
1472                                 continue;
1473                         if (current->tgid != li->pid)
1474                                 continue;
1475                         if (types[i] != li->type)
1476                                 continue;
1477                         if (cinode->can_cache_brlcks) {
1478                                 /*
1479                                  * We can cache brlock requests - simply remove
1480                                  * a lock from the file's list.
1481                                  */
1482                                 list_del(&li->llist);
1483                                 cifs_del_lock_waiters(li);
1484                                 kfree(li);
1485                                 continue;
1486                         }
1487                         cur->Pid = cpu_to_le16(li->pid);
1488                         cur->LengthLow = cpu_to_le32((u32)li->length);
1489                         cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1490                         cur->OffsetLow = cpu_to_le32((u32)li->offset);
1491                         cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1492                         /*
1493                          * We need to save a lock here to let us add it again to
1494                          * the file's list if the unlock range request fails on
1495                          * the server.
1496                          */
1497                         list_move(&li->llist, &tmp_llist);
1498                         if (++num == max_num) {
1499                                 stored_rc = cifs_lockv(xid, tcon,
1500                                                        cfile->fid.netfid,
1501                                                        li->type, num, 0, buf);
1502                                 if (stored_rc) {
1503                                         /*
1504                                          * We failed on the unlock range
1505                                          * request - add all locks from the tmp
1506                                          * list to the head of the file's list.
1507                                          */
1508                                         cifs_move_llist(&tmp_llist,
1509                                                         &cfile->llist->locks);
1510                                         rc = stored_rc;
1511                                 } else
1512                                         /*
1513                                          * The unlock range request succeed -
1514                                          * free the tmp list.
1515                                          */
1516                                         cifs_free_llist(&tmp_llist);
1517                                 cur = buf;
1518                                 num = 0;
1519                         } else
1520                                 cur++;
1521                 }
1522                 if (num) {
1523                         stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1524                                                types[i], num, 0, buf);
1525                         if (stored_rc) {
1526                                 cifs_move_llist(&tmp_llist,
1527                                                 &cfile->llist->locks);
1528                                 rc = stored_rc;
1529                         } else
1530                                 cifs_free_llist(&tmp_llist);
1531                 }
1532         }
1533
1534         up_write(&cinode->lock_sem);
1535         kfree(buf);
1536         return rc;
1537 }
1538
1539 static int
1540 cifs_setlk(struct file *file, struct file_lock *flock, __u32 type,
1541            bool wait_flag, bool posix_lck, int lock, int unlock,
1542            unsigned int xid)
1543 {
1544         int rc = 0;
1545         __u64 length = 1 + flock->fl_end - flock->fl_start;
1546         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1547         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1548         struct TCP_Server_Info *server = tcon->ses->server;
1549         struct inode *inode = d_inode(cfile->dentry);
1550
1551         if (posix_lck) {
1552                 int posix_lock_type;
1553
1554                 rc = cifs_posix_lock_set(file, flock);
1555                 if (!rc || rc < 0)
1556                         return rc;
1557
1558                 if (type & server->vals->shared_lock_type)
1559                         posix_lock_type = CIFS_RDLCK;
1560                 else
1561                         posix_lock_type = CIFS_WRLCK;
1562
1563                 if (unlock == 1)
1564                         posix_lock_type = CIFS_UNLCK;
1565
1566                 rc = CIFSSMBPosixLock(xid, tcon, cfile->fid.netfid,
1567                                       hash_lockowner(flock->fl_owner),
1568                                       flock->fl_start, length,
1569                                       NULL, posix_lock_type, wait_flag);
1570                 goto out;
1571         }
1572
1573         if (lock) {
1574                 struct cifsLockInfo *lock;
1575
1576                 lock = cifs_lock_init(flock->fl_start, length, type);
1577                 if (!lock)
1578                         return -ENOMEM;
1579
1580                 rc = cifs_lock_add_if(cfile, lock, wait_flag);
1581                 if (rc < 0) {
1582                         kfree(lock);
1583                         return rc;
1584                 }
1585                 if (!rc)
1586                         goto out;
1587
1588                 /*
1589                  * Windows 7 server can delay breaking lease from read to None
1590                  * if we set a byte-range lock on a file - break it explicitly
1591                  * before sending the lock to the server to be sure the next
1592                  * read won't conflict with non-overlapted locks due to
1593                  * pagereading.
1594                  */
1595                 if (!CIFS_CACHE_WRITE(CIFS_I(inode)) &&
1596                                         CIFS_CACHE_READ(CIFS_I(inode))) {
1597                         cifs_zap_mapping(inode);
1598                         cifs_dbg(FYI, "Set no oplock for inode=%p due to mand locks\n",
1599                                  inode);
1600                         CIFS_I(inode)->oplock = 0;
1601                 }
1602
1603                 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1604                                             type, 1, 0, wait_flag);
1605                 if (rc) {
1606                         kfree(lock);
1607                         return rc;
1608                 }
1609
1610                 cifs_lock_add(cfile, lock);
1611         } else if (unlock)
1612                 rc = server->ops->mand_unlock_range(cfile, flock, xid);
1613
1614 out:
1615         if (flock->fl_flags & FL_POSIX && !rc)
1616                 rc = locks_lock_file_wait(file, flock);
1617         return rc;
1618 }
1619
1620 int cifs_lock(struct file *file, int cmd, struct file_lock *flock)
1621 {
1622         int rc, xid;
1623         int lock = 0, unlock = 0;
1624         bool wait_flag = false;
1625         bool posix_lck = false;
1626         struct cifs_sb_info *cifs_sb;
1627         struct cifs_tcon *tcon;
1628         struct cifsInodeInfo *cinode;
1629         struct cifsFileInfo *cfile;
1630         __u16 netfid;
1631         __u32 type;
1632
1633         rc = -EACCES;
1634         xid = get_xid();
1635
1636         cifs_dbg(FYI, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld end: %lld\n",
1637                  cmd, flock->fl_flags, flock->fl_type,
1638                  flock->fl_start, flock->fl_end);
1639
1640         cfile = (struct cifsFileInfo *)file->private_data;
1641         tcon = tlink_tcon(cfile->tlink);
1642
1643         cifs_read_flock(flock, &type, &lock, &unlock, &wait_flag,
1644                         tcon->ses->server);
1645
1646         cifs_sb = CIFS_FILE_SB(file);
1647         netfid = cfile->fid.netfid;
1648         cinode = CIFS_I(file_inode(file));
1649
1650         if (cap_unix(tcon->ses) &&
1651             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1652             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1653                 posix_lck = true;
1654         /*
1655          * BB add code here to normalize offset and length to account for
1656          * negative length which we can not accept over the wire.
1657          */
1658         if (IS_GETLK(cmd)) {
1659                 rc = cifs_getlk(file, flock, type, wait_flag, posix_lck, xid);
1660                 free_xid(xid);
1661                 return rc;
1662         }
1663
1664         if (!lock && !unlock) {
1665                 /*
1666                  * if no lock or unlock then nothing to do since we do not
1667                  * know what it is
1668                  */
1669                 free_xid(xid);
1670                 return -EOPNOTSUPP;
1671         }
1672
1673         rc = cifs_setlk(file, flock, type, wait_flag, posix_lck, lock, unlock,
1674                         xid);
1675         free_xid(xid);
1676         return rc;
1677 }
1678
1679 /*
1680  * update the file size (if needed) after a write. Should be called with
1681  * the inode->i_lock held
1682  */
1683 void
1684 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
1685                       unsigned int bytes_written)
1686 {
1687         loff_t end_of_write = offset + bytes_written;
1688
1689         if (end_of_write > cifsi->server_eof)
1690                 cifsi->server_eof = end_of_write;
1691 }
1692
1693 static ssize_t
1694 cifs_write(struct cifsFileInfo *open_file, __u32 pid, const char *write_data,
1695            size_t write_size, loff_t *offset)
1696 {
1697         int rc = 0;
1698         unsigned int bytes_written = 0;
1699         unsigned int total_written;
1700         struct cifs_sb_info *cifs_sb;
1701         struct cifs_tcon *tcon;
1702         struct TCP_Server_Info *server;
1703         unsigned int xid;
1704         struct dentry *dentry = open_file->dentry;
1705         struct cifsInodeInfo *cifsi = CIFS_I(d_inode(dentry));
1706         struct cifs_io_parms io_parms;
1707
1708         cifs_sb = CIFS_SB(dentry->d_sb);
1709
1710         cifs_dbg(FYI, "write %zd bytes to offset %lld of %pd\n",
1711                  write_size, *offset, dentry);
1712
1713         tcon = tlink_tcon(open_file->tlink);
1714         server = tcon->ses->server;
1715
1716         if (!server->ops->sync_write)
1717                 return -ENOSYS;
1718
1719         xid = get_xid();
1720
1721         for (total_written = 0; write_size > total_written;
1722              total_written += bytes_written) {
1723                 rc = -EAGAIN;
1724                 while (rc == -EAGAIN) {
1725                         struct kvec iov[2];
1726                         unsigned int len;
1727
1728                         if (open_file->invalidHandle) {
1729                                 /* we could deadlock if we called
1730                                    filemap_fdatawait from here so tell
1731                                    reopen_file not to flush data to
1732                                    server now */
1733                                 rc = cifs_reopen_file(open_file, false);
1734                                 if (rc != 0)
1735                                         break;
1736                         }
1737
1738                         len = min(server->ops->wp_retry_size(d_inode(dentry)),
1739                                   (unsigned int)write_size - total_written);
1740                         /* iov[0] is reserved for smb header */
1741                         iov[1].iov_base = (char *)write_data + total_written;
1742                         iov[1].iov_len = len;
1743                         io_parms.pid = pid;
1744                         io_parms.tcon = tcon;
1745                         io_parms.offset = *offset;
1746                         io_parms.length = len;
1747                         rc = server->ops->sync_write(xid, &open_file->fid,
1748                                         &io_parms, &bytes_written, iov, 1);
1749                 }
1750                 if (rc || (bytes_written == 0)) {
1751                         if (total_written)
1752                                 break;
1753                         else {
1754                                 free_xid(xid);
1755                                 return rc;
1756                         }
1757                 } else {
1758                         spin_lock(&d_inode(dentry)->i_lock);
1759                         cifs_update_eof(cifsi, *offset, bytes_written);
1760                         spin_unlock(&d_inode(dentry)->i_lock);
1761                         *offset += bytes_written;
1762                 }
1763         }
1764
1765         cifs_stats_bytes_written(tcon, total_written);
1766
1767         if (total_written > 0) {
1768                 spin_lock(&d_inode(dentry)->i_lock);
1769                 if (*offset > d_inode(dentry)->i_size)
1770                         i_size_write(d_inode(dentry), *offset);
1771                 spin_unlock(&d_inode(dentry)->i_lock);
1772         }
1773         mark_inode_dirty_sync(d_inode(dentry));
1774         free_xid(xid);
1775         return total_written;
1776 }
1777
1778 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1779                                         bool fsuid_only)
1780 {
1781         struct cifsFileInfo *open_file = NULL;
1782         struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1783         struct cifs_tcon *tcon = cifs_sb_master_tcon(cifs_sb);
1784
1785         /* only filter by fsuid on multiuser mounts */
1786         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1787                 fsuid_only = false;
1788
1789         spin_lock(&tcon->open_file_lock);
1790         /* we could simply get the first_list_entry since write-only entries
1791            are always at the end of the list but since the first entry might
1792            have a close pending, we go through the whole list */
1793         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1794                 if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
1795                         continue;
1796                 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1797                         if (!open_file->invalidHandle) {
1798                                 /* found a good file */
1799                                 /* lock it so it will not be closed on us */
1800                                 cifsFileInfo_get(open_file);
1801                                 spin_unlock(&tcon->open_file_lock);
1802                                 return open_file;
1803                         } /* else might as well continue, and look for
1804                              another, or simply have the caller reopen it
1805                              again rather than trying to fix this handle */
1806                 } else /* write only file */
1807                         break; /* write only files are last so must be done */
1808         }
1809         spin_unlock(&tcon->open_file_lock);
1810         return NULL;
1811 }
1812
1813 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1814                                         bool fsuid_only)
1815 {
1816         struct cifsFileInfo *open_file, *inv_file = NULL;
1817         struct cifs_sb_info *cifs_sb;
1818         struct cifs_tcon *tcon;
1819         bool any_available = false;
1820         int rc;
1821         unsigned int refind = 0;
1822
1823         /* Having a null inode here (because mapping->host was set to zero by
1824         the VFS or MM) should not happen but we had reports of on oops (due to
1825         it being zero) during stress testcases so we need to check for it */
1826
1827         if (cifs_inode == NULL) {
1828                 cifs_dbg(VFS, "Null inode passed to cifs_writeable_file\n");
1829                 dump_stack();
1830                 return NULL;
1831         }
1832
1833         cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1834         tcon = cifs_sb_master_tcon(cifs_sb);
1835
1836         /* only filter by fsuid on multiuser mounts */
1837         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1838                 fsuid_only = false;
1839
1840         spin_lock(&tcon->open_file_lock);
1841 refind_writable:
1842         if (refind > MAX_REOPEN_ATT) {
1843                 spin_unlock(&tcon->open_file_lock);
1844                 return NULL;
1845         }
1846         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1847                 if (!any_available && open_file->pid != current->tgid)
1848                         continue;
1849                 if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
1850                         continue;
1851                 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1852                         if (!open_file->invalidHandle) {
1853                                 /* found a good writable file */
1854                                 cifsFileInfo_get(open_file);
1855                                 spin_unlock(&tcon->open_file_lock);
1856                                 return open_file;
1857                         } else {
1858                                 if (!inv_file)
1859                                         inv_file = open_file;
1860                         }
1861                 }
1862         }
1863         /* couldn't find useable FH with same pid, try any available */
1864         if (!any_available) {
1865                 any_available = true;
1866                 goto refind_writable;
1867         }
1868
1869         if (inv_file) {
1870                 any_available = false;
1871                 cifsFileInfo_get(inv_file);
1872         }
1873
1874         spin_unlock(&tcon->open_file_lock);
1875
1876         if (inv_file) {
1877                 rc = cifs_reopen_file(inv_file, false);
1878                 if (!rc)
1879                         return inv_file;
1880                 else {
1881                         spin_lock(&tcon->open_file_lock);
1882                         list_move_tail(&inv_file->flist,
1883                                         &cifs_inode->openFileList);
1884                         spin_unlock(&tcon->open_file_lock);
1885                         cifsFileInfo_put(inv_file);
1886                         ++refind;
1887                         inv_file = NULL;
1888                         spin_lock(&tcon->open_file_lock);
1889                         goto refind_writable;
1890                 }
1891         }
1892
1893         return NULL;
1894 }
1895
1896 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1897 {
1898         struct address_space *mapping = page->mapping;
1899         loff_t offset = (loff_t)page->index << PAGE_SHIFT;
1900         char *write_data;
1901         int rc = -EFAULT;
1902         int bytes_written = 0;
1903         struct inode *inode;
1904         struct cifsFileInfo *open_file;
1905
1906         if (!mapping || !mapping->host)
1907                 return -EFAULT;
1908
1909         inode = page->mapping->host;
1910
1911         offset += (loff_t)from;
1912         write_data = kmap(page);
1913         write_data += from;
1914
1915         if ((to > PAGE_SIZE) || (from > to)) {
1916                 kunmap(page);
1917                 return -EIO;
1918         }
1919
1920         /* racing with truncate? */
1921         if (offset > mapping->host->i_size) {
1922                 kunmap(page);
1923                 return 0; /* don't care */
1924         }
1925
1926         /* check to make sure that we are not extending the file */
1927         if (mapping->host->i_size - offset < (loff_t)to)
1928                 to = (unsigned)(mapping->host->i_size - offset);
1929
1930         open_file = find_writable_file(CIFS_I(mapping->host), false);
1931         if (open_file) {
1932                 bytes_written = cifs_write(open_file, open_file->pid,
1933                                            write_data, to - from, &offset);
1934                 cifsFileInfo_put(open_file);
1935                 /* Does mm or vfs already set times? */
1936                 inode->i_atime = inode->i_mtime = current_time(inode);
1937                 if ((bytes_written > 0) && (offset))
1938                         rc = 0;
1939                 else if (bytes_written < 0)
1940                         rc = bytes_written;
1941         } else {
1942                 cifs_dbg(FYI, "No writeable filehandles for inode\n");
1943                 rc = -EIO;
1944         }
1945
1946         kunmap(page);
1947         return rc;
1948 }
1949
1950 static struct cifs_writedata *
1951 wdata_alloc_and_fillpages(pgoff_t tofind, struct address_space *mapping,
1952                           pgoff_t end, pgoff_t *index,
1953                           unsigned int *found_pages)
1954 {
1955         unsigned int nr_pages;
1956         struct page **pages;
1957         struct cifs_writedata *wdata;
1958
1959         wdata = cifs_writedata_alloc((unsigned int)tofind,
1960                                      cifs_writev_complete);
1961         if (!wdata)
1962                 return NULL;
1963
1964         /*
1965          * find_get_pages_tag seems to return a max of 256 on each
1966          * iteration, so we must call it several times in order to
1967          * fill the array or the wsize is effectively limited to
1968          * 256 * PAGE_SIZE.
1969          */
1970         *found_pages = 0;
1971         pages = wdata->pages;
1972         do {
1973                 nr_pages = find_get_pages_tag(mapping, index,
1974                                               PAGECACHE_TAG_DIRTY, tofind,
1975                                               pages);
1976                 *found_pages += nr_pages;
1977                 tofind -= nr_pages;
1978                 pages += nr_pages;
1979         } while (nr_pages && tofind && *index <= end);
1980
1981         return wdata;
1982 }
1983
1984 static unsigned int
1985 wdata_prepare_pages(struct cifs_writedata *wdata, unsigned int found_pages,
1986                     struct address_space *mapping,
1987                     struct writeback_control *wbc,
1988                     pgoff_t end, pgoff_t *index, pgoff_t *next, bool *done)
1989 {
1990         unsigned int nr_pages = 0, i;
1991         struct page *page;
1992
1993         for (i = 0; i < found_pages; i++) {
1994                 page = wdata->pages[i];
1995                 /*
1996                  * At this point we hold neither mapping->tree_lock nor
1997                  * lock on the page itself: the page may be truncated or
1998                  * invalidated (changing page->mapping to NULL), or even
1999                  * swizzled back from swapper_space to tmpfs file
2000                  * mapping
2001                  */
2002
2003                 if (nr_pages == 0)
2004                         lock_page(page);
2005                 else if (!trylock_page(page))
2006                         break;
2007
2008                 if (unlikely(page->mapping != mapping)) {
2009                         unlock_page(page);
2010                         break;
2011                 }
2012
2013                 if (!wbc->range_cyclic && page->index > end) {
2014                         *done = true;
2015                         unlock_page(page);
2016                         break;
2017                 }
2018
2019                 if (*next && (page->index != *next)) {
2020                         /* Not next consecutive page */
2021                         unlock_page(page);
2022                         break;
2023                 }
2024
2025                 if (wbc->sync_mode != WB_SYNC_NONE)
2026                         wait_on_page_writeback(page);
2027
2028                 if (PageWriteback(page) ||
2029                                 !clear_page_dirty_for_io(page)) {
2030                         unlock_page(page);
2031                         break;
2032                 }
2033
2034                 /*
2035                  * This actually clears the dirty bit in the radix tree.
2036                  * See cifs_writepage() for more commentary.
2037                  */
2038                 set_page_writeback(page);
2039                 if (page_offset(page) >= i_size_read(mapping->host)) {
2040                         *done = true;
2041                         unlock_page(page);
2042                         end_page_writeback(page);
2043                         break;
2044                 }
2045
2046                 wdata->pages[i] = page;
2047                 *next = page->index + 1;
2048                 ++nr_pages;
2049         }
2050
2051         /* reset index to refind any pages skipped */
2052         if (nr_pages == 0)
2053                 *index = wdata->pages[0]->index + 1;
2054
2055         /* put any pages we aren't going to use */
2056         for (i = nr_pages; i < found_pages; i++) {
2057                 put_page(wdata->pages[i]);
2058                 wdata->pages[i] = NULL;
2059         }
2060
2061         return nr_pages;
2062 }
2063
2064 static int
2065 wdata_send_pages(struct cifs_writedata *wdata, unsigned int nr_pages,
2066                  struct address_space *mapping, struct writeback_control *wbc)
2067 {
2068         int rc = 0;
2069         struct TCP_Server_Info *server;
2070         unsigned int i;
2071
2072         wdata->sync_mode = wbc->sync_mode;
2073         wdata->nr_pages = nr_pages;
2074         wdata->offset = page_offset(wdata->pages[0]);
2075         wdata->pagesz = PAGE_SIZE;
2076         wdata->tailsz = min(i_size_read(mapping->host) -
2077                         page_offset(wdata->pages[nr_pages - 1]),
2078                         (loff_t)PAGE_SIZE);
2079         wdata->bytes = ((nr_pages - 1) * PAGE_SIZE) + wdata->tailsz;
2080
2081         if (wdata->cfile != NULL)
2082                 cifsFileInfo_put(wdata->cfile);
2083         wdata->cfile = find_writable_file(CIFS_I(mapping->host), false);
2084         if (!wdata->cfile) {
2085                 cifs_dbg(VFS, "No writable handles for inode\n");
2086                 rc = -EBADF;
2087         } else {
2088                 wdata->pid = wdata->cfile->pid;
2089                 server = tlink_tcon(wdata->cfile->tlink)->ses->server;
2090                 rc = server->ops->async_writev(wdata, cifs_writedata_release);
2091         }
2092
2093         for (i = 0; i < nr_pages; ++i)
2094                 unlock_page(wdata->pages[i]);
2095
2096         return rc;
2097 }
2098
2099 static int cifs_writepages(struct address_space *mapping,
2100                            struct writeback_control *wbc)
2101 {
2102         struct cifs_sb_info *cifs_sb = CIFS_SB(mapping->host->i_sb);
2103         struct TCP_Server_Info *server;
2104         bool done = false, scanned = false, range_whole = false;
2105         pgoff_t end, index;
2106         struct cifs_writedata *wdata;
2107         int rc = 0;
2108
2109         /*
2110          * If wsize is smaller than the page cache size, default to writing
2111          * one page at a time via cifs_writepage
2112          */
2113         if (cifs_sb->wsize < PAGE_SIZE)
2114                 return generic_writepages(mapping, wbc);
2115
2116         if (wbc->range_cyclic) {
2117                 index = mapping->writeback_index; /* Start from prev offset */
2118                 end = -1;
2119         } else {
2120                 index = wbc->range_start >> PAGE_SHIFT;
2121                 end = wbc->range_end >> PAGE_SHIFT;
2122                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2123                         range_whole = true;
2124                 scanned = true;
2125         }
2126         server = cifs_sb_master_tcon(cifs_sb)->ses->server;
2127 retry:
2128         while (!done && index <= end) {
2129                 unsigned int i, nr_pages, found_pages, wsize, credits;
2130                 pgoff_t next = 0, tofind, saved_index = index;
2131
2132                 rc = server->ops->wait_mtu_credits(server, cifs_sb->wsize,
2133                                                    &wsize, &credits);
2134                 if (rc)
2135                         break;
2136
2137                 tofind = min((wsize / PAGE_SIZE) - 1, end - index) + 1;
2138
2139                 wdata = wdata_alloc_and_fillpages(tofind, mapping, end, &index,
2140                                                   &found_pages);
2141                 if (!wdata) {
2142                         rc = -ENOMEM;
2143                         add_credits_and_wake_if(server, credits, 0);
2144                         break;
2145                 }
2146
2147                 if (found_pages == 0) {
2148                         kref_put(&wdata->refcount, cifs_writedata_release);
2149                         add_credits_and_wake_if(server, credits, 0);
2150                         break;
2151                 }
2152
2153                 nr_pages = wdata_prepare_pages(wdata, found_pages, mapping, wbc,
2154                                                end, &index, &next, &done);
2155
2156                 /* nothing to write? */
2157                 if (nr_pages == 0) {
2158                         kref_put(&wdata->refcount, cifs_writedata_release);
2159                         add_credits_and_wake_if(server, credits, 0);
2160                         continue;
2161                 }
2162
2163                 wdata->credits = credits;
2164
2165                 rc = wdata_send_pages(wdata, nr_pages, mapping, wbc);
2166
2167                 /* send failure -- clean up the mess */
2168                 if (rc != 0) {
2169                         add_credits_and_wake_if(server, wdata->credits, 0);
2170                         for (i = 0; i < nr_pages; ++i) {
2171                                 if (rc == -EAGAIN)
2172                                         redirty_page_for_writepage(wbc,
2173                                                            wdata->pages[i]);
2174                                 else
2175                                         SetPageError(wdata->pages[i]);
2176                                 end_page_writeback(wdata->pages[i]);
2177                                 put_page(wdata->pages[i]);
2178                         }
2179                         if (rc != -EAGAIN)
2180                                 mapping_set_error(mapping, rc);
2181                 }
2182                 kref_put(&wdata->refcount, cifs_writedata_release);
2183
2184                 if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN) {
2185                         index = saved_index;
2186                         continue;
2187                 }
2188
2189                 wbc->nr_to_write -= nr_pages;
2190                 if (wbc->nr_to_write <= 0)
2191                         done = true;
2192
2193                 index = next;
2194         }
2195
2196         if (!scanned && !done) {
2197                 /*
2198                  * We hit the last page and there is more work to be done: wrap
2199                  * back to the start of the file
2200                  */
2201                 scanned = true;
2202                 index = 0;
2203                 goto retry;
2204         }
2205
2206         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2207                 mapping->writeback_index = index;
2208
2209         return rc;
2210 }
2211
2212 static int
2213 cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
2214 {
2215         int rc;
2216         unsigned int xid;
2217
2218         xid = get_xid();
2219 /* BB add check for wbc flags */
2220         get_page(page);
2221         if (!PageUptodate(page))
2222                 cifs_dbg(FYI, "ppw - page not up to date\n");
2223
2224         /*
2225          * Set the "writeback" flag, and clear "dirty" in the radix tree.
2226          *
2227          * A writepage() implementation always needs to do either this,
2228          * or re-dirty the page with "redirty_page_for_writepage()" in
2229          * the case of a failure.
2230          *
2231          * Just unlocking the page will cause the radix tree tag-bits
2232          * to fail to update with the state of the page correctly.
2233          */
2234         set_page_writeback(page);
2235 retry_write:
2236         rc = cifs_partialpagewrite(page, 0, PAGE_SIZE);
2237         if (rc == -EAGAIN && wbc->sync_mode == WB_SYNC_ALL)
2238                 goto retry_write;
2239         else if (rc == -EAGAIN)
2240                 redirty_page_for_writepage(wbc, page);
2241         else if (rc != 0)
2242                 SetPageError(page);
2243         else
2244                 SetPageUptodate(page);
2245         end_page_writeback(page);
2246         put_page(page);
2247         free_xid(xid);
2248         return rc;
2249 }
2250
2251 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
2252 {
2253         int rc = cifs_writepage_locked(page, wbc);
2254         unlock_page(page);
2255         return rc;
2256 }
2257
2258 static int cifs_write_end(struct file *file, struct address_space *mapping,
2259                         loff_t pos, unsigned len, unsigned copied,
2260                         struct page *page, void *fsdata)
2261 {
2262         int rc;
2263         struct inode *inode = mapping->host;
2264         struct cifsFileInfo *cfile = file->private_data;
2265         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
2266         __u32 pid;
2267
2268         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2269                 pid = cfile->pid;
2270         else
2271                 pid = current->tgid;
2272
2273         cifs_dbg(FYI, "write_end for page %p from pos %lld with %d bytes\n",
2274                  page, pos, copied);
2275
2276         if (PageChecked(page)) {
2277                 if (copied == len)
2278                         SetPageUptodate(page);
2279                 ClearPageChecked(page);
2280         } else if (!PageUptodate(page) && copied == PAGE_SIZE)
2281                 SetPageUptodate(page);
2282
2283         if (!PageUptodate(page)) {
2284                 char *page_data;
2285                 unsigned offset = pos & (PAGE_SIZE - 1);
2286                 unsigned int xid;
2287
2288                 xid = get_xid();
2289                 /* this is probably better than directly calling
2290                    partialpage_write since in this function the file handle is
2291                    known which we might as well leverage */
2292                 /* BB check if anything else missing out of ppw
2293                    such as updating last write time */
2294                 page_data = kmap(page);
2295                 rc = cifs_write(cfile, pid, page_data + offset, copied, &pos);
2296                 /* if (rc < 0) should we set writebehind rc? */
2297                 kunmap(page);
2298
2299                 free_xid(xid);
2300         } else {
2301                 rc = copied;
2302                 pos += copied;
2303                 set_page_dirty(page);
2304         }
2305
2306         if (rc > 0) {
2307                 spin_lock(&inode->i_lock);
2308                 if (pos > inode->i_size)
2309                         i_size_write(inode, pos);
2310                 spin_unlock(&inode->i_lock);
2311         }
2312
2313         unlock_page(page);
2314         put_page(page);
2315
2316         return rc;
2317 }
2318
2319 int cifs_strict_fsync(struct file *file, loff_t start, loff_t end,
2320                       int datasync)
2321 {
2322         unsigned int xid;
2323         int rc = 0;
2324         struct cifs_tcon *tcon;
2325         struct TCP_Server_Info *server;
2326         struct cifsFileInfo *smbfile = file->private_data;
2327         struct inode *inode = file_inode(file);
2328         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2329
2330         rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2331         if (rc)
2332                 return rc;
2333         inode_lock(inode);
2334
2335         xid = get_xid();
2336
2337         cifs_dbg(FYI, "Sync file - name: %pD datasync: 0x%x\n",
2338                  file, datasync);
2339
2340         if (!CIFS_CACHE_READ(CIFS_I(inode))) {
2341                 rc = cifs_zap_mapping(inode);
2342                 if (rc) {
2343                         cifs_dbg(FYI, "rc: %d during invalidate phase\n", rc);
2344                         rc = 0; /* don't care about it in fsync */
2345                 }
2346         }
2347
2348         tcon = tlink_tcon(smbfile->tlink);
2349         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2350                 server = tcon->ses->server;
2351                 if (server->ops->flush)
2352                         rc = server->ops->flush(xid, tcon, &smbfile->fid);
2353                 else
2354                         rc = -ENOSYS;
2355         }
2356
2357         free_xid(xid);
2358         inode_unlock(inode);
2359         return rc;
2360 }
2361
2362 int cifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2363 {
2364         unsigned int xid;
2365         int rc = 0;
2366         struct cifs_tcon *tcon;
2367         struct TCP_Server_Info *server;
2368         struct cifsFileInfo *smbfile = file->private_data;
2369         struct cifs_sb_info *cifs_sb = CIFS_FILE_SB(file);
2370         struct inode *inode = file->f_mapping->host;
2371
2372         rc = filemap_write_and_wait_range(inode->i_mapping, start, end);
2373         if (rc)
2374                 return rc;
2375         inode_lock(inode);
2376
2377         xid = get_xid();
2378
2379         cifs_dbg(FYI, "Sync file - name: %pD datasync: 0x%x\n",
2380                  file, datasync);
2381
2382         tcon = tlink_tcon(smbfile->tlink);
2383         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2384                 server = tcon->ses->server;
2385                 if (server->ops->flush)
2386                         rc = server->ops->flush(xid, tcon, &smbfile->fid);
2387                 else
2388                         rc = -ENOSYS;
2389         }
2390
2391         free_xid(xid);
2392         inode_unlock(inode);
2393         return rc;
2394 }
2395
2396 /*
2397  * As file closes, flush all cached write data for this inode checking
2398  * for write behind errors.
2399  */
2400 int cifs_flush(struct file *file, fl_owner_t id)
2401 {
2402         struct inode *inode = file_inode(file);
2403         int rc = 0;
2404
2405         if (file->f_mode & FMODE_WRITE)
2406                 rc = filemap_write_and_wait(inode->i_mapping);
2407
2408         cifs_dbg(FYI, "Flush inode %p file %p rc %d\n", inode, file, rc);
2409
2410         return rc;
2411 }
2412
2413 static int
2414 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
2415 {
2416         int rc = 0;
2417         unsigned long i;
2418
2419         for (i = 0; i < num_pages; i++) {
2420                 pages[i] = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2421                 if (!pages[i]) {
2422                         /*
2423                          * save number of pages we have already allocated and
2424                          * return with ENOMEM error
2425                          */
2426                         num_pages = i;
2427                         rc = -ENOMEM;
2428                         break;
2429                 }
2430         }
2431
2432         if (rc) {
2433                 for (i = 0; i < num_pages; i++)
2434                         put_page(pages[i]);
2435         }
2436         return rc;
2437 }
2438
2439 static inline
2440 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
2441 {
2442         size_t num_pages;
2443         size_t clen;
2444
2445         clen = min_t(const size_t, len, wsize);
2446         num_pages = DIV_ROUND_UP(clen, PAGE_SIZE);
2447
2448         if (cur_len)
2449                 *cur_len = clen;
2450
2451         return num_pages;
2452 }
2453
2454 static void
2455 cifs_uncached_writedata_release(struct kref *refcount)
2456 {
2457         int i;
2458         struct cifs_writedata *wdata = container_of(refcount,
2459                                         struct cifs_writedata, refcount);
2460
2461         kref_put(&wdata->ctx->refcount, cifs_aio_ctx_release);
2462         for (i = 0; i < wdata->nr_pages; i++)
2463                 put_page(wdata->pages[i]);
2464         cifs_writedata_release(refcount);
2465 }
2466
2467 static void collect_uncached_write_data(struct cifs_aio_ctx *ctx);
2468
2469 static void
2470 cifs_uncached_writev_complete(struct work_struct *work)
2471 {
2472         struct cifs_writedata *wdata = container_of(work,
2473                                         struct cifs_writedata, work);
2474         struct inode *inode = d_inode(wdata->cfile->dentry);
2475         struct cifsInodeInfo *cifsi = CIFS_I(inode);
2476
2477         spin_lock(&inode->i_lock);
2478         cifs_update_eof(cifsi, wdata->offset, wdata->bytes);
2479         if (cifsi->server_eof > inode->i_size)
2480                 i_size_write(inode, cifsi->server_eof);
2481         spin_unlock(&inode->i_lock);
2482
2483         complete(&wdata->done);
2484         collect_uncached_write_data(wdata->ctx);
2485         /* the below call can possibly free the last ref to aio ctx */
2486         kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2487 }
2488
2489 static int
2490 wdata_fill_from_iovec(struct cifs_writedata *wdata, struct iov_iter *from,
2491                       size_t *len, unsigned long *num_pages)
2492 {
2493         size_t save_len, copied, bytes, cur_len = *len;
2494         unsigned long i, nr_pages = *num_pages;
2495
2496         save_len = cur_len;
2497         for (i = 0; i < nr_pages; i++) {
2498                 bytes = min_t(const size_t, cur_len, PAGE_SIZE);
2499                 copied = copy_page_from_iter(wdata->pages[i], 0, bytes, from);
2500                 cur_len -= copied;
2501                 /*
2502                  * If we didn't copy as much as we expected, then that
2503                  * may mean we trod into an unmapped area. Stop copying
2504                  * at that point. On the next pass through the big
2505                  * loop, we'll likely end up getting a zero-length
2506                  * write and bailing out of it.
2507                  */
2508                 if (copied < bytes)
2509                         break;
2510         }
2511         cur_len = save_len - cur_len;
2512         *len = cur_len;
2513
2514         /*
2515          * If we have no data to send, then that probably means that
2516          * the copy above failed altogether. That's most likely because
2517          * the address in the iovec was bogus. Return -EFAULT and let
2518          * the caller free anything we allocated and bail out.
2519          */
2520         if (!cur_len)
2521                 return -EFAULT;
2522
2523         /*
2524          * i + 1 now represents the number of pages we actually used in
2525          * the copy phase above.
2526          */
2527         *num_pages = i + 1;
2528         return 0;
2529 }
2530
2531 static int
2532 cifs_write_from_iter(loff_t offset, size_t len, struct iov_iter *from,
2533                      struct cifsFileInfo *open_file,
2534                      struct cifs_sb_info *cifs_sb, struct list_head *wdata_list,
2535                      struct cifs_aio_ctx *ctx)
2536 {
2537         int rc = 0;
2538         size_t cur_len;
2539         unsigned long nr_pages, num_pages, i;
2540         struct cifs_writedata *wdata;
2541         struct iov_iter saved_from = *from;
2542         loff_t saved_offset = offset;
2543         pid_t pid;
2544         struct TCP_Server_Info *server;
2545
2546         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2547                 pid = open_file->pid;
2548         else
2549                 pid = current->tgid;
2550
2551         server = tlink_tcon(open_file->tlink)->ses->server;
2552
2553         do {
2554                 unsigned int wsize, credits;
2555
2556                 rc = server->ops->wait_mtu_credits(server, cifs_sb->wsize,
2557                                                    &wsize, &credits);
2558                 if (rc)
2559                         break;
2560
2561                 nr_pages = get_numpages(wsize, len, &cur_len);
2562                 wdata = cifs_writedata_alloc(nr_pages,
2563                                              cifs_uncached_writev_complete);
2564                 if (!wdata) {
2565                         rc = -ENOMEM;
2566                         add_credits_and_wake_if(server, credits, 0);
2567                         break;
2568                 }
2569
2570                 rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
2571                 if (rc) {
2572                         kfree(wdata);
2573                         add_credits_and_wake_if(server, credits, 0);
2574                         break;
2575                 }
2576
2577                 num_pages = nr_pages;
2578                 rc = wdata_fill_from_iovec(wdata, from, &cur_len, &num_pages);
2579                 if (rc) {
2580                         for (i = 0; i < nr_pages; i++)
2581                                 put_page(wdata->pages[i]);
2582                         kfree(wdata);
2583                         add_credits_and_wake_if(server, credits, 0);
2584                         break;
2585                 }
2586
2587                 /*
2588                  * Bring nr_pages down to the number of pages we actually used,
2589                  * and free any pages that we didn't use.
2590                  */
2591                 for ( ; nr_pages > num_pages; nr_pages--)
2592                         put_page(wdata->pages[nr_pages - 1]);
2593
2594                 wdata->sync_mode = WB_SYNC_ALL;
2595                 wdata->nr_pages = nr_pages;
2596                 wdata->offset = (__u64)offset;
2597                 wdata->cfile = cifsFileInfo_get(open_file);
2598                 wdata->pid = pid;
2599                 wdata->bytes = cur_len;
2600                 wdata->pagesz = PAGE_SIZE;
2601                 wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
2602                 wdata->credits = credits;
2603                 wdata->ctx = ctx;
2604                 kref_get(&ctx->refcount);
2605
2606                 if (!wdata->cfile->invalidHandle ||
2607                     !(rc = cifs_reopen_file(wdata->cfile, false)))
2608                         rc = server->ops->async_writev(wdata,
2609                                         cifs_uncached_writedata_release);
2610                 if (rc) {
2611                         add_credits_and_wake_if(server, wdata->credits, 0);
2612                         kref_put(&wdata->refcount,
2613                                  cifs_uncached_writedata_release);
2614                         if (rc == -EAGAIN) {
2615                                 *from = saved_from;
2616                                 iov_iter_advance(from, offset - saved_offset);
2617                                 continue;
2618                         }
2619                         break;
2620                 }
2621
2622                 list_add_tail(&wdata->list, wdata_list);
2623                 offset += cur_len;
2624                 len -= cur_len;
2625         } while (len > 0);
2626
2627         return rc;
2628 }
2629
2630 static void collect_uncached_write_data(struct cifs_aio_ctx *ctx)
2631 {
2632         struct cifs_writedata *wdata, *tmp;
2633         struct cifs_tcon *tcon;
2634         struct cifs_sb_info *cifs_sb;
2635         struct dentry *dentry = ctx->cfile->dentry;
2636         unsigned int i;
2637         int rc;
2638
2639         tcon = tlink_tcon(ctx->cfile->tlink);
2640         cifs_sb = CIFS_SB(dentry->d_sb);
2641
2642         mutex_lock(&ctx->aio_mutex);
2643
2644         if (list_empty(&ctx->list)) {
2645                 mutex_unlock(&ctx->aio_mutex);
2646                 return;
2647         }
2648
2649         rc = ctx->rc;
2650         /*
2651          * Wait for and collect replies for any successful sends in order of
2652          * increasing offset. Once an error is hit, then return without waiting
2653          * for any more replies.
2654          */
2655 restart_loop:
2656         list_for_each_entry_safe(wdata, tmp, &ctx->list, list) {
2657                 if (!rc) {
2658                         if (!try_wait_for_completion(&wdata->done)) {
2659                                 mutex_unlock(&ctx->aio_mutex);
2660                                 return;
2661                         }
2662
2663                         if (wdata->result)
2664                                 rc = wdata->result;
2665                         else
2666                                 ctx->total_len += wdata->bytes;
2667
2668                         /* resend call if it's a retryable error */
2669                         if (rc == -EAGAIN) {
2670                                 struct list_head tmp_list;
2671                                 struct iov_iter tmp_from = ctx->iter;
2672
2673                                 INIT_LIST_HEAD(&tmp_list);
2674                                 list_del_init(&wdata->list);
2675
2676                                 iov_iter_advance(&tmp_from,
2677                                                  wdata->offset - ctx->pos);
2678
2679                                 rc = cifs_write_from_iter(wdata->offset,
2680                                                 wdata->bytes, &tmp_from,
2681                                                 ctx->cfile, cifs_sb, &tmp_list,
2682                                                 ctx);
2683
2684                                 list_splice(&tmp_list, &ctx->list);
2685
2686                                 kref_put(&wdata->refcount,
2687                                          cifs_uncached_writedata_release);
2688                                 goto restart_loop;
2689                         }
2690                 }
2691                 list_del_init(&wdata->list);
2692                 kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2693         }
2694
2695         for (i = 0; i < ctx->npages; i++)
2696                 put_page(ctx->bv[i].bv_page);
2697
2698         cifs_stats_bytes_written(tcon, ctx->total_len);
2699         set_bit(CIFS_INO_INVALID_MAPPING, &CIFS_I(dentry->d_inode)->flags);
2700
2701         ctx->rc = (rc == 0) ? ctx->total_len : rc;
2702
2703         mutex_unlock(&ctx->aio_mutex);
2704
2705         if (ctx->iocb && ctx->iocb->ki_complete)
2706                 ctx->iocb->ki_complete(ctx->iocb, ctx->rc, 0);
2707         else
2708                 complete(&ctx->done);
2709 }
2710
2711 ssize_t cifs_user_writev(struct kiocb *iocb, struct iov_iter *from)
2712 {
2713         struct file *file = iocb->ki_filp;
2714         ssize_t total_written = 0;
2715         struct cifsFileInfo *cfile;
2716         struct cifs_tcon *tcon;
2717         struct cifs_sb_info *cifs_sb;
2718         struct cifs_aio_ctx *ctx;
2719         struct iov_iter saved_from = *from;
2720         int rc;
2721
2722         /*
2723          * BB - optimize the way when signing is disabled. We can drop this
2724          * extra memory-to-memory copying and use iovec buffers for constructing
2725          * write request.
2726          */
2727
2728         rc = generic_write_checks(iocb, from);
2729         if (rc <= 0)
2730                 return rc;
2731
2732         cifs_sb = CIFS_FILE_SB(file);
2733         cfile = file->private_data;
2734         tcon = tlink_tcon(cfile->tlink);
2735
2736         if (!tcon->ses->server->ops->async_writev)
2737                 return -ENOSYS;
2738
2739         ctx = cifs_aio_ctx_alloc();
2740         if (!ctx)
2741                 return -ENOMEM;
2742
2743         ctx->cfile = cifsFileInfo_get(cfile);
2744
2745         if (!is_sync_kiocb(iocb))
2746                 ctx->iocb = iocb;
2747
2748         ctx->pos = iocb->ki_pos;
2749
2750         rc = setup_aio_ctx_iter(ctx, from, WRITE);
2751         if (rc) {
2752                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
2753                 return rc;
2754         }
2755
2756         /* grab a lock here due to read response handlers can access ctx */
2757         mutex_lock(&ctx->aio_mutex);
2758
2759         rc = cifs_write_from_iter(iocb->ki_pos, ctx->len, &saved_from,
2760                                   cfile, cifs_sb, &ctx->list, ctx);
2761
2762         /*
2763          * If at least one write was successfully sent, then discard any rc
2764          * value from the later writes. If the other write succeeds, then
2765          * we'll end up returning whatever was written. If it fails, then
2766          * we'll get a new rc value from that.
2767          */
2768         if (!list_empty(&ctx->list))
2769                 rc = 0;
2770
2771         mutex_unlock(&ctx->aio_mutex);
2772
2773         if (rc) {
2774                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
2775                 return rc;
2776         }
2777
2778         if (!is_sync_kiocb(iocb)) {
2779                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
2780                 return -EIOCBQUEUED;
2781         }
2782
2783         rc = wait_for_completion_killable(&ctx->done);
2784         if (rc) {
2785                 mutex_lock(&ctx->aio_mutex);
2786                 ctx->rc = rc = -EINTR;
2787                 total_written = ctx->total_len;
2788                 mutex_unlock(&ctx->aio_mutex);
2789         } else {
2790                 rc = ctx->rc;
2791                 total_written = ctx->total_len;
2792         }
2793
2794         kref_put(&ctx->refcount, cifs_aio_ctx_release);
2795
2796         if (unlikely(!total_written))
2797                 return rc;
2798
2799         iocb->ki_pos += total_written;
2800         return total_written;
2801 }
2802
2803 static ssize_t
2804 cifs_writev(struct kiocb *iocb, struct iov_iter *from)
2805 {
2806         struct file *file = iocb->ki_filp;
2807         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
2808         struct inode *inode = file->f_mapping->host;
2809         struct cifsInodeInfo *cinode = CIFS_I(inode);
2810         struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
2811         ssize_t rc;
2812
2813         /*
2814          * We need to hold the sem to be sure nobody modifies lock list
2815          * with a brlock that prevents writing.
2816          */
2817         down_read(&cinode->lock_sem);
2818         inode_lock(inode);
2819
2820         rc = generic_write_checks(iocb, from);
2821         if (rc <= 0)
2822                 goto out;
2823
2824         if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(from),
2825                                      server->vals->exclusive_lock_type, NULL,
2826                                      CIFS_WRITE_OP))
2827                 rc = __generic_file_write_iter(iocb, from);
2828         else
2829                 rc = -EACCES;
2830 out:
2831         inode_unlock(inode);
2832
2833         if (rc > 0)
2834                 rc = generic_write_sync(iocb, rc);
2835         up_read(&cinode->lock_sem);
2836         return rc;
2837 }
2838
2839 ssize_t
2840 cifs_strict_writev(struct kiocb *iocb, struct iov_iter *from)
2841 {
2842         struct inode *inode = file_inode(iocb->ki_filp);
2843         struct cifsInodeInfo *cinode = CIFS_I(inode);
2844         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2845         struct cifsFileInfo *cfile = (struct cifsFileInfo *)
2846                                                 iocb->ki_filp->private_data;
2847         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
2848         ssize_t written;
2849
2850         written = cifs_get_writer(cinode);
2851         if (written)
2852                 return written;
2853
2854         if (CIFS_CACHE_WRITE(cinode)) {
2855                 if (cap_unix(tcon->ses) &&
2856                 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability))
2857                   && ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0)) {
2858                         written = generic_file_write_iter(iocb, from);
2859                         goto out;
2860                 }
2861                 written = cifs_writev(iocb, from);
2862                 goto out;
2863         }
2864         /*
2865          * For non-oplocked files in strict cache mode we need to write the data
2866          * to the server exactly from the pos to pos+len-1 rather than flush all
2867          * affected pages because it may cause a error with mandatory locks on
2868          * these pages but not on the region from pos to ppos+len-1.
2869          */
2870         written = cifs_user_writev(iocb, from);
2871         if (written > 0 && CIFS_CACHE_READ(cinode)) {
2872                 /*
2873                  * Windows 7 server can delay breaking level2 oplock if a write
2874                  * request comes - break it on the client to prevent reading
2875                  * an old data.
2876                  */
2877                 cifs_zap_mapping(inode);
2878                 cifs_dbg(FYI, "Set no oplock for inode=%p after a write operation\n",
2879                          inode);
2880                 cinode->oplock = 0;
2881         }
2882 out:
2883         cifs_put_writer(cinode);
2884         return written;
2885 }
2886
2887 static struct cifs_readdata *
2888 cifs_readdata_alloc(unsigned int nr_pages, work_func_t complete)
2889 {
2890         struct cifs_readdata *rdata;
2891
2892         rdata = kzalloc(sizeof(*rdata) + (sizeof(struct page *) * nr_pages),
2893                         GFP_KERNEL);
2894         if (rdata != NULL) {
2895                 kref_init(&rdata->refcount);
2896                 INIT_LIST_HEAD(&rdata->list);
2897                 init_completion(&rdata->done);
2898                 INIT_WORK(&rdata->work, complete);
2899         }
2900
2901         return rdata;
2902 }
2903
2904 void
2905 cifs_readdata_release(struct kref *refcount)
2906 {
2907         struct cifs_readdata *rdata = container_of(refcount,
2908                                         struct cifs_readdata, refcount);
2909
2910         if (rdata->cfile)
2911                 cifsFileInfo_put(rdata->cfile);
2912
2913         kfree(rdata);
2914 }
2915
2916 static int
2917 cifs_read_allocate_pages(struct cifs_readdata *rdata, unsigned int nr_pages)
2918 {
2919         int rc = 0;
2920         struct page *page;
2921         unsigned int i;
2922
2923         for (i = 0; i < nr_pages; i++) {
2924                 page = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2925                 if (!page) {
2926                         rc = -ENOMEM;
2927                         break;
2928                 }
2929                 rdata->pages[i] = page;
2930         }
2931
2932         if (rc) {
2933                 for (i = 0; i < nr_pages; i++) {
2934                         put_page(rdata->pages[i]);
2935                         rdata->pages[i] = NULL;
2936                 }
2937         }
2938         return rc;
2939 }
2940
2941 static void
2942 cifs_uncached_readdata_release(struct kref *refcount)
2943 {
2944         struct cifs_readdata *rdata = container_of(refcount,
2945                                         struct cifs_readdata, refcount);
2946         unsigned int i;
2947
2948         kref_put(&rdata->ctx->refcount, cifs_aio_ctx_release);
2949         for (i = 0; i < rdata->nr_pages; i++) {
2950                 put_page(rdata->pages[i]);
2951                 rdata->pages[i] = NULL;
2952         }
2953         cifs_readdata_release(refcount);
2954 }
2955
2956 /**
2957  * cifs_readdata_to_iov - copy data from pages in response to an iovec
2958  * @rdata:      the readdata response with list of pages holding data
2959  * @iter:       destination for our data
2960  *
2961  * This function copies data from a list of pages in a readdata response into
2962  * an array of iovecs. It will first calculate where the data should go
2963  * based on the info in the readdata and then copy the data into that spot.
2964  */
2965 static int
2966 cifs_readdata_to_iov(struct cifs_readdata *rdata, struct iov_iter *iter)
2967 {
2968         size_t remaining = rdata->got_bytes;
2969         unsigned int i;
2970
2971         for (i = 0; i < rdata->nr_pages; i++) {
2972                 struct page *page = rdata->pages[i];
2973                 size_t copy = min_t(size_t, remaining, PAGE_SIZE);
2974                 size_t written;
2975
2976                 if (unlikely(iter->type & ITER_PIPE)) {
2977                         void *addr = kmap_atomic(page);
2978
2979                         written = copy_to_iter(addr, copy, iter);
2980                         kunmap_atomic(addr);
2981                 } else
2982                         written = copy_page_to_iter(page, 0, copy, iter);
2983                 remaining -= written;
2984                 if (written < copy && iov_iter_count(iter) > 0)
2985                         break;
2986         }
2987         return remaining ? -EFAULT : 0;
2988 }
2989
2990 static void collect_uncached_read_data(struct cifs_aio_ctx *ctx);
2991
2992 static void
2993 cifs_uncached_readv_complete(struct work_struct *work)
2994 {
2995         struct cifs_readdata *rdata = container_of(work,
2996                                                 struct cifs_readdata, work);
2997
2998         complete(&rdata->done);
2999         collect_uncached_read_data(rdata->ctx);
3000         /* the below call can possibly free the last ref to aio ctx */
3001         kref_put(&rdata->refcount, cifs_uncached_readdata_release);
3002 }
3003
3004 static int
3005 uncached_fill_pages(struct TCP_Server_Info *server,
3006                     struct cifs_readdata *rdata, struct iov_iter *iter,
3007                     unsigned int len)
3008 {
3009         int result = 0;
3010         unsigned int i;
3011         unsigned int nr_pages = rdata->nr_pages;
3012
3013         rdata->got_bytes = 0;
3014         rdata->tailsz = PAGE_SIZE;
3015         for (i = 0; i < nr_pages; i++) {
3016                 struct page *page = rdata->pages[i];
3017                 size_t n;
3018
3019                 if (len <= 0) {
3020                         /* no need to hold page hostage */
3021                         rdata->pages[i] = NULL;
3022                         rdata->nr_pages--;
3023                         put_page(page);
3024                         continue;
3025                 }
3026                 n = len;
3027                 if (len >= PAGE_SIZE) {
3028                         /* enough data to fill the page */
3029                         n = PAGE_SIZE;
3030                         len -= n;
3031                 } else {
3032                         zero_user(page, len, PAGE_SIZE - len);
3033                         rdata->tailsz = len;
3034                         len = 0;
3035                 }
3036                 if (iter)
3037                         result = copy_page_from_iter(page, 0, n, iter);
3038                 else
3039                         result = cifs_read_page_from_socket(server, page, n);
3040                 if (result < 0)
3041                         break;
3042
3043                 rdata->got_bytes += result;
3044         }
3045
3046         return rdata->got_bytes > 0 && result != -ECONNABORTED ?
3047                                                 rdata->got_bytes : result;
3048 }
3049
3050 static int
3051 cifs_uncached_read_into_pages(struct TCP_Server_Info *server,
3052                               struct cifs_readdata *rdata, unsigned int len)
3053 {
3054         return uncached_fill_pages(server, rdata, NULL, len);
3055 }
3056
3057 static int
3058 cifs_uncached_copy_into_pages(struct TCP_Server_Info *server,
3059                               struct cifs_readdata *rdata,
3060                               struct iov_iter *iter)
3061 {
3062         return uncached_fill_pages(server, rdata, iter, iter->count);
3063 }
3064
3065 static int
3066 cifs_send_async_read(loff_t offset, size_t len, struct cifsFileInfo *open_file,
3067                      struct cifs_sb_info *cifs_sb, struct list_head *rdata_list,
3068                      struct cifs_aio_ctx *ctx)
3069 {
3070         struct cifs_readdata *rdata;
3071         unsigned int npages, rsize, credits;
3072         size_t cur_len;
3073         int rc;
3074         pid_t pid;
3075         struct TCP_Server_Info *server;
3076
3077         server = tlink_tcon(open_file->tlink)->ses->server;
3078
3079         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3080                 pid = open_file->pid;
3081         else
3082                 pid = current->tgid;
3083
3084         do {
3085                 rc = server->ops->wait_mtu_credits(server, cifs_sb->rsize,
3086                                                    &rsize, &credits);
3087                 if (rc)
3088                         break;
3089
3090                 cur_len = min_t(const size_t, len, rsize);
3091                 npages = DIV_ROUND_UP(cur_len, PAGE_SIZE);
3092
3093                 /* allocate a readdata struct */
3094                 rdata = cifs_readdata_alloc(npages,
3095                                             cifs_uncached_readv_complete);
3096                 if (!rdata) {
3097                         add_credits_and_wake_if(server, credits, 0);
3098                         rc = -ENOMEM;
3099                         break;
3100                 }
3101
3102                 rc = cifs_read_allocate_pages(rdata, npages);
3103                 if (rc)
3104                         goto error;
3105
3106                 rdata->cfile = cifsFileInfo_get(open_file);
3107                 rdata->nr_pages = npages;
3108                 rdata->offset = offset;
3109                 rdata->bytes = cur_len;
3110                 rdata->pid = pid;
3111                 rdata->pagesz = PAGE_SIZE;
3112                 rdata->read_into_pages = cifs_uncached_read_into_pages;
3113                 rdata->copy_into_pages = cifs_uncached_copy_into_pages;
3114                 rdata->credits = credits;
3115                 rdata->ctx = ctx;
3116                 kref_get(&ctx->refcount);
3117
3118                 if (!rdata->cfile->invalidHandle ||
3119                     !(rc = cifs_reopen_file(rdata->cfile, true)))
3120                         rc = server->ops->async_readv(rdata);
3121 error:
3122                 if (rc) {
3123                         add_credits_and_wake_if(server, rdata->credits, 0);
3124                         kref_put(&rdata->refcount,
3125                                  cifs_uncached_readdata_release);
3126                         if (rc == -EAGAIN)
3127                                 continue;
3128                         break;
3129                 }
3130
3131                 list_add_tail(&rdata->list, rdata_list);
3132                 offset += cur_len;
3133                 len -= cur_len;
3134         } while (len > 0);
3135
3136         return rc;
3137 }
3138
3139 static void
3140 collect_uncached_read_data(struct cifs_aio_ctx *ctx)
3141 {
3142         struct cifs_readdata *rdata, *tmp;
3143         struct iov_iter *to = &ctx->iter;
3144         struct cifs_sb_info *cifs_sb;
3145         struct cifs_tcon *tcon;
3146         unsigned int i;
3147         int rc;
3148
3149         tcon = tlink_tcon(ctx->cfile->tlink);
3150         cifs_sb = CIFS_SB(ctx->cfile->dentry->d_sb);
3151
3152         mutex_lock(&ctx->aio_mutex);
3153
3154         if (list_empty(&ctx->list)) {
3155                 mutex_unlock(&ctx->aio_mutex);
3156                 return;
3157         }
3158
3159         rc = ctx->rc;
3160         /* the loop below should proceed in the order of increasing offsets */
3161 again:
3162         list_for_each_entry_safe(rdata, tmp, &ctx->list, list) {
3163                 if (!rc) {
3164                         if (!try_wait_for_completion(&rdata->done)) {
3165                                 mutex_unlock(&ctx->aio_mutex);
3166                                 return;
3167                         }
3168
3169                         if (rdata->result == -EAGAIN) {
3170                                 /* resend call if it's a retryable error */
3171                                 struct list_head tmp_list;
3172                                 unsigned int got_bytes = rdata->got_bytes;
3173
3174                                 list_del_init(&rdata->list);
3175                                 INIT_LIST_HEAD(&tmp_list);
3176
3177                                 /*
3178                                  * Got a part of data and then reconnect has
3179                                  * happened -- fill the buffer and continue
3180                                  * reading.
3181                                  */
3182                                 if (got_bytes && got_bytes < rdata->bytes) {
3183                                         rc = cifs_readdata_to_iov(rdata, to);
3184                                         if (rc) {
3185                                                 kref_put(&rdata->refcount,
3186                                                 cifs_uncached_readdata_release);
3187                                                 continue;
3188                                         }
3189                                 }
3190
3191                                 rc = cifs_send_async_read(
3192                                                 rdata->offset + got_bytes,
3193                                                 rdata->bytes - got_bytes,
3194                                                 rdata->cfile, cifs_sb,
3195                                                 &tmp_list, ctx);
3196
3197                                 list_splice(&tmp_list, &ctx->list);
3198
3199                                 kref_put(&rdata->refcount,
3200                                          cifs_uncached_readdata_release);
3201                                 goto again;
3202                         } else if (rdata->result)
3203                                 rc = rdata->result;
3204                         else
3205                                 rc = cifs_readdata_to_iov(rdata, to);
3206
3207                         /* if there was a short read -- discard anything left */
3208                         if (rdata->got_bytes && rdata->got_bytes < rdata->bytes)
3209                                 rc = -ENODATA;
3210                 }
3211                 list_del_init(&rdata->list);
3212                 kref_put(&rdata->refcount, cifs_uncached_readdata_release);
3213         }
3214
3215         for (i = 0; i < ctx->npages; i++) {
3216                 if (ctx->should_dirty)
3217                         set_page_dirty(ctx->bv[i].bv_page);
3218                 put_page(ctx->bv[i].bv_page);
3219         }
3220
3221         ctx->total_len = ctx->len - iov_iter_count(to);
3222
3223         cifs_stats_bytes_read(tcon, ctx->total_len);
3224
3225         /* mask nodata case */
3226         if (rc == -ENODATA)
3227                 rc = 0;
3228
3229         ctx->rc = (rc == 0) ? ctx->total_len : rc;
3230
3231         mutex_unlock(&ctx->aio_mutex);
3232
3233         if (ctx->iocb && ctx->iocb->ki_complete)
3234                 ctx->iocb->ki_complete(ctx->iocb, ctx->rc, 0);
3235         else
3236                 complete(&ctx->done);
3237 }
3238
3239 ssize_t cifs_user_readv(struct kiocb *iocb, struct iov_iter *to)
3240 {
3241         struct file *file = iocb->ki_filp;
3242         ssize_t rc;
3243         size_t len;
3244         ssize_t total_read = 0;
3245         loff_t offset = iocb->ki_pos;
3246         struct cifs_sb_info *cifs_sb;
3247         struct cifs_tcon *tcon;
3248         struct cifsFileInfo *cfile;
3249         struct cifs_aio_ctx *ctx;
3250
3251         len = iov_iter_count(to);
3252         if (!len)
3253                 return 0;
3254
3255         cifs_sb = CIFS_FILE_SB(file);
3256         cfile = file->private_data;
3257         tcon = tlink_tcon(cfile->tlink);
3258
3259         if (!tcon->ses->server->ops->async_readv)
3260                 return -ENOSYS;
3261
3262         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3263                 cifs_dbg(FYI, "attempting read on write only file instance\n");
3264
3265         ctx = cifs_aio_ctx_alloc();
3266         if (!ctx)
3267                 return -ENOMEM;
3268
3269         ctx->cfile = cifsFileInfo_get(cfile);
3270
3271         if (!is_sync_kiocb(iocb))
3272                 ctx->iocb = iocb;
3273
3274         if (to->type & ITER_IOVEC)
3275                 ctx->should_dirty = true;
3276
3277         rc = setup_aio_ctx_iter(ctx, to, READ);
3278         if (rc) {
3279                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
3280                 return rc;
3281         }
3282
3283         len = ctx->len;
3284
3285         /* grab a lock here due to read response handlers can access ctx */
3286         mutex_lock(&ctx->aio_mutex);
3287
3288         rc = cifs_send_async_read(offset, len, cfile, cifs_sb, &ctx->list, ctx);
3289
3290         /* if at least one read request send succeeded, then reset rc */
3291         if (!list_empty(&ctx->list))
3292                 rc = 0;
3293
3294         mutex_unlock(&ctx->aio_mutex);
3295
3296         if (rc) {
3297                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
3298                 return rc;
3299         }
3300
3301         if (!is_sync_kiocb(iocb)) {
3302                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
3303                 return -EIOCBQUEUED;
3304         }
3305
3306         rc = wait_for_completion_killable(&ctx->done);
3307         if (rc) {
3308                 mutex_lock(&ctx->aio_mutex);
3309                 ctx->rc = rc = -EINTR;
3310                 total_read = ctx->total_len;
3311                 mutex_unlock(&ctx->aio_mutex);
3312         } else {
3313                 rc = ctx->rc;
3314                 total_read = ctx->total_len;
3315         }
3316
3317         kref_put(&ctx->refcount, cifs_aio_ctx_release);
3318
3319         if (total_read) {
3320                 iocb->ki_pos += total_read;
3321                 return total_read;
3322         }
3323         return rc;
3324 }
3325
3326 ssize_t
3327 cifs_strict_readv(struct kiocb *iocb, struct iov_iter *to)
3328 {
3329         struct inode *inode = file_inode(iocb->ki_filp);
3330         struct cifsInodeInfo *cinode = CIFS_I(inode);
3331         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
3332         struct cifsFileInfo *cfile = (struct cifsFileInfo *)
3333                                                 iocb->ki_filp->private_data;
3334         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
3335         int rc = -EACCES;
3336
3337         /*
3338          * In strict cache mode we need to read from the server all the time
3339          * if we don't have level II oplock because the server can delay mtime
3340          * change - so we can't make a decision about inode invalidating.
3341          * And we can also fail with pagereading if there are mandatory locks
3342          * on pages affected by this read but not on the region from pos to
3343          * pos+len-1.
3344          */
3345         if (!CIFS_CACHE_READ(cinode))
3346                 return cifs_user_readv(iocb, to);
3347
3348         if (cap_unix(tcon->ses) &&
3349             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
3350             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
3351                 return generic_file_read_iter(iocb, to);
3352
3353         /*
3354          * We need to hold the sem to be sure nobody modifies lock list
3355          * with a brlock that prevents reading.
3356          */
3357         down_read(&cinode->lock_sem);
3358         if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(to),
3359                                      tcon->ses->server->vals->shared_lock_type,
3360                                      NULL, CIFS_READ_OP))
3361                 rc = generic_file_read_iter(iocb, to);
3362         up_read(&cinode->lock_sem);
3363         return rc;
3364 }
3365
3366 static ssize_t
3367 cifs_read(struct file *file, char *read_data, size_t read_size, loff_t *offset)
3368 {
3369         int rc = -EACCES;
3370         unsigned int bytes_read = 0;
3371         unsigned int total_read;
3372         unsigned int current_read_size;
3373         unsigned int rsize;
3374         struct cifs_sb_info *cifs_sb;
3375         struct cifs_tcon *tcon;
3376         struct TCP_Server_Info *server;
3377         unsigned int xid;
3378         char *cur_offset;
3379         struct cifsFileInfo *open_file;
3380         struct cifs_io_parms io_parms;
3381         int buf_type = CIFS_NO_BUFFER;
3382         __u32 pid;
3383
3384         xid = get_xid();
3385         cifs_sb = CIFS_FILE_SB(file);
3386
3387         /* FIXME: set up handlers for larger reads and/or convert to async */
3388         rsize = min_t(unsigned int, cifs_sb->rsize, CIFSMaxBufSize);
3389
3390         if (file->private_data == NULL) {
3391                 rc = -EBADF;
3392                 free_xid(xid);
3393                 return rc;
3394         }
3395         open_file = file->private_data;
3396         tcon = tlink_tcon(open_file->tlink);
3397         server = tcon->ses->server;
3398
3399         if (!server->ops->sync_read) {
3400                 free_xid(xid);
3401                 return -ENOSYS;
3402         }
3403
3404         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3405                 pid = open_file->pid;
3406         else
3407                 pid = current->tgid;
3408
3409         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3410                 cifs_dbg(FYI, "attempting read on write only file instance\n");
3411
3412         for (total_read = 0, cur_offset = read_data; read_size > total_read;
3413              total_read += bytes_read, cur_offset += bytes_read) {
3414                 do {
3415                         current_read_size = min_t(uint, read_size - total_read,
3416                                                   rsize);
3417                         /*
3418                          * For windows me and 9x we do not want to request more
3419                          * than it negotiated since it will refuse the read
3420                          * then.
3421                          */
3422                         if ((tcon->ses) && !(tcon->ses->capabilities &
3423                                 tcon->ses->server->vals->cap_large_files)) {
3424                                 current_read_size = min_t(uint,
3425                                         current_read_size, CIFSMaxBufSize);
3426                         }
3427                         if (open_file->invalidHandle) {
3428                                 rc = cifs_reopen_file(open_file, true);
3429                                 if (rc != 0)
3430                                         break;
3431                         }
3432                         io_parms.pid = pid;
3433                         io_parms.tcon = tcon;
3434                         io_parms.offset = *offset;
3435                         io_parms.length = current_read_size;
3436                         rc = server->ops->sync_read(xid, &open_file->fid, &io_parms,
3437                                                     &bytes_read, &cur_offset,
3438                                                     &buf_type);
3439                 } while (rc == -EAGAIN);
3440
3441                 if (rc || (bytes_read == 0)) {
3442                         if (total_read) {
3443                                 break;
3444                         } else {
3445                                 free_xid(xid);
3446                                 return rc;
3447                         }
3448                 } else {
3449                         cifs_stats_bytes_read(tcon, total_read);
3450                         *offset += bytes_read;
3451                 }
3452         }
3453         free_xid(xid);
3454         return total_read;
3455 }
3456
3457 /*
3458  * If the page is mmap'ed into a process' page tables, then we need to make
3459  * sure that it doesn't change while being written back.
3460  */
3461 static int
3462 cifs_page_mkwrite(struct vm_fault *vmf)
3463 {
3464         struct page *page = vmf->page;
3465
3466         lock_page(page);
3467         return VM_FAULT_LOCKED;
3468 }
3469
3470 static const struct vm_operations_struct cifs_file_vm_ops = {
3471         .fault = filemap_fault,
3472         .map_pages = filemap_map_pages,
3473         .page_mkwrite = cifs_page_mkwrite,
3474 };
3475
3476 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
3477 {
3478         int rc, xid;
3479         struct inode *inode = file_inode(file);
3480
3481         xid = get_xid();
3482
3483         if (!CIFS_CACHE_READ(CIFS_I(inode))) {
3484                 rc = cifs_zap_mapping(inode);
3485                 if (rc)
3486                         return rc;
3487         }
3488
3489         rc = generic_file_mmap(file, vma);
3490         if (rc == 0)
3491                 vma->vm_ops = &cifs_file_vm_ops;
3492         free_xid(xid);
3493         return rc;
3494 }
3495
3496 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
3497 {
3498         int rc, xid;
3499
3500         xid = get_xid();
3501         rc = cifs_revalidate_file(file);
3502         if (rc) {
3503                 cifs_dbg(FYI, "Validation prior to mmap failed, error=%d\n",
3504                          rc);
3505                 free_xid(xid);
3506                 return rc;
3507         }
3508         rc = generic_file_mmap(file, vma);
3509         if (rc == 0)
3510                 vma->vm_ops = &cifs_file_vm_ops;
3511         free_xid(xid);
3512         return rc;
3513 }
3514
3515 static void
3516 cifs_readv_complete(struct work_struct *work)
3517 {
3518         unsigned int i, got_bytes;
3519         struct cifs_readdata *rdata = container_of(work,
3520                                                 struct cifs_readdata, work);
3521
3522         got_bytes = rdata->got_bytes;
3523         for (i = 0; i < rdata->nr_pages; i++) {
3524                 struct page *page = rdata->pages[i];
3525
3526                 lru_cache_add_file(page);
3527
3528                 if (rdata->result == 0 ||
3529                     (rdata->result == -EAGAIN && got_bytes)) {
3530                         flush_dcache_page(page);
3531                         SetPageUptodate(page);
3532                 }
3533
3534                 unlock_page(page);
3535
3536                 if (rdata->result == 0 ||
3537                     (rdata->result == -EAGAIN && got_bytes))
3538                         cifs_readpage_to_fscache(rdata->mapping->host, page);
3539
3540                 got_bytes -= min_t(unsigned int, PAGE_SIZE, got_bytes);
3541
3542                 put_page(page);
3543                 rdata->pages[i] = NULL;
3544         }
3545         kref_put(&rdata->refcount, cifs_readdata_release);
3546 }
3547
3548 static int
3549 readpages_fill_pages(struct TCP_Server_Info *server,
3550                      struct cifs_readdata *rdata, struct iov_iter *iter,
3551                      unsigned int len)
3552 {
3553         int result = 0;
3554         unsigned int i;
3555         u64 eof;
3556         pgoff_t eof_index;
3557         unsigned int nr_pages = rdata->nr_pages;
3558
3559         /* determine the eof that the server (probably) has */
3560         eof = CIFS_I(rdata->mapping->host)->server_eof;
3561         eof_index = eof ? (eof - 1) >> PAGE_SHIFT : 0;
3562         cifs_dbg(FYI, "eof=%llu eof_index=%lu\n", eof, eof_index);
3563
3564         rdata->got_bytes = 0;
3565         rdata->tailsz = PAGE_SIZE;
3566         for (i = 0; i < nr_pages; i++) {
3567                 struct page *page = rdata->pages[i];
3568                 size_t n = PAGE_SIZE;
3569
3570                 if (len >= PAGE_SIZE) {
3571                         len -= PAGE_SIZE;
3572                 } else if (len > 0) {
3573                         /* enough for partial page, fill and zero the rest */
3574                         zero_user(page, len, PAGE_SIZE - len);
3575                         n = rdata->tailsz = len;
3576                         len = 0;
3577                 } else if (page->index > eof_index) {
3578                         /*
3579                          * The VFS will not try to do readahead past the
3580                          * i_size, but it's possible that we have outstanding
3581                          * writes with gaps in the middle and the i_size hasn't
3582                          * caught up yet. Populate those with zeroed out pages
3583                          * to prevent the VFS from repeatedly attempting to
3584                          * fill them until the writes are flushed.
3585                          */
3586                         zero_user(page, 0, PAGE_SIZE);
3587                         lru_cache_add_file(page);
3588                         flush_dcache_page(page);
3589                         SetPageUptodate(page);
3590                         unlock_page(page);
3591                         put_page(page);
3592                         rdata->pages[i] = NULL;
3593                         rdata->nr_pages--;
3594                         continue;
3595                 } else {
3596                         /* no need to hold page hostage */
3597                         lru_cache_add_file(page);
3598                         unlock_page(page);
3599                         put_page(page);
3600                         rdata->pages[i] = NULL;
3601                         rdata->nr_pages--;
3602                         continue;
3603                 }
3604
3605                 if (iter)
3606                         result = copy_page_from_iter(page, 0, n, iter);
3607                 else
3608                         result = cifs_read_page_from_socket(server, page, n);
3609                 if (result < 0)
3610                         break;
3611
3612                 rdata->got_bytes += result;
3613         }
3614
3615         return rdata->got_bytes > 0 && result != -ECONNABORTED ?
3616                                                 rdata->got_bytes : result;
3617 }
3618
3619 static int
3620 cifs_readpages_read_into_pages(struct TCP_Server_Info *server,
3621                                struct cifs_readdata *rdata, unsigned int len)
3622 {
3623         return readpages_fill_pages(server, rdata, NULL, len);
3624 }
3625
3626 static int
3627 cifs_readpages_copy_into_pages(struct TCP_Server_Info *server,
3628                                struct cifs_readdata *rdata,
3629                                struct iov_iter *iter)
3630 {
3631         return readpages_fill_pages(server, rdata, iter, iter->count);
3632 }
3633
3634 static int
3635 readpages_get_pages(struct address_space *mapping, struct list_head *page_list,
3636                     unsigned int rsize, struct list_head *tmplist,
3637                     unsigned int *nr_pages, loff_t *offset, unsigned int *bytes)
3638 {
3639         struct page *page, *tpage;
3640         unsigned int expected_index;
3641         int rc;
3642         gfp_t gfp = readahead_gfp_mask(mapping);
3643
3644         INIT_LIST_HEAD(tmplist);
3645
3646         page = list_entry(page_list->prev, struct page, lru);
3647
3648         /*
3649          * Lock the page and put it in the cache. Since no one else
3650          * should have access to this page, we're safe to simply set
3651          * PG_locked without checking it first.
3652          */
3653         __SetPageLocked(page);
3654         rc = add_to_page_cache_locked(page, mapping,
3655                                       page->index, gfp);
3656
3657         /* give up if we can't stick it in the cache */
3658         if (rc) {
3659                 __ClearPageLocked(page);
3660                 return rc;
3661         }
3662
3663         /* move first page to the tmplist */
3664         *offset = (loff_t)page->index << PAGE_SHIFT;
3665         *bytes = PAGE_SIZE;
3666         *nr_pages = 1;
3667         list_move_tail(&page->lru, tmplist);
3668
3669         /* now try and add more pages onto the request */
3670         expected_index = page->index + 1;
3671         list_for_each_entry_safe_reverse(page, tpage, page_list, lru) {
3672                 /* discontinuity ? */
3673                 if (page->index != expected_index)
3674                         break;
3675
3676                 /* would this page push the read over the rsize? */
3677                 if (*bytes + PAGE_SIZE > rsize)
3678                         break;
3679
3680                 __SetPageLocked(page);
3681                 if (add_to_page_cache_locked(page, mapping, page->index, gfp)) {
3682                         __ClearPageLocked(page);
3683                         break;
3684                 }
3685                 list_move_tail(&page->lru, tmplist);
3686                 (*bytes) += PAGE_SIZE;
3687                 expected_index++;
3688                 (*nr_pages)++;
3689         }
3690         return rc;
3691 }
3692
3693 static int cifs_readpages(struct file *file, struct address_space *mapping,
3694         struct list_head *page_list, unsigned num_pages)
3695 {
3696         int rc;
3697         struct list_head tmplist;
3698         struct cifsFileInfo *open_file = file->private_data;
3699         struct cifs_sb_info *cifs_sb = CIFS_FILE_SB(file);
3700         struct TCP_Server_Info *server;
3701         pid_t pid;
3702
3703         /*
3704          * Reads as many pages as possible from fscache. Returns -ENOBUFS
3705          * immediately if the cookie is negative
3706          *
3707          * After this point, every page in the list might have PG_fscache set,
3708          * so we will need to clean that up off of every page we don't use.
3709          */
3710         rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
3711                                          &num_pages);
3712         if (rc == 0)
3713                 return rc;
3714
3715         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3716                 pid = open_file->pid;
3717         else
3718                 pid = current->tgid;
3719
3720         rc = 0;
3721         server = tlink_tcon(open_file->tlink)->ses->server;
3722
3723         cifs_dbg(FYI, "%s: file=%p mapping=%p num_pages=%u\n",
3724                  __func__, file, mapping, num_pages);
3725
3726         /*
3727          * Start with the page at end of list and move it to private
3728          * list. Do the same with any following pages until we hit
3729          * the rsize limit, hit an index discontinuity, or run out of
3730          * pages. Issue the async read and then start the loop again
3731          * until the list is empty.
3732          *
3733          * Note that list order is important. The page_list is in
3734          * the order of declining indexes. When we put the pages in
3735          * the rdata->pages, then we want them in increasing order.
3736          */
3737         while (!list_empty(page_list)) {
3738                 unsigned int i, nr_pages, bytes, rsize;
3739                 loff_t offset;
3740                 struct page *page, *tpage;
3741                 struct cifs_readdata *rdata;
3742                 unsigned credits;
3743
3744                 rc = server->ops->wait_mtu_credits(server, cifs_sb->rsize,
3745                                                    &rsize, &credits);
3746                 if (rc)
3747                         break;
3748
3749                 /*
3750                  * Give up immediately if rsize is too small to read an entire
3751                  * page. The VFS will fall back to readpage. We should never
3752                  * reach this point however since we set ra_pages to 0 when the
3753                  * rsize is smaller than a cache page.
3754                  */
3755                 if (unlikely(rsize < PAGE_SIZE)) {
3756                         add_credits_and_wake_if(server, credits, 0);
3757                         return 0;
3758                 }
3759
3760                 rc = readpages_get_pages(mapping, page_list, rsize, &tmplist,
3761                                          &nr_pages, &offset, &bytes);
3762                 if (rc) {
3763                         add_credits_and_wake_if(server, credits, 0);
3764                         break;
3765                 }
3766
3767                 rdata = cifs_readdata_alloc(nr_pages, cifs_readv_complete);
3768                 if (!rdata) {
3769                         /* best to give up if we're out of mem */
3770                         list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3771                                 list_del(&page->lru);
3772                                 lru_cache_add_file(page);
3773                                 unlock_page(page);
3774                                 put_page(page);
3775                         }
3776                         rc = -ENOMEM;
3777                         add_credits_and_wake_if(server, credits, 0);
3778                         break;
3779                 }
3780
3781                 rdata->cfile = cifsFileInfo_get(open_file);
3782                 rdata->mapping = mapping;
3783                 rdata->offset = offset;
3784                 rdata->bytes = bytes;
3785                 rdata->pid = pid;
3786                 rdata->pagesz = PAGE_SIZE;
3787                 rdata->read_into_pages = cifs_readpages_read_into_pages;
3788                 rdata->copy_into_pages = cifs_readpages_copy_into_pages;
3789                 rdata->credits = credits;
3790
3791                 list_for_each_entry_safe(page, tpage, &tmplist, lru) {
3792                         list_del(&page->lru);
3793                         rdata->pages[rdata->nr_pages++] = page;
3794                 }
3795
3796                 if (!rdata->cfile->invalidHandle ||
3797                     !(rc = cifs_reopen_file(rdata->cfile, true)))
3798                         rc = server->ops->async_readv(rdata);
3799                 if (rc) {
3800                         add_credits_and_wake_if(server, rdata->credits, 0);
3801                         for (i = 0; i < rdata->nr_pages; i++) {
3802                                 page = rdata->pages[i];
3803                                 lru_cache_add_file(page);
3804                                 unlock_page(page);
3805                                 put_page(page);
3806                         }
3807                         /* Fallback to the readpage in error/reconnect cases */
3808                         kref_put(&rdata->refcount, cifs_readdata_release);
3809                         break;
3810                 }
3811
3812                 kref_put(&rdata->refcount, cifs_readdata_release);
3813         }
3814
3815         /* Any pages that have been shown to fscache but didn't get added to
3816          * the pagecache must be uncached before they get returned to the
3817          * allocator.
3818          */
3819         cifs_fscache_readpages_cancel(mapping->host, page_list);
3820         return rc;
3821 }
3822
3823 /*
3824  * cifs_readpage_worker must be called with the page pinned
3825  */
3826 static int cifs_readpage_worker(struct file *file, struct page *page,
3827         loff_t *poffset)
3828 {
3829         char *read_data;
3830         int rc;
3831
3832         /* Is the page cached? */
3833         rc = cifs_readpage_from_fscache(file_inode(file), page);
3834         if (rc == 0)
3835                 goto read_complete;
3836
3837         read_data = kmap(page);
3838         /* for reads over a certain size could initiate async read ahead */
3839
3840         rc = cifs_read(file, read_data, PAGE_SIZE, poffset);
3841
3842         if (rc < 0)
3843                 goto io_error;
3844         else
3845                 cifs_dbg(FYI, "Bytes read %d\n", rc);
3846
3847         file_inode(file)->i_atime =
3848                 current_time(file_inode(file));
3849
3850         if (PAGE_SIZE > rc)
3851                 memset(read_data + rc, 0, PAGE_SIZE - rc);
3852
3853         flush_dcache_page(page);
3854         SetPageUptodate(page);
3855
3856         /* send this page to the cache */
3857         cifs_readpage_to_fscache(file_inode(file), page);
3858
3859         rc = 0;
3860
3861 io_error:
3862         kunmap(page);
3863         unlock_page(page);
3864
3865 read_complete:
3866         return rc;
3867 }
3868
3869 static int cifs_readpage(struct file *file, struct page *page)
3870 {
3871         loff_t offset = (loff_t)page->index << PAGE_SHIFT;
3872         int rc = -EACCES;
3873         unsigned int xid;
3874
3875         xid = get_xid();
3876
3877         if (file->private_data == NULL) {
3878                 rc = -EBADF;
3879                 free_xid(xid);
3880                 return rc;
3881         }
3882
3883         cifs_dbg(FYI, "readpage %p at offset %d 0x%x\n",
3884                  page, (int)offset, (int)offset);
3885
3886         rc = cifs_readpage_worker(file, page, &offset);
3887
3888         free_xid(xid);
3889         return rc;
3890 }
3891
3892 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
3893 {
3894         struct cifsFileInfo *open_file;
3895         struct cifs_tcon *tcon =
3896                 cifs_sb_master_tcon(CIFS_SB(cifs_inode->vfs_inode.i_sb));
3897
3898         spin_lock(&tcon->open_file_lock);
3899         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
3900                 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
3901                         spin_unlock(&tcon->open_file_lock);
3902                         return 1;
3903                 }
3904         }
3905         spin_unlock(&tcon->open_file_lock);
3906         return 0;
3907 }
3908
3909 /* We do not want to update the file size from server for inodes
3910    open for write - to avoid races with writepage extending
3911    the file - in the future we could consider allowing
3912    refreshing the inode only on increases in the file size
3913    but this is tricky to do without racing with writebehind
3914    page caching in the current Linux kernel design */
3915 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
3916 {
3917         if (!cifsInode)
3918                 return true;
3919
3920         if (is_inode_writable(cifsInode)) {
3921                 /* This inode is open for write at least once */
3922                 struct cifs_sb_info *cifs_sb;
3923
3924                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
3925                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
3926                         /* since no page cache to corrupt on directio
3927                         we can change size safely */
3928                         return true;
3929                 }
3930
3931                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
3932                         return true;
3933
3934                 return false;
3935         } else
3936                 return true;
3937 }
3938
3939 static int cifs_write_begin(struct file *file, struct address_space *mapping,
3940                         loff_t pos, unsigned len, unsigned flags,
3941                         struct page **pagep, void **fsdata)
3942 {
3943         int oncethru = 0;
3944         pgoff_t index = pos >> PAGE_SHIFT;
3945         loff_t offset = pos & (PAGE_SIZE - 1);
3946         loff_t page_start = pos & PAGE_MASK;
3947         loff_t i_size;
3948         struct page *page;
3949         int rc = 0;
3950
3951         cifs_dbg(FYI, "write_begin from %lld len %d\n", (long long)pos, len);
3952
3953 start:
3954         page = grab_cache_page_write_begin(mapping, index, flags);
3955         if (!page) {
3956                 rc = -ENOMEM;
3957                 goto out;
3958         }
3959
3960         if (PageUptodate(page))
3961                 goto out;
3962
3963         /*
3964          * If we write a full page it will be up to date, no need to read from
3965          * the server. If the write is short, we'll end up doing a sync write
3966          * instead.
3967          */
3968         if (len == PAGE_SIZE)
3969                 goto out;
3970
3971         /*
3972          * optimize away the read when we have an oplock, and we're not
3973          * expecting to use any of the data we'd be reading in. That
3974          * is, when the page lies beyond the EOF, or straddles the EOF
3975          * and the write will cover all of the existing data.
3976          */
3977         if (CIFS_CACHE_READ(CIFS_I(mapping->host))) {
3978                 i_size = i_size_read(mapping->host);
3979                 if (page_start >= i_size ||
3980                     (offset == 0 && (pos + len) >= i_size)) {
3981                         zero_user_segments(page, 0, offset,
3982                                            offset + len,
3983                                            PAGE_SIZE);
3984                         /*
3985                          * PageChecked means that the parts of the page
3986                          * to which we're not writing are considered up
3987                          * to date. Once the data is copied to the
3988                          * page, it can be set uptodate.
3989                          */
3990                         SetPageChecked(page);
3991                         goto out;
3992                 }
3993         }
3994
3995         if ((file->f_flags & O_ACCMODE) != O_WRONLY && !oncethru) {
3996                 /*
3997                  * might as well read a page, it is fast enough. If we get
3998                  * an error, we don't need to return it. cifs_write_end will
3999                  * do a sync write instead since PG_uptodate isn't set.
4000                  */
4001                 cifs_readpage_worker(file, page, &page_start);
4002                 put_page(page);
4003                 oncethru = 1;
4004                 goto start;
4005         } else {
4006                 /* we could try using another file handle if there is one -
4007                    but how would we lock it to prevent close of that handle
4008                    racing with this read? In any case
4009                    this will be written out by write_end so is fine */
4010         }
4011 out:
4012         *pagep = page;
4013         return rc;
4014 }
4015
4016 static int cifs_release_page(struct page *page, gfp_t gfp)
4017 {
4018         if (PagePrivate(page))
4019                 return 0;
4020
4021         return cifs_fscache_release_page(page, gfp);
4022 }
4023
4024 static void cifs_invalidate_page(struct page *page, unsigned int offset,
4025                                  unsigned int length)
4026 {
4027         struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
4028
4029         if (offset == 0 && length == PAGE_SIZE)
4030                 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
4031 }
4032
4033 static int cifs_launder_page(struct page *page)
4034 {
4035         int rc = 0;
4036         loff_t range_start = page_offset(page);
4037         loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1);
4038         struct writeback_control wbc = {
4039                 .sync_mode = WB_SYNC_ALL,
4040                 .nr_to_write = 0,
4041                 .range_start = range_start,
4042                 .range_end = range_end,
4043         };
4044
4045         cifs_dbg(FYI, "Launder page: %p\n", page);
4046
4047         if (clear_page_dirty_for_io(page))
4048                 rc = cifs_writepage_locked(page, &wbc);
4049
4050         cifs_fscache_invalidate_page(page, page->mapping->host);
4051         return rc;
4052 }
4053
4054 void cifs_oplock_break(struct work_struct *work)
4055 {
4056         struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
4057                                                   oplock_break);
4058         struct inode *inode = d_inode(cfile->dentry);
4059         struct cifsInodeInfo *cinode = CIFS_I(inode);
4060         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
4061         struct TCP_Server_Info *server = tcon->ses->server;
4062         int rc = 0;
4063
4064         wait_on_bit(&cinode->flags, CIFS_INODE_PENDING_WRITERS,
4065                         TASK_UNINTERRUPTIBLE);
4066
4067         server->ops->downgrade_oplock(server, cinode,
4068                 test_bit(CIFS_INODE_DOWNGRADE_OPLOCK_TO_L2, &cinode->flags));
4069
4070         if (!CIFS_CACHE_WRITE(cinode) && CIFS_CACHE_READ(cinode) &&
4071                                                 cifs_has_mand_locks(cinode)) {
4072                 cifs_dbg(FYI, "Reset oplock to None for inode=%p due to mand locks\n",
4073                          inode);
4074                 cinode->oplock = 0;
4075         }
4076
4077         if (inode && S_ISREG(inode->i_mode)) {
4078                 if (CIFS_CACHE_READ(cinode))
4079                         break_lease(inode, O_RDONLY);
4080                 else
4081                         break_lease(inode, O_WRONLY);
4082                 rc = filemap_fdatawrite(inode->i_mapping);
4083                 if (!CIFS_CACHE_READ(cinode)) {
4084                         rc = filemap_fdatawait(inode->i_mapping);
4085                         mapping_set_error(inode->i_mapping, rc);
4086                         cifs_zap_mapping(inode);
4087                 }
4088                 cifs_dbg(FYI, "Oplock flush inode %p rc %d\n", inode, rc);
4089         }
4090
4091         rc = cifs_push_locks(cfile);
4092         if (rc)
4093                 cifs_dbg(VFS, "Push locks rc = %d\n", rc);
4094
4095         /*
4096          * releasing stale oplock after recent reconnect of smb session using
4097          * a now incorrect file handle is not a data integrity issue but do
4098          * not bother sending an oplock release if session to server still is
4099          * disconnected since oplock already released by the server
4100          */
4101         if (!cfile->oplock_break_cancelled) {
4102                 rc = tcon->ses->server->ops->oplock_response(tcon, &cfile->fid,
4103                                                              cinode);
4104                 cifs_dbg(FYI, "Oplock release rc = %d\n", rc);
4105         }
4106         cifs_done_oplock_break(cinode);
4107 }
4108
4109 /*
4110  * The presence of cifs_direct_io() in the address space ops vector
4111  * allowes open() O_DIRECT flags which would have failed otherwise.
4112  *
4113  * In the non-cached mode (mount with cache=none), we shunt off direct read and write requests
4114  * so this method should never be called.
4115  *
4116  * Direct IO is not yet supported in the cached mode.
4117  */
4118 static ssize_t
4119 cifs_direct_io(struct kiocb *iocb, struct iov_iter *iter)
4120 {
4121         /*
4122          * FIXME
4123          * Eventually need to support direct IO for non forcedirectio mounts
4124          */
4125         return -EINVAL;
4126 }
4127
4128
4129 const struct address_space_operations cifs_addr_ops = {
4130         .readpage = cifs_readpage,
4131         .readpages = cifs_readpages,
4132         .writepage = cifs_writepage,
4133         .writepages = cifs_writepages,
4134         .write_begin = cifs_write_begin,
4135         .write_end = cifs_write_end,
4136         .set_page_dirty = __set_page_dirty_nobuffers,
4137         .releasepage = cifs_release_page,
4138         .direct_IO = cifs_direct_io,
4139         .invalidatepage = cifs_invalidate_page,
4140         .launder_page = cifs_launder_page,
4141 };
4142
4143 /*
4144  * cifs_readpages requires the server to support a buffer large enough to
4145  * contain the header plus one complete page of data.  Otherwise, we need
4146  * to leave cifs_readpages out of the address space operations.
4147  */
4148 const struct address_space_operations cifs_addr_ops_smallbuf = {
4149         .readpage = cifs_readpage,
4150         .writepage = cifs_writepage,
4151         .writepages = cifs_writepages,
4152         .write_begin = cifs_write_begin,
4153         .write_end = cifs_write_end,
4154         .set_page_dirty = __set_page_dirty_nobuffers,
4155         .releasepage = cifs_release_page,
4156         .invalidatepage = cifs_invalidate_page,
4157         .launder_page = cifs_launder_page,
4158 };