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)) {
 274                 if (VN_CACHED(vp))
 275                         bhv_vop_flushinval_pages(vp, ctooff(offtoct(*offset)),
 276                                                  -1, FI_REMAPF_LOCKED);
 277                 mutex_unlock(&inode->i_mutex);
 278         }
 279
 280         xfs_rw_enter_trace(XFS_READ_ENTER, &ip->i_iocore,
 281                                 (void *)iovp, segs, *offset, ioflags);
 282
 283         iocb->ki_pos = *offset;
 284         ret = generic_file_aio_read(iocb, iovp, segs, *offset);
 285         if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO))
 286                 ret = wait_on_sync_kiocb(iocb);
 287         if (ret > 0)
 288                 XFS_STATS_ADD(xs_read_bytes, ret);
 289
 290         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 291         return ret;
 292 }
 293
 294 ssize_t
 295 xfs_sendfile(
 296         bhv_desc_t              *bdp,
 297         struct file             *filp,
 298         loff_t                  *offset,
 299         int                     ioflags,
 300         size_t                  count,
 301         read_actor_t            actor,
 302         void                    *target,
 303         cred_t                  *credp)
 304 {
 305         xfs_inode_t             *ip = XFS_BHVTOI(bdp);
 306         xfs_mount_t             *mp = ip->i_mount;
 307         ssize_t                 ret;
 308
 309         XFS_STATS_INC(xs_read_calls);
 310         if (XFS_FORCED_SHUTDOWN(mp))
 311                 return -EIO;
 312
 313         xfs_ilock(ip, XFS_IOLOCK_SHARED);
 314
 315         if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_READ) &&
 316             (!(ioflags & IO_INVIS))) {
 317                 bhv_vrwlock_t locktype = VRWLOCK_READ;
 318                 int error;
 319
 320                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, BHV_TO_VNODE(bdp),
 321                                       *offset, count,
 322                                       FILP_DELAY_FLAG(filp), &locktype);
 323                 if (error) {
 324                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 325                         return -error;
 326                 }
 327         }
 328         xfs_rw_enter_trace(XFS_SENDFILE_ENTER, &ip->i_iocore,
 329                    (void *)(unsigned long)target, count, *offset, ioflags);
 330         ret = generic_file_sendfile(filp, offset, count, actor, target);
 331         if (ret > 0)
 332                 XFS_STATS_ADD(xs_read_bytes, ret);
 333
 334         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 335         return ret;
 336 }
 337
 338 ssize_t
 339 xfs_splice_read(
 340         bhv_desc_t              *bdp,
 341         struct file             *infilp,
 342         loff_t                  *ppos,
 343         struct pipe_inode_info  *pipe,
 344         size_t                  count,
 345         int                     flags,
 346         int                     ioflags,
 347         cred_t                  *credp)
 348 {
 349         xfs_inode_t             *ip = XFS_BHVTOI(bdp);
 350         xfs_mount_t             *mp = ip->i_mount;
 351         ssize_t                 ret;
 352
 353         XFS_STATS_INC(xs_read_calls);
 354         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 355                 return -EIO;
 356
 357         xfs_ilock(ip, XFS_IOLOCK_SHARED);
 358
 359         if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_READ) &&
 360             (!(ioflags & IO_INVIS))) {
 361                 bhv_vrwlock_t locktype = VRWLOCK_READ;
 362                 int error;
 363
 364                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, BHV_TO_VNODE(bdp),
 365                                         *ppos, count,
 366                                         FILP_DELAY_FLAG(infilp), &locktype);
 367                 if (error) {
 368                         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 369                         return -error;
 370                 }
 371         }
 372         xfs_rw_enter_trace(XFS_SPLICE_READ_ENTER, &ip->i_iocore,
 373                            pipe, count, *ppos, ioflags);
 374         ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
 375         if (ret > 0)
 376                 XFS_STATS_ADD(xs_read_bytes, ret);
 377
 378         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 379         return ret;
 380 }
 381
 382 ssize_t
 383 xfs_splice_write(
 384         bhv_desc_t              *bdp,
 385         struct pipe_inode_info  *pipe,
 386         struct file             *outfilp,
 387         loff_t                  *ppos,
 388         size_t                  count,
 389         int                     flags,
 390         int                     ioflags,
 391         cred_t                  *credp)
 392 {
 393         xfs_inode_t             *ip = XFS_BHVTOI(bdp);
 394         xfs_mount_t             *mp = ip->i_mount;
 395         ssize_t                 ret;
 396         struct inode            *inode = outfilp->f_mapping->host;
 397         xfs_fsize_t             isize;
 398
 399         XFS_STATS_INC(xs_write_calls);
 400         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 401                 return -EIO;
 402
 403         xfs_ilock(ip, XFS_IOLOCK_EXCL);
 404
 405         if (DM_EVENT_ENABLED(BHV_TO_VNODE(bdp)->v_vfsp, ip, DM_EVENT_WRITE) &&
 406             (!(ioflags & IO_INVIS))) {
 407                 bhv_vrwlock_t locktype = VRWLOCK_WRITE;
 408                 int error;
 409
 410                 error = XFS_SEND_DATA(mp, DM_EVENT_WRITE, BHV_TO_VNODE(bdp),
 411                                         *ppos, count,
 412                                         FILP_DELAY_FLAG(outfilp), &locktype);
 413                 if (error) {
 414                         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 415                         return -error;
 416                 }
 417         }
 418         xfs_rw_enter_trace(XFS_SPLICE_WRITE_ENTER, &ip->i_iocore,
 419                            pipe, count, *ppos, ioflags);
 420         ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
 421         if (ret > 0)
 422                 XFS_STATS_ADD(xs_write_bytes, ret);
 423
 424         isize = i_size_read(inode);
 425         if (unlikely(ret < 0 && ret != -EFAULT && *ppos > isize))
 426                 *ppos = isize;
 427
 428         if (*ppos > ip->i_d.di_size) {
 429                 xfs_ilock(ip, XFS_ILOCK_EXCL);
 430                 if (*ppos > ip->i_d.di_size) {
 431                         ip->i_d.di_size = *ppos;
 432                         i_size_write(inode, *ppos);
 433                         ip->i_update_core = 1;
 434                         ip->i_update_size = 1;
 435                 }
 436                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
 437         }
 438         xfs_iunlock(ip, XFS_IOLOCK_EXCL);
 439         return ret;
 440 }
 441
 442 /*
 443  * This routine is called to handle zeroing any space in the last
 444  * block of the file that is beyond the EOF.  We do this since the
 445  * size is being increased without writing anything to that block
 446  * and we don't want anyone to read the garbage on the disk.
 447  */
 448 STATIC int                              /* error (positive) */
 449 xfs_zero_last_block(
 450         struct inode    *ip,
 451         xfs_iocore_t    *io,
 452         xfs_fsize_t     isize,
 453         xfs_fsize_t     end_size)
 454 {
 455         xfs_fileoff_t   last_fsb;
 456         xfs_mount_t     *mp = io->io_mount;
 457         int             nimaps;
 458         int             zero_offset;
 459         int             zero_len;
 460         int             error = 0;
 461         xfs_bmbt_irec_t imap;
 462         loff_t          loff;
 463
 464         ASSERT(ismrlocked(io->io_lock, MR_UPDATE) != 0);
 465
 466         zero_offset = XFS_B_FSB_OFFSET(mp, isize);
 467         if (zero_offset == 0) {
 468                 /*
 469                  * There are no extra bytes in the last block on disk to
 470                  * zero, so return.
 471                  */
 472                 return 0;
 473         }
 474
 475         last_fsb = XFS_B_TO_FSBT(mp, isize);
 476         nimaps = 1;
 477         error = XFS_BMAPI(mp, NULL, io, last_fsb, 1, 0, NULL, 0, &imap,
 478                           &nimaps, NULL, NULL);
 479         if (error) {
 480                 return error;
 481         }
 482         ASSERT(nimaps > 0);
 483         /*
 484          * If the block underlying isize is just a hole, then there
 485          * is nothing to zero.
 486          */
 487         if (imap.br_startblock == HOLESTARTBLOCK) {
 488                 return 0;
 489         }
 490         /*
 491          * Zero the part of the last block beyond the EOF, and write it
 492          * out sync.  We need to drop the ilock while we do this so we
 493          * don't deadlock when the buffer cache calls back to us.
 494          */
 495         XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL| XFS_EXTSIZE_RD);
 496
 497         loff = XFS_FSB_TO_B(mp, last_fsb);
 498         zero_len = mp->m_sb.sb_blocksize - zero_offset;
 499         error = xfs_iozero(ip, loff + zero_offset, zero_len, end_size);
 500
 501         XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
 502         ASSERT(error >= 0);
 503         return error;
 504 }
 505
 506 /*
 507  * Zero any on disk space between the current EOF and the new,
 508  * larger EOF.  This handles the normal case of zeroing the remainder
 509  * of the last block in the file and the unusual case of zeroing blocks
 510  * out beyond the size of the file.  This second case only happens
 511  * with fixed size extents and when the system crashes before the inode
 512  * size was updated but after blocks were allocated.  If fill is set,
 513  * then any holes in the range are filled and zeroed.  If not, the holes
 514  * are left alone as holes.
 515  */
 516
 517 int                                     /* error (positive) */
 518 xfs_zero_eof(
 519         bhv_vnode_t     *vp,
 520         xfs_iocore_t    *io,
 521         xfs_off_t       offset,         /* starting I/O offset */
 522         xfs_fsize_t     isize,          /* current inode size */
 523         xfs_fsize_t     end_size)       /* terminal inode size */
 524 {
 525         struct inode    *ip = vn_to_inode(vp);
 526         xfs_fileoff_t   start_zero_fsb;
 527         xfs_fileoff_t   end_zero_fsb;
 528         xfs_fileoff_t   zero_count_fsb;
 529         xfs_fileoff_t   last_fsb;
 530         xfs_mount_t     *mp = io->io_mount;
 531         int             nimaps;
 532         int             error = 0;
 533         xfs_bmbt_irec_t imap;
 534
 535         ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
 536         ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
 537         ASSERT(offset > isize);
 538
 539         /*
 540          * First handle zeroing the block on which isize resides.
 541          * We only zero a part of that block so it is handled specially.
 542          */
 543         error = xfs_zero_last_block(ip, io, isize, end_size);
 544         if (error) {
 545                 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
 546                 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
 547                 return error;
 548         }
 549
 550         /*
 551          * Calculate the range between the new size and the old
 552          * where blocks needing to be zeroed may exist.  To get the
 553          * block where the last byte in the file currently resides,
 554          * we need to subtract one from the size and truncate back
 555          * to a block boundary.  We subtract 1 in case the size is
 556          * exactly on a block boundary.
 557          */
 558         last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
 559         start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
 560         end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
 561         ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
 562         if (last_fsb == end_zero_fsb) {
 563                 /*
 564                  * The size was only incremented on its last block.
 565                  * We took care of that above, so just return.
 566                  */
 567                 return 0;
 568         }
 569
 570         ASSERT(start_zero_fsb <= end_zero_fsb);
 571         while (start_zero_fsb <= end_zero_fsb) {
 572                 nimaps = 1;
 573                 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
 574                 error = XFS_BMAPI(mp, NULL, io, start_zero_fsb, zero_count_fsb,
 575                                   0, NULL, 0, &imap, &nimaps, NULL, NULL);
 576                 if (error) {
 577                         ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
 578                         ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
 579                         return error;
 580                 }
 581                 ASSERT(nimaps > 0);
 582
 583                 if (imap.br_state == XFS_EXT_UNWRITTEN ||
 584                     imap.br_startblock == HOLESTARTBLOCK) {
 585                         /*
 586                          * This loop handles initializing pages that were
 587                          * partially initialized by the code below this
 588                          * loop. It basically zeroes the part of the page
 589                          * that sits on a hole and sets the page as P_HOLE
 590                          * and calls remapf if it is a mapped file.
 591                          */
 592                         start_zero_fsb = imap.br_startoff + imap.br_blockcount;
 593                         ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 594                         continue;
 595                 }
 596
 597                 /*
 598                  * There are blocks we need to zero.
 599                  * Drop the inode lock while we're doing the I/O.
 600                  * We'll still have the iolock to protect us.
 601                  */
 602                 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
 603
 604                 error = xfs_iozero(ip,
 605                                    XFS_FSB_TO_B(mp, start_zero_fsb),
 606                                    XFS_FSB_TO_B(mp, imap.br_blockcount),
 607                                    end_size);
 608                 if (error) {
 609                         goto out_lock;
 610                 }
 611
 612                 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
 613                 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
 614
 615                 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
 616         }
 617
 618         return 0;
 619
 620 out_lock:
 621
 622         XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
 623         ASSERT(error >= 0);
 624         return error;
 625 }
 626
 627 ssize_t                         /* bytes written, or (-) error */
 628 xfs_write(
 629         bhv_desc_t              *bdp,
 630         struct kiocb            *iocb,
 631         const struct iovec      *iovp,
 632         unsigned int            nsegs,
 633         loff_t                  *offset,
 634         int                     ioflags,
 635         cred_t                  *credp)
 636 {
 637         struct file             *file = iocb->ki_filp;
 638         struct address_space    *mapping = file->f_mapping;
 639         struct inode            *inode = mapping->host;
 640         unsigned long           segs = nsegs;
 641         xfs_inode_t             *xip;
 642         xfs_mount_t             *mp;
 643         ssize_t                 ret = 0, error = 0;
 644         xfs_fsize_t             isize, new_size;
 645         xfs_iocore_t            *io;
 646         bhv_vnode_t             *vp;
 647         unsigned long           seg;
 648         int                     iolock;
 649         int                     eventsent = 0;
 650         bhv_vrwlock_t           locktype;
 651         size_t                  ocount = 0, count;
 652         loff_t                  pos;
 653         int                     need_i_mutex = 1, need_flush = 0;
 654
 655         XFS_STATS_INC(xs_write_calls);
 656
 657         vp = BHV_TO_VNODE(bdp);
 658         xip = XFS_BHVTOI(bdp);
 659
 660         for (seg = 0; seg < segs; seg++) {
 661                 const struct iovec *iv = &iovp[seg];
 662
 663                 /*
 664                  * If any segment has a negative length, or the cumulative
 665                  * length ever wraps negative then return -EINVAL.
 666                  */
 667                 ocount += iv->iov_len;
 668                 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0))
 669                         return -EINVAL;
 670                 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
 671                         continue;
 672                 if (seg == 0)
 673                         return -EFAULT;
 674                 segs = seg;
 675                 ocount -= iv->iov_len;  /* This segment is no good */
 676                 break;
 677         }
 678
 679         count = ocount;
 680         pos = *offset;
 681
 682         if (count == 0)
 683                 return 0;
 684
 685         io = &xip->i_iocore;
 686         mp = io->io_mount;
 687
 688         vfs_wait_for_freeze(vp->v_vfsp, SB_FREEZE_WRITE);
 689
 690         if (XFS_FORCED_SHUTDOWN(mp))
 691                 return -EIO;
 692
 693         if (ioflags & IO_ISDIRECT) {
 694                 xfs_buftarg_t   *target =
 695                         (xip->i_d.di_flags & XFS_DIFLAG_REALTIME) ?
 696                                 mp->m_rtdev_targp : mp->m_ddev_targp;
 697
 698                 if ((pos & target->bt_smask) || (count & target->bt_smask))
 699                         return XFS_ERROR(-EINVAL);
 700
 701                 if (!VN_CACHED(vp) && pos < i_size_read(inode))
 702                         need_i_mutex = 0;
 703
 704                 if (VN_CACHED(vp))
 705                         need_flush = 1;
 706         }
 707
 708 relock:
 709         if (need_i_mutex) {
 710                 iolock = XFS_IOLOCK_EXCL;
 711                 locktype = VRWLOCK_WRITE;
 712
 713                 mutex_lock(&inode->i_mutex);
 714         } else {
 715                 iolock = XFS_IOLOCK_SHARED;
 716                 locktype = VRWLOCK_WRITE_DIRECT;
 717         }
 718
 719         xfs_ilock(xip, XFS_ILOCK_EXCL|iolock);
 720
 721         isize = i_size_read(inode);
 722
 723         if (file->f_flags & O_APPEND)
 724                 *offset = isize;
 725
 726 start:
 727         error = -generic_write_checks(file, &pos, &count,
 728                                         S_ISBLK(inode->i_mode));
 729         if (error) {
 730                 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
 731                 goto out_unlock_mutex;
 732         }
 733
 734         new_size = pos + count;
 735         if (new_size > isize)
 736                 io->io_new_size = new_size;
 737
 738         if ((DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_WRITE) &&
 739             !(ioflags & IO_INVIS) && !eventsent)) {
 740                 loff_t          savedsize = pos;
 741                 int             dmflags = FILP_DELAY_FLAG(file);
 742
 743                 if (need_i_mutex)
 744                         dmflags |= DM_FLAGS_IMUX;
 745
 746                 xfs_iunlock(xip, XFS_ILOCK_EXCL);
 747                 error = XFS_SEND_DATA(xip->i_mount, DM_EVENT_WRITE, vp,
 748                                       pos, count,
 749                                       dmflags, &locktype);
 750                 if (error) {
 751                         xfs_iunlock(xip, iolock);
 752                         goto out_unlock_mutex;
 753                 }
 754                 xfs_ilock(xip, XFS_ILOCK_EXCL);
 755                 eventsent = 1;
 756
 757                 /*
 758                  * The iolock was dropped and reacquired in XFS_SEND_DATA
 759                  * so we have to recheck the size when appending.
 760                  * We will only "goto start;" once, since having sent the
 761                  * event prevents another call to XFS_SEND_DATA, which is
 762                  * what allows the size to change in the first place.
 763                  */
 764                 if ((file->f_flags & O_APPEND) && savedsize != isize) {
 765                         pos = isize = xip->i_d.di_size;
 766                         goto start;
 767                 }
 768         }
 769
 770         if (likely(!(ioflags & IO_INVIS))) {
 771                 file_update_time(file);
 772                 xfs_ichgtime_fast(xip, inode,
 773                                   XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
 774         }
 775
 776         /*
 777          * If the offset is beyond the size of the file, we have a couple
 778          * of things to do. First, if there is already space allocated
 779          * we need to either create holes or zero the disk or ...
 780          *
 781          * If there is a page where the previous size lands, we need
 782          * to zero it out up to the new size.
 783          */
 784
 785         if (pos > isize) {
 786                 error = xfs_zero_eof(BHV_TO_VNODE(bdp), io, pos,
 787                                         isize, pos + count);
 788                 if (error) {
 789                         xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
 790                         goto out_unlock_mutex;
 791                 }
 792         }
 793         xfs_iunlock(xip, XFS_ILOCK_EXCL);
 794
 795         /*
 796          * If we're writing the file then make sure to clear the
 797          * setuid and setgid bits if the process is not being run
 798          * by root.  This keeps people from modifying setuid and
 799          * setgid binaries.
 800          */
 801
 802         if (((xip->i_d.di_mode & S_ISUID) ||
 803             ((xip->i_d.di_mode & (S_ISGID | S_IXGRP)) ==
 804                 (S_ISGID | S_IXGRP))) &&
 805              !capable(CAP_FSETID)) {
 806                 error = xfs_write_clear_setuid(xip);
 807                 if (likely(!error))
 808                         error = -remove_suid(file->f_dentry);
 809                 if (unlikely(error)) {
 810                         xfs_iunlock(xip, iolock);
 811                         goto out_unlock_mutex;
 812                 }
 813         }
 814
 815 retry:
 816         /* We can write back this queue in page reclaim */
 817         current->backing_dev_info = mapping->backing_dev_info;
 818
 819         if ((ioflags & IO_ISDIRECT)) {
 820                 if (need_flush) {
 821                         xfs_inval_cached_trace(io, pos, -1,
 822                                         ctooff(offtoct(pos)), -1);
 823                         bhv_vop_flushinval_pages(vp, ctooff(offtoct(pos)),
 824                                         -1, FI_REMAPF_LOCKED);
 825                 }
 826
 827                 if (need_i_mutex) {
 828                         /* demote the lock now the cached pages are gone */
 829                         XFS_ILOCK_DEMOTE(mp, io, XFS_IOLOCK_EXCL);
 830                         mutex_unlock(&inode->i_mutex);
 831
 832                         iolock = XFS_IOLOCK_SHARED;
 833                         locktype = VRWLOCK_WRITE_DIRECT;
 834                         need_i_mutex = 0;
 835                 }
 836
 837                 xfs_rw_enter_trace(XFS_DIOWR_ENTER, io, (void *)iovp, segs,
 838                                 *offset, ioflags);
 839                 ret = generic_file_direct_write(iocb, iovp,
 840                                 &segs, pos, offset, count, ocount);
 841
 842                 /*
 843                  * direct-io write to a hole: fall through to buffered I/O
 844                  * for completing the rest of the request.
 845                  */
 846                 if (ret >= 0 && ret != count) {
 847                         XFS_STATS_ADD(xs_write_bytes, ret);
 848
 849                         pos += ret;
 850                         count -= ret;
 851
 852                         need_i_mutex = 1;
 853                         ioflags &= ~IO_ISDIRECT;
 854                         xfs_iunlock(xip, iolock);
 855                         goto relock;
 856                 }
 857         } else {
 858                 xfs_rw_enter_trace(XFS_WRITE_ENTER, io, (void *)iovp, segs,
 859                                 *offset, ioflags);
 860                 ret = generic_file_buffered_write(iocb, iovp, segs,
 861                                 pos, offset, count, ret);
 862         }
 863
 864         current->backing_dev_info = NULL;
 865
 866         if (ret == -EIOCBQUEUED && !(ioflags & IO_ISAIO))
 867                 ret = wait_on_sync_kiocb(iocb);
 868
 869         if ((ret == -ENOSPC) &&
 870             DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_NOSPACE) &&
 871             !(ioflags & IO_INVIS)) {
 872
 873                 xfs_rwunlock(bdp, locktype);
 874                 if (need_i_mutex)
 875                         mutex_unlock(&inode->i_mutex);
 876                 error = XFS_SEND_NAMESP(xip->i_mount, DM_EVENT_NOSPACE, vp,
 877                                 DM_RIGHT_NULL, vp, DM_RIGHT_NULL, NULL, NULL,
 878                                 0, 0, 0); /* Delay flag intentionally  unused */
 879                 if (error)
 880                         goto out_nounlocks;
 881                 if (need_i_mutex)
 882                         mutex_lock(&inode->i_mutex);
 883                 xfs_rwlock(bdp, locktype);
 884                 pos = xip->i_d.di_size;
 885                 ret = 0;
 886                 goto retry;
 887         }
 888
 889         isize = i_size_read(inode);
 890         if (unlikely(ret < 0 && ret != -EFAULT && *offset > isize))
 891                 *offset = isize;
 892
 893         if (*offset > xip->i_d.di_size) {
 894                 xfs_ilock(xip, XFS_ILOCK_EXCL);
 895                 if (*offset > xip->i_d.di_size) {
 896                         xip->i_d.di_size = *offset;
 897                         i_size_write(inode, *offset);
 898                         xip->i_update_core = 1;
 899                         xip->i_update_size = 1;
 900                 }
 901                 xfs_iunlock(xip, XFS_ILOCK_EXCL);
 902         }
 903
 904         error = -ret;
 905         if (ret <= 0)
 906                 goto out_unlock_internal;
 907
 908         XFS_STATS_ADD(xs_write_bytes, ret);
 909
 910         /* Handle various SYNC-type writes */
 911         if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
 912                 error = xfs_write_sync_logforce(mp, xip);
 913                 if (error)
 914                         goto out_unlock_internal;
 915
 916                 xfs_rwunlock(bdp, locktype);
 917                 if (need_i_mutex)
 918                         mutex_unlock(&inode->i_mutex);
 919
 920                 error = sync_page_range(inode, mapping, pos, ret);
 921                 if (!error)
 922                         error = ret;
 923                 return error;
 924         }
 925
 926  out_unlock_internal:
 927         xfs_rwunlock(bdp, locktype);
 928  out_unlock_mutex:
 929         if (need_i_mutex)
 930                 mutex_unlock(&inode->i_mutex);
 931  out_nounlocks:
 932         return -error;
 933 }
 934
 935 /*
 936  * All xfs metadata buffers except log state machine buffers
 937  * get this attached as their b_bdstrat callback function.
 938  * This is so that we can catch a buffer
 939  * after prematurely unpinning it to forcibly shutdown the filesystem.
 940  */
 941 int
 942 xfs_bdstrat_cb(struct xfs_buf *bp)
 943 {
 944         xfs_mount_t     *mp;
 945
 946         mp = XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *);
 947         if (!XFS_FORCED_SHUTDOWN(mp)) {
 948                 xfs_buf_iorequest(bp);
 949                 return 0;
 950         } else {
 951                 xfs_buftrace("XFS__BDSTRAT IOERROR", bp);
 952                 /*
 953                  * Metadata write that didn't get logged but
 954                  * written delayed anyway. These aren't associated
 955                  * with a transaction, and can be ignored.
 956                  */
 957                 if (XFS_BUF_IODONE_FUNC(bp) == NULL &&
 958                     (XFS_BUF_ISREAD(bp)) == 0)
 959                         return (xfs_bioerror_relse(bp));
 960                 else
 961                         return (xfs_bioerror(bp));
 962         }
 963 }
 964
 965
 966 int
 967 xfs_bmap(bhv_desc_t     *bdp,
 968         xfs_off_t       offset,
 969         ssize_t         count,
 970         int             flags,
 971         xfs_iomap_t     *iomapp,
 972         int             *niomaps)
 973 {
 974         xfs_inode_t     *ip = XFS_BHVTOI(bdp);
 975         xfs_iocore_t    *io = &ip->i_iocore;
 976
 977         ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
 978         ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
 979                ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
 980
 981         return xfs_iomap(io, offset, count, flags, iomapp, niomaps);
 982 }
 983
 984 /*
 985  * Wrapper around bdstrat so that we can stop data
 986  * from going to disk in case we are shutting down the filesystem.
 987  * Typically user data goes thru this path; one of the exceptions
 988  * is the superblock.
 989  */
 990 int
 991 xfsbdstrat(
 992         struct xfs_mount        *mp,
 993         struct xfs_buf          *bp)
 994 {
 995         ASSERT(mp);
 996         if (!XFS_FORCED_SHUTDOWN(mp)) {
 997                 /* Grio redirection would go here
 998                  * if (XFS_BUF_IS_GRIO(bp)) {
 999                  */
1000
1001                 xfs_buf_iorequest(bp);
1002                 return 0;
1003         }
1004
1005         xfs_buftrace("XFSBDSTRAT IOERROR", bp);
1006         return (xfs_bioerror_relse(bp));
1007 }
1008
1009 /*
1010  * If the underlying (data/log/rt) device is readonly, there are some
1011  * operations that cannot proceed.
1012  */
1013 int
1014 xfs_dev_is_read_only(
1015         xfs_mount_t             *mp,
1016         char                    *message)
1017 {
1018         if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1019             xfs_readonly_buftarg(mp->m_logdev_targp) ||
1020             (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1021                 cmn_err(CE_NOTE,
1022                         "XFS: %s required on read-only device.", message);
1023                 cmn_err(CE_NOTE,
1024                         "XFS: write access unavailable, cannot proceed.");
1025                 return EROFS;
1026         }
1027         return 0;
1028 }