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