fs/cifs/file.c

   1 /*
   2  *   fs/cifs/file.c
   3  *
   4  *   vfs operations that deal with files
   5  *
   6  *   Copyright (C) International Business Machines  Corp., 2002,2010
   7  *   Author(s): Steve French (sfrench@us.ibm.com)
   8  *              Jeremy Allison (jra@samba.org)
   9  *
  10  *   This library is free software; you can redistribute it and/or modify
  11  *   it under the terms of the GNU Lesser General Public License as published
  12  *   by the Free Software Foundation; either version 2.1 of the License, or
  13  *   (at your option) any later version.
  14  *
  15  *   This library is distributed in the hope that it will be useful,
  16  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
  18  *   the GNU Lesser General Public License for more details.
  19  *
  20  *   You should have received a copy of the GNU Lesser General Public License
  21  *   along with this library; if not, write to the Free Software
  22  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  23  */
  24 #include <linux/fs.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/stat.h>
  27 #include <linux/fcntl.h>
  28 #include <linux/pagemap.h>
  29 #include <linux/pagevec.h>
  30 #include <linux/writeback.h>
  31 #include <linux/task_io_accounting_ops.h>
  32 #include <linux/delay.h>
  33 #include <linux/mount.h>
  34 #include <linux/slab.h>
  35 #include <linux/swap.h>
  36 #include <asm/div64.h>
  37 #include "cifsfs.h"
  38 #include "cifspdu.h"
  39 #include "cifsglob.h"
  40 #include "cifsproto.h"
  41 #include "cifs_unicode.h"
  42 #include "cifs_debug.h"
  43 #include "cifs_fs_sb.h"
  44 #include "fscache.h"
  45 #include "smbdirect.h"
  46
  47 static inline int cifs_convert_flags(unsigned int flags)
  48 {
  49         if ((flags & O_ACCMODE) == O_RDONLY)
  50                 return GENERIC_READ;
  51         else if ((flags & O_ACCMODE) == O_WRONLY)
  52                 return GENERIC_WRITE;
  53         else if ((flags & O_ACCMODE) == O_RDWR) {
  54                 /* GENERIC_ALL is too much permission to request
  55                    can cause unnecessary access denied on create */
  56                 /* return GENERIC_ALL; */
  57                 return (GENERIC_READ | GENERIC_WRITE);
  58         }
  59
  60         return (READ_CONTROL | FILE_WRITE_ATTRIBUTES | FILE_READ_ATTRIBUTES |
  61                 FILE_WRITE_EA | FILE_APPEND_DATA | FILE_WRITE_DATA |
  62                 FILE_READ_DATA);
  63 }
  64
  65 static u32 cifs_posix_convert_flags(unsigned int flags)
  66 {
  67         u32 posix_flags = 0;
  68
  69         if ((flags & O_ACCMODE) == O_RDONLY)
  70                 posix_flags = SMB_O_RDONLY;
  71         else if ((flags & O_ACCMODE) == O_WRONLY)
  72                 posix_flags = SMB_O_WRONLY;
  73         else if ((flags & O_ACCMODE) == O_RDWR)
  74                 posix_flags = SMB_O_RDWR;
  75
  76         if (flags & O_CREAT) {
  77                 posix_flags |= SMB_O_CREAT;
  78                 if (flags & O_EXCL)
  79                         posix_flags |= SMB_O_EXCL;
  80         } else if (flags & O_EXCL)
  81                 cifs_dbg(FYI, "Application %s pid %d has incorrectly set O_EXCL flag but not O_CREAT on file open. Ignoring O_EXCL\n",
  82                          current->comm, current->tgid);
  83
  84         if (flags & O_TRUNC)
  85                 posix_flags |= SMB_O_TRUNC;
  86         /* be safe and imply O_SYNC for O_DSYNC */
  87         if (flags & O_DSYNC)
  88                 posix_flags |= SMB_O_SYNC;
  89         if (flags & O_DIRECTORY)
  90                 posix_flags |= SMB_O_DIRECTORY;
  91         if (flags & O_NOFOLLOW)
  92                 posix_flags |= SMB_O_NOFOLLOW;
  93         if (flags & O_DIRECT)
  94                 posix_flags |= SMB_O_DIRECT;
  95
  96         return posix_flags;
  97 }
  98
  99 static inline int cifs_get_disposition(unsigned int flags)
 100 {
 101         if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
 102                 return FILE_CREATE;
 103         else if ((flags & (O_CREAT | O_TRUNC)) == (O_CREAT | O_TRUNC))
 104                 return FILE_OVERWRITE_IF;
 105         else if ((flags & O_CREAT) == O_CREAT)
 106                 return FILE_OPEN_IF;
 107         else if ((flags & O_TRUNC) == O_TRUNC)
 108                 return FILE_OVERWRITE;
 109         else
 110                 return FILE_OPEN;
 111 }
 112
 113 int cifs_posix_open(char *full_path, struct inode **pinode,
 114                         struct super_block *sb, int mode, unsigned int f_flags,
 115                         __u32 *poplock, __u16 *pnetfid, unsigned int xid)
 116 {
 117         int rc;
 118         FILE_UNIX_BASIC_INFO *presp_data;
 119         __u32 posix_flags = 0;
 120         struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
 121         struct cifs_fattr fattr;
 122         struct tcon_link *tlink;
 123         struct cifs_tcon *tcon;
 124
 125         cifs_dbg(FYI, "posix open %s\n", full_path);
 126
 127         presp_data = kzalloc(sizeof(FILE_UNIX_BASIC_INFO), GFP_KERNEL);
 128         if (presp_data == NULL)
 129                 return -ENOMEM;
 130
 131         tlink = cifs_sb_tlink(cifs_sb);
 132         if (IS_ERR(tlink)) {
 133                 rc = PTR_ERR(tlink);
 134                 goto posix_open_ret;
 135         }
 136
 137         tcon = tlink_tcon(tlink);
 138         mode &= ~current_umask();
 139
 140         posix_flags = cifs_posix_convert_flags(f_flags);
 141         rc = CIFSPOSIXCreate(xid, tcon, posix_flags, mode, pnetfid, presp_data,
 142                              poplock, full_path, cifs_sb->local_nls,
 143                              cifs_remap(cifs_sb));
 144         cifs_put_tlink(tlink);
 145
 146         if (rc)
 147                 goto posix_open_ret;
 148
 149         if (presp_data->Type == cpu_to_le32(-1))
 150                 goto posix_open_ret; /* open ok, caller does qpathinfo */
 151
 152         if (!pinode)
 153                 goto posix_open_ret; /* caller does not need info */
 154
 155         cifs_unix_basic_to_fattr(&fattr, presp_data, cifs_sb);
 156
 157         /* get new inode and set it up */
 158         if (*pinode == NULL) {
 159                 cifs_fill_uniqueid(sb, &fattr);
 160                 *pinode = cifs_iget(sb, &fattr);
 161                 if (!*pinode) {
 162                         rc = -ENOMEM;
 163                         goto posix_open_ret;
 164                 }
 165         } else {
 166                 cifs_fattr_to_inode(*pinode, &fattr);
 167         }
 168
 169 posix_open_ret:
 170         kfree(presp_data);
 171         return rc;
 172 }
 173
 174 static int
 175 cifs_nt_open(char *full_path, struct inode *inode, struct cifs_sb_info *cifs_sb,
 176              struct cifs_tcon *tcon, unsigned int f_flags, __u32 *oplock,
 177              struct cifs_fid *fid, unsigned int xid)
 178 {
 179         int rc;
 180         int desired_access;
 181         int disposition;
 182         int create_options = CREATE_NOT_DIR;
 183         FILE_ALL_INFO *buf;
 184         struct TCP_Server_Info *server = tcon->ses->server;
 185         struct cifs_open_parms oparms;
 186
 187         if (!server->ops->open)
 188                 return -ENOSYS;
 189
 190         desired_access = cifs_convert_flags(f_flags);
 191
 192 /*********************************************************************
 193  *  open flag mapping table:
 194  *
 195  *      POSIX Flag            CIFS Disposition
 196  *      ----------            ----------------
 197  *      O_CREAT               FILE_OPEN_IF
 198  *      O_CREAT | O_EXCL      FILE_CREATE
 199  *      O_CREAT | O_TRUNC     FILE_OVERWRITE_IF
 200  *      O_TRUNC               FILE_OVERWRITE
 201  *      none of the above     FILE_OPEN
 202  *
 203  *      Note that there is not a direct match between disposition
 204  *      FILE_SUPERSEDE (ie create whether or not file exists although
 205  *      O_CREAT | O_TRUNC is similar but truncates the existing
 206  *      file rather than creating a new file as FILE_SUPERSEDE does
 207  *      (which uses the attributes / metadata passed in on open call)
 208  *?
 209  *?  O_SYNC is a reasonable match to CIFS writethrough flag
 210  *?  and the read write flags match reasonably.  O_LARGEFILE
 211  *?  is irrelevant because largefile support is always used
 212  *?  by this client. Flags O_APPEND, O_DIRECT, O_DIRECTORY,
 213  *       O_FASYNC, O_NOFOLLOW, O_NONBLOCK need further investigation
 214  *********************************************************************/
 215
 216         disposition = cifs_get_disposition(f_flags);
 217
 218         /* BB pass O_SYNC flag through on file attributes .. BB */
 219
 220         buf = kmalloc(sizeof(FILE_ALL_INFO), GFP_KERNEL);
 221         if (!buf)
 222                 return -ENOMEM;
 223
 224         if (backup_cred(cifs_sb))
 225                 create_options |= CREATE_OPEN_BACKUP_INTENT;
 226
 227         /* O_SYNC also has bit for O_DSYNC so following check picks up either */
 228         if (f_flags & O_SYNC)
 229                 create_options |= CREATE_WRITE_THROUGH;
 230
 231         if (f_flags & O_DIRECT)
 232                 create_options |= CREATE_NO_BUFFER;
 233
 234         oparms.tcon = tcon;
 235         oparms.cifs_sb = cifs_sb;
 236         oparms.desired_access = desired_access;
 237         oparms.create_options = create_options;
 238         oparms.disposition = disposition;
 239         oparms.path = full_path;
 240         oparms.fid = fid;
 241         oparms.reconnect = false;
 242
 243         rc = server->ops->open(xid, &oparms, oplock, buf);
 244
 245         if (rc)
 246                 goto out;
 247
 248         if (tcon->unix_ext)
 249                 rc = cifs_get_inode_info_unix(&inode, full_path, inode->i_sb,
 250                                               xid);
 251         else
 252                 rc = cifs_get_inode_info(&inode, full_path, buf, inode->i_sb,
 253                                          xid, fid);
 254
 255 out:
 256         kfree(buf);
 257         return rc;
 258 }
 259
 260 static bool
 261 cifs_has_mand_locks(struct cifsInodeInfo *cinode)
 262 {
 263         struct cifs_fid_locks *cur;
 264         bool has_locks = false;
 265
 266         down_read(&cinode->lock_sem);
 267         list_for_each_entry(cur, &cinode->llist, llist) {
 268                 if (!list_empty(&cur->locks)) {
 269                         has_locks = true;
 270                         break;
 271                 }
 272         }
 273         up_read(&cinode->lock_sem);
 274         return has_locks;
 275 }
 276
 277 struct cifsFileInfo *
 278 cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,
 279                   struct tcon_link *tlink, __u32 oplock)
 280 {
 281         struct dentry *dentry = file_dentry(file);
 282         struct inode *inode = d_inode(dentry);
 283         struct cifsInodeInfo *cinode = CIFS_I(inode);
 284         struct cifsFileInfo *cfile;
 285         struct cifs_fid_locks *fdlocks;
 286         struct cifs_tcon *tcon = tlink_tcon(tlink);
 287         struct TCP_Server_Info *server = tcon->ses->server;
 288
 289         cfile = kzalloc(sizeof(struct cifsFileInfo), GFP_KERNEL);
 290         if (cfile == NULL)
 291                 return cfile;
 292
 293         fdlocks = kzalloc(sizeof(struct cifs_fid_locks), GFP_KERNEL);
 294         if (!fdlocks) {
 295                 kfree(cfile);
 296                 return NULL;
 297         }
 298
 299         INIT_LIST_HEAD(&fdlocks->locks);
 300         fdlocks->cfile = cfile;
 301         cfile->llist = fdlocks;
 302         down_write(&cinode->lock_sem);
 303         list_add(&fdlocks->llist, &cinode->llist);
 304         up_write(&cinode->lock_sem);
 305
 306         cfile->count = 1;
 307         cfile->pid = current->tgid;
 308         cfile->uid = current_fsuid();
 309         cfile->dentry = dget(dentry);
 310         cfile->f_flags = file->f_flags;
 311         cfile->invalidHandle = false;
 312         cfile->tlink = cifs_get_tlink(tlink);
 313         INIT_WORK(&cfile->oplock_break, cifs_oplock_break);
 314         mutex_init(&cfile->fh_mutex);
 315         spin_lock_init(&cfile->file_info_lock);
 316
 317         cifs_sb_active(inode->i_sb);
 318
 319         /*
 320          * If the server returned a read oplock and we have mandatory brlocks,
 321          * set oplock level to None.
 322          */
 323         if (server->ops->is_read_op(oplock) && cifs_has_mand_locks(cinode)) {
 324                 cifs_dbg(FYI, "Reset oplock val from read to None due to mand locks\n");
 325                 oplock = 0;
 326         }
 327
 328         spin_lock(&tcon->open_file_lock);
 329         if (fid->pending_open->oplock != CIFS_OPLOCK_NO_CHANGE && oplock)
 330                 oplock = fid->pending_open->oplock;
 331         list_del(&fid->pending_open->olist);
 332
 333         fid->purge_cache = false;
 334         server->ops->set_fid(cfile, fid, oplock);
 335
 336         list_add(&cfile->tlist, &tcon->openFileList);
 337         atomic_inc(&tcon->num_local_opens);
 338
 339         /* if readable file instance put first in list*/
 340         if (file->f_mode & FMODE_READ)
 341                 list_add(&cfile->flist, &cinode->openFileList);
 342         else
 343                 list_add_tail(&cfile->flist, &cinode->openFileList);
 344         spin_unlock(&tcon->open_file_lock);
 345
 346         if (fid->purge_cache)
 347                 cifs_zap_mapping(inode);
 348
 349         file->private_data = cfile;
 350         return cfile;
 351 }
 352
 353 struct cifsFileInfo *
 354 cifsFileInfo_get(struct cifsFileInfo *cifs_file)
 355 {
 356         spin_lock(&cifs_file->file_info_lock);
 357         cifsFileInfo_get_locked(cifs_file);
 358         spin_unlock(&cifs_file->file_info_lock);
 359         return cifs_file;
 360 }
 361
 362 /*
 363  * Release a reference on the file private data. This may involve closing
 364  * the filehandle out on the server. Must be called without holding
 365  * tcon->open_file_lock and cifs_file->file_info_lock.
 366  */
 367 void cifsFileInfo_put(struct cifsFileInfo *cifs_file)
 368 {
 369         struct inode *inode = d_inode(cifs_file->dentry);
 370         struct cifs_tcon *tcon = tlink_tcon(cifs_file->tlink);
 371         struct TCP_Server_Info *server = tcon->ses->server;
 372         struct cifsInodeInfo *cifsi = CIFS_I(inode);
 373         struct super_block *sb = inode->i_sb;
 374         struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
 375         struct cifsLockInfo *li, *tmp;
 376         struct cifs_fid fid;
 377         struct cifs_pending_open open;
 378         bool oplock_break_cancelled;
 379
 380         spin_lock(&tcon->open_file_lock);
 381
 382         spin_lock(&cifs_file->file_info_lock);
 383         if (--cifs_file->count > 0) {
 384                 spin_unlock(&cifs_file->file_info_lock);
 385                 spin_unlock(&tcon->open_file_lock);
 386                 return;
 387         }
 388         spin_unlock(&cifs_file->file_info_lock);
 389
 390         if (server->ops->get_lease_key)
 391                 server->ops->get_lease_key(inode, &fid);
 392
 393         /* store open in pending opens to make sure we don't miss lease break */
 394         cifs_add_pending_open_locked(&fid, cifs_file->tlink, &open);
 395
 396         /* remove it from the lists */
 397         list_del(&cifs_file->flist);
 398         list_del(&cifs_file->tlist);
 399         atomic_dec(&tcon->num_local_opens);
 400
 401         if (list_empty(&cifsi->openFileList)) {
 402                 cifs_dbg(FYI, "closing last open instance for inode %p\n",
 403                          d_inode(cifs_file->dentry));
 404                 /*
 405                  * In strict cache mode we need invalidate mapping on the last
 406                  * close  because it may cause a error when we open this file
 407                  * again and get at least level II oplock.
 408                  */
 409                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO)
 410                         set_bit(CIFS_INO_INVALID_MAPPING, &cifsi->flags);
 411                 cifs_set_oplock_level(cifsi, 0);
 412         }
 413
 414         spin_unlock(&tcon->open_file_lock);
 415
 416         oplock_break_cancelled = cancel_work_sync(&cifs_file->oplock_break);
 417
 418         if (!tcon->need_reconnect && !cifs_file->invalidHandle) {
 419                 struct TCP_Server_Info *server = tcon->ses->server;
 420                 unsigned int xid;
 421
 422                 xid = get_xid();
 423                 if (server->ops->close)
 424                         server->ops->close(xid, tcon, &cifs_file->fid);
 425                 _free_xid(xid);
 426         }
 427
 428         if (oplock_break_cancelled)
 429                 cifs_done_oplock_break(cifsi);
 430
 431         cifs_del_pending_open(&open);
 432
 433         /*
 434          * Delete any outstanding lock records. We'll lose them when the file
 435          * is closed anyway.
 436          */
 437         down_write(&cifsi->lock_sem);
 438         list_for_each_entry_safe(li, tmp, &cifs_file->llist->locks, llist) {
 439                 list_del(&li->llist);
 440                 cifs_del_lock_waiters(li);
 441                 kfree(li);
 442         }
 443         list_del(&cifs_file->llist->llist);
 444         kfree(cifs_file->llist);
 445         up_write(&cifsi->lock_sem);
 446
 447         cifs_put_tlink(cifs_file->tlink);
 448         dput(cifs_file->dentry);
 449         cifs_sb_deactive(sb);
 450         kfree(cifs_file);
 451 }
 452
 453 int cifs_open(struct inode *inode, struct file *file)
 454
 455 {
 456         int rc = -EACCES;
 457         unsigned int xid;
 458         __u32 oplock;
 459         struct cifs_sb_info *cifs_sb;
 460         struct TCP_Server_Info *server;
 461         struct cifs_tcon *tcon;
 462         struct tcon_link *tlink;
 463         struct cifsFileInfo *cfile = NULL;
 464         char *full_path = NULL;
 465         bool posix_open_ok = false;
 466         struct cifs_fid fid;
 467         struct cifs_pending_open open;
 468
 469         xid = get_xid();
 470
 471         cifs_sb = CIFS_SB(inode->i_sb);
 472         tlink = cifs_sb_tlink(cifs_sb);
 473         if (IS_ERR(tlink)) {
 474                 free_xid(xid);
 475                 return PTR_ERR(tlink);
 476         }
 477         tcon = tlink_tcon(tlink);
 478         server = tcon->ses->server;
 479
 480         full_path = build_path_from_dentry(file_dentry(file));
 481         if (full_path == NULL) {
 482                 rc = -ENOMEM;
 483                 goto out;
 484         }
 485
 486         cifs_dbg(FYI, "inode = 0x%p file flags are 0x%x for %s\n",
 487                  inode, file->f_flags, full_path);
 488
 489         if (file->f_flags & O_DIRECT &&
 490             cifs_sb->mnt_cifs_flags & CIFS_MOUNT_STRICT_IO) {
 491                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NO_BRL)
 492                         file->f_op = &cifs_file_direct_nobrl_ops;
 493                 else
 494                         file->f_op = &cifs_file_direct_ops;
 495         }
 496
 497         if (server->oplocks)
 498                 oplock = REQ_OPLOCK;
 499         else
 500                 oplock = 0;
 501
 502         if (!tcon->broken_posix_open && tcon->unix_ext &&
 503             cap_unix(tcon->ses) && (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 504                                 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 505                 /* can not refresh inode info since size could be stale */
 506                 rc = cifs_posix_open(full_path, &inode, inode->i_sb,
 507                                 cifs_sb->mnt_file_mode /* ignored */,
 508                                 file->f_flags, &oplock, &fid.netfid, xid);
 509                 if (rc == 0) {
 510                         cifs_dbg(FYI, "posix open succeeded\n");
 511                         posix_open_ok = true;
 512                 } else if ((rc == -EINVAL) || (rc == -EOPNOTSUPP)) {
 513                         if (tcon->ses->serverNOS)
 514                                 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",
 515                                          tcon->ses->serverName,
 516                                          tcon->ses->serverNOS);
 517                         tcon->broken_posix_open = true;
 518                 } else if ((rc != -EIO) && (rc != -EREMOTE) &&
 519                          (rc != -EOPNOTSUPP)) /* path not found or net err */
 520                         goto out;
 521                 /*
 522                  * Else fallthrough to retry open the old way on network i/o
 523                  * or DFS errors.
 524                  */
 525         }
 526
 527         if (server->ops->get_lease_key)
 528                 server->ops->get_lease_key(inode, &fid);
 529
 530         cifs_add_pending_open(&fid, tlink, &open);
 531
 532         if (!posix_open_ok) {
 533                 if (server->ops->get_lease_key)
 534                         server->ops->get_lease_key(inode, &fid);
 535
 536                 rc = cifs_nt_open(full_path, inode, cifs_sb, tcon,
 537                                   file->f_flags, &oplock, &fid, xid);
 538                 if (rc) {
 539                         cifs_del_pending_open(&open);
 540                         goto out;
 541                 }
 542         }
 543
 544         cfile = cifs_new_fileinfo(&fid, file, tlink, oplock);
 545         if (cfile == NULL) {
 546                 if (server->ops->close)
 547                         server->ops->close(xid, tcon, &fid);
 548                 cifs_del_pending_open(&open);
 549                 rc = -ENOMEM;
 550                 goto out;
 551         }
 552
 553         cifs_fscache_set_inode_cookie(inode, file);
 554
 555         if ((oplock & CIFS_CREATE_ACTION) && !posix_open_ok && tcon->unix_ext) {
 556                 /*
 557                  * Time to set mode which we can not set earlier due to
 558                  * problems creating new read-only files.
 559                  */
 560                 struct cifs_unix_set_info_args args = {
 561                         .mode   = inode->i_mode,
 562                         .uid    = INVALID_UID, /* no change */
 563                         .gid    = INVALID_GID, /* no change */
 564                         .ctime  = NO_CHANGE_64,
 565                         .atime  = NO_CHANGE_64,
 566                         .mtime  = NO_CHANGE_64,
 567                         .device = 0,
 568                 };
 569                 CIFSSMBUnixSetFileInfo(xid, tcon, &args, fid.netfid,
 570                                        cfile->pid);
 571         }
 572
 573 out:
 574         kfree(full_path);
 575         free_xid(xid);
 576         cifs_put_tlink(tlink);
 577         return rc;
 578 }
 579
 580 static int cifs_push_posix_locks(struct cifsFileInfo *cfile);
 581
 582 /*
 583  * Try to reacquire byte range locks that were released when session
 584  * to server was lost.
 585  */
 586 static int
 587 cifs_relock_file(struct cifsFileInfo *cfile)
 588 {
 589         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
 590         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 591         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
 592         int rc = 0;
 593
 594         down_read_nested(&cinode->lock_sem, SINGLE_DEPTH_NESTING);
 595         if (cinode->can_cache_brlcks) {
 596                 /* can cache locks - no need to relock */
 597                 up_read(&cinode->lock_sem);
 598                 return rc;
 599         }
 600
 601         if (cap_unix(tcon->ses) &&
 602             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
 603             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
 604                 rc = cifs_push_posix_locks(cfile);
 605         else
 606                 rc = tcon->ses->server->ops->push_mand_locks(cfile);
 607
 608         up_read(&cinode->lock_sem);
 609         return rc;
 610 }
 611
 612 static int
 613 cifs_reopen_file(struct cifsFileInfo *cfile, bool can_flush)
 614 {
 615         int rc = -EACCES;
 616         unsigned int xid;
 617         __u32 oplock;
 618         struct cifs_sb_info *cifs_sb;
 619         struct cifs_tcon *tcon;
 620         struct TCP_Server_Info *server;
 621         struct cifsInodeInfo *cinode;
 622         struct inode *inode;
 623         char *full_path = NULL;
 624         int desired_access;
 625         int disposition = FILE_OPEN;
 626         int create_options = CREATE_NOT_DIR;
 627         struct cifs_open_parms oparms;
 628
 629         xid = get_xid();
 630         mutex_lock(&cfile->fh_mutex);
 631         if (!cfile->invalidHandle) {
 632                 mutex_unlock(&cfile->fh_mutex);
 633                 rc = 0;
 634                 free_xid(xid);
 635                 return rc;
 636         }
 637
 638         inode = d_inode(cfile->dentry);
 639         cifs_sb = CIFS_SB(inode->i_sb);
 640         tcon = tlink_tcon(cfile->tlink);
 641         server = tcon->ses->server;
 642
 643         /*
 644          * Can not grab rename sem here because various ops, including those
 645          * that already have the rename sem can end up causing writepage to get
 646          * called and if the server was down that means we end up here, and we
 647          * can never tell if the caller already has the rename_sem.
 648          */
 649         full_path = build_path_from_dentry(cfile->dentry);
 650         if (full_path == NULL) {
 651                 rc = -ENOMEM;
 652                 mutex_unlock(&cfile->fh_mutex);
 653                 free_xid(xid);
 654                 return rc;
 655         }
 656
 657         cifs_dbg(FYI, "inode = 0x%p file flags 0x%x for %s\n",
 658                  inode, cfile->f_flags, full_path);
 659
 660         if (tcon->ses->server->oplocks)
 661                 oplock = REQ_OPLOCK;
 662         else
 663                 oplock = 0;
 664
 665         if (tcon->unix_ext && cap_unix(tcon->ses) &&
 666             (CIFS_UNIX_POSIX_PATH_OPS_CAP &
 667                                 le64_to_cpu(tcon->fsUnixInfo.Capability))) {
 668                 /*
 669                  * O_CREAT, O_EXCL and O_TRUNC already had their effect on the
 670                  * original open. Must mask them off for a reopen.
 671                  */
 672                 unsigned int oflags = cfile->f_flags &
 673                                                 ~(O_CREAT | O_EXCL | O_TRUNC);
 674
 675                 rc = cifs_posix_open(full_path, NULL, inode->i_sb,
 676                                      cifs_sb->mnt_file_mode /* ignored */,
 677                                      oflags, &oplock, &cfile->fid.netfid, xid);
 678                 if (rc == 0) {
 679                         cifs_dbg(FYI, "posix reopen succeeded\n");
 680                         oparms.reconnect = true;
 681                         goto reopen_success;
 682                 }
 683                 /*
 684                  * fallthrough to retry open the old way on errors, especially
 685                  * in the reconnect path it is important to retry hard
 686                  */
 687         }
 688
 689         desired_access = cifs_convert_flags(cfile->f_flags);
 690
 691         if (backup_cred(cifs_sb))
 692                 create_options |= CREATE_OPEN_BACKUP_INTENT;
 693
 694         if (server->ops->get_lease_key)
 695                 server->ops->get_lease_key(inode, &cfile->fid);
 696
 697         oparms.tcon = tcon;
 698         oparms.cifs_sb = cifs_sb;
 699         oparms.desired_access = desired_access;
 700         oparms.create_options = create_options;
 701         oparms.disposition = disposition;
 702         oparms.path = full_path;
 703         oparms.fid = &cfile->fid;
 704         oparms.reconnect = true;
 705
 706         /*
 707          * Can not refresh inode by passing in file_info buf to be returned by
 708          * ops->open and then calling get_inode_info with returned buf since
 709          * file might have write behind data that needs to be flushed and server
 710          * version of file size can be stale. If we knew for sure that inode was
 711          * not dirty locally we could do this.
 712          */
 713         rc = server->ops->open(xid, &oparms, &oplock, NULL);
 714         if (rc == -ENOENT && oparms.reconnect == false) {
 715                 /* durable handle timeout is expired - open the file again */
 716                 rc = server->ops->open(xid, &oparms, &oplock, NULL);
 717                 /* indicate that we need to relock the file */
 718                 oparms.reconnect = true;
 719         }
 720
 721         if (rc) {
 722                 mutex_unlock(&cfile->fh_mutex);
 723                 cifs_dbg(FYI, "cifs_reopen returned 0x%x\n", rc);
 724                 cifs_dbg(FYI, "oplock: %d\n", oplock);
 725                 goto reopen_error_exit;
 726         }
 727
 728 reopen_success:
 729         cfile->invalidHandle = false;
 730         mutex_unlock(&cfile->fh_mutex);
 731         cinode = CIFS_I(inode);
 732
 733         if (can_flush) {
 734                 rc = filemap_write_and_wait(inode->i_mapping);
 735                 mapping_set_error(inode->i_mapping, rc);
 736
 737                 if (tcon->unix_ext)
 738                         rc = cifs_get_inode_info_unix(&inode, full_path,
 739                                                       inode->i_sb, xid);
 740                 else
 741                         rc = cifs_get_inode_info(&inode, full_path, NULL,
 742                                                  inode->i_sb, xid, NULL);
 743         }
 744         /*
 745          * Else we are writing out data to server already and could deadlock if
 746          * we tried to flush data, and since we do not know if we have data that
 747          * would invalidate the current end of file on the server we can not go
 748          * to the server to get the new inode info.
 749          */
 750
 751         /*
 752          * If the server returned a read oplock and we have mandatory brlocks,
 753          * set oplock level to None.
 754          */
 755         if (server->ops->is_read_op(oplock) && cifs_has_mand_locks(cinode)) {
 756                 cifs_dbg(FYI, "Reset oplock val from read to None due to mand locks\n");
 757                 oplock = 0;
 758         }
 759
 760         server->ops->set_fid(cfile, &cfile->fid, oplock);
 761         if (oparms.reconnect)
 762                 cifs_relock_file(cfile);
 763
 764 reopen_error_exit:
 765         kfree(full_path);
 766         free_xid(xid);
 767         return rc;
 768 }
 769
 770 int cifs_close(struct inode *inode, struct file *file)
 771 {
 772         if (file->private_data != NULL) {
 773                 cifsFileInfo_put(file->private_data);
 774                 file->private_data = NULL;
 775         }
 776
 777         /* return code from the ->release op is always ignored */
 778         return 0;
 779 }
 780
 781 void
 782 cifs_reopen_persistent_handles(struct cifs_tcon *tcon)
 783 {
 784         struct cifsFileInfo *open_file;
 785         struct list_head *tmp;
 786         struct list_head *tmp1;
 787         struct list_head tmp_list;
 788
 789         if (!tcon->use_persistent || !tcon->need_reopen_files)
 790                 return;
 791
 792         tcon->need_reopen_files = false;
 793
 794         cifs_dbg(FYI, "Reopen persistent handles");
 795         INIT_LIST_HEAD(&tmp_list);
 796
 797         /* list all files open on tree connection, reopen resilient handles  */
 798         spin_lock(&tcon->open_file_lock);
 799         list_for_each(tmp, &tcon->openFileList) {
 800                 open_file = list_entry(tmp, struct cifsFileInfo, tlist);
 801                 if (!open_file->invalidHandle)
 802                         continue;
 803                 cifsFileInfo_get(open_file);
 804                 list_add_tail(&open_file->rlist, &tmp_list);
 805         }
 806         spin_unlock(&tcon->open_file_lock);
 807
 808         list_for_each_safe(tmp, tmp1, &tmp_list) {
 809                 open_file = list_entry(tmp, struct cifsFileInfo, rlist);
 810                 if (cifs_reopen_file(open_file, false /* do not flush */))
 811                         tcon->need_reopen_files = true;
 812                 list_del_init(&open_file->rlist);
 813                 cifsFileInfo_put(open_file);
 814         }
 815 }
 816
 817 int cifs_closedir(struct inode *inode, struct file *file)
 818 {
 819         int rc = 0;
 820         unsigned int xid;
 821         struct cifsFileInfo *cfile = file->private_data;
 822         struct cifs_tcon *tcon;
 823         struct TCP_Server_Info *server;
 824         char *buf;
 825
 826         cifs_dbg(FYI, "Closedir inode = 0x%p\n", inode);
 827
 828         if (cfile == NULL)
 829                 return rc;
 830
 831         xid = get_xid();
 832         tcon = tlink_tcon(cfile->tlink);
 833         server = tcon->ses->server;
 834
 835         cifs_dbg(FYI, "Freeing private data in close dir\n");
 836         spin_lock(&cfile->file_info_lock);
 837         if (server->ops->dir_needs_close(cfile)) {
 838                 cfile->invalidHandle = true;
 839                 spin_unlock(&cfile->file_info_lock);
 840                 if (server->ops->close_dir)
 841                         rc = server->ops->close_dir(xid, tcon, &cfile->fid);
 842                 else
 843                         rc = -ENOSYS;
 844                 cifs_dbg(FYI, "Closing uncompleted readdir with rc %d\n", rc);
 845                 /* not much we can do if it fails anyway, ignore rc */
 846                 rc = 0;
 847         } else
 848                 spin_unlock(&cfile->file_info_lock);
 849
 850         buf = cfile->srch_inf.ntwrk_buf_start;
 851         if (buf) {
 852                 cifs_dbg(FYI, "closedir free smb buf in srch struct\n");
 853                 cfile->srch_inf.ntwrk_buf_start = NULL;
 854                 if (cfile->srch_inf.smallBuf)
 855                         cifs_small_buf_release(buf);
 856                 else
 857                         cifs_buf_release(buf);
 858         }
 859
 860         cifs_put_tlink(cfile->tlink);
 861         kfree(file->private_data);
 862         file->private_data = NULL;
 863         /* BB can we lock the filestruct while this is going on? */
 864         free_xid(xid);
 865         return rc;
 866 }
 867
 868 static struct cifsLockInfo *
 869 cifs_lock_init(__u64 offset, __u64 length, __u8 type, __u16 flags)
 870 {
 871         struct cifsLockInfo *lock =
 872                 kmalloc(sizeof(struct cifsLockInfo), GFP_KERNEL);
 873         if (!lock)
 874                 return lock;
 875         lock->offset = offset;
 876         lock->length = length;
 877         lock->type = type;
 878         lock->pid = current->tgid;
 879         lock->flags = flags;
 880         INIT_LIST_HEAD(&lock->blist);
 881         init_waitqueue_head(&lock->block_q);
 882         return lock;
 883 }
 884
 885 void
 886 cifs_del_lock_waiters(struct cifsLockInfo *lock)
 887 {
 888         struct cifsLockInfo *li, *tmp;
 889         list_for_each_entry_safe(li, tmp, &lock->blist, blist) {
 890                 list_del_init(&li->blist);
 891                 wake_up(&li->block_q);
 892         }
 893 }
 894
 895 #define CIFS_LOCK_OP    0
 896 #define CIFS_READ_OP    1
 897 #define CIFS_WRITE_OP   2
 898
 899 /* @rw_check : 0 - no op, 1 - read, 2 - write */
 900 static bool
 901 cifs_find_fid_lock_conflict(struct cifs_fid_locks *fdlocks, __u64 offset,
 902                             __u64 length, __u8 type, __u16 flags,
 903                             struct cifsFileInfo *cfile,
 904                             struct cifsLockInfo **conf_lock, int rw_check)
 905 {
 906         struct cifsLockInfo *li;
 907         struct cifsFileInfo *cur_cfile = fdlocks->cfile;
 908         struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
 909
 910         list_for_each_entry(li, &fdlocks->locks, llist) {
 911                 if (offset + length <= li->offset ||
 912                     offset >= li->offset + li->length)
 913                         continue;
 914                 if (rw_check != CIFS_LOCK_OP && current->tgid == li->pid &&
 915                     server->ops->compare_fids(cfile, cur_cfile)) {
 916                         /* shared lock prevents write op through the same fid */
 917                         if (!(li->type & server->vals->shared_lock_type) ||
 918                             rw_check != CIFS_WRITE_OP)
 919                                 continue;
 920                 }
 921                 if ((type & server->vals->shared_lock_type) &&
 922                     ((server->ops->compare_fids(cfile, cur_cfile) &&
 923                      current->tgid == li->pid) || type == li->type))
 924                         continue;
 925                 if (rw_check == CIFS_LOCK_OP &&
 926                     (flags & FL_OFDLCK) && (li->flags & FL_OFDLCK) &&
 927                     server->ops->compare_fids(cfile, cur_cfile))
 928                         continue;
 929                 if (conf_lock)
 930                         *conf_lock = li;
 931                 return true;
 932         }
 933         return false;
 934 }
 935
 936 bool
 937 cifs_find_lock_conflict(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
 938                         __u8 type, __u16 flags,
 939                         struct cifsLockInfo **conf_lock, int rw_check)
 940 {
 941         bool rc = false;
 942         struct cifs_fid_locks *cur;
 943         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 944
 945         list_for_each_entry(cur, &cinode->llist, llist) {
 946                 rc = cifs_find_fid_lock_conflict(cur, offset, length, type,
 947                                                  flags, cfile, conf_lock,
 948                                                  rw_check);
 949                 if (rc)
 950                         break;
 951         }
 952
 953         return rc;
 954 }
 955
 956 /*
 957  * Check if there is another lock that prevents us to set the lock (mandatory
 958  * style). If such a lock exists, update the flock structure with its
 959  * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
 960  * or leave it the same if we can't. Returns 0 if we don't need to request to
 961  * the server or 1 otherwise.
 962  */
 963 static int
 964 cifs_lock_test(struct cifsFileInfo *cfile, __u64 offset, __u64 length,
 965                __u8 type, struct file_lock *flock)
 966 {
 967         int rc = 0;
 968         struct cifsLockInfo *conf_lock;
 969         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 970         struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
 971         bool exist;
 972
 973         down_read(&cinode->lock_sem);
 974
 975         exist = cifs_find_lock_conflict(cfile, offset, length, type,
 976                                         flock->fl_flags, &conf_lock,
 977                                         CIFS_LOCK_OP);
 978         if (exist) {
 979                 flock->fl_start = conf_lock->offset;
 980                 flock->fl_end = conf_lock->offset + conf_lock->length - 1;
 981                 flock->fl_pid = conf_lock->pid;
 982                 if (conf_lock->type & server->vals->shared_lock_type)
 983                         flock->fl_type = F_RDLCK;
 984                 else
 985                         flock->fl_type = F_WRLCK;
 986         } else if (!cinode->can_cache_brlcks)
 987                 rc = 1;
 988         else
 989                 flock->fl_type = F_UNLCK;
 990
 991         up_read(&cinode->lock_sem);
 992         return rc;
 993 }
 994
 995 static void
 996 cifs_lock_add(struct cifsFileInfo *cfile, struct cifsLockInfo *lock)
 997 {
 998         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
 999         down_write(&cinode->lock_sem);
1000         list_add_tail(&lock->llist, &cfile->llist->locks);
1001         up_write(&cinode->lock_sem);
1002 }
1003
1004 /*
1005  * Set the byte-range lock (mandatory style). Returns:
1006  * 1) 0, if we set the lock and don't need to request to the server;
1007  * 2) 1, if no locks prevent us but we need to request to the server;
1008  * 3) -EACCES, if there is a lock that prevents us and wait is false.
1009  */
1010 static int
1011 cifs_lock_add_if(struct cifsFileInfo *cfile, struct cifsLockInfo *lock,
1012                  bool wait)
1013 {
1014         struct cifsLockInfo *conf_lock;
1015         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1016         bool exist;
1017         int rc = 0;
1018
1019 try_again:
1020         exist = false;
1021         down_write(&cinode->lock_sem);
1022
1023         exist = cifs_find_lock_conflict(cfile, lock->offset, lock->length,
1024                                         lock->type, lock->flags, &conf_lock,
1025                                         CIFS_LOCK_OP);
1026         if (!exist && cinode->can_cache_brlcks) {
1027                 list_add_tail(&lock->llist, &cfile->llist->locks);
1028                 up_write(&cinode->lock_sem);
1029                 return rc;
1030         }
1031
1032         if (!exist)
1033                 rc = 1;
1034         else if (!wait)
1035                 rc = -EACCES;
1036         else {
1037                 list_add_tail(&lock->blist, &conf_lock->blist);
1038                 up_write(&cinode->lock_sem);
1039                 rc = wait_event_interruptible(lock->block_q,
1040                                         (lock->blist.prev == &lock->blist) &&
1041                                         (lock->blist.next == &lock->blist));
1042                 if (!rc)
1043                         goto try_again;
1044                 down_write(&cinode->lock_sem);
1045                 list_del_init(&lock->blist);
1046         }
1047
1048         up_write(&cinode->lock_sem);
1049         return rc;
1050 }
1051
1052 /*
1053  * Check if there is another lock that prevents us to set the lock (posix
1054  * style). If such a lock exists, update the flock structure with its
1055  * properties. Otherwise, set the flock type to F_UNLCK if we can cache brlocks
1056  * or leave it the same if we can't. Returns 0 if we don't need to request to
1057  * the server or 1 otherwise.
1058  */
1059 static int
1060 cifs_posix_lock_test(struct file *file, struct file_lock *flock)
1061 {
1062         int rc = 0;
1063         struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
1064         unsigned char saved_type = flock->fl_type;
1065
1066         if ((flock->fl_flags & FL_POSIX) == 0)
1067                 return 1;
1068
1069         down_read(&cinode->lock_sem);
1070         posix_test_lock(file, flock);
1071
1072         if (flock->fl_type == F_UNLCK && !cinode->can_cache_brlcks) {
1073                 flock->fl_type = saved_type;
1074                 rc = 1;
1075         }
1076
1077         up_read(&cinode->lock_sem);
1078         return rc;
1079 }
1080
1081 /*
1082  * Set the byte-range lock (posix style). Returns:
1083  * 1) 0, if we set the lock and don't need to request to the server;
1084  * 2) 1, if we need to request to the server;
1085  * 3) <0, if the error occurs while setting the lock.
1086  */
1087 static int
1088 cifs_posix_lock_set(struct file *file, struct file_lock *flock)
1089 {
1090         struct cifsInodeInfo *cinode = CIFS_I(file_inode(file));
1091         int rc = 1;
1092
1093         if ((flock->fl_flags & FL_POSIX) == 0)
1094                 return rc;
1095
1096 try_again:
1097         down_write(&cinode->lock_sem);
1098         if (!cinode->can_cache_brlcks) {
1099                 up_write(&cinode->lock_sem);
1100                 return rc;
1101         }
1102
1103         rc = posix_lock_file(file, flock, NULL);
1104         up_write(&cinode->lock_sem);
1105         if (rc == FILE_LOCK_DEFERRED) {
1106                 rc = wait_event_interruptible(flock->fl_wait, !flock->fl_blocker);
1107                 if (!rc)
1108                         goto try_again;
1109                 locks_delete_block(flock);
1110         }
1111         return rc;
1112 }
1113
1114 int
1115 cifs_push_mandatory_locks(struct cifsFileInfo *cfile)
1116 {
1117         unsigned int xid;
1118         int rc = 0, stored_rc;
1119         struct cifsLockInfo *li, *tmp;
1120         struct cifs_tcon *tcon;
1121         unsigned int num, max_num, max_buf;
1122         LOCKING_ANDX_RANGE *buf, *cur;
1123         static const int types[] = {
1124                 LOCKING_ANDX_LARGE_FILES,
1125                 LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES
1126         };
1127         int i;
1128
1129         xid = get_xid();
1130         tcon = tlink_tcon(cfile->tlink);
1131
1132         /*
1133          * Accessing maxBuf is racy with cifs_reconnect - need to store value
1134          * and check it for zero before using.
1135          */
1136         max_buf = tcon->ses->server->maxBuf;
1137         if (!max_buf) {
1138                 free_xid(xid);
1139                 return -EINVAL;
1140         }
1141
1142         max_num = (max_buf - sizeof(struct smb_hdr)) /
1143                                                 sizeof(LOCKING_ANDX_RANGE);
1144         buf = kcalloc(max_num, sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1145         if (!buf) {
1146                 free_xid(xid);
1147                 return -ENOMEM;
1148         }
1149
1150         for (i = 0; i < 2; i++) {
1151                 cur = buf;
1152                 num = 0;
1153                 list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1154                         if (li->type != types[i])
1155                                 continue;
1156                         cur->Pid = cpu_to_le16(li->pid);
1157                         cur->LengthLow = cpu_to_le32((u32)li->length);
1158                         cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1159                         cur->OffsetLow = cpu_to_le32((u32)li->offset);
1160                         cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1161                         if (++num == max_num) {
1162                                 stored_rc = cifs_lockv(xid, tcon,
1163                                                        cfile->fid.netfid,
1164                                                        (__u8)li->type, 0, num,
1165                                                        buf);
1166                                 if (stored_rc)
1167                                         rc = stored_rc;
1168                                 cur = buf;
1169                                 num = 0;
1170                         } else
1171                                 cur++;
1172                 }
1173
1174                 if (num) {
1175                         stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1176                                                (__u8)types[i], 0, num, buf);
1177                         if (stored_rc)
1178                                 rc = stored_rc;
1179                 }
1180         }
1181
1182         kfree(buf);
1183         free_xid(xid);
1184         return rc;
1185 }
1186
1187 static __u32
1188 hash_lockowner(fl_owner_t owner)
1189 {
1190         return cifs_lock_secret ^ hash32_ptr((const void *)owner);
1191 }
1192
1193 struct lock_to_push {
1194         struct list_head llist;
1195         __u64 offset;
1196         __u64 length;
1197         __u32 pid;
1198         __u16 netfid;
1199         __u8 type;
1200 };
1201
1202 static int
1203 cifs_push_posix_locks(struct cifsFileInfo *cfile)
1204 {
1205         struct inode *inode = d_inode(cfile->dentry);
1206         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1207         struct file_lock *flock;
1208         struct file_lock_context *flctx = inode->i_flctx;
1209         unsigned int count = 0, i;
1210         int rc = 0, xid, type;
1211         struct list_head locks_to_send, *el;
1212         struct lock_to_push *lck, *tmp;
1213         __u64 length;
1214
1215         xid = get_xid();
1216
1217         if (!flctx)
1218                 goto out;
1219
1220         spin_lock(&flctx->flc_lock);
1221         list_for_each(el, &flctx->flc_posix) {
1222                 count++;
1223         }
1224         spin_unlock(&flctx->flc_lock);
1225
1226         INIT_LIST_HEAD(&locks_to_send);
1227
1228         /*
1229          * Allocating count locks is enough because no FL_POSIX locks can be
1230          * added to the list while we are holding cinode->lock_sem that
1231          * protects locking operations of this inode.
1232          */
1233         for (i = 0; i < count; i++) {
1234                 lck = kmalloc(sizeof(struct lock_to_push), GFP_KERNEL);
1235                 if (!lck) {
1236                         rc = -ENOMEM;
1237                         goto err_out;
1238                 }
1239                 list_add_tail(&lck->llist, &locks_to_send);
1240         }
1241
1242         el = locks_to_send.next;
1243         spin_lock(&flctx->flc_lock);
1244         list_for_each_entry(flock, &flctx->flc_posix, fl_list) {
1245                 if (el == &locks_to_send) {
1246                         /*
1247                          * The list ended. We don't have enough allocated
1248                          * structures - something is really wrong.
1249                          */
1250                         cifs_dbg(VFS, "Can't push all brlocks!\n");
1251                         break;
1252                 }
1253                 length = 1 + flock->fl_end - flock->fl_start;
1254                 if (flock->fl_type == F_RDLCK || flock->fl_type == F_SHLCK)
1255                         type = CIFS_RDLCK;
1256                 else
1257                         type = CIFS_WRLCK;
1258                 lck = list_entry(el, struct lock_to_push, llist);
1259                 lck->pid = hash_lockowner(flock->fl_owner);
1260                 lck->netfid = cfile->fid.netfid;
1261                 lck->length = length;
1262                 lck->type = type;
1263                 lck->offset = flock->fl_start;
1264         }
1265         spin_unlock(&flctx->flc_lock);
1266
1267         list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1268                 int stored_rc;
1269
1270                 stored_rc = CIFSSMBPosixLock(xid, tcon, lck->netfid, lck->pid,
1271                                              lck->offset, lck->length, NULL,
1272                                              lck->type, 0);
1273                 if (stored_rc)
1274                         rc = stored_rc;
1275                 list_del(&lck->llist);
1276                 kfree(lck);
1277         }
1278
1279 out:
1280         free_xid(xid);
1281         return rc;
1282 err_out:
1283         list_for_each_entry_safe(lck, tmp, &locks_to_send, llist) {
1284                 list_del(&lck->llist);
1285                 kfree(lck);
1286         }
1287         goto out;
1288 }
1289
1290 static int
1291 cifs_push_locks(struct cifsFileInfo *cfile)
1292 {
1293         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
1294         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1295         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1296         int rc = 0;
1297
1298         /* we are going to update can_cache_brlcks here - need a write access */
1299         down_write(&cinode->lock_sem);
1300         if (!cinode->can_cache_brlcks) {
1301                 up_write(&cinode->lock_sem);
1302                 return rc;
1303         }
1304
1305         if (cap_unix(tcon->ses) &&
1306             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1307             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1308                 rc = cifs_push_posix_locks(cfile);
1309         else
1310                 rc = tcon->ses->server->ops->push_mand_locks(cfile);
1311
1312         cinode->can_cache_brlcks = false;
1313         up_write(&cinode->lock_sem);
1314         return rc;
1315 }
1316
1317 static void
1318 cifs_read_flock(struct file_lock *flock, __u32 *type, int *lock, int *unlock,
1319                 bool *wait_flag, struct TCP_Server_Info *server)
1320 {
1321         if (flock->fl_flags & FL_POSIX)
1322                 cifs_dbg(FYI, "Posix\n");
1323         if (flock->fl_flags & FL_FLOCK)
1324                 cifs_dbg(FYI, "Flock\n");
1325         if (flock->fl_flags & FL_SLEEP) {
1326                 cifs_dbg(FYI, "Blocking lock\n");
1327                 *wait_flag = true;
1328         }
1329         if (flock->fl_flags & FL_ACCESS)
1330                 cifs_dbg(FYI, "Process suspended by mandatory locking - not implemented yet\n");
1331         if (flock->fl_flags & FL_LEASE)
1332                 cifs_dbg(FYI, "Lease on file - not implemented yet\n");
1333         if (flock->fl_flags &
1334             (~(FL_POSIX | FL_FLOCK | FL_SLEEP |
1335                FL_ACCESS | FL_LEASE | FL_CLOSE | FL_OFDLCK)))
1336                 cifs_dbg(FYI, "Unknown lock flags 0x%x\n", flock->fl_flags);
1337
1338         *type = server->vals->large_lock_type;
1339         if (flock->fl_type == F_WRLCK) {
1340                 cifs_dbg(FYI, "F_WRLCK\n");
1341                 *type |= server->vals->exclusive_lock_type;
1342                 *lock = 1;
1343         } else if (flock->fl_type == F_UNLCK) {
1344                 cifs_dbg(FYI, "F_UNLCK\n");
1345                 *type |= server->vals->unlock_lock_type;
1346                 *unlock = 1;
1347                 /* Check if unlock includes more than one lock range */
1348         } else if (flock->fl_type == F_RDLCK) {
1349                 cifs_dbg(FYI, "F_RDLCK\n");
1350                 *type |= server->vals->shared_lock_type;
1351                 *lock = 1;
1352         } else if (flock->fl_type == F_EXLCK) {
1353                 cifs_dbg(FYI, "F_EXLCK\n");
1354                 *type |= server->vals->exclusive_lock_type;
1355                 *lock = 1;
1356         } else if (flock->fl_type == F_SHLCK) {
1357                 cifs_dbg(FYI, "F_SHLCK\n");
1358                 *type |= server->vals->shared_lock_type;
1359                 *lock = 1;
1360         } else
1361                 cifs_dbg(FYI, "Unknown type of lock\n");
1362 }
1363
1364 static int
1365 cifs_getlk(struct file *file, struct file_lock *flock, __u32 type,
1366            bool wait_flag, bool posix_lck, unsigned int xid)
1367 {
1368         int rc = 0;
1369         __u64 length = 1 + flock->fl_end - flock->fl_start;
1370         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1371         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1372         struct TCP_Server_Info *server = tcon->ses->server;
1373         __u16 netfid = cfile->fid.netfid;
1374
1375         if (posix_lck) {
1376                 int posix_lock_type;
1377
1378                 rc = cifs_posix_lock_test(file, flock);
1379                 if (!rc)
1380                         return rc;
1381
1382                 if (type & server->vals->shared_lock_type)
1383                         posix_lock_type = CIFS_RDLCK;
1384                 else
1385                         posix_lock_type = CIFS_WRLCK;
1386                 rc = CIFSSMBPosixLock(xid, tcon, netfid,
1387                                       hash_lockowner(flock->fl_owner),
1388                                       flock->fl_start, length, flock,
1389                                       posix_lock_type, wait_flag);
1390                 return rc;
1391         }
1392
1393         rc = cifs_lock_test(cfile, flock->fl_start, length, type, flock);
1394         if (!rc)
1395                 return rc;
1396
1397         /* BB we could chain these into one lock request BB */
1398         rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length, type,
1399                                     1, 0, false);
1400         if (rc == 0) {
1401                 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1402                                             type, 0, 1, false);
1403                 flock->fl_type = F_UNLCK;
1404                 if (rc != 0)
1405                         cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1406                                  rc);
1407                 return 0;
1408         }
1409
1410         if (type & server->vals->shared_lock_type) {
1411                 flock->fl_type = F_WRLCK;
1412                 return 0;
1413         }
1414
1415         type &= ~server->vals->exclusive_lock_type;
1416
1417         rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1418                                     type | server->vals->shared_lock_type,
1419                                     1, 0, false);
1420         if (rc == 0) {
1421                 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1422                         type | server->vals->shared_lock_type, 0, 1, false);
1423                 flock->fl_type = F_RDLCK;
1424                 if (rc != 0)
1425                         cifs_dbg(VFS, "Error unlocking previously locked range %d during test of lock\n",
1426                                  rc);
1427         } else
1428                 flock->fl_type = F_WRLCK;
1429
1430         return 0;
1431 }
1432
1433 void
1434 cifs_move_llist(struct list_head *source, struct list_head *dest)
1435 {
1436         struct list_head *li, *tmp;
1437         list_for_each_safe(li, tmp, source)
1438                 list_move(li, dest);
1439 }
1440
1441 void
1442 cifs_free_llist(struct list_head *llist)
1443 {
1444         struct cifsLockInfo *li, *tmp;
1445         list_for_each_entry_safe(li, tmp, llist, llist) {
1446                 cifs_del_lock_waiters(li);
1447                 list_del(&li->llist);
1448                 kfree(li);
1449         }
1450 }
1451
1452 int
1453 cifs_unlock_range(struct cifsFileInfo *cfile, struct file_lock *flock,
1454                   unsigned int xid)
1455 {
1456         int rc = 0, stored_rc;
1457         static const int types[] = {
1458                 LOCKING_ANDX_LARGE_FILES,
1459                 LOCKING_ANDX_SHARED_LOCK | LOCKING_ANDX_LARGE_FILES
1460         };
1461         unsigned int i;
1462         unsigned int max_num, num, max_buf;
1463         LOCKING_ANDX_RANGE *buf, *cur;
1464         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1465         struct cifsInodeInfo *cinode = CIFS_I(d_inode(cfile->dentry));
1466         struct cifsLockInfo *li, *tmp;
1467         __u64 length = 1 + flock->fl_end - flock->fl_start;
1468         struct list_head tmp_llist;
1469
1470         INIT_LIST_HEAD(&tmp_llist);
1471
1472         /*
1473          * Accessing maxBuf is racy with cifs_reconnect - need to store value
1474          * and check it for zero before using.
1475          */
1476         max_buf = tcon->ses->server->maxBuf;
1477         if (!max_buf)
1478                 return -EINVAL;
1479
1480         max_num = (max_buf - sizeof(struct smb_hdr)) /
1481                                                 sizeof(LOCKING_ANDX_RANGE);
1482         buf = kcalloc(max_num, sizeof(LOCKING_ANDX_RANGE), GFP_KERNEL);
1483         if (!buf)
1484                 return -ENOMEM;
1485
1486         down_write(&cinode->lock_sem);
1487         for (i = 0; i < 2; i++) {
1488                 cur = buf;
1489                 num = 0;
1490                 list_for_each_entry_safe(li, tmp, &cfile->llist->locks, llist) {
1491                         if (flock->fl_start > li->offset ||
1492                             (flock->fl_start + length) <
1493                             (li->offset + li->length))
1494                                 continue;
1495                         if (current->tgid != li->pid)
1496                                 continue;
1497                         if (types[i] != li->type)
1498                                 continue;
1499                         if (cinode->can_cache_brlcks) {
1500                                 /*
1501                                  * We can cache brlock requests - simply remove
1502                                  * a lock from the file's list.
1503                                  */
1504                                 list_del(&li->llist);
1505                                 cifs_del_lock_waiters(li);
1506                                 kfree(li);
1507                                 continue;
1508                         }
1509                         cur->Pid = cpu_to_le16(li->pid);
1510                         cur->LengthLow = cpu_to_le32((u32)li->length);
1511                         cur->LengthHigh = cpu_to_le32((u32)(li->length>>32));
1512                         cur->OffsetLow = cpu_to_le32((u32)li->offset);
1513                         cur->OffsetHigh = cpu_to_le32((u32)(li->offset>>32));
1514                         /*
1515                          * We need to save a lock here to let us add it again to
1516                          * the file's list if the unlock range request fails on
1517                          * the server.
1518                          */
1519                         list_move(&li->llist, &tmp_llist);
1520                         if (++num == max_num) {
1521                                 stored_rc = cifs_lockv(xid, tcon,
1522                                                        cfile->fid.netfid,
1523                                                        li->type, num, 0, buf);
1524                                 if (stored_rc) {
1525                                         /*
1526                                          * We failed on the unlock range
1527                                          * request - add all locks from the tmp
1528                                          * list to the head of the file's list.
1529                                          */
1530                                         cifs_move_llist(&tmp_llist,
1531                                                         &cfile->llist->locks);
1532                                         rc = stored_rc;
1533                                 } else
1534                                         /*
1535                                          * The unlock range request succeed -
1536                                          * free the tmp list.
1537                                          */
1538                                         cifs_free_llist(&tmp_llist);
1539                                 cur = buf;
1540                                 num = 0;
1541                         } else
1542                                 cur++;
1543                 }
1544                 if (num) {
1545                         stored_rc = cifs_lockv(xid, tcon, cfile->fid.netfid,
1546                                                types[i], num, 0, buf);
1547                         if (stored_rc) {
1548                                 cifs_move_llist(&tmp_llist,
1549                                                 &cfile->llist->locks);
1550                                 rc = stored_rc;
1551                         } else
1552                                 cifs_free_llist(&tmp_llist);
1553                 }
1554         }
1555
1556         up_write(&cinode->lock_sem);
1557         kfree(buf);
1558         return rc;
1559 }
1560
1561 static int
1562 cifs_setlk(struct file *file, struct file_lock *flock, __u32 type,
1563            bool wait_flag, bool posix_lck, int lock, int unlock,
1564            unsigned int xid)
1565 {
1566         int rc = 0;
1567         __u64 length = 1 + flock->fl_end - flock->fl_start;
1568         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
1569         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
1570         struct TCP_Server_Info *server = tcon->ses->server;
1571         struct inode *inode = d_inode(cfile->dentry);
1572
1573         if (posix_lck) {
1574                 int posix_lock_type;
1575
1576                 rc = cifs_posix_lock_set(file, flock);
1577                 if (!rc || rc < 0)
1578                         return rc;
1579
1580                 if (type & server->vals->shared_lock_type)
1581                         posix_lock_type = CIFS_RDLCK;
1582                 else
1583                         posix_lock_type = CIFS_WRLCK;
1584
1585                 if (unlock == 1)
1586                         posix_lock_type = CIFS_UNLCK;
1587
1588                 rc = CIFSSMBPosixLock(xid, tcon, cfile->fid.netfid,
1589                                       hash_lockowner(flock->fl_owner),
1590                                       flock->fl_start, length,
1591                                       NULL, posix_lock_type, wait_flag);
1592                 goto out;
1593         }
1594
1595         if (lock) {
1596                 struct cifsLockInfo *lock;
1597
1598                 lock = cifs_lock_init(flock->fl_start, length, type,
1599                                       flock->fl_flags);
1600                 if (!lock)
1601                         return -ENOMEM;
1602
1603                 rc = cifs_lock_add_if(cfile, lock, wait_flag);
1604                 if (rc < 0) {
1605                         kfree(lock);
1606                         return rc;
1607                 }
1608                 if (!rc)
1609                         goto out;
1610
1611                 /*
1612                  * Windows 7 server can delay breaking lease from read to None
1613                  * if we set a byte-range lock on a file - break it explicitly
1614                  * before sending the lock to the server to be sure the next
1615                  * read won't conflict with non-overlapted locks due to
1616                  * pagereading.
1617                  */
1618                 if (!CIFS_CACHE_WRITE(CIFS_I(inode)) &&
1619                                         CIFS_CACHE_READ(CIFS_I(inode))) {
1620                         cifs_zap_mapping(inode);
1621                         cifs_dbg(FYI, "Set no oplock for inode=%p due to mand locks\n",
1622                                  inode);
1623                         CIFS_I(inode)->oplock = 0;
1624                 }
1625
1626                 rc = server->ops->mand_lock(xid, cfile, flock->fl_start, length,
1627                                             type, 1, 0, wait_flag);
1628                 if (rc) {
1629                         kfree(lock);
1630                         return rc;
1631                 }
1632
1633                 cifs_lock_add(cfile, lock);
1634         } else if (unlock)
1635                 rc = server->ops->mand_unlock_range(cfile, flock, xid);
1636
1637 out:
1638         if (flock->fl_flags & FL_POSIX && !rc)
1639                 rc = locks_lock_file_wait(file, flock);
1640         return rc;
1641 }
1642
1643 int cifs_lock(struct file *file, int cmd, struct file_lock *flock)
1644 {
1645         int rc, xid;
1646         int lock = 0, unlock = 0;
1647         bool wait_flag = false;
1648         bool posix_lck = false;
1649         struct cifs_sb_info *cifs_sb;
1650         struct cifs_tcon *tcon;
1651         struct cifsInodeInfo *cinode;
1652         struct cifsFileInfo *cfile;
1653         __u16 netfid;
1654         __u32 type;
1655
1656         rc = -EACCES;
1657         xid = get_xid();
1658
1659         cifs_dbg(FYI, "Lock parm: 0x%x flockflags: 0x%x flocktype: 0x%x start: %lld end: %lld\n",
1660                  cmd, flock->fl_flags, flock->fl_type,
1661                  flock->fl_start, flock->fl_end);
1662
1663         cfile = (struct cifsFileInfo *)file->private_data;
1664         tcon = tlink_tcon(cfile->tlink);
1665
1666         cifs_read_flock(flock, &type, &lock, &unlock, &wait_flag,
1667                         tcon->ses->server);
1668         cifs_sb = CIFS_FILE_SB(file);
1669         netfid = cfile->fid.netfid;
1670         cinode = CIFS_I(file_inode(file));
1671
1672         if (cap_unix(tcon->ses) &&
1673             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
1674             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
1675                 posix_lck = true;
1676         /*
1677          * BB add code here to normalize offset and length to account for
1678          * negative length which we can not accept over the wire.
1679          */
1680         if (IS_GETLK(cmd)) {
1681                 rc = cifs_getlk(file, flock, type, wait_flag, posix_lck, xid);
1682                 free_xid(xid);
1683                 return rc;
1684         }
1685
1686         if (!lock && !unlock) {
1687                 /*
1688                  * if no lock or unlock then nothing to do since we do not
1689                  * know what it is
1690                  */
1691                 free_xid(xid);
1692                 return -EOPNOTSUPP;
1693         }
1694
1695         rc = cifs_setlk(file, flock, type, wait_flag, posix_lck, lock, unlock,
1696                         xid);
1697         free_xid(xid);
1698         return rc;
1699 }
1700
1701 /*
1702  * update the file size (if needed) after a write. Should be called with
1703  * the inode->i_lock held
1704  */
1705 void
1706 cifs_update_eof(struct cifsInodeInfo *cifsi, loff_t offset,
1707                       unsigned int bytes_written)
1708 {
1709         loff_t end_of_write = offset + bytes_written;
1710
1711         if (end_of_write > cifsi->server_eof)
1712                 cifsi->server_eof = end_of_write;
1713 }
1714
1715 static ssize_t
1716 cifs_write(struct cifsFileInfo *open_file, __u32 pid, const char *write_data,
1717            size_t write_size, loff_t *offset)
1718 {
1719         int rc = 0;
1720         unsigned int bytes_written = 0;
1721         unsigned int total_written;
1722         struct cifs_sb_info *cifs_sb;
1723         struct cifs_tcon *tcon;
1724         struct TCP_Server_Info *server;
1725         unsigned int xid;
1726         struct dentry *dentry = open_file->dentry;
1727         struct cifsInodeInfo *cifsi = CIFS_I(d_inode(dentry));
1728         struct cifs_io_parms io_parms;
1729
1730         cifs_sb = CIFS_SB(dentry->d_sb);
1731
1732         cifs_dbg(FYI, "write %zd bytes to offset %lld of %pd\n",
1733                  write_size, *offset, dentry);
1734
1735         tcon = tlink_tcon(open_file->tlink);
1736         server = tcon->ses->server;
1737
1738         if (!server->ops->sync_write)
1739                 return -ENOSYS;
1740
1741         xid = get_xid();
1742
1743         for (total_written = 0; write_size > total_written;
1744              total_written += bytes_written) {
1745                 rc = -EAGAIN;
1746                 while (rc == -EAGAIN) {
1747                         struct kvec iov[2];
1748                         unsigned int len;
1749
1750                         if (open_file->invalidHandle) {
1751                                 /* we could deadlock if we called
1752                                    filemap_fdatawait from here so tell
1753                                    reopen_file not to flush data to
1754                                    server now */
1755                                 rc = cifs_reopen_file(open_file, false);
1756                                 if (rc != 0)
1757                                         break;
1758                         }
1759
1760                         len = min(server->ops->wp_retry_size(d_inode(dentry)),
1761                                   (unsigned int)write_size - total_written);
1762                         /* iov[0] is reserved for smb header */
1763                         iov[1].iov_base = (char *)write_data + total_written;
1764                         iov[1].iov_len = len;
1765                         io_parms.pid = pid;
1766                         io_parms.tcon = tcon;
1767                         io_parms.offset = *offset;
1768                         io_parms.length = len;
1769                         rc = server->ops->sync_write(xid, &open_file->fid,
1770                                         &io_parms, &bytes_written, iov, 1);
1771                 }
1772                 if (rc || (bytes_written == 0)) {
1773                         if (total_written)
1774                                 break;
1775                         else {
1776                                 free_xid(xid);
1777                                 return rc;
1778                         }
1779                 } else {
1780                         spin_lock(&d_inode(dentry)->i_lock);
1781                         cifs_update_eof(cifsi, *offset, bytes_written);
1782                         spin_unlock(&d_inode(dentry)->i_lock);
1783                         *offset += bytes_written;
1784                 }
1785         }
1786
1787         cifs_stats_bytes_written(tcon, total_written);
1788
1789         if (total_written > 0) {
1790                 spin_lock(&d_inode(dentry)->i_lock);
1791                 if (*offset > d_inode(dentry)->i_size)
1792                         i_size_write(d_inode(dentry), *offset);
1793                 spin_unlock(&d_inode(dentry)->i_lock);
1794         }
1795         mark_inode_dirty_sync(d_inode(dentry));
1796         free_xid(xid);
1797         return total_written;
1798 }
1799
1800 struct cifsFileInfo *find_readable_file(struct cifsInodeInfo *cifs_inode,
1801                                         bool fsuid_only)
1802 {
1803         struct cifsFileInfo *open_file = NULL;
1804         struct cifs_sb_info *cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1805         struct cifs_tcon *tcon = cifs_sb_master_tcon(cifs_sb);
1806
1807         /* only filter by fsuid on multiuser mounts */
1808         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1809                 fsuid_only = false;
1810
1811         spin_lock(&tcon->open_file_lock);
1812         /* we could simply get the first_list_entry since write-only entries
1813            are always at the end of the list but since the first entry might
1814            have a close pending, we go through the whole list */
1815         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1816                 if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
1817                         continue;
1818                 if (OPEN_FMODE(open_file->f_flags) & FMODE_READ) {
1819                         if (!open_file->invalidHandle) {
1820                                 /* found a good file */
1821                                 /* lock it so it will not be closed on us */
1822                                 cifsFileInfo_get(open_file);
1823                                 spin_unlock(&tcon->open_file_lock);
1824                                 return open_file;
1825                         } /* else might as well continue, and look for
1826                              another, or simply have the caller reopen it
1827                              again rather than trying to fix this handle */
1828                 } else /* write only file */
1829                         break; /* write only files are last so must be done */
1830         }
1831         spin_unlock(&tcon->open_file_lock);
1832         return NULL;
1833 }
1834
1835 struct cifsFileInfo *find_writable_file(struct cifsInodeInfo *cifs_inode,
1836                                         bool fsuid_only)
1837 {
1838         struct cifsFileInfo *open_file, *inv_file = NULL;
1839         struct cifs_sb_info *cifs_sb;
1840         struct cifs_tcon *tcon;
1841         bool any_available = false;
1842         int rc;
1843         unsigned int refind = 0;
1844
1845         /* Having a null inode here (because mapping->host was set to zero by
1846         the VFS or MM) should not happen but we had reports of on oops (due to
1847         it being zero) during stress testcases so we need to check for it */
1848
1849         if (cifs_inode == NULL) {
1850                 cifs_dbg(VFS, "Null inode passed to cifs_writeable_file\n");
1851                 dump_stack();
1852                 return NULL;
1853         }
1854
1855         cifs_sb = CIFS_SB(cifs_inode->vfs_inode.i_sb);
1856         tcon = cifs_sb_master_tcon(cifs_sb);
1857
1858         /* only filter by fsuid on multiuser mounts */
1859         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_MULTIUSER))
1860                 fsuid_only = false;
1861
1862         spin_lock(&tcon->open_file_lock);
1863 refind_writable:
1864         if (refind > MAX_REOPEN_ATT) {
1865                 spin_unlock(&tcon->open_file_lock);
1866                 return NULL;
1867         }
1868         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
1869                 if (!any_available && open_file->pid != current->tgid)
1870                         continue;
1871                 if (fsuid_only && !uid_eq(open_file->uid, current_fsuid()))
1872                         continue;
1873                 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
1874                         if (!open_file->invalidHandle) {
1875                                 /* found a good writable file */
1876                                 cifsFileInfo_get(open_file);
1877                                 spin_unlock(&tcon->open_file_lock);
1878                                 return open_file;
1879                         } else {
1880                                 if (!inv_file)
1881                                         inv_file = open_file;
1882                         }
1883                 }
1884         }
1885         /* couldn't find useable FH with same pid, try any available */
1886         if (!any_available) {
1887                 any_available = true;
1888                 goto refind_writable;
1889         }
1890
1891         if (inv_file) {
1892                 any_available = false;
1893                 cifsFileInfo_get(inv_file);
1894         }
1895
1896         spin_unlock(&tcon->open_file_lock);
1897
1898         if (inv_file) {
1899                 rc = cifs_reopen_file(inv_file, false);
1900                 if (!rc)
1901                         return inv_file;
1902                 else {
1903                         spin_lock(&tcon->open_file_lock);
1904                         list_move_tail(&inv_file->flist,
1905                                         &cifs_inode->openFileList);
1906                         spin_unlock(&tcon->open_file_lock);
1907                         cifsFileInfo_put(inv_file);
1908                         ++refind;
1909                         inv_file = NULL;
1910                         spin_lock(&tcon->open_file_lock);
1911                         goto refind_writable;
1912                 }
1913         }
1914
1915         return NULL;
1916 }
1917
1918 static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
1919 {
1920         struct address_space *mapping = page->mapping;
1921         loff_t offset = (loff_t)page->index << PAGE_SHIFT;
1922         char *write_data;
1923         int rc = -EFAULT;
1924         int bytes_written = 0;
1925         struct inode *inode;
1926         struct cifsFileInfo *open_file;
1927
1928         if (!mapping || !mapping->host)
1929                 return -EFAULT;
1930
1931         inode = page->mapping->host;
1932
1933         offset += (loff_t)from;
1934         write_data = kmap(page);
1935         write_data += from;
1936
1937         if ((to > PAGE_SIZE) || (from > to)) {
1938                 kunmap(page);
1939                 return -EIO;
1940         }
1941
1942         /* racing with truncate? */
1943         if (offset > mapping->host->i_size) {
1944                 kunmap(page);
1945                 return 0; /* don't care */
1946         }
1947
1948         /* check to make sure that we are not extending the file */
1949         if (mapping->host->i_size - offset < (loff_t)to)
1950                 to = (unsigned)(mapping->host->i_size - offset);
1951
1952         open_file = find_writable_file(CIFS_I(mapping->host), false);
1953         if (open_file) {
1954                 bytes_written = cifs_write(open_file, open_file->pid,
1955                                            write_data, to - from, &offset);
1956                 cifsFileInfo_put(open_file);
1957                 /* Does mm or vfs already set times? */
1958                 inode->i_atime = inode->i_mtime = current_time(inode);
1959                 if ((bytes_written > 0) && (offset))
1960                         rc = 0;
1961                 else if (bytes_written < 0)
1962                         rc = bytes_written;
1963         } else {
1964                 cifs_dbg(FYI, "No writeable filehandles for inode\n");
1965                 rc = -EIO;
1966         }
1967
1968         kunmap(page);
1969         return rc;
1970 }
1971
1972 static struct cifs_writedata *
1973 wdata_alloc_and_fillpages(pgoff_t tofind, struct address_space *mapping,
1974                           pgoff_t end, pgoff_t *index,
1975                           unsigned int *found_pages)
1976 {
1977         struct cifs_writedata *wdata;
1978
1979         wdata = cifs_writedata_alloc((unsigned int)tofind,
1980                                      cifs_writev_complete);
1981         if (!wdata)
1982                 return NULL;
1983
1984         *found_pages = find_get_pages_range_tag(mapping, index, end,
1985                                 PAGECACHE_TAG_DIRTY, tofind, wdata->pages);
1986         return wdata;
1987 }
1988
1989 static unsigned int
1990 wdata_prepare_pages(struct cifs_writedata *wdata, unsigned int found_pages,
1991                     struct address_space *mapping,
1992                     struct writeback_control *wbc,
1993                     pgoff_t end, pgoff_t *index, pgoff_t *next, bool *done)
1994 {
1995         unsigned int nr_pages = 0, i;
1996         struct page *page;
1997
1998         for (i = 0; i < found_pages; i++) {
1999                 page = wdata->pages[i];
2000                 /*
2001                  * At this point we hold neither the i_pages lock nor the
2002                  * page lock: the page may be truncated or invalidated
2003                  * (changing page->mapping to NULL), or even swizzled
2004                  * back from swapper_space to tmpfs file mapping
2005                  */
2006
2007                 if (nr_pages == 0)
2008                         lock_page(page);
2009                 else if (!trylock_page(page))
2010                         break;
2011
2012                 if (unlikely(page->mapping != mapping)) {
2013                         unlock_page(page);
2014                         break;
2015                 }
2016
2017                 if (!wbc->range_cyclic && page->index > end) {
2018                         *done = true;
2019                         unlock_page(page);
2020                         break;
2021                 }
2022
2023                 if (*next && (page->index != *next)) {
2024                         /* Not next consecutive page */
2025                         unlock_page(page);
2026                         break;
2027                 }
2028
2029                 if (wbc->sync_mode != WB_SYNC_NONE)
2030                         wait_on_page_writeback(page);
2031
2032                 if (PageWriteback(page) ||
2033                                 !clear_page_dirty_for_io(page)) {
2034                         unlock_page(page);
2035                         break;
2036                 }
2037
2038                 /*
2039                  * This actually clears the dirty bit in the radix tree.
2040                  * See cifs_writepage() for more commentary.
2041                  */
2042                 set_page_writeback(page);
2043                 if (page_offset(page) >= i_size_read(mapping->host)) {
2044                         *done = true;
2045                         unlock_page(page);
2046                         end_page_writeback(page);
2047                         break;
2048                 }
2049
2050                 wdata->pages[i] = page;
2051                 *next = page->index + 1;
2052                 ++nr_pages;
2053         }
2054
2055         /* reset index to refind any pages skipped */
2056         if (nr_pages == 0)
2057                 *index = wdata->pages[0]->index + 1;
2058
2059         /* put any pages we aren't going to use */
2060         for (i = nr_pages; i < found_pages; i++) {
2061                 put_page(wdata->pages[i]);
2062                 wdata->pages[i] = NULL;
2063         }
2064
2065         return nr_pages;
2066 }
2067
2068 static int
2069 wdata_send_pages(struct cifs_writedata *wdata, unsigned int nr_pages,
2070                  struct address_space *mapping, struct writeback_control *wbc)
2071 {
2072         int rc = 0;
2073         struct TCP_Server_Info *server;
2074         unsigned int i;
2075
2076         wdata->sync_mode = wbc->sync_mode;
2077         wdata->nr_pages = nr_pages;
2078         wdata->offset = page_offset(wdata->pages[0]);
2079         wdata->pagesz = PAGE_SIZE;
2080         wdata->tailsz = min(i_size_read(mapping->host) -
2081                         page_offset(wdata->pages[nr_pages - 1]),
2082                         (loff_t)PAGE_SIZE);
2083         wdata->bytes = ((nr_pages - 1) * PAGE_SIZE) + wdata->tailsz;
2084
2085         if (wdata->cfile != NULL)
2086                 cifsFileInfo_put(wdata->cfile);
2087         wdata->cfile = find_writable_file(CIFS_I(mapping->host), false);
2088         if (!wdata->cfile) {
2089                 cifs_dbg(VFS, "No writable handles for inode\n");
2090                 rc = -EBADF;
2091         } else {
2092                 wdata->pid = wdata->cfile->pid;
2093                 server = tlink_tcon(wdata->cfile->tlink)->ses->server;
2094                 rc = server->ops->async_writev(wdata, cifs_writedata_release);
2095         }
2096
2097         for (i = 0; i < nr_pages; ++i)
2098                 unlock_page(wdata->pages[i]);
2099
2100         return rc;
2101 }
2102
2103 static int cifs_writepages(struct address_space *mapping,
2104                            struct writeback_control *wbc)
2105 {
2106         struct cifs_sb_info *cifs_sb = CIFS_SB(mapping->host->i_sb);
2107         struct TCP_Server_Info *server;
2108         bool done = false, scanned = false, range_whole = false;
2109         pgoff_t end, index;
2110         struct cifs_writedata *wdata;
2111         int rc = 0;
2112         unsigned int xid;
2113
2114         /*
2115          * If wsize is smaller than the page cache size, default to writing
2116          * one page at a time via cifs_writepage
2117          */
2118         if (cifs_sb->wsize < PAGE_SIZE)
2119                 return generic_writepages(mapping, wbc);
2120
2121         xid = get_xid();
2122         if (wbc->range_cyclic) {
2123                 index = mapping->writeback_index; /* Start from prev offset */
2124                 end = -1;
2125         } else {
2126                 index = wbc->range_start >> PAGE_SHIFT;
2127                 end = wbc->range_end >> PAGE_SHIFT;
2128                 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2129                         range_whole = true;
2130                 scanned = true;
2131         }
2132         server = cifs_sb_master_tcon(cifs_sb)->ses->server;
2133 retry:
2134         while (!done && index <= end) {
2135                 unsigned int i, nr_pages, found_pages, wsize, credits;
2136                 pgoff_t next = 0, tofind, saved_index = index;
2137
2138                 rc = server->ops->wait_mtu_credits(server, cifs_sb->wsize,
2139                                                    &wsize, &credits);
2140                 if (rc)
2141                         break;
2142
2143                 tofind = min((wsize / PAGE_SIZE) - 1, end - index) + 1;
2144
2145                 wdata = wdata_alloc_and_fillpages(tofind, mapping, end, &index,
2146                                                   &found_pages);
2147                 if (!wdata) {
2148                         rc = -ENOMEM;
2149                         add_credits_and_wake_if(server, credits, 0);
2150                         break;
2151                 }
2152
2153                 if (found_pages == 0) {
2154                         kref_put(&wdata->refcount, cifs_writedata_release);
2155                         add_credits_and_wake_if(server, credits, 0);
2156                         break;
2157                 }
2158
2159                 nr_pages = wdata_prepare_pages(wdata, found_pages, mapping, wbc,
2160                                                end, &index, &next, &done);
2161
2162                 /* nothing to write? */
2163                 if (nr_pages == 0) {
2164                         kref_put(&wdata->refcount, cifs_writedata_release);
2165                         add_credits_and_wake_if(server, credits, 0);
2166                         continue;
2167                 }
2168
2169                 wdata->credits = credits;
2170
2171                 rc = wdata_send_pages(wdata, nr_pages, mapping, wbc);
2172
2173                 /* send failure -- clean up the mess */
2174                 if (rc != 0) {
2175                         add_credits_and_wake_if(server, wdata->credits, 0);
2176                         for (i = 0; i < nr_pages; ++i) {
2177                                 if (rc == -EAGAIN)
2178                                         redirty_page_for_writepage(wbc,
2179                                                            wdata->pages[i]);
2180                                 else
2181                                         SetPageError(wdata->pages[i]);
2182                                 end_page_writeback(wdata->pages[i]);
2183                                 put_page(wdata->pages[i]);
2184                         }
2185                         if (rc != -EAGAIN)
2186                                 mapping_set_error(mapping, rc);
2187                 }
2188                 kref_put(&wdata->refcount, cifs_writedata_release);
2189
2190                 if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN) {
2191                         index = saved_index;
2192                         continue;
2193                 }
2194
2195                 wbc->nr_to_write -= nr_pages;
2196                 if (wbc->nr_to_write <= 0)
2197                         done = true;
2198
2199                 index = next;
2200         }
2201
2202         if (!scanned && !done) {
2203                 /*
2204                  * We hit the last page and there is more work to be done: wrap
2205                  * back to the start of the file
2206                  */
2207                 scanned = true;
2208                 index = 0;
2209                 goto retry;
2210         }
2211
2212         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2213                 mapping->writeback_index = index;
2214
2215         free_xid(xid);
2216         return rc;
2217 }
2218
2219 static int
2220 cifs_writepage_locked(struct page *page, struct writeback_control *wbc)
2221 {
2222         int rc;
2223         unsigned int xid;
2224
2225         xid = get_xid();
2226 /* BB add check for wbc flags */
2227         get_page(page);
2228         if (!PageUptodate(page))
2229                 cifs_dbg(FYI, "ppw - page not up to date\n");
2230
2231         /*
2232          * Set the "writeback" flag, and clear "dirty" in the radix tree.
2233          *
2234          * A writepage() implementation always needs to do either this,
2235          * or re-dirty the page with "redirty_page_for_writepage()" in
2236          * the case of a failure.
2237          *
2238          * Just unlocking the page will cause the radix tree tag-bits
2239          * to fail to update with the state of the page correctly.
2240          */
2241         set_page_writeback(page);
2242 retry_write:
2243         rc = cifs_partialpagewrite(page, 0, PAGE_SIZE);
2244         if (rc == -EAGAIN) {
2245                 if (wbc->sync_mode == WB_SYNC_ALL)
2246                         goto retry_write;
2247                 redirty_page_for_writepage(wbc, page);
2248         } else if (rc != 0) {
2249                 SetPageError(page);
2250                 mapping_set_error(page->mapping, rc);
2251         } else {
2252                 SetPageUptodate(page);
2253         }
2254         end_page_writeback(page);
2255         put_page(page);
2256         free_xid(xid);
2257         return rc;
2258 }
2259
2260 static int cifs_writepage(struct page *page, struct writeback_control *wbc)
2261 {
2262         int rc = cifs_writepage_locked(page, wbc);
2263         unlock_page(page);
2264         return rc;
2265 }
2266
2267 static int cifs_write_end(struct file *file, struct address_space *mapping,
2268                         loff_t pos, unsigned len, unsigned copied,
2269                         struct page *page, void *fsdata)
2270 {
2271         int rc;
2272         struct inode *inode = mapping->host;
2273         struct cifsFileInfo *cfile = file->private_data;
2274         struct cifs_sb_info *cifs_sb = CIFS_SB(cfile->dentry->d_sb);
2275         __u32 pid;
2276
2277         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2278                 pid = cfile->pid;
2279         else
2280                 pid = current->tgid;
2281
2282         cifs_dbg(FYI, "write_end for page %p from pos %lld with %d bytes\n",
2283                  page, pos, copied);
2284
2285         if (PageChecked(page)) {
2286                 if (copied == len)
2287                         SetPageUptodate(page);
2288                 ClearPageChecked(page);
2289         } else if (!PageUptodate(page) && copied == PAGE_SIZE)
2290                 SetPageUptodate(page);
2291
2292         if (!PageUptodate(page)) {
2293                 char *page_data;
2294                 unsigned offset = pos & (PAGE_SIZE - 1);
2295                 unsigned int xid;
2296
2297                 xid = get_xid();
2298                 /* this is probably better than directly calling
2299                    partialpage_write since in this function the file handle is
2300                    known which we might as well leverage */
2301                 /* BB check if anything else missing out of ppw
2302                    such as updating last write time */
2303                 page_data = kmap(page);
2304                 rc = cifs_write(cfile, pid, page_data + offset, copied, &pos);
2305                 /* if (rc < 0) should we set writebehind rc? */
2306                 kunmap(page);
2307
2308                 free_xid(xid);
2309         } else {
2310                 rc = copied;
2311                 pos += copied;
2312                 set_page_dirty(page);
2313         }
2314
2315         if (rc > 0) {
2316                 spin_lock(&inode->i_lock);
2317                 if (pos > inode->i_size)
2318                         i_size_write(inode, pos);
2319                 spin_unlock(&inode->i_lock);
2320         }
2321
2322         unlock_page(page);
2323         put_page(page);
2324
2325         return rc;
2326 }
2327
2328 int cifs_strict_fsync(struct file *file, loff_t start, loff_t end,
2329                       int datasync)
2330 {
2331         unsigned int xid;
2332         int rc = 0;
2333         struct cifs_tcon *tcon;
2334         struct TCP_Server_Info *server;
2335         struct cifsFileInfo *smbfile = file->private_data;
2336         struct inode *inode = file_inode(file);
2337         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2338
2339         rc = file_write_and_wait_range(file, start, end);
2340         if (rc)
2341                 return rc;
2342         inode_lock(inode);
2343
2344         xid = get_xid();
2345
2346         cifs_dbg(FYI, "Sync file - name: %pD datasync: 0x%x\n",
2347                  file, datasync);
2348
2349         if (!CIFS_CACHE_READ(CIFS_I(inode))) {
2350                 rc = cifs_zap_mapping(inode);
2351                 if (rc) {
2352                         cifs_dbg(FYI, "rc: %d during invalidate phase\n", rc);
2353                         rc = 0; /* don't care about it in fsync */
2354                 }
2355         }
2356
2357         tcon = tlink_tcon(smbfile->tlink);
2358         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2359                 server = tcon->ses->server;
2360                 if (server->ops->flush)
2361                         rc = server->ops->flush(xid, tcon, &smbfile->fid);
2362                 else
2363                         rc = -ENOSYS;
2364         }
2365
2366         free_xid(xid);
2367         inode_unlock(inode);
2368         return rc;
2369 }
2370
2371 int cifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
2372 {
2373         unsigned int xid;
2374         int rc = 0;
2375         struct cifs_tcon *tcon;
2376         struct TCP_Server_Info *server;
2377         struct cifsFileInfo *smbfile = file->private_data;
2378         struct cifs_sb_info *cifs_sb = CIFS_FILE_SB(file);
2379         struct inode *inode = file->f_mapping->host;
2380
2381         rc = file_write_and_wait_range(file, start, end);
2382         if (rc)
2383                 return rc;
2384         inode_lock(inode);
2385
2386         xid = get_xid();
2387
2388         cifs_dbg(FYI, "Sync file - name: %pD datasync: 0x%x\n",
2389                  file, datasync);
2390
2391         tcon = tlink_tcon(smbfile->tlink);
2392         if (!(cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOSSYNC)) {
2393                 server = tcon->ses->server;
2394                 if (server->ops->flush)
2395                         rc = server->ops->flush(xid, tcon, &smbfile->fid);
2396                 else
2397                         rc = -ENOSYS;
2398         }
2399
2400         free_xid(xid);
2401         inode_unlock(inode);
2402         return rc;
2403 }
2404
2405 /*
2406  * As file closes, flush all cached write data for this inode checking
2407  * for write behind errors.
2408  */
2409 int cifs_flush(struct file *file, fl_owner_t id)
2410 {
2411         struct inode *inode = file_inode(file);
2412         int rc = 0;
2413
2414         if (file->f_mode & FMODE_WRITE)
2415                 rc = filemap_write_and_wait(inode->i_mapping);
2416
2417         cifs_dbg(FYI, "Flush inode %p file %p rc %d\n", inode, file, rc);
2418
2419         return rc;
2420 }
2421
2422 static int
2423 cifs_write_allocate_pages(struct page **pages, unsigned long num_pages)
2424 {
2425         int rc = 0;
2426         unsigned long i;
2427
2428         for (i = 0; i < num_pages; i++) {
2429                 pages[i] = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
2430                 if (!pages[i]) {
2431                         /*
2432                          * save number of pages we have already allocated and
2433                          * return with ENOMEM error
2434                          */
2435                         num_pages = i;
2436                         rc = -ENOMEM;
2437                         break;
2438                 }
2439         }
2440
2441         if (rc) {
2442                 for (i = 0; i < num_pages; i++)
2443                         put_page(pages[i]);
2444         }
2445         return rc;
2446 }
2447
2448 static inline
2449 size_t get_numpages(const size_t wsize, const size_t len, size_t *cur_len)
2450 {
2451         size_t num_pages;
2452         size_t clen;
2453
2454         clen = min_t(const size_t, len, wsize);
2455         num_pages = DIV_ROUND_UP(clen, PAGE_SIZE);
2456
2457         if (cur_len)
2458                 *cur_len = clen;
2459
2460         return num_pages;
2461 }
2462
2463 static void
2464 cifs_uncached_writedata_release(struct kref *refcount)
2465 {
2466         int i;
2467         struct cifs_writedata *wdata = container_of(refcount,
2468                                         struct cifs_writedata, refcount);
2469
2470         kref_put(&wdata->ctx->refcount, cifs_aio_ctx_release);
2471         for (i = 0; i < wdata->nr_pages; i++)
2472                 put_page(wdata->pages[i]);
2473         cifs_writedata_release(refcount);
2474 }
2475
2476 static void collect_uncached_write_data(struct cifs_aio_ctx *ctx);
2477
2478 static void
2479 cifs_uncached_writev_complete(struct work_struct *work)
2480 {
2481         struct cifs_writedata *wdata = container_of(work,
2482                                         struct cifs_writedata, work);
2483         struct inode *inode = d_inode(wdata->cfile->dentry);
2484         struct cifsInodeInfo *cifsi = CIFS_I(inode);
2485
2486         spin_lock(&inode->i_lock);
2487         cifs_update_eof(cifsi, wdata->offset, wdata->bytes);
2488         if (cifsi->server_eof > inode->i_size)
2489                 i_size_write(inode, cifsi->server_eof);
2490         spin_unlock(&inode->i_lock);
2491
2492         complete(&wdata->done);
2493         collect_uncached_write_data(wdata->ctx);
2494         /* the below call can possibly free the last ref to aio ctx */
2495         kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2496 }
2497
2498 static int
2499 wdata_fill_from_iovec(struct cifs_writedata *wdata, struct iov_iter *from,
2500                       size_t *len, unsigned long *num_pages)
2501 {
2502         size_t save_len, copied, bytes, cur_len = *len;
2503         unsigned long i, nr_pages = *num_pages;
2504
2505         save_len = cur_len;
2506         for (i = 0; i < nr_pages; i++) {
2507                 bytes = min_t(const size_t, cur_len, PAGE_SIZE);
2508                 copied = copy_page_from_iter(wdata->pages[i], 0, bytes, from);
2509                 cur_len -= copied;
2510                 /*
2511                  * If we didn't copy as much as we expected, then that
2512                  * may mean we trod into an unmapped area. Stop copying
2513                  * at that point. On the next pass through the big
2514                  * loop, we'll likely end up getting a zero-length
2515                  * write and bailing out of it.
2516                  */
2517                 if (copied < bytes)
2518                         break;
2519         }
2520         cur_len = save_len - cur_len;
2521         *len = cur_len;
2522
2523         /*
2524          * If we have no data to send, then that probably means that
2525          * the copy above failed altogether. That's most likely because
2526          * the address in the iovec was bogus. Return -EFAULT and let
2527          * the caller free anything we allocated and bail out.
2528          */
2529         if (!cur_len)
2530                 return -EFAULT;
2531
2532         /*
2533          * i + 1 now represents the number of pages we actually used in
2534          * the copy phase above.
2535          */
2536         *num_pages = i + 1;
2537         return 0;
2538 }
2539
2540 static int
2541 cifs_resend_wdata(struct cifs_writedata *wdata, struct list_head *wdata_list,
2542         struct cifs_aio_ctx *ctx)
2543 {
2544         unsigned int wsize, credits;
2545         int rc;
2546         struct TCP_Server_Info *server =
2547                 tlink_tcon(wdata->cfile->tlink)->ses->server;
2548
2549         /*
2550          * Wait for credits to resend this wdata.
2551          * Note: we are attempting to resend the whole wdata not in segments
2552          */
2553         do {
2554                 rc = server->ops->wait_mtu_credits(
2555                         server, wdata->bytes, &wsize, &credits);
2556
2557                 if (rc)
2558                         goto out;
2559
2560                 if (wsize < wdata->bytes) {
2561                         add_credits_and_wake_if(server, credits, 0);
2562                         msleep(1000);
2563                 }
2564         } while (wsize < wdata->bytes);
2565
2566         rc = -EAGAIN;
2567         while (rc == -EAGAIN) {
2568                 rc = 0;
2569                 if (wdata->cfile->invalidHandle)
2570                         rc = cifs_reopen_file(wdata->cfile, false);
2571                 if (!rc)
2572                         rc = server->ops->async_writev(wdata,
2573                                         cifs_uncached_writedata_release);
2574         }
2575
2576         if (!rc) {
2577                 list_add_tail(&wdata->list, wdata_list);
2578                 return 0;
2579         }
2580
2581         add_credits_and_wake_if(server, wdata->credits, 0);
2582 out:
2583         kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2584
2585         return rc;
2586 }
2587
2588 static int
2589 cifs_write_from_iter(loff_t offset, size_t len, struct iov_iter *from,
2590                      struct cifsFileInfo *open_file,
2591                      struct cifs_sb_info *cifs_sb, struct list_head *wdata_list,
2592                      struct cifs_aio_ctx *ctx)
2593 {
2594         int rc = 0;
2595         size_t cur_len;
2596         unsigned long nr_pages, num_pages, i;
2597         struct cifs_writedata *wdata;
2598         struct iov_iter saved_from = *from;
2599         loff_t saved_offset = offset;
2600         pid_t pid;
2601         struct TCP_Server_Info *server;
2602         struct page **pagevec;
2603         size_t start;
2604
2605         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
2606                 pid = open_file->pid;
2607         else
2608                 pid = current->tgid;
2609
2610         server = tlink_tcon(open_file->tlink)->ses->server;
2611
2612         do {
2613                 unsigned int wsize, credits;
2614
2615                 rc = server->ops->wait_mtu_credits(server, cifs_sb->wsize,
2616                                                    &wsize, &credits);
2617                 if (rc)
2618                         break;
2619
2620                 cur_len = min_t(const size_t, len, wsize);
2621
2622                 if (ctx->direct_io) {
2623                         ssize_t result;
2624
2625                         result = iov_iter_get_pages_alloc(
2626                                 from, &pagevec, cur_len, &start);
2627                         if (result < 0) {
2628                                 cifs_dbg(VFS,
2629                                         "direct_writev couldn't get user pages "
2630                                         "(rc=%zd) iter type %d iov_offset %zd "
2631                                         "count %zd\n",
2632                                         result, from->type,
2633                                         from->iov_offset, from->count);
2634                                 dump_stack();
2635
2636                                 rc = result;
2637                                 add_credits_and_wake_if(server, credits, 0);
2638                                 break;
2639                         }
2640                         cur_len = (size_t)result;
2641                         iov_iter_advance(from, cur_len);
2642
2643                         nr_pages =
2644                                 (cur_len + start + PAGE_SIZE - 1) / PAGE_SIZE;
2645
2646                         wdata = cifs_writedata_direct_alloc(pagevec,
2647                                              cifs_uncached_writev_complete);
2648                         if (!wdata) {
2649                                 rc = -ENOMEM;
2650                                 add_credits_and_wake_if(server, credits, 0);
2651                                 break;
2652                         }
2653
2654
2655                         wdata->page_offset = start;
2656                         wdata->tailsz =
2657                                 nr_pages > 1 ?
2658                                         cur_len - (PAGE_SIZE - start) -
2659                                         (nr_pages - 2) * PAGE_SIZE :
2660                                         cur_len;
2661                 } else {
2662                         nr_pages = get_numpages(wsize, len, &cur_len);
2663                         wdata = cifs_writedata_alloc(nr_pages,
2664                                              cifs_uncached_writev_complete);
2665                         if (!wdata) {
2666                                 rc = -ENOMEM;
2667                                 add_credits_and_wake_if(server, credits, 0);
2668                                 break;
2669                         }
2670
2671                         rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
2672                         if (rc) {
2673                                 kfree(wdata);
2674                                 add_credits_and_wake_if(server, credits, 0);
2675                                 break;
2676                         }
2677
2678                         num_pages = nr_pages;
2679                         rc = wdata_fill_from_iovec(
2680                                 wdata, from, &cur_len, &num_pages);
2681                         if (rc) {
2682                                 for (i = 0; i < nr_pages; i++)
2683                                         put_page(wdata->pages[i]);
2684                                 kfree(wdata);
2685                                 add_credits_and_wake_if(server, credits, 0);
2686                                 break;
2687                         }
2688
2689                         /*
2690                          * Bring nr_pages down to the number of pages we
2691                          * actually used, and free any pages that we didn't use.
2692                          */
2693                         for ( ; nr_pages > num_pages; nr_pages--)
2694                                 put_page(wdata->pages[nr_pages - 1]);
2695
2696                         wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
2697                 }
2698
2699                 wdata->sync_mode = WB_SYNC_ALL;
2700                 wdata->nr_pages = nr_pages;
2701                 wdata->offset = (__u64)offset;
2702                 wdata->cfile = cifsFileInfo_get(open_file);
2703                 wdata->pid = pid;
2704                 wdata->bytes = cur_len;
2705                 wdata->pagesz = PAGE_SIZE;
2706                 wdata->credits = credits;
2707                 wdata->ctx = ctx;
2708                 kref_get(&ctx->refcount);
2709
2710                 if (!wdata->cfile->invalidHandle ||
2711                     !(rc = cifs_reopen_file(wdata->cfile, false)))
2712                         rc = server->ops->async_writev(wdata,
2713                                         cifs_uncached_writedata_release);
2714                 if (rc) {
2715                         add_credits_and_wake_if(server, wdata->credits, 0);
2716                         kref_put(&wdata->refcount,
2717                                  cifs_uncached_writedata_release);
2718                         if (rc == -EAGAIN) {
2719                                 *from = saved_from;
2720                                 iov_iter_advance(from, offset - saved_offset);
2721                                 continue;
2722                         }
2723                         break;
2724                 }
2725
2726                 list_add_tail(&wdata->list, wdata_list);
2727                 offset += cur_len;
2728                 len -= cur_len;
2729         } while (len > 0);
2730
2731         return rc;
2732 }
2733
2734 static void collect_uncached_write_data(struct cifs_aio_ctx *ctx)
2735 {
2736         struct cifs_writedata *wdata, *tmp;
2737         struct cifs_tcon *tcon;
2738         struct cifs_sb_info *cifs_sb;
2739         struct dentry *dentry = ctx->cfile->dentry;
2740         unsigned int i;
2741         int rc;
2742
2743         tcon = tlink_tcon(ctx->cfile->tlink);
2744         cifs_sb = CIFS_SB(dentry->d_sb);
2745
2746         mutex_lock(&ctx->aio_mutex);
2747
2748         if (list_empty(&ctx->list)) {
2749                 mutex_unlock(&ctx->aio_mutex);
2750                 return;
2751         }
2752
2753         rc = ctx->rc;
2754         /*
2755          * Wait for and collect replies for any successful sends in order of
2756          * increasing offset. Once an error is hit, then return without waiting
2757          * for any more replies.
2758          */
2759 restart_loop:
2760         list_for_each_entry_safe(wdata, tmp, &ctx->list, list) {
2761                 if (!rc) {
2762                         if (!try_wait_for_completion(&wdata->done)) {
2763                                 mutex_unlock(&ctx->aio_mutex);
2764                                 return;
2765                         }
2766
2767                         if (wdata->result)
2768                                 rc = wdata->result;
2769                         else
2770                                 ctx->total_len += wdata->bytes;
2771
2772                         /* resend call if it's a retryable error */
2773                         if (rc == -EAGAIN) {
2774                                 struct list_head tmp_list;
2775                                 struct iov_iter tmp_from = ctx->iter;
2776
2777                                 INIT_LIST_HEAD(&tmp_list);
2778                                 list_del_init(&wdata->list);
2779
2780                                 if (ctx->direct_io)
2781                                         rc = cifs_resend_wdata(
2782                                                 wdata, &tmp_list, ctx);
2783                                 else {
2784                                         iov_iter_advance(&tmp_from,
2785                                                  wdata->offset - ctx->pos);
2786
2787                                         rc = cifs_write_from_iter(wdata->offset,
2788                                                 wdata->bytes, &tmp_from,
2789                                                 ctx->cfile, cifs_sb, &tmp_list,
2790                                                 ctx);
2791                                 }
2792
2793                                 list_splice(&tmp_list, &ctx->list);
2794
2795                                 kref_put(&wdata->refcount,
2796                                          cifs_uncached_writedata_release);
2797                                 goto restart_loop;
2798                         }
2799                 }
2800                 list_del_init(&wdata->list);
2801                 kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2802         }
2803
2804         if (!ctx->direct_io)
2805                 for (i = 0; i < ctx->npages; i++)
2806                         put_page(ctx->bv[i].bv_page);
2807
2808         cifs_stats_bytes_written(tcon, ctx->total_len);
2809         set_bit(CIFS_INO_INVALID_MAPPING, &CIFS_I(dentry->d_inode)->flags);
2810
2811         ctx->rc = (rc == 0) ? ctx->total_len : rc;
2812
2813         mutex_unlock(&ctx->aio_mutex);
2814
2815         if (ctx->iocb && ctx->iocb->ki_complete)
2816                 ctx->iocb->ki_complete(ctx->iocb, ctx->rc, 0);
2817         else
2818                 complete(&ctx->done);
2819 }
2820
2821 static ssize_t __cifs_writev(
2822         struct kiocb *iocb, struct iov_iter *from, bool direct)
2823 {
2824         struct file *file = iocb->ki_filp;
2825         ssize_t total_written = 0;
2826         struct cifsFileInfo *cfile;
2827         struct cifs_tcon *tcon;
2828         struct cifs_sb_info *cifs_sb;
2829         struct cifs_aio_ctx *ctx;
2830         struct iov_iter saved_from = *from;
2831         size_t len = iov_iter_count(from);
2832         int rc;
2833
2834         /*
2835          * iov_iter_get_pages_alloc doesn't work with ITER_KVEC.
2836          * In this case, fall back to non-direct write function.
2837          * this could be improved by getting pages directly in ITER_KVEC
2838          */
2839         if (direct && from->type & ITER_KVEC) {
2840                 cifs_dbg(FYI, "use non-direct cifs_writev for kvec I/O\n");
2841                 direct = false;
2842         }
2843
2844         rc = generic_write_checks(iocb, from);
2845         if (rc <= 0)
2846                 return rc;
2847
2848         cifs_sb = CIFS_FILE_SB(file);
2849         cfile = file->private_data;
2850         tcon = tlink_tcon(cfile->tlink);
2851
2852         if (!tcon->ses->server->ops->async_writev)
2853                 return -ENOSYS;
2854
2855         ctx = cifs_aio_ctx_alloc();
2856         if (!ctx)
2857                 return -ENOMEM;
2858
2859         ctx->cfile = cifsFileInfo_get(cfile);
2860
2861         if (!is_sync_kiocb(iocb))
2862                 ctx->iocb = iocb;
2863
2864         ctx->pos = iocb->ki_pos;
2865
2866         if (direct) {
2867                 ctx->direct_io = true;
2868                 ctx->iter = *from;
2869                 ctx->len = len;
2870         } else {
2871                 rc = setup_aio_ctx_iter(ctx, from, WRITE);
2872                 if (rc) {
2873                         kref_put(&ctx->refcount, cifs_aio_ctx_release);
2874                         return rc;
2875                 }
2876         }
2877
2878         /* grab a lock here due to read response handlers can access ctx */
2879         mutex_lock(&ctx->aio_mutex);
2880
2881         rc = cifs_write_from_iter(iocb->ki_pos, ctx->len, &saved_from,
2882                                   cfile, cifs_sb, &ctx->list, ctx);
2883
2884         /*
2885          * If at least one write was successfully sent, then discard any rc
2886          * value from the later writes. If the other write succeeds, then
2887          * we'll end up returning whatever was written. If it fails, then
2888          * we'll get a new rc value from that.
2889          */
2890         if (!list_empty(&ctx->list))
2891                 rc = 0;
2892
2893         mutex_unlock(&ctx->aio_mutex);
2894
2895         if (rc) {
2896                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
2897                 return rc;
2898         }
2899
2900         if (!is_sync_kiocb(iocb)) {
2901                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
2902                 return -EIOCBQUEUED;
2903         }
2904
2905         rc = wait_for_completion_killable(&ctx->done);
2906         if (rc) {
2907                 mutex_lock(&ctx->aio_mutex);
2908                 ctx->rc = rc = -EINTR;
2909                 total_written = ctx->total_len;
2910                 mutex_unlock(&ctx->aio_mutex);
2911         } else {
2912                 rc = ctx->rc;
2913                 total_written = ctx->total_len;
2914         }
2915
2916         kref_put(&ctx->refcount, cifs_aio_ctx_release);
2917
2918         if (unlikely(!total_written))
2919                 return rc;
2920
2921         iocb->ki_pos += total_written;
2922         return total_written;
2923 }
2924
2925 ssize_t cifs_direct_writev(struct kiocb *iocb, struct iov_iter *from)
2926 {
2927         return __cifs_writev(iocb, from, true);
2928 }
2929
2930 ssize_t cifs_user_writev(struct kiocb *iocb, struct iov_iter *from)
2931 {
2932         return __cifs_writev(iocb, from, false);
2933 }
2934
2935 static ssize_t
2936 cifs_writev(struct kiocb *iocb, struct iov_iter *from)
2937 {
2938         struct file *file = iocb->ki_filp;
2939         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
2940         struct inode *inode = file->f_mapping->host;
2941         struct cifsInodeInfo *cinode = CIFS_I(inode);
2942         struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
2943         ssize_t rc;
2944
2945         inode_lock(inode);
2946         /*
2947          * We need to hold the sem to be sure nobody modifies lock list
2948          * with a brlock that prevents writing.
2949          */
2950         down_read(&cinode->lock_sem);
2951
2952         rc = generic_write_checks(iocb, from);
2953         if (rc <= 0)
2954                 goto out;
2955
2956         if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(from),
2957                                      server->vals->exclusive_lock_type, 0,
2958                                      NULL, CIFS_WRITE_OP))
2959                 rc = __generic_file_write_iter(iocb, from);
2960         else
2961                 rc = -EACCES;
2962 out:
2963         up_read(&cinode->lock_sem);
2964         inode_unlock(inode);
2965
2966         if (rc > 0)
2967                 rc = generic_write_sync(iocb, rc);
2968         return rc;
2969 }
2970
2971 ssize_t
2972 cifs_strict_writev(struct kiocb *iocb, struct iov_iter *from)
2973 {
2974         struct inode *inode = file_inode(iocb->ki_filp);
2975         struct cifsInodeInfo *cinode = CIFS_I(inode);
2976         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2977         struct cifsFileInfo *cfile = (struct cifsFileInfo *)
2978                                                 iocb->ki_filp->private_data;
2979         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
2980         ssize_t written;
2981
2982         written = cifs_get_writer(cinode);
2983         if (written)
2984                 return written;
2985
2986         if (CIFS_CACHE_WRITE(cinode)) {
2987                 if (cap_unix(tcon->ses) &&
2988                 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability))
2989                   && ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0)) {
2990                         written = generic_file_write_iter(iocb, from);
2991                         goto out;
2992                 }
2993                 written = cifs_writev(iocb, from);
2994                 goto out;
2995         }
2996         /*
2997          * For non-oplocked files in strict cache mode we need to write the data
2998          * to the server exactly from the pos to pos+len-1 rather than flush all
2999          * affected pages because it may cause a error with mandatory locks on
3000          * these pages but not on the region from pos to ppos+len-1.
3001          */
3002         written = cifs_user_writev(iocb, from);
3003         if (written > 0 && CIFS_CACHE_READ(cinode)) {
3004                 /*
3005                  * Windows 7 server can delay breaking level2 oplock if a write
3006                  * request comes - break it on the client to prevent reading
3007                  * an old data.
3008                  */
3009                 cifs_zap_mapping(inode);
3010                 cifs_dbg(FYI, "Set no oplock for inode=%p after a write operation\n",
3011                          inode);
3012                 cinode->oplock = 0;
3013         }
3014 out:
3015         cifs_put_writer(cinode);
3016         return written;
3017 }
3018
3019 static struct cifs_readdata *
3020 cifs_readdata_direct_alloc(struct page **pages, work_func_t complete)
3021 {
3022         struct cifs_readdata *rdata;
3023
3024         rdata = kzalloc(sizeof(*rdata), GFP_KERNEL);
3025         if (rdata != NULL) {
3026                 rdata->pages = pages;
3027                 kref_init(&rdata->refcount);
3028                 INIT_LIST_HEAD(&rdata->list);
3029                 init_completion(&rdata->done);
3030                 INIT_WORK(&rdata->work, complete);
3031         }
3032
3033         return rdata;
3034 }
3035
3036 static struct cifs_readdata *
3037 cifs_readdata_alloc(unsigned int nr_pages, work_func_t complete)
3038 {
3039         struct page **pages =
3040                 kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
3041         struct cifs_readdata *ret = NULL;
3042
3043         if (pages) {
3044                 ret = cifs_readdata_direct_alloc(pages, complete);
3045                 if (!ret)
3046                         kfree(pages);
3047         }
3048
3049         return ret;
3050 }
3051
3052 void
3053 cifs_readdata_release(struct kref *refcount)
3054 {
3055         struct cifs_readdata *rdata = container_of(refcount,
3056                                         struct cifs_readdata, refcount);
3057 #ifdef CONFIG_CIFS_SMB_DIRECT
3058         if (rdata->mr) {
3059                 smbd_deregister_mr(rdata->mr);
3060                 rdata->mr = NULL;
3061         }
3062 #endif
3063         if (rdata->cfile)
3064                 cifsFileInfo_put(rdata->cfile);
3065
3066         kvfree(rdata->pages);
3067         kfree(rdata);
3068 }
3069
3070 static int
3071 cifs_read_allocate_pages(struct cifs_readdata *rdata, unsigned int nr_pages)
3072 {
3073         int rc = 0;
3074         struct page *page;
3075         unsigned int i;
3076
3077         for (i = 0; i < nr_pages; i++) {
3078                 page = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
3079                 if (!page) {
3080                         rc = -ENOMEM;
3081                         break;
3082                 }
3083                 rdata->pages[i] = page;
3084         }
3085
3086         if (rc) {
3087                 for (i = 0; i < nr_pages; i++) {
3088                         put_page(rdata->pages[i]);
3089                         rdata->pages[i] = NULL;
3090                 }
3091         }
3092         return rc;
3093 }
3094
3095 static void
3096 cifs_uncached_readdata_release(struct kref *refcount)
3097 {
3098         struct cifs_readdata *rdata = container_of(refcount,
3099                                         struct cifs_readdata, refcount);
3100         unsigned int i;
3101
3102         kref_put(&rdata->ctx->refcount, cifs_aio_ctx_release);
3103         for (i = 0; i < rdata->nr_pages; i++) {
3104                 put_page(rdata->pages[i]);
3105         }
3106         cifs_readdata_release(refcount);
3107 }
3108
3109 /**
3110  * cifs_readdata_to_iov - copy data from pages in response to an iovec
3111  * @rdata:      the readdata response with list of pages holding data
3112  * @iter:       destination for our data
3113  *
3114  * This function copies data from a list of pages in a readdata response into
3115  * an array of iovecs. It will first calculate where the data should go
3116  * based on the info in the readdata and then copy the data into that spot.
3117  */
3118 static int
3119 cifs_readdata_to_iov(struct cifs_readdata *rdata, struct iov_iter *iter)
3120 {
3121         size_t remaining = rdata->got_bytes;
3122         unsigned int i;
3123
3124         for (i = 0; i < rdata->nr_pages; i++) {
3125                 struct page *page = rdata->pages[i];
3126                 size_t copy = min_t(size_t, remaining, PAGE_SIZE);
3127                 size_t written;
3128
3129                 if (unlikely(iov_iter_is_pipe(iter))) {
3130                         void *addr = kmap_atomic(page);
3131
3132                         written = copy_to_iter(addr, copy, iter);
3133                         kunmap_atomic(addr);
3134                 } else
3135                         written = copy_page_to_iter(page, 0, copy, iter);
3136                 remaining -= written;
3137                 if (written < copy && iov_iter_count(iter) > 0)
3138                         break;
3139         }
3140         return remaining ? -EFAULT : 0;
3141 }
3142
3143 static void collect_uncached_read_data(struct cifs_aio_ctx *ctx);
3144
3145 static void
3146 cifs_uncached_readv_complete(struct work_struct *work)
3147 {
3148         struct cifs_readdata *rdata = container_of(work,
3149                                                 struct cifs_readdata, work);
3150
3151         complete(&rdata->done);
3152         collect_uncached_read_data(rdata->ctx);
3153         /* the below call can possibly free the last ref to aio ctx */
3154         kref_put(&rdata->refcount, cifs_uncached_readdata_release);
3155 }
3156
3157 static int
3158 uncached_fill_pages(struct TCP_Server_Info *server,
3159                     struct cifs_readdata *rdata, struct iov_iter *iter,
3160                     unsigned int len)
3161 {
3162         int result = 0;
3163         unsigned int i;
3164         unsigned int nr_pages = rdata->nr_pages;
3165         unsigned int page_offset = rdata->page_offset;
3166
3167         rdata->got_bytes = 0;
3168         rdata->tailsz = PAGE_SIZE;
3169         for (i = 0; i < nr_pages; i++) {
3170                 struct page *page = rdata->pages[i];
3171                 size_t n;
3172                 unsigned int segment_size = rdata->pagesz;
3173
3174                 if (i == 0)
3175                         segment_size -= page_offset;
3176                 else
3177                         page_offset = 0;
3178
3179
3180                 if (len <= 0) {
3181                         /* no need to hold page hostage */
3182                         rdata->pages[i] = NULL;
3183                         rdata->nr_pages--;
3184                         put_page(page);
3185                         continue;
3186                 }
3187
3188                 n = len;
3189                 if (len >= segment_size)
3190                         /* enough data to fill the page */
3191                         n = segment_size;
3192                 else
3193                         rdata->tailsz = len;
3194                 len -= n;
3195
3196                 if (iter)
3197                         result = copy_page_from_iter(
3198                                         page, page_offset, n, iter);
3199 #ifdef CONFIG_CIFS_SMB_DIRECT
3200                 else if (rdata->mr)
3201                         result = n;
3202 #endif
3203                 else
3204                         result = cifs_read_page_from_socket(
3205                                         server, page, page_offset, n);
3206                 if (result < 0)
3207                         break;
3208
3209                 rdata->got_bytes += result;
3210         }
3211
3212         return rdata->got_bytes > 0 && result != -ECONNABORTED ?
3213                                                 rdata->got_bytes : result;
3214 }
3215
3216 static int
3217 cifs_uncached_read_into_pages(struct TCP_Server_Info *server,
3218                               struct cifs_readdata *rdata, unsigned int len)
3219 {
3220         return uncached_fill_pages(server, rdata, NULL, len);
3221 }
3222
3223 static int
3224 cifs_uncached_copy_into_pages(struct TCP_Server_Info *server,
3225                               struct cifs_readdata *rdata,
3226                               struct iov_iter *iter)
3227 {
3228         return uncached_fill_pages(server, rdata, iter, iter->count);
3229 }
3230
3231 static int cifs_resend_rdata(struct cifs_readdata *rdata,
3232                         struct list_head *rdata_list,
3233                         struct cifs_aio_ctx *ctx)
3234 {
3235         unsigned int rsize, credits;
3236         int rc;
3237         struct TCP_Server_Info *server =
3238                 tlink_tcon(rdata->cfile->tlink)->ses->server;
3239
3240         /*
3241          * Wait for credits to resend this rdata.
3242          * Note: we are attempting to resend the whole rdata not in segments
3243          */
3244         do {
3245                 rc = server->ops->wait_mtu_credits(server, rdata->bytes,
3246                                                 &rsize, &credits);
3247
3248                 if (rc)
3249                         goto out;
3250
3251                 if (rsize < rdata->bytes) {
3252                         add_credits_and_wake_if(server, credits, 0);
3253                         msleep(1000);
3254                 }
3255         } while (rsize < rdata->bytes);
3256
3257         rc = -EAGAIN;
3258         while (rc == -EAGAIN) {
3259                 rc = 0;
3260                 if (rdata->cfile->invalidHandle)
3261                         rc = cifs_reopen_file(rdata->cfile, true);
3262                 if (!rc)
3263                         rc = server->ops->async_readv(rdata);
3264         }
3265
3266         if (!rc) {
3267                 /* Add to aio pending list */
3268                 list_add_tail(&rdata->list, rdata_list);
3269                 return 0;
3270         }
3271
3272         add_credits_and_wake_if(server, rdata->credits, 0);
3273 out:
3274         kref_put(&rdata->refcount,
3275                 cifs_uncached_readdata_release);
3276
3277         return rc;
3278 }
3279
3280 static int
3281 cifs_send_async_read(loff_t offset, size_t len, struct cifsFileInfo *open_file,
3282                      struct cifs_sb_info *cifs_sb, struct list_head *rdata_list,
3283                      struct cifs_aio_ctx *ctx)
3284 {
3285         struct cifs_readdata *rdata;
3286         unsigned int npages, rsize, credits;
3287         size_t cur_len;
3288         int rc;
3289         pid_t pid;
3290         struct TCP_Server_Info *server;
3291         struct page **pagevec;
3292         size_t start;
3293         struct iov_iter direct_iov = ctx->iter;
3294
3295         server = tlink_tcon(open_file->tlink)->ses->server;
3296
3297         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3298                 pid = open_file->pid;
3299         else
3300                 pid = current->tgid;
3301
3302         if (ctx->direct_io)
3303                 iov_iter_advance(&direct_iov, offset - ctx->pos);
3304
3305         do {
3306                 rc = server->ops->wait_mtu_credits(server, cifs_sb->rsize,
3307                                                    &rsize, &credits);
3308                 if (rc)
3309                         break;
3310
3311                 cur_len = min_t(const size_t, len, rsize);
3312
3313                 if (ctx->direct_io) {
3314                         ssize_t result;
3315
3316                         result = iov_iter_get_pages_alloc(
3317                                         &direct_iov, &pagevec,
3318                                         cur_len, &start);
3319                         if (result < 0) {
3320                                 cifs_dbg(VFS,
3321                                         "couldn't get user pages (rc=%zd)"
3322                                         " iter type %d"
3323                                         " iov_offset %zd count %zd\n",
3324                                         result, direct_iov.type,
3325                                         direct_iov.iov_offset,
3326                                         direct_iov.count);
3327                                 dump_stack();
3328
3329                                 rc = result;
3330                                 add_credits_and_wake_if(server, credits, 0);
3331                                 break;
3332                         }
3333                         cur_len = (size_t)result;
3334                         iov_iter_advance(&direct_iov, cur_len);
3335
3336                         rdata = cifs_readdata_direct_alloc(
3337                                         pagevec, cifs_uncached_readv_complete);
3338                         if (!rdata) {
3339                                 add_credits_and_wake_if(server, credits, 0);
3340                                 rc = -ENOMEM;
3341                                 break;
3342                         }
3343
3344                         npages = (cur_len + start + PAGE_SIZE-1) / PAGE_SIZE;
3345                         rdata->page_offset = start;
3346                         rdata->tailsz = npages > 1 ?
3347                                 cur_len-(PAGE_SIZE-start)-(npages-2)*PAGE_SIZE :
3348                                 cur_len;
3349
3350                 } else {
3351
3352                         npages = DIV_ROUND_UP(cur_len, PAGE_SIZE);
3353                         /* allocate a readdata struct */
3354                         rdata = cifs_readdata_alloc(npages,
3355                                             cifs_uncached_readv_complete);
3356                         if (!rdata) {
3357                                 add_credits_and_wake_if(server, credits, 0);
3358                                 rc = -ENOMEM;
3359                                 break;
3360                         }
3361
3362                         rc = cifs_read_allocate_pages(rdata, npages);
3363                         if (rc)
3364                                 goto error;
3365
3366                         rdata->tailsz = PAGE_SIZE;
3367                 }
3368
3369                 rdata->cfile = cifsFileInfo_get(open_file);
3370                 rdata->nr_pages = npages;
3371                 rdata->offset = offset;
3372                 rdata->bytes = cur_len;
3373                 rdata->pid = pid;
3374                 rdata->pagesz = PAGE_SIZE;
3375                 rdata->read_into_pages = cifs_uncached_read_into_pages;
3376                 rdata->copy_into_pages = cifs_uncached_copy_into_pages;
3377                 rdata->credits = credits;
3378                 rdata->ctx = ctx;
3379                 kref_get(&ctx->refcount);
3380
3381                 if (!rdata->cfile->invalidHandle ||
3382                     !(rc = cifs_reopen_file(rdata->cfile, true)))
3383                         rc = server->ops->async_readv(rdata);
3384 error:
3385                 if (rc) {
3386                         add_credits_and_wake_if(server, rdata->credits, 0);
3387                         kref_put(&rdata->refcount,
3388                                 cifs_uncached_readdata_release);
3389                         if (rc == -EAGAIN) {
3390                                 iov_iter_revert(&direct_iov, cur_len);
3391                                 continue;
3392                         }
3393                         break;
3394                 }
3395
3396                 list_add_tail(&rdata->list, rdata_list);
3397                 offset += cur_len;
3398                 len -= cur_len;
3399         } while (len > 0);
3400
3401         return rc;
3402 }
3403
3404 static void
3405 collect_uncached_read_data(struct cifs_aio_ctx *ctx)
3406 {
3407         struct cifs_readdata *rdata, *tmp;
3408         struct iov_iter *to = &ctx->iter;
3409         struct cifs_sb_info *cifs_sb;
3410         struct cifs_tcon *tcon;
3411         unsigned int i;
3412         int rc;
3413
3414         tcon = tlink_tcon(ctx->cfile->tlink);
3415         cifs_sb = CIFS_SB(ctx->cfile->dentry->d_sb);
3416
3417         mutex_lock(&ctx->aio_mutex);
3418
3419         if (list_empty(&ctx->list)) {
3420                 mutex_unlock(&ctx->aio_mutex);
3421                 return;
3422         }
3423
3424         rc = ctx->rc;
3425         /* the loop below should proceed in the order of increasing offsets */
3426 again:
3427         list_for_each_entry_safe(rdata, tmp, &ctx->list, list) {
3428                 if (!rc) {
3429                         if (!try_wait_for_completion(&rdata->done)) {
3430                                 mutex_unlock(&ctx->aio_mutex);
3431                                 return;
3432                         }
3433
3434                         if (rdata->result == -EAGAIN) {
3435                                 /* resend call if it's a retryable error */
3436                                 struct list_head tmp_list;
3437                                 unsigned int got_bytes = rdata->got_bytes;
3438
3439                                 list_del_init(&rdata->list);
3440                                 INIT_LIST_HEAD(&tmp_list);
3441
3442                                 /*
3443                                  * Got a part of data and then reconnect has
3444                                  * happened -- fill the buffer and continue
3445                                  * reading.
3446                                  */
3447                                 if (got_bytes && got_bytes < rdata->bytes) {
3448                                         rc = 0;
3449                                         if (!ctx->direct_io)
3450                                                 rc = cifs_readdata_to_iov(rdata, to);
3451                                         if (rc) {
3452                                                 kref_put(&rdata->refcount,
3453                                                         cifs_uncached_readdata_release);
3454                                                 continue;
3455                                         }
3456                                 }
3457
3458                                 if (ctx->direct_io) {
3459                                         /*
3460                                          * Re-use rdata as this is a
3461                                          * direct I/O
3462                                          */
3463                                         rc = cifs_resend_rdata(
3464                                                 rdata,
3465                                                 &tmp_list, ctx);
3466                                 } else {
3467                                         rc = cifs_send_async_read(
3468                                                 rdata->offset + got_bytes,
3469                                                 rdata->bytes - got_bytes,
3470                                                 rdata->cfile, cifs_sb,
3471                                                 &tmp_list, ctx);
3472
3473                                         kref_put(&rdata->refcount,
3474                                                 cifs_uncached_readdata_release);
3475                                 }
3476
3477                                 list_splice(&tmp_list, &ctx->list);
3478
3479                                 goto again;
3480                         } else if (rdata->result)
3481                                 rc = rdata->result;
3482                         else if (!ctx->direct_io)
3483                                 rc = cifs_readdata_to_iov(rdata, to);
3484
3485                         /* if there was a short read -- discard anything left */
3486                         if (rdata->got_bytes && rdata->got_bytes < rdata->bytes)
3487                                 rc = -ENODATA;
3488
3489                         ctx->total_len += rdata->got_bytes;
3490                 }
3491                 list_del_init(&rdata->list);
3492                 kref_put(&rdata->refcount, cifs_uncached_readdata_release);
3493         }
3494
3495         if (!ctx->direct_io) {
3496                 for (i = 0; i < ctx->npages; i++) {
3497                         if (ctx->should_dirty)
3498                                 set_page_dirty(ctx->bv[i].bv_page);
3499                         put_page(ctx->bv[i].bv_page);
3500                 }
3501
3502                 ctx->total_len = ctx->len - iov_iter_count(to);
3503         }
3504
3505         cifs_stats_bytes_read(tcon, ctx->total_len);
3506
3507         /* mask nodata case */
3508         if (rc == -ENODATA)
3509                 rc = 0;
3510
3511         ctx->rc = (rc == 0) ? ctx->total_len : rc;
3512
3513         mutex_unlock(&ctx->aio_mutex);
3514
3515         if (ctx->iocb && ctx->iocb->ki_complete)
3516                 ctx->iocb->ki_complete(ctx->iocb, ctx->rc, 0);
3517         else
3518                 complete(&ctx->done);
3519 }
3520
3521 static ssize_t __cifs_readv(
3522         struct kiocb *iocb, struct iov_iter *to, bool direct)
3523 {
3524         size_t len;
3525         struct file *file = iocb->ki_filp;
3526         struct cifs_sb_info *cifs_sb;
3527         struct cifsFileInfo *cfile;
3528         struct cifs_tcon *tcon;
3529         ssize_t rc, total_read = 0;
3530         loff_t offset = iocb->ki_pos;
3531         struct cifs_aio_ctx *ctx;
3532
3533         /*
3534          * iov_iter_get_pages_alloc() doesn't work with ITER_KVEC,
3535          * fall back to data copy read path
3536          * this could be improved by getting pages directly in ITER_KVEC
3537          */
3538         if (direct && to->type & ITER_KVEC) {
3539                 cifs_dbg(FYI, "use non-direct cifs_user_readv for kvec I/O\n");
3540                 direct = false;
3541         }
3542
3543         len = iov_iter_count(to);
3544         if (!len)
3545                 return 0;
3546
3547         cifs_sb = CIFS_FILE_SB(file);
3548         cfile = file->private_data;
3549         tcon = tlink_tcon(cfile->tlink);
3550
3551         if (!tcon->ses->server->ops->async_readv)
3552                 return -ENOSYS;
3553
3554         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3555                 cifs_dbg(FYI, "attempting read on write only file instance\n");
3556
3557         ctx = cifs_aio_ctx_alloc();
3558         if (!ctx)
3559                 return -ENOMEM;
3560
3561         ctx->cfile = cifsFileInfo_get(cfile);
3562
3563         if (!is_sync_kiocb(iocb))
3564                 ctx->iocb = iocb;
3565
3566         if (iter_is_iovec(to))
3567                 ctx->should_dirty = true;
3568
3569         if (direct) {
3570                 ctx->pos = offset;
3571                 ctx->direct_io = true;
3572                 ctx->iter = *to;
3573                 ctx->len = len;
3574         } else {
3575                 rc = setup_aio_ctx_iter(ctx, to, READ);
3576                 if (rc) {
3577                         kref_put(&ctx->refcount, cifs_aio_ctx_release);
3578                         return rc;
3579                 }
3580                 len = ctx->len;
3581         }
3582
3583         /* grab a lock here due to read response handlers can access ctx */
3584         mutex_lock(&ctx->aio_mutex);
3585
3586         rc = cifs_send_async_read(offset, len, cfile, cifs_sb, &ctx->list, ctx);
3587
3588         /* if at least one read request send succeeded, then reset rc */
3589         if (!list_empty(&ctx->list))
3590                 rc = 0;
3591
3592         mutex_unlock(&ctx->aio_mutex);
3593
3594         if (rc) {
3595                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
3596                 return rc;
3597         }
3598
3599         if (!is_sync_kiocb(iocb)) {
3600                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
3601                 return -EIOCBQUEUED;
3602         }
3603
3604         rc = wait_for_completion_killable(&ctx->done);
3605         if (rc) {
3606                 mutex_lock(&ctx->aio_mutex);
3607                 ctx->rc = rc = -EINTR;
3608                 total_read = ctx->total_len;
3609                 mutex_unlock(&ctx->aio_mutex);
3610         } else {
3611                 rc = ctx->rc;
3612                 total_read = ctx->total_len;
3613         }
3614
3615         kref_put(&ctx->refcount, cifs_aio_ctx_release);
3616
3617         if (total_read) {
3618                 iocb->ki_pos += total_read;
3619                 return total_read;
3620         }
3621         return rc;
3622 }
3623
3624 ssize_t cifs_direct_readv(struct kiocb *iocb, struct iov_iter *to)
3625 {
3626         return __cifs_readv(iocb, to, true);
3627 }
3628
3629 ssize_t cifs_user_readv(struct kiocb *iocb, struct iov_iter *to)
3630 {
3631         return __cifs_readv(iocb, to, false);
3632 }
3633
3634 ssize_t
3635 cifs_strict_readv(struct kiocb *iocb, struct iov_iter *to)
3636 {
3637         struct inode *inode = file_inode(iocb->ki_filp);
3638         struct cifsInodeInfo *cinode = CIFS_I(inode);
3639         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
3640         struct cifsFileInfo *cfile = (struct cifsFileInfo *)
3641                                                 iocb->ki_filp->private_data;
3642         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
3643         int rc = -EACCES;
3644
3645         /*
3646          * In strict cache mode we need to read from the server all the time
3647          * if we don't have level II oplock because the server can delay mtime
3648          * change - so we can't make a decision about inode invalidating.
3649          * And we can also fail with pagereading if there are mandatory locks
3650          * on pages affected by this read but not on the region from pos to
3651          * pos+len-1.
3652          */
3653         if (!CIFS_CACHE_READ(cinode))
3654                 return cifs_user_readv(iocb, to);
3655
3656         if (cap_unix(tcon->ses) &&
3657             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
3658             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
3659                 return generic_file_read_iter(iocb, to);
3660
3661         /*
3662          * We need to hold the sem to be sure nobody modifies lock list
3663          * with a brlock that prevents reading.
3664          */
3665         down_read(&cinode->lock_sem);
3666         if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(to),
3667                                      tcon->ses->server->vals->shared_lock_type,
3668                                      0, NULL, CIFS_READ_OP))
3669                 rc = generic_file_read_iter(iocb, to);
3670         up_read(&cinode->lock_sem);
3671         return rc;
3672 }
3673
3674 static ssize_t
3675 cifs_read(struct file *file, char *read_data, size_t read_size, loff_t *offset)
3676 {
3677         int rc = -EACCES;
3678         unsigned int bytes_read = 0;
3679         unsigned int total_read;
3680         unsigned int current_read_size;
3681         unsigned int rsize;
3682         struct cifs_sb_info *cifs_sb;
3683         struct cifs_tcon *tcon;
3684         struct TCP_Server_Info *server;
3685         unsigned int xid;
3686         char *cur_offset;
3687         struct cifsFileInfo *open_file;
3688         struct cifs_io_parms io_parms;
3689         int buf_type = CIFS_NO_BUFFER;
3690         __u32 pid;
3691
3692         xid = get_xid();
3693         cifs_sb = CIFS_FILE_SB(file);
3694
3695         /* FIXME: set up handlers for larger reads and/or convert to async */
3696         rsize = min_t(unsigned int, cifs_sb->rsize, CIFSMaxBufSize);
3697
3698         if (file->private_data == NULL) {
3699                 rc = -EBADF;
3700                 free_xid(xid);
3701                 return rc;
3702         }
3703         open_file = file->private_data;
3704         tcon = tlink_tcon(open_file->tlink);
3705         server = tcon->ses->server;
3706
3707         if (!server->ops->sync_read) {
3708                 free_xid(xid);
3709                 return -ENOSYS;
3710         }
3711
3712         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3713                 pid = open_file->pid;
3714         else
3715                 pid = current->tgid;
3716
3717         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3718                 cifs_dbg(FYI, "attempting read on write only file instance\n");
3719
3720         for (total_read = 0, cur_offset = read_data; read_size > total_read;
3721              total_read += bytes_read, cur_offset += bytes_read) {
3722                 do {
3723                         current_read_size = min_t(uint, read_size - total_read,
3724                                                   rsize);
3725                         /*
3726                          * For windows me and 9x we do not want to request more
3727                          * than it negotiated since it will refuse the read
3728                          * then.
3729                          */
3730                         if ((tcon->ses) && !(tcon->ses->capabilities &
3731                                 tcon->ses->server->vals->cap_large_files)) {
3732                                 current_read_size = min_t(uint,
3733                                         current_read_size, CIFSMaxBufSize);
3734                         }
3735                         if (open_file->invalidHandle) {
3736                                 rc = cifs_reopen_file(open_file, true);
3737                                 if (rc != 0)
3738                                         break;
3739                         }
3740                         io_parms.pid = pid;
3741                         io_parms.tcon = tcon;
3742                         io_parms.offset = *offset;
3743                         io_parms.length = current_read_size;
3744                         rc = server->ops->sync_read(xid, &open_file->fid, &io_parms,
3745                                                     &bytes_read, &cur_offset,
3746                                                     &buf_type);
3747                 } while (rc == -EAGAIN);
3748
3749                 if (rc || (bytes_read == 0)) {
3750                         if (total_read) {
3751                                 break;
3752                         } else {
3753                                 free_xid(xid);
3754                                 return rc;
3755                         }
3756                 } else {
3757                         cifs_stats_bytes_read(tcon, total_read);
3758                         *offset += bytes_read;
3759                 }
3760         }
3761         free_xid(xid);
3762         return total_read;
3763 }
3764
3765 /*
3766  * If the page is mmap'ed into a process' page tables, then we need to make
3767  * sure that it doesn't change while being written back.
3768  */
3769 static vm_fault_t
3770 cifs_page_mkwrite(struct vm_fault *vmf)
3771 {
3772         struct page *page = vmf->page;
3773
3774         lock_page(page);
3775         return VM_FAULT_LOCKED;
3776 }
3777
3778 static const struct vm_operations_struct cifs_file_vm_ops = {
3779         .fault = filemap_fault,
3780         .map_pages = filemap_map_pages,
3781         .page_mkwrite = cifs_page_mkwrite,
3782 };
3783
3784 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
3785 {
3786         int xid, rc = 0;
3787         struct inode *inode = file_inode(file);
3788
3789         xid = get_xid();
3790
3791         if (!CIFS_CACHE_READ(CIFS_I(inode)))
3792                 rc = cifs_zap_mapping(inode);
3793         if (!rc)
3794                 rc = generic_file_mmap(file, vma);
3795         if (!rc)
3796                 vma->vm_ops = &cifs_file_vm_ops;
3797
3798         free_xid(xid);
3799         return rc;
3800 }
3801
3802 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
3803 {
3804         int rc, xid;
3805
3806         xid = get_xid();
3807
3808         rc = cifs_revalidate_file(file);
3809         if (rc)
3810                 cifs_dbg(FYI, "Validation prior to mmap failed, error=%d\n",
3811                          rc);
3812         if (!rc)
3813                 rc = generic_file_mmap(file, vma);
3814         if (!rc)
3815                 vma->vm_ops = &cifs_file_vm_ops;
3816
3817         free_xid(xid);
3818         return rc;
3819 }
3820
3821 static void
3822 cifs_readv_complete(struct work_struct *work)
3823 {
3824         unsigned int i, got_bytes;
3825         struct cifs_readdata *rdata = container_of(work,
3826                                                 struct cifs_readdata, work);
3827
3828         got_bytes = rdata->got_bytes;
3829         for (i = 0; i < rdata->nr_pages; i++) {
3830                 struct page *page = rdata->pages[i];
3831
3832                 lru_cache_add_file(page);
3833
3834                 if (rdata->result == 0 ||
3835                     (rdata->result == -EAGAIN && got_bytes)) {
3836                         flush_dcache_page(page);
3837                         SetPageUptodate(page);
3838                 }
3839
3840                 unlock_page(page);
3841
3842                 if (rdata->result == 0 ||
3843                     (rdata->result == -EAGAIN && got_bytes))
3844                         cifs_readpage_to_fscache(rdata->mapping->host, page);
3845
3846                 got_bytes -= min_t(unsigned int, PAGE_SIZE, got_bytes);
3847
3848                 put_page(page);
3849                 rdata->pages[i] = NULL;
3850         }
3851         kref_put(&rdata->refcount, cifs_readdata_release);
3852 }
3853
3854 static int
3855 readpages_fill_pages(struct TCP_Server_Info *server,
3856                      struct cifs_readdata *rdata, struct iov_iter *iter,
3857                      unsigned int len)
3858 {
3859         int result = 0;
3860         unsigned int i;
3861         u64 eof;
3862         pgoff_t eof_index;
3863         unsigned int nr_pages = rdata->nr_pages;
3864         unsigned int page_offset = rdata->page_offset;
3865
3866         /* determine the eof that the server (probably) has */
3867         eof = CIFS_I(rdata->mapping->host)->server_eof;
3868         eof_index = eof ? (eof - 1) >> PAGE_SHIFT : 0;
3869         cifs_dbg(FYI, "eof=%llu eof_index=%lu\n", eof, eof_index);
3870
3871         rdata->got_bytes = 0;
3872         rdata->tailsz = PAGE_SIZE;
3873         for (i = 0; i < nr_pages; i++) {
3874                 struct page *page = rdata->pages[i];
3875                 unsigned int to_read = rdata->pagesz;
3876                 size_t n;
3877
3878                 if (i == 0)
3879                         to_read -= page_offset;
3880                 else
3881                         page_offset = 0;
3882
3883                 n = to_read;
3884
3885                 if (len >= to_read) {
3886                         len -= to_read;
3887                 } else if (len > 0) {
3888                         /* enough for partial page, fill and zero the rest */
3889                         zero_user(page, len + page_offset, to_read - len);
3890                         n = rdata->tailsz = len;
3891                         len = 0;
3892                 } else if (page->index > eof_index) {
3893                         /*
3894                          * The VFS will not try to do readahead past the
3895                          * i_size, but it's possible that we have outstanding
3896                          * writes with gaps in the middle and the i_size hasn't
3897                          * caught up yet. Populate those with zeroed out pages
3898                          * to prevent the VFS from repeatedly attempting to
3899                          * fill them until the writes are flushed.
3900                          */
3901                         zero_user(page, 0, PAGE_SIZE);
3902                         lru_cache_add_file(page);
3903                         flush_dcache_page(page);
3904                         SetPageUptodate(page);
3905                         unlock_page(page);
3906                         put_page(page);
3907                         rdata->pages[i] = NULL;
3908                         rdata->nr_pages--;
3909                         continue;
3910                 } else {
3911                         /* no need to hold page hostage */
3912                         lru_cache_add_file(page);
3913                         unlock_page(page);
3914                         put_page(page);
3915                         rdata->pages[i] = NULL;
3916                         rdata->nr_pages--;
3917                         continue;
3918                 }
3919
3920                 if (iter)
3921                         result = copy_page_from_iter(
3922                                         page, page_offset, n, iter);
3923 #ifdef CONFIG_CIFS_SMB_DIRECT
3924                 else if (rdata->mr)
3925                         result = n;
3926 #endif
3927                 else
3928                         result = cifs_read_page_from_socket(
3929                                         server, page, page_offset, n);
3930                 if (result < 0)
3931                         break;
3932
3933                 rdata->got_bytes += result;
3934         }
3935
3936         return rdata->got_bytes > 0 && result != -ECONNABORTED ?
3937                                                 rdata->got_bytes : result;
3938 }
3939
3940 static int
3941 cifs_readpages_read_into_pages(struct TCP_Server_Info *server,
3942                                struct cifs_readdata *rdata, unsigned int len)
3943 {
3944         return readpages_fill_pages(server, rdata, NULL, len);
3945 }
3946
3947 static int
3948 cifs_readpages_copy_into_pages(struct TCP_Server_Info *server,
3949                                struct cifs_readdata *rdata,
3950                                struct iov_iter *iter)
3951 {
3952         return readpages_fill_pages(server, rdata, iter, iter->count);
3953 }
3954
3955 static int
3956 readpages_get_pages(struct address_space *mapping, struct list_head *page_list,
3957                     unsigned int rsize, struct list_head *tmplist,
3958                     unsigned int *nr_pages, loff_t *offset, unsigned int *bytes)
3959 {
3960         struct page *page, *tpage;
3961         unsigned int expected_index;
3962         int rc;
3963         gfp_t gfp = readahead_gfp_mask(mapping);
3964
3965         INIT_LIST_HEAD(tmplist);
3966
3967         page = list_entry(page_list->prev, struct page, lru);
3968
3969         /*
3970          * Lock the page and put it in the cache. Since no one else
3971          * should have access to this page, we're safe to simply set
3972          * PG_locked without checking it first.
3973          */
3974         __SetPageLocked(page);
3975         rc = add_to_page_cache_locked(page, mapping,
3976                                       page->index, gfp);
3977
3978         /* give up if we can't stick it in the cache */
3979         if (rc) {
3980                 __ClearPageLocked(page);
3981                 return rc;
3982         }
3983
3984         /* move first page to the tmplist */
3985         *offset = (loff_t)page->index << PAGE_SHIFT;
3986         *bytes = PAGE_SIZE;
3987         *nr_pages = 1;
3988         list_move_tail(&page->lru, tmplist);
3989
3990         /* now try and add more pages onto the request */
3991         expected_index = page->index + 1;
3992         list_for_each_entry_safe_reverse(page, tpage, page_list, lru) {
3993                 /* discontinuity ? */
3994                 if (page->index != expected_index)
3995                         break;
3996
3997                 /* would this page push the read over the rsize? */
3998                 if (*bytes + PAGE_SIZE > rsize)
3999                         break;
4000
4001                 __SetPageLocked(page);
4002                 if (add_to_page_cache_locked(page, mapping, page->index, gfp)) {
4003                         __ClearPageLocked(page);
4004                         break;
4005                 }
4006                 list_move_tail(&page->lru, tmplist);
4007                 (*bytes) += PAGE_SIZE;
4008                 expected_index++;
4009                 (*nr_pages)++;
4010         }
4011         return rc;
4012 }
4013
4014 static int cifs_readpages(struct file *file, struct address_space *mapping,
4015         struct list_head *page_list, unsigned num_pages)
4016 {
4017         int rc;
4018         struct list_head tmplist;
4019         struct cifsFileInfo *open_file = file->private_data;
4020         struct cifs_sb_info *cifs_sb = CIFS_FILE_SB(file);
4021         struct TCP_Server_Info *server;
4022         pid_t pid;
4023         unsigned int xid;
4024
4025         xid = get_xid();
4026         /*
4027          * Reads as many pages as possible from fscache. Returns -ENOBUFS
4028          * immediately if the cookie is negative
4029          *
4030          * After this point, every page in the list might have PG_fscache set,
4031          * so we will need to clean that up off of every page we don't use.
4032          */
4033         rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
4034                                          &num_pages);
4035         if (rc == 0) {
4036                 free_xid(xid);
4037                 return rc;
4038         }
4039
4040         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
4041                 pid = open_file->pid;
4042         else
4043                 pid = current->tgid;
4044
4045         rc = 0;
4046         server = tlink_tcon(open_file->tlink)->ses->server;
4047
4048         cifs_dbg(FYI, "%s: file=%p mapping=%p num_pages=%u\n",
4049                  __func__, file, mapping, num_pages);
4050
4051         /*
4052          * Start with the page at end of list and move it to private
4053          * list. Do the same with any following pages until we hit
4054          * the rsize limit, hit an index discontinuity, or run out of
4055          * pages. Issue the async read and then start the loop again
4056          * until the list is empty.
4057          *
4058          * Note that list order is important. The page_list is in
4059          * the order of declining indexes. When we put the pages in
4060          * the rdata->pages, then we want them in increasing order.
4061          */
4062         while (!list_empty(page_list)) {
4063                 unsigned int i, nr_pages, bytes, rsize;
4064                 loff_t offset;
4065                 struct page *page, *tpage;
4066                 struct cifs_readdata *rdata;
4067                 unsigned credits;
4068
4069                 rc = server->ops->wait_mtu_credits(server, cifs_sb->rsize,
4070                                                    &rsize, &credits);
4071                 if (rc)
4072                         break;
4073
4074                 /*
4075                  * Give up immediately if rsize is too small to read an entire
4076                  * page. The VFS will fall back to readpage. We should never
4077                  * reach this point however since we set ra_pages to 0 when the
4078                  * rsize is smaller than a cache page.
4079                  */
4080                 if (unlikely(rsize < PAGE_SIZE)) {
4081                         add_credits_and_wake_if(server, credits, 0);
4082                         free_xid(xid);
4083                         return 0;
4084                 }
4085
4086                 rc = readpages_get_pages(mapping, page_list, rsize, &tmplist,
4087                                          &nr_pages, &offset, &bytes);
4088                 if (rc) {
4089                         add_credits_and_wake_if(server, credits, 0);
4090                         break;
4091                 }
4092
4093                 rdata = cifs_readdata_alloc(nr_pages, cifs_readv_complete);
4094                 if (!rdata) {
4095                         /* best to give up if we're out of mem */
4096                         list_for_each_entry_safe(page, tpage, &tmplist, lru) {
4097                                 list_del(&page->lru);
4098                                 lru_cache_add_file(page);
4099                                 unlock_page(page);
4100                                 put_page(page);
4101                         }
4102                         rc = -ENOMEM;
4103                         add_credits_and_wake_if(server, credits, 0);
4104                         break;
4105                 }
4106
4107                 rdata->cfile = cifsFileInfo_get(open_file);
4108                 rdata->mapping = mapping;
4109                 rdata->offset = offset;
4110                 rdata->bytes = bytes;
4111                 rdata->pid = pid;
4112                 rdata->pagesz = PAGE_SIZE;
4113                 rdata->tailsz = PAGE_SIZE;
4114                 rdata->read_into_pages = cifs_readpages_read_into_pages;
4115                 rdata->copy_into_pages = cifs_readpages_copy_into_pages;
4116                 rdata->credits = credits;
4117
4118                 list_for_each_entry_safe(page, tpage, &tmplist, lru) {
4119                         list_del(&page->lru);
4120                         rdata->pages[rdata->nr_pages++] = page;
4121                 }
4122
4123                 if (!rdata->cfile->invalidHandle ||
4124                     !(rc = cifs_reopen_file(rdata->cfile, true)))
4125                         rc = server->ops->async_readv(rdata);
4126                 if (rc) {
4127                         add_credits_and_wake_if(server, rdata->credits, 0);
4128                         for (i = 0; i < rdata->nr_pages; i++) {
4129                                 page = rdata->pages[i];
4130                                 lru_cache_add_file(page);
4131                                 unlock_page(page);
4132                                 put_page(page);
4133                         }
4134                         /* Fallback to the readpage in error/reconnect cases */
4135                         kref_put(&rdata->refcount, cifs_readdata_release);
4136                         break;
4137                 }
4138
4139                 kref_put(&rdata->refcount, cifs_readdata_release);
4140         }
4141
4142         /* Any pages that have been shown to fscache but didn't get added to
4143          * the pagecache must be uncached before they get returned to the
4144          * allocator.
4145          */
4146         cifs_fscache_readpages_cancel(mapping->host, page_list);
4147         free_xid(xid);
4148         return rc;
4149 }
4150
4151 /*
4152  * cifs_readpage_worker must be called with the page pinned
4153  */
4154 static int cifs_readpage_worker(struct file *file, struct page *page,
4155         loff_t *poffset)
4156 {
4157         char *read_data;
4158         int rc;
4159
4160         /* Is the page cached? */
4161         rc = cifs_readpage_from_fscache(file_inode(file), page);
4162         if (rc == 0)
4163                 goto read_complete;
4164
4165         read_data = kmap(page);
4166         /* for reads over a certain size could initiate async read ahead */
4167
4168         rc = cifs_read(file, read_data, PAGE_SIZE, poffset);
4169
4170         if (rc < 0)
4171                 goto io_error;
4172         else
4173                 cifs_dbg(FYI, "Bytes read %d\n", rc);
4174
4175         /* we do not want atime to be less than mtime, it broke some apps */
4176         file_inode(file)->i_atime = current_time(file_inode(file));
4177         if (timespec64_compare(&(file_inode(file)->i_atime), &(file_inode(file)->i_mtime)))
4178                 file_inode(file)->i_atime = file_inode(file)->i_mtime;
4179         else
4180                 file_inode(file)->i_atime = current_time(file_inode(file));
4181
4182         if (PAGE_SIZE > rc)
4183                 memset(read_data + rc, 0, PAGE_SIZE - rc);
4184
4185         flush_dcache_page(page);
4186         SetPageUptodate(page);
4187
4188         /* send this page to the cache */
4189         cifs_readpage_to_fscache(file_inode(file), page);
4190
4191         rc = 0;
4192
4193 io_error:
4194         kunmap(page);
4195         unlock_page(page);
4196
4197 read_complete:
4198         return rc;
4199 }
4200
4201 static int cifs_readpage(struct file *file, struct page *page)
4202 {
4203         loff_t offset = (loff_t)page->index << PAGE_SHIFT;
4204         int rc = -EACCES;
4205         unsigned int xid;
4206
4207         xid = get_xid();
4208
4209         if (file->private_data == NULL) {
4210                 rc = -EBADF;
4211                 free_xid(xid);
4212                 return rc;
4213         }
4214
4215         cifs_dbg(FYI, "readpage %p at offset %d 0x%x\n",
4216                  page, (int)offset, (int)offset);
4217
4218         rc = cifs_readpage_worker(file, page, &offset);
4219
4220         free_xid(xid);
4221         return rc;
4222 }
4223
4224 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
4225 {
4226         struct cifsFileInfo *open_file;
4227         struct cifs_tcon *tcon =
4228                 cifs_sb_master_tcon(CIFS_SB(cifs_inode->vfs_inode.i_sb));
4229
4230         spin_lock(&tcon->open_file_lock);
4231         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
4232                 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
4233                         spin_unlock(&tcon->open_file_lock);
4234                         return 1;
4235                 }
4236         }
4237         spin_unlock(&tcon->open_file_lock);
4238         return 0;
4239 }
4240
4241 /* We do not want to update the file size from server for inodes
4242    open for write - to avoid races with writepage extending
4243    the file - in the future we could consider allowing
4244    refreshing the inode only on increases in the file size
4245    but this is tricky to do without racing with writebehind
4246    page caching in the current Linux kernel design */
4247 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
4248 {
4249         if (!cifsInode)
4250                 return true;
4251
4252         if (is_inode_writable(cifsInode)) {
4253                 /* This inode is open for write at least once */
4254                 struct cifs_sb_info *cifs_sb;
4255
4256                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
4257                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
4258                         /* since no page cache to corrupt on directio
4259                         we can change size safely */
4260                         return true;
4261                 }
4262
4263                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
4264                         return true;
4265
4266                 return false;
4267         } else
4268                 return true;
4269 }
4270
4271 static int cifs_write_begin(struct file *file, struct address_space *mapping,
4272                         loff_t pos, unsigned len, unsigned flags,
4273                         struct page **pagep, void **fsdata)
4274 {
4275         int oncethru = 0;
4276         pgoff_t index = pos >> PAGE_SHIFT;
4277         loff_t offset = pos & (PAGE_SIZE - 1);
4278         loff_t page_start = pos & PAGE_MASK;
4279         loff_t i_size;
4280         struct page *page;
4281         int rc = 0;
4282
4283         cifs_dbg(FYI, "write_begin from %lld len %d\n", (long long)pos, len);
4284
4285 start:
4286         page = grab_cache_page_write_begin(mapping, index, flags);
4287         if (!page) {
4288                 rc = -ENOMEM;
4289                 goto out;
4290         }
4291
4292         if (PageUptodate(page))
4293                 goto out;
4294
4295         /*
4296          * If we write a full page it will be up to date, no need to read from
4297          * the server. If the write is short, we'll end up doing a sync write
4298          * instead.
4299          */
4300         if (len == PAGE_SIZE)
4301                 goto out;
4302
4303         /*
4304          * optimize away the read when we have an oplock, and we're not
4305          * expecting to use any of the data we'd be reading in. That
4306          * is, when the page lies beyond the EOF, or straddles the EOF
4307          * and the write will cover all of the existing data.
4308          */
4309         if (CIFS_CACHE_READ(CIFS_I(mapping->host))) {
4310                 i_size = i_size_read(mapping->host);
4311                 if (page_start >= i_size ||
4312                     (offset == 0 && (pos + len) >= i_size)) {
4313                         zero_user_segments(page, 0, offset,
4314                                            offset + len,
4315                                            PAGE_SIZE);
4316                         /*
4317                          * PageChecked means that the parts of the page
4318                          * to which we're not writing are considered up
4319                          * to date. Once the data is copied to the
4320                          * page, it can be set uptodate.
4321                          */
4322                         SetPageChecked(page);
4323                         goto out;
4324                 }
4325         }
4326
4327         if ((file->f_flags & O_ACCMODE) != O_WRONLY && !oncethru) {
4328                 /*
4329                  * might as well read a page, it is fast enough. If we get
4330                  * an error, we don't need to return it. cifs_write_end will
4331                  * do a sync write instead since PG_uptodate isn't set.
4332                  */
4333                 cifs_readpage_worker(file, page, &page_start);
4334                 put_page(page);
4335                 oncethru = 1;
4336                 goto start;
4337         } else {
4338                 /* we could try using another file handle if there is one -
4339                    but how would we lock it to prevent close of that handle
4340                    racing with this read? In any case
4341                    this will be written out by write_end so is fine */
4342         }
4343 out:
4344         *pagep = page;
4345         return rc;
4346 }
4347
4348 static int cifs_release_page(struct page *page, gfp_t gfp)
4349 {
4350         if (PagePrivate(page))
4351                 return 0;
4352
4353         return cifs_fscache_release_page(page, gfp);
4354 }
4355
4356 static void cifs_invalidate_page(struct page *page, unsigned int offset,
4357                                  unsigned int length)
4358 {
4359         struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
4360
4361         if (offset == 0 && length == PAGE_SIZE)
4362                 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
4363 }
4364
4365 static int cifs_launder_page(struct page *page)
4366 {
4367         int rc = 0;
4368         loff_t range_start = page_offset(page);
4369         loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1);
4370         struct writeback_control wbc = {
4371                 .sync_mode = WB_SYNC_ALL,
4372                 .nr_to_write = 0,
4373                 .range_start = range_start,
4374                 .range_end = range_end,
4375         };
4376
4377         cifs_dbg(FYI, "Launder page: %p\n", page);
4378
4379         if (clear_page_dirty_for_io(page))
4380                 rc = cifs_writepage_locked(page, &wbc);
4381
4382         cifs_fscache_invalidate_page(page, page->mapping->host);
4383         return rc;
4384 }
4385
4386 void cifs_oplock_break(struct work_struct *work)
4387 {
4388         struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
4389                                                   oplock_break);
4390         struct inode *inode = d_inode(cfile->dentry);
4391         struct cifsInodeInfo *cinode = CIFS_I(inode);
4392         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
4393         struct TCP_Server_Info *server = tcon->ses->server;
4394         int rc = 0;
4395
4396         wait_on_bit(&cinode->flags, CIFS_INODE_PENDING_WRITERS,
4397                         TASK_UNINTERRUPTIBLE);
4398
4399         server->ops->downgrade_oplock(server, cinode,
4400                 test_bit(CIFS_INODE_DOWNGRADE_OPLOCK_TO_L2, &cinode->flags));
4401
4402         if (!CIFS_CACHE_WRITE(cinode) && CIFS_CACHE_READ(cinode) &&
4403                                                 cifs_has_mand_locks(cinode)) {
4404                 cifs_dbg(FYI, "Reset oplock to None for inode=%p due to mand locks\n",
4405                          inode);
4406                 cinode->oplock = 0;
4407         }
4408
4409         if (inode && S_ISREG(inode->i_mode)) {
4410                 if (CIFS_CACHE_READ(cinode))
4411                         break_lease(inode, O_RDONLY);
4412                 else
4413                         break_lease(inode, O_WRONLY);
4414                 rc = filemap_fdatawrite(inode->i_mapping);
4415                 if (!CIFS_CACHE_READ(cinode)) {
4416                         rc = filemap_fdatawait(inode->i_mapping);
4417                         mapping_set_error(inode->i_mapping, rc);
4418                         cifs_zap_mapping(inode);
4419                 }
4420                 cifs_dbg(FYI, "Oplock flush inode %p rc %d\n", inode, rc);
4421         }
4422
4423         rc = cifs_push_locks(cfile);
4424         if (rc)
4425                 cifs_dbg(VFS, "Push locks rc = %d\n", rc);
4426
4427         /*
4428          * releasing stale oplock after recent reconnect of smb session using
4429          * a now incorrect file handle is not a data integrity issue but do
4430          * not bother sending an oplock release if session to server still is
4431          * disconnected since oplock already released by the server
4432          */
4433         if (!cfile->oplock_break_cancelled) {
4434                 rc = tcon->ses->server->ops->oplock_response(tcon, &cfile->fid,
4435                                                              cinode);
4436                 cifs_dbg(FYI, "Oplock release rc = %d\n", rc);
4437         }
4438         cifs_done_oplock_break(cinode);
4439 }
4440
4441 /*
4442  * The presence of cifs_direct_io() in the address space ops vector
4443  * allowes open() O_DIRECT flags which would have failed otherwise.
4444  *
4445  * In the non-cached mode (mount with cache=none), we shunt off direct read and write requests
4446  * so this method should never be called.
4447  *
4448  * Direct IO is not yet supported in the cached mode.
4449  */
4450 static ssize_t
4451 cifs_direct_io(struct kiocb *iocb, struct iov_iter *iter)
4452 {
4453         /*
4454          * FIXME
4455          * Eventually need to support direct IO for non forcedirectio mounts
4456          */
4457         return -EINVAL;
4458 }
4459
4460
4461 const struct address_space_operations cifs_addr_ops = {
4462         .readpage = cifs_readpage,
4463         .readpages = cifs_readpages,
4464         .writepage = cifs_writepage,
4465         .writepages = cifs_writepages,
4466         .write_begin = cifs_write_begin,
4467         .write_end = cifs_write_end,
4468         .set_page_dirty = __set_page_dirty_nobuffers,
4469         .releasepage = cifs_release_page,
4470         .direct_IO = cifs_direct_io,
4471         .invalidatepage = cifs_invalidate_page,
4472         .launder_page = cifs_launder_page,
4473 };
4474
4475 /*
4476  * cifs_readpages requires the server to support a buffer large enough to
4477  * contain the header plus one complete page of data.  Otherwise, we need
4478  * to leave cifs_readpages out of the address space operations.
4479  */
4480 const struct address_space_operations cifs_addr_ops_smallbuf = {
4481         .readpage = cifs_readpage,
4482         .writepage = cifs_writepage,
4483         .writepages = cifs_writepages,
4484         .write_begin = cifs_write_begin,
4485         .write_end = cifs_write_end,
4486         .set_page_dirty = __set_page_dirty_nobuffers,
4487         .releasepage = cifs_release_page,
4488         .invalidatepage = cifs_invalidate_page,
4489         .launder_page = cifs_launder_page,
4490 };