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_next);
1107                 if (!rc)
1108                         goto try_again;
1109                 posix_unblock_lock(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                 if (ctx->direct_io) {
2621                         ssize_t result;
2622
2623                         result = iov_iter_get_pages_alloc(
2624                                 from, &pagevec, wsize, &start);
2625                         if (result < 0) {
2626                                 cifs_dbg(VFS,
2627                                         "direct_writev couldn't get user pages "
2628                                         "(rc=%zd) iter type %d iov_offset %zd "
2629                                         "count %zd\n",
2630                                         result, from->type,
2631                                         from->iov_offset, from->count);
2632                                 dump_stack();
2633                                 break;
2634                         }
2635                         cur_len = (size_t)result;
2636                         iov_iter_advance(from, cur_len);
2637
2638                         nr_pages =
2639                                 (cur_len + start + PAGE_SIZE - 1) / PAGE_SIZE;
2640
2641                         wdata = cifs_writedata_direct_alloc(pagevec,
2642                                              cifs_uncached_writev_complete);
2643                         if (!wdata) {
2644                                 rc = -ENOMEM;
2645                                 add_credits_and_wake_if(server, credits, 0);
2646                                 break;
2647                         }
2648
2649
2650                         wdata->page_offset = start;
2651                         wdata->tailsz =
2652                                 nr_pages > 1 ?
2653                                         cur_len - (PAGE_SIZE - start) -
2654                                         (nr_pages - 2) * PAGE_SIZE :
2655                                         cur_len;
2656                 } else {
2657                         nr_pages = get_numpages(wsize, len, &cur_len);
2658                         wdata = cifs_writedata_alloc(nr_pages,
2659                                              cifs_uncached_writev_complete);
2660                         if (!wdata) {
2661                                 rc = -ENOMEM;
2662                                 add_credits_and_wake_if(server, credits, 0);
2663                                 break;
2664                         }
2665
2666                         rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
2667                         if (rc) {
2668                                 kfree(wdata);
2669                                 add_credits_and_wake_if(server, credits, 0);
2670                                 break;
2671                         }
2672
2673                         num_pages = nr_pages;
2674                         rc = wdata_fill_from_iovec(
2675                                 wdata, from, &cur_len, &num_pages);
2676                         if (rc) {
2677                                 for (i = 0; i < nr_pages; i++)
2678                                         put_page(wdata->pages[i]);
2679                                 kfree(wdata);
2680                                 add_credits_and_wake_if(server, credits, 0);
2681                                 break;
2682                         }
2683
2684                         /*
2685                          * Bring nr_pages down to the number of pages we
2686                          * actually used, and free any pages that we didn't use.
2687                          */
2688                         for ( ; nr_pages > num_pages; nr_pages--)
2689                                 put_page(wdata->pages[nr_pages - 1]);
2690
2691                         wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
2692                 }
2693
2694                 wdata->sync_mode = WB_SYNC_ALL;
2695                 wdata->nr_pages = nr_pages;
2696                 wdata->offset = (__u64)offset;
2697                 wdata->cfile = cifsFileInfo_get(open_file);
2698                 wdata->pid = pid;
2699                 wdata->bytes = cur_len;
2700                 wdata->pagesz = PAGE_SIZE;
2701                 wdata->credits = credits;
2702                 wdata->ctx = ctx;
2703                 kref_get(&ctx->refcount);
2704
2705                 if (!wdata->cfile->invalidHandle ||
2706                     !(rc = cifs_reopen_file(wdata->cfile, false)))
2707                         rc = server->ops->async_writev(wdata,
2708                                         cifs_uncached_writedata_release);
2709                 if (rc) {
2710                         add_credits_and_wake_if(server, wdata->credits, 0);
2711                         kref_put(&wdata->refcount,
2712                                  cifs_uncached_writedata_release);
2713                         if (rc == -EAGAIN) {
2714                                 *from = saved_from;
2715                                 iov_iter_advance(from, offset - saved_offset);
2716                                 continue;
2717                         }
2718                         break;
2719                 }
2720
2721                 list_add_tail(&wdata->list, wdata_list);
2722                 offset += cur_len;
2723                 len -= cur_len;
2724         } while (len > 0);
2725
2726         return rc;
2727 }
2728
2729 static void collect_uncached_write_data(struct cifs_aio_ctx *ctx)
2730 {
2731         struct cifs_writedata *wdata, *tmp;
2732         struct cifs_tcon *tcon;
2733         struct cifs_sb_info *cifs_sb;
2734         struct dentry *dentry = ctx->cfile->dentry;
2735         unsigned int i;
2736         int rc;
2737
2738         tcon = tlink_tcon(ctx->cfile->tlink);
2739         cifs_sb = CIFS_SB(dentry->d_sb);
2740
2741         mutex_lock(&ctx->aio_mutex);
2742
2743         if (list_empty(&ctx->list)) {
2744                 mutex_unlock(&ctx->aio_mutex);
2745                 return;
2746         }
2747
2748         rc = ctx->rc;
2749         /*
2750          * Wait for and collect replies for any successful sends in order of
2751          * increasing offset. Once an error is hit, then return without waiting
2752          * for any more replies.
2753          */
2754 restart_loop:
2755         list_for_each_entry_safe(wdata, tmp, &ctx->list, list) {
2756                 if (!rc) {
2757                         if (!try_wait_for_completion(&wdata->done)) {
2758                                 mutex_unlock(&ctx->aio_mutex);
2759                                 return;
2760                         }
2761
2762                         if (wdata->result)
2763                                 rc = wdata->result;
2764                         else
2765                                 ctx->total_len += wdata->bytes;
2766
2767                         /* resend call if it's a retryable error */
2768                         if (rc == -EAGAIN) {
2769                                 struct list_head tmp_list;
2770                                 struct iov_iter tmp_from = ctx->iter;
2771
2772                                 INIT_LIST_HEAD(&tmp_list);
2773                                 list_del_init(&wdata->list);
2774
2775                                 if (ctx->direct_io)
2776                                         rc = cifs_resend_wdata(
2777                                                 wdata, &tmp_list, ctx);
2778                                 else {
2779                                         iov_iter_advance(&tmp_from,
2780                                                  wdata->offset - ctx->pos);
2781
2782                                         rc = cifs_write_from_iter(wdata->offset,
2783                                                 wdata->bytes, &tmp_from,
2784                                                 ctx->cfile, cifs_sb, &tmp_list,
2785                                                 ctx);
2786                                 }
2787
2788                                 list_splice(&tmp_list, &ctx->list);
2789
2790                                 kref_put(&wdata->refcount,
2791                                          cifs_uncached_writedata_release);
2792                                 goto restart_loop;
2793                         }
2794                 }
2795                 list_del_init(&wdata->list);
2796                 kref_put(&wdata->refcount, cifs_uncached_writedata_release);
2797         }
2798
2799         if (!ctx->direct_io)
2800                 for (i = 0; i < ctx->npages; i++)
2801                         put_page(ctx->bv[i].bv_page);
2802
2803         cifs_stats_bytes_written(tcon, ctx->total_len);
2804         set_bit(CIFS_INO_INVALID_MAPPING, &CIFS_I(dentry->d_inode)->flags);
2805
2806         ctx->rc = (rc == 0) ? ctx->total_len : rc;
2807
2808         mutex_unlock(&ctx->aio_mutex);
2809
2810         if (ctx->iocb && ctx->iocb->ki_complete)
2811                 ctx->iocb->ki_complete(ctx->iocb, ctx->rc, 0);
2812         else
2813                 complete(&ctx->done);
2814 }
2815
2816 static ssize_t __cifs_writev(
2817         struct kiocb *iocb, struct iov_iter *from, bool direct)
2818 {
2819         struct file *file = iocb->ki_filp;
2820         ssize_t total_written = 0;
2821         struct cifsFileInfo *cfile;
2822         struct cifs_tcon *tcon;
2823         struct cifs_sb_info *cifs_sb;
2824         struct cifs_aio_ctx *ctx;
2825         struct iov_iter saved_from = *from;
2826         size_t len = iov_iter_count(from);
2827         int rc;
2828
2829         /*
2830          * iov_iter_get_pages_alloc doesn't work with ITER_KVEC.
2831          * In this case, fall back to non-direct write function.
2832          * this could be improved by getting pages directly in ITER_KVEC
2833          */
2834         if (direct && from->type & ITER_KVEC) {
2835                 cifs_dbg(FYI, "use non-direct cifs_writev for kvec I/O\n");
2836                 direct = false;
2837         }
2838
2839         rc = generic_write_checks(iocb, from);
2840         if (rc <= 0)
2841                 return rc;
2842
2843         cifs_sb = CIFS_FILE_SB(file);
2844         cfile = file->private_data;
2845         tcon = tlink_tcon(cfile->tlink);
2846
2847         if (!tcon->ses->server->ops->async_writev)
2848                 return -ENOSYS;
2849
2850         ctx = cifs_aio_ctx_alloc();
2851         if (!ctx)
2852                 return -ENOMEM;
2853
2854         ctx->cfile = cifsFileInfo_get(cfile);
2855
2856         if (!is_sync_kiocb(iocb))
2857                 ctx->iocb = iocb;
2858
2859         ctx->pos = iocb->ki_pos;
2860
2861         if (direct) {
2862                 ctx->direct_io = true;
2863                 ctx->iter = *from;
2864                 ctx->len = len;
2865         } else {
2866                 rc = setup_aio_ctx_iter(ctx, from, WRITE);
2867                 if (rc) {
2868                         kref_put(&ctx->refcount, cifs_aio_ctx_release);
2869                         return rc;
2870                 }
2871         }
2872
2873         /* grab a lock here due to read response handlers can access ctx */
2874         mutex_lock(&ctx->aio_mutex);
2875
2876         rc = cifs_write_from_iter(iocb->ki_pos, ctx->len, &saved_from,
2877                                   cfile, cifs_sb, &ctx->list, ctx);
2878
2879         /*
2880          * If at least one write was successfully sent, then discard any rc
2881          * value from the later writes. If the other write succeeds, then
2882          * we'll end up returning whatever was written. If it fails, then
2883          * we'll get a new rc value from that.
2884          */
2885         if (!list_empty(&ctx->list))
2886                 rc = 0;
2887
2888         mutex_unlock(&ctx->aio_mutex);
2889
2890         if (rc) {
2891                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
2892                 return rc;
2893         }
2894
2895         if (!is_sync_kiocb(iocb)) {
2896                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
2897                 return -EIOCBQUEUED;
2898         }
2899
2900         rc = wait_for_completion_killable(&ctx->done);
2901         if (rc) {
2902                 mutex_lock(&ctx->aio_mutex);
2903                 ctx->rc = rc = -EINTR;
2904                 total_written = ctx->total_len;
2905                 mutex_unlock(&ctx->aio_mutex);
2906         } else {
2907                 rc = ctx->rc;
2908                 total_written = ctx->total_len;
2909         }
2910
2911         kref_put(&ctx->refcount, cifs_aio_ctx_release);
2912
2913         if (unlikely(!total_written))
2914                 return rc;
2915
2916         iocb->ki_pos += total_written;
2917         return total_written;
2918 }
2919
2920 ssize_t cifs_direct_writev(struct kiocb *iocb, struct iov_iter *from)
2921 {
2922         return __cifs_writev(iocb, from, true);
2923 }
2924
2925 ssize_t cifs_user_writev(struct kiocb *iocb, struct iov_iter *from)
2926 {
2927         return __cifs_writev(iocb, from, false);
2928 }
2929
2930 static ssize_t
2931 cifs_writev(struct kiocb *iocb, struct iov_iter *from)
2932 {
2933         struct file *file = iocb->ki_filp;
2934         struct cifsFileInfo *cfile = (struct cifsFileInfo *)file->private_data;
2935         struct inode *inode = file->f_mapping->host;
2936         struct cifsInodeInfo *cinode = CIFS_I(inode);
2937         struct TCP_Server_Info *server = tlink_tcon(cfile->tlink)->ses->server;
2938         ssize_t rc;
2939
2940         inode_lock(inode);
2941         /*
2942          * We need to hold the sem to be sure nobody modifies lock list
2943          * with a brlock that prevents writing.
2944          */
2945         down_read(&cinode->lock_sem);
2946
2947         rc = generic_write_checks(iocb, from);
2948         if (rc <= 0)
2949                 goto out;
2950
2951         if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(from),
2952                                      server->vals->exclusive_lock_type, 0,
2953                                      NULL, CIFS_WRITE_OP))
2954                 rc = __generic_file_write_iter(iocb, from);
2955         else
2956                 rc = -EACCES;
2957 out:
2958         up_read(&cinode->lock_sem);
2959         inode_unlock(inode);
2960
2961         if (rc > 0)
2962                 rc = generic_write_sync(iocb, rc);
2963         return rc;
2964 }
2965
2966 ssize_t
2967 cifs_strict_writev(struct kiocb *iocb, struct iov_iter *from)
2968 {
2969         struct inode *inode = file_inode(iocb->ki_filp);
2970         struct cifsInodeInfo *cinode = CIFS_I(inode);
2971         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
2972         struct cifsFileInfo *cfile = (struct cifsFileInfo *)
2973                                                 iocb->ki_filp->private_data;
2974         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
2975         ssize_t written;
2976
2977         written = cifs_get_writer(cinode);
2978         if (written)
2979                 return written;
2980
2981         if (CIFS_CACHE_WRITE(cinode)) {
2982                 if (cap_unix(tcon->ses) &&
2983                 (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability))
2984                   && ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0)) {
2985                         written = generic_file_write_iter(iocb, from);
2986                         goto out;
2987                 }
2988                 written = cifs_writev(iocb, from);
2989                 goto out;
2990         }
2991         /*
2992          * For non-oplocked files in strict cache mode we need to write the data
2993          * to the server exactly from the pos to pos+len-1 rather than flush all
2994          * affected pages because it may cause a error with mandatory locks on
2995          * these pages but not on the region from pos to ppos+len-1.
2996          */
2997         written = cifs_user_writev(iocb, from);
2998         if (written > 0 && CIFS_CACHE_READ(cinode)) {
2999                 /*
3000                  * Windows 7 server can delay breaking level2 oplock if a write
3001                  * request comes - break it on the client to prevent reading
3002                  * an old data.
3003                  */
3004                 cifs_zap_mapping(inode);
3005                 cifs_dbg(FYI, "Set no oplock for inode=%p after a write operation\n",
3006                          inode);
3007                 cinode->oplock = 0;
3008         }
3009 out:
3010         cifs_put_writer(cinode);
3011         return written;
3012 }
3013
3014 static struct cifs_readdata *
3015 cifs_readdata_direct_alloc(struct page **pages, work_func_t complete)
3016 {
3017         struct cifs_readdata *rdata;
3018
3019         rdata = kzalloc(sizeof(*rdata), GFP_KERNEL);
3020         if (rdata != NULL) {
3021                 rdata->pages = pages;
3022                 kref_init(&rdata->refcount);
3023                 INIT_LIST_HEAD(&rdata->list);
3024                 init_completion(&rdata->done);
3025                 INIT_WORK(&rdata->work, complete);
3026         }
3027
3028         return rdata;
3029 }
3030
3031 static struct cifs_readdata *
3032 cifs_readdata_alloc(unsigned int nr_pages, work_func_t complete)
3033 {
3034         struct page **pages =
3035                 kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
3036         struct cifs_readdata *ret = NULL;
3037
3038         if (pages) {
3039                 ret = cifs_readdata_direct_alloc(pages, complete);
3040                 if (!ret)
3041                         kfree(pages);
3042         }
3043
3044         return ret;
3045 }
3046
3047 void
3048 cifs_readdata_release(struct kref *refcount)
3049 {
3050         struct cifs_readdata *rdata = container_of(refcount,
3051                                         struct cifs_readdata, refcount);
3052 #ifdef CONFIG_CIFS_SMB_DIRECT
3053         if (rdata->mr) {
3054                 smbd_deregister_mr(rdata->mr);
3055                 rdata->mr = NULL;
3056         }
3057 #endif
3058         if (rdata->cfile)
3059                 cifsFileInfo_put(rdata->cfile);
3060
3061         kvfree(rdata->pages);
3062         kfree(rdata);
3063 }
3064
3065 static int
3066 cifs_read_allocate_pages(struct cifs_readdata *rdata, unsigned int nr_pages)
3067 {
3068         int rc = 0;
3069         struct page *page;
3070         unsigned int i;
3071
3072         for (i = 0; i < nr_pages; i++) {
3073                 page = alloc_page(GFP_KERNEL|__GFP_HIGHMEM);
3074                 if (!page) {
3075                         rc = -ENOMEM;
3076                         break;
3077                 }
3078                 rdata->pages[i] = page;
3079         }
3080
3081         if (rc) {
3082                 for (i = 0; i < nr_pages; i++) {
3083                         put_page(rdata->pages[i]);
3084                         rdata->pages[i] = NULL;
3085                 }
3086         }
3087         return rc;
3088 }
3089
3090 static void
3091 cifs_uncached_readdata_release(struct kref *refcount)
3092 {
3093         struct cifs_readdata *rdata = container_of(refcount,
3094                                         struct cifs_readdata, refcount);
3095         unsigned int i;
3096
3097         kref_put(&rdata->ctx->refcount, cifs_aio_ctx_release);
3098         for (i = 0; i < rdata->nr_pages; i++) {
3099                 put_page(rdata->pages[i]);
3100         }
3101         cifs_readdata_release(refcount);
3102 }
3103
3104 /**
3105  * cifs_readdata_to_iov - copy data from pages in response to an iovec
3106  * @rdata:      the readdata response with list of pages holding data
3107  * @iter:       destination for our data
3108  *
3109  * This function copies data from a list of pages in a readdata response into
3110  * an array of iovecs. It will first calculate where the data should go
3111  * based on the info in the readdata and then copy the data into that spot.
3112  */
3113 static int
3114 cifs_readdata_to_iov(struct cifs_readdata *rdata, struct iov_iter *iter)
3115 {
3116         size_t remaining = rdata->got_bytes;
3117         unsigned int i;
3118
3119         for (i = 0; i < rdata->nr_pages; i++) {
3120                 struct page *page = rdata->pages[i];
3121                 size_t copy = min_t(size_t, remaining, PAGE_SIZE);
3122                 size_t written;
3123
3124                 if (unlikely(iov_iter_is_pipe(iter))) {
3125                         void *addr = kmap_atomic(page);
3126
3127                         written = copy_to_iter(addr, copy, iter);
3128                         kunmap_atomic(addr);
3129                 } else
3130                         written = copy_page_to_iter(page, 0, copy, iter);
3131                 remaining -= written;
3132                 if (written < copy && iov_iter_count(iter) > 0)
3133                         break;
3134         }
3135         return remaining ? -EFAULT : 0;
3136 }
3137
3138 static void collect_uncached_read_data(struct cifs_aio_ctx *ctx);
3139
3140 static void
3141 cifs_uncached_readv_complete(struct work_struct *work)
3142 {
3143         struct cifs_readdata *rdata = container_of(work,
3144                                                 struct cifs_readdata, work);
3145
3146         complete(&rdata->done);
3147         collect_uncached_read_data(rdata->ctx);
3148         /* the below call can possibly free the last ref to aio ctx */
3149         kref_put(&rdata->refcount, cifs_uncached_readdata_release);
3150 }
3151
3152 static int
3153 uncached_fill_pages(struct TCP_Server_Info *server,
3154                     struct cifs_readdata *rdata, struct iov_iter *iter,
3155                     unsigned int len)
3156 {
3157         int result = 0;
3158         unsigned int i;
3159         unsigned int nr_pages = rdata->nr_pages;
3160         unsigned int page_offset = rdata->page_offset;
3161
3162         rdata->got_bytes = 0;
3163         rdata->tailsz = PAGE_SIZE;
3164         for (i = 0; i < nr_pages; i++) {
3165                 struct page *page = rdata->pages[i];
3166                 size_t n;
3167                 unsigned int segment_size = rdata->pagesz;
3168
3169                 if (i == 0)
3170                         segment_size -= page_offset;
3171                 else
3172                         page_offset = 0;
3173
3174
3175                 if (len <= 0) {
3176                         /* no need to hold page hostage */
3177                         rdata->pages[i] = NULL;
3178                         rdata->nr_pages--;
3179                         put_page(page);
3180                         continue;
3181                 }
3182
3183                 n = len;
3184                 if (len >= segment_size)
3185                         /* enough data to fill the page */
3186                         n = segment_size;
3187                 else
3188                         rdata->tailsz = len;
3189                 len -= n;
3190
3191                 if (iter)
3192                         result = copy_page_from_iter(
3193                                         page, page_offset, n, iter);
3194 #ifdef CONFIG_CIFS_SMB_DIRECT
3195                 else if (rdata->mr)
3196                         result = n;
3197 #endif
3198                 else
3199                         result = cifs_read_page_from_socket(
3200                                         server, page, page_offset, n);
3201                 if (result < 0)
3202                         break;
3203
3204                 rdata->got_bytes += result;
3205         }
3206
3207         return rdata->got_bytes > 0 && result != -ECONNABORTED ?
3208                                                 rdata->got_bytes : result;
3209 }
3210
3211 static int
3212 cifs_uncached_read_into_pages(struct TCP_Server_Info *server,
3213                               struct cifs_readdata *rdata, unsigned int len)
3214 {
3215         return uncached_fill_pages(server, rdata, NULL, len);
3216 }
3217
3218 static int
3219 cifs_uncached_copy_into_pages(struct TCP_Server_Info *server,
3220                               struct cifs_readdata *rdata,
3221                               struct iov_iter *iter)
3222 {
3223         return uncached_fill_pages(server, rdata, iter, iter->count);
3224 }
3225
3226 static int cifs_resend_rdata(struct cifs_readdata *rdata,
3227                         struct list_head *rdata_list,
3228                         struct cifs_aio_ctx *ctx)
3229 {
3230         unsigned int rsize, credits;
3231         int rc;
3232         struct TCP_Server_Info *server =
3233                 tlink_tcon(rdata->cfile->tlink)->ses->server;
3234
3235         /*
3236          * Wait for credits to resend this rdata.
3237          * Note: we are attempting to resend the whole rdata not in segments
3238          */
3239         do {
3240                 rc = server->ops->wait_mtu_credits(server, rdata->bytes,
3241                                                 &rsize, &credits);
3242
3243                 if (rc)
3244                         goto out;
3245
3246                 if (rsize < rdata->bytes) {
3247                         add_credits_and_wake_if(server, credits, 0);
3248                         msleep(1000);
3249                 }
3250         } while (rsize < rdata->bytes);
3251
3252         rc = -EAGAIN;
3253         while (rc == -EAGAIN) {
3254                 rc = 0;
3255                 if (rdata->cfile->invalidHandle)
3256                         rc = cifs_reopen_file(rdata->cfile, true);
3257                 if (!rc)
3258                         rc = server->ops->async_readv(rdata);
3259         }
3260
3261         if (!rc) {
3262                 /* Add to aio pending list */
3263                 list_add_tail(&rdata->list, rdata_list);
3264                 return 0;
3265         }
3266
3267         add_credits_and_wake_if(server, rdata->credits, 0);
3268 out:
3269         kref_put(&rdata->refcount,
3270                 cifs_uncached_readdata_release);
3271
3272         return rc;
3273 }
3274
3275 static int
3276 cifs_send_async_read(loff_t offset, size_t len, struct cifsFileInfo *open_file,
3277                      struct cifs_sb_info *cifs_sb, struct list_head *rdata_list,
3278                      struct cifs_aio_ctx *ctx)
3279 {
3280         struct cifs_readdata *rdata;
3281         unsigned int npages, rsize, credits;
3282         size_t cur_len;
3283         int rc;
3284         pid_t pid;
3285         struct TCP_Server_Info *server;
3286         struct page **pagevec;
3287         size_t start;
3288         struct iov_iter direct_iov = ctx->iter;
3289
3290         server = tlink_tcon(open_file->tlink)->ses->server;
3291
3292         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3293                 pid = open_file->pid;
3294         else
3295                 pid = current->tgid;
3296
3297         if (ctx->direct_io)
3298                 iov_iter_advance(&direct_iov, offset - ctx->pos);
3299
3300         do {
3301                 rc = server->ops->wait_mtu_credits(server, cifs_sb->rsize,
3302                                                    &rsize, &credits);
3303                 if (rc)
3304                         break;
3305
3306                 cur_len = min_t(const size_t, len, rsize);
3307
3308                 if (ctx->direct_io) {
3309                         ssize_t result;
3310
3311                         result = iov_iter_get_pages_alloc(
3312                                         &direct_iov, &pagevec,
3313                                         cur_len, &start);
3314                         if (result < 0) {
3315                                 cifs_dbg(VFS,
3316                                         "couldn't get user pages (cur_len=%zd)"
3317                                         " iter type %d"
3318                                         " iov_offset %zd count %zd\n",
3319                                         result, direct_iov.type,
3320                                         direct_iov.iov_offset,
3321                                         direct_iov.count);
3322                                 dump_stack();
3323                                 break;
3324                         }
3325                         cur_len = (size_t)result;
3326                         iov_iter_advance(&direct_iov, cur_len);
3327
3328                         rdata = cifs_readdata_direct_alloc(
3329                                         pagevec, cifs_uncached_readv_complete);
3330                         if (!rdata) {
3331                                 add_credits_and_wake_if(server, credits, 0);
3332                                 rc = -ENOMEM;
3333                                 break;
3334                         }
3335
3336                         npages = (cur_len + start + PAGE_SIZE-1) / PAGE_SIZE;
3337                         rdata->page_offset = start;
3338                         rdata->tailsz = npages > 1 ?
3339                                 cur_len-(PAGE_SIZE-start)-(npages-2)*PAGE_SIZE :
3340                                 cur_len;
3341
3342                 } else {
3343
3344                         npages = DIV_ROUND_UP(cur_len, PAGE_SIZE);
3345                         /* allocate a readdata struct */
3346                         rdata = cifs_readdata_alloc(npages,
3347                                             cifs_uncached_readv_complete);
3348                         if (!rdata) {
3349                                 add_credits_and_wake_if(server, credits, 0);
3350                                 rc = -ENOMEM;
3351                                 break;
3352                         }
3353
3354                         rc = cifs_read_allocate_pages(rdata, npages);
3355                         if (rc)
3356                                 goto error;
3357
3358                         rdata->tailsz = PAGE_SIZE;
3359                 }
3360
3361                 rdata->cfile = cifsFileInfo_get(open_file);
3362                 rdata->nr_pages = npages;
3363                 rdata->offset = offset;
3364                 rdata->bytes = cur_len;
3365                 rdata->pid = pid;
3366                 rdata->pagesz = PAGE_SIZE;
3367                 rdata->read_into_pages = cifs_uncached_read_into_pages;
3368                 rdata->copy_into_pages = cifs_uncached_copy_into_pages;
3369                 rdata->credits = credits;
3370                 rdata->ctx = ctx;
3371                 kref_get(&ctx->refcount);
3372
3373                 if (!rdata->cfile->invalidHandle ||
3374                     !(rc = cifs_reopen_file(rdata->cfile, true)))
3375                         rc = server->ops->async_readv(rdata);
3376 error:
3377                 if (rc) {
3378                         add_credits_and_wake_if(server, rdata->credits, 0);
3379                         kref_put(&rdata->refcount,
3380                                 cifs_uncached_readdata_release);
3381                         if (rc == -EAGAIN) {
3382                                 iov_iter_revert(&direct_iov, cur_len);
3383                                 continue;
3384                         }
3385                         break;
3386                 }
3387
3388                 list_add_tail(&rdata->list, rdata_list);
3389                 offset += cur_len;
3390                 len -= cur_len;
3391         } while (len > 0);
3392
3393         return rc;
3394 }
3395
3396 static void
3397 collect_uncached_read_data(struct cifs_aio_ctx *ctx)
3398 {
3399         struct cifs_readdata *rdata, *tmp;
3400         struct iov_iter *to = &ctx->iter;
3401         struct cifs_sb_info *cifs_sb;
3402         struct cifs_tcon *tcon;
3403         unsigned int i;
3404         int rc;
3405
3406         tcon = tlink_tcon(ctx->cfile->tlink);
3407         cifs_sb = CIFS_SB(ctx->cfile->dentry->d_sb);
3408
3409         mutex_lock(&ctx->aio_mutex);
3410
3411         if (list_empty(&ctx->list)) {
3412                 mutex_unlock(&ctx->aio_mutex);
3413                 return;
3414         }
3415
3416         rc = ctx->rc;
3417         /* the loop below should proceed in the order of increasing offsets */
3418 again:
3419         list_for_each_entry_safe(rdata, tmp, &ctx->list, list) {
3420                 if (!rc) {
3421                         if (!try_wait_for_completion(&rdata->done)) {
3422                                 mutex_unlock(&ctx->aio_mutex);
3423                                 return;
3424                         }
3425
3426                         if (rdata->result == -EAGAIN) {
3427                                 /* resend call if it's a retryable error */
3428                                 struct list_head tmp_list;
3429                                 unsigned int got_bytes = rdata->got_bytes;
3430
3431                                 list_del_init(&rdata->list);
3432                                 INIT_LIST_HEAD(&tmp_list);
3433
3434                                 /*
3435                                  * Got a part of data and then reconnect has
3436                                  * happened -- fill the buffer and continue
3437                                  * reading.
3438                                  */
3439                                 if (got_bytes && got_bytes < rdata->bytes) {
3440                                         rc = 0;
3441                                         if (!ctx->direct_io)
3442                                                 rc = cifs_readdata_to_iov(rdata, to);
3443                                         if (rc) {
3444                                                 kref_put(&rdata->refcount,
3445                                                         cifs_uncached_readdata_release);
3446                                                 continue;
3447                                         }
3448                                 }
3449
3450                                 if (ctx->direct_io) {
3451                                         /*
3452                                          * Re-use rdata as this is a
3453                                          * direct I/O
3454                                          */
3455                                         rc = cifs_resend_rdata(
3456                                                 rdata,
3457                                                 &tmp_list, ctx);
3458                                 } else {
3459                                         rc = cifs_send_async_read(
3460                                                 rdata->offset + got_bytes,
3461                                                 rdata->bytes - got_bytes,
3462                                                 rdata->cfile, cifs_sb,
3463                                                 &tmp_list, ctx);
3464
3465                                         kref_put(&rdata->refcount,
3466                                                 cifs_uncached_readdata_release);
3467                                 }
3468
3469                                 list_splice(&tmp_list, &ctx->list);
3470
3471                                 goto again;
3472                         } else if (rdata->result)
3473                                 rc = rdata->result;
3474                         else if (!ctx->direct_io)
3475                                 rc = cifs_readdata_to_iov(rdata, to);
3476
3477                         /* if there was a short read -- discard anything left */
3478                         if (rdata->got_bytes && rdata->got_bytes < rdata->bytes)
3479                                 rc = -ENODATA;
3480
3481                         ctx->total_len += rdata->got_bytes;
3482                 }
3483                 list_del_init(&rdata->list);
3484                 kref_put(&rdata->refcount, cifs_uncached_readdata_release);
3485         }
3486
3487         if (!ctx->direct_io) {
3488                 for (i = 0; i < ctx->npages; i++) {
3489                         if (ctx->should_dirty)
3490                                 set_page_dirty(ctx->bv[i].bv_page);
3491                         put_page(ctx->bv[i].bv_page);
3492                 }
3493
3494                 ctx->total_len = ctx->len - iov_iter_count(to);
3495         }
3496
3497         cifs_stats_bytes_read(tcon, ctx->total_len);
3498
3499         /* mask nodata case */
3500         if (rc == -ENODATA)
3501                 rc = 0;
3502
3503         ctx->rc = (rc == 0) ? ctx->total_len : rc;
3504
3505         mutex_unlock(&ctx->aio_mutex);
3506
3507         if (ctx->iocb && ctx->iocb->ki_complete)
3508                 ctx->iocb->ki_complete(ctx->iocb, ctx->rc, 0);
3509         else
3510                 complete(&ctx->done);
3511 }
3512
3513 static ssize_t __cifs_readv(
3514         struct kiocb *iocb, struct iov_iter *to, bool direct)
3515 {
3516         size_t len;
3517         struct file *file = iocb->ki_filp;
3518         struct cifs_sb_info *cifs_sb;
3519         struct cifsFileInfo *cfile;
3520         struct cifs_tcon *tcon;
3521         ssize_t rc, total_read = 0;
3522         loff_t offset = iocb->ki_pos;
3523         struct cifs_aio_ctx *ctx;
3524
3525         /*
3526          * iov_iter_get_pages_alloc() doesn't work with ITER_KVEC,
3527          * fall back to data copy read path
3528          * this could be improved by getting pages directly in ITER_KVEC
3529          */
3530         if (direct && to->type & ITER_KVEC) {
3531                 cifs_dbg(FYI, "use non-direct cifs_user_readv for kvec I/O\n");
3532                 direct = false;
3533         }
3534
3535         len = iov_iter_count(to);
3536         if (!len)
3537                 return 0;
3538
3539         cifs_sb = CIFS_FILE_SB(file);
3540         cfile = file->private_data;
3541         tcon = tlink_tcon(cfile->tlink);
3542
3543         if (!tcon->ses->server->ops->async_readv)
3544                 return -ENOSYS;
3545
3546         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3547                 cifs_dbg(FYI, "attempting read on write only file instance\n");
3548
3549         ctx = cifs_aio_ctx_alloc();
3550         if (!ctx)
3551                 return -ENOMEM;
3552
3553         ctx->cfile = cifsFileInfo_get(cfile);
3554
3555         if (!is_sync_kiocb(iocb))
3556                 ctx->iocb = iocb;
3557
3558         if (iter_is_iovec(to))
3559                 ctx->should_dirty = true;
3560
3561         if (direct) {
3562                 ctx->pos = offset;
3563                 ctx->direct_io = true;
3564                 ctx->iter = *to;
3565                 ctx->len = len;
3566         } else {
3567                 rc = setup_aio_ctx_iter(ctx, to, READ);
3568                 if (rc) {
3569                         kref_put(&ctx->refcount, cifs_aio_ctx_release);
3570                         return rc;
3571                 }
3572                 len = ctx->len;
3573         }
3574
3575         /* grab a lock here due to read response handlers can access ctx */
3576         mutex_lock(&ctx->aio_mutex);
3577
3578         rc = cifs_send_async_read(offset, len, cfile, cifs_sb, &ctx->list, ctx);
3579
3580         /* if at least one read request send succeeded, then reset rc */
3581         if (!list_empty(&ctx->list))
3582                 rc = 0;
3583
3584         mutex_unlock(&ctx->aio_mutex);
3585
3586         if (rc) {
3587                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
3588                 return rc;
3589         }
3590
3591         if (!is_sync_kiocb(iocb)) {
3592                 kref_put(&ctx->refcount, cifs_aio_ctx_release);
3593                 return -EIOCBQUEUED;
3594         }
3595
3596         rc = wait_for_completion_killable(&ctx->done);
3597         if (rc) {
3598                 mutex_lock(&ctx->aio_mutex);
3599                 ctx->rc = rc = -EINTR;
3600                 total_read = ctx->total_len;
3601                 mutex_unlock(&ctx->aio_mutex);
3602         } else {
3603                 rc = ctx->rc;
3604                 total_read = ctx->total_len;
3605         }
3606
3607         kref_put(&ctx->refcount, cifs_aio_ctx_release);
3608
3609         if (total_read) {
3610                 iocb->ki_pos += total_read;
3611                 return total_read;
3612         }
3613         return rc;
3614 }
3615
3616 ssize_t cifs_direct_readv(struct kiocb *iocb, struct iov_iter *to)
3617 {
3618         return __cifs_readv(iocb, to, true);
3619 }
3620
3621 ssize_t cifs_user_readv(struct kiocb *iocb, struct iov_iter *to)
3622 {
3623         return __cifs_readv(iocb, to, false);
3624 }
3625
3626 ssize_t
3627 cifs_strict_readv(struct kiocb *iocb, struct iov_iter *to)
3628 {
3629         struct inode *inode = file_inode(iocb->ki_filp);
3630         struct cifsInodeInfo *cinode = CIFS_I(inode);
3631         struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb);
3632         struct cifsFileInfo *cfile = (struct cifsFileInfo *)
3633                                                 iocb->ki_filp->private_data;
3634         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
3635         int rc = -EACCES;
3636
3637         /*
3638          * In strict cache mode we need to read from the server all the time
3639          * if we don't have level II oplock because the server can delay mtime
3640          * change - so we can't make a decision about inode invalidating.
3641          * And we can also fail with pagereading if there are mandatory locks
3642          * on pages affected by this read but not on the region from pos to
3643          * pos+len-1.
3644          */
3645         if (!CIFS_CACHE_READ(cinode))
3646                 return cifs_user_readv(iocb, to);
3647
3648         if (cap_unix(tcon->ses) &&
3649             (CIFS_UNIX_FCNTL_CAP & le64_to_cpu(tcon->fsUnixInfo.Capability)) &&
3650             ((cifs_sb->mnt_cifs_flags & CIFS_MOUNT_NOPOSIXBRL) == 0))
3651                 return generic_file_read_iter(iocb, to);
3652
3653         /*
3654          * We need to hold the sem to be sure nobody modifies lock list
3655          * with a brlock that prevents reading.
3656          */
3657         down_read(&cinode->lock_sem);
3658         if (!cifs_find_lock_conflict(cfile, iocb->ki_pos, iov_iter_count(to),
3659                                      tcon->ses->server->vals->shared_lock_type,
3660                                      0, NULL, CIFS_READ_OP))
3661                 rc = generic_file_read_iter(iocb, to);
3662         up_read(&cinode->lock_sem);
3663         return rc;
3664 }
3665
3666 static ssize_t
3667 cifs_read(struct file *file, char *read_data, size_t read_size, loff_t *offset)
3668 {
3669         int rc = -EACCES;
3670         unsigned int bytes_read = 0;
3671         unsigned int total_read;
3672         unsigned int current_read_size;
3673         unsigned int rsize;
3674         struct cifs_sb_info *cifs_sb;
3675         struct cifs_tcon *tcon;
3676         struct TCP_Server_Info *server;
3677         unsigned int xid;
3678         char *cur_offset;
3679         struct cifsFileInfo *open_file;
3680         struct cifs_io_parms io_parms;
3681         int buf_type = CIFS_NO_BUFFER;
3682         __u32 pid;
3683
3684         xid = get_xid();
3685         cifs_sb = CIFS_FILE_SB(file);
3686
3687         /* FIXME: set up handlers for larger reads and/or convert to async */
3688         rsize = min_t(unsigned int, cifs_sb->rsize, CIFSMaxBufSize);
3689
3690         if (file->private_data == NULL) {
3691                 rc = -EBADF;
3692                 free_xid(xid);
3693                 return rc;
3694         }
3695         open_file = file->private_data;
3696         tcon = tlink_tcon(open_file->tlink);
3697         server = tcon->ses->server;
3698
3699         if (!server->ops->sync_read) {
3700                 free_xid(xid);
3701                 return -ENOSYS;
3702         }
3703
3704         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
3705                 pid = open_file->pid;
3706         else
3707                 pid = current->tgid;
3708
3709         if ((file->f_flags & O_ACCMODE) == O_WRONLY)
3710                 cifs_dbg(FYI, "attempting read on write only file instance\n");
3711
3712         for (total_read = 0, cur_offset = read_data; read_size > total_read;
3713              total_read += bytes_read, cur_offset += bytes_read) {
3714                 do {
3715                         current_read_size = min_t(uint, read_size - total_read,
3716                                                   rsize);
3717                         /*
3718                          * For windows me and 9x we do not want to request more
3719                          * than it negotiated since it will refuse the read
3720                          * then.
3721                          */
3722                         if ((tcon->ses) && !(tcon->ses->capabilities &
3723                                 tcon->ses->server->vals->cap_large_files)) {
3724                                 current_read_size = min_t(uint,
3725                                         current_read_size, CIFSMaxBufSize);
3726                         }
3727                         if (open_file->invalidHandle) {
3728                                 rc = cifs_reopen_file(open_file, true);
3729                                 if (rc != 0)
3730                                         break;
3731                         }
3732                         io_parms.pid = pid;
3733                         io_parms.tcon = tcon;
3734                         io_parms.offset = *offset;
3735                         io_parms.length = current_read_size;
3736                         rc = server->ops->sync_read(xid, &open_file->fid, &io_parms,
3737                                                     &bytes_read, &cur_offset,
3738                                                     &buf_type);
3739                 } while (rc == -EAGAIN);
3740
3741                 if (rc || (bytes_read == 0)) {
3742                         if (total_read) {
3743                                 break;
3744                         } else {
3745                                 free_xid(xid);
3746                                 return rc;
3747                         }
3748                 } else {
3749                         cifs_stats_bytes_read(tcon, total_read);
3750                         *offset += bytes_read;
3751                 }
3752         }
3753         free_xid(xid);
3754         return total_read;
3755 }
3756
3757 /*
3758  * If the page is mmap'ed into a process' page tables, then we need to make
3759  * sure that it doesn't change while being written back.
3760  */
3761 static vm_fault_t
3762 cifs_page_mkwrite(struct vm_fault *vmf)
3763 {
3764         struct page *page = vmf->page;
3765
3766         lock_page(page);
3767         return VM_FAULT_LOCKED;
3768 }
3769
3770 static const struct vm_operations_struct cifs_file_vm_ops = {
3771         .fault = filemap_fault,
3772         .map_pages = filemap_map_pages,
3773         .page_mkwrite = cifs_page_mkwrite,
3774 };
3775
3776 int cifs_file_strict_mmap(struct file *file, struct vm_area_struct *vma)
3777 {
3778         int xid, rc = 0;
3779         struct inode *inode = file_inode(file);
3780
3781         xid = get_xid();
3782
3783         if (!CIFS_CACHE_READ(CIFS_I(inode)))
3784                 rc = cifs_zap_mapping(inode);
3785         if (!rc)
3786                 rc = generic_file_mmap(file, vma);
3787         if (!rc)
3788                 vma->vm_ops = &cifs_file_vm_ops;
3789
3790         free_xid(xid);
3791         return rc;
3792 }
3793
3794 int cifs_file_mmap(struct file *file, struct vm_area_struct *vma)
3795 {
3796         int rc, xid;
3797
3798         xid = get_xid();
3799
3800         rc = cifs_revalidate_file(file);
3801         if (rc)
3802                 cifs_dbg(FYI, "Validation prior to mmap failed, error=%d\n",
3803                          rc);
3804         if (!rc)
3805                 rc = generic_file_mmap(file, vma);
3806         if (!rc)
3807                 vma->vm_ops = &cifs_file_vm_ops;
3808
3809         free_xid(xid);
3810         return rc;
3811 }
3812
3813 static void
3814 cifs_readv_complete(struct work_struct *work)
3815 {
3816         unsigned int i, got_bytes;
3817         struct cifs_readdata *rdata = container_of(work,
3818                                                 struct cifs_readdata, work);
3819
3820         got_bytes = rdata->got_bytes;
3821         for (i = 0; i < rdata->nr_pages; i++) {
3822                 struct page *page = rdata->pages[i];
3823
3824                 lru_cache_add_file(page);
3825
3826                 if (rdata->result == 0 ||
3827                     (rdata->result == -EAGAIN && got_bytes)) {
3828                         flush_dcache_page(page);
3829                         SetPageUptodate(page);
3830                 }
3831
3832                 unlock_page(page);
3833
3834                 if (rdata->result == 0 ||
3835                     (rdata->result == -EAGAIN && got_bytes))
3836                         cifs_readpage_to_fscache(rdata->mapping->host, page);
3837
3838                 got_bytes -= min_t(unsigned int, PAGE_SIZE, got_bytes);
3839
3840                 put_page(page);
3841                 rdata->pages[i] = NULL;
3842         }
3843         kref_put(&rdata->refcount, cifs_readdata_release);
3844 }
3845
3846 static int
3847 readpages_fill_pages(struct TCP_Server_Info *server,
3848                      struct cifs_readdata *rdata, struct iov_iter *iter,
3849                      unsigned int len)
3850 {
3851         int result = 0;
3852         unsigned int i;
3853         u64 eof;
3854         pgoff_t eof_index;
3855         unsigned int nr_pages = rdata->nr_pages;
3856         unsigned int page_offset = rdata->page_offset;
3857
3858         /* determine the eof that the server (probably) has */
3859         eof = CIFS_I(rdata->mapping->host)->server_eof;
3860         eof_index = eof ? (eof - 1) >> PAGE_SHIFT : 0;
3861         cifs_dbg(FYI, "eof=%llu eof_index=%lu\n", eof, eof_index);
3862
3863         rdata->got_bytes = 0;
3864         rdata->tailsz = PAGE_SIZE;
3865         for (i = 0; i < nr_pages; i++) {
3866                 struct page *page = rdata->pages[i];
3867                 unsigned int to_read = rdata->pagesz;
3868                 size_t n;
3869
3870                 if (i == 0)
3871                         to_read -= page_offset;
3872                 else
3873                         page_offset = 0;
3874
3875                 n = to_read;
3876
3877                 if (len >= to_read) {
3878                         len -= to_read;
3879                 } else if (len > 0) {
3880                         /* enough for partial page, fill and zero the rest */
3881                         zero_user(page, len + page_offset, to_read - len);
3882                         n = rdata->tailsz = len;
3883                         len = 0;
3884                 } else if (page->index > eof_index) {
3885                         /*
3886                          * The VFS will not try to do readahead past the
3887                          * i_size, but it's possible that we have outstanding
3888                          * writes with gaps in the middle and the i_size hasn't
3889                          * caught up yet. Populate those with zeroed out pages
3890                          * to prevent the VFS from repeatedly attempting to
3891                          * fill them until the writes are flushed.
3892                          */
3893                         zero_user(page, 0, PAGE_SIZE);
3894                         lru_cache_add_file(page);
3895                         flush_dcache_page(page);
3896                         SetPageUptodate(page);
3897                         unlock_page(page);
3898                         put_page(page);
3899                         rdata->pages[i] = NULL;
3900                         rdata->nr_pages--;
3901                         continue;
3902                 } else {
3903                         /* no need to hold page hostage */
3904                         lru_cache_add_file(page);
3905                         unlock_page(page);
3906                         put_page(page);
3907                         rdata->pages[i] = NULL;
3908                         rdata->nr_pages--;
3909                         continue;
3910                 }
3911
3912                 if (iter)
3913                         result = copy_page_from_iter(
3914                                         page, page_offset, n, iter);
3915 #ifdef CONFIG_CIFS_SMB_DIRECT
3916                 else if (rdata->mr)
3917                         result = n;
3918 #endif
3919                 else
3920                         result = cifs_read_page_from_socket(
3921                                         server, page, page_offset, n);
3922                 if (result < 0)
3923                         break;
3924
3925                 rdata->got_bytes += result;
3926         }
3927
3928         return rdata->got_bytes > 0 && result != -ECONNABORTED ?
3929                                                 rdata->got_bytes : result;
3930 }
3931
3932 static int
3933 cifs_readpages_read_into_pages(struct TCP_Server_Info *server,
3934                                struct cifs_readdata *rdata, unsigned int len)
3935 {
3936         return readpages_fill_pages(server, rdata, NULL, len);
3937 }
3938
3939 static int
3940 cifs_readpages_copy_into_pages(struct TCP_Server_Info *server,
3941                                struct cifs_readdata *rdata,
3942                                struct iov_iter *iter)
3943 {
3944         return readpages_fill_pages(server, rdata, iter, iter->count);
3945 }
3946
3947 static int
3948 readpages_get_pages(struct address_space *mapping, struct list_head *page_list,
3949                     unsigned int rsize, struct list_head *tmplist,
3950                     unsigned int *nr_pages, loff_t *offset, unsigned int *bytes)
3951 {
3952         struct page *page, *tpage;
3953         unsigned int expected_index;
3954         int rc;
3955         gfp_t gfp = readahead_gfp_mask(mapping);
3956
3957         INIT_LIST_HEAD(tmplist);
3958
3959         page = list_entry(page_list->prev, struct page, lru);
3960
3961         /*
3962          * Lock the page and put it in the cache. Since no one else
3963          * should have access to this page, we're safe to simply set
3964          * PG_locked without checking it first.
3965          */
3966         __SetPageLocked(page);
3967         rc = add_to_page_cache_locked(page, mapping,
3968                                       page->index, gfp);
3969
3970         /* give up if we can't stick it in the cache */
3971         if (rc) {
3972                 __ClearPageLocked(page);
3973                 return rc;
3974         }
3975
3976         /* move first page to the tmplist */
3977         *offset = (loff_t)page->index << PAGE_SHIFT;
3978         *bytes = PAGE_SIZE;
3979         *nr_pages = 1;
3980         list_move_tail(&page->lru, tmplist);
3981
3982         /* now try and add more pages onto the request */
3983         expected_index = page->index + 1;
3984         list_for_each_entry_safe_reverse(page, tpage, page_list, lru) {
3985                 /* discontinuity ? */
3986                 if (page->index != expected_index)
3987                         break;
3988
3989                 /* would this page push the read over the rsize? */
3990                 if (*bytes + PAGE_SIZE > rsize)
3991                         break;
3992
3993                 __SetPageLocked(page);
3994                 if (add_to_page_cache_locked(page, mapping, page->index, gfp)) {
3995                         __ClearPageLocked(page);
3996                         break;
3997                 }
3998                 list_move_tail(&page->lru, tmplist);
3999                 (*bytes) += PAGE_SIZE;
4000                 expected_index++;
4001                 (*nr_pages)++;
4002         }
4003         return rc;
4004 }
4005
4006 static int cifs_readpages(struct file *file, struct address_space *mapping,
4007         struct list_head *page_list, unsigned num_pages)
4008 {
4009         int rc;
4010         struct list_head tmplist;
4011         struct cifsFileInfo *open_file = file->private_data;
4012         struct cifs_sb_info *cifs_sb = CIFS_FILE_SB(file);
4013         struct TCP_Server_Info *server;
4014         pid_t pid;
4015         unsigned int xid;
4016
4017         xid = get_xid();
4018         /*
4019          * Reads as many pages as possible from fscache. Returns -ENOBUFS
4020          * immediately if the cookie is negative
4021          *
4022          * After this point, every page in the list might have PG_fscache set,
4023          * so we will need to clean that up off of every page we don't use.
4024          */
4025         rc = cifs_readpages_from_fscache(mapping->host, mapping, page_list,
4026                                          &num_pages);
4027         if (rc == 0) {
4028                 free_xid(xid);
4029                 return rc;
4030         }
4031
4032         if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
4033                 pid = open_file->pid;
4034         else
4035                 pid = current->tgid;
4036
4037         rc = 0;
4038         server = tlink_tcon(open_file->tlink)->ses->server;
4039
4040         cifs_dbg(FYI, "%s: file=%p mapping=%p num_pages=%u\n",
4041                  __func__, file, mapping, num_pages);
4042
4043         /*
4044          * Start with the page at end of list and move it to private
4045          * list. Do the same with any following pages until we hit
4046          * the rsize limit, hit an index discontinuity, or run out of
4047          * pages. Issue the async read and then start the loop again
4048          * until the list is empty.
4049          *
4050          * Note that list order is important. The page_list is in
4051          * the order of declining indexes. When we put the pages in
4052          * the rdata->pages, then we want them in increasing order.
4053          */
4054         while (!list_empty(page_list)) {
4055                 unsigned int i, nr_pages, bytes, rsize;
4056                 loff_t offset;
4057                 struct page *page, *tpage;
4058                 struct cifs_readdata *rdata;
4059                 unsigned credits;
4060
4061                 rc = server->ops->wait_mtu_credits(server, cifs_sb->rsize,
4062                                                    &rsize, &credits);
4063                 if (rc)
4064                         break;
4065
4066                 /*
4067                  * Give up immediately if rsize is too small to read an entire
4068                  * page. The VFS will fall back to readpage. We should never
4069                  * reach this point however since we set ra_pages to 0 when the
4070                  * rsize is smaller than a cache page.
4071                  */
4072                 if (unlikely(rsize < PAGE_SIZE)) {
4073                         add_credits_and_wake_if(server, credits, 0);
4074                         free_xid(xid);
4075                         return 0;
4076                 }
4077
4078                 rc = readpages_get_pages(mapping, page_list, rsize, &tmplist,
4079                                          &nr_pages, &offset, &bytes);
4080                 if (rc) {
4081                         add_credits_and_wake_if(server, credits, 0);
4082                         break;
4083                 }
4084
4085                 rdata = cifs_readdata_alloc(nr_pages, cifs_readv_complete);
4086                 if (!rdata) {
4087                         /* best to give up if we're out of mem */
4088                         list_for_each_entry_safe(page, tpage, &tmplist, lru) {
4089                                 list_del(&page->lru);
4090                                 lru_cache_add_file(page);
4091                                 unlock_page(page);
4092                                 put_page(page);
4093                         }
4094                         rc = -ENOMEM;
4095                         add_credits_and_wake_if(server, credits, 0);
4096                         break;
4097                 }
4098
4099                 rdata->cfile = cifsFileInfo_get(open_file);
4100                 rdata->mapping = mapping;
4101                 rdata->offset = offset;
4102                 rdata->bytes = bytes;
4103                 rdata->pid = pid;
4104                 rdata->pagesz = PAGE_SIZE;
4105                 rdata->tailsz = PAGE_SIZE;
4106                 rdata->read_into_pages = cifs_readpages_read_into_pages;
4107                 rdata->copy_into_pages = cifs_readpages_copy_into_pages;
4108                 rdata->credits = credits;
4109
4110                 list_for_each_entry_safe(page, tpage, &tmplist, lru) {
4111                         list_del(&page->lru);
4112                         rdata->pages[rdata->nr_pages++] = page;
4113                 }
4114
4115                 if (!rdata->cfile->invalidHandle ||
4116                     !(rc = cifs_reopen_file(rdata->cfile, true)))
4117                         rc = server->ops->async_readv(rdata);
4118                 if (rc) {
4119                         add_credits_and_wake_if(server, rdata->credits, 0);
4120                         for (i = 0; i < rdata->nr_pages; i++) {
4121                                 page = rdata->pages[i];
4122                                 lru_cache_add_file(page);
4123                                 unlock_page(page);
4124                                 put_page(page);
4125                         }
4126                         /* Fallback to the readpage in error/reconnect cases */
4127                         kref_put(&rdata->refcount, cifs_readdata_release);
4128                         break;
4129                 }
4130
4131                 kref_put(&rdata->refcount, cifs_readdata_release);
4132         }
4133
4134         /* Any pages that have been shown to fscache but didn't get added to
4135          * the pagecache must be uncached before they get returned to the
4136          * allocator.
4137          */
4138         cifs_fscache_readpages_cancel(mapping->host, page_list);
4139         free_xid(xid);
4140         return rc;
4141 }
4142
4143 /*
4144  * cifs_readpage_worker must be called with the page pinned
4145  */
4146 static int cifs_readpage_worker(struct file *file, struct page *page,
4147         loff_t *poffset)
4148 {
4149         char *read_data;
4150         int rc;
4151
4152         /* Is the page cached? */
4153         rc = cifs_readpage_from_fscache(file_inode(file), page);
4154         if (rc == 0)
4155                 goto read_complete;
4156
4157         read_data = kmap(page);
4158         /* for reads over a certain size could initiate async read ahead */
4159
4160         rc = cifs_read(file, read_data, PAGE_SIZE, poffset);
4161
4162         if (rc < 0)
4163                 goto io_error;
4164         else
4165                 cifs_dbg(FYI, "Bytes read %d\n", rc);
4166
4167         /* we do not want atime to be less than mtime, it broke some apps */
4168         file_inode(file)->i_atime = current_time(file_inode(file));
4169         if (timespec64_compare(&(file_inode(file)->i_atime), &(file_inode(file)->i_mtime)))
4170                 file_inode(file)->i_atime = file_inode(file)->i_mtime;
4171         else
4172                 file_inode(file)->i_atime = current_time(file_inode(file));
4173
4174         if (PAGE_SIZE > rc)
4175                 memset(read_data + rc, 0, PAGE_SIZE - rc);
4176
4177         flush_dcache_page(page);
4178         SetPageUptodate(page);
4179
4180         /* send this page to the cache */
4181         cifs_readpage_to_fscache(file_inode(file), page);
4182
4183         rc = 0;
4184
4185 io_error:
4186         kunmap(page);
4187         unlock_page(page);
4188
4189 read_complete:
4190         return rc;
4191 }
4192
4193 static int cifs_readpage(struct file *file, struct page *page)
4194 {
4195         loff_t offset = (loff_t)page->index << PAGE_SHIFT;
4196         int rc = -EACCES;
4197         unsigned int xid;
4198
4199         xid = get_xid();
4200
4201         if (file->private_data == NULL) {
4202                 rc = -EBADF;
4203                 free_xid(xid);
4204                 return rc;
4205         }
4206
4207         cifs_dbg(FYI, "readpage %p at offset %d 0x%x\n",
4208                  page, (int)offset, (int)offset);
4209
4210         rc = cifs_readpage_worker(file, page, &offset);
4211
4212         free_xid(xid);
4213         return rc;
4214 }
4215
4216 static int is_inode_writable(struct cifsInodeInfo *cifs_inode)
4217 {
4218         struct cifsFileInfo *open_file;
4219         struct cifs_tcon *tcon =
4220                 cifs_sb_master_tcon(CIFS_SB(cifs_inode->vfs_inode.i_sb));
4221
4222         spin_lock(&tcon->open_file_lock);
4223         list_for_each_entry(open_file, &cifs_inode->openFileList, flist) {
4224                 if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
4225                         spin_unlock(&tcon->open_file_lock);
4226                         return 1;
4227                 }
4228         }
4229         spin_unlock(&tcon->open_file_lock);
4230         return 0;
4231 }
4232
4233 /* We do not want to update the file size from server for inodes
4234    open for write - to avoid races with writepage extending
4235    the file - in the future we could consider allowing
4236    refreshing the inode only on increases in the file size
4237    but this is tricky to do without racing with writebehind
4238    page caching in the current Linux kernel design */
4239 bool is_size_safe_to_change(struct cifsInodeInfo *cifsInode, __u64 end_of_file)
4240 {
4241         if (!cifsInode)
4242                 return true;
4243
4244         if (is_inode_writable(cifsInode)) {
4245                 /* This inode is open for write at least once */
4246                 struct cifs_sb_info *cifs_sb;
4247
4248                 cifs_sb = CIFS_SB(cifsInode->vfs_inode.i_sb);
4249                 if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_DIRECT_IO) {
4250                         /* since no page cache to corrupt on directio
4251                         we can change size safely */
4252                         return true;
4253                 }
4254
4255                 if (i_size_read(&cifsInode->vfs_inode) < end_of_file)
4256                         return true;
4257
4258                 return false;
4259         } else
4260                 return true;
4261 }
4262
4263 static int cifs_write_begin(struct file *file, struct address_space *mapping,
4264                         loff_t pos, unsigned len, unsigned flags,
4265                         struct page **pagep, void **fsdata)
4266 {
4267         int oncethru = 0;
4268         pgoff_t index = pos >> PAGE_SHIFT;
4269         loff_t offset = pos & (PAGE_SIZE - 1);
4270         loff_t page_start = pos & PAGE_MASK;
4271         loff_t i_size;
4272         struct page *page;
4273         int rc = 0;
4274
4275         cifs_dbg(FYI, "write_begin from %lld len %d\n", (long long)pos, len);
4276
4277 start:
4278         page = grab_cache_page_write_begin(mapping, index, flags);
4279         if (!page) {
4280                 rc = -ENOMEM;
4281                 goto out;
4282         }
4283
4284         if (PageUptodate(page))
4285                 goto out;
4286
4287         /*
4288          * If we write a full page it will be up to date, no need to read from
4289          * the server. If the write is short, we'll end up doing a sync write
4290          * instead.
4291          */
4292         if (len == PAGE_SIZE)
4293                 goto out;
4294
4295         /*
4296          * optimize away the read when we have an oplock, and we're not
4297          * expecting to use any of the data we'd be reading in. That
4298          * is, when the page lies beyond the EOF, or straddles the EOF
4299          * and the write will cover all of the existing data.
4300          */
4301         if (CIFS_CACHE_READ(CIFS_I(mapping->host))) {
4302                 i_size = i_size_read(mapping->host);
4303                 if (page_start >= i_size ||
4304                     (offset == 0 && (pos + len) >= i_size)) {
4305                         zero_user_segments(page, 0, offset,
4306                                            offset + len,
4307                                            PAGE_SIZE);
4308                         /*
4309                          * PageChecked means that the parts of the page
4310                          * to which we're not writing are considered up
4311                          * to date. Once the data is copied to the
4312                          * page, it can be set uptodate.
4313                          */
4314                         SetPageChecked(page);
4315                         goto out;
4316                 }
4317         }
4318
4319         if ((file->f_flags & O_ACCMODE) != O_WRONLY && !oncethru) {
4320                 /*
4321                  * might as well read a page, it is fast enough. If we get
4322                  * an error, we don't need to return it. cifs_write_end will
4323                  * do a sync write instead since PG_uptodate isn't set.
4324                  */
4325                 cifs_readpage_worker(file, page, &page_start);
4326                 put_page(page);
4327                 oncethru = 1;
4328                 goto start;
4329         } else {
4330                 /* we could try using another file handle if there is one -
4331                    but how would we lock it to prevent close of that handle
4332                    racing with this read? In any case
4333                    this will be written out by write_end so is fine */
4334         }
4335 out:
4336         *pagep = page;
4337         return rc;
4338 }
4339
4340 static int cifs_release_page(struct page *page, gfp_t gfp)
4341 {
4342         if (PagePrivate(page))
4343                 return 0;
4344
4345         return cifs_fscache_release_page(page, gfp);
4346 }
4347
4348 static void cifs_invalidate_page(struct page *page, unsigned int offset,
4349                                  unsigned int length)
4350 {
4351         struct cifsInodeInfo *cifsi = CIFS_I(page->mapping->host);
4352
4353         if (offset == 0 && length == PAGE_SIZE)
4354                 cifs_fscache_invalidate_page(page, &cifsi->vfs_inode);
4355 }
4356
4357 static int cifs_launder_page(struct page *page)
4358 {
4359         int rc = 0;
4360         loff_t range_start = page_offset(page);
4361         loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1);
4362         struct writeback_control wbc = {
4363                 .sync_mode = WB_SYNC_ALL,
4364                 .nr_to_write = 0,
4365                 .range_start = range_start,
4366                 .range_end = range_end,
4367         };
4368
4369         cifs_dbg(FYI, "Launder page: %p\n", page);
4370
4371         if (clear_page_dirty_for_io(page))
4372                 rc = cifs_writepage_locked(page, &wbc);
4373
4374         cifs_fscache_invalidate_page(page, page->mapping->host);
4375         return rc;
4376 }
4377
4378 void cifs_oplock_break(struct work_struct *work)
4379 {
4380         struct cifsFileInfo *cfile = container_of(work, struct cifsFileInfo,
4381                                                   oplock_break);
4382         struct inode *inode = d_inode(cfile->dentry);
4383         struct cifsInodeInfo *cinode = CIFS_I(inode);
4384         struct cifs_tcon *tcon = tlink_tcon(cfile->tlink);
4385         struct TCP_Server_Info *server = tcon->ses->server;
4386         int rc = 0;
4387
4388         wait_on_bit(&cinode->flags, CIFS_INODE_PENDING_WRITERS,
4389                         TASK_UNINTERRUPTIBLE);
4390
4391         server->ops->downgrade_oplock(server, cinode,
4392                 test_bit(CIFS_INODE_DOWNGRADE_OPLOCK_TO_L2, &cinode->flags));
4393
4394         if (!CIFS_CACHE_WRITE(cinode) && CIFS_CACHE_READ(cinode) &&
4395                                                 cifs_has_mand_locks(cinode)) {
4396                 cifs_dbg(FYI, "Reset oplock to None for inode=%p due to mand locks\n",
4397                          inode);
4398                 cinode->oplock = 0;
4399         }
4400
4401         if (inode && S_ISREG(inode->i_mode)) {
4402                 if (CIFS_CACHE_READ(cinode))
4403                         break_lease(inode, O_RDONLY);
4404                 else
4405                         break_lease(inode, O_WRONLY);
4406                 rc = filemap_fdatawrite(inode->i_mapping);
4407                 if (!CIFS_CACHE_READ(cinode)) {
4408                         rc = filemap_fdatawait(inode->i_mapping);
4409                         mapping_set_error(inode->i_mapping, rc);
4410                         cifs_zap_mapping(inode);
4411                 }
4412                 cifs_dbg(FYI, "Oplock flush inode %p rc %d\n", inode, rc);
4413         }
4414
4415         rc = cifs_push_locks(cfile);
4416         if (rc)
4417                 cifs_dbg(VFS, "Push locks rc = %d\n", rc);
4418
4419         /*
4420          * releasing stale oplock after recent reconnect of smb session using
4421          * a now incorrect file handle is not a data integrity issue but do
4422          * not bother sending an oplock release if session to server still is
4423          * disconnected since oplock already released by the server
4424          */
4425         if (!cfile->oplock_break_cancelled) {
4426                 rc = tcon->ses->server->ops->oplock_response(tcon, &cfile->fid,
4427                                                              cinode);
4428                 cifs_dbg(FYI, "Oplock release rc = %d\n", rc);
4429         }
4430         cifs_done_oplock_break(cinode);
4431 }
4432
4433 /*
4434  * The presence of cifs_direct_io() in the address space ops vector
4435  * allowes open() O_DIRECT flags which would have failed otherwise.
4436  *
4437  * In the non-cached mode (mount with cache=none), we shunt off direct read and write requests
4438  * so this method should never be called.
4439  *
4440  * Direct IO is not yet supported in the cached mode.
4441  */
4442 static ssize_t
4443 cifs_direct_io(struct kiocb *iocb, struct iov_iter *iter)
4444 {
4445         /*
4446          * FIXME
4447          * Eventually need to support direct IO for non forcedirectio mounts
4448          */
4449         return -EINVAL;
4450 }
4451
4452
4453 const struct address_space_operations cifs_addr_ops = {
4454         .readpage = cifs_readpage,
4455         .readpages = cifs_readpages,
4456         .writepage = cifs_writepage,
4457         .writepages = cifs_writepages,
4458         .write_begin = cifs_write_begin,
4459         .write_end = cifs_write_end,
4460         .set_page_dirty = __set_page_dirty_nobuffers,
4461         .releasepage = cifs_release_page,
4462         .direct_IO = cifs_direct_io,
4463         .invalidatepage = cifs_invalidate_page,
4464         .launder_page = cifs_launder_page,
4465 };
4466
4467 /*
4468  * cifs_readpages requires the server to support a buffer large enough to
4469  * contain the header plus one complete page of data.  Otherwise, we need
4470  * to leave cifs_readpages out of the address space operations.
4471  */
4472 const struct address_space_operations cifs_addr_ops_smallbuf = {
4473         .readpage = cifs_readpage,
4474         .writepage = cifs_writepage,
4475         .writepages = cifs_writepages,
4476         .write_begin = cifs_write_begin,
4477         .write_end = cifs_write_end,
4478         .set_page_dirty = __set_page_dirty_nobuffers,
4479         .releasepage = cifs_release_page,
4480         .invalidatepage = cifs_invalidate_page,
4481         .launder_page = cifs_launder_page,
4482 };