git.oblomov.eu Git - linux-2.6/blob - fs/xfs/linux-2.6/xfs_lrw.c

   1 /*
   2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_bit.h"
  21 #include "xfs_log.h"
  22 #include "xfs_inum.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_dir2.h"
  27 #include "xfs_alloc.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_quota.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_alloc_btree.h"
  33 #include "xfs_ialloc_btree.h"
  34 #include "xfs_dir2_sf.h"
  35 #include "xfs_attr_sf.h"
  36 #include "xfs_dinode.h"
  37 #include "xfs_inode.h"
  38 #include "xfs_bmap.h"
  39 #include "xfs_btree.h"
  40 #include "xfs_ialloc.h"
  41 #include "xfs_rtalloc.h"
  42 #include "xfs_error.h"
  43 #include "xfs_itable.h"
  44 #include "xfs_rw.h"
  45 #include "xfs_acl.h"
  46 #include "xfs_attr.h"
  47 #include "xfs_inode_item.h"
  48 #include "xfs_buf_item.h"
  49 #include "xfs_utils.h"
  50 #include "xfs_iomap.h"
  51
  52 #include <linux/capability.h>
  53 #include <linux/writeback.h>
  54
  55
  56 #if defined(XFS_RW_TRACE)
  57 void
  58 xfs_rw_enter_trace(
  59         int                     tag,
  60         xfs_iocore_t            *io,
  61         void                    *data,
  62         size_t                  segs,
  63         loff_t                  offset,
  64         int                     ioflags)
  65 {
  66         xfs_inode_t     *ip = XFS_IO_INODE(io);
  67
  68         if (ip->i_rwtrace == NULL)
  69                 return;
  70         ktrace_enter(ip->i_rwtrace,
  71                 (void *)(unsigned long)tag,
  72                 (void *)ip,
  73                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
  74                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
  75                 (void *)data,
  76                 (void *)((unsigned long)segs),
  77                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
  78                 (void *)((unsigned long)(offset & 0xffffffff)),
  79                 (void *)((unsigned long)ioflags),
  80                 (void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
  81                 (void *)((unsigned long)(io->io_new_size & 0xffffffff)),
  82                 (void *)((unsigned long)current_pid()),
  83                 (void *)NULL,
  84                 (void *)NULL,
  85                 (void *)NULL,
  86                 (void *)NULL);
  87 }
  88
  89 void
  90 xfs_inval_cached_trace(
  91         xfs_iocore_t    *io,
  92         xfs_off_t       offset,
  93         xfs_off_t       len,
  94         xfs_off_t       first,
  95         xfs_off_t       last)
  96 {
  97         xfs_inode_t     *ip = XFS_IO_INODE(io);
  98
  99         if (ip->i_rwtrace == NULL)
 100                 return;
 101         ktrace_enter(ip->i_rwtrace,
 102                 (void *)(__psint_t)XFS_INVAL_CACHED,
 103                 (void *)ip,
 104                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
 105                 (void *)((unsigned long)(offset & 0xffffffff)),
 106                 (void *)((unsigned long)((len >> 32) & 0xffffffff)),
 107                 (void *)((unsigned long)(len & 0xffffffff)),
 108                 (void *)((unsigned long)((first >> 32) & 0xffffffff)),
 109                 (void *)((unsigned long)(first & 0xffffffff)),
 110                 (void *)((unsigned long)((last >> 32) & 0xffffffff)),
 111                 (void *)((unsigned long)(last & 0xffffffff)),
 112                 (void *)((unsigned long)current_pid()),
 113                 (void *)NULL,
 114                 (void *)NULL,
 115                 (void *)NULL,
 116                 (void *)NULL,
 117                 (void *)NULL);
 118 }
 119 #endif
 120
 121 /*
 122  *      xfs_iozero
 123  *
 124  *      xfs_iozero clears the specified range of buffer supplied,
 125  *      and marks all the affected blocks as valid and modified.  If
 126  *      an affected block is not allocated, it will be allocated.  If
 127  *      an affected block is not completely overwritten, and is not
 128  *      valid before the operation, it will be read from disk before
 129  *      being partially zeroed.
 130  */
 131 STATIC int
 132 xfs_iozero(
 133         struct inode            *ip,    /* inode                        */
 134         loff_t                  pos,    /* offset in file               */
 135         size_t                  count)  /* size of data to zero         */
 136 {
 137         struct page             *page;
 138         struct address_space    *mapping;
 139         int                     status;
 140
 141         mapping = ip->i_mapping;
 142         do {
 143                 unsigned offset, bytes;
 144                 void *fsdata;
 145
 146                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
 147                 bytes = PAGE_CACHE_SIZE - offset;
 148                 if (bytes > count)
 149                         bytes = count;
 150
 151                 status = pagecache_write_begin(NULL, mapping, pos, bytes,
 152                                         AOP_FLAG_UNINTERRUPTIBLE,
 153                                         &page, &fsdata);
 154                 if (status)
 155                         break;
 156
 157                 zero_user_page(page, offset, bytes, KM_USER0);
 158
 159                 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
 160                                         page, fsdata);
 161                 WARN_ON(status <= 0); /* can't return less than zero! */
 162                 pos += bytes;
 163                 count -= bytes;
 164                 status = 0;
 165         } while (count);
 166
 167         return (-status);
 168 }
 169
 170 ssize_t                 /* bytes read, or (-)  error */
 171 xfs_read(
 172         bhv_desc_t              *bdp,
 173         struct kiocb            *iocb,
 174         const struct iovec      *iovp,
 175         unsigned int            segs,
 176         loff_t                  *offset,
 177         int                     ioflags,
 178         cred_t                  *credp)
 179 {
 180         struct file             *file = iocb->ki_filp;
 181         struct inode            *inode = file->f_mapping->host;
 182         size_t                  size = 0;
 183         ssize_t                 ret = 0;
 184         xfs_fsize_t             n;
 185         xfs_inode_t             *ip;
 186         xfs_mount_t             *mp;
 187         bhv_vnode_t             *vp;
 188         unsigned long           seg;
 189
 190         ip = XFS_BHVTOI(bdp);
 191         vp = BHV_TO_VNODE(bdp);
 192         mp = ip->i_mount;
 193
 194         XFS_STATS_INC(xs_read_calls);
 195
 196         /* START copy & waste from filemap.c */
 197         for (seg = 0; seg < segs; seg++) {
 198                 const struct iovec *iv = &iovp[seg];
 199
 200                 /*
 201                  * If any segment has a negative length, or the cumulative
 202                  * length ever wraps negative then return -EINVAL.
 203                  */
 204                 size += iv->iov_len;
 205                 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
 206                         return XFS_ERROR(-EINVAL);
 207         }
 208         /* END copy & waste from filemap.c */
 209
 210         if (unlikely(ioflags & IO_ISDIRECT)) {
 211                 xfs_buftarg_t   *target =
 212                         (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
 213                                 mp->m_rtdev_targp : mp->m_ddev_targp;
 214                 if ((*offset & target->bt_smask) ||
 215                     (size & target->bt_smask)) {
 216                         if (*offset == ip->i_size) {
 217                                 return (0);
 218                         }
 219                         return -XFS_ERROR(EINVAL);
 220                 }
 221         }
 222
 223         n = XFS_MAXIOFFSET(mp) - *offset;
 224         if ((n <= 0) || (size == 0))
 225                 return 0;
 226
 227         if (n < size)
 228                 size = n;
 229
 230         if (XFS_FORCED_SHUTDOWN(mp))
 231                 return -EIO;
 232
 233         if (unlikely(ioflags & IO_ISDIRECT))
 234                 mutex_lock(&inode->i_mutex);
 235         xfs_ilock(ip, XFS_IOLOCK_SHARED);
 236
 237         if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) &&
 238             !(ioflags & IO_INVIS)) {
 239                 bhv_vrwlock_t locktype = VRWLOCK_READ;
 240                 int dmflags = FILP_DELAY_FLAG(file) | DM_SEM_FLAG_RD(ioflags);
 241
 242                 ret = -XFS_SEND_DATA(mp, DM_EVENT_READ,
 243                                         BHV_TO_VNODE(bdp), *offset, size,
 244                                         dmflags, &locktype);
 245                 if (ret) {
 246                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 247                         if (unlikely(ioflags & IO_ISDIRECT))
 248                                 mutex_unlock(&inode->i_mutex);
 249                         return ret;
 250                 }
 251         }
 252
 253         if (unlikely(ioflags & IO_ISDIRECT)) {
 254                 if (VN_CACHED(vp))
 255                         ret = bhv_vop_flushinval_pages(vp, ctooff(offtoct(*offset)),
 256                                                  -1, FI_REMAPF_LOCKED);
 257                 mutex_unlock(&inode->i_mutex);
 258                 if (ret) {
 259                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 260                         return ret;
 261                 }
 262         }
 263
 264         xfs_rw_enter_trace(XFS_READ_ENTER, &ip->i_iocore,
 265                                 (void *)iovp, segs, *offset, ioflags);
 266
 267         iocb->ki_pos = *offset;
 268         ret = generic_file_aio_read(iocb, iovp, segs, *offset);
 269         if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO))
 270                 ret = wait_on_sync_kiocb(iocb);
 271         if (ret > 0)
 272                 XFS_STATS_ADD(xs_read_bytes, ret);
 273
 274         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 275         return ret;
 276 }
 277
 278 ssize_t
 279 xfs_splice_read(
 280         bhv_desc_t              *bdp,
 281         struct file             *infilp,
 282         loff_t                  *ppos,
 283         struct pipe_inode_info  *pipe,
 284         size_t                  count,
 285         int                     flags,
 286         int                     ioflags,
 287         cred_t                  *credp)
 288 {
 289         xfs_inode_t             *ip = XFS_BHVTOI(bdp);
 290         xfs_mount_t             *mp = ip->i_mount;
 291         ssize_t                 ret;
 292
 293         XFS_STATS_INC(xs_read_calls);
 294         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 295                 return -EIO;
 296
 297         xfs_ilock(ip, XFS_IOLOCK_SHARED);
 298
 299         if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_READ) &&
 300             (!(ioflags & IO_INVIS))) {
 301                 bhv_vrwlock_t locktype = VRWLOCK_READ;
 302                 int error;
 303
 304                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, BHV_TO_VNODE(bdp),
 305                                         *ppos, count,
 306                                         FILP_DELAY_FLAG(infilp), &locktype);
 307                 if (error) {
 308                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 309                         return -error;
 310                 }
 311         }
 312         xfs_rw_enter_trace(XFS_SPLICE_READ_ENTER, &ip->i_iocore,
 313                            pipe, count, *ppos, ioflags);
 314         ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
 315         if (ret > 0)
 316                 XFS_STATS_ADD(xs_read_bytes, ret);
 317
 318         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 319         return ret;
 320 }
 321
 322 ssize_t
 323 xfs_splice_write(
 324         bhv_desc_t              *bdp,
 325         struct pipe_inode_info  *pipe,
 326         struct file             *outfilp,
 327         loff_t                  *ppos,
 328         size_t                  count,
 329         int                     flags,
 330         int                     ioflags,
 331         cred_t                  *credp)
 332 {
 333         xfs_inode_t             *ip = XFS_BHVTOI(bdp);
 334         xfs_mount_t             *mp = ip->i_mount;
 335         xfs_iocore_t            *io = &ip->i_iocore;
 336         ssize_t                 ret;
 337         struct inode            *inode = outfilp->f_mapping->host;
 338         xfs_fsize_t             isize, new_size;
 339
 340         XFS_STATS_INC(xs_write_calls);
 341         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 342                 return -EIO;
 343
 344         xfs_ilock(ip, XFS_IOLOCK_EXCL);
 345
 346         if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_WRITE) &&
 347             (!(ioflags & IO_INVIS))) {
 348                 bhv_vrwlock_t locktype = VRWLOCK_WRITE;
 349                 int error;
 350
 351                 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, BHV_TO_VNODE(bdp),
 352                                         *ppos, count,
 353                                         FILP_DELAY_FLAG(outfilp), &locktype);
 354                 if (error) {
 355                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 356                         return -error;
 357                 }
 358         }
 359
 360         new_size = *ppos + count;
 361
 362         xfs_ilock(ip, XFS_ILOCK_EXCL);
 363         if (new_size > ip->i_size)
 364                 io->io_new_size = new_size;
 365         xfs_iunlock(ip, XFS_ILOCK_EXCL);
 366
 367         xfs_rw_enter_trace(XFS_SPLICE_WRITE_ENTER, &ip->i_iocore,
 368                            pipe, count, *ppos, ioflags);
 369         ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
 370         if (ret > 0)
 371                 XFS_STATS_ADD(xs_write_bytes, ret);
 372
 373         isize = i_size_read(inode);
 374         if (unlikely(ret < 0 && ret != -EFAULT && *ppos > isize))
 375                 *ppos = isize;
 376
 377         if (*ppos > ip->i_size) {
 378                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 379                 if (*ppos > ip->i_size)
 380                         ip->i_size = *ppos;
 381                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 382         }
 383
 384         if (io->io_new_size) {
 385                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 386                 io->io_new_size = 0;
 387                 if (ip->i_d.di_size > ip->i_size)
 388                         ip->i_d.di_size = ip->i_size;
 389                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 390         }
 391         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 392         return ret;
 393 }
 394
 395 /*
 396  * This routine is called to handle zeroing any space in the last
 397  * block of the file that is beyond the EOF.  We do this since the
 398  * size is being increased without writing anything to that block
 399  * and we don't want anyone to read the garbage on the disk.
 400  */
 401 STATIC int                              /* error (positive) */
 402 xfs_zero_last_block(
 403         struct inode    *ip,
 404         xfs_iocore_t    *io,
 405         xfs_fsize_t     offset,
 406         xfs_fsize_t     isize)
 407 {
 408         xfs_fileoff_t   last_fsb;
 409         xfs_mount_t     *mp = io->io_mount;
 410         int             nimaps;
 411         int             zero_offset;
 412         int             zero_len;
 413         int             error = 0;
 414         xfs_bmbt_irec_t imap;
 415
 416         ASSERT(ismrlocked(io->io_lock, MR_UPDATE) != 0);
 417
 418         zero_offset = XFS_B_FSB_OFFSET(mp, isize);
 419         if (zero_offset == 0) {
 420                 /*
 421                  * There are no extra bytes in the last block on disk to
 422                  * zero, so return.
 423                  */
 424                 return 0;
 425         }
 426
 427         last_fsb = XFS_B_TO_FSBT(mp, isize);
 428         nimaps = 1;
 429         error = XFS_BMAPI(mp, NULL, io, last_fsb, 1, 0, NULL, 0, &imap,
 430                           &nimaps, NULL, NULL);
 431         if (error) {
 432                 return error;
 433         }
 434         ASSERT(nimaps > 0);
 435         /*
 436          * If the block underlying isize is just a hole, then there
 437          * is nothing to zero.
 438          */
 439         if (imap.br_startblock == HOLESTARTBLOCK) {
 440                 return 0;
 441         }
 442         /*
 443          * Zero the part of the last block beyond the EOF, and write it
 444          * out sync.  We need to drop the ilock while we do this so we
 445          * don't deadlock when the buffer cache calls back to us.
 446          */
 447         XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL| XFS_EXTSIZE_RD);
 448
 449         zero_len = mp->m_sb.sb_blocksize - zero_offset;
 450         if (isize + zero_len > offset)
 451                 zero_len = offset - isize;
 452         error = xfs_iozero(ip, isize, zero_len);
 453
 454         XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
 455         ASSERT(error >= 0);
 456         return error;
 457 }
 458
 459 /*
 460  * Zero any on disk space between the current EOF and the new,
 461  * larger EOF.  This handles the normal case of zeroing the remainder
 462  * of the last block in the file and the unusual case of zeroing blocks
 463  * out beyond the size of the file.  This second case only happens
 464  * with fixed size extents and when the system crashes before the inode
 465  * size was updated but after blocks were allocated.  If fill is set,
 466  * then any holes in the range are filled and zeroed.  If not, the holes
 467  * are left alone as holes.
 468  */
 469
 470 int                                     /* error (positive) */
 471 xfs_zero_eof(
 472         bhv_vnode_t     *vp,
 473         xfs_iocore_t    *io,
 474         xfs_off_t       offset,         /* starting I/O offset */
 475         xfs_fsize_t     isize)          /* current inode size */
 476 {
 477         struct inode    *ip = vn_to_inode(vp);
 478         xfs_fileoff_t   start_zero_fsb;
 479         xfs_fileoff_t   end_zero_fsb;
 480         xfs_fileoff_t   zero_count_fsb;
 481         xfs_fileoff_t   last_fsb;
 482         xfs_fileoff_t   zero_off;
 483         xfs_fsize_t     zero_len;
 484         xfs_mount_t     *mp = io->io_mount;
 485         int             nimaps;
 486         int             error = 0;
 487         xfs_bmbt_irec_t imap;
 488
 489         ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
 490         ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
 491         ASSERT(offset > isize);
 492
 493         /*
 494          * First handle zeroing the block on which isize resides.
 495          * We only zero a part of that block so it is handled specially.
 496          */
 497         error = xfs_zero_last_block(ip, io, offset, isize);
 498         if (error) {
 499                 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
 500                 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
 501                 return error;
 502         }
 503
 504         /*
 505          * Calculate the range between the new size and the old
 506          * where blocks needing to be zeroed may exist.  To get the
 507          * block where the last byte in the file currently resides,
 508          * we need to subtract one from the size and truncate back
 509          * to a block boundary.  We subtract 1 in case the size is
 510          * exactly on a block boundary.
 511          */
 512         last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
 513         start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
 514         end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
 515         ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
 516         if (last_fsb == end_zero_fsb) {
 517                 /*
 518                  * The size was only incremented on its last block.
 519                  * We took care of that above, so just return.
 520                  */
 521                 return 0;
 522         }
 523
 524         ASSERT(start_zero_fsb <= end_zero_fsb);
 525         while (start_zero_fsb <= end_zero_fsb) {
 526                 nimaps = 1;
 527                 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
 528                 error = XFS_BMAPI(mp, NULL, io, start_zero_fsb, zero_count_fsb,
 529                                   0, NULL, 0, &imap, &nimaps, NULL, NULL);
 530                 if (error) {
 531                         ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
 532                         ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
 533                         return error;
 534                 }
 535                 ASSERT(nimaps > 0);
 536
 537                 if (imap.br_state == XFS_EXT_UNWRITTEN ||
 538                     imap.br_startblock == HOLESTARTBLOCK) {
 539                         /*
 540                          * This loop handles initializing pages that were
 541                          * partially initialized by the code below this
 542                          * loop. It basically zeroes the part of the page
 543                          * that sits on a hole and sets the page as P_HOLE
 544                          * and calls remapf if it is a mapped file.
 545                          */
 546                         start_zero_fsb = imap.br_startoff + imap.br_blockcount;
 547                         ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 548                         continue;
 549                 }
 550
 551                 /*
 552                  * There are blocks we need to zero.
 553                  * Drop the inode lock while we're doing the I/O.
 554                  * We'll still have the iolock to protect us.
 555                  */
 556                 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
 557
 558                 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
 559                 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
 560
 561                 if ((zero_off + zero_len) > offset)
 562                         zero_len = offset - zero_off;
 563
 564                 error = xfs_iozero(ip, zero_off, zero_len);
 565                 if (error) {
 566                         goto out_lock;
 567                 }
 568
 569                 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
 570                 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 571
 572                 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
 573         }
 574
 575         return 0;
 576
 577 out_lock:
 578
 579         XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
 580         ASSERT(error >= 0);
 581         return error;
 582 }
 583
 584 ssize_t                         /* bytes written, or (-) error */
 585 xfs_write(
 586         bhv_desc_t              *bdp,
 587         struct kiocb            *iocb,
 588         const struct iovec      *iovp,
 589         unsigned int            nsegs,
 590         loff_t                  *offset,
 591         int                     ioflags,
 592         cred_t                  *credp)
 593 {
 594         struct file             *file = iocb->ki_filp;
 595         struct address_space    *mapping = file->f_mapping;
 596         struct inode            *inode = mapping->host;
 597         unsigned long           segs = nsegs;
 598         xfs_inode_t             *xip;
 599         xfs_mount_t             *mp;
 600         ssize_t                 ret = 0, error = 0;
 601         xfs_fsize_t             isize, new_size;
 602         xfs_iocore_t            *io;
 603         bhv_vnode_t             *vp;
 604         int                     iolock;
 605         int                     eventsent = 0;
 606         bhv_vrwlock_t           locktype;
 607         size_t                  ocount = 0, count;
 608         loff_t                  pos;
 609         int                     need_i_mutex;
 610
 611         XFS_STATS_INC(xs_write_calls);
 612
 613         vp = BHV_TO_VNODE(bdp);
 614         xip = XFS_BHVTOI(bdp);
 615
 616         error = generic_segment_checks(iovp, &segs, &ocount, VERIFY_READ);
 617         if (error)
 618                 return error;
 619
 620         count = ocount;
 621         pos = *offset;
 622
 623         if (count == 0)
 624                 return 0;
 625
 626         io = &xip->i_iocore;
 627         mp = io->io_mount;
 628
 629         vfs_wait_for_freeze(vp->v_vfsp, SB_FREEZE_WRITE);
 630
 631         if (XFS_FORCED_SHUTDOWN(mp))
 632                 return -EIO;
 633
 634 relock:
 635         if (ioflags & IO_ISDIRECT) {
 636                 iolock = XFS_IOLOCK_SHARED;
 637                 locktype = VRWLOCK_WRITE_DIRECT;
 638                 need_i_mutex = 0;
 639         } else {
 640                 iolock = XFS_IOLOCK_EXCL;
 641                 locktype = VRWLOCK_WRITE;
 642                 need_i_mutex = 1;
 643                 mutex_lock(&inode->i_mutex);
 644         }
 645
 646         xfs_ilock(xip, XFS_ILOCK_EXCL|iolock);
 647
 648 start:
 649         error = -generic_write_checks(file, &pos, &count,
 650                                         S_ISBLK(inode->i_mode));
 651         if (error) {
 652                 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
 653                 goto out_unlock_mutex;
 654         }
 655
 656         if ((DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_WRITE) &&
 657             !(ioflags & IO_INVIS) && !eventsent)) {
 658                 int             dmflags = FILP_DELAY_FLAG(file);
 659
 660                 if (need_i_mutex)
 661                         dmflags |= DM_FLAGS_IMUX;
 662
 663                 xfs_iunlock(xip, XFS_ILOCK_EXCL);
 664                 error = XFS_SEND_DATA(xip->i_mount, DM_EVENT_WRITE, vp,
 665                                       pos, count,
 666                                       dmflags, &locktype);
 667                 if (error) {
 668                         goto out_unlock_internal;
 669                 }
 670                 xfs_ilock(xip, XFS_ILOCK_EXCL);
 671                 eventsent = 1;
 672
 673                 /*
 674                  * The iolock was dropped and reacquired in XFS_SEND_DATA
 675                  * so we have to recheck the size when appending.
 676                  * We will only "goto start;" once, since having sent the
 677                  * event prevents another call to XFS_SEND_DATA, which is
 678                  * what allows the size to change in the first place.
 679                  */
 680                 if ((file->f_flags & O_APPEND) && pos != xip->i_size)
 681                         goto start;
 682         }
 683
 684         if (ioflags & IO_ISDIRECT) {
 685                 xfs_buftarg_t   *target =
 686                         (xip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
 687                                 mp->m_rtdev_targp : mp->m_ddev_targp;
 688
 689                 if ((pos & target->bt_smask) || (count & target->bt_smask)) {
 690                         xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
 691                         return XFS_ERROR(-EINVAL);
 692                 }
 693
 694                 if (!need_i_mutex && (VN_CACHED(vp) || pos > xip->i_size)) {
 695                         xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
 696                         iolock = XFS_IOLOCK_EXCL;
 697                         locktype = VRWLOCK_WRITE;
 698                         need_i_mutex = 1;
 699                         mutex_lock(&inode->i_mutex);
 700                         xfs_ilock(xip, XFS_ILOCK_EXCL|iolock);
 701                         goto start;
 702                 }
 703         }
 704
 705         new_size = pos + count;
 706         if (new_size > xip->i_size)
 707                 io->io_new_size = new_size;
 708
 709         if (likely(!(ioflags & IO_INVIS))) {
 710                 file_update_time(file);
 711                 xfs_ichgtime_fast(xip, inode,
 712                                   XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
 713         }
 714
 715         /*
 716          * If the offset is beyond the size of the file, we have a couple
 717          * of things to do. First, if there is already space allocated
 718          * we need to either create holes or zero the disk or ...
 719          *
 720          * If there is a page where the previous size lands, we need
 721          * to zero it out up to the new size.
 722          */
 723
 724         if (pos > xip->i_size) {
 725                 error = xfs_zero_eof(BHV_TO_VNODE(bdp), io, pos, xip->i_size);
 726                 if (error) {
 727                         xfs_iunlock(xip, XFS_ILOCK_EXCL);
 728                         goto out_unlock_internal;
 729                 }
 730         }
 731         xfs_iunlock(xip, XFS_ILOCK_EXCL);
 732
 733         /*
 734          * If we're writing the file then make sure to clear the
 735          * setuid and setgid bits if the process is not being run
 736          * by root.  This keeps people from modifying setuid and
 737          * setgid binaries.
 738          */
 739
 740         if (((xip->i_d.di_mode & S_ISUID) ||
 741             ((xip->i_d.di_mode & (S_ISGID | S_IXGRP)) ==
 742                 (S_ISGID | S_IXGRP))) &&
 743              !capable(CAP_FSETID)) {
 744                 error = xfs_write_clear_setuid(xip);
 745                 if (likely(!error))
 746                         error = -remove_suid(file->f_path.dentry);
 747                 if (unlikely(error)) {
 748                         goto out_unlock_internal;
 749                 }
 750         }
 751
 752 retry:
 753         /* We can write back this queue in page reclaim */
 754         current->backing_dev_info = mapping->backing_dev_info;
 755
 756         if ((ioflags & IO_ISDIRECT)) {
 757                 if (VN_CACHED(vp)) {
 758                         WARN_ON(need_i_mutex == 0);
 759                         xfs_inval_cached_trace(io, pos, -1,
 760                                         ctooff(offtoct(pos)), -1);
 761                         error = bhv_vop_flushinval_pages(vp, ctooff(offtoct(pos)),
 762                                         -1, FI_REMAPF_LOCKED);
 763                         if (error)
 764                                 goto out_unlock_internal;
 765                 }
 766
 767                 if (need_i_mutex) {
 768                         /* demote the lock now the cached pages are gone */
 769                         XFS_ILOCK_DEMOTE(mp, io, XFS_IOLOCK_EXCL);
 770                         mutex_unlock(&inode->i_mutex);
 771
 772                         iolock = XFS_IOLOCK_SHARED;
 773                         locktype = VRWLOCK_WRITE_DIRECT;
 774                         need_i_mutex = 0;
 775                 }
 776
 777                 xfs_rw_enter_trace(XFS_DIOWR_ENTER, io, (void *)iovp, segs,
 778                                 *offset, ioflags);
 779                 ret = generic_file_direct_write(iocb, iovp,
 780                                 &segs, pos, offset, count, ocount);
 781
 782                 /*
 783                  * direct-io write to a hole: fall through to buffered I/O
 784                  * for completing the rest of the request.
 785                  */
 786                 if (ret >= 0 && ret != count) {
 787                         XFS_STATS_ADD(xs_write_bytes, ret);
 788
 789                         pos += ret;
 790                         count -= ret;
 791
 792                         ioflags &= ~IO_ISDIRECT;
 793                         xfs_iunlock(xip, iolock);
 794                         goto relock;
 795                 }
 796         } else {
 797                 xfs_rw_enter_trace(XFS_WRITE_ENTER, io, (void *)iovp, segs,
 798                                 *offset, ioflags);
 799                 ret = generic_file_buffered_write(iocb, iovp, segs,
 800                                 pos, offset, count, ret);
 801         }
 802
 803         current->backing_dev_info = NULL;
 804
 805         if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO))
 806                 ret = wait_on_sync_kiocb(iocb);
 807
 808         if ((ret == -ENOSPC) &&
 809             DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_NOSPACE) &&
 810             !(ioflags & IO_INVIS)) {
 811
 812                 xfs_rwunlock(bdp, locktype);
 813                 if (need_i_mutex)
 814                         mutex_unlock(&inode->i_mutex);
 815                 error = XFS_SEND_NAMESP(xip->i_mount, DM_EVENT_NOSPACE, vp,
 816                                 DM_RIGHT_NULL, vp, DM_RIGHT_NULL, NULL, NULL,
 817                                 0, 0, 0); /* Delay flag intentionally  unused */
 818                 if (need_i_mutex)
 819                         mutex_lock(&inode->i_mutex);
 820                 xfs_rwlock(bdp, locktype);
 821                 if (error)
 822                         goto out_unlock_internal;
 823                 pos = xip->i_size;
 824                 ret = 0;
 825                 goto retry;
 826         }
 827
 828         isize = i_size_read(inode);
 829         if (unlikely(ret < 0 && ret != -EFAULT && *offset > isize))
 830                 *offset = isize;
 831
 832         if (*offset > xip->i_size) {
 833                 xfs_ilock(xip, XFS_ILOCK_EXCL);
 834                 if (*offset > xip->i_size)
 835                         xip->i_size = *offset;
 836                 xfs_iunlock(xip, XFS_ILOCK_EXCL);
 837         }
 838
 839         error = -ret;
 840         if (ret <= 0)
 841                 goto out_unlock_internal;
 842
 843         XFS_STATS_ADD(xs_write_bytes, ret);
 844
 845         /* Handle various SYNC-type writes */
 846         if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
 847                 error = xfs_write_sync_logforce(mp, xip);
 848                 if (error)
 849                         goto out_unlock_internal;
 850
 851                 xfs_rwunlock(bdp, locktype);
 852                 if (need_i_mutex)
 853                         mutex_unlock(&inode->i_mutex);
 854
 855                 error = sync_page_range(inode, mapping, pos, ret);
 856                 if (!error)
 857                         error = -ret;
 858                 if (need_i_mutex)
 859                         mutex_lock(&inode->i_mutex);
 860                 xfs_rwlock(bdp, locktype);
 861         }
 862
 863  out_unlock_internal:
 864         if (io->io_new_size) {
 865                 xfs_ilock(xip, XFS_ILOCK_EXCL);
 866                 io->io_new_size = 0;
 867                 /*
 868                  * If this was a direct or synchronous I/O that failed (such
 869                  * as ENOSPC) then part of the I/O may have been written to
 870                  * disk before the error occured.  In this case the on-disk
 871                  * file size may have been adjusted beyond the in-memory file
 872                  * size and now needs to be truncated back.
 873                  */
 874                 if (xip->i_d.di_size > xip->i_size)
 875                         xip->i_d.di_size = xip->i_size;
 876                 xfs_iunlock(xip, XFS_ILOCK_EXCL);
 877         }
 878         xfs_rwunlock(bdp, locktype);
 879  out_unlock_mutex:
 880         if (need_i_mutex)
 881                 mutex_unlock(&inode->i_mutex);
 882         return -error;
 883 }
 884
 885 /*
 886  * All xfs metadata buffers except log state machine buffers
 887  * get this attached as their b_bdstrat callback function.
 888  * This is so that we can catch a buffer
 889  * after prematurely unpinning it to forcibly shutdown the filesystem.
 890  */
 891 int
 892 xfs_bdstrat_cb(struct xfs_buf *bp)
 893 {
 894         xfs_mount_t     *mp;
 895
 896         mp = XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *);
 897         if (!XFS_FORCED_SHUTDOWN(mp)) {
 898                 xfs_buf_iorequest(bp);
 899                 return 0;
 900         } else {
 901                 xfs_buftrace("XFS__BDSTRAT IOERROR", bp);
 902                 /*
 903                  * Metadata write that didn't get logged but
 904                  * written delayed anyway. These aren't associated
 905                  * with a transaction, and can be ignored.
 906                  */
 907                 if (XFS_BUF_IODONE_FUNC(bp) == NULL &&
 908                     (XFS_BUF_ISREAD(bp)) == 0)
 909                         return (xfs_bioerror_relse(bp));
 910                 else
 911                         return (xfs_bioerror(bp));
 912         }
 913 }
 914
 915
 916 int
 917 xfs_bmap(bhv_desc_t     *bdp,
 918         xfs_off_t       offset,
 919         ssize_t         count,
 920         int             flags,
 921         xfs_iomap_t     *iomapp,
 922         int             *niomaps)
 923 {
 924         xfs_inode_t     *ip = XFS_BHVTOI(bdp);
 925         xfs_iocore_t    *io = &ip->i_iocore;
 926
 927         ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
 928         ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
 929                ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
 930
 931         return xfs_iomap(io, offset, count, flags, iomapp, niomaps);
 932 }
 933
 934 /*
 935  * Wrapper around bdstrat so that we can stop data
 936  * from going to disk in case we are shutting down the filesystem.
 937  * Typically user data goes thru this path; one of the exceptions
 938  * is the superblock.
 939  */
 940 int
 941 xfsbdstrat(
 942         struct xfs_mount        *mp,
 943         struct xfs_buf          *bp)
 944 {
 945         ASSERT(mp);
 946         if (!XFS_FORCED_SHUTDOWN(mp)) {
 947                 /* Grio redirection would go here
 948                  * if (XFS_BUF_IS_GRIO(bp)) {
 949                  */
 950
 951                 xfs_buf_iorequest(bp);
 952                 return 0;
 953         }
 954
 955         xfs_buftrace("XFSBDSTRAT IOERROR", bp);
 956         return (xfs_bioerror_relse(bp));
 957 }
 958
 959 /*
 960  * If the underlying (data/log/rt) device is readonly, there are some
 961  * operations that cannot proceed.
 962  */
 963 int
 964 xfs_dev_is_read_only(
 965         xfs_mount_t             *mp,
 966         char                    *message)
 967 {
 968         if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
 969             xfs_readonly_buftarg(mp->m_logdev_targp) ||
 970             (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
 971                 cmn_err(CE_NOTE,
 972                         "XFS: %s required on read-only device.", message);
 973                 cmn_err(CE_NOTE,
 974                         "XFS: write access unavailable, cannot proceed.");
 975                 return EROFS;
 976         }
 977         return 0;
 978 }