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_cap.h"
  47 #include "xfs_mac.h"
  48 #include "xfs_attr.h"
  49 #include "xfs_inode_item.h"
  50 #include "xfs_buf_item.h"
  51 #include "xfs_utils.h"
  52 #include "xfs_iomap.h"
  53
  54 #include <linux/capability.h>
  55 #include <linux/writeback.h>
  56
  57
  58 #if defined(XFS_RW_TRACE)
  59 void
  60 xfs_rw_enter_trace(
  61         int                     tag,
  62         xfs_iocore_t            *io,
  63         void                    *data,
  64         size_t                  segs,
  65         loff_t                  offset,
  66         int                     ioflags)
  67 {
  68         xfs_inode_t     *ip = XFS_IO_INODE(io);
  69
  70         if (ip->i_rwtrace == NULL)
  71                 return;
  72         ktrace_enter(ip->i_rwtrace,
  73                 (void *)(unsigned long)tag,
  74                 (void *)ip,
  75                 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
  76                 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
  77                 (void *)data,
  78                 (void *)((unsigned long)segs),
  79                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
  80                 (void *)((unsigned long)(offset & 0xffffffff)),
  81                 (void *)((unsigned long)ioflags),
  82                 (void *)((unsigned long)((io->io_new_size >> 32) & 0xffffffff)),
  83                 (void *)((unsigned long)(io->io_new_size & 0xffffffff)),
  84                 (void *)((unsigned long)current_pid()),
  85                 (void *)NULL,
  86                 (void *)NULL,
  87                 (void *)NULL,
  88                 (void *)NULL);
  89 }
  90
  91 void
  92 xfs_inval_cached_trace(
  93         xfs_iocore_t    *io,
  94         xfs_off_t       offset,
  95         xfs_off_t       len,
  96         xfs_off_t       first,
  97         xfs_off_t       last)
  98 {
  99         xfs_inode_t     *ip = XFS_IO_INODE(io);
 100
 101         if (ip->i_rwtrace == NULL)
 102                 return;
 103         ktrace_enter(ip->i_rwtrace,
 104                 (void *)(__psint_t)XFS_INVAL_CACHED,
 105                 (void *)ip,
 106                 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
 107                 (void *)((unsigned long)(offset & 0xffffffff)),
 108                 (void *)((unsigned long)((len >> 32) & 0xffffffff)),
 109                 (void *)((unsigned long)(len & 0xffffffff)),
 110                 (void *)((unsigned long)((first >> 32) & 0xffffffff)),
 111                 (void *)((unsigned long)(first & 0xffffffff)),
 112                 (void *)((unsigned long)((last >> 32) & 0xffffffff)),
 113                 (void *)((unsigned long)(last & 0xffffffff)),
 114                 (void *)((unsigned long)current_pid()),
 115                 (void *)NULL,
 116                 (void *)NULL,
 117                 (void *)NULL,
 118                 (void *)NULL,
 119                 (void *)NULL);
 120 }
 121 #endif
 122
 123 /*
 124  *      xfs_iozero
 125  *
 126  *      xfs_iozero clears the specified range of buffer supplied,
 127  *      and marks all the affected blocks as valid and modified.  If
 128  *      an affected block is not allocated, it will be allocated.  If
 129  *      an affected block is not completely overwritten, and is not
 130  *      valid before the operation, it will be read from disk before
 131  *      being partially zeroed.
 132  */
 133 STATIC int
 134 xfs_iozero(
 135         struct inode            *ip,    /* inode                        */
 136         loff_t                  pos,    /* offset in file               */
 137         size_t                  count,  /* size of data to zero         */
 138         loff_t                  end_size)       /* max file size to set */
 139 {
 140         unsigned                bytes;
 141         struct page             *page;
 142         struct address_space    *mapping;
 143         char                    *kaddr;
 144         int                     status;
 145
 146         mapping = ip->i_mapping;
 147         do {
 148                 unsigned long index, offset;
 149
 150                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
 151                 index = pos >> PAGE_CACHE_SHIFT;
 152                 bytes = PAGE_CACHE_SIZE - offset;
 153                 if (bytes > count)
 154                         bytes = count;
 155
 156                 status = -ENOMEM;
 157                 page = grab_cache_page(mapping, index);
 158                 if (!page)
 159                         break;
 160
 161                 kaddr = kmap(page);
 162                 status = mapping->a_ops->prepare_write(NULL, page, offset,
 163                                                         offset + bytes);
 164                 if (status) {
 165                         goto unlock;
 166                 }
 167
 168                 memset((void *) (kaddr + offset), 0, bytes);
 169                 flush_dcache_page(page);
 170                 status = mapping->a_ops->commit_write(NULL, page, offset,
 171                                                         offset + bytes);
 172                 if (!status) {
 173                         pos += bytes;
 174                         count -= bytes;
 175                         if (pos > i_size_read(ip))
 176                                 i_size_write(ip, pos < end_size ? pos : end_size);
 177                 }
 178
 179 unlock:
 180                 kunmap(page);
 181                 unlock_page(page);
 182                 page_cache_release(page);
 183                 if (status)
 184                         break;
 185         } while (count);
 186
 187         return (-status);
 188 }
 189
 190 ssize_t                 /* bytes read, or (-)  error */
 191 xfs_read(
 192         bhv_desc_t              *bdp,
 193         struct kiocb            *iocb,
 194         const struct iovec      *iovp,
 195         unsigned int            segs,
 196         loff_t                  *offset,
 197         int                     ioflags,
 198         cred_t                  *credp)
 199 {
 200         struct file             *file = iocb->ki_filp;
 201         struct inode            *inode = file->f_mapping->host;
 202         size_t                  size = 0;
 203         ssize_t                 ret;
 204         xfs_fsize_t             n;
 205         xfs_inode_t             *ip;
 206         xfs_mount_t             *mp;
 207         bhv_vnode_t             *vp;
 208         unsigned long           seg;
 209
 210         ip = XFS_BHVTOI(bdp);
 211         vp = BHV_TO_VNODE(bdp);
 212         mp = ip->i_mount;
 213
 214         XFS_STATS_INC(xs_read_calls);
 215
 216         /* START copy & waste from filemap.c */
 217         for (seg = 0; seg < segs; seg++) {
 218                 const struct iovec *iv = &iovp[seg];
 219
 220                 /*
 221                  * If any segment has a negative length, or the cumulative
 222                  * length ever wraps negative then return -EINVAL.
 223                  */
 224                 size += iv->iov_len;
 225                 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
 226                         return XFS_ERROR(-EINVAL);
 227         }
 228         /* END copy & waste from filemap.c */
 229
 230         if (unlikely(ioflags & IO_ISDIRECT)) {
 231                 xfs_buftarg_t   *target =
 232                         (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
 233                                 mp->m_rtdev_targp : mp->m_ddev_targp;
 234                 if ((*offset & target->bt_smask) ||
 235                     (size & target->bt_smask)) {
 236                         if (*offset == ip->i_d.di_size) {
 237                                 return (0);
 238                         }
 239                         return -XFS_ERROR(EINVAL);
 240                 }
 241         }
 242
 243         n = XFS_MAXIOFFSET(mp) - *offset;
 244         if ((n <= 0) || (size == 0))
 245                 return 0;
 246
 247         if (n < size)
 248                 size = n;
 249
 250         if (XFS_FORCED_SHUTDOWN(mp))
 251                 return -EIO;
 252
 253         if (unlikely(ioflags & IO_ISDIRECT))
 254                 mutex_lock(&inode->i_mutex);
 255         xfs_ilock(ip, XFS_IOLOCK_SHARED);
 256
 257         if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) &&
 258             !(ioflags & IO_INVIS)) {
 259                 bhv_vrwlock_t locktype = VRWLOCK_READ;
 260                 int dmflags = FILP_DELAY_FLAG(file) | DM_SEM_FLAG_RD(ioflags);
 261
 262                 ret = -XFS_SEND_DATA(mp, DM_EVENT_READ,
 263                                         BHV_TO_VNODE(bdp), *offset, size,
 264                                         dmflags, &locktype);
 265                 if (ret) {
 266                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 267                         if (unlikely(ioflags & IO_ISDIRECT))
 268                                 mutex_unlock(&inode->i_mutex);
 269                         return ret;
 270                 }
 271         }
 272
 273         if (unlikely((ioflags & IO_ISDIRECT) && VN_CACHED(vp)))
 274                 bhv_vop_flushinval_pages(vp, ctooff(offtoct(*offset)),
 275                                                 -1, FI_REMAPF_LOCKED);
 276
 277         if (unlikely(ioflags & IO_ISDIRECT))
 278                 mutex_unlock(&inode->i_mutex);
 279
 280         xfs_rw_enter_trace(XFS_READ_ENTER, &ip->i_iocore,
 281                                 (void *)iovp, segs, *offset, ioflags);
 282         ret = __generic_file_aio_read(iocb, iovp, segs, offset);
 283         if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO))
 284                 ret = wait_on_sync_kiocb(iocb);
 285         if (ret > 0)
 286                 XFS_STATS_ADD(xs_read_bytes, ret);
 287
 288         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 289         return ret;
 290 }
 291
 292 ssize_t
 293 xfs_sendfile(
 294         bhv_desc_t              *bdp,
 295         struct file             *filp,
 296         loff_t                  *offset,
 297         int                     ioflags,
 298         size_t                  count,
 299         read_actor_t            actor,
 300         void                    *target,
 301         cred_t                  *credp)
 302 {
 303         xfs_inode_t             *ip = XFS_BHVTOI(bdp);
 304         xfs_mount_t             *mp = ip->i_mount;
 305         ssize_t                 ret;
 306
 307         XFS_STATS_INC(xs_read_calls);
 308         if (XFS_FORCED_SHUTDOWN(mp))
 309                 return -EIO;
 310
 311         xfs_ilock(ip, XFS_IOLOCK_SHARED);
 312
 313         if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_READ) &&
 314             (!(ioflags & IO_INVIS))) {
 315                 bhv_vrwlock_t locktype = VRWLOCK_READ;
 316                 int error;
 317
 318                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, BHV_TO_VNODE(bdp),
 319                                       *offset, count,
 320                                       FILP_DELAY_FLAG(filp), &locktype);
 321                 if (error) {
 322                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 323                         return -error;
 324                 }
 325         }
 326         xfs_rw_enter_trace(XFS_SENDFILE_ENTER, &ip->i_iocore,
 327                    (void *)(unsigned long)target, count, *offset, ioflags);
 328         ret = generic_file_sendfile(filp, offset, count, actor, target);
 329         if (ret > 0)
 330                 XFS_STATS_ADD(xs_read_bytes, ret);
 331
 332         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 333         return ret;
 334 }
 335
 336 ssize_t
 337 xfs_splice_read(
 338         bhv_desc_t              *bdp,
 339         struct file             *infilp,
 340         loff_t                  *ppos,
 341         struct pipe_inode_info  *pipe,
 342         size_t                  count,
 343         int                     flags,
 344         int                     ioflags,
 345         cred_t                  *credp)
 346 {
 347         xfs_inode_t             *ip = XFS_BHVTOI(bdp);
 348         xfs_mount_t             *mp = ip->i_mount;
 349         ssize_t                 ret;
 350
 351         XFS_STATS_INC(xs_read_calls);
 352         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 353                 return -EIO;
 354
 355         xfs_ilock(ip, XFS_IOLOCK_SHARED);
 356
 357         if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_READ) &&
 358             (!(ioflags & IO_INVIS))) {
 359                 bhv_vrwlock_t locktype = VRWLOCK_READ;
 360                 int error;
 361
 362                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, BHV_TO_VNODE(bdp),
 363                                         *ppos, count,
 364                                         FILP_DELAY_FLAG(infilp), &locktype);
 365                 if (error) {
 366                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 367                         return -error;
 368                 }
 369         }
 370         xfs_rw_enter_trace(XFS_SPLICE_READ_ENTER, &ip->i_iocore,
 371                            pipe, count, *ppos, ioflags);
 372         ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
 373         if (ret > 0)
 374                 XFS_STATS_ADD(xs_read_bytes, ret);
 375
 376         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 377         return ret;
 378 }
 379
 380 ssize_t
 381 xfs_splice_write(
 382         bhv_desc_t              *bdp,
 383         struct pipe_inode_info  *pipe,
 384         struct file             *outfilp,
 385         loff_t                  *ppos,
 386         size_t                  count,
 387         int                     flags,
 388         int                     ioflags,
 389         cred_t                  *credp)
 390 {
 391         xfs_inode_t             *ip = XFS_BHVTOI(bdp);
 392         xfs_mount_t             *mp = ip->i_mount;
 393         ssize_t                 ret;
 394
 395         XFS_STATS_INC(xs_write_calls);
 396         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 397                 return -EIO;
 398
 399         xfs_ilock(ip, XFS_IOLOCK_EXCL);
 400
 401         if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_WRITE) &&
 402             (!(ioflags & IO_INVIS))) {
 403                 bhv_vrwlock_t locktype = VRWLOCK_WRITE;
 404                 int error;
 405
 406                 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, BHV_TO_VNODE(bdp),
 407                                         *ppos, count,
 408                                         FILP_DELAY_FLAG(outfilp), &locktype);
 409                 if (error) {
 410                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 411                         return -error;
 412                 }
 413         }
 414         xfs_rw_enter_trace(XFS_SPLICE_WRITE_ENTER, &ip->i_iocore,
 415                            pipe, count, *ppos, ioflags);
 416         ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
 417         if (ret > 0)
 418                 XFS_STATS_ADD(xs_write_bytes, ret);
 419
 420         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 421         return ret;
 422 }
 423
 424 /*
 425  * This routine is called to handle zeroing any space in the last
 426  * block of the file that is beyond the EOF.  We do this since the
 427  * size is being increased without writing anything to that block
 428  * and we don't want anyone to read the garbage on the disk.
 429  */
 430 STATIC int                              /* error (positive) */
 431 xfs_zero_last_block(
 432         struct inode    *ip,
 433         xfs_iocore_t    *io,
 434         xfs_fsize_t     isize,
 435         xfs_fsize_t     end_size)
 436 {
 437         xfs_fileoff_t   last_fsb;
 438         xfs_mount_t     *mp = io->io_mount;
 439         int             nimaps;
 440         int             zero_offset;
 441         int             zero_len;
 442         int             error = 0;
 443         xfs_bmbt_irec_t imap;
 444         loff_t          loff;
 445
 446         ASSERT(ismrlocked(io->io_lock, MR_UPDATE) != 0);
 447
 448         zero_offset = XFS_B_FSB_OFFSET(mp, isize);
 449         if (zero_offset == 0) {
 450                 /*
 451                  * There are no extra bytes in the last block on disk to
 452                  * zero, so return.
 453                  */
 454                 return 0;
 455         }
 456
 457         last_fsb = XFS_B_TO_FSBT(mp, isize);
 458         nimaps = 1;
 459         error = XFS_BMAPI(mp, NULL, io, last_fsb, 1, 0, NULL, 0, &imap,
 460                           &nimaps, NULL, NULL);
 461         if (error) {
 462                 return error;
 463         }
 464         ASSERT(nimaps > 0);
 465         /*
 466          * If the block underlying isize is just a hole, then there
 467          * is nothing to zero.
 468          */
 469         if (imap.br_startblock == HOLESTARTBLOCK) {
 470                 return 0;
 471         }
 472         /*
 473          * Zero the part of the last block beyond the EOF, and write it
 474          * out sync.  We need to drop the ilock while we do this so we
 475          * don't deadlock when the buffer cache calls back to us.
 476          */
 477         XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL| XFS_EXTSIZE_RD);
 478
 479         loff = XFS_FSB_TO_B(mp, last_fsb);
 480         zero_len = mp->m_sb.sb_blocksize - zero_offset;
 481         error = xfs_iozero(ip, loff + zero_offset, zero_len, end_size);
 482
 483         XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
 484         ASSERT(error >= 0);
 485         return error;
 486 }
 487
 488 /*
 489  * Zero any on disk space between the current EOF and the new,
 490  * larger EOF.  This handles the normal case of zeroing the remainder
 491  * of the last block in the file and the unusual case of zeroing blocks
 492  * out beyond the size of the file.  This second case only happens
 493  * with fixed size extents and when the system crashes before the inode
 494  * size was updated but after blocks were allocated.  If fill is set,
 495  * then any holes in the range are filled and zeroed.  If not, the holes
 496  * are left alone as holes.
 497  */
 498
 499 int                                     /* error (positive) */
 500 xfs_zero_eof(
 501         bhv_vnode_t     *vp,
 502         xfs_iocore_t    *io,
 503         xfs_off_t       offset,         /* starting I/O offset */
 504         xfs_fsize_t     isize,          /* current inode size */
 505         xfs_fsize_t     end_size)       /* terminal inode size */
 506 {
 507         struct inode    *ip = vn_to_inode(vp);
 508         xfs_fileoff_t   start_zero_fsb;
 509         xfs_fileoff_t   end_zero_fsb;
 510         xfs_fileoff_t   zero_count_fsb;
 511         xfs_fileoff_t   last_fsb;
 512         xfs_mount_t     *mp = io->io_mount;
 513         int             nimaps;
 514         int             error = 0;
 515         xfs_bmbt_irec_t imap;
 516
 517         ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
 518         ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
 519         ASSERT(offset > isize);
 520
 521         /*
 522          * First handle zeroing the block on which isize resides.
 523          * We only zero a part of that block so it is handled specially.
 524          */
 525         error = xfs_zero_last_block(ip, io, isize, end_size);
 526         if (error) {
 527                 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
 528                 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
 529                 return error;
 530         }
 531
 532         /*
 533          * Calculate the range between the new size and the old
 534          * where blocks needing to be zeroed may exist.  To get the
 535          * block where the last byte in the file currently resides,
 536          * we need to subtract one from the size and truncate back
 537          * to a block boundary.  We subtract 1 in case the size is
 538          * exactly on a block boundary.
 539          */
 540         last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
 541         start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
 542         end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
 543         ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
 544         if (last_fsb == end_zero_fsb) {
 545                 /*
 546                  * The size was only incremented on its last block.
 547                  * We took care of that above, so just return.
 548                  */
 549                 return 0;
 550         }
 551
 552         ASSERT(start_zero_fsb <= end_zero_fsb);
 553         while (start_zero_fsb <= end_zero_fsb) {
 554                 nimaps = 1;
 555                 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
 556                 error = XFS_BMAPI(mp, NULL, io, start_zero_fsb, zero_count_fsb,
 557                                   0, NULL, 0, &imap, &nimaps, NULL, NULL);
 558                 if (error) {
 559                         ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
 560                         ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
 561                         return error;
 562                 }
 563                 ASSERT(nimaps > 0);
 564
 565                 if (imap.br_state == XFS_EXT_UNWRITTEN ||
 566                     imap.br_startblock == HOLESTARTBLOCK) {
 567                         /*
 568                          * This loop handles initializing pages that were
 569                          * partially initialized by the code below this
 570                          * loop. It basically zeroes the part of the page
 571                          * that sits on a hole and sets the page as P_HOLE
 572                          * and calls remapf if it is a mapped file.
 573                          */
 574                         start_zero_fsb = imap.br_startoff + imap.br_blockcount;
 575                         ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 576                         continue;
 577                 }
 578
 579                 /*
 580                  * There are blocks we need to zero.
 581                  * Drop the inode lock while we're doing the I/O.
 582                  * We'll still have the iolock to protect us.
 583                  */
 584                 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
 585
 586                 error = xfs_iozero(ip,
 587                                    XFS_FSB_TO_B(mp, start_zero_fsb),
 588                                    XFS_FSB_TO_B(mp, imap.br_blockcount),
 589                                    end_size);
 590                 if (error) {
 591                         goto out_lock;
 592                 }
 593
 594                 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
 595                 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 596
 597                 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
 598         }
 599
 600         return 0;
 601
 602 out_lock:
 603
 604         XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
 605         ASSERT(error >= 0);
 606         return error;
 607 }
 608
 609 ssize_t                         /* bytes written, or (-) error */
 610 xfs_write(
 611         bhv_desc_t              *bdp,
 612         struct kiocb            *iocb,
 613         const struct iovec      *iovp,
 614         unsigned int            nsegs,
 615         loff_t                  *offset,
 616         int                     ioflags,
 617         cred_t                  *credp)
 618 {
 619         struct file             *file = iocb->ki_filp;
 620         struct address_space    *mapping = file->f_mapping;
 621         struct inode            *inode = mapping->host;
 622         unsigned long           segs = nsegs;
 623         xfs_inode_t             *xip;
 624         xfs_mount_t             *mp;
 625         ssize_t                 ret = 0, error = 0;
 626         xfs_fsize_t             isize, new_size;
 627         xfs_iocore_t            *io;
 628         bhv_vnode_t             *vp;
 629         unsigned long           seg;
 630         int                     iolock;
 631         int                     eventsent = 0;
 632         bhv_vrwlock_t           locktype;
 633         size_t                  ocount = 0, count;
 634         loff_t                  pos;
 635         int                     need_i_mutex = 1, need_flush = 0;
 636
 637         XFS_STATS_INC(xs_write_calls);
 638
 639         vp = BHV_TO_VNODE(bdp);
 640         xip = XFS_BHVTOI(bdp);
 641
 642         for (seg = 0; seg < segs; seg++) {
 643                 const struct iovec *iv = &iovp[seg];
 644
 645                 /*
 646                  * If any segment has a negative length, or the cumulative
 647                  * length ever wraps negative then return -EINVAL.
 648                  */
 649                 ocount += iv->iov_len;
 650                 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
 651                         return -EINVAL;
 652                 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
 653                         continue;
 654                 if (seg == 0)
 655                         return -EFAULT;
 656                 segs = seg;
 657                 ocount -= iv->iov_len;  /* This segment is no good */
 658                 break;
 659         }
 660
 661         count = ocount;
 662         pos = *offset;
 663
 664         if (count == 0)
 665                 return 0;
 666
 667         io = &xip->i_iocore;
 668         mp = io->io_mount;
 669
 670         vfs_wait_for_freeze(vp->v_vfsp, SB_FREEZE_WRITE);
 671
 672         if (XFS_FORCED_SHUTDOWN(mp))
 673                 return -EIO;
 674
 675         if (ioflags & IO_ISDIRECT) {
 676                 xfs_buftarg_t   *target =
 677                         (xip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
 678                                 mp->m_rtdev_targp : mp->m_ddev_targp;
 679
 680                 if ((pos & target->bt_smask) || (count & target->bt_smask))
 681                         return XFS_ERROR(-EINVAL);
 682
 683                 if (!VN_CACHED(vp) && pos < i_size_read(inode))
 684                         need_i_mutex = 0;
 685
 686                 if (VN_CACHED(vp))
 687                         need_flush = 1;
 688         }
 689
 690 relock:
 691         if (need_i_mutex) {
 692                 iolock = XFS_IOLOCK_EXCL;
 693                 locktype = VRWLOCK_WRITE;
 694
 695                 mutex_lock(&inode->i_mutex);
 696         } else {
 697                 iolock = XFS_IOLOCK_SHARED;
 698                 locktype = VRWLOCK_WRITE_DIRECT;
 699         }
 700
 701         xfs_ilock(xip, XFS_ILOCK_EXCL|iolock);
 702
 703         isize = i_size_read(inode);
 704
 705         if (file->f_flags & O_APPEND)
 706                 *offset = isize;
 707
 708 start:
 709         error = -generic_write_checks(file, &pos, &count,
 710                                         S_ISBLK(inode->i_mode));
 711         if (error) {
 712                 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
 713                 goto out_unlock_mutex;
 714         }
 715
 716         new_size = pos + count;
 717         if (new_size > isize)
 718                 io->io_new_size = new_size;
 719
 720         if ((DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_WRITE) &&
 721             !(ioflags & IO_INVIS) && !eventsent)) {
 722                 loff_t          savedsize = pos;
 723                 int             dmflags = FILP_DELAY_FLAG(file);
 724
 725                 if (need_i_mutex)
 726                         dmflags |= DM_FLAGS_IMUX;
 727
 728                 xfs_iunlock(xip, XFS_ILOCK_EXCL);
 729                 error = XFS_SEND_DATA(xip->i_mount, DM_EVENT_WRITE, vp,
 730                                       pos, count,
 731                                       dmflags, &locktype);
 732                 if (error) {
 733                         xfs_iunlock(xip, iolock);
 734                         goto out_unlock_mutex;
 735                 }
 736                 xfs_ilock(xip, XFS_ILOCK_EXCL);
 737                 eventsent = 1;
 738
 739                 /*
 740                  * The iolock was dropped and reacquired in XFS_SEND_DATA
 741                  * so we have to recheck the size when appending.
 742                  * We will only "goto start;" once, since having sent the
 743                  * event prevents another call to XFS_SEND_DATA, which is
 744                  * what allows the size to change in the first place.
 745                  */
 746                 if ((file->f_flags & O_APPEND) && savedsize != isize) {
 747                         pos = isize = xip->i_d.di_size;
 748                         goto start;
 749                 }
 750         }
 751
 752         if (likely(!(ioflags & IO_INVIS))) {
 753                 file_update_time(file);
 754                 xfs_ichgtime_fast(xip, inode,
 755                                   XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
 756         }
 757
 758         /*
 759          * If the offset is beyond the size of the file, we have a couple
 760          * of things to do. First, if there is already space allocated
 761          * we need to either create holes or zero the disk or ...
 762          *
 763          * If there is a page where the previous size lands, we need
 764          * to zero it out up to the new size.
 765          */
 766
 767         if (pos > isize) {
 768                 error = xfs_zero_eof(BHV_TO_VNODE(bdp), io, pos,
 769                                         isize, pos + count);
 770                 if (error) {
 771                         xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
 772                         goto out_unlock_mutex;
 773                 }
 774         }
 775         xfs_iunlock(xip, XFS_ILOCK_EXCL);
 776
 777         /*
 778          * If we're writing the file then make sure to clear the
 779          * setuid and setgid bits if the process is not being run
 780          * by root.  This keeps people from modifying setuid and
 781          * setgid binaries.
 782          */
 783
 784         if (((xip->i_d.di_mode & S_ISUID) ||
 785             ((xip->i_d.di_mode & (S_ISGID | S_IXGRP)) ==
 786                 (S_ISGID | S_IXGRP))) &&
 787              !capable(CAP_FSETID)) {
 788                 error = xfs_write_clear_setuid(xip);
 789                 if (likely(!error))
 790                         error = -remove_suid(file->f_dentry);
 791                 if (unlikely(error)) {
 792                         xfs_iunlock(xip, iolock);
 793                         goto out_unlock_mutex;
 794                 }
 795         }
 796
 797 retry:
 798         /* We can write back this queue in page reclaim */
 799         current->backing_dev_info = mapping->backing_dev_info;
 800
 801         if ((ioflags & IO_ISDIRECT)) {
 802                 if (need_flush) {
 803                         xfs_inval_cached_trace(io, pos, -1,
 804                                         ctooff(offtoct(pos)), -1);
 805                         bhv_vop_flushinval_pages(vp, ctooff(offtoct(pos)),
 806                                         -1, FI_REMAPF_LOCKED);
 807                 }
 808
 809                 if (need_i_mutex) {
 810                         /* demote the lock now the cached pages are gone */
 811                         XFS_ILOCK_DEMOTE(mp, io, XFS_IOLOCK_EXCL);
 812                         mutex_unlock(&inode->i_mutex);
 813
 814                         iolock = XFS_IOLOCK_SHARED;
 815                         locktype = VRWLOCK_WRITE_DIRECT;
 816                         need_i_mutex = 0;
 817                 }
 818
 819                 xfs_rw_enter_trace(XFS_DIOWR_ENTER, io, (void *)iovp, segs,
 820                                 *offset, ioflags);
 821                 ret = generic_file_direct_write(iocb, iovp,
 822                                 &segs, pos, offset, count, ocount);
 823
 824                 /*
 825                  * direct-io write to a hole: fall through to buffered I/O
 826                  * for completing the rest of the request.
 827                  */
 828                 if (ret >= 0 && ret != count) {
 829                         XFS_STATS_ADD(xs_write_bytes, ret);
 830
 831                         pos += ret;
 832                         count -= ret;
 833
 834                         need_i_mutex = 1;
 835                         ioflags &= ~IO_ISDIRECT;
 836                         xfs_iunlock(xip, iolock);
 837                         goto relock;
 838                 }
 839         } else {
 840                 xfs_rw_enter_trace(XFS_WRITE_ENTER, io, (void *)iovp, segs,
 841                                 *offset, ioflags);
 842                 ret = generic_file_buffered_write(iocb, iovp, segs,
 843                                 pos, offset, count, ret);
 844         }
 845
 846         current->backing_dev_info = NULL;
 847
 848         if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO))
 849                 ret = wait_on_sync_kiocb(iocb);
 850
 851         if ((ret == -ENOSPC) &&
 852             DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_NOSPACE) &&
 853             !(ioflags & IO_INVIS)) {
 854
 855                 xfs_rwunlock(bdp, locktype);
 856                 if (need_i_mutex)
 857                         mutex_unlock(&inode->i_mutex);
 858                 error = XFS_SEND_NAMESP(xip->i_mount, DM_EVENT_NOSPACE, vp,
 859                                 DM_RIGHT_NULL, vp, DM_RIGHT_NULL, NULL, NULL,
 860                                 0, 0, 0); /* Delay flag intentionally  unused */
 861                 if (error)
 862                         goto out_nounlocks;
 863                 if (need_i_mutex)
 864                         mutex_lock(&inode->i_mutex);
 865                 xfs_rwlock(bdp, locktype);
 866                 pos = xip->i_d.di_size;
 867                 ret = 0;
 868                 goto retry;
 869         }
 870
 871         isize = i_size_read(inode);
 872         if (unlikely(ret < 0 && ret != -EFAULT && *offset > isize))
 873                 *offset = isize;
 874
 875         if (*offset > xip->i_d.di_size) {
 876                 xfs_ilock(xip, XFS_ILOCK_EXCL);
 877                 if (*offset > xip->i_d.di_size) {
 878                         xip->i_d.di_size = *offset;
 879                         i_size_write(inode, *offset);
 880                         xip->i_update_core = 1;
 881                         xip->i_update_size = 1;
 882                 }
 883                 xfs_iunlock(xip, XFS_ILOCK_EXCL);
 884         }
 885
 886         error = -ret;
 887         if (ret <= 0)
 888                 goto out_unlock_internal;
 889
 890         XFS_STATS_ADD(xs_write_bytes, ret);
 891
 892         /* Handle various SYNC-type writes */
 893         if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
 894                 error = xfs_write_sync_logforce(mp, xip);
 895                 if (error)
 896                         goto out_unlock_internal;
 897
 898                 xfs_rwunlock(bdp, locktype);
 899                 if (need_i_mutex)
 900                         mutex_unlock(&inode->i_mutex);
 901
 902                 error = sync_page_range(inode, mapping, pos, ret);
 903                 if (!error)
 904                         error = ret;
 905                 return error;
 906         }
 907
 908  out_unlock_internal:
 909         xfs_rwunlock(bdp, locktype);
 910  out_unlock_mutex:
 911         if (need_i_mutex)
 912                 mutex_unlock(&inode->i_mutex);
 913  out_nounlocks:
 914         return -error;
 915 }
 916
 917 /*
 918  * All xfs metadata buffers except log state machine buffers
 919  * get this attached as their b_bdstrat callback function.
 920  * This is so that we can catch a buffer
 921  * after prematurely unpinning it to forcibly shutdown the filesystem.
 922  */
 923 int
 924 xfs_bdstrat_cb(struct xfs_buf *bp)
 925 {
 926         xfs_mount_t     *mp;
 927
 928         mp = XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *);
 929         if (!XFS_FORCED_SHUTDOWN(mp)) {
 930                 xfs_buf_iorequest(bp);
 931                 return 0;
 932         } else {
 933                 xfs_buftrace("XFS__BDSTRAT IOERROR", bp);
 934                 /*
 935                  * Metadata write that didn't get logged but
 936                  * written delayed anyway. These aren't associated
 937                  * with a transaction, and can be ignored.
 938                  */
 939                 if (XFS_BUF_IODONE_FUNC(bp) == NULL &&
 940                     (XFS_BUF_ISREAD(bp)) == 0)
 941                         return (xfs_bioerror_relse(bp));
 942                 else
 943                         return (xfs_bioerror(bp));
 944         }
 945 }
 946
 947
 948 int
 949 xfs_bmap(bhv_desc_t     *bdp,
 950         xfs_off_t       offset,
 951         ssize_t         count,
 952         int             flags,
 953         xfs_iomap_t     *iomapp,
 954         int             *niomaps)
 955 {
 956         xfs_inode_t     *ip = XFS_BHVTOI(bdp);
 957         xfs_iocore_t    *io = &ip->i_iocore;
 958
 959         ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
 960         ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
 961                ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
 962
 963         return xfs_iomap(io, offset, count, flags, iomapp, niomaps);
 964 }
 965
 966 /*
 967  * Wrapper around bdstrat so that we can stop data
 968  * from going to disk in case we are shutting down the filesystem.
 969  * Typically user data goes thru this path; one of the exceptions
 970  * is the superblock.
 971  */
 972 int
 973 xfsbdstrat(
 974         struct xfs_mount        *mp,
 975         struct xfs_buf          *bp)
 976 {
 977         ASSERT(mp);
 978         if (!XFS_FORCED_SHUTDOWN(mp)) {
 979                 /* Grio redirection would go here
 980                  * if (XFS_BUF_IS_GRIO(bp)) {
 981                  */
 982
 983                 xfs_buf_iorequest(bp);
 984                 return 0;
 985         }
 986
 987         xfs_buftrace("XFSBDSTRAT IOERROR", bp);
 988         return (xfs_bioerror_relse(bp));
 989 }
 990
 991 /*
 992  * If the underlying (data/log/rt) device is readonly, there are some
 993  * operations that cannot proceed.
 994  */
 995 int
 996 xfs_dev_is_read_only(
 997         xfs_mount_t             *mp,
 998         char                    *message)
 999 {
1000         if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1001             xfs_readonly_buftarg(mp->m_logdev_targp) ||
1002             (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1003                 cmn_err(CE_NOTE,
1004                         "XFS: %s required on read-only device.", message);
1005                 cmn_err(CE_NOTE,
1006                         "XFS: write access unavailable, cannot proceed.");
1007                 return EROFS;
1008         }
1009         return 0;
1010 }