File: [Development] / xfs-linux / xfs_rw.c (download)
Revision 1.256, Sun Jul 12 03:57:42 1998 UTC (19 years, 3 months ago) by bbowen
Branch: MAIN
Changes since 1.255: +975 -96
lines
566573 - add support for file i/o using recursively
mmapped buffers to the buffered read/write paths.
|
#ident "$Revision$"
#include <sys/param.h>
#include <sys/buf.h>
#include <sys/uio.h>
#include <sys/vfs.h>
#include <sys/vnode.h>
#include <sys/cred.h>
#include <sys/sysmacros.h>
#include <sys/pfdat.h>
#include <sys/uuid.h>
#include <sys/major.h>
#include <sys/grio.h>
#include <sys/pda.h>
#include <sys/dmi_kern.h>
#include <sys/cmn_err.h>
#include <sys/debug.h>
#include <sys/errno.h>
#include <sys/fcntl.h>
#include <sys/var.h>
#include <sys/conf.h>
#include <sys/systm.h>
#include <sys/uthread.h>
#include <ksys/as.h>
#include <sys/kmem.h>
#include <sys/sema.h>
#include <ksys/vfile.h>
#include <sys/flock.h>
#include <sys/fs_subr.h>
#include <sys/dmi.h>
#include <sys/dmi_kern.h>
#include <sys/schedctl.h>
#include <sys/atomic_ops.h>
#include <sys/ktrace.h>
#include <sys/sysinfo.h>
#include <sys/ksa.h>
#include <ksys/sthread.h>
#include "xfs_macros.h"
#include "xfs_types.h"
#include "xfs_inum.h"
#include "xfs_log.h"
#include "xfs_trans.h"
#include "xfs_sb.h"
#include "xfs_ag.h"
#include "xfs_mount.h"
#include "xfs_alloc_btree.h"
#include "xfs_bmap_btree.h"
#include "xfs_ialloc_btree.h"
#include "xfs_itable.h"
#include "xfs_btree.h"
#include "xfs_alloc.h"
#include "xfs_bmap.h"
#include "xfs_ialloc.h"
#include "xfs_attr_sf.h"
#include "xfs_dir_sf.h"
#include "xfs_dinode.h"
#include "xfs_inode_item.h"
#include "xfs_inode.h"
#include "xfs_error.h"
#include "xfs_bit.h"
#include "xfs_rw.h"
#include "xfs_quota.h"
#include "xfs_trans_space.h"
#include "xfs_dmapi.h"
#include <limits.h>
/*
* turning on UIOSZ_DEBUG in a DEBUG kernel causes each xfs_write/xfs_read
* to set the write/read i/o size to a random valid value and instruments
* the distribution.
*
#define UIOSZ_DEBUG
*/
#ifdef UIOSZ_DEBUG
int uiodbg = 0;
int uiodbg_readiolog[XFS_UIO_MAX_READIO_LOG - XFS_UIO_MIN_READIO_LOG + 1] =
{0, 0, 0, 0};
int uiodbg_writeiolog[XFS_UIO_MAX_WRITEIO_LOG - XFS_UIO_MIN_WRITEIO_LOG + 1] =
{0, 0, 0, 0};
int uiodbg_switch = 0;
#endif
extern int xfs_nfs_io_units;
/*
* This lock is used by xfs_strat_write().
* The xfs_strat_lock is initialized in xfs_init().
*/
lock_t xfs_strat_lock;
/*
* Variables for coordination with the xfs_strat daemon.
* The xfsc_lock and xfsc_wait variables are initialized
* in xfs_init();
*/
static int xfsc_count;
static buf_t *xfsc_list;
static int xfsc_bufcount;
lock_t xfsc_lock;
sv_t xfsc_wait;
/*
* Variables for coordination with the xfsd daemons.
* The xfsd_lock and xfsd_wait variables are initialized
* in xfs_init();
*/
static int xfsd_count;
static buf_t *xfsd_list;
static int xfsd_bufcount;
lock_t xfsd_lock;
sv_t xfsd_wait;
/*
* Zone allocator for xfs_gap_t structures.
*/
zone_t *xfs_gap_zone;
#ifdef DEBUG
/*
* Global trace buffer for xfs_strat_write() tracing.
*/
ktrace_t *xfs_strat_trace_buf;
#endif
#if !defined(XFS_STRAT_TRACE)
#define xfs_strat_write_bp_trace(tag, ip, bp)
#define xfs_strat_write_subbp_trace(tag, ip, bp, rbp, loff, lcnt, lblk)
#endif /* !XFS_STRAT_TRACE */
STATIC int
xfs_zero_last_block(
xfs_inode_t *ip,
off_t offset,
xfs_fsize_t isize,
cred_t *credp,
struct pm *pmp);
STATIC void
xfs_zero_bp(
buf_t *bp,
int data_offset,
int data_len);
STATIC int
xfs_retrieved(
uint available,
off_t offset,
size_t count,
uint *total_retrieved,
xfs_fsize_t isize);
#ifndef DEBUG
#define xfs_strat_write_check(ip,off,count,imap,nimap)
#define xfs_check_rbp(ip,bp,rbp,locked)
#define xfs_check_bp(ip,bp)
#define xfs_check_gap_list(ip)
#else /* DEBUG */
STATIC void
xfs_strat_write_check(
xfs_inode_t *ip,
xfs_fileoff_t offset_fsb,
xfs_filblks_t buf_fsb,
xfs_bmbt_irec_t *imap,
int imap_count);
STATIC void
xfs_check_rbp(
xfs_inode_t *ip,
buf_t *bp,
buf_t *rbp,
int locked);
STATIC void
xfs_check_bp(
xfs_inode_t *ip,
buf_t *bp);
STATIC void
xfs_check_gap_list(
xfs_inode_t *ip);
#endif /* DEBUG */
STATIC int
xfs_build_gap_list(
xfs_inode_t *ip,
off_t offset,
size_t count);
STATIC void
xfs_free_gap_list(
xfs_inode_t *ip);
STATIC void
xfs_cmp_gap_list_and_zero(
xfs_inode_t *ip,
buf_t *bp);
STATIC void
xfs_delete_gap_list(
xfs_inode_t *ip,
xfs_fileoff_t offset_fsb,
xfs_extlen_t count_fsb);
STATIC int
xfs_strat_write_unwritten(
bhv_desc_t *bdp,
buf_t *bp);
STATIC void
xfs_strat_write_iodone(
buf_t *bp);
STATIC void
xfs_strat_comp(void);
STATIC int
xfsd(void);
int
xfs_diordwr(
bhv_desc_t *bdp,
uio_t *uiop,
int ioflag,
cred_t *credp,
uint64_t rw);
extern int
grio_io_is_guaranteed(
vfile_t *,
stream_id_t *);
extern void
griostrategy(
buf_t *);
extern int
grio_monitor_io_start(
stream_id_t *,
__int64_t);
extern int
grio_monitor_io_end(
stream_id_t *,
int);
extern void xfs_error(
xfs_mount_t *,
int);
STATIC void
xfs_delalloc_cleanup(
xfs_inode_t *ip,
xfs_fileoff_t start_fsb,
xfs_filblks_t count_fsb);
extern void xfs_buf_iodone_callbacks(struct buf *);
extern void xlog_iodone(struct buf *);
int xfs_bioerror_relse(buf_t *bp);
/*
* Round the given file offset down to the nearest read/write
* size boundary.
*/
#define XFS_READIO_ALIGN(ip,off) (((off) >> ip->i_readio_log) \
<< ip->i_readio_log)
#define XFS_WRITEIO_ALIGN(ip,off) (((off) >> ip->i_writeio_log) \
<< ip->i_writeio_log)
#if !defined(XFS_RW_TRACE)
#define xfs_rw_enter_trace(tag, ip, uiop, ioflags)
#define xfs_iomap_enter_trace(tag, ip, offset, count);
#define xfs_iomap_map_trace(tag, ip, offset, count, bmapp, imapp)
#define xfs_inval_cached_trace(ip, offset, len, first, last)
#else
/*
* Trace routine for the read/write path. This is the routine entry trace.
*/
static void
xfs_rw_enter_trace(
int tag,
xfs_inode_t *ip,
uio_t *uiop,
int ioflags)
{
if (ip->i_rwtrace == NULL) {
return;
}
ktrace_enter(ip->i_rwtrace,
(void*)((unsigned long)tag),
(void*)ip,
(void*)((ip->i_d.di_size >> 32) & 0xffffffff),
(void*)(ip->i_d.di_size & 0xffffffff),
(void*)(((__uint64_t)uiop->uio_offset >> 32) &
0xffffffff),
(void*)(uiop->uio_offset & 0xffffffff),
(void*)uiop->uio_resid,
(void*)((unsigned long)ioflags),
(void*)((ip->i_next_offset >> 32) & 0xffffffff),
(void*)(ip->i_next_offset & 0xffffffff),
(void*)((unsigned long)((ip->i_io_offset >> 32) &
0xffffffff)),
(void*)(ip->i_io_offset & 0xffffffff),
(void*)((unsigned long)(ip->i_io_size)),
(void*)((unsigned long)(ip->i_last_req_sz)),
(void*)((unsigned long)((ip->i_new_size >> 32) &
0xffffffff)),
(void*)(ip->i_new_size & 0xffffffff));
}
static void
xfs_iomap_enter_trace(
int tag,
xfs_inode_t *ip,
off_t offset,
size_t count)
{
if (ip->i_rwtrace == NULL) {
return;
}
ktrace_enter(ip->i_rwtrace,
(void*)((unsigned long)tag),
(void*)ip,
(void*)((ip->i_d.di_size >> 32) & 0xffffffff),
(void*)(ip->i_d.di_size & 0xffffffff),
(void*)(((__uint64_t)offset >> 32) & 0xffffffff),
(void*)(offset & 0xffffffff),
(void*)((unsigned long)count),
(void*)((ip->i_next_offset >> 32) & 0xffffffff),
(void*)(ip->i_next_offset & 0xffffffff),
(void*)((ip->i_io_offset >> 32) & 0xffffffff),
(void*)(ip->i_io_offset & 0xffffffff),
(void*)((unsigned long)(ip->i_io_size)),
(void*)((unsigned long)(ip->i_last_req_sz)),
(void*)((ip->i_new_size >> 32) & 0xffffffff),
(void*)(ip->i_new_size & 0xffffffff),
(void*)0);
}
void
xfs_iomap_map_trace(
int tag,
xfs_inode_t *ip,
off_t offset,
size_t count,
struct bmapval *bmapp,
xfs_bmbt_irec_t *imapp)
{
if (ip->i_rwtrace == NULL) {
return;
}
ktrace_enter(ip->i_rwtrace,
(void*)((unsigned long)tag),
(void*)ip,
(void*)((ip->i_d.di_size >> 32) & 0xffffffff),
(void*)(ip->i_d.di_size & 0xffffffff),
(void*)(((__uint64_t)offset >> 32) & 0xffffffff),
(void*)(offset & 0xffffffff),
(void*)((unsigned long)count),
(void*)((bmapp->offset >> 32) & 0xffffffff),
(void*)(bmapp->offset & 0xffffffff),
(void*)((unsigned long)(bmapp->length)),
(void*)((unsigned long)(bmapp->pboff)),
(void*)((unsigned long)(bmapp->pbsize)),
(void*)(bmapp->bn),
(void*)(__psint_t)(imapp->br_startoff),
(void*)((unsigned long)(imapp->br_blockcount)),
(void*)(__psint_t)(imapp->br_startblock));
}
static void
xfs_inval_cached_trace(
xfs_inode_t *ip,
off_t offset,
off_t len,
off_t first,
off_t last)
{
if (ip->i_rwtrace == NULL)
return;
ktrace_enter(ip->i_rwtrace,
(void *)(__psint_t)XFS_INVAL_CACHED,
(void *)ip,
(void *)(((__uint64_t)offset >> 32) & 0xffffffff),
(void *)(offset & 0xffffffff),
(void *)(((__uint64_t)len >> 32) & 0xffffffff),
(void *)(len & 0xffffffff),
(void *)(((__uint64_t)first >> 32) & 0xffffffff),
(void *)(first & 0xffffffff),
(void *)(((__uint64_t)last >> 32) & 0xffffffff),
(void *)(last & 0xffffffff),
(void *)0,
(void *)0,
(void *)0,
(void *)0,
(void *)0,
(void *)0);
}
#endif /* XFS_RW_TRACE */
#ifdef DEBUG
/* ARGSUSED */
void
debug_print_vnmaps(vnmap_t *vnmap, int numvnmaps, int vnmap_flags)
{
int i;
for (i = 0; i < numvnmaps; i++, vnmap++) {
cmn_err(CE_DEBUG,
"vaddr = 0x%llx, len = 0x%llx, flags = 0x%x\n ovvaddr = 0x%llx len = 0x%llx offset = %lld\n",
(uint64_t)vnmap->vnmap_vaddr,
(uint64_t)vnmap->vnmap_len,
vnmap->vnmap_flags,
(uint64_t)vnmap->vnmap_ovvaddr,
(uint64_t)vnmap->vnmap_ovlen,
vnmap->vnmap_ovoffset);
}
}
#endif
/*
* Fill in the bmap structure to indicate how the next bp
* should fit over the given extent.
*
* Everything here is in terms of file system blocks, not BBs.
*/
void
xfs_next_bmap(
xfs_mount_t *mp,
xfs_bmbt_irec_t *imapp,
struct bmapval *bmapp,
int iosize,
int last_iosize,
xfs_fileoff_t ioalign,
xfs_fileoff_t last_offset,
xfs_fileoff_t req_offset,
xfs_fsize_t isize)
{
__int64_t extra_blocks;
xfs_fileoff_t size_diff;
xfs_fileoff_t ext_offset;
xfs_fileoff_t last_file_fsb;
xfs_fsblock_t start_block;
/*
* Make sure that the request offset lies in the extent given.
*/
ASSERT(req_offset >= imapp->br_startoff);
ASSERT(req_offset < (imapp->br_startoff + imapp->br_blockcount));
if (last_offset == -1) {
ASSERT(ioalign != -1);
if (ioalign < imapp->br_startoff) {
/*
* The alignment we guessed at can't
* happen on this extent, so align
* to the beginning of this extent.
* Subtract whatever we drop from the
* iosize so that we stay aligned on
* our iosize boundaries.
*/
size_diff = imapp->br_startoff - ioalign;
iosize -= size_diff;
ASSERT(iosize > 0);
ext_offset = 0;
bmapp->offset = imapp->br_startoff;
ASSERT(bmapp->offset <= req_offset);
} else {
/*
* The alignment requested fits on this
* extent, so use it.
*/
ext_offset = ioalign - imapp->br_startoff;
bmapp->offset = ioalign;
ASSERT(bmapp->offset <= req_offset);
}
} else {
/*
* This is one of a series of sequential access to the
* file. Make sure to line up the buffer we specify
* so that it doesn't overlap the last one. It should
* either be the same as the last one (if we need data
* from it) or follow immediately after the last one.
*/
ASSERT(ioalign == -1);
if (last_offset >= imapp->br_startoff) {
/*
* The last I/O was from the same extent
* that this one will at least start on.
* This assumes that we're going sequentially.
*/
if (req_offset < (last_offset + last_iosize)) {
/*
* This request overlaps the buffer
* we used for the last request. Just
* get that buffer again.
*/
ext_offset = last_offset -
imapp->br_startoff;
bmapp->offset = last_offset;
iosize = last_iosize;
} else {
/*
* This request does not overlap the buffer
* used for the last one. Get it its own.
*/
ext_offset = req_offset - imapp->br_startoff;
bmapp->offset = req_offset;
}
} else {
/*
* The last I/O was on a different extent than
* this one. We start at the beginning of this one.
* This assumes that we're going sequentially.
*/
ext_offset = req_offset - imapp->br_startoff;
bmapp->offset = req_offset;
}
}
start_block = imapp->br_startblock;
if (start_block == HOLESTARTBLOCK) {
bmapp->bn = -1;
bmapp->eof = BMAP_HOLE;
} else if (start_block == DELAYSTARTBLOCK) {
bmapp->bn = -1;
bmapp->eof = BMAP_DELAY;
} else {
bmapp->bn = start_block + ext_offset;
bmapp->eof = 0;
if (imapp->br_state == XFS_EXT_UNWRITTEN)
bmapp->eof |= BMAP_UNWRITTEN;
}
bmapp->length = iosize;
/*
* If the iosize from our offset extends beyond the
* end of the extent, then trim down the length
* to match that of the extent.
*/
extra_blocks = (off_t)(bmapp->offset + bmapp->length) -
(__uint64_t)(imapp->br_startoff +
imapp->br_blockcount);
if (extra_blocks > 0) {
bmapp->length -= extra_blocks;
ASSERT(bmapp->length > 0);
}
/*
* If the iosize from our offset extends beyond the end
* of the file and the current extent is simply a hole,
* then trim down the length to match the
* size of the file. This keeps us from going out too
* far into hole at the EOF that extends to infinity.
*/
if (start_block == HOLESTARTBLOCK) {
last_file_fsb = XFS_B_TO_FSB(mp, isize);
extra_blocks = (off_t)(bmapp->offset + bmapp->length) -
(__uint64_t)last_file_fsb;
if (extra_blocks > 0) {
bmapp->length -= extra_blocks;
ASSERT(bmapp->length > 0);
}
ASSERT((bmapp->offset + bmapp->length) <= last_file_fsb);
}
bmapp->bsize = XFS_FSB_TO_B(mp, bmapp->length);
}
/*
* xfs_retrieved() is a utility function used to calculate the
* value of bmap.pbsize.
*
* Available is the number of bytes mapped by the current bmap.
* Offset is the file offset of the current request by the user.
* Count is the size of the current request by the user.
* Total_retrieved is a running total of the number of bytes
* which have been setup for the user in this call so far.
* Isize is the current size of the file being read.
*/
STATIC int
xfs_retrieved(
uint available,
off_t offset,
size_t count,
uint *total_retrieved,
xfs_fsize_t isize)
{
uint retrieved;
xfs_fsize_t file_bytes_left;
if ((available + *total_retrieved) > count) {
/*
* This buffer will return more bytes
* than we asked for. Trim retrieved
* so we can set bmapp->pbsize correctly.
*/
retrieved = count - *total_retrieved;
} else {
retrieved = available;
}
file_bytes_left = isize - (offset + *total_retrieved);
if (file_bytes_left < retrieved) {
/*
* The user has requested more bytes
* than there are in the file. Trim
* down the number to those left in
* the file.
*/
retrieved = file_bytes_left;
}
*total_retrieved += retrieved;
return retrieved;
}
/*
* xfs_iomap_extra()
*
* This is called when the VM/chunk cache is trying to create a buffer
* for a page which is beyond the end of the file. If we're at the
* start of the page we give it as much of a mapping as we can, but
* if it comes back for the rest of the page we say there is nothing there.
* This behavior is tied to the code in the VM/chunk cache (do_pdflush())
* that will call here.
*/
STATIC int /* error */
xfs_iomap_extra(
xfs_inode_t *ip,
off_t offset,
size_t count,
struct bmapval *bmapp,
int *nbmaps,
struct pm *pmp)
{
xfs_fileoff_t offset_fsb;
xfs_fileoff_t end_fsb;
xfs_fsize_t nisize;
xfs_mount_t *mp;
int nimaps;
xfs_fsblock_t firstblock;
int error;
xfs_bmbt_irec_t imap;
nisize = ip->i_new_size;
if (nisize < ip->i_d.di_size) {
nisize = ip->i_d.di_size;
}
mp = ip->i_mount;
offset_fsb = XFS_B_TO_FSBT(mp, offset);
if (poff(offset) != 0) {
/*
* This is the 'remainder' of a page being mapped out.
* Since it is already beyond the EOF, there is no reason
* to bother.
*/
ASSERT(count < NBPP);
*nbmaps = 1;
bmapp->eof = BMAP_EOF;
bmapp->bn = -1;
bmapp->offset = XFS_FSB_TO_BB(mp, offset_fsb);
bmapp->length = 0;
bmapp->bsize = 0;
bmapp->pboff = 0;
bmapp->pbsize = 0;
bmapp->pmp = pmp;
if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) {
bmapp->pbdev = mp->m_rtdev;
} else {
bmapp->pbdev = mp->m_dev;
}
} else {
/*
* A page is being mapped out so that it can be flushed.
* The page is beyond the EOF, but we need to return
* something to keep the chunk cache happy.
*/
ASSERT(count <= NBPP);
end_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
nimaps = 1;
firstblock = NULLFSBLOCK;
error = xfs_bmapi(NULL, ip, offset_fsb,
(xfs_filblks_t)(end_fsb - offset_fsb),
0, &firstblock, 0, &imap,
&nimaps, NULL);
if (error) {
return error;
}
ASSERT(nimaps == 1);
*nbmaps = 1;
bmapp->eof = BMAP_EOF;
if (imap.br_startblock == HOLESTARTBLOCK) {
bmapp->eof |= BMAP_HOLE;
bmapp->bn = -1;
} else if (imap.br_startblock == DELAYSTARTBLOCK) {
bmapp->eof |= BMAP_DELAY;
bmapp->bn = -1;
} else {
bmapp->bn = XFS_FSB_TO_DB(ip, imap.br_startblock);
if (imap.br_state == XFS_EXT_UNWRITTEN)
bmapp->eof |= BMAP_UNWRITTEN;
}
bmapp->offset = XFS_FSB_TO_BB(mp, offset_fsb);
bmapp->length = XFS_FSB_TO_BB(mp, imap.br_blockcount);
ASSERT(bmapp->length > 0);
bmapp->bsize = BBTOB(bmapp->length);
bmapp->pboff = offset - BBTOOFF(bmapp->offset);
ASSERT(bmapp->pboff >= 0);
bmapp->pbsize = bmapp->bsize - bmapp->pboff;
ASSERT(bmapp->pbsize > 0);
bmapp->pmp = pmp;
if (bmapp->pbsize > count) {
bmapp->pbsize = count;
}
if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) {
bmapp->pbdev = mp->m_rtdev;
} else {
bmapp->pbdev = mp->m_dev;
}
}
return 0;
}
/*
* xfs_iomap_read()
*
* This is the main I/O policy routine for reads. It fills in
* the given bmapval structures to indicate what I/O requests
* should be used to read in the portion of the file for the given
* offset and count.
*
* The inode's I/O lock may be held SHARED here, but the inode lock
* must be held EXCL because it protects the read ahead state variables
* in the inode.
* Bug 516806: The readahead state is now maintained by i_rlock therefore,
* the inode lock can be held in SHARED mode. The only time we need it
* in EXCL mode is when it is being read in the first time.
*/
int /* error */
xfs_iomap_read(
xfs_inode_t *ip,
off_t offset,
size_t count,
struct bmapval *bmapp,
int *nbmaps,
struct pm *pmp,
int *unlocked,
unsigned int lockmode)
{
xfs_fileoff_t offset_fsb;
xfs_fileoff_t ioalign;
xfs_fileoff_t last_offset;
xfs_fileoff_t last_required_offset;
xfs_fileoff_t next_offset;
xfs_fileoff_t last_fsb;
xfs_fileoff_t max_fsb;
xfs_fsize_t nisize;
off_t offset_page;
off_t aligned_offset;
xfs_fsblock_t firstblock;
int nimaps;
int error;
unsigned int iosize;
unsigned int last_iosize;
unsigned int retrieved_bytes;
unsigned int total_retrieved_bytes;
short filled_bmaps;
short read_aheads;
int x;
xfs_mount_t *mp;
struct bmapval *curr_bmapp;
struct bmapval *next_bmapp;
struct bmapval *last_bmapp;
struct bmapval *first_read_ahead_bmapp;
struct bmapval *next_read_ahead_bmapp;
xfs_bmbt_irec_t *curr_imapp;
xfs_bmbt_irec_t *last_imapp;
xfs_bmbt_irec_t imap[XFS_MAX_RW_NBMAPS];
ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE | MR_ACCESS) != 0);
ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE | MR_ACCESS) != 0);
xfs_iomap_enter_trace(XFS_IOMAP_READ_ENTER, ip, offset, count);
mp = ip->i_mount;
nisize = ip->i_new_size;
if (nisize < ip->i_d.di_size) {
nisize = ip->i_d.di_size;
}
offset_fsb = XFS_B_TO_FSBT(mp, offset);
nimaps = sizeof(imap) / sizeof(imap[0]);
last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)nisize));
if (offset >= nisize) {
/*
* The VM/chunk code is trying to map a page or part
* of a page to be pushed out that is beyond the end
* of the file. We handle these cases separately so
* that they do not interfere with the normal path
* code.
*/
error = xfs_iomap_extra(ip, offset, count, bmapp, nbmaps, pmp);
return error;
}
/*
* Map out to the maximum possible file size. This will return
* an extra hole we don't really care about at the end, but we
* won't do any read-ahead beyond the EOF anyway. We do this
* so that the buffers we create here line up well with those
* created in xfs_iomap_write() which extend beyond the end of
* the file.
*/
max_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAX_FILE_OFFSET);
firstblock = NULLFSBLOCK;
error = xfs_bmapi(NULL, ip, offset_fsb,
(xfs_filblks_t)(max_fsb - offset_fsb),
XFS_BMAPI_ENTIRE, &firstblock, 0, imap,
&nimaps, NULL);
if (error) {
return error;
}
xfs_iunlock_map_shared(ip, lockmode);
*unlocked = 1;
mutex_lock(&ip->i_rlock, PINOD);
if ((offset == ip->i_next_offset) &&
(count <= ip->i_last_req_sz)) {
/*
* Sequential I/O of same size as last time.
*/
ASSERT(ip->i_io_size > 0);
iosize = ip->i_io_size;
ASSERT(iosize <= XFS_BB_TO_FSBT(mp, XFS_MAX_BMAP_LEN_BB));
last_offset = ip->i_io_offset;
ioalign = -1;
} else {
/*
* The I/O size for the file has not yet been
* determined, so figure it out.
*/
if (XFS_B_TO_FSB(mp, count) <= ip->i_readio_blocks) {
/*
* The request is smaller than our
* minimum I/O size, so default to
* the minimum. For these size requests
* we always want to align the requests
* to XFS_READ_SIZE boundaries as well.
*/
iosize = ip->i_readio_blocks;
ASSERT(iosize <=
XFS_BB_TO_FSBT(mp, XFS_MAX_BMAP_LEN_BB));
aligned_offset = XFS_READIO_ALIGN(ip, offset);
ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
} else {
/*
* The request is bigger than our
* minimum I/O size and it's the
* first one in this sequence, so
* set the I/O size for the file
* now.
*
* In calculating the offset rounded down
* to a page, make sure to round down the
* fs block offset rather than the byte
* offset for the case where our block size
* is greater than the page size. This way
* offset_page will always align to a fs block
* as well as a page.
*
* For the end of the I/O we need to round
* offset + count up to a page boundary and
* then round that up to a file system block
* boundary.
*/
offset_page = ctooff(offtoct(XFS_FSB_TO_B(mp,
offset_fsb)));
last_fsb = XFS_B_TO_FSB(mp,
ctooff(offtoc(offset + count)));
iosize = last_fsb - XFS_B_TO_FSBT(mp, offset_page);
if (iosize >
XFS_BB_TO_FSBT(mp, XFS_MAX_BMAP_LEN_BB)) {
iosize = XFS_BB_TO_FSBT(mp,
XFS_MAX_BMAP_LEN_BB);
}
ioalign = XFS_B_TO_FSB(mp, offset_page);
}
last_offset = -1;
}
/*
* Now we've got the I/O size and the last offset,
* so start figuring out how to align our
* buffers.
*/
xfs_next_bmap(mp, imap, bmapp, iosize, iosize, ioalign,
last_offset, offset_fsb, nisize);
ASSERT((bmapp->length > 0) &&
(offset >= XFS_FSB_TO_B(mp, bmapp->offset)));
if (XFS_FSB_TO_B(mp, bmapp->offset + bmapp->length) >= nisize) {
bmapp->eof |= BMAP_EOF;
}
bmapp->pboff = offset - XFS_FSB_TO_B(mp, bmapp->offset);
retrieved_bytes = bmapp->bsize - bmapp->pboff;
total_retrieved_bytes = 0;
bmapp->pbsize = xfs_retrieved(retrieved_bytes, offset, count,
&total_retrieved_bytes, nisize);
xfs_iomap_map_trace(XFS_IOMAP_READ_MAP,
ip, offset, count, bmapp, imap);
/*
* Only map additional buffers if they've been asked for
* and the I/O being done is sequential and has reached the
* point where we need to initiate more read ahead or we didn't get
* the whole request in the first bmap.
*/
last_fsb = XFS_B_TO_FSB(mp, nisize);
filled_bmaps = 1;
last_required_offset = bmapp[0].offset;
first_read_ahead_bmapp = NULL;
if ((*nbmaps > 1) &&
(((offset == ip->i_next_offset) &&
(offset != 0) &&
(offset_fsb >= ip->i_reada_blkno)) ||
retrieved_bytes < count)) {
curr_bmapp = &bmapp[0];
next_bmapp = &bmapp[1];
last_bmapp = &bmapp[*nbmaps - 1];
curr_imapp = &imap[0];
last_imapp = &imap[nimaps - 1];
/*
* curr_bmap is always the last one we filled
* in, and next_bmapp is always the next one
* to be filled in.
*/
while (next_bmapp <= last_bmapp) {
next_offset = curr_bmapp->offset +
curr_bmapp->length;
if (next_offset >= last_fsb) {
/*
* We've mapped all the way to the EOF.
* Everything beyond there is inaccessible,
* so get out now.
*/
break;
}
last_iosize = curr_bmapp->length;
if (next_offset <
(curr_imapp->br_startoff +
curr_imapp->br_blockcount)) {
xfs_next_bmap(mp, curr_imapp,
next_bmapp, iosize, last_iosize, -1,
curr_bmapp->offset, next_offset,
nisize);
} else {
curr_imapp++;
if (curr_imapp <= last_imapp) {
xfs_next_bmap(mp,
curr_imapp, next_bmapp,
iosize, last_iosize, -1,
curr_bmapp->offset, next_offset,
nisize);
} else {
/*
* We're out of imaps. We
* either hit the end of
* file or just didn't give
* enough of them to bmapi.
* The caller will just come
* back if we haven't done
* enough yet.
*/
break;
}
}
filled_bmaps++;
curr_bmapp = next_bmapp;
next_bmapp++;
ASSERT(curr_bmapp->length > 0);
/*
* Make sure to fill in the pboff and pbsize
* fields as long as the bmaps are required for
* the request (as opposed to strictly read-ahead).
*/
if (total_retrieved_bytes < count) {
curr_bmapp->pboff = 0;
curr_bmapp->pbsize =
xfs_retrieved(curr_bmapp->bsize,
offset, count,
&total_retrieved_bytes,
nisize);
}
if (XFS_FSB_TO_B(mp, curr_bmapp->offset +
curr_bmapp->length) >= nisize) {
curr_bmapp->eof |= BMAP_EOF;
}
xfs_iomap_map_trace(XFS_IOMAP_READ_MAP, ip, offset,
count, curr_bmapp, curr_imapp);
/*
* Keep track of the offset of the last buffer
* needed to satisfy the I/O request. This will
* be used for i_io_offset later. Also record
* the first bmapp used to track a read ahead.
*/
if (XFS_FSB_TO_B(mp, curr_bmapp->offset) <
(offset + count)) {
last_required_offset = curr_bmapp->offset;
} else if (first_read_ahead_bmapp == NULL) {
first_read_ahead_bmapp = curr_bmapp;
}
}
/*
* If we're describing any read-ahead here, then move
* the read-ahead blkno up to the point where we've
* gone through half the read-ahead described here.
* This way we don't issue more read-ahead until we
* are half-way through the last read-ahead.
*
* If we're not describing any read-ahead because the
* request is just fragmented, then move up the
* read-ahead blkno to just past what we're returning
* so that we can maybe start it on the next request.
*/
if (first_read_ahead_bmapp != NULL) {
read_aheads = curr_bmapp - first_read_ahead_bmapp +1;
next_read_ahead_bmapp = first_read_ahead_bmapp +
(read_aheads / 2);
ip->i_reada_blkno = next_read_ahead_bmapp->offset;
} else {
ip->i_reada_blkno = curr_bmapp->offset +
curr_bmapp->length;
}
} else if ((*nbmaps > 1) && (offset != ip->i_io_offset)) {
/*
* In this case the caller is not yet accessing the
* file sequentially, but set the read-ahead blkno
* so that we can tell if they start doing so.
*/
ip->i_reada_blkno = bmapp[0].offset + bmapp[0].length;
}
ASSERT(iosize <= XFS_BB_TO_FSBT(mp, XFS_MAX_BMAP_LEN_BB));
ip->i_io_size = iosize;
ip->i_io_offset = last_required_offset;
if (count > ip->i_last_req_sz) {
/*
* Record the "last request size" for the file.
* We don't let it shrink so that big requests
* that are not satisfied in one call here still
* record the full request size (not the smaller
* one that comes in to finish mapping the request).
*/
ip->i_last_req_sz = count;
}
if (total_retrieved_bytes >= count) {
/*
* We've mapped all of the caller's request, so
* the next request in a sequential read will
* come in the block this one ended on or the
* next if we consumed up to the very end of
* a block.
*/
ip->i_next_offset = offset + count;
} else {
/*
* We didn't satisfy the entire request, so we
* can expect xfs_read_file() to come back with
* what is left of the request.
*/
ip->i_next_offset = offset + total_retrieved_bytes;
}
mutex_unlock(&ip->i_rlock);
*nbmaps = filled_bmaps;
for (x = 0; x < filled_bmaps; x++) {
curr_bmapp = &bmapp[x];
if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) {
curr_bmapp->pbdev = mp->m_rtdev;
} else {
curr_bmapp->pbdev = mp->m_dev;
}
ASSERT(curr_bmapp->offset <= XFS_B_TO_FSB(mp, nisize));
curr_bmapp->offset = XFS_FSB_TO_BB(mp, curr_bmapp->offset);
curr_bmapp->length = XFS_FSB_TO_BB(mp, curr_bmapp->length);
ASSERT(curr_bmapp->length > 0);
ASSERT((x == 0) ||
((bmapp[x - 1].offset + bmapp[x - 1].length) ==
curr_bmapp->offset));
if (curr_bmapp->bn != -1) {
curr_bmapp->bn = XFS_FSB_TO_DB(ip, curr_bmapp->bn);
}
curr_bmapp->pmp = pmp;
}
return 0;
}
/* ARGSUSED */
int
xfs_vop_readbuf(bhv_desc_t *bdp,
off_t offset,
ssize_t len,
int ioflags,
struct cred *creds,
flid_t *fl,
buf_t **rbuf,
int *pboff,
int *pbsize)
{
vnode_t *vp;
xfs_inode_t *ip;
int error;
struct bmapval bmaps[2];
int nmaps;
buf_t *bp;
extern void chunkrelse(buf_t *bp);
int unlocked;
int lockmode;
vp = BHV_TO_VNODE(bdp);
ip = XFS_BHVTOI(bdp);
*rbuf = NULL;
*pboff = *pbsize = -1;
error = 0;
xfs_rwlockf(bdp, VRWLOCK_READ, 0);
if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
error = XFS_ERROR(EIO);
goto out;
}
/*
* blow this off if mandatory locking or DMI are involved
*/
if ((vp->v_flag & (VENF_LOCKING|VFRLOCKS)) == (VENF_LOCKING|VFRLOCKS))
goto out;
if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) && !(ioflags & IO_INVIS))
goto out;
unlocked = 0;
lockmode = xfs_ilock_map_shared(ip);
if (offset >= ip->i_d.di_size) {
xfs_iunlock_map_shared(ip, lockmode);
goto out;
}
/*
* prohibit iomap read from giving us back our data in
* two buffers but let it set up read-ahead. Turn off
* read-ahead for NFSv2. It's I/O sizes are too small
* to be of any real benefit (8K reads, 32K read buffers).
*/
nmaps = (ioflags & IO_NFS) ? 1 : 2;
error = xfs_iomap_read(ip, offset, len, bmaps, &nmaps, NULL, &unlocked,
lockmode);
if (!unlocked)
xfs_iunlock_map_shared(ip, lockmode);
/*
* if the first bmap doesn't match the I/O request, forget it.
* This means that we can't fit the request into one buffer.
*/
if (error ||
((bmaps[0].pbsize != len) &&
(bmaps[0].eof & BMAP_EOF) == 0))
goto out;
/*
* if the caller has specified that the I/O fit into
* one page and it doesn't, forget it. The caller won't
* be able to use it.
*/
if ((ioflags & IO_ONEPAGE)
&& pnum(offset) != pnum(offset + bmaps[0].pbsize-1)) {
goto out;
}
bp = chunkread(vp, bmaps, nmaps, creds);
if (bp->b_flags & B_ERROR) {
error = geterror(bp);
ASSERT(error != EINVAL);
/*
* b_relse functions like chunkhold
* expect B_DONE to be there.
*/
bp->b_flags |= B_DONE|B_STALE;
brelse(bp);
goto out;
}
if ((bmaps[0].pboff + bmaps[0].pbsize) == bmaps[0].bsize)
bp->b_relse = chunkrelse;
*rbuf = bp;
*pboff = bmaps[0].pboff;
*pbsize = bmaps[0].pbsize;
xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
out:
xfs_rwunlockf(bdp, VRWLOCK_READ, 0);
return XFS_ERROR(error);
}
/*
* set of routines (xfs_lockdown_iopages, xfs_unlock_iopages,
* xfs_mapped_biomove) to deal with i/o to or from mmapped files where
* some or all of the user buffer is mmapped to the file passed in
* as the target of the read/write system call.
*
* there are 6 sets of deadlocks and possible problems.
*
* 1) the i/o is a read, the user buffer lies in a region
* that is mapped to the file and not paged in. the fault
* code calls VOP_READ. But we deadlock if another thread
* has tried to write the file in the meantime because he's
* now waiting on the lock and the i/o lock has to be a barrier
* lock to prevent writer starvation. this is addressed by calling
* the new VASOP, verifyvnmap that returns a set of maps indicating
* which ranges of the biomove'd virtual addresses are mmapped to
* the file. the i/o path then breaks up the biomove (using
* xfs_mapped_biomove) into pieces and enables nested locking
* on the i/o lock during the biomove calls that could result
* in page faults that need to read data from the file.
*
* 2) like above only the i/o is a write. the page fault deadlocks
* instantly since we already hold the i/o lock in write mode and
* the xfs_read called by the page fault needs it in read mode.
* this one is handled as above by enabling nested locking around
* the appropriate biomove calls.
*
* 3) the i/o is a read, the user buffer lies in a region
* that is mapped autogrow,shared, and the biomove filling
* the user buffer causes the file to grow. the page fault
* triggered by the biomove needs to run xfs_write() and take
* the i/o lock in write mode. this is addressed by making the
* read path smart enough to detect this condition using the
* information returned by the verifyvnmap VASOP and take the
* i/o lock in update mode at the beginning. we then enable
* recursive locking (see xfs_ilock) to allow the fault to obtain
* the i/o lock regardless of whose waiting on it.
*
* 4) deadlock in pagewait(). if the biomove faults and needs a
* page that exists but that page was created by the chunkread
* issued by xfs_read/xfs_write, the page won't be marked P_DONE
* (and usable by the fault code) until the i/o finishes and releases
* the buffer. so the fault code will find the page and wait forever
* waiting for it to be marked P_DONE. this case is handled by
* useracc'ing the dangerous pages before the chunkread so that
* the pages exist prior to the chunkread. this case only happens
* if the range of file offsets touched by the i/o overlap with
* the range of file offsets associated with the mmapped virtual
* addresses touched by the biomove.
*
* 5) buffer deadlock. in the autogrow case, it's possible that
* that there can be no page overlap but that the file blocks
* that need to be initialized for the autogrow by xfs_zeroeof_blocks()
* and the file blocks in the i/o will both wind up living in the same
* buffer set up by the i/o path. this will deadlock because the
* the buffer will be held by the i/o path when the biomove faults
* causing the autogrow process to deadlock on the buffer semaphore.
* this case is handled like case #2. we create the pages and cause
* the autogrow to happen before the chunkread so that the fault/autogrow
* code and the io path code don't have to use the same buffer at the
* same time.
*
* 6) a write i/o that passes in a data buffer that is mmapped beyond
* the current EOF. Even if nested locking is enabled, the write path
* assumes that because buffered writes are single-threaded only one
* buffered writer can use the gap list and other inode fields at once.
* this is addressed by faulting in the user virtual address associated
* mapped to the largest file offset mapped by the buffer. the fault
* occurs after the i/o lock is taken in write mode but before any real
* work is done. this allows the VM system to issue a VOP_WRITE to
* grow the file to the appropriate point. then the real write call
* executes after the file has been grown. the fault is issued with
* nested locking enabled so the write-path inode fields are used
* serially while still holding onto the i/o lock to prevent a race
* with truncate().
*
* Situations this code does NOT attempt to deal with:
*
* - the above situations only we're doing direct I/O instead of
* buffered I/O.
*
* - situations arising from the mappings changing while our i/o
* is in progress. it's possible that someone could remap part
* of the buffer to make it dangerous after we've called to
* verifyvnmap but before we've done all our biomoves. fixing
* this would require serializing i/o's and mmaps/munmaps
* and I'd rather not do that.
*/
/*
* routines to enable/disable/query nested iolock locking and isolate
* the curuthread references to make the Cellular Irix merge easier
*/
void
xfs_enable_nested_locking(void)
{
ASSERT_ALWAYS(curuthread->ut_vnlock == 0);
curuthread->ut_vnlock = UT_FSNESTED;
}
void
xfs_disable_nested_locking(void)
{
ASSERT_ALWAYS(curuthread->ut_vnlock == UT_FSNESTED);
curuthread->ut_vnlock = 0;
}
int
xfs_is_nested_locking_enabled(void)
{
return curuthread->ut_vnlock & UT_FSNESTED;
}
/*
* xfs_lockdown_iopages() - lock down any mmapped user pages required for
* this i/o. this is either
*
* 1) pages which will be referenced by the biomove and whose backing
* file blocks will be directly read/written by the i/o
* 2) pages which will be referenced by the biomove, whose backing
* file blocks will reside in the same buffer used by the i/o,
* and whose file blocks are beyond the current EOF causing
* the file to be grown as part of the biomove
*
* if any pages are locked down, *useracced is set to 1 and a set
* of xfs_uaccmap_t's are returned indicating the set of address ranges
* were locked down. these pages should be unlocked as soon as
* possible after they are biomoved.
*
* returns ENOMEM if the number of pages being locked down exceeds
* maxdmasz. the pages should be unlocked ASAP after the biomove
* using xfs_unlock_iopages().
*/
int
xfs_lockdown_iopages(
struct bmapval *bmapp,
xfs_fsize_t isize, /* in - current inode size/eof */
int vnmapflags, /* in - map flags */
vnmap_t **vnmapp, /* in/out - vmaps array */
int *nvnmapp, /* in/out - number of valid maps left */
xfs_uaccmap_t *uaccmap, /* in - caller supplied useracc maps */
int *nuaccmapp, /* out - number of filled in uaccmaps */
int *useracced) /* out - did we useracc anything */
{
int nuaccmaps;
vnmap_t *vnmap = *vnmapp;
uvaddr_t useracc_startvaddr;
uvaddr_t useracc_endvaddr;
size_t useracc_len;
__psunsigned_t uacc_startshift;
__psunsigned_t uacc_endshift;
uvaddr_t agac_vaddr_start;
uvaddr_t agac_vaddr_end;
__psunsigned_t start_trim;
off_t bmap_end;
off_t bmap_start;
off_t vnmap_end;
off_t overlap_off_start;
off_t overlap_off_end;
off_t iomove_off_start;
off_t iomove_off_end;
off_t curr_offset;
int error;
int numpages;
/*
* do we have overlapping pages we need to
* useracc? don't have to worry about readahead
* buffers since those pages will be marked
* P_DONE by the buffer release function out of
* biodone when the i/o to the buffer finishes.
*/
ASSERT(*nuaccmapp >= *nvnmapp);
ASSERT((vnmapflags & (AS_VNMAP_OVERLAP|AS_VNMAP_AUTOGROW)));
error = numpages = *useracced = *nuaccmapp = nuaccmaps = 0;
curr_offset = 0;
bmap_start = BBTOOFF(bmapp[0].offset);
bmap_end = bmap_start + BBTOOFF(bmapp[0].length);
/*
* process all vnmaps up through the end of the current
* buffer
*/
while (vnmap && curr_offset < bmap_end &&
vnmap->vnmap_ovoffset < bmap_end) {
/*
* skip over maps that aren't marked as overlap or
* autogrow
*/
if (!(vnmap->vnmap_flags &
(AS_VNMAP_OVERLAP|AS_VNMAP_AUTOGROW))) {
(*nvnmapp)--;
if (*nvnmapp >= 0)
vnmap++;
else
vnmap = NULL;
continue;
}
/*
* if we haven an autogrow region, if the iomove
* grows the file, we need to calculate if any part
* of the iomove that is beyond the EOF will land
* in this buffer. if so, we need to lock those
* pages down now otherwise the xfs_write called by
* the fault code will need to zero the area of the
* file between the current and new EOF but it has to
* to get the buffer to do that. the problem is
* the buffer will be held (already locked) by the
* i/o so we'll deadlock.
*/
if (isize >= 0 && vnmap->vnmap_flags & AS_VNMAP_AUTOGROW) {
/*
* calculate file offsets touched by the iomove
* indicated in this vnmap
*/
start_trim = vnmap->vnmap_ovvaddr - vnmap->vnmap_vaddr;
iomove_off_start = vnmap->vnmap_ovoffset - start_trim;
iomove_off_end = iomove_off_start + vnmap->vnmap_len;
/*
* determine if any of the uimoved pages are in
* the buffer that will be set up by this bmap.
* we have to lock down any pages in the buffer
* between eof and the end of the uiomove to
* grow the file out to the end of the uiomove.
*/
if (isize < bmap_start) {
overlap_off_start = MAX(bmap_start,
iomove_off_start);
} else {
overlap_off_start = MIN(isize,
iomove_off_start);
}
overlap_off_end = MIN(iomove_off_end, bmap_end);
/*
* if so, set up the useracc range to span those
* pages. the useracc range can only grow larger
* than that range. it can't get smaller.
*/
if (overlap_off_end - overlap_off_start > 0) {
agac_vaddr_start = vnmap->vnmap_vaddr +
(iomove_off_start > overlap_off_start ?
iomove_off_start - overlap_off_start :
0);
agac_vaddr_end = vnmap->vnmap_vaddr +
vnmap->vnmap_len -
(iomove_off_end > overlap_off_end ?
iomove_off_end - overlap_off_end :
0);
} else {
agac_vaddr_start = (uvaddr_t) -1LL;
agac_vaddr_end = (uvaddr_t) 0LL;
}
} else {
agac_vaddr_start = (uvaddr_t) -1LL;
agac_vaddr_end = (uvaddr_t) 0LL;
}
/*
* useracc the smallest possible range. don't bother
* unless the overlap specified in the vnmap overlaps
* the file offsets specified for the buffer by the bmap.
*/
vnmap_end = vnmap->vnmap_ovoffset + vnmap->vnmap_ovlen;
overlap_off_start = MAX(vnmap->vnmap_ovoffset, bmap_start);
overlap_off_end = MIN(vnmap_end, bmap_end);
if (vnmap->vnmap_flags & AS_VNMAP_OVERLAP &&
overlap_off_end - overlap_off_start > 0) {
uacc_startshift = overlap_off_start -
vnmap->vnmap_ovoffset;
uacc_endshift = vnmap_end - overlap_off_end;
ASSERT(overlap_off_start - vnmap->vnmap_ovoffset >= 0);
ASSERT(vnmap_end - overlap_off_end >= 0);
useracc_startvaddr = vnmap->vnmap_ovvaddr +
uacc_startshift;
useracc_endvaddr = vnmap->vnmap_ovvaddr +
vnmap->vnmap_ovlen -
uacc_endshift;
} else {
useracc_startvaddr = (uvaddr_t) -1LL;
useracc_endvaddr = (uvaddr_t) 0LL;
}
/*
* enlarge range if necessary to include
* pages that have to be pinned because they
* would cause a zero-eof autogrow scenario
*/
useracc_startvaddr = MIN(useracc_startvaddr, agac_vaddr_start);
useracc_endvaddr = MAX(useracc_endvaddr, agac_vaddr_end);
if (useracc_startvaddr != (uvaddr_t) -1LL &&
useracc_endvaddr != (uvaddr_t) 0LL) {
/*
* enable recursive locking so the fault handler won't
* block on the i/o lock when setting up the pages.
* round the lockdown range to page boundaries.
* make sure we don't pin more than maxdmasz pages
* per i/o. that's not allowed.
*/
ASSERT(useracc_endvaddr > useracc_startvaddr);
useracc_startvaddr = (uvaddr_t)
ctob(btoct(useracc_startvaddr));
useracc_endvaddr = (uvaddr_t)
ctob(btoc(useracc_endvaddr));
useracc_len = useracc_endvaddr - useracc_startvaddr;
numpages += btoc(useracc_len);
if (numpages > maxdmasz) {
cmn_err(CE_WARN,
"xfs_lockdown_iopages needed to pin %d pages, maxdmasz = %d\nPlease increase maxdmasz and try again.\n",
numpages, maxdmasz);
error = XFS_ERROR(ENOMEM);
break;
}
xfs_enable_nested_locking();
error = useracc((void *)useracc_startvaddr, useracc_len,
B_READ|B_PHYS, NULL);
xfs_disable_nested_locking();
if (!error) {
*useracced = 1;
uaccmap->xfs_uacstart = useracc_startvaddr;
uaccmap->xfs_uaclen = useracc_len;
uaccmap++;
nuaccmaps++;
}
}
/*
* have to check next vmap if this vmap ends
* before the buffer does
*/
if (bmap_end >= vnmap_end) {
(*nvnmapp)--;
if (*nvnmapp >= 0)
vnmap++;
else
vnmap = NULL;
curr_offset = vnmap_end;
} else
curr_offset = bmap_end;
}
*nuaccmapp = nuaccmaps;
*vnmapp = vnmap;
return error;
}
/*
* xfs_unlock_iopages() - unlock the set of pages specified in the
* xfs_uaccmap_t's set up by xfs_lockdown_iopages().
*/
void
xfs_unlock_iopages(
xfs_uaccmap_t *uaccmap, /* useracc maps */
int nuaccmaps) /* number of filled in uaccmaps */
{
int i;
for (i = 0; i < nuaccmaps; i++, uaccmap++) {
ASSERT(uaccmap->xfs_uaclen <= ((size_t)-1));
unuseracc((void *)uaccmap->xfs_uacstart,
(size_t)uaccmap->xfs_uaclen,
B_PHYS|B_READ);
}
return;
}
/*
* handles biomoves of data where some of the user addresses are mapped to
* the file being read/written. each vnmap_t represents a range of addresses
* mapped to a file. that range needs to be biomoved with recursive locking
* enabled so we don't deadlock on the i/o lock when faulting in the biomove.
* however, we don't want recursive locking enabled on any other page since
* we could screw up the locking on other inodes if we try and take the
* i/o lock on a different inode where we don't hold the lock and we have
* recursive locking enabled.
*
* note -- we could do one biomove if we were willing to add an
* inode pointer in the uthread along with the vnlocks field. this
* code trades off increased complexity and slower execution speed
* (the extra biomoves plus the cycles required in pas_verifyvnmap
* to set up the additional vnmap's that we might not need if we
* weren't breaking up our biomoves) suffered only in the danger cases
* against memory bloat in the uthread structure that would be suffered by
* all uthreads.
*
* vnmapp and nvnmapp are set to the first map that hasn't completely been
* moved and the count of remaining valid maps respectively.
*/
int
xfs_mapped_biomove(
struct buf *bp,
u_int pboff,
size_t io_len,
enum uio_rw rw,
struct uio *uiop,
vnmap_t **vnmapp,
int *nvnmapp)
{
uvaddr_t io_end;
uvaddr_t current_vaddr;
int numvnmaps = *nvnmapp;
vnmap_t *vnmap = *vnmapp;
size_t partial_io_len;
size_t vmap_io_len;
int error = 0;
ASSERT(uiop->uio_iovcnt == 1);
/*
* do nothing if the requested biomove is entirely before
* the first vnmap address range
*/
current_vaddr = uiop->uio_iov->iov_base;
io_end = current_vaddr + io_len;
if (io_end < vnmap->vnmap_vaddr)
return XFS_ERROR(biomove(bp, pboff, io_len, rw, uiop));
/*
* move as much as we can (up to the first vnmap)
*/
if (current_vaddr < vnmap->vnmap_vaddr) {
partial_io_len = MIN(io_len,
(__psunsigned_t) vnmap->vnmap_vaddr -
(__psunsigned_t) uiop->uio_iov->iov_base);
if (error = biomove(bp, pboff, partial_io_len, rw, uiop))
return XFS_ERROR(error);
pboff += partial_io_len;
io_len -= partial_io_len;
current_vaddr += partial_io_len;
}
while (vnmap && current_vaddr < io_end) {
/*
* move what we can of the first vnmap. allow recursive
* io-lock locking in the biomove.
*/
ASSERT(uiop->uio_iov->iov_base >= vnmap->vnmap_vaddr);
ASSERT(current_vaddr >= vnmap->vnmap_vaddr);
vmap_io_len = vnmap->vnmap_len -
((__psunsigned_t) uiop->uio_iov->iov_base -
(__psunsigned_t) vnmap->vnmap_vaddr);
partial_io_len = MIN(vmap_io_len, io_len);
xfs_enable_nested_locking();
if (error = biomove(bp, pboff, partial_io_len, rw, uiop)) {
xfs_disable_nested_locking();
error = XFS_ERROR(error);
break;
}
xfs_disable_nested_locking();
pboff += partial_io_len;
io_len -= partial_io_len;
current_vaddr += partial_io_len;
/*
* did we move the entire vnmap? if so, look at the next map
* and move the data up to the next vnmap if there is one
* (or finish up the I/O if we're out of maps)
*/
ASSERT(current_vaddr <= vnmap->vnmap_vaddr + vnmap->vnmap_len);
if (current_vaddr == vnmap->vnmap_vaddr + vnmap->vnmap_len) {
numvnmaps--;
if (numvnmaps > 0) {
vnmap++;
partial_io_len = MIN(
(__psunsigned_t)vnmap->vnmap_vaddr -
(__psunsigned_t)uiop->uio_iov->iov_base,
io_len);
} else {
vnmap = NULL;
partial_io_len = io_len;
}
if (partial_io_len > 0) {
if (error = biomove(bp, pboff, partial_io_len,
rw, uiop)) {
error = XFS_ERROR(error);
break;
}
pboff += partial_io_len;
io_len -= partial_io_len;
current_vaddr += partial_io_len;
}
}
}
*nvnmapp = numvnmaps;
*vnmapp = vnmap;
return error;
}
/* ARGSUSED */
int
xfs_read_file(
bhv_desc_t *bdp,
uio_t *uiop,
int ioflag,
cred_t *credp,
vnmap_t *vnmaps,
int numvnmaps,
const uint vnmapflags,
xfs_uaccmap_t *uaccmaps,
xfs_fsize_t isize)
{
struct bmapval bmaps[XFS_MAX_RW_NBMAPS];
struct bmapval *bmapp;
int nbmaps;
vnode_t *vp;
buf_t *bp;
int read_bmaps;
int buffer_bytes_ok;
xfs_inode_t *ip;
int error;
int unlocked;
unsigned int lockmode;
int useracced = 0;
vnmap_t *cur_ldvnmap = vnmaps;
int num_ldvnmaps = numvnmaps;
int num_biovnmaps = numvnmaps;
int nuaccmaps;
int do_lockdown = vnmapflags & (AS_VNMAP_OVERLAP |
AS_VNMAP_AUTOGROW);
vnmap_t *cur_biovnmap = vnmaps;
vp = BHV_TO_VNODE(bdp);
ip = XFS_BHVTOI(bdp);
if (XFS_FORCED_SHUTDOWN(ip->i_mount))
return XFS_ERROR(EIO);
error = 0;
buffer_bytes_ok = 0;
XFSSTATS.xs_read_calls++;
XFSSTATS.xs_read_bytes += uiop->uio_resid;
/*
* Loop until uio->uio_resid, which is the number of bytes the
* caller has requested, goes to 0 or we get an error. Each
* call to xfs_iomap_read tries to map as much of the request
* plus read-ahead as it can. We must hold the inode lock
* exclusively when calling xfs_iomap_read.
* Bug 516806: Introduced i_rlock to protect the readahead state
* therefore do not need to hold the inode lock in exclusive
* mode except when we first read in the file and the extents
* are in btree format - xfs_ilock_map_shared takes care of it.
*/
do {
lockmode = xfs_ilock_map_shared(ip);
xfs_rw_enter_trace(XFS_READ_ENTER, ip, uiop, ioflag);
/*
* We've fallen off the end of the file, so
* just return with what we've done so far.
*/
if (uiop->uio_offset >= ip->i_d.di_size) {
xfs_iunlock_map_shared(ip, lockmode);
break;
}
unlocked = 0;
nbmaps = ip->i_mount->m_nreadaheads ;
ASSERT(nbmaps <= sizeof(bmaps) / sizeof(bmaps[0]));
/*
* XXX - rcc - we could make sure that if an overlap
* exists, that we don't set up bmaps that are > maxdmasz
* pages
*/
error = xfs_iomap_read(ip, uiop->uio_offset, uiop->uio_resid,
bmaps, &nbmaps, uiop->uio_pmp, &unlocked, lockmode);
if (!unlocked)
xfs_iunlock_map_shared(ip, lockmode);
if (error || (bmaps[0].pbsize == 0)) {
break;
}
bmapp = &bmaps[0];
read_bmaps = nbmaps;
ASSERT(BBTOOFF(bmapp->offset) <= uiop->uio_offset);
/*
* The first time through this loop we kick off I/O on
* all the bmaps described by the iomap_read call.
* Subsequent passes just wait for each buffer needed
* to satisfy this request to complete. Buffers which
* are started in the first pass but are actually just
* read ahead buffers are never waited for, since uio_resid
* will go to 0 before we get to them.
*
* This works OK because iomap_read always tries to
* describe all the buffers we need to satisfy this
* read call plus the necessary read-ahead in the
* first call to it.
*/
while ((uiop->uio_resid != 0) && (nbmaps > 0)) {
/*
* do we have overlapping pages we need to
* useracc? don't have to worry about readahead
* buffers since those pages will be marked
* P_DONE by the buffer release function out of
* biodone when the i/o to the buffer finishes.
*/
if (cur_ldvnmap && do_lockdown) {
nuaccmaps = numvnmaps;
if (xfs_lockdown_iopages(bmapp, isize,
vnmapflags,
&cur_ldvnmap,
&num_ldvnmaps,
uaccmaps, &nuaccmaps,
&useracced)) {
if (useracced)
xfs_unlock_iopages(uaccmaps,
nuaccmaps);
error = XFS_ERROR(ENOMEM);
useracced = 0;
break;
}
}
bp = chunkread(vp, bmapp, read_bmaps, credp);
if (bp->b_flags & B_ERROR) {
error = geterror(bp);
ASSERT(error != EINVAL);
/*
* b_relse functions like chunkhold
* expect B_DONE to be there.
*/
bp->b_flags |= B_DONE|B_STALE;
brelse(bp);
break;
} else if (bp->b_resid != 0) {
buffer_bytes_ok = 0;
bp->b_flags |= B_DONE|B_STALE;
brelse(bp);
break;
} else {
buffer_bytes_ok = 1;
ASSERT((BBTOOFF(bmapp->offset) + bmapp->pboff)
== uiop->uio_offset);
if (!cur_biovnmap) {
error = biomove(bp, bmapp->pboff,
bmapp->pbsize, UIO_READ,
uiop);
} else {
#pragma mips_frequency_hint NEVER
/*
* break up the biomoves so that
* we never biomove across a region
* that might fault on more than
* one inode
*/
error = xfs_mapped_biomove(bp,
bmapp->pboff,
bmapp->pbsize,
UIO_READ, uiop,
&cur_biovnmap,
&num_biovnmaps);
}
if (error) {
bp->b_flags |= B_DONE|B_STALE;
brelse(bp);
break;
}
}
brelse(bp);
if (useracced) {
xfs_unlock_iopages(uaccmaps, nuaccmaps);
useracced = 0;
}
XFSSTATS.xs_read_bufs++;
read_bmaps = 1;
nbmaps--;
bmapp++;
}
if (useracced) {
xfs_unlock_iopages(uaccmaps, nuaccmaps);
useracced = 0;
}
} while (!error && (uiop->uio_resid != 0) && buffer_bytes_ok);
return error;
}
/*
* xfs_read
*
* This is the XFS VOP_READ entry point. It does some minimal
* error checking and then switches out based on the file type.
*/
int
xfs_read(
bhv_desc_t *bdp,
uio_t *uiop,
int ioflag,
cred_t *credp,
flid_t *fl)
{
xfs_inode_t *ip;
int type;
off_t offset;
off_t n;
size_t count;
int error;
int lflag;
vrwlock_t lmode;
xfs_mount_t *mp;
size_t resid;
vnode_t *vp;
vasid_t vasid;
as_verifyvnmap_t vnmap_args;
as_verifyvnmapres_t vnmap_res;
const int numvnmaps = 10;
vnmap_t vnmaps[numvnmaps];
vnmap_t *rvnmaps;
int num_rvnmaps;
int rvnmap_flags;
int rvnmap_size = 0;
xfs_uaccmap_t uaccmap_array[numvnmaps];
xfs_uaccmap_t *uaccmaps;
xfs_fsize_t isize;
#if defined(DEBUG) && defined(UIOSZ_DEBUG)
/*
* Randomly set io size
*/
extern ulong_t random(void);
extern int srandom(int);
timespec_t now; /* current time */
static int seed = 0; /* randomizing seed value */
if (!seed) {
nanotime(&now);
seed = (int)now.tv_sec ^ (int)now.tv_nsec;
srandom(seed);
}
ioflag |= IO_UIOSZ;
uiop->uio_readiolog = (random() & 0x3) + XFS_UIO_MIN_READIO_LOG;
#endif
vp = BHV_TO_VNODE(bdp);
ip = XFS_BHVTOI(bdp);
lmode = VRWLOCK_READ;
lflag = 0;
/*
* need to protect against deadlocks that can occur if the
* biomove touches a virtual address in user space that is
* mapped to the file being read. This only works for
* read/write and pread/pwrite. readv/writev lose.
* direct i/o loses too for now.
*
* note that if someone remaps the user buffer to this file
* while the I/O in progess, we lose, too. instant deadlock.
*/
rvnmaps = NULL;
num_rvnmaps = 0;
rvnmap_flags = 0;
uaccmaps = NULL;
if (uiop->uio_segflg == UIO_USERSPACE && uiop->uio_iovcnt == 1 &&
!(ioflag & IO_DIRECT) && VN_MAPPED(vp)) {
#pragma mips_frequency_hint NEVER
as_lookup_current(&vasid);
VAS_LOCK(vasid, AS_SHARED);
vnmap_args.as_vp = vp;
vnmap_args.as_vaddr = (uvaddr_t)
ctob(btoct(uiop->uio_iov->iov_base));
vnmap_args.as_len = uiop->uio_iov->iov_len +
((__psunsigned_t) uiop->uio_iov->iov_base -
(__psunsigned_t) vnmap_args.as_vaddr);
vnmap_args.as_offstart = uiop->uio_offset;
vnmap_args.as_offend = uiop->uio_offset + uiop->uio_resid;
vnmap_args.as_vnmap = vnmaps;
vnmap_args.as_nmaps = numvnmaps;
if (error = VAS_VERIFYVNMAP(vasid, &vnmap_args, &vnmap_res)) {
VAS_UNLOCK(vasid);
return error;
}
VAS_UNLOCK(vasid);
if (vnmap_res.as_rescodes) {
rvnmaps = (vnmap_res.as_multimaps == NULL)
? vnmaps
: vnmap_res.as_multimaps;
rvnmap_flags = vnmap_res.as_rescodes;
num_rvnmaps = vnmap_res.as_nmaps;
rvnmap_size = vnmap_res.as_mapsize;
if (num_rvnmaps <= numvnmaps)
uaccmaps = uaccmap_array;
else if ((uaccmaps = kmem_alloc(num_rvnmaps *
sizeof(xfs_uaccmap_t),
KM_SLEEP)) == NULL) {
error = XFS_ERROR(ENOMEM);
goto out;
}
}
}
/*
* check if we're in recursive lock mode (a read inside a biomove
* to a page that is mapped to ip that faulted)
*/
lflag = xfs_is_nested_locking_enabled()
? XFS_IOLOCK_NESTED
: 0;
if (!(ioflag & IO_ISLOCKED)) {
/* For calls from the paging system where the faulting
* process is multithreaded, try to grab the I/O lock,
* if it is already held, then we ask the paging system
* to try again by returning EAGAIN. It's safe to return
* directly since the UIO_NOSPACE i/o never takes the
* aspacelock (VAS_LOCK) above.
*/
if ((uiop->uio_segflg == UIO_NOSPACE) &&
(ioflag & IO_MTTHREAD) && VN_MAPPED(vp)) {
if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED|lflag)) {
return XFS_ERROR(EAGAIN);
}
} else {
xfs_rwlockf(bdp, VRWLOCK_READ, lflag);
}
}
/*
* if we're in a possible mmap autogrow case, check to
* see if a biomove is going to have to grow the file.
* if so, drop the iolock and re-obtain it in write mode.
* it's possible that someone might have grown the file
* while we were re-acquiring the lock. if so, then we
* demote the iolock from exclusive back to shared and
* proceed onwards.
*/
isize = -1;
if (rvnmap_flags & AS_VNMAP_AUTOGROW) {
xfs_ilock(ip, XFS_ILOCK_SHARED);
if (vnmap_res.as_maxoffset <= ip->i_d.di_size)
xfs_iunlock(ip, XFS_ILOCK_SHARED);
else {
/*
* note, we don't have to worry about the
* multi-threaded ilock_nowait case above
* because the fault path will never biomove
* a page and cause an autogrow fault
*/
xfs_iunlock(ip, XFS_ILOCK_SHARED);
xfs_rwunlockf(bdp, VRWLOCK_READ, lflag);
xfs_rwlockf(bdp, VRWLOCK_WRITE, lflag);
xfs_ilock(ip, XFS_ILOCK_SHARED);
if (vnmap_res.as_maxoffset > ip->i_d.di_size) {
isize = ip->i_d.di_size;
xfs_iunlock(ip, XFS_ILOCK_SHARED);
lmode = VRWLOCK_WRITE;
} else {
xfs_iunlock(ip, XFS_ILOCK_SHARED);
xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
}
}
}
type = ip->i_d.di_mode & IFMT;
ASSERT(type == IFDIR ||
ismrlocked(&ip->i_iolock, MR_ACCESS | MR_UPDATE) != 0);
ASSERT(type == IFREG || type == IFDIR ||
type == IFLNK || type == IFSOCK);
offset = uiop->uio_offset;
count = uiop->uio_resid;
/* check for locks if some exist and mandatory locking is enabled */
if ((vp->v_flag & (VENF_LOCKING|VFRLOCKS)) ==
(VENF_LOCKING|VFRLOCKS)) {
error = fs_checklock(vp, FREAD, offset, count, uiop->uio_fmode,
credp, fl, VRWLOCK_READ);
if (error)
goto out;
}
if (offset < 0) {
error = XFS_ERROR(EINVAL);
goto out;
}
if (count <= 0) {
error = 0;
goto out;
}
if (ioflag & IO_RSYNC) {
/* First we sync the data */
if ((ioflag & IO_SYNC) || (ioflag & IO_DSYNC)) {
chunkpush(vp, offset, offset + count , 0);
}
/* Now we sync the timestamps */
if (ioflag & IO_SYNC) {
xfs_ilock(ip, XFS_ILOCK_SHARED);
xfs_iflock(ip);
error = xfs_iflush(ip, XFS_IFLUSH_SYNC);
xfs_iunlock(ip, XFS_ILOCK_SHARED);
if (error == EFSCORRUPTED)
goto out;
} else {
if (ioflag & IO_DSYNC) {
mp = ip->i_mount;
xfs_log_force(mp, (xfs_lsn_t)0,
XFS_LOG_FORCE | XFS_LOG_SYNC );
}
}
}
switch (type) {
case IFREG:
/*
* Don't allow reads to pass down counts which could
* overflow. Be careful to record the part that we
* refuse so that we can add it back into uio_resid
* so that the caller will see a short read.
*/
n = XFS_MAX_FILE_OFFSET - offset;
if (n <= 0) {
error = 0;
goto out;
}
if (n < uiop->uio_resid) {
resid = uiop->uio_resid - n;
uiop->uio_resid = n;
} else {
resid = 0;
}
/*
* Not ready for in-line files yet.
*/
ASSERT((ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS) ||
(ip->i_d.di_format == XFS_DINODE_FMT_BTREE));
if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) &&
!(ioflag & IO_INVIS)) {
vrwlock_t locktype = VRWLOCK_READ;
error = xfs_dm_send_data_event(DM_EVENT_READ, bdp,
offset, count,
UIO_DELAY_FLAG(uiop), &locktype);
if (error)
goto out;
}
/*
* Respect preferred read size if indicated in uio structure.
* But if the read size has already been set, go with the
* smallest value. Silently ignore requests that aren't
* within valid I/O size limits.
*/
mp = ip->i_mount;
if ((ioflag & IO_UIOSZ) &&
uiop->uio_readiolog != ip->i_readio_log &&
uiop->uio_readiolog >= mp->m_sb.sb_blocklog &&
uiop->uio_readiolog >= XFS_UIO_MIN_READIO_LOG &&
uiop->uio_readiolog <= XFS_UIO_MAX_READIO_LOG) {
xfs_ilock(ip, XFS_ILOCK_EXCL);
#if !(defined(DEBUG) && defined(UIOSZ_DEBUG))
if (!(ip->i_flags & XFS_IUIOSZ) ||
uiop->uio_readiolog < ip->i_readio_log) {
#endif /* ! (DEBUG && UIOSZ_DEBUG) */
ip->i_readio_log = uiop->uio_readiolog;
ip->i_readio_blocks = 1 <<
(int) (ip->i_readio_log -
mp->m_sb.sb_blocklog);
/*
* set inode max io field to largest
* possible value that could have been
* applied to the inode
*/
if (!(ip->i_flags & XFS_IUIOSZ)) {
ip->i_max_io_log = MAX(ip->i_max_io_log,
MAX(mp->m_readio_log,
ip->i_readio_log));
ip->i_flags |= XFS_IUIOSZ;
}
#if defined(DEBUG) && defined(UIOSZ_DEBUG)
atomicAddInt(&uiodbg_switch, 1);
atomicAddInt(
&(uiodbg_readiolog[ip->i_readio_log -
XFS_UIO_MIN_READIO_LOG]),
1);
#endif
#if !(defined(DEBUG) && defined(UIOSZ_DEBUG))
}
#endif /* ! (DEBUG && UIOSZ_DEBUG) */
xfs_iunlock(ip, XFS_ILOCK_EXCL);
}
if (ioflag & IO_DIRECT) {
error = xfs_diordwr(bdp, uiop, ioflag, credp, B_READ);
} else {
error = xfs_read_file(bdp, uiop, ioflag, credp,
rvnmaps, num_rvnmaps,
rvnmap_flags, uaccmaps, isize);
}
ASSERT(ismrlocked(&ip->i_iolock, MR_ACCESS | MR_UPDATE) != 0);
/* don't update timestamps if doing invisible I/O */
if (!(ioflag & IO_INVIS)) {
xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
}
/*
* Add back whatever we refused to do because of file
* size limitations.
*/
uiop->uio_resid += resid;
break;
case IFDIR:
error = XFS_ERROR(EISDIR);
break;
case IFLNK:
error = XFS_ERROR(EINVAL);
break;
case IFSOCK:
error = XFS_ERROR(ENODEV);
break;
default:
ASSERT(0);
error = XFS_ERROR(EINVAL);
break;
}
out:
if (rvnmap_size > 0)
kmem_free(rvnmaps, rvnmap_size);
if (num_rvnmaps > numvnmaps)
kmem_free(uaccmaps, num_rvnmaps * sizeof(xfs_uaccmap_t));
if (!(ioflag & IO_ISLOCKED))
xfs_rwunlockf(bdp, lmode, lflag);
return error;
}
/*
* Map the given I/O size and I/O alignment over the given extent.
* If we're at the end of the file and the underlying extent is
* delayed alloc, make sure we extend out to the
* next i_writeio_blocks boundary. Otherwise make sure that we
* are confined to the given extent.
*/
/*ARGSUSED*/
STATIC void
xfs_write_bmap(
xfs_mount_t *mp,
xfs_bmbt_irec_t *imapp,
struct bmapval *bmapp,
int iosize,
xfs_fileoff_t ioalign,
xfs_fsize_t isize)
{
__int64_t extra_blocks;
xfs_fileoff_t size_diff;
xfs_fileoff_t ext_offset;
xfs_fsblock_t start_block;
if (ioalign < imapp->br_startoff) {
/*
* The desired alignment doesn't end up on this
* extent. Move up to the beginning of the extent.
* Subtract whatever we drop from the iosize so that
* we stay aligned on iosize boundaries.
*/
size_diff = imapp->br_startoff - ioalign;
iosize -= (int)size_diff;
ASSERT(iosize > 0);
ext_offset = 0;
bmapp->offset = imapp->br_startoff;
} else {
/*
* The alignment requested fits on this extent,
* so use it.
*/
ext_offset = ioalign - imapp->br_startoff;
bmapp->offset = ioalign;
}
start_block = imapp->br_startblock;
ASSERT(start_block != HOLESTARTBLOCK);
if (start_block != DELAYSTARTBLOCK) {
bmapp->bn = start_block + ext_offset;
bmapp->eof = (imapp->br_state != XFS_EXT_UNWRITTEN) ?
0 : BMAP_UNWRITTEN;
} else {
bmapp->bn = -1;
bmapp->eof = BMAP_DELAY;
}
bmapp->length = iosize;
/*
* If the iosize from our offset extends beyond the end of
* the extent, then trim down length to match that of the extent.
*/
extra_blocks = (off_t)(bmapp->offset + bmapp->length) -
(__uint64_t)(imapp->br_startoff +
imapp->br_blockcount);
if (extra_blocks > 0) {
bmapp->length -= extra_blocks;
ASSERT(bmapp->length > 0);
}
bmapp->bsize = XFS_FSB_TO_B(mp, bmapp->length);
}
/*
* This routine is called to handle zeroing any space in the last
* block of the file that is beyond the EOF. We do this since the
* size is being increased without writing anything to that block
* and we don't want anyone to read the garbage on the disk.
*/
/* ARGSUSED */
STATIC int /* error */
xfs_zero_last_block(
xfs_inode_t *ip,
off_t offset,
xfs_fsize_t isize,
cred_t *credp,
struct pm *pmp)
{
xfs_fileoff_t last_fsb;
xfs_fileoff_t next_fsb;
xfs_fileoff_t end_fsb;
xfs_fsblock_t firstblock;
xfs_mount_t *mp;
buf_t *bp;
vnode_t *vp;
int nimaps;
int zero_offset;
int zero_len;
int isize_fsb_offset;
int i;
int error;
int hole;
pfd_t *pfdp;
xfs_bmbt_irec_t imap;
struct bmapval bmap;
ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
ASSERT(offset > isize);
mp = ip->i_mount;
vp = XFS_ITOV(ip);
/*
* If the file system block size is less than the page size,
* then there could be bytes in the last page after the last
* fsblock containing isize which have not been initialized.
* Since if such a page is in memory it will be marked P_DONE
* and may now be fully accessible, we need to zero any part of
* it which is beyond the old file size. We don't need to send
* this out to disk, we're just iniitializing it to zeroes like
* we would have done in xfs_strat_read() had the size been bigger.
*/
if ((mp->m_sb.sb_blocksize < NBPP) && ((i = poff(isize)) != 0)) {
pfdp = pfind(vp, offtoct(isize), VM_ATTACH);
if (pfdp != NULL) {
page_zero(pfdp, NOCOLOR, i, (NBPP - i));
/*
* Now we check to see if there are any holes in the
* page over the end of the file that are beyond the
* end of the file. If so, we want to set the P_HOLE
* flag in the page and blow away any active mappings
* to it so that future faults on the page will cause
* the space where the holes are to be allocated.
* This keeps us from losing updates that are beyond
* the current end of file when the page is already
* in memory.
*/
next_fsb = XFS_B_TO_FSBT(mp, isize);
end_fsb = XFS_B_TO_FSB(mp, ctooff(offtoc(isize)));
hole = 0;
while (next_fsb < end_fsb) {
nimaps = 1;
firstblock = NULLFSBLOCK;
error = xfs_bmapi(NULL, ip, next_fsb, 1, 0,
&firstblock, 0, &imap,
&nimaps, NULL);
if (error) {
pagefree(pfdp);
return error;
}
ASSERT(nimaps > 0);
if (imap.br_startblock == HOLESTARTBLOCK) {
hole = 1;
break;
}
next_fsb++;
}
if (hole) {
/*
* In order to make processes notice the
* newly set P_HOLE flag, blow away any
* mappings to the file. We have to drop
* the inode lock while doing this to avoid
* deadlocks with the chunk cache.
*/
if (VN_MAPPED(vp)) {
xfs_iunlock(ip, XFS_ILOCK_EXCL);
VOP_PAGES_SETHOLE(vp, pfdp, 1, 1,
ctooff(offtoct(isize)));
xfs_ilock(ip, XFS_ILOCK_EXCL);
}
}
pagefree(pfdp);
}
}
isize_fsb_offset = XFS_B_FSB_OFFSET(mp, isize);
if (isize_fsb_offset == 0) {
/*
* There are not extra bytes in the last block to
* zero, so return.
*/
return 0;
}
last_fsb = XFS_B_TO_FSBT(mp, isize);
nimaps = 1;
firstblock = NULLFSBLOCK;
error = xfs_bmapi(NULL, ip, last_fsb, 1, 0, &firstblock, 0, &imap,
&nimaps, NULL);
if (error) {
return error;
}
ASSERT(nimaps > 0);
/*
* If the block underlying isize is just a hole, then there
* is nothing to zero.
*/
if (imap.br_startblock == HOLESTARTBLOCK) {
return 0;
}
/*
* Get a buffer for the last block, zero the part beyond the
* EOF, and write it out sync. We need to drop the ilock
* while we do this so we don't deadlock when the buffer cache
* calls back to us.
*/
xfs_iunlock(ip, XFS_ILOCK_EXCL);
bmap.offset = XFS_FSB_TO_BB(mp, last_fsb);
bmap.length = XFS_FSB_TO_BB(mp, 1);
bmap.bsize = BBTOB(bmap.length);
bmap.pboff = 0;
bmap.pbsize = bmap.bsize;
if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) {
bmap.pbdev = mp->m_rtdev;
} else {
bmap.pbdev = mp->m_dev;
}
bmap.eof = BMAP_EOF;
bmap.pmp = pmp;
if (imap.br_startblock != DELAYSTARTBLOCK) {
bmap.bn = XFS_FSB_TO_DB(ip, imap.br_startblock);
if (imap.br_state == XFS_EXT_UNWRITTEN)
bmap.eof |= BMAP_UNWRITTEN;
} else {
bmap.bn = -1;
bmap.eof |= BMAP_DELAY;
}
bp = chunkread(vp, &bmap, 1, credp);
if (bp->b_flags & B_ERROR) {
error = geterror(bp);
bp->b_flags |= B_DONE|B_STALE;
brelse(bp);
xfs_ilock(ip, XFS_ILOCK_EXCL);
return error;
}
zero_offset = isize_fsb_offset;
zero_len = mp->m_sb.sb_blocksize - isize_fsb_offset;
xfs_zero_bp(bp, zero_offset, zero_len);
/*
* We don't want to start a transaction here, so don't
* push out a buffer over a delayed allocation extent.
* Also, we can get away with it since the space isn't
* allocated so it's faster anyway.
*
* We don't bother to call xfs_b*write here since this is
* just userdata, and we don't want to bring the filesystem
* down if they hit an error. Since these will go through
* xfsstrategy anyway, we have control over whether to let the
* buffer go thru or not, in case of a forced shutdown.
*/
ASSERT(bp->b_vp);
if (imap.br_startblock == DELAYSTARTBLOCK ||
imap.br_state == XFS_EXT_UNWRITTEN) {
bdwrite(bp);
} else {
error = bwrite(bp);
}
xfs_ilock(ip, XFS_ILOCK_EXCL);
return error;
}
/*
* Zero any on disk space between the current EOF and the new,
* larger EOF. This handles the normal case of zeroing the remainder
* of the last block in the file and the unusual case of zeroing blocks
* out beyond the size of the file. This second case only happens
* with fixed size extents and when the system crashes before the inode
* size was updated but after blocks were allocated. If fill is set,
* then any holes in the range are filled and zeroed. If not, the holes
* are left alone as holes.
*/
int /* error */
xfs_zero_eof(
xfs_inode_t *ip,
off_t offset,
xfs_fsize_t isize,
cred_t *credp,
struct pm *pmp)
{
xfs_fileoff_t start_zero_fsb;
xfs_fileoff_t end_zero_fsb;
xfs_fileoff_t prev_zero_fsb;
xfs_fileoff_t zero_count_fsb;
xfs_fileoff_t last_fsb;
xfs_fsblock_t firstblock;
xfs_extlen_t buf_len_fsb;
xfs_extlen_t prev_zero_count;
xfs_mount_t *mp;
buf_t *bp;
int nimaps;
int error;
xfs_bmbt_irec_t imap;
struct bmapval bmap;
pfd_t *pfdp;
int i;
vnode_t *vp;
int length;
ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE));
ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE));
mp = ip->i_mount;
vp = XFS_ITOV(ip);
/*
* First handle zeroing the block on which isize resides.
* We only zero a part of that block so it is handled specially.
*/
error = xfs_zero_last_block(ip, offset, isize, credp, pmp);
if (error) {
ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE));
ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE));
return error;
}
/*
* Calculate the range between the new size and the old
* where blocks needing to be zeroed may exist. To get the
* block where the last byte in the file currently resides,
* we need to subtract one from the size and truncate back
* to a block boundary. We subtract 1 in case the size is
* exactly on a block boundary.
*/
last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
if (last_fsb == end_zero_fsb) {
/*
* The size was only incremented on its last block.
* We took care of that above, so just return.
*/
return 0;
}
ASSERT(start_zero_fsb <= end_zero_fsb);
prev_zero_fsb = NULLFILEOFF;
prev_zero_count = 0;
while (start_zero_fsb <= end_zero_fsb) {
nimaps = 1;
zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
firstblock = NULLFSBLOCK;
error = xfs_bmapi(NULL, ip, start_zero_fsb, zero_count_fsb,
0, &firstblock, 0, &imap, &nimaps, NULL);
if (error) {
ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE));
ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE));
return error;
}
ASSERT(nimaps > 0);
if (imap.br_startblock == HOLESTARTBLOCK) {
/*
* This loop handles initializing pages that were
* partially initialized by the code below this
* loop. It basically zeroes the part of the page
* that sits on a hole and sets the page as P_HOLE
* and calls remapf if it is a mapped file.
*/
if ((prev_zero_fsb != NULLFILEOFF) &&
(dtopt(XFS_FSB_TO_BB(mp, prev_zero_fsb)) ==
dtopt(XFS_FSB_TO_BB(mp, imap.br_startoff)) ||
dtopt(XFS_FSB_TO_BB(mp, prev_zero_fsb +
prev_zero_count)) ==
dtopt(XFS_FSB_TO_BB(mp, imap.br_startoff)))) {
pfdp = pfind(vp, dtopt(XFS_FSB_TO_BB(mp,
imap.br_startoff)), VM_ATTACH);
if (pfdp != NULL) {
i = poff(XFS_FSB_TO_B(mp,
imap.br_startoff));
length = MIN(NBPP - i, XFS_FSB_TO_B(mp,
imap.br_blockcount));
page_zero(pfdp, NOCOLOR, i, length);
if (VN_MAPPED(vp))
VOP_PAGES_SETHOLE(vp, pfdp, 1, 1,
ctooff(offtoct(XFS_FSB_TO_B(mp,
imap.br_startoff))));
pagefree(pfdp);
}
}
prev_zero_fsb = NULLFILEOFF;
prev_zero_count = 0;
start_zero_fsb = imap.br_startoff +
imap.br_blockcount;
ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
continue;
}
/*
* There are blocks in the range requested.
* Zero them a single write at a time. We actually
* don't zero the entire range returned if it is
* too big and simply loop around to get the rest.
* That is not the most efficient thing to do, but it
* is simple and this path should not be exercised often.
*/
buf_len_fsb = XFS_FILBLKS_MIN(imap.br_blockcount,
ip->i_writeio_blocks);
/*
* Drop the inode lock while we're doing the I/O.
* We'll still have the iolock to protect us.
*/
xfs_iunlock(ip, XFS_ILOCK_EXCL);
bmap.offset = XFS_FSB_TO_BB(mp, imap.br_startoff);
bmap.length = XFS_FSB_TO_BB(mp, buf_len_fsb);
bmap.bsize = BBTOB(bmap.length);
bmap.eof = BMAP_EOF;
bmap.pmp = pmp;
if (imap.br_startblock == DELAYSTARTBLOCK) {
bmap.eof |= BMAP_DELAY;
bmap.bn = -1;
} else {
bmap.bn = XFS_FSB_TO_DB(ip, imap.br_startblock);
if (imap.br_state == XFS_EXT_UNWRITTEN)
bmap.eof |= BMAP_UNWRITTEN;
}
if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) {
bmap.pbdev = mp->m_rtdev;
} else {
bmap.pbdev = mp->m_dev;
}
bp = getchunk(XFS_ITOV(ip), &bmap, credp);
#ifdef _VCE_AVOIDANCE
if (vce_avoidance) {
extern void biozero(struct buf *, u_int, int);
biozero(bp, 0, bmap.bsize);
} else
#endif
{
bp_mapin(bp);
bzero(bp->b_un.b_addr, bp->b_bcount);
}
ASSERT(bp->b_vp);
buftrace("XFS ZERO EOF", bp);
if (imap.br_startblock == DELAYSTARTBLOCK ||
imap.br_state == XFS_EXT_UNWRITTEN) {
bdwrite(bp);
} else {
error = bwrite(bp);
if (error) {
xfs_ilock(ip, XFS_ILOCK_EXCL);
return error;
}
}
prev_zero_fsb = start_zero_fsb;
prev_zero_count = buf_len_fsb;
start_zero_fsb = imap.br_startoff + buf_len_fsb;
ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
xfs_ilock(ip, XFS_ILOCK_EXCL);
}
return 0;
}
STATIC int
xfs_iomap_write(
xfs_inode_t *ip,
off_t offset,
size_t count,
struct bmapval *bmapp,
int *nbmaps,
int ioflag,
struct pm *pmp)
{
xfs_fileoff_t offset_fsb;
xfs_fileoff_t ioalign;
xfs_fileoff_t next_offset_fsb;
xfs_fileoff_t last_fsb;
xfs_fileoff_t bmap_end_fsb;
xfs_fileoff_t last_file_fsb;
xfs_fileoff_t start_fsb;
xfs_filblks_t count_fsb;
off_t aligned_offset;
xfs_fsize_t isize;
xfs_fsblock_t firstblock;
__uint64_t last_page_offset;
int nimaps;
int error;
int n;
unsigned int iosize;
unsigned int writing_bytes;
short filled_bmaps;
short x;
short small_write;
size_t count_remaining;
xfs_mount_t *mp;
struct bmapval *curr_bmapp;
struct bmapval *next_bmapp;
struct bmapval *last_bmapp;
xfs_bmbt_irec_t *curr_imapp;
xfs_bmbt_irec_t *last_imapp;
#define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
int aeof;
ASSERT(ismrlocked(&ip->i_lock, MR_UPDATE) != 0);
xfs_iomap_enter_trace(XFS_IOMAP_WRITE_ENTER, ip, offset, count);
mp = ip->i_mount;
ASSERT(! XFS_NOT_DQATTACHED(mp, ip));
if (ip->i_new_size > ip->i_d.di_size) {
isize = ip->i_new_size;
} else {
isize = ip->i_d.di_size;
}
aeof = 0;
offset_fsb = XFS_B_TO_FSBT(mp, offset);
last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
/*
* If the caller is doing a write at the end of the file,
* then extend the allocation (and the buffer used for the write)
* out to the file system's write iosize. We clean up any extra
* space left over when the file is closed in xfs_inactive().
* We can only do this if we are sure that we will create buffers
* over all of the space we allocate beyond the end of the file.
* Not doing so would allow us to create delalloc blocks with
* no pages in memory covering them. So, we need to check that
* there are not any real blocks in the area beyond the end of
* the file which we are optimistically going to preallocate. If
* there are then our buffers will stop when they encounter them
* and we may accidentally create delalloc blocks beyond them
* that we never cover with a buffer. All of this is because
* we are not actually going to write the extra blocks preallocated
* at this point.
*
* We don't bother with this for sync writes, because we need
* to minimize the amount we write for good performance.
*/
if (!(ioflag & IO_SYNC) && ((offset + count) > ip->i_d.di_size)) {
start_fsb = XFS_B_TO_FSBT(mp,
((xfs_ufsize_t)(offset + count - 1)));
count_fsb = ip->i_writeio_blocks;
while (count_fsb > 0) {
nimaps = XFS_WRITE_IMAPS;
firstblock = NULLFSBLOCK;
error = xfs_bmapi(NULL, ip, start_fsb, count_fsb,
0, &firstblock, 0, imap, &nimaps,
NULL);
if (error) {
return error;
}
for (n = 0; n < nimaps; n++) {
if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
(imap[n].br_startblock != DELAYSTARTBLOCK)) {
goto write_map;
}
start_fsb += imap[n].br_blockcount;
count_fsb -= imap[n].br_blockcount;
ASSERT(count_fsb < 0xffff000);
}
}
iosize = ip->i_writeio_blocks;
aligned_offset = XFS_WRITEIO_ALIGN(ip, (offset + count - 1));
ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
last_fsb = ioalign + iosize;
aeof = 1;
}
write_map:
nimaps = XFS_WRITE_IMAPS;
firstblock = NULLFSBLOCK;
/*
* roundup the allocation request to m_dalign boundary if file size
* is greater that 512K and we are allocating past the allocation eof
*/
if (mp->m_dalign && (ip->i_d.di_size >= 524288) && aeof) {
int eof;
xfs_fileoff_t new_last_fsb;
new_last_fsb = roundup(last_fsb, mp->m_dalign);
error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
if (error) {
return error;
}
if (eof)
last_fsb = new_last_fsb;
}
error = xfs_bmapi(NULL, ip, offset_fsb,
(xfs_filblks_t)(last_fsb - offset_fsb),
XFS_BMAPI_DELAY | XFS_BMAPI_WRITE |
XFS_BMAPI_ENTIRE, &firstblock, 1, imap,
&nimaps, NULL);
/*
* This can be EDQUOT, if nimaps == 0
*/
if (error) {
return error;
}
/*
* If bmapi returned us nothing, and if we didn't get back EDQUOT,
* then we must have run out of space.
*/
if (nimaps == 0) {
xfs_iomap_enter_trace(XFS_IOMAP_WRITE_NOSPACE,
ip, offset, count);
return XFS_ERROR(ENOSPC);
}
if (!(ioflag & IO_SYNC) ||
((last_fsb - offset_fsb) >= ip->i_writeio_blocks)) {
/*
* For normal or large sync writes, align everything
* into i_writeio_blocks sized chunks.
*/
iosize = ip->i_writeio_blocks;
aligned_offset = XFS_WRITEIO_ALIGN(ip, offset);
ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
small_write = 0;
} else {
/*
* For small sync writes try to minimize the amount
* of I/O we do. Round down and up to the larger of
* page or block boundaries. Set the small_write
* variable to 1 to indicate to the code below that
* we are not using the normal buffer alignment scheme.
*/
if (NBPP > mp->m_sb.sb_blocksize) {
aligned_offset = ctooff(offtoct(offset));
ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
last_page_offset = ctob64(btoc64(offset + count));
iosize = XFS_B_TO_FSBT(mp, last_page_offset -
aligned_offset);
} else {
ioalign = offset_fsb;
iosize = last_fsb - offset_fsb;
}
small_write = 1;
}
/*
* Now map our desired I/O size and alignment over the
* extents returned by xfs_bmapi().
*/
xfs_write_bmap(mp, imap, bmapp, iosize, ioalign, isize);
ASSERT((bmapp->length > 0) &&
(offset >= XFS_FSB_TO_B(mp, bmapp->offset)));
/*
* A bmap is the EOF bmap when it reaches to or beyond the new
* inode size.
*/
bmap_end_fsb = bmapp->offset + bmapp->length;
if (XFS_FSB_TO_B(mp, bmap_end_fsb) >= isize) {
bmapp->eof |= BMAP_EOF;
}
bmapp->pboff = offset - XFS_FSB_TO_B(mp, bmapp->offset);
writing_bytes = bmapp->bsize - bmapp->pboff;
if (writing_bytes > count) {
/*
* The mapping is for more bytes than we're actually
* going to write, so trim writing_bytes so we can
* get bmapp->pbsize right.
*/
writing_bytes = count;
}
bmapp->pbsize = writing_bytes;
xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP,
ip, offset, count, bmapp, imap);
/*
* Map more buffers if the first does not map the entire
* request. We do this until we run out of bmaps, imaps,
* or bytes to write.
*/
last_file_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)isize));
filled_bmaps = 1;
if ((*nbmaps > 1) &&
((nimaps > 1) || (bmapp->offset + bmapp->length <
imap[0].br_startoff + imap[0].br_blockcount)) &&
(writing_bytes < count)) {
curr_bmapp = &bmapp[0];
next_bmapp = &bmapp[1];
last_bmapp = &bmapp[*nbmaps - 1];
curr_imapp = &imap[0];
last_imapp = &imap[nimaps - 1];
count_remaining = count - writing_bytes;
/*
* curr_bmapp is always the last one we filled
* in, and next_bmapp is always the next one to
* be filled in.
*/
while (next_bmapp <= last_bmapp) {
next_offset_fsb = curr_bmapp->offset +
curr_bmapp->length;
if (next_offset_fsb >= last_file_fsb) {
/*
* We've gone beyond the region asked for
* by the caller, so we're done.
*/
break;
}
if (small_write) {
iosize -= curr_bmapp->length;
ASSERT((iosize > 0) ||
(curr_imapp == last_imapp));
/*
* We have nothing more to write, so
* we're done.
*/
if (iosize == 0) {
break;
}
}
if (next_offset_fsb <
(curr_imapp->br_startoff +
curr_imapp->br_blockcount)) {
/*
* I'm still on the same extent, so
* the last bmap must have ended on
* a writeio_blocks boundary. Thus,
* we just start where the last one
* left off.
*/
ASSERT((XFS_FSB_TO_B(mp, next_offset_fsb) &
((1 << (int) ip->i_writeio_log) - 1))
==0);
xfs_write_bmap(mp, curr_imapp, next_bmapp,
iosize, next_offset_fsb,
isize);
} else {
curr_imapp++;
if (curr_imapp <= last_imapp) {
/*
* We're moving on to the next
* extent. Since we try to end
* all buffers on writeio_blocks
* boundaries, round next_offset
* down to a writeio_blocks boundary
* before calling xfs_write_bmap().
*
* For small, sync writes we don't
* bother with the alignment stuff.
*
* XXXajs
* Adding a macro to writeio align
* fsblocks would be good to reduce
* the bit shifting here.
*/
if (small_write) {
ioalign = next_offset_fsb;
} else {
aligned_offset =
XFS_FSB_TO_B(mp,
next_offset_fsb);
aligned_offset =
XFS_WRITEIO_ALIGN(ip,
aligned_offset);
ioalign = XFS_B_TO_FSBT(mp,
aligned_offset);
}
xfs_write_bmap(mp, curr_imapp,
next_bmapp, iosize,
ioalign, isize);
} else {
/*
* We're out of imaps. The caller
* will have to call again to map
* the rest of the write request.
*/
break;
}
}
/*
* The write must start at offset 0 in this bmap
* since we're just continuing from the last
* buffer. Thus the request offset in the buffer
* indicated by pboff must be 0.
*/
next_bmapp->pboff = 0;
/*
* The request size within this buffer is the
* entire buffer unless the count of bytes to
* write runs out.
*/
writing_bytes = next_bmapp->bsize;
if (writing_bytes > count_remaining) {
writing_bytes = count_remaining;
}
next_bmapp->pbsize = writing_bytes;
count_remaining -= writing_bytes;
ASSERT(((long)count_remaining) >= 0);
filled_bmaps++;
curr_bmapp++;
next_bmapp++;
/*
* A bmap is the EOF bmap when it reaches to
* or beyond the new inode size.
*/
bmap_end_fsb = curr_bmapp->offset +
curr_bmapp->length;
if (((xfs_ufsize_t)XFS_FSB_TO_B(mp, bmap_end_fsb)) >=
(xfs_ufsize_t)isize) {
curr_bmapp->eof |= BMAP_EOF;
}
xfs_iomap_map_trace(XFS_IOMAP_WRITE_MAP, ip, offset,
count, curr_bmapp, curr_imapp);
}
}
*nbmaps = filled_bmaps;
for (x = 0; x < filled_bmaps; x++) {
curr_bmapp = &bmapp[x];
if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) {
curr_bmapp->pbdev = mp->m_rtdev;
} else {
curr_bmapp->pbdev = mp->m_dev;
}
curr_bmapp->offset = XFS_FSB_TO_BB(mp, curr_bmapp->offset);
curr_bmapp->length = XFS_FSB_TO_BB(mp, curr_bmapp->length);
ASSERT((x == 0) ||
((bmapp[x - 1].offset + bmapp[x - 1].length) ==
curr_bmapp->offset));
if (curr_bmapp->bn != -1) {
curr_bmapp->bn = XFS_FSB_TO_DB(ip, curr_bmapp->bn);
}
curr_bmapp->pmp = pmp;
}
return 0;
}
int
xfs_write_file(
bhv_desc_t *bdp,
uio_t *uiop,
int ioflag,
cred_t *credp,
xfs_lsn_t *commit_lsn_p,
vnmap_t *vnmaps,
int numvnmaps,
const uint vnmapflags,
xfs_uaccmap_t *uaccmaps)
{
struct bmapval bmaps[XFS_MAX_RW_NBMAPS];
struct bmapval *bmapp;
int nbmaps;
vnode_t *vp;
buf_t *bp;
xfs_inode_t *ip;
int error;
int eof_zeroed;
int fillhole;
int gaps_mapped;
off_t offset;
ssize_t count;
int read;
xfs_fsize_t isize;
xfs_fsize_t new_size;
xfs_mount_t *mp;
int fsynced;
extern void chunkrelse(buf_t*);
int useracced = 0;
vnmap_t *cur_ldvnmap = vnmaps;
int num_ldvnmaps = numvnmaps;
int num_biovnmaps = numvnmaps;
int nuaccmaps;
vnmap_t *cur_biovnmap = vnmaps;
vp = BHV_TO_VNODE(bdp);
ip = XFS_BHVTOI(bdp);
mp = ip->i_mount;
if (XFS_FORCED_SHUTDOWN(mp))
return XFS_ERROR(EIO);
/*
* If the file has fixed size extents or is a real time file,
* buffered I/O cannot be performed.
* This check will be removed in the future.
*/
if ((ip->i_d.di_extsize) ||
(ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
return (XFS_ERROR(EINVAL));
}
/*
* Make sure that the dquots are there
*/
if (XFS_IS_QUOTA_ON(mp)) {
if (XFS_NOT_DQATTACHED(mp, ip)) {
if (error = xfs_qm_dqattach(ip, 0))
return (error);
}
}
error = 0;
eof_zeroed = 0;
gaps_mapped = 0;
XFSSTATS.xs_write_calls++;
XFSSTATS.xs_write_bytes += uiop->uio_resid;
/*
* i_new_size is used by xfs_iomap_read() when the chunk
* cache code calls back into the file system through
* xfs_bmap(). This way we can tell where the end of
* file is going to be even though we haven't yet updated
* ip->i_d.di_size. This is guarded by the iolock and the
* inode lock. Either is sufficient for reading the value.
*/
new_size = uiop->uio_offset + uiop->uio_resid;
/*
* i_write_offset is used by xfs_strat_read() when the chunk
* cache code calls back into the file system through
* xfs_strategy() to initialize a buffer. We use it there
* to know how much of the buffer needs to be zeroed and how
* much will be initialize here by the write or not need to
* be initialized because it will be beyond the inode size.
* This is protected by the io lock.
*/
ip->i_write_offset = uiop->uio_offset;
/*
* Loop until uiop->uio_resid, which is the number of bytes the
* caller has requested to write, goes to 0 or we get an error.
* Each call to xfs_iomap_write() tries to map as much of the
* request as it can in ip->i_writeio_blocks sized chunks.
*/
if (!((ioflag & (IO_NFS3|IO_NFS)) &&
uiop->uio_offset > ip->i_d.di_size &&
uiop->uio_offset - ip->i_d.di_size <= (xfs_nfs_io_units *
(1 << (int) MAX(ip->i_writeio_log,
uiop->uio_writeiolog))))) {
fillhole = 0;
offset = uiop->uio_offset;
count = uiop->uio_resid;
} else {
/*
* Cope with NFS out-of-order writes. If we're
* extending eof to a point within the indicated
* window, fill any holes between old and new eof.
* Set up offset/count so we deal with all the bytes
* between current eof and end of the new write.
*/
fillhole = 1;
offset = ip->i_d.di_size;
count = uiop->uio_offset + uiop->uio_resid - offset;
}
fsynced = 0;
do {
xfs_ilock(ip, XFS_ILOCK_EXCL);
isize = ip->i_d.di_size;
if (new_size > isize) {
ip->i_new_size = new_size;
}
xfs_rw_enter_trace(XFS_WRITE_ENTER, ip, uiop, ioflag);
/*
* If this is the first pass through the loop, then map
* out all of the holes we might fill in with this write
* and list them in the inode's gap list. This is for
* use by xfs_strat_read() in determining if the real
* blocks underlying a delalloc buffer have been initialized
* or not. Since writes are single threaded, if the blocks
* were holes when we started and xfs_strat_read() is asked
* to read one in while we're still here in xfs_write_file(),
* then the block is not initialized. Only we can
* initialize it and once we write out a buffer we remove
* any entries in the gap list which overlap that buffer.
*/
if (!gaps_mapped) {
error = xfs_build_gap_list(ip, offset, count);
if (error) {
goto error0;
}
gaps_mapped = 1;
}
/*
* If we've seeked passed the EOF to do this write,
* then we need to make sure that any buffer overlapping
* the EOF is zeroed beyond the EOF.
*/
if (!eof_zeroed && uiop->uio_offset > isize && isize != 0) {
error = xfs_zero_eof(ip, uiop->uio_offset, isize,
credp, uiop->uio_pmp);
if (error) {
goto error0;
}
eof_zeroed = 1;
}
nbmaps = sizeof(bmaps) / sizeof(bmaps[0]);
error = xfs_iomap_write(ip, offset, count, bmaps, &nbmaps,
ioflag, uiop->uio_pmp);
xfs_iunlock(ip, XFS_ILOCK_EXCL);
/*
* Clear out any read-ahead info since the write may
* have made it invalid.
*/
if (!error)
XFS_INODE_CLEAR_READ_AHEAD(ip);
if (error == ENOSPC) {
switch (fsynced) {
case 0:
VOP_FLUSH_PAGES(vp, 0,
(off_t)xfs_file_last_byte(ip), 0,
FI_NONE, error);
error = 0;
fsynced = 1;
continue;
case 1:
fsynced = 2;
if (!(ioflag & IO_SYNC)) {
ioflag |= IO_SYNC;
error = 0;
continue;
}
/* FALLTHROUGH */
case 2:
case 3:
VFS_SYNC(vp->v_vfsp,
SYNC_NOWAIT|SYNC_BDFLUSH|SYNC_FSDATA,
get_current_cred(), error);
error = 0;
delay(HZ);
fsynced++;
continue;
}
}
if (error || (bmaps[0].pbsize == 0)) {
break;
}
fsynced = 0;
bmapp = &bmaps[0];
/*
* Each pass through the loop writes another buffer
* to the file. For big requests, iomap_write will
* have given up multiple bmaps to use so we make fewer
* calls to it on big requests than if it only gave
* us one at a time.
*
* Error handling is a bit tricky because of delayed
* allocation. We need to make sure that we create
* dirty buffers over all the delayed allocation
* extents created in xfs_iomap_write(). Thus, when
* we get an error we continue to process each of
* the bmaps returned by xfs_iomap_write(). Each is
* read in so that it is fully initialized and then
* written out without actually copying in the user's
* data.
*/
while (((uiop->uio_resid != 0) || (error != 0)) &&
(nbmaps > 0)) {
/*
* do we have overlapping pages we need to
* useracc? don't have to worry about autogrow
* case here as those have been failed earlier
*/
if (cur_ldvnmap && vnmapflags & AS_VNMAP_OVERLAP) {
nuaccmaps = numvnmaps;
/*
* tell xfs_lockdown_iopages to skip
* the autogrow checking since any
* necessary file growing was handled
* at the beginning of xfs_write()
*/
if (error = xfs_lockdown_iopages(bmapp, -1,
vnmapflags,
&cur_ldvnmap,
&num_ldvnmaps,
uaccmaps, &nuaccmaps,
&useracced)) {
if (useracced)
xfs_unlock_iopages(uaccmaps,
nuaccmaps);
error = XFS_ERROR(ENOMEM);
useracced = 0;
}
}
/*
* If the write doesn't completely overwrite
* the buffer and we're not writing from
* the beginning of the buffer to the end
* of the file then we need to read the
* buffer. We also always want to read the
* buffer if we've encountered an error and
* we're just cleaning up.
*
* Reading the buffer will send it to xfs_strategy
* which will take care of zeroing the holey
* parts of the buffer and coordinating with
* other, simultaneous writers.
*/
if ((error != 0) ||
((bmapp->pbsize != bmapp->bsize) &&
!((bmapp->pboff == 0) &&
(uiop->uio_offset >= isize)))) {
bp = chunkread(vp, bmapp, 1, credp);
read = 1;
} else {
bp = getchunk(vp, bmapp, credp);
read = 0;
}
/*
* There is not much we can do with buffer errors.
* The assumption here is that the space underlying
* the buffer must now be allocated (even if it
* wasn't when we mapped the buffer) and we got an
* error reading from it. In this case the blocks
* will remain unreadable, so we just toss the buffer
* and its associated pages.
*/
if (bp->b_flags & B_ERROR) {
error = geterror(bp);
ASSERT(error != EINVAL);
bp->b_flags |= B_DONE|B_STALE;
bp->b_flags &= ~(B_DELWRI);
brelse(bp);
bmapp++;
nbmaps--;
continue;
}
/*
* If we've already encountered an error, then
* write the buffers out without copying the user's
* data into them. This way we get dirty buffers
* over our delayed allocation extents which
* have been initialized by xfs_strategy() since
* we forced the chunkread() above.
* We write the data out synchronously here so that
* we don't have to worry about having buffers
* possibly out beyond the EOF when we later flush
* or truncate the file. We set the B_STALE bit so
* that the buffer will be decommissioned after it
* is synced out.
*/
if (error != 0) {
bp->b_flags |= B_STALE;
(void) bwrite(bp);
bmapp++;
nbmaps--;
continue;
}
ASSERT(fillhole || fillhole == 0 &&
BBTOOFF(bmapp->offset) + bmapp->pboff
== uiop->uio_offset);
/*
* zero the bytes up to the data being written
* but don't overwrite data we read in. This
* zero-fills the buffers we set up for the NFS
* case to fill the holes between EOF and the new
* write.
*/
if (!read && BBTOOFF(bmapp->offset) + bmapp->pboff
< uiop->uio_offset) {
xfs_zero_bp(bp, bmapp->pboff,
MIN(bmapp->pbsize, uiop->uio_offset -
(BBTOOFF(bmapp->offset) +
bmapp->pboff)));
}
/*
* biomove the data in the region to be written.
* In the NFS hole-filling case, don't fill
* anything until we hit the first buffer with
* data that we have to write.
*/
if (!fillhole) {
if (!cur_biovnmap) {
error = biomove(bp, bmapp->pboff,
bmapp->pbsize,
UIO_WRITE,
uiop);
} else {
#pragma mips_frequency_hint NEVER
/*
* break up the biomoves so that
* we never biomove across a region
* that might fault on more than
* one inode
*/
error = xfs_mapped_biomove(bp,
bmapp->pboff,
bmapp->pbsize,
UIO_WRITE, uiop,
&cur_biovnmap,
&num_biovnmaps);
}
} else if (BBTOOFF(bmapp->offset) + bmapp->pboff +
bmapp->pbsize >= uiop->uio_offset) {
/*
* NFS - first buffer to be written. biomove
* into the portion of the buffer that the
* user originally asked to write to.
*/
ASSERT(BBTOOFF(bmapp->offset) + bmapp->pboff
<= uiop->uio_offset);
if (!cur_biovnmap) {
error = biomove(bp,
uiop->uio_offset -
BBTOOFF(bmapp->offset),
bmapp->pbsize -
(int)(uiop->uio_offset -
(BBTOOFF(bmapp->offset) +
bmapp->pboff)),
UIO_WRITE,
uiop);
} else {
#pragma mips_frequency_hint NEVER
/*
* break up the biomoves so that
* we never biomove across a region
* that might fault on more than
* one inode
*/
error = xfs_mapped_biomove(bp,
uiop->uio_offset -
BBTOOFF(bmapp->offset),
bmapp->pbsize -
(int)(uiop->uio_offset -
(BBTOOFF(bmapp->offset) +
bmapp->pboff)),
UIO_WRITE,
uiop,
&cur_biovnmap,
&num_biovnmaps);
}
/*
* turn off hole-filling code. The rest
* of the buffers can be handled as per
* the usual write path.
*/
fillhole = 0;
}
if (error) {
/*
* If the buffer is already done then just
* mark it dirty without copying any more
* data into it. It is already fully
* initialized.
* Otherwise, we must have getchunk()'d
* the buffer above. Use chunkreread()
* to get it initialized by xfs_strategy()
* and then write it out.
* We write the data out synchronously here
* so that we don't have to worry about
* having buffers possibly out beyond the
* EOF when we later flush or truncate
* the file. We set the B_STALE bit so
* that the buffer will be decommissioned
* after it is synced out.
*/
if (useracced) {
xfs_unlock_iopages(uaccmaps,
nuaccmaps);
useracced = 0;
}
if (!(bp->b_flags & B_DONE)) {
chunkreread(bp);
}
bp->b_flags |= B_STALE;
(void) bwrite(bp);
bmapp++;
nbmaps--;
continue;
}
/*
* reset offset/count to reflect the biomove
*/
if (!fillhole) {
offset = uiop->uio_offset;
count = uiop->uio_resid;
} else {
/*
* NFS - set offset to the beginning
* of the next area in the file to be
* copied or zero-filled. Drop count
* by the the amount we just zero-filled.
*/
offset = BBTOOFF(bmapp->offset) +
bmapp->pboff + bmapp->pbsize;
count -= bmapp->pbsize;
}
/*
* Make sure that any gap list entries overlapping
* the buffer being written are removed now that
* we know that the blocks underlying the buffer
* will be initialized. We don't need the inode
* lock to manipulate the gap list here, because
* we have the io lock held exclusively so noone
* else can get to xfs_strat_read() where we look
* at the list.
*/
xfs_delete_gap_list(ip,
XFS_BB_TO_FSBT(mp, bp->b_offset),
XFS_B_TO_FSBT(mp, bp->b_bcount));
ASSERT(bp->b_vp);
if ((ioflag & IO_SYNC) || (ioflag & IO_DSYNC)) {
if ((bmapp->pboff + bmapp->pbsize) ==
bmapp->bsize) {
bp->b_relse = chunkrelse;
}
/*
* save the commit lsn for dsync writes
* so xfs_write can force the log up to
* the appropriate point. don't bother
* doing this for sync writes since xfs_write
* will have to kick off a sync xaction to
* log the timestamp updates anyway.
*/
if (ioflag & IO_DSYNC) {
bp->b_fsprivate3 = commit_lsn_p;
bp->b_flags |= B_HOLD;
}
error = bwrite(bp);
if (ioflag & IO_DSYNC) {
bp->b_fsprivate3 = NULL;
bp->b_flags &= ~B_HOLD;
brelse(bp);
}
} else {
bdwrite(bp);
}
if (useracced) {
xfs_unlock_iopages(uaccmaps, nuaccmaps);
useracced = 0;
}
/*
* If we've grown the file, get back the
* inode lock and move di_size up to the
* new size. It may be that someone else
* made it even bigger, so be careful not
* to shrink it.
*
* No one could have shrunk the file, because
* we are holding the iolock exclusive.
*
* Have to update di_size after brelsing the buffer
* because if we are running low on buffers and
* xfsd is trying to push out a delalloc buffer for
* our inode, then it grabs the ilock in exclusive
* mode to do an allocation, and calls get_buf to
* get to read in a metabuffer (agf, agfl). If the
* metabuffer is in the buffer cache, but it gets
* reused before we can grab the cpsema(), then
* we will sleep in get_buf waiting for it to be
* released whilst holding the ilock.
* If it so happens that the buffer was reused by
* the above code path, then we end up holding this
* buffer locked whilst we try to get the ilock so we
* end up deadlocking. (bug 504578).
* For the IO_SYNC writes, the di_size now gets logged
* and synced to disk in the transaction in xfs_write().
*/
if (offset > isize) {
xfs_ilock(ip, XFS_ILOCK_EXCL);
if (offset > ip->i_d.di_size) {
ip->i_d.di_size = offset;
ip->i_update_core = 1;
ip->i_update_size = 1;
isize = offset;
}
xfs_iunlock(ip, XFS_ILOCK_EXCL);
}
XFSSTATS.xs_write_bufs++;
bmapp++;
nbmaps--;
}
} while ((uiop->uio_resid > 0) && !error);
/*
* Free up any remaining entries in the gap list, because the
* list only applies to this write call. Also clear the new_size
* field of the inode while we've go it locked.
*/
xfs_ilock(ip, XFS_ILOCK_EXCL);
error0:
xfs_free_gap_list(ip);
ip->i_new_size = 0;
xfs_iunlock(ip, XFS_ILOCK_EXCL);
ip->i_write_offset = 0;
return error;
}
/*
* This is a subroutine for xfs_write() and other writers
* (xfs_fcntl) which clears the setuid and setgid bits when a file is written.
*/
int
xfs_write_clear_setuid(
xfs_inode_t *ip)
{
xfs_mount_t *mp;
xfs_trans_t *tp;
int error;
mp = ip->i_mount;
tp = xfs_trans_alloc(mp, XFS_TRANS_WRITEID);
if (error = xfs_trans_reserve(tp, 0,
XFS_WRITEID_LOG_RES(mp),
0, 0, 0)) {
xfs_trans_cancel(tp, 0);
return error;
}
xfs_ilock(ip, XFS_ILOCK_EXCL);
xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
xfs_trans_ihold(tp, ip);
ip->i_d.di_mode &= ~ISUID;
/*
* Note that we don't have to worry about mandatory
* file locking being disabled here because we only
* clear the ISGID bit if the Group execute bit is
* on, but if it was on then mandatory locking wouldn't
* have been enabled.
*/
if (ip->i_d.di_mode & (IEXEC >> 3)) {
ip->i_d.di_mode &= ~ISGID;
}
xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
xfs_trans_set_sync(tp);
error = xfs_trans_commit(tp, 0, NULL);
xfs_iunlock(ip, XFS_ILOCK_EXCL);
return 0;
}
/*
* xfs_write
*
* This is the XFS VOP_WRITE entry point. It does some minimal error
* checking and then switches out based on the file type.
*/
int
xfs_write(
bhv_desc_t *bdp,
uio_t *uiop,
int ioflag,
cred_t *credp,
flid_t *fl)
{
xfs_inode_t *ip;
xfs_mount_t *mp;
xfs_trans_t *tp;
int type;
off_t offset;
size_t count;
int error, transerror;
int lflag;
off_t n;
int resid;
off_t savedsize;
xfs_fsize_t limit;
int eventsent;
vnode_t *vp;
xfs_lsn_t commit_lsn;
vasid_t vasid;
const int numvnmaps = 10;
vnmap_t vnmaps[numvnmaps];
vnmap_t *rvnmaps;
vnmap_t *map;
int num_rvnmaps;
int rvnmap_flags;
int rvnmap_size = 0;
int i;
xfs_uaccmap_t uaccmap_array[numvnmaps];
xfs_uaccmap_t *uaccmaps;
as_verifyvnmap_t vnmap_args;
as_verifyvnmapres_t vnmap_res;
#if defined(DEBUG) && defined(UIOSZ_DEBUG)
/*
* Randomly set io size
*/
extern ulong_t random(void);
extern int srandom(int);
timespec_t now; /* current time */
static int seed = 0; /* randomizing seed value */
if (!seed) {
nanotime(&now);
seed = (int)now.tv_sec ^ (int)now.tv_nsec;
srandom(seed);
}
ioflag |= IO_UIOSZ;
uiop->uio_writeiolog = (random() & 0x3) + XFS_UIO_MIN_WRITEIO_LOG;
#endif
vp = BHV_TO_VNODE(bdp);
ip = XFS_BHVTOI(bdp);
mp = ip->i_mount;
eventsent = 0;
commit_lsn = -1;
if (XFS_FORCED_SHUTDOWN(ip->i_mount))
return (EIO);
/*
* need to protect against deadlocks that can occur if the
* biomove touches a virtual address in user space that is
* mapped to the file being read. This only works for
* read/write and pread/pwrite. readv/writev lose.
* direct i/o loses too for now.
*
* note that if someone remaps the user buffer to this file
* while the I/O in progess, we lose too. instant deadlock.
*/
rvnmaps = NULL;
num_rvnmaps = 0;
rvnmap_flags = 0;
uaccmaps = NULL;
if (uiop->uio_segflg == UIO_USERSPACE && uiop->uio_iovcnt == 1 &&
!(ioflag & IO_DIRECT) && VN_MAPPED(vp)) {
#pragma mips_frequency_hint NEVER
as_lookup_current(&vasid);
VAS_LOCK(vasid, AS_SHARED);
vnmap_args.as_vp = vp;
vnmap_args.as_vaddr = (uvaddr_t)
ctob(btoct(uiop->uio_iov->iov_base));
vnmap_args.as_len = uiop->uio_iov->iov_len +
((__psunsigned_t) uiop->uio_iov->iov_base -
(__psunsigned_t) vnmap_args.as_vaddr);
vnmap_args.as_offstart = uiop->uio_offset;
vnmap_args.as_offend = uiop->uio_offset + uiop->uio_resid;
vnmap_args.as_vnmap = vnmaps;
vnmap_args.as_nmaps = numvnmaps;
if (error = VAS_VERIFYVNMAP(vasid, &vnmap_args, &vnmap_res)) {
VAS_UNLOCK(vasid);
return error;
}
VAS_UNLOCK(vasid);
if (vnmap_res.as_rescodes) {
rvnmaps = (vnmap_res.as_multimaps == NULL)
? vnmaps
: vnmap_res.as_multimaps;
rvnmap_flags = vnmap_res.as_rescodes;
num_rvnmaps = vnmap_res.as_nmaps;
rvnmap_size = vnmap_res.as_mapsize;
if (num_rvnmaps <= numvnmaps)
uaccmaps = uaccmap_array;
else if ((uaccmaps = kmem_alloc(num_rvnmaps *
sizeof(xfs_uaccmap_t),
KM_SLEEP)) == NULL) {
error = XFS_ERROR(ENOMEM);
goto out;
}
}
}
/*
* check if we're in recursive lock mode (a read inside a biomove
* to a page that is mapped to ip that faulted)
*/
lflag = xfs_is_nested_locking_enabled()
? XFS_IOLOCK_NESTED
: 0;
if (!(ioflag & IO_ISLOCKED))
xfs_rwlockf(bdp, (ioflag & IO_DIRECT) ?
VRWLOCK_WRITE_DIRECT : VRWLOCK_WRITE,
lflag);
/*
* if the write will grow the file beyond the current
* eof, grow the file now by faulting in the page with
* the largest file offset. That will zero-fill all
* pages in the buffer containing eof and will enable
* faults between the current eof and the new eof to
* read in zero'ed pages.
*/
if (rvnmap_flags & AS_VNMAP_AUTOGROW) {
#pragma mips_frequency_hint NEVER
xfs_ilock(ip, XFS_ILOCK_SHARED);
if (vnmap_res.as_maxoffset <= ip->i_d.di_size) {
xfs_iunlock(ip, XFS_ILOCK_SHARED);
} else {
xfs_iunlock(ip, XFS_ILOCK_SHARED);
map = rvnmaps;
for (i = 0; i < num_rvnmaps; i++) {
if (!(map->vnmap_flags & AS_VNMAP_AUTOGROW))
continue;
xfs_enable_nested_locking();
error = fubyte((char *) map->vnmap_ovvaddr +
map->vnmap_ovlen - 1);
xfs_disable_nested_locking();
if (error == -1) {
error = XFS_ERROR(EFAULT);
goto out;
}
}
}
}
type = ip->i_d.di_mode & IFMT;
ASSERT(!(vp->v_vfsp->vfs_flag & VFS_RDONLY));
ASSERT(type == IFDIR ||
ismrlocked(&ip->i_iolock, MR_UPDATE) ||
(ismrlocked(&ip->i_iolock, MR_ACCESS) &&
(ioflag & IO_DIRECT)));
ASSERT(type == IFREG || type == IFDIR ||
type == IFLNK || type == IFSOCK);
start:
if (ioflag & IO_APPEND) {
/*
* In append mode, start at the end of the file.
* Since I've got the iolock exclusive I can look
* at di_size.
*/
uiop->uio_offset = savedsize = ip->i_d.di_size;
}
offset = uiop->uio_offset;
count = uiop->uio_resid;
/* check for locks if some exist and mandatory locking is enabled */
if ((vp->v_flag & (VENF_LOCKING|VFRLOCKS)) ==
(VENF_LOCKING|VFRLOCKS)) {
error = fs_checklock(vp, FWRITE, offset, count,
uiop->uio_fmode, credp, fl,
VRWLOCK_WRITE);
if (error)
goto out;
}
if (offset < 0) {
error = XFS_ERROR(EINVAL);
goto out;
}
if (count <= 0) {
error = 0;
goto out;
}
switch (type) {
case IFREG:
limit = ((uiop->uio_limit < XFS_MAX_FILE_OFFSET) ?
uiop->uio_limit : XFS_MAX_FILE_OFFSET);
n = limit - uiop->uio_offset;
if (n <= 0) {
error = XFS_ERROR(EFBIG);
goto out;
}
if (n < uiop->uio_resid) {
resid = uiop->uio_resid - n;
uiop->uio_resid = n;
} else {
resid = 0;
}
if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_WRITE) &&
!(ioflag & IO_INVIS) && !eventsent) {
vrwlock_t locktype;
locktype = (ioflag & IO_DIRECT) ?
VRWLOCK_WRITE_DIRECT:VRWLOCK_WRITE;
error = xfs_dm_send_data_event(DM_EVENT_WRITE, bdp,
offset, count,
UIO_DELAY_FLAG(uiop), &locktype);
if (error)
goto out;
eventsent = 1;
}
/*
* The iolock was dropped and reaquired in
* xfs_dm_send_data_event so we have to recheck the size
* when appending. We will only "goto start;" once,
* since having sent the event prevents another call
* to xfs_dm_send_data_event, which is what
* allows the size to change in the first place.
*/
if ((ioflag & IO_APPEND) && savedsize != ip->i_d.di_size)
goto start;
/*
* implement osync == dsync option
*/
if (ioflag & IO_SYNC && mp->m_flags & XFS_MOUNT_OSYNCISDSYNC) {
ioflag &= ~IO_SYNC;
ioflag |= IO_DSYNC;
}
/*
* If we're writing the file then make sure to clear the
* setuid and setgid bits if the process is not being run
* by root. This keeps people from modifying setuid and
* setgid binaries. Don't allow this to happen if this
* file is a swap file (I know, weird).
*/
if (((ip->i_d.di_mode & ISUID) ||
((ip->i_d.di_mode & (ISGID | (IEXEC >> 3))) ==
(ISGID | (IEXEC >> 3)))) &&
!(vp->v_flag & VISSWAP) &&
!cap_able_cred(credp, CAP_FSETID)) {
error = xfs_write_clear_setuid(ip);
if (error) {
goto out;
}
}
/*
* Respect preferred write size if indicated in uio structure.
* But if the write size has already been set, go with the
* smallest value. Silently ignore requests that aren't
* within valid I/O size limits.
*/
if ((ioflag & IO_UIOSZ) &&
uiop->uio_writeiolog != ip->i_writeio_log &&
uiop->uio_writeiolog >= mp->m_sb.sb_blocklog &&
uiop->uio_writeiolog >= XFS_UIO_MIN_WRITEIO_LOG &&
uiop->uio_writeiolog <= XFS_UIO_MAX_WRITEIO_LOG) {
xfs_ilock(ip, XFS_ILOCK_EXCL);
#if !(defined(DEBUG) && defined(UIOSZ_DEBUG))
if (!(ip->i_flags & XFS_IUIOSZ) ||
uiop->uio_writeiolog < ip->i_writeio_log) {
#endif /* ! (DEBUG && UIOSZ_DEBUG) */
ip->i_writeio_log = uiop->uio_writeiolog;
ip->i_writeio_blocks = 1 <<
(int) (ip->i_writeio_log -
mp->m_sb.sb_blocklog);
/*
* set inode max io size to largest value
* that the inode could ever have had
*/
if (!(ip->i_flags & XFS_IUIOSZ)) {
ip->i_max_io_log = MAX(ip->i_max_io_log,
MAX(mp->m_writeio_log,
ip->i_writeio_log));
ip->i_flags |= XFS_IUIOSZ;
}
#if defined(DEBUG) && defined(UIOSZ_DEBUG)
atomicAddInt(&uiodbg_switch, 1);
atomicAddInt(
&(uiodbg_writeiolog[ip->i_writeio_log -
XFS_UIO_MIN_WRITEIO_LOG]),
1);
#endif
#if !(defined(DEBUG) && defined(UIOSZ_DEBUG))
}
#endif /* ! (DEBUG && UIOSZ_DEBUG) */
xfs_iunlock(ip, XFS_ILOCK_EXCL);
}
retry:
if (ioflag & IO_DIRECT) {
error = xfs_diordwr(bdp, uiop, ioflag, credp, B_WRITE);
} else {
error = xfs_write_file(bdp, uiop, ioflag, credp,
&commit_lsn,
rvnmaps, num_rvnmaps,
rvnmap_flags, uaccmaps);
}
if (error == ENOSPC &&
DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_NOSPACE) &&
!(ioflag & IO_INVIS)) {
vrwlock_t locktype;
locktype = (ioflag & IO_DIRECT) ?
VRWLOCK_WRITE_DIRECT:VRWLOCK_WRITE;
VOP_RWUNLOCK(vp, locktype);
error = dm_send_namesp_event(DM_EVENT_NOSPACE, bdp,
DM_RIGHT_NULL, bdp, DM_RIGHT_NULL, NULL, NULL,
0, 0, 0); /* Delay flag intentionally unused */
VOP_RWLOCK(vp, locktype);
if (error)
goto out;
offset = uiop->uio_offset;
goto retry;
} else if (error == ENOSPC) {
if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) {
xfs_error(mp, 2);
} else {
xfs_error(mp, 1);
}
}
/*
* Add back whatever we refused to do because of
* uio_limit.
*/
uiop->uio_resid += resid;
/*
* We've done at least a partial write, so don't
* return an error on this call. Also update the
* timestamps since we changed the file.
*/
if (count != uiop->uio_resid) {
error = 0;
/* don't update timestamps if doing invisible I/O */
if (!(ioflag & IO_INVIS))
xfs_ichgtime(ip,
XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
}
/*
* If the write was synchronous then we need to make
* sure that the inode modification time is permanent.
* We'll have update the timestamp above, so here
* we use a synchronous transaction to log the inode.
* It's not fast, but it's necessary.
*
* If this a dsync write and the size got changed
* non-transactionally, then we need to ensure that
* the size change gets logged in a synchronous
* transaction. If an allocation transaction occurred
* without extending the size, then we have to force
* the log up the proper point to ensure that the
* allocation is permanent. We can't count on
* the fact that buffered writes lock out direct I/O
* writes because the direct I/O write could have extended
* the size non-transactionally and then finished just before
* we started. xfs_write_file will think that the file
* didn't grow but the update isn't safe unless the
* size change is logged.
*
* If the vnode is a swap vnode, then don't do anything
* which could require allocating memory.
*/
if ((ioflag & IO_SYNC ||
(ioflag & IO_DSYNC && ip->i_update_size)) &&
!(vp->v_flag & VISSWAP)) {
tp = xfs_trans_alloc(mp, XFS_TRANS_WRITE_SYNC);
if (transerror = xfs_trans_reserve(tp, 0,
XFS_SWRITE_LOG_RES(mp),
0, 0, 0)) {
xfs_trans_cancel(tp, 0);
error = transerror;
break;
}
xfs_ilock(ip, XFS_ILOCK_EXCL);
xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
xfs_trans_ihold(tp, ip);
xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
xfs_trans_set_sync(tp);
transerror = xfs_trans_commit(tp, 0, &commit_lsn);
if ( transerror )
error = transerror;
xfs_iunlock(ip, XFS_ILOCK_EXCL);
} else if ((ioflag & IO_DSYNC) && !(vp->v_flag & VISSWAP)) {
/*
* force the log if we've committed a transaction
* against the inode or if someone else has and
* the commit record hasn't gone to disk (e.g.
* the inode is pinned). This guarantees that
* all changes affecting the inode are permanent
* when we return.
*/
if (commit_lsn != -1)
xfs_log_force(mp, (xfs_lsn_t)commit_lsn,
XFS_LOG_FORCE | XFS_LOG_SYNC );
else if (xfs_ipincount(ip) > 0)
xfs_log_force(mp, (xfs_lsn_t)0,
XFS_LOG_FORCE | XFS_LOG_SYNC );
}
if (ioflag & (IO_NFS|IO_NFS3)) {
xfs_refcache_insert(ip);
}
break;
case IFDIR:
error = XFS_ERROR(EISDIR);
break;
case IFLNK:
error = XFS_ERROR(EINVAL);
break;
case IFSOCK:
error = XFS_ERROR(ENODEV);
break;
default:
ASSERT(0);
error = XFS_ERROR(EINVAL);
break;
}
out:
if (rvnmap_size > 0)
kmem_free(rvnmaps, rvnmap_size);
if (num_rvnmaps > numvnmaps)
kmem_free(uaccmaps, num_rvnmaps * sizeof(xfs_uaccmap_t));
if (!(ioflag & IO_ISLOCKED))
xfs_rwunlockf(bdp, (ioflag & IO_DIRECT) ?
VRWLOCK_WRITE_DIRECT : VRWLOCK_WRITE,
lflag);
return error;
}
/*
* This is the XFS entry point for VOP_BMAP().
* It simply switches based on the given flags
* to either xfs_iomap_read() or xfs_iomap_write().
* This cannot be used to grow a file or to read
* beyond the end of the file.
*
* The caller is required to be holding the inode's
* iolock in at least shared mode for a read mapping
* and exclusively for a write mapping.
*/
/*ARGSUSED*/
int
xfs_bmap(
bhv_desc_t *bdp,
off_t offset,
ssize_t count,
int flags,
cred_t *credp,
struct bmapval *bmapp,
int *nbmaps)
{
xfs_inode_t *ip;
int error;
int unlocked;
int lockmode;
ip = XFS_BHVTOI(bdp);
ASSERT((ip->i_d.di_mode & IFMT) == IFREG);
ASSERT((flags == B_READ) || (flags == B_WRITE));
if (XFS_FORCED_SHUTDOWN(ip->i_mount))
return (EIO);
if (flags == B_READ) {
ASSERT(ismrlocked(&ip->i_iolock, MR_ACCESS | MR_UPDATE) != 0);
unlocked = 0;
lockmode = xfs_ilock_map_shared(ip);
error = xfs_iomap_read(ip, offset, count, bmapp,
nbmaps, NULL, &unlocked, lockmode);
if (!unlocked)
xfs_iunlock_map_shared(ip, lockmode);
} else {
ASSERT(ismrlocked(&ip->i_iolock, MR_UPDATE) != 0);
xfs_ilock(ip, XFS_ILOCK_EXCL);
ASSERT(ip->i_d.di_size >= (offset + count));
/*
* Make sure that the dquots are there. This doesn't hold
* the ilock across a disk read.
*/
if (XFS_IS_QUOTA_ON(ip->i_mount)) {
if (XFS_NOT_DQATTACHED(ip->i_mount, ip)) {
if (error = xfs_qm_dqattach(ip,
XFS_QMOPT_ILOCKED)){
xfs_iunlock(ip, XFS_ILOCK_EXCL);
goto error0;
}
}
}
error = xfs_iomap_write(ip, offset, count, bmapp,
nbmaps, 0, NULL);
xfs_iunlock(ip, XFS_ILOCK_EXCL);
if (!error)
XFS_INODE_CLEAR_READ_AHEAD(ip);
}
error0:
return error;
}
/*
* Set up rbp so that it points to the memory attached to bp
* from rbp_offset from the start of bp for rbp_len bytes.
*/
STATIC void
xfs_overlap_bp(
buf_t *bp,
buf_t *rbp,
uint rbp_offset,
uint rbp_len)
{
int pgbboff;
int bytes_off;
pfd_t *pfdp;
if (BP_ISMAPPED(bp)) {
/*
* The real buffer is already mapped, so just use
* its virtual memory for ourselves.
*/
rbp->b_un.b_addr = bp->b_un.b_addr + rbp_offset;
rbp->b_bcount = rbp_len;
rbp->b_bufsize = rbp_len;
} else {
/*
* The real buffer is not yet mapped to virtual memory.
* Just get the subordinate buffer's page pointers
* set up and make it a PAGEIO buffer like the real one.
*
* First find the first page of rbp. We do this by
* walking the list of pages in bp until we find the
* one containing the start of rbp. Note that neither
* bp nor rbp are required to start on page boundaries.
*/
bytes_off = rbp_offset + BBTOOFF(dpoff(bp->b_offset));
pfdp = NULL;
pfdp = getnextpg(bp, pfdp);
ASSERT(pfdp != NULL);
while (bytes_off >= NBPP) {
pfdp = getnextpg(bp, pfdp);
ASSERT(pfdp != NULL);
bytes_off -= NBPP;
}
rbp->b_pages = pfdp;
rbp->b_bcount = rbp_len;
rbp->b_offset = bp->b_offset + BTOBB(rbp_offset);
pgbboff = dpoff(rbp->b_offset);
rbp->b_bufsize = ctob(dtop(pgbboff + BTOBB(rbp_len)));
rbp->b_flags |= B_PAGEIO;
if (pgbboff != 0) {
bp_mapin(rbp);
}
}
rbp->b_blkno = bp->b_blkno + BTOBB(rbp_offset);
rbp->b_remain = 0;
rbp->b_vp = bp->b_vp;
rbp->b_edev = bp->b_edev;
rbp->b_flags |= (bp->b_flags & (B_UNCACHED | B_ASYNC));
}
/*
* Zero the given bp from data_offset from the start of it for data_len
* bytes.
*/
STATIC void
xfs_zero_bp(
buf_t *bp,
int data_offset,
int data_len)
{
pfd_t *pfdp;
caddr_t page_addr;
int len;
if (BP_ISMAPPED(bp)) {
bzero(bp->b_un.b_addr + data_offset, data_len);
return;
}
data_offset += BBTOOFF(dpoff(bp->b_offset));
pfdp = NULL;
pfdp = getnextpg(bp, pfdp);
ASSERT(pfdp != NULL);
while (data_offset >= NBPP) {
pfdp = getnextpg(bp, pfdp);
ASSERT(pfdp != NULL);
data_offset -= NBPP;
}
ASSERT(data_offset >= 0);
while (data_len > 0) {
page_addr = page_mapin(pfdp, (bp->b_flags & B_UNCACHED ?
VM_UNCACHED : 0), 0);
len = MIN(data_len, NBPP - data_offset);
bzero(page_addr + data_offset, len);
data_len -= len;
data_offset = 0;
page_mapout(page_addr);
pfdp = getnextpg(bp, pfdp);
}
}
/*
* Verify that the gap list is properly sorted and that no entries
* overlap.
*/
#ifdef DEBUG
STATIC void
xfs_check_gap_list(
xfs_inode_t *ip)
{
xfs_gap_t *last_gap;
xfs_gap_t *curr_gap;
int loops;
last_gap = NULL;
curr_gap = ip->i_gap_list;
loops = 0;
while (curr_gap != NULL) {
ASSERT(curr_gap->xg_count_fsb > 0);
if (last_gap != NULL) {
ASSERT((last_gap->xg_offset_fsb +
last_gap->xg_count_fsb) <
curr_gap->xg_offset_fsb);
}
last_gap = curr_gap;
curr_gap = curr_gap->xg_next;
ASSERT(loops++ < 1000);
}
}
#endif
/*
* For the given inode, offset, and count of bytes, build a list
* of xfs_gap_t structures in the inode's gap list describing the
* holes in the file in the range described by the offset and count.
*
* The list must be empty when we start, and the inode lock must
* be held exclusively.
*/
STATIC int /* error */
xfs_build_gap_list(
xfs_inode_t *ip,
off_t offset,
size_t count)
{
xfs_fileoff_t offset_fsb;
xfs_fileoff_t last_fsb;
xfs_filblks_t count_fsb;
xfs_fsblock_t firstblock;
xfs_gap_t *new_gap;
xfs_gap_t *last_gap;
xfs_mount_t *mp;
int i;
int error;
int nimaps;
#define XFS_BGL_NIMAPS 8
xfs_bmbt_irec_t imaps[XFS_BGL_NIMAPS];
xfs_bmbt_irec_t *imapp;
ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
ASSERT(ip->i_gap_list == NULL);
mp = ip->i_mount;
offset_fsb = XFS_B_TO_FSBT(mp, offset);
last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
count_fsb = (xfs_filblks_t)(last_fsb - offset_fsb);
ASSERT(count_fsb > 0);
last_gap = NULL;
while (count_fsb > 0) {
nimaps = XFS_BGL_NIMAPS;
firstblock = NULLFSBLOCK;
error = xfs_bmapi(NULL, ip, offset_fsb, count_fsb,
0, &firstblock, 0, imaps, &nimaps, NULL);
if (error) {
return error;
}
ASSERT(nimaps != 0);
/*
* Look for the holes in the mappings returned by bmapi.
* Decrement count_fsb and increment offset_fsb as we go.
*/
for (i = 0; i < nimaps; i++) {
imapp = &imaps[i];
count_fsb -= imapp->br_blockcount;
ASSERT(((long)count_fsb) >= 0);
ASSERT(offset_fsb == imapp->br_startoff);
offset_fsb += imapp->br_blockcount;
ASSERT(offset_fsb <= last_fsb);
ASSERT((offset_fsb < last_fsb) || (count_fsb == 0));
/*
* Skip anything that is not a hole or
* unwritten.
*/
if (imapp->br_startblock != HOLESTARTBLOCK ||
imapp->br_state == XFS_EXT_UNWRITTEN) {
continue;
}
/*
* We found a hole. Now add an entry to the inode's
* gap list corresponding to it. The list is
* a singly linked, NULL terminated list. We add
* each entry to the end of the list so that it is
* sorted by file offset.
*/
new_gap = kmem_zone_alloc(xfs_gap_zone, KM_SLEEP);
new_gap->xg_offset_fsb = imapp->br_startoff;
new_gap->xg_count_fsb = imapp->br_blockcount;
new_gap->xg_next = NULL;
if (last_gap == NULL) {
ip->i_gap_list = new_gap;
} else {
last_gap->xg_next = new_gap;
}
last_gap = new_gap;
}
}
xfs_check_gap_list(ip);
return 0;
}
/*
* Remove or trim any entries in the inode's gap list which overlap
* the given range. I'm going to assume for now that we never give
* a range which is actually in the middle of an entry (i.e. we'd need
* to split it in two). This is a valid assumption for now given the
* use of this in xfs_write_file() where we start at the front and
* move sequentially forward.
*/
STATIC void
xfs_delete_gap_list(
xfs_inode_t *ip,
xfs_fileoff_t offset_fsb,
xfs_extlen_t count_fsb)
{
xfs_gap_t *curr_gap;
xfs_gap_t *last_gap;
xfs_gap_t *next_gap;
xfs_fileoff_t gap_offset_fsb;
xfs_extlen_t gap_count_fsb;
xfs_fileoff_t gap_end_fsb;
xfs_fileoff_t end_fsb;
last_gap = NULL;
curr_gap = ip->i_gap_list;
while (curr_gap != NULL) {
gap_offset_fsb = curr_gap->xg_offset_fsb;
gap_count_fsb = curr_gap->xg_count_fsb;
/*
* The entries are sorted by offset, so if we see
* one beyond our range we're done.
*/
end_fsb = offset_fsb + count_fsb;
if (gap_offset_fsb >= end_fsb) {
return;
}
gap_end_fsb = gap_offset_fsb + gap_count_fsb;
if (gap_end_fsb <= offset_fsb) {
/*
* This shouldn't be able to happen for now.
*/
ASSERT(0);
last_gap = curr_gap;
curr_gap = curr_gap->xg_next;
continue;
}
/*
* We've go an overlap. If the gap is entirely contained
* in the region then remove it. If not, then shrink it
* by the amount overlapped.
*/
if (gap_end_fsb > end_fsb) {
/*
* The region does not extend to the end of the gap.
* Shorten the gap by the amount in the region,
* and then we're done since we've reached the
* end of the region.
*/
ASSERT(gap_offset_fsb >= offset_fsb);
curr_gap->xg_offset_fsb = end_fsb;
curr_gap->xg_count_fsb = gap_end_fsb - end_fsb;
return;
}
next_gap = curr_gap->xg_next;
if (last_gap == NULL) {
ip->i_gap_list = next_gap;
} else {
ASSERT(0);
ASSERT(last_gap->xg_next == curr_gap);
last_gap->xg_next = next_gap;
}
kmem_zone_free(xfs_gap_zone, curr_gap);
curr_gap = next_gap;
}
}
/*
* Free up all of the entries in the inode's gap list. This requires
* the inode lock to be held exclusively.
*/
STATIC void
xfs_free_gap_list(
xfs_inode_t *ip)
{
xfs_gap_t *curr_gap;
xfs_gap_t *next_gap;
ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE) != 0);
xfs_check_gap_list(ip);
curr_gap = ip->i_gap_list;
while (curr_gap != NULL) {
next_gap = curr_gap->xg_next;
kmem_zone_free(xfs_gap_zone, curr_gap);
curr_gap = next_gap;
}
ip->i_gap_list = NULL;
}
/*
* Zero the parts of the buffer which overlap gaps in the inode's gap list.
* Deal with everything in BBs since the buffer is not guaranteed to be block
* aligned.
*/
STATIC void
xfs_cmp_gap_list_and_zero(
xfs_inode_t *ip,
buf_t *bp)
{
off_t bp_offset_bb;
int bp_len_bb;
off_t gap_offset_bb;
int gap_len_bb;
int zero_offset_bb;
int zero_len_bb;
xfs_gap_t *curr_gap;
xfs_mount_t *mp;
ASSERT(ismrlocked(&(ip->i_lock), MR_UPDATE | MR_ACCESS) != 0);
xfs_check_gap_list(ip);
bp_offset_bb = bp->b_offset;
bp_len_bb = BTOBB(bp->b_bcount);
mp = ip->i_mount;
curr_gap = ip->i_gap_list;
while (curr_gap != NULL) {
gap_offset_bb = XFS_FSB_TO_BB(mp, curr_gap->xg_offset_fsb);
gap_len_bb = XFS_FSB_TO_BB(mp, curr_gap->xg_count_fsb);
/*
* Check to see if this gap is before the buffer starts.
*/
if ((gap_offset_bb + gap_len_bb) <= bp_offset_bb) {
curr_gap = curr_gap->xg_next;
continue;
}
/*
* Check to see if this gap is after th buffer ends.
* If it is then we're done since the list is sorted
* by gap offset.
*/
if (gap_offset_bb >= (bp_offset_bb + bp_len_bb)) {
break;
}
/*
* We found a gap which overlaps the buffer. Zero
* the portion of the buffer overlapping the gap.
*/
if (gap_offset_bb < bp_offset_bb) {
/*
* The gap starts before the buffer, so we start
* zeroing from the start of the buffer.
*/
zero_offset_bb = 0;
/*
* To calculate the amount of overlap. First
* subtract the portion of the gap which is before
* the buffer. If the length is still longer than
* the buffer, then just zero the entire buffer.
*/
zero_len_bb = gap_len_bb -
(bp_offset_bb - gap_offset_bb);
if (zero_len_bb > bp_len_bb) {
zero_len_bb = bp_len_bb;
}
ASSERT(zero_len_bb > 0);
} else {
/*
* The gap starts at the beginning or in the middle
* of the buffer. The offset into the buffer is
* the difference between the two offsets.
*/
zero_offset_bb = gap_offset_bb - bp_offset_bb;
/*
* Figure out the length of the overlap. If the
* gap extends beyond the end of the buffer, then
* just zero to the end of the buffer. Otherwise
* just zero the length of the gap.
*/
if ((gap_offset_bb + gap_len_bb) >
(bp_offset_bb + bp_len_bb)) {
zero_len_bb = bp_len_bb - zero_offset_bb;
} else {
zero_len_bb = gap_len_bb;
}
}
/*
* Now that we've calculated the range of the buffer to
* zero, do it.
*/
xfs_zero_bp(bp, BBTOB(zero_offset_bb), BBTOB(zero_len_bb));
curr_gap = curr_gap->xg_next;
}
}
/*
* "Read" in a buffer whose b_blkno is -1 or uninitialized.
* If b_blkno is -1, this means that at the time the buffer was created
* there was no underlying backing store for the range of the file covered
* by the bp. An uninitialized buffer (with B_UNINITIAL set) indicates
* that there is allocated storage, but all or some portions of the
* underlying blocks have never been written.
*
* To figure out the current state of things, we lock the inode
* and call xfs_bmapi() to look at the current extents format.
* If we're over a hole, delayed allocation or uninitialized space we
* simply zero the corresponding portions of the buffer. For parts
* over real disk space we need to read in the stuff from disk.
*
* We know that we can just use xfs_ilock(SHARED) rather than
* xfs_ilock_map_shared() here, because the extents had to be
* read in in order to create the buffer we're trying to write out.
*/
STATIC int
xfs_strat_read(
bhv_desc_t *bdp,
buf_t *bp)
{
xfs_fileoff_t offset_fsb;
xfs_fileoff_t map_start_fsb;
xfs_fileoff_t imap_offset;
xfs_fsblock_t last_bp_bb;
xfs_fsblock_t last_map_bb;
xfs_fsblock_t firstblock;
xfs_filblks_t count_fsb;
xfs_extlen_t imap_blocks;
xfs_fsize_t isize;
off_t offset;
off_t end_offset;
off_t init_limit;
int x;
caddr_t datap;
buf_t *rbp;
xfs_mount_t *mp;
xfs_inode_t *ip;
int count;
int block_off;
int data_bytes;
int data_offset;
int nimaps;
int error;
#define XFS_STRAT_READ_IMAPS XFS_BMAP_MAX_NMAP
xfs_bmbt_irec_t imap[XFS_STRAT_READ_IMAPS];
ASSERT((bp->b_blkno == -1) || (bp->b_flags & B_UNINITIAL));
ip = XFS_BHVTOI(bdp);
mp = ip->i_mount;
offset_fsb = XFS_BB_TO_FSBT(mp, bp->b_offset);
/*
* Only read up to the EOF or the current write offset.
* The idea here is to avoid initializing pages which are
* going to be immediately overwritten in xfs_write_file().
* The most important case is the sequential write case, where
* the new pages at the end of the file are sent here by
* chunk_patch(). We don't want to zero them since they
* are about to be overwritten.
*
* The ip->i_write_off tells us the offset of any write in
* progress. If it is 0 then we assume that no write is
* in progress. If the write offset is within the file size,
* the the file size is the upper limit. If the write offset
* is beyond the file size, then we only want to initialize the
* buffer up to the write offset. Beyond that will either be
* overwritten or be beyond the new EOF.
*/
isize = ip->i_d.di_size;
offset = BBTOOFF(bp->b_offset);
end_offset = offset + bp->b_bcount;
if (ip->i_write_offset == 0) {
init_limit = isize;
} else if (ip->i_write_offset <= isize) {
init_limit = isize;
} else {
init_limit = ip->i_write_offset;
}
if (end_offset <= init_limit) {
count = bp->b_bcount;
} else {
count = init_limit - offset;
}
if (count <= 0) {
iodone(bp);
return 0;
}
/*
* Since the buffer may not be file system block aligned, we
* can't do a simple shift to find the number of blocks underlying
* it. Instead we subtract the last block it sits on from the first.
*/
count_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count))) -
XFS_B_TO_FSBT(mp, offset);
map_start_fsb = offset_fsb;
xfs_ilock(ip, XFS_ILOCK_SHARED);
while (count_fsb != 0) {
nimaps = XFS_STRAT_READ_IMAPS;
firstblock = NULLFSBLOCK;
error = xfs_bmapi(NULL, ip, map_start_fsb, count_fsb, 0,
&firstblock, 0, imap, &nimaps, NULL);
if (error) {
xfs_iunlock(ip, XFS_ILOCK_SHARED);
xfs_bioerror_relse(bp);
return error;
}
ASSERT(nimaps >= 1);
for (x = 0; x < nimaps; x++) {
imap_offset = imap[x].br_startoff;
ASSERT(imap_offset == map_start_fsb);
imap_blocks = imap[x].br_blockcount;
ASSERT(imap_blocks <= count_fsb);
/*
* Calculate the offset of this mapping in the
* buffer and the number of bytes of this mapping
* that are in the buffer. If the block size is
* greater than the page size, then the buffer may
* not line up on file system block boundaries
* (e.g. pages being read in from chunk_patch()).
* In that case we need to account for the space
* in the file system blocks underlying the buffer
* that is not actually a part of the buffer. This
* space is the space in the first block before the
* start of the buffer and the space in the last
* block after the end of the buffer.
*/
data_offset = XFS_FSB_TO_B(mp,
imap_offset - offset_fsb);
data_bytes = XFS_FSB_TO_B(mp, imap_blocks);
block_off = 0;
if (mp->m_sb.sb_blocksize > NBPP) {
/*
* If the buffer is actually fsb
* aligned then this will simply
* subtract 0 and do no harm. If the
* current mapping is for the start of
* the buffer, then data offset will be
* zero so we don't need to subtract out
* any space at the beginning.
*/
if (data_offset > 0) {
data_offset -= BBTOB(
XFS_BB_FSB_OFFSET(mp,
bp->b_offset));
}
if (map_start_fsb == offset_fsb) {
ASSERT(data_offset == 0);
/*
* This is on the first block
* mapped, so it must be the start
* of the buffer. Subtract out from
* the number of bytes the bytes
* between the start of the block
* and the start of the buffer.
*/
data_bytes -=
BBTOB(XFS_BB_FSB_OFFSET(
mp, bp->b_offset));
/*
* Set block_off to the number of
* BBs that the buffer is offset
* from the start of this mapping.
*/
block_off = XFS_BB_FSB_OFFSET(mp,
bp->b_offset);
ASSERT(block_off >= 0);
}
if (imap_blocks == count_fsb) {
/*
* This mapping includes the last
* block to be mapped. Subtract out
* from the number of bytes the bytes
* between the end of the buffer and
* the end of the block. It may
* be the case that the buffer
* extends beyond the mapping (if
* it is beyond the end of the file),
* in which case no adjustment
* is necessary.
*/
last_bp_bb = bp->b_offset +
BTOBB(bp->b_bcount);
last_map_bb =
XFS_FSB_TO_BB(mp,
(imap_offset +
imap_blocks));
if (last_map_bb > last_bp_bb) {
data_bytes -=
BBTOB(last_map_bb -
last_bp_bb);
}
}
}
ASSERT(data_bytes > 0);
ASSERT(data_offset >= 0);
if ((imap[x].br_startblock == DELAYSTARTBLOCK) ||
(imap[x].br_startblock == HOLESTARTBLOCK) ||
(imap[x].br_state == XFS_EXT_UNWRITTEN)) {
/*
* This is either a hole,a delayed alloc
* extent or uninitialized allocated space.
* Either way, just fill it with zeroes.
*/
datap = bp_mapin(bp);
datap += data_offset;
bzero(datap, data_bytes);
if (!dpoff(bp->b_offset)) {
bp_mapout(bp);
}
} else {
/*
* The extent really exists on disk, so
* read it in.
*/
rbp = getrbuf(KM_SLEEP);
xfs_overlap_bp(bp, rbp, data_offset,
data_bytes);
rbp->b_blkno = XFS_FSB_TO_DB(ip,
imap[x].br_startblock) +
block_off;
rbp->b_offset = XFS_FSB_TO_BB(mp,
imap_offset) +
block_off;
rbp->b_flags |= B_READ;
rbp->b_flags &= ~B_ASYNC;
rbp->b_target = bp->b_target;
xfs_check_rbp(ip, bp, rbp, 1);
(void) xfsbdstrat(mp, rbp);
error = iowait(rbp);
if (error) {
bp->b_flags |= B_ERROR;
bp->b_error = error;
ASSERT(bp->b_error != EINVAL);
}
/*
* Check to see if the block extent (or parts
* of it) have not yet been initialized and
* should therefore be zeroed.
*/
xfs_cmp_gap_list_and_zero(ip, rbp);
if (BP_ISMAPPED(rbp)) {
bp_mapout(rbp);
}
freerbuf(rbp);
}
count_fsb -= imap_blocks;
map_start_fsb += imap_blocks;
}
}
xfs_iunlock(ip, XFS_ILOCK_SHARED);
iodone(bp);
return error;
}
#if defined(XFS_STRAT_TRACE)
void
xfs_strat_write_bp_trace(
int tag,
xfs_inode_t *ip,
buf_t *bp)
{
if (ip->i_strat_trace == NULL) {
return;
}
ktrace_enter(ip->i_strat_trace,
(void*)((__psunsigned_t)tag),
(void*)ip,
(void*)((__psunsigned_t)((ip->i_d.di_size >> 32) &
0xffffffff)),
(void*)(ip->i_d.di_size & 0xffffffff),
(void*)bp,
(void*)((__psunsigned_t)((bp->b_offset >> 32) &
0xffffffff)),
(void*)(bp->b_offset & 0xffffffff),
(void*)((__psunsigned_t)(bp->b_bcount)),
(void*)((__psunsigned_t)(bp->b_bufsize)),
(void*)(bp->b_blkno),
(void*)(__psunsigned_t)((bp->b_flags >> 32) & 0xffffffff),
(void*)(bp->b_flags & 0xffffffff),
(void*)(bp->b_pages),
(void*)(bp->b_pages->pf_pageno),
(void*)0,
(void*)0);
ktrace_enter(xfs_strat_trace_buf,
(void*)((__psunsigned_t)tag),
(void*)ip,
(void*)((__psunsigned_t)((ip->i_d.di_size >> 32) &
0xffffffff)),
(void*)(ip->i_d.di_size & 0xffffffff),
(void*)bp,
(void*)((__psunsigned_t)((bp->b_offset >> 32) &
0xffffffff)),
(void*)(bp->b_offset & 0xffffffff),
(void*)((__psunsigned_t)(bp->b_bcount)),
(void*)((__psunsigned_t)(bp->b_bufsize)),
(void*)(bp->b_blkno),
(void*)(__psunsigned_t)((bp->b_flags >> 32) & 0xffffffff),
(void*)(bp->b_flags & 0xffffffff),
(void*)(bp->b_pages),
(void*)(bp->b_pages->pf_pageno),
(void*)0,
(void*)0);
}
void
xfs_strat_write_subbp_trace(
int tag,
xfs_inode_t *ip,
buf_t *bp,
buf_t *rbp,
off_t last_off,
int last_bcount,
daddr_t last_blkno)
{
if (ip->i_strat_trace == NULL) {
return;
}
ktrace_enter(ip->i_strat_trace,
(void*)((unsigned long)tag),
(void*)ip,
(void*)((unsigned long)((ip->i_d.di_size >> 32) &
0xffffffff)),
(void*)(ip->i_d.di_size & 0xffffffff),
(void*)bp,
(void*)rbp,
(void*)((unsigned long)((rbp->b_offset >> 32) &
0xffffffff)),
(void*)(rbp->b_offset & 0xffffffff),
(void*)((unsigned long)(rbp->b_bcount)),
(void*)(rbp->b_blkno),
(void*)((__psunsigned_t)(rbp->b_flags)), /* lower 32 flags only */
(void*)(rbp->b_un.b_addr),
(void*)(bp->b_pages),
(void*)(last_off),
(void*)((unsigned long)(last_bcount)),
(void*)(last_blkno));
ktrace_enter(xfs_strat_trace_buf,
(void*)((unsigned long)tag),
(void*)ip,
(void*)((unsigned long)((ip->i_d.di_size >> 32) &
0xffffffff)),
(void*)(ip->i_d.di_size & 0xffffffff),
(void*)bp,
(void*)rbp,
(void*)((unsigned long)((rbp->b_offset >> 32) &
0xffffffff)),
(void*)(rbp->b_offset & 0xffffffff),
(void*)((unsigned long)(rbp->b_bcount)),
(void*)(rbp->b_blkno),
(void*)((__psunsigned_t)(rbp->b_flags)), /* lower 32 flags only */
(void*)(rbp->b_un.b_addr),
(void*)(bp->b_pages),
(void*)(last_off),
(void*)((unsigned long)(last_bcount)),
(void*)(last_blkno));
}
#endif /* XFS_STRAT_TRACE */
#ifdef DEBUG
/*
* xfs_strat_write_check
*
* Make sure that there are blocks or delayed allocation blocks
* underlying the entire area given. The imap parameter is simply
* given as a scratch area in order to reduce stack space. No
* values are returned within it.
*/
STATIC void
xfs_strat_write_check(
xfs_inode_t *ip,
xfs_fileoff_t offset_fsb,
xfs_filblks_t buf_fsb,
xfs_bmbt_irec_t *imap,
int imap_count)
{
xfs_filblks_t count_fsb;
xfs_fsblock_t firstblock;
int nimaps;
int n;
int error;
xfs_ilock(ip, XFS_ILOCK_SHARED);
count_fsb = 0;
while (count_fsb < buf_fsb) {
nimaps = imap_count;
firstblock = NULLFSBLOCK;
error = xfs_bmapi(NULL, ip, (offset_fsb + count_fsb),
(buf_fsb - count_fsb), 0, &firstblock, 0,
imap, &nimaps, NULL);
if (error) {
xfs_iunlock(ip, XFS_ILOCK_SHARED);
return;
}
ASSERT(nimaps > 0);
n = 0;
while (n < nimaps) {
ASSERT(imap[n].br_startblock != HOLESTARTBLOCK);
count_fsb += imap[n].br_blockcount;
ASSERT(count_fsb <= buf_fsb);
n++;
}
}
xfs_iunlock(ip, XFS_ILOCK_SHARED);
return;
}
#endif /* DEBUG */
/*
* xfs_strat_write_iodone -
* I/O completion for the first write of unwritten buffered data.
* Since this occurs in an interrupt thread, massage some bp info
* and queue to the xfs_strat daemon.
*/
STATIC void
xfs_strat_write_iodone(
buf_t *bp)
{
bhv_desc_t *bdp;
/* REFERENCED */
xfs_inode_t *ip;
int s;
ASSERT(bp->b_flags & B_UNINITIAL);
ASSERT(bp->b_vp);
bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(bp->b_vp), &xfs_vnodeops);
ASSERT(bdp);
ASSERT(xfsc_count > 0);
ip = XFS_BHVTOI(bdp);
/*
* Delay I/O done work until the transaction is completed.
*/
bp->b_iodone = NULL;
ASSERT(bp->b_flags & B_BUSY);
ASSERT(valusema(&bp->b_lock) <= 0);
ASSERT(!(bp->b_flags & B_DONE));
/*
* Queue to the xfsc_list.
*/
s = mp_mutex_spinlock(&xfsc_lock);
/*
* Queue the buffer at the end of the list.
* Bump the inode count of the number of queued buffers.
*/
if (xfsc_list == NULL) {
bp->av_forw = bp;
bp->av_back = bp;
xfsc_list = bp;
} else {
bp->av_back = xfsc_list->av_back;
xfsc_list->av_back->av_forw = bp;
xfsc_list->av_back = bp;
bp->av_forw = xfsc_list;
}
buftrace("STRAT_WRITE_IODONE", bp);
xfsc_bufcount++;
xfs_strat_write_bp_trace(XFS_STRAT_ENTER, ip, bp);
(void)sv_signal(&xfsc_wait);
mp_mutex_spinunlock(&xfsc_lock, s);
return;
}
STATIC void
xfs_strat_comp(void)
{
bhv_desc_t *bdp;
xfs_inode_t *ip;
xfs_trans_t *tp;
xfs_mount_t *mp;
buf_t *bp;
buf_t *forw;
buf_t *back;
xfs_fileoff_t offset_fsb;
xfs_filblks_t count_fsb;
xfs_filblks_t numblks_fsb;
xfs_bmbt_irec_t imap;
int nimaps;
int nres;
int s;
int error;
int committed;
xfs_fsblock_t firstfsb;
xfs_bmap_free_t free_list;
s = mp_mutex_spinlock(&xfsc_lock);
xfsc_count++;
while (1) {
while (xfsc_list == NULL) {
mp_sv_wait(&xfsc_wait, PRIBIO, &xfsc_lock, s);
s = mp_mutex_spinlock(&xfsc_lock);
}
/*
* Pull a buffer off of the list.
*/
bp = xfsc_list;
ASSERT(bp);
forw = bp->av_forw;
back = bp->av_back;
forw->av_back = back;
back->av_forw = forw;
if (forw == bp) {
xfsc_list = NULL;
} else {
xfsc_list = forw;
}
xfsc_bufcount--;;
ASSERT(xfsc_bufcount >= 0);
mp_mutex_spinunlock(&xfsc_lock, s);
bp->av_forw = bp;
bp->av_back = bp;
ASSERT((bp->b_flags & B_UNINITIAL) == B_UNINITIAL);
ASSERT(bp->b_vp);
buftrace("STRAT_WRITE_CMPL", bp);
bdp = vn_bhv_lookup_unlocked(VN_BHV_HEAD(bp->b_vp),
&xfs_vnodeops);
ASSERT(bdp);
ip = XFS_BHVTOI(bdp);
mp = ip->i_mount;
offset_fsb = XFS_BB_TO_FSBT(mp, bp->b_offset);
count_fsb = XFS_B_TO_FSB(mp, bp->b_bcount);
do {
nres = XFS_DIOSTRAT_SPACE_RES(mp, 0);
xfs_strat_write_bp_trace(XFS_STRAT_UNINT_CMPL, ip, bp);
/*
* Set up a transaction with which to allocate the
* backing store for the file. Do allocations in a
* loop until we get some space in the range we are
* interested in. The other space that might be
* allocated is in the delayed allocation extent
* on which we sit but before our buffer starts.
*/
tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
error = xfs_trans_reserve(tp, nres,
XFS_WRITE_LOG_RES(mp), 0,
XFS_TRANS_PERM_LOG_RES,
XFS_WRITE_LOG_COUNT);
if (error) {
xfs_trans_cancel(tp, 0);
goto error0;
}
xfs_ilock(ip, XFS_ILOCK_EXCL);
xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
xfs_trans_ihold(tp, ip);
xfs_strat_write_bp_trace(XFS_STRAT_ENTER, ip, bp);
/*
* Modify the unwritten extent state of the buffer.
*/
XFS_BMAP_INIT(&free_list, &firstfsb);
nimaps = 1;
error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
XFS_BMAPI_WRITE, &firstfsb,
1, &imap, &nimaps, &free_list);
if (error)
goto error_on_bmapi_transaction;
error = xfs_bmap_finish(&(tp), &(free_list),
firstfsb, &committed);
if (error)
goto error_on_bmapi_transaction;
error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES,
NULL);
xfs_iunlock(ip, XFS_ILOCK_EXCL);
if (error)
goto error0;
if ((numblks_fsb = imap.br_blockcount) == 0) {
/*
* The extent size should alway get bigger
* otherwise the loop is stuck.
*/
ASSERT(imap.br_blockcount);
break;
}
offset_fsb += numblks_fsb;
count_fsb -= numblks_fsb;
} while (count_fsb > 0);
bp->b_flags &= ~(B_UNINITIAL);
xfs_strat_write_bp_trace(XFS_STRAT_UNINT_CMPL, ip, bp);
buftrace("STRAT_WRITE_CMPL", bp);
/* fall through on normal completion */
error0:
if (error) {
bp->b_flags |= B_ERROR;
bp->b_error = error;
}
biodone(bp);
error = 0;
s = mp_mutex_spinlock(&xfsc_lock);
}
error_on_bmapi_transaction:
xfs_bmap_cancel(&free_list);
xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
xfs_iunlock(ip, XFS_ILOCK_EXCL);
goto error0;
}
/*
* This is the completion routine for the heap-allocated buffers
* used to write out a buffer which becomes fragmented during
* xfs_strat_write(). It must coordinate with xfs_strat_write()
* to properly mark the lead buffer as done when necessary and
* to free the subordinate buffer.
*/
STATIC void
xfs_strat_write_relse(
buf_t *rbp)
{
int s;
buf_t *leader;
buf_t *forw;
buf_t *back;
s = mutex_spinlock(&xfs_strat_lock);
ASSERT(rbp->b_flags & B_DONE);
forw = (buf_t*)rbp->b_fsprivate2;
back = (buf_t*)rbp->b_fsprivate;
ASSERT(back != NULL);
ASSERT(((buf_t *)back->b_fsprivate2) == rbp);
ASSERT((forw == NULL) || (((buf_t *)forw->b_fsprivate) == rbp));
/*
* Pull ourselves from the list.
*/
back->b_fsprivate2 = forw;
if (forw != NULL) {
forw->b_fsprivate = back;
}
if ((forw == NULL) &&
(back->b_flags & B_LEADER) &&
!(back->b_flags & B_PARTIAL)) {
/*
* We are the only buffer in the list and the lead buffer
* has cleared the B_PARTIAL bit to indicate that all
* subordinate buffers have been issued. That means it
* is time to finish off the lead buffer.
*/
leader = back;
if (rbp->b_flags & B_ERROR) {
leader->b_flags |= B_ERROR;
leader->b_error = XFS_ERROR(rbp->b_error);
ASSERT(leader->b_error != EINVAL);
}
leader->b_flags &= ~B_LEADER;
mutex_spinunlock(&xfs_strat_lock, s);
iodone(leader);
} else {
/*
* Either there are still other buffers in the list or
* not all of the subordinate buffers have yet been issued.
* In this case just pass any errors on to the lead buffer.
*/
while (!(back->b_flags & B_LEADER)) {
back = (buf_t*)back->b_fsprivate;
}
ASSERT(back != NULL);
ASSERT(back->b_flags & B_LEADER);
leader = back;
if (rbp->b_flags & B_ERROR) {
leader->b_flags |= B_ERROR;
leader->b_error = XFS_ERROR(rbp->b_error);
ASSERT(leader->b_error != EINVAL);
}
mutex_spinunlock(&xfs_strat_lock, s);
}
rbp->b_fsprivate = NULL;
rbp->b_fsprivate2 = NULL;
rbp->b_relse = NULL;
if (BP_ISMAPPED(rbp)) {
bp_mapout(rbp);
}
freerbuf(rbp);
}
#ifdef DEBUG
/*ARGSUSED*/
void
xfs_check_rbp(
xfs_inode_t *ip,
buf_t *bp,
buf_t *rbp,
int locked)
{
xfs_mount_t *mp;
int nimaps;
xfs_bmbt_irec_t imap;
xfs_fileoff_t rbp_offset_fsb;
xfs_filblks_t rbp_len_fsb;
pfd_t *pfdp;
xfs_fsblock_t firstblock;
int error;
mp = ip->i_mount;
rbp_offset_fsb = XFS_BB_TO_FSBT(mp, rbp->b_offset);
rbp_len_fsb = XFS_BB_TO_FSB(mp, rbp->b_offset+BTOBB(rbp->b_bcount)) -
XFS_BB_TO_FSBT(mp, rbp->b_offset);
nimaps = 1;
if (!locked) {
xfs_ilock(ip, XFS_ILOCK_SHARED);
}
firstblock = NULLFSBLOCK;
error = xfs_bmapi(NULL, ip, rbp_offset_fsb, rbp_len_fsb, 0,
&firstblock, 0, &imap, &nimaps, NULL);
if (!locked) {
xfs_iunlock(ip, XFS_ILOCK_SHARED);
}
if (error) {
return;
}
ASSERT(imap.br_startoff == rbp_offset_fsb);
ASSERT(imap.br_blockcount == rbp_len_fsb);
ASSERT((XFS_FSB_TO_DB(ip, imap.br_startblock) +
XFS_BB_FSB_OFFSET(mp, rbp->b_offset)) ==
rbp->b_blkno);
if (rbp->b_flags & B_PAGEIO) {
pfdp = NULL;
pfdp = getnextpg(rbp, pfdp);
ASSERT(pfdp != NULL);
ASSERT(dtopt(rbp->b_offset) == pfdp->pf_pageno);
}
if (rbp->b_flags & B_MAPPED) {
ASSERT(BTOBB(poff(rbp->b_un.b_addr)) ==
dpoff(rbp->b_offset));
}
}
/*
* Verify that the given buffer is going to the right place in its
* file. Also check that it is properly mapped and points to the
* right page. We can only do a trylock from here in order to prevent
* deadlocks, since this is called from the strategy routine.
*/
void
xfs_check_bp(
xfs_inode_t *ip,
buf_t *bp)
{
xfs_mount_t *mp;
int nimaps;
xfs_bmbt_irec_t imap[2];
xfs_fileoff_t bp_offset_fsb;
xfs_filblks_t bp_len_fsb;
pfd_t *pfdp;
int locked;
xfs_fsblock_t firstblock;
int error;
int bmapi_flags;
mp = ip->i_mount;
if (bp->b_flags & B_PAGEIO) {
pfdp = NULL;
pfdp = getnextpg(bp, pfdp);
ASSERT(pfdp != NULL);
ASSERT(dtopt(bp->b_offset) == pfdp->pf_pageno);
if (dpoff(bp->b_offset)) {
ASSERT(bp->b_flags & B_MAPPED);
}
}
if (bp->b_flags & B_MAPPED) {
ASSERT(BTOBB(poff(bp->b_un.b_addr)) ==
dpoff(bp->b_offset));
}
bp_offset_fsb = XFS_BB_TO_FSBT(mp, bp->b_offset);
bp_len_fsb = XFS_BB_TO_FSB(mp, bp->b_offset + BTOBB(bp->b_bcount)) -
XFS_BB_TO_FSBT(mp, bp->b_offset);
ASSERT(bp_len_fsb > 0);
if (bp->b_flags & B_UNINITIAL) {
nimaps = 2;
bmapi_flags = XFS_BMAPI_WRITE|XFS_BMAPI_IGSTATE;
} else {
nimaps = 1;
bmapi_flags = 0;
}
locked = xfs_ilock_nowait(ip, XFS_ILOCK_SHARED);
if (!locked) {
return;
}
firstblock = NULLFSBLOCK;
error = xfs_bmapi(NULL, ip, bp_offset_fsb, bp_len_fsb, bmapi_flags,
&firstblock, 0, imap, &nimaps, NULL);
xfs_iunlock(ip, XFS_ILOCK_SHARED);
if (error) {
return;
}
ASSERT(nimaps == 1);
ASSERT(imap->br_startoff == bp_offset_fsb);
ASSERT(imap->br_blockcount == bp_len_fsb);
ASSERT((XFS_FSB_TO_DB(ip, imap->br_startblock) +
XFS_BB_FSB_OFFSET(mp, bp->b_offset)) ==
bp->b_blkno);
}
#endif /* DEBUG */
/*
* xfs_strat_write_unwritten is called for buffered
* writes of preallocated but unwritten extents. These
* require a transaction to indicate the extent has been
* written.
* The write is set up by a call to xfs_bmapi with
* an "ignore state" flag. After the I/O has completed,
* xfs_strat_write_iodone is called to queue the completed
* buffer to xfs_strat_write_complete. That routine
* calls xfs_bmapi() with a write flag, and issues the
* required transaction.
*/
STATIC int
xfs_strat_write_unwritten(
bhv_desc_t *bdp,
buf_t *bp)
{
xfs_fileoff_t offset_fsb;
off_t offset_fsb_bb;
xfs_filblks_t count_fsb;
/* REFERENCED */
xfs_mount_t *mp;
xfs_inode_t *ip;
xfs_bmap_free_t free_list;
xfs_fsblock_t first_block;
int error;
int nimaps;
xfs_bmbt_irec_t imap[XFS_BMAP_MAX_NMAP];
#define XFS_STRAT_WRITE_IMAPS 2
/*
* If XFS_STRAT_WRITE_IMAPS is changed then the definition
* of XFS_STRATW_LOG_RES in xfs_trans.h must be changed to
* reflect the new number of extents that can actually be
* allocated in a single transaction.
*/
ip = XFS_BHVTOI(bdp);
mp = ip->i_mount;
error = 0;
/*
* Drop the count of queued buffers. We need to do
* this before the bdstrat because callers of
* VOP_FLUSHINVAL_PAGES(), for example, may expect the queued_buf
* count to be down when it rturns. See xfs_itruncate_start.
*/
atomicAddInt(&(ip->i_queued_bufs), -1);
ASSERT(ip->i_queued_bufs >= 0);
/*
* Don't proceed if we're forcing a shutdown.
*/
if (XFS_FORCED_SHUTDOWN(mp)) {
return xfs_bioerror_relse(bp);
}
if (XFS_IS_QUOTA_ON(mp)) {
if (XFS_NOT_DQATTACHED(mp, ip)) {
if (error = xfs_qm_dqattach(ip, 0)) {
return xfs_bioerror_relse(bp);
}
}
}
/*
*
*/
offset_fsb = XFS_BB_TO_FSBT(mp, bp->b_offset);
count_fsb = XFS_B_TO_FSB(mp, bp->b_bcount);
offset_fsb_bb = XFS_FSB_TO_BB(mp, offset_fsb);
ASSERT((offset_fsb_bb == bp->b_offset) || (count_fsb == 1));
xfs_ilock(ip, XFS_ILOCK_EXCL);
xfs_strat_write_bp_trace(XFS_STRAT_ENTER, ip, bp);
/*
* Modify the unwritten extent state of the buffer.
*/
XFS_BMAP_INIT(&(free_list), &(first_block));
nimaps = 2;
error = xfs_bmapi(NULL, ip, offset_fsb, count_fsb,
XFS_BMAPI_WRITE|XFS_BMAPI_IGSTATE, &first_block,
0, imap, &nimaps, &free_list);
if (error) {
xfs_iunlock(ip, XFS_ILOCK_EXCL);
goto error0;
}
ASSERT(nimaps == 1);
if (bp->b_blkno < 0) {
bp->b_blkno = XFS_FSB_TO_DB(ip, imap->br_startblock) +
(bp->b_offset - offset_fsb_bb);
}
/*
* Before dropping the inode lock, clear any
* read-ahead state since in allocating space
* here we may have made it invalid.
*/
xfs_iunlock(ip, XFS_ILOCK_EXCL);
XFS_INODE_CLEAR_READ_AHEAD(ip);
/*
* For an unwritten buffer, do a write of the buffer,
* and commit the transaction which modified the
* extent state.
*/
ASSERT((imap[0].br_startoff <= offset_fsb) &&
(imap[0].br_blockcount >=
(offset_fsb + count_fsb - imap[0].br_startoff)));
ASSERT(bp->b_iodone == NULL);
bp->b_iodone = xfs_strat_write_iodone;
xfs_strat_write_bp_trace(XFS_STRAT_UNINT, ip, bp);
xfs_check_bp(ip, bp);
xfsbdstrat(mp, bp);
return 0;
error0:
bp->b_flags |= B_ERROR;
bp->b_error = error;
biodone(bp);
return error;
}
/*
* This is called to convert all delayed allocation blocks in the given
* range back to 'holes' in the file. It is used when a buffer will not
* be able to be written out due to disk errors in the allocation calls.
*/
STATIC void
xfs_delalloc_cleanup(
xfs_inode_t *ip,
xfs_fileoff_t start_fsb,
xfs_filblks_t count_fsb)
{
xfs_fsblock_t first_block;
int nimaps;
int done;
int error;
int n;
#define XFS_CLEANUP_MAPS 4
xfs_bmbt_irec_t imap[XFS_CLEANUP_MAPS];
ASSERT(count_fsb < 0xffff000);
xfs_ilock(ip, XFS_ILOCK_EXCL);
while (count_fsb != 0) {
first_block = NULLFSBLOCK;
nimaps = XFS_CLEANUP_MAPS;
error = xfs_bmapi(NULL, ip, start_fsb, count_fsb, 0,
&first_block, 1, imap, &nimaps, NULL);
if (error) {
xfs_iunlock(ip, XFS_ILOCK_EXCL);
return;
}
ASSERT(nimaps > 0);
n = 0;
while (n < nimaps) {
if (imap[n].br_startblock == DELAYSTARTBLOCK) {
if (!XFS_FORCED_SHUTDOWN(ip->i_mount))
xfs_force_shutdown(ip->i_mount,
XFS_METADATA_IO_ERROR);
error = xfs_bunmapi(NULL, ip,
imap[n].br_startoff,
imap[n].br_blockcount,
0, 1, &first_block, NULL,
&done);
if (error) {
xfs_iunlock(ip, XFS_ILOCK_EXCL);
return;
}
ASSERT(done);
}
start_fsb += imap[n].br_blockcount;
count_fsb -= imap[n].br_blockcount;
ASSERT(count_fsb < 0xffff000);
n++;
}
}
xfs_iunlock(ip, XFS_ILOCK_EXCL);
}
/*
* xfs_strat_write is called for buffered writes which
* require a transaction. These cases are:
* - Delayed allocation (since allocation now takes place).
* - Writing a previously unwritten extent.
*/
STATIC int
xfs_strat_write(
bhv_desc_t *bdp,
buf_t *bp)
{
xfs_fileoff_t offset_fsb;
off_t offset_fsb_bb;
xfs_fileoff_t map_start_fsb;
xfs_fileoff_t imap_offset;
xfs_fsblock_t first_block;
xfs_filblks_t count_fsb;
xfs_extlen_t imap_blocks;
#ifdef DEBUG
off_t last_rbp_offset;
xfs_extlen_t last_rbp_bcount;
daddr_t last_rbp_blkno;
#endif
/* REFERENCED */
int rbp_count;
buf_t *rbp;
xfs_mount_t *mp;
xfs_inode_t *ip;
xfs_trans_t *tp;
int error;
xfs_bmap_free_t free_list;
xfs_bmbt_irec_t *imapp;
int rbp_offset;
int rbp_len;
int set_lead;
int s;
/* REFERENCED */
int loops;
int imap_index;
int nimaps;
int committed;
xfs_lsn_t commit_lsn;
xfs_bmbt_irec_t imap[XFS_BMAP_MAX_NMAP];
#define XFS_STRAT_WRITE_IMAPS 2
/*
* If XFS_STRAT_WRITE_IMAPS is changed then the definition
* of XFS_STRATW_LOG_RES in xfs_trans.h must be changed to
* reflect the new number of extents that can actually be
* allocated in a single transaction.
*/
XFSSTATS.xs_xstrat_bytes += bp->b_bcount;
if ((bp->b_flags & B_UNINITIAL) == B_UNINITIAL)
return xfs_strat_write_unwritten(bdp, bp);
ip = XFS_BHVTOI(bdp);
mp = ip->i_mount;
set_lead = 0;
rbp_count = 0;
error = 0;
bp->b_flags |= B_STALE;
/*
* Drop the count of queued buffers. We need to do
* this before the bdstrat(s) because callers of
* VOP_FLUSHINVAL_PAGES(), for example, may expect the queued_buf
* count to be down when it rturns. See xfs_itruncate_start.
*/
atomicAddInt(&(ip->i_queued_bufs), -1);
ASSERT(ip->i_queued_bufs >= 0);
/*
* Don't proceed if we're forcing a shutdown.
* We may not have bmap'd all the blocks needed.
*/
if (XFS_FORCED_SHUTDOWN(mp)) {
return xfs_bioerror_relse(bp);
}
if (XFS_IS_QUOTA_ON(mp)) {
if (XFS_NOT_DQATTACHED(mp, ip)) {
if (error = xfs_qm_dqattach(ip, 0)) {
return xfs_bioerror_relse(bp);
}
}
}
/*
* It is possible that the buffer does not start on a block
* boundary in the case where the system page size is less
* than the file system block size. In this case, the buffer
* is guaranteed to be only a single page long, so we know
* that we will allocate the block for it in a single extent.
* Thus, the looping code below does not have to worry about
* this case. It is only handled in the fast path code.
*/
ASSERT(bp->b_blkno == -1);
offset_fsb = XFS_BB_TO_FSBT(mp, bp->b_offset);
count_fsb = XFS_B_TO_FSB(mp, bp->b_bcount);
offset_fsb_bb = XFS_FSB_TO_BB(mp, offset_fsb);
ASSERT((offset_fsb_bb == bp->b_offset) || (count_fsb == 1));
xfs_strat_write_check(ip, offset_fsb,
count_fsb, imap,
XFS_STRAT_WRITE_IMAPS);
map_start_fsb = offset_fsb;
while (count_fsb != 0) {
/*
* Set up a transaction with which to allocate the
* backing store for the file. Do allocations in a
* loop until we get some space in the range we are
* interested in. The other space that might be allocated
* is in the delayed allocation extent on which we sit
* but before our buffer starts.
*/
nimaps = 0;
loops = 0;
while (nimaps == 0) {
tp = xfs_trans_alloc(mp,
XFS_TRANS_STRAT_WRITE);
error = xfs_trans_reserve(tp, 0,
XFS_WRITE_LOG_RES(mp),
0, XFS_TRANS_PERM_LOG_RES,
XFS_WRITE_LOG_COUNT);
if (error) {
xfs_trans_cancel(tp, 0);
bp->b_flags |= B_ERROR;
bp->b_error = error;
goto error0;
}
ASSERT(error == 0);
xfs_ilock(ip, XFS_ILOCK_EXCL);
xfs_trans_ijoin(tp, ip,
XFS_ILOCK_EXCL);
xfs_trans_ihold(tp, ip);
xfs_strat_write_bp_trace(XFS_STRAT_ENTER,
ip, bp);
/*
* Allocate the backing store for the file.
*/
XFS_BMAP_INIT(&(free_list),
&(first_block));
nimaps = XFS_STRAT_WRITE_IMAPS;
error = xfs_bmapi(tp, ip, map_start_fsb, count_fsb,
XFS_BMAPI_WRITE, &first_block, 1,
imap, &nimaps, &free_list);
if (error) {
xfs_bmap_cancel(&free_list);
xfs_trans_cancel(tp,
(XFS_TRANS_RELEASE_LOG_RES |
XFS_TRANS_ABORT));
xfs_iunlock(ip, XFS_ILOCK_EXCL);
bp->b_flags |= B_ERROR;
bp->b_error = error;
goto error0;
}
ASSERT(loops++ <=
(offset_fsb +
XFS_B_TO_FSB(mp, bp->b_bcount)));
error = xfs_bmap_finish(&(tp), &(free_list),
first_block, &committed);
if (error) {
xfs_bmap_cancel(&free_list);
xfs_trans_cancel(tp,
(XFS_TRANS_RELEASE_LOG_RES |
XFS_TRANS_ABORT));
xfs_iunlock(ip, XFS_ILOCK_EXCL);
bp->b_flags |= B_ERROR;
bp->b_error = error;
goto error0;
}
error = xfs_trans_commit(tp,
XFS_TRANS_RELEASE_LOG_RES,
&commit_lsn);
if (error) {
xfs_iunlock(ip, XFS_ILOCK_EXCL);
bp->b_flags |= B_ERROR;
bp->b_error = error;
goto error0;
}
/*
* write the commit lsn if requested into the
* place pointed at by the buffer. This is
* used by IO_DSYNC writes and b_fsprivate3
* should be a pointer to a stack (automatic)
* variable. So be *very* careful if you muck
* with b_fsprivate3.
*/
if (bp->b_fsprivate3)
*(xfs_lsn_t *)bp->b_fsprivate3 = commit_lsn;
/*
* Before dropping the lock, clear any read-ahead
* state since in allocating space here we may have
* made it invalid.
*/
xfs_iunlock(ip, XFS_ILOCK_EXCL);
XFS_INODE_CLEAR_READ_AHEAD(ip);
}
/*
* This is a quick check to see if the first time through
* was able to allocate a single extent over which to
* write.
*/
if ((map_start_fsb == offset_fsb) &&
(imap[0].br_blockcount == count_fsb)) {
ASSERT(nimaps == 1);
/*
* Set the buffer's block number to match
* what we allocated. If the buffer does
* not start on a block boundary (can only
* happen if the block size is larger than
* the page size), then make sure to add in
* the offset of the buffer into the file system
* block to the disk block number to write.
*/
bp->b_blkno =
XFS_FSB_TO_DB(ip, imap[0].br_startblock) +
(bp->b_offset - offset_fsb_bb);
xfs_strat_write_bp_trace(XFS_STRAT_FAST,
ip, bp);
xfs_check_bp(ip, bp);
#ifdef XFSRACEDEBUG
delay_for_intr();
delay(100);
#endif
xfsbdstrat(mp, bp);
XFSSTATS.xs_xstrat_quick++;
return 0;
}
/*
* Bmap couldn't manage to lay the buffer out as
* one extent, so we need to do multiple writes
* to push the data to the multiple extents.
* Write out the subordinate bps asynchronously
* and have their completion functions coordinate
* with the code at the end of this function to
* deal with marking our bp as done when they have
* ALL completed.
*/
XFSSTATS.xs_xstrat_split++;
imap_index = 0;
if (!set_lead) {
bp->b_flags |= B_LEADER | B_PARTIAL;
set_lead = 1;
}
while (imap_index < nimaps) {
rbp = getrbuf(KM_SLEEP);
imapp = &(imap[imap_index]);
ASSERT((imapp->br_startblock !=
DELAYSTARTBLOCK) &&
(imapp->br_startblock !=
HOLESTARTBLOCK));
imap_offset = imapp->br_startoff;
rbp_offset = XFS_FSB_TO_B(mp,
imap_offset -
offset_fsb);
imap_blocks = imapp->br_blockcount;
ASSERT((imap_offset + imap_blocks) <=
(offset_fsb +
XFS_B_TO_FSB(mp, bp->b_bcount)));
rbp_len = XFS_FSB_TO_B(mp,
imap_blocks);
xfs_overlap_bp(bp, rbp, rbp_offset,
rbp_len);
rbp->b_blkno =
XFS_FSB_TO_DB(ip, imapp->br_startblock);
rbp->b_offset = XFS_FSB_TO_BB(mp,
imap_offset);
rbp->b_target = bp->b_target;
xfs_strat_write_subbp_trace(XFS_STRAT_SUB,
ip, bp,
rbp,
last_rbp_offset,
last_rbp_bcount,
last_rbp_blkno);
#ifdef DEBUG
xfs_check_rbp(ip, bp, rbp, 0);
if (rbp_count > 0) {
ASSERT((last_rbp_offset +
BTOBB(last_rbp_bcount)) ==
rbp->b_offset);
ASSERT((rbp->b_blkno <
last_rbp_blkno) ||
(rbp->b_blkno >=
(last_rbp_blkno +
BTOBB(last_rbp_bcount))));
if (rbp->b_blkno <
last_rbp_blkno) {
ASSERT((rbp->b_blkno +
BTOBB(rbp->b_bcount)) <
last_rbp_blkno);
}
}
last_rbp_offset = rbp->b_offset;
last_rbp_bcount = rbp->b_bcount;
last_rbp_blkno = rbp->b_blkno;
#endif
/*
* Link the buffer into the list of subordinate
* buffers started at bp->b_fsprivate2. The
* subordinate buffers use b_fsprivate and
* b_fsprivate2 for back and forw pointers, but
* the lead buffer cannot use b_fsprivate.
* A subordinate buffer can always find the lead
* buffer by searching back through the fsprivate
* fields until it finds the buffer marked with
* B_LEADER.
*/
s = mutex_spinlock(&xfs_strat_lock);
rbp->b_fsprivate = bp;
rbp->b_fsprivate2 = bp->b_fsprivate2;
if (bp->b_fsprivate2 != NULL) {
((buf_t*)(bp->b_fsprivate2))->b_fsprivate =
rbp;
}
bp->b_fsprivate2 = rbp;
mutex_spinunlock(&xfs_strat_lock, s);
rbp->b_relse = xfs_strat_write_relse;
rbp->b_flags |= B_ASYNC;
#ifdef XFSRACEDEBUG
delay_for_intr();
delay(100);
#endif
xfsbdstrat(mp, rbp);
map_start_fsb +=
imapp->br_blockcount;
count_fsb -= imapp->br_blockcount;
ASSERT(count_fsb < 0xffff000);
imap_index++;
}
}
/*
* Now that we've issued all the partial I/Os, check to see
* if they've all completed. If they have then mark the buffer
* as done, otherwise clear the B_PARTIAL flag in the buffer to
* indicate that the last subordinate buffer to complete should
* mark the buffer done. Also, drop the count of queued buffers
* now that we know that all the space underlying the buffer has
* been allocated and it has really been sent out to disk.
*
* Use set_lead to tell whether we kicked off any partial I/Os
* or whether we jumped here after an error before issuing any.
*/
error0:
if (error) {
ASSERT(count_fsb != 0);
/*
* Since we're never going to convert the remaining
* delalloc blocks beneath this buffer into real block,
* get rid of them now.
*/
xfs_delalloc_cleanup(ip, map_start_fsb, count_fsb);
}
if (set_lead) {
s = mutex_spinlock(&xfs_strat_lock);
ASSERT((bp->b_flags & (B_DONE | B_PARTIAL)) == B_PARTIAL);
ASSERT(bp->b_flags & B_LEADER);
if (bp->b_fsprivate2 == NULL) {
/*
* All of the subordinate buffers have completed.
* Call iodone() to note that the I/O has completed.
*/
bp->b_flags &= ~(B_PARTIAL | B_LEADER);
mutex_spinunlock(&xfs_strat_lock, s);
biodone(bp);
return error;
}
bp->b_flags &= ~B_PARTIAL;
mutex_spinunlock(&xfs_strat_lock, s);
} else {
biodone(bp);
}
return error;
}
/*
* Force a shutdown of the filesystem instantly while keeping
* the filesystem consistent. We don't do an unmount here; just shutdown
* the shop, make sure that absolutely nothing persistent happens to
* this filesystem after this point.
*/
void
xfs_force_shutdown(
struct xfs_mount *mp,
int flags)
{
int ntries;
int logerror;
extern dev_t rootdev; /* from sys/systm.h */
#define XFS_MAX_DRELSE_RETRIES 10
logerror = flags & XFS_LOG_IO_ERROR;
/*
* No need to duplicate efforts.
*/
if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
return;
if (XFS_MTOVFS(mp)->vfs_dev == rootdev)
cmn_err(CE_PANIC, "Fatal error on root filesystem");
/*
* This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
* queue up anybody new on the log reservations, and wakes up
* everybody who's sleeping on log reservations and tells
* them the bad news.
*/
if (xfs_log_force_umount(mp, logerror))
return;
if (flags & XFS_CORRUPT_INCORE)
cmn_err(CE_ALERT,
"Corruption of in-memory data detected. Shutting down filesystem: %s",
mp->m_fsname);
else
cmn_err(CE_ALERT,
"I/O Error Detected. Shutting down filesystem: %s",
mp->m_fsname);
cmn_err(CE_ALERT,
"Please umount the filesystem, and rectify the problem(s)");
/*
* Release all delayed write buffers for this device.
* It wouldn't be a fatal error if we couldn't release all
* delwri bufs; in general they all get unpinned eventually.
*/
ntries = 0;
#ifdef XFSERRORDEBUG
{
int nbufs;
while (nbufs = incore_relse(mp->m_dev, 1, 0)) {
printf("XFS: released 0x%x bufs\n", nbufs);
if (ntries >= XFS_MAX_DRELSE_RETRIES) {
printf("XFS: ntries 0x%x\n", ntries);
debug("ntries");
break;
}
delay(++ntries * 5);
}
}
#else
while (incore_relse(mp->m_dev, 1, 0)) {
if (ntries >= XFS_MAX_DRELSE_RETRIES)
break;
delay(++ntries * 5);
}
#endif
}
/*
* Called when we want to stop a buffer from getting written or read.
* We attach the EIO error, muck with its flags, and call biodone
* so that the proper iodone callbacks get called.
*/
int
xfs_bioerror(
buf_t *bp)
{
#ifdef XFSERRORDEBUG
ASSERT(bp->b_flags & B_READ || bp->b_iodone);
#endif
/*
* No need to wait until the buffer is unpinned.
* We aren't flushing it.
*/
buftrace("XFS IOERROR", bp);
bioerror(bp, EIO);
/*
* We're calling biodone, so delete B_DONE flag. Either way
* we have to call the iodone callback, and calling biodone
* probably is the best way since it takes care of
* GRIO as well.
*/
bp->b_flags &= ~(B_READ|B_DELWRI|B_DONE);
bp->b_flags |= B_STALE;
bp->b_bdstrat = NULL;
biodone(bp);
return (EIO);
}
/*
* Same as xfs_bioerror, except that we are releasing the buffer
* here ourselves, and avoiding the biodone call.
* This is meant for userdata errors; metadata bufs come with
* iodone functions attached, so that we can track down errors.
*/
int
xfs_bioerror_relse(
buf_t *bp)
{
int64_t fl;
ASSERT(bp->b_iodone != xfs_buf_iodone_callbacks);
ASSERT(bp->b_iodone != xlog_iodone);
buftrace("XFS IOERRELSE", bp);
fl = bp->b_flags;
/*
* No need to wait until the buffer is unpinned.
* We aren't flushing it.
*
* chunkhold expects B_DONE to be set, whether
* we actually finish the I/O or not. We don't want to
* change that interface.
*/
bp->b_flags &= ~(B_READ|B_DELWRI);
bp->b_flags |= B_DONE|B_STALE;
bp->b_iodone = NULL;
bp->b_bdstrat = NULL;
if (!(fl & B_ASYNC)) {
/*
* Mark b_error and B_ERROR _both_.
* Lot's of chunkcache code assumes that.
* There's no reason to mark error for
* ASYNC buffers.
*/
bioerror(bp, EIO);
vsema(&bp->b_iodonesema);
} else {
brelse(bp);
}
return (EIO);
}
/*
* Prints out an ALERT message about I/O error.
*/
void
xfs_ioerror_alert(
char *func,
struct xfs_mount *mp,
dev_t dev,
daddr_t blkno)
{
cmn_err(CE_ALERT,
"I/O error in filesystem (\"%s\") meta-data dev 0x%x block 0x%x (\"%s\")",
mp->m_fsname,
dev,
blkno,
func);
}
/*
* This isn't an absolute requirement, but it is
* just a good idea to call xfs_read_buf instead of
* directly doing a read_buf call. For one, we shouldn't
* be doing this disk read if we are in SHUTDOWN state anyway,
* so this stops that from happening. Secondly, this does all
* the error checking stuff and the brelse if appropriate for
* the caller, so the code can be a little leaner.
*/
int
xfs_read_buf(
struct xfs_mount *mp,
buftarg_t *target,
daddr_t blkno,
int len,
uint flags,
buf_t **bpp)
{
buf_t *bp;
int error;
bp = read_buf_targ(target, blkno, len, flags);
error = geterror(bp);
if (bp && !error && !XFS_FORCED_SHUTDOWN(mp)) {
*bpp = bp;
} else {
*bpp = NULL;
if (!error)
error = XFS_ERROR(EIO);
if (bp) {
bp->b_flags &= ~(B_DONE|B_DELWRI);
bp->b_flags |= B_STALE;
/*
* brelse clears B_ERROR and b_error
*/
brelse(bp);
}
}
return (error);
}
/*
* Wrapper around bwrite() so that we can trap
* write errors, and act accordingly.
*/
int
xfs_bwrite(
struct xfs_mount *mp,
struct buf *bp)
{
int error;
/*
* XXXsup how does this work for quotas.
*/
ASSERT(bp->b_target);
ASSERT(bp->b_vp == NULL);
bp->b_bdstrat = xfs_bdstrat_cb;
bp->b_fsprivate3 = mp;
if (error = bwrite(bp)) {
ASSERT(mp);
/*
* Cannot put a buftrace here since if the buffer is not
* B_HOLD then we will brelse() the buffer before returning
* from bwrite and we could be tracing a buffer that has
* been reused.
*/
xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
}
return (error);
}
/*
* All xfs metadata buffers except log state machine buffers
* get this attached as their b_bdstrat callback function.
* This is so that we can catch a buffer
* after prematurely unpinning it to forcibly shutdown the filesystem.
*/
int
xfs_bdstrat_cb(struct buf *bp)
{
xfs_mount_t *mp;
mp = bp->b_fsprivate3;
ASSERT(bp->b_target);
if (!XFS_FORCED_SHUTDOWN(mp)) {
#if CELL_IRIX
vnode_t *vp;
vp = bp->b_target->specvp;
ASSERT(vp);
bp->b_bdstrat = NULL;
VOP_STRATEGY(vp, bp);
#else
struct bdevsw *my_bdevsw;
my_bdevsw = bp->b_target->bdevsw;
ASSERT(my_bdevsw != NULL);
bp->b_bdstrat = NULL;
bdstrat(my_bdevsw, bp);
#endif /* CELL_IRIX */
return 0;
} else {
buftrace("XFS__BDSTRAT IOERROR", bp);
/*
* Metadata write that didn't get logged but
* written delayed anyway. These aren't associated
* with a transaction, and can be ignored.
*/
if (bp->b_iodone == NULL &&
(bp->b_flags & B_READ) == 0)
return (xfs_bioerror_relse(bp));
else
return (xfs_bioerror(bp));
}
}
/*
* Wrapper around bdstrat so that we can stop data
* from going to disk in case we are shutting down the filesystem.
* Typically user data goes thru this path; one of the exceptions
* is the superblock.
*/
int
xfsbdstrat(
struct xfs_mount *mp,
struct buf *bp)
{
int dev_major = emajor(bp->b_edev);
ASSERT(mp);
ASSERT(bp->b_target);
if (!XFS_FORCED_SHUTDOWN(mp)) {
/*
* We want priority I/Os to non-XLV disks to go thru'
* griostrategy(). The rest of the I/Os follow the normal
* path, and are uncontrolled. If we want to rectify
* that, use griostrategy2.
*/
if ( (BUF_IS_GRIO(bp)) &&
(dev_major != XLV_MAJOR) ) {
griostrategy(bp);
} else {
#if CELL_IRIX
vnode_t *vp;
vp = bp->b_target->specvp;
ASSERT(vp);
VOP_STRATEGY(vp, bp);
#else
struct bdevsw *my_bdevsw;
my_bdevsw = bp->b_target->bdevsw;
ASSERT(my_bdevsw != NULL);
bdstrat(my_bdevsw, bp);
#endif
}
return 0;
}
buftrace("XFSBDSTRAT IOERROR", bp);
return (xfs_bioerror_relse(bp));
}
/*
* xfs_strategy
*
* This is where all the I/O and all the REAL allocations take
* place. For buffers with -1 for their b_blkno field, we need
* to do a bmap to figure out what to do with them. If it's a
* write we may need to do an allocation, while if it's a read
* we may either need to read from disk or do some block zeroing.
* If b_blkno specifies a real but prevously unwritten block,
* a transaction needs to be initiated to mark the block as
* initialized. If b_blkno specifies a real block, then all
* we need to do is pass the buffer on to the underlying driver.
*/
void
xfs_strategy(
bhv_desc_t *bdp,
buf_t *bp)
{
int s;
xfs_inode_t *ip;
xfs_mount_t *mp;
ip = XFS_BHVTOI(bdp);
mp = ip->i_mount;
bp->b_target = (bp->b_edev == mp->m_dev) ? mp->m_ddev_targp :
&mp->m_rtdev_targ;
/*
* If this is just a buffer whose underlying disk space
* is already allocated, then just do the requested I/O.
*/
buftrace("XFS_STRATEGY", bp);
if (bp->b_blkno >= 0 && !(bp->b_flags & B_UNINITIAL)) {
xfs_check_bp(ip, bp);
/*
* XXXsup We should probably ignore FORCED_SHUTDOWN
* in this case. The disk space is already allocated,
* and this seems like data loss that can be avoided.
* But I need to test it first.
*/
(void) xfsbdstrat(ip->i_mount, bp);
return;
}
/*
* If we're reading, then we need to find out how the
* portion of the file required for this buffer is layed
* out and zero/read in the appropriate data.
*/
if (bp->b_flags & B_READ) {
xfs_strat_read(bdp, bp);
return;
}
if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
xfs_bioerror_relse(bp);
return;
}
/*
* Here we're writing the file and probably need to allocate
* some underlying disk space or to mark it as initialized.
* If the buffer is being written asynchronously by bdflush()
* then we queue if for the xfsds so that we won't put
* bdflush() to sleep.
*/
if ((bp->b_flags & (B_ASYNC | B_BDFLUSH)) == (B_ASYNC | B_BDFLUSH) &&
(xfsd_count > 0)) {
s = mp_mutex_spinlock(&xfsd_lock);
/*
* Queue the buffer at the end of the list.
* Bump the inode count of the number of queued buffers.
*/
if (xfsd_list == NULL) {
bp->av_forw = bp;
bp->av_back = bp;
xfsd_list = bp;
} else {
bp->av_back = xfsd_list->av_back;
xfsd_list->av_back->av_forw = bp;
xfsd_list->av_back = bp;
bp->av_forw = xfsd_list;
}
/*
* Store the behavior pointer where the xfsds can find
* it so that we don't have to lookup the XFS behavior
* from the vnode in the buffer again.
*/
bp->b_private = bdp;
xfsd_bufcount++;
ASSERT(ip->i_queued_bufs >= 0);
atomicAddInt(&(ip->i_queued_bufs), 1);
(void)sv_signal(&xfsd_wait);
mp_mutex_spinunlock(&xfsd_lock, s);
} else {
/*
* We're not going to queue it for the xfsds, but bump the
* inode's count anyway so that we can tell that this
* buffer is still on its way out.
*/
ASSERT(ip->i_queued_bufs >= 0);
atomicAddInt(&(ip->i_queued_bufs), 1);
xfs_strat_write(bdp, bp);
}
}
/*
* This is called from xfs_init() to start the xfs daemons.
* We'll start with a minimum of 4 of them, and add 1
* for each 128 MB of memory up to 1 GB. That should
* be enough.
*/
void
xfs_start_daemons(void)
{
int num_daemons;
int i;
int num_pages;
extern int xfsd_pri;
num_daemons = 4;
#if _MIPS_SIM != _ABI64
/*
* For small memory systems we reduce the number of daemons
* to conserve memory. For systems with less than 32 MB of
* memory we go with 2 daemons, and for systems with less
* than 48 MB of memory we go with 3.
*/
if (physmem < 8192) {
num_daemons = 2;
} else if (physmem < 12288) {
num_daemons = 3;
}
#endif
num_pages = (int)physmem - 32768;
while ((num_pages > 0) && (num_daemons < 13)) {
num_pages -= 32768;
num_daemons++;
}
ASSERT(num_daemons <= 13);
#define XFSD_SSIZE (2*KTHREAD_DEF_STACKSZ)
/*
* Start the xfs_strat_completion daemon, and the
* xfsds. For now, use the same priority.
*/
sthread_create("xfsc", 0, XFSD_SSIZE, 0, xfsd_pri, KT_PS,
(st_func_t *)xfs_strat_comp, 0, 0, 0, 0);
for (i = 0; i < num_daemons; i++) {
sthread_create("xfsd", 0, XFSD_SSIZE, 0, xfsd_pri, KT_PS,
(st_func_t *)xfsd, 0, 0, 0, 0);
}
#undef XFSD_SSIZE
return;
}
/*
* This is the main loop for the xfs daemons.
* From here they wait in a loop for buffers which will
* require transactions to write out and process them as they come.
* This way we never force bdflush() to wait on one of our transactions,
* thereby keeping the system happier and preventing buffer deadlocks.
*/
STATIC int
xfsd(void)
{
int s;
buf_t *bp;
buf_t *forw;
buf_t *back;
bhv_desc_t *bdp;
s = mp_mutex_spinlock(&xfsd_lock);
xfsd_count++;
while (1) {
while (xfsd_list == NULL) {
mp_sv_wait(&xfsd_wait, PRIBIO, &xfsd_lock, s);
s = mp_mutex_spinlock(&xfsd_lock);
}
/*
* Pull a buffer off of the list.
*/
bp = xfsd_list;
forw = bp->av_forw;
back = bp->av_back;
forw->av_back = back;
back->av_forw = forw;
if (forw == bp) {
xfsd_list = NULL;
} else {
xfsd_list = forw;
}
xfsd_bufcount--;;
ASSERT(xfsd_bufcount >= 0);
mp_mutex_spinunlock(&xfsd_lock, s);
bp->av_forw = bp;
bp->av_back = bp;
ASSERT((bp->b_flags & (B_BUSY | B_ASYNC | B_READ)) ==
(B_BUSY | B_ASYNC));
XFSSTATS.xs_xfsd_bufs++;
bdp = bp->b_private;
bp->b_private = NULL;
xfs_strat_write(bdp, bp);
s = mp_mutex_spinlock(&xfsd_lock);
}
}
struct dio_s {
bhv_desc_t *bdp;
cred_t *cr;
int ioflag;
};
/*
* xfs_inval_cached_pages()
* This routine is responsible for keeping direct I/O and buffered I/O
* somewhat coherent. From here we make sure that we're at least
* temporarily holding the inode I/O lock exclusively and then call
* the page cache to flush and invalidate any cached pages. If there
* are no cached pages this routine will be very quick.
*/
void
xfs_inval_cached_pages(
xfs_inode_t *ip,
off_t offset,
off_t len)
{
vnode_t *vp;
int relock;
__uint64_t flush_end;
vp = XFS_ITOV(ip);
if (!VN_CACHED(vp)) {
return;
}
/*
* We need to get the I/O lock exclusively in order
* to safely invalidate pages and mappings.
*/
relock = ismrlocked(&(ip->i_iolock), MR_ACCESS);
if (relock) {
xfs_iunlock(ip, XFS_IOLOCK_SHARED);
xfs_ilock(ip, XFS_IOLOCK_EXCL);
}
/*
* Round up to the next page boundary and then back
* off by one byte. We back off by one because this
* is a first byte/last byte interface rather than
* a start/len interface. We round up to a page
* boundary because the page/chunk cache code is
* slightly broken and won't invalidate all the right
* buffers otherwise.
*
* We also have to watch out for overflow, so if we
* go over the maximum off_t value we just pull back
* to that max.
*/
flush_end = (__uint64_t)ctooff(offtoc(offset + len)) - 1;
if (flush_end > (__uint64_t)LONGLONG_MAX) {
flush_end = LONGLONG_MAX;
}
xfs_inval_cached_trace(ip, offset, len, ctooff(offtoct(offset)),
flush_end);
VOP_FLUSHINVAL_PAGES(vp, ctooff(offtoct(offset)), (off_t)flush_end,
FI_REMAPF_LOCKED);
if (relock) {
xfs_iunlock(ip, XFS_IOLOCK_EXCL);
xfs_ilock(ip, XFS_IOLOCK_SHARED);
}
}
/*
* xfs_dio_read()
* This routine issues the calls to the disk device strategy routine
* for file system read made using direct I/O from user space.
* The I/Os for each extent involved are issued at once.
*
* RETURNS:
* error
*/
STATIC int
xfs_dio_read(
xfs_dio_t *diop)
{
buf_t *bp;
xfs_inode_t *ip;
xfs_trans_t *tp;
xfs_mount_t *mp;
xfs_bmbt_irec_t imaps[XFS_BMAP_MAX_NMAP], *imapp;
buf_t *bps[XFS_BMAP_MAX_NMAP], *nbp;
xfs_fileoff_t offset_fsb;
xfs_fsblock_t firstfsb;
xfs_filblks_t count_fsb;
xfs_bmap_free_t free_list;
caddr_t base;
ssize_t resid, count, totxfer;
off_t offset, offset_this_req, bytes_this_req, trail = 0;
int i, j, error, reccount;
int end_of_file, bufsissued, totresid;
int blk_algn, rt;
int unwritten;
uint lock_mode;
xfs_fsize_t new_size;
CHECK_GRIO_TIMESTAMP(bp, 40);
bp = diop->xd_bp;
ip = diop->xd_ip;
mp = diop->xd_mp;
blk_algn = diop->xd_blkalgn;
base = bp->b_un.b_addr;
error = resid = totxfer = end_of_file = 0;
offset = BBTOOFF((off_t)bp->b_blkno);
totresid = count = bp->b_bcount;
/*
* Determine if this file is using the realtime volume.
*/
rt = (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
/*
* Process the request until:
* 1) an I/O error occurs
* 2) end of file is reached.
* 3) end of device (driver error) occurs
* 4) request is completed.
*/
while (!error && !end_of_file && !resid && count) {
offset_fsb = XFS_B_TO_FSBT(mp, offset);
count_fsb = XFS_B_TO_FSB(mp, count);
tp = NULL;
unwritten = 0;
retry:
XFS_BMAP_INIT(&free_list, &firstfsb);
/*
* Read requests will be issued
* up to XFS_BMAP_MAX_MAP at a time.
*/
reccount = XFS_BMAP_MAX_NMAP;
imapp = &imaps[0];
CHECK_GRIO_TIMESTAMP(bp, 40);
lock_mode = xfs_ilock_map_shared(ip);
CHECK_GRIO_TIMESTAMP(bp, 40);
/*
* Issue the bmapi() call to get the extent info.
*/
CHECK_GRIO_TIMESTAMP(bp, 40);
error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
0, &firstfsb, 0, imapp,
&reccount, &free_list);
CHECK_GRIO_TIMESTAMP(bp, 40);
xfs_iunlock_map_shared(ip, lock_mode);
if (error)
break;
/*
* xfs_bmapi() did not return an error but the
* reccount was zero. This means that a delayed write is
* in progress and it is necessary to call xfs_bmapi() again
* to map the correct portion of the file.
*/
if ((!error) && (reccount == 0)) {
goto retry;
}
/*
* Run through each extent.
*/
bufsissued = 0;
for (i = 0; (i < reccount) && (!end_of_file) && (count);
i++) {
imapp = &imaps[i];
unwritten = !(imapp->br_state == XFS_EXT_NORM);
bytes_this_req =
XFS_FSB_TO_B(mp, imapp->br_blockcount) -
BBTOB(blk_algn);
ASSERT(bytes_this_req);
offset_this_req =
XFS_FSB_TO_B(mp, imapp->br_startoff) +
BBTOB(blk_algn);
/*
* Reduce request size, if it
* is longer than user buffer.
*/
if (bytes_this_req > count) {
bytes_this_req = count;
}
/*
* Check if this is the end of the file.
*/
new_size = offset_this_req + bytes_this_req;
if (new_size >ip->i_d.di_size) {
/*
* If trying to read past end of
* file, shorten the request size.
*/
xfs_ilock(ip, XFS_ILOCK_SHARED);
if (new_size > ip->i_d.di_size) {
if (ip->i_d.di_size > offset_this_req) {
trail = ip->i_d.di_size -
offset_this_req;
bytes_this_req = trail;
bytes_this_req &= ~BBMASK;
bytes_this_req += BBSIZE;
} else {
bytes_this_req = 0;
}
end_of_file = 1;
if (!bytes_this_req) {
xfs_iunlock(ip, XFS_ILOCK_SHARED);
break;
}
}
xfs_iunlock(ip, XFS_ILOCK_SHARED);
}
/*
* Do not do I/O if there is a hole in the file.
* Do not read if the blocks are unwritten.
*/
if ((imapp->br_startblock == HOLESTARTBLOCK) ||
unwritten) {
/*
* Physio() has already mapped user address.
*/
bzero(base, bytes_this_req);
/*
* Bump the transfer count.
*/
if (trail)
totxfer += trail;
else
totxfer += bytes_this_req;
} else {
/*
* Setup I/O request for this extent.
*/
CHECK_GRIO_TIMESTAMP(bp, 40);
bps[bufsissued++]= nbp = getphysbuf(bp->b_edev);
CHECK_GRIO_TIMESTAMP(bp, 40);
nbp->b_flags = bp->b_flags;
nbp->b_grio_private = bp->b_grio_private;
/* b_iopri */
nbp->b_error = 0;
nbp->b_target = bp->b_target;
if (rt) {
nbp->b_blkno = XFS_FSB_TO_BB(mp,
imapp->br_startblock);
} else {
nbp->b_blkno = XFS_FSB_TO_DADDR(mp,
imapp->br_startblock) +
blk_algn;
}
ASSERT(bytes_this_req);
nbp->b_bcount = bytes_this_req;
nbp->b_un.b_addr = base;
/*
* Issue I/O request.
*/
CHECK_GRIO_TIMESTAMP(nbp, 40);
(void) xfsbdstrat(mp, nbp);
if (error = geterror(nbp)) {
biowait(nbp);
nbp->b_flags = 0;
nbp->b_un.b_addr = 0;
nbp->b_grio_private = 0; /* b_iopri */
putphysbuf( nbp );
bps[bufsissued--] = 0;
break;
}
}
/*
* update pointers for next round.
*/
base += bytes_this_req;
offset += bytes_this_req;
count -= bytes_this_req;
blk_algn= 0;
} /* end of for loop */
/*
* Wait for I/O completion and recover buffers.
*/
for (j = 0; j < bufsissued ; j++) {
nbp = bps[j];
biowait(nbp);
nbp->b_flags &= ~B_GR_BUF; /* Why? B_PRV_BUF? */
if (!error)
error = geterror(nbp);
if (!error && !resid) {
resid = nbp->b_resid;
/*
* prevent adding up partial xfers
*/
if (trail && (j == (bufsissued -1 ))) {
if (resid <= (nbp->b_bcount - trail) )
totxfer += trail;
} else {
totxfer += (nbp->b_bcount - resid);
}
}
nbp->b_flags = 0;
nbp->b_bcount = 0;
nbp->b_un.b_addr = 0;
nbp->b_grio_private = 0; /* b_iopri */
putphysbuf( nbp );
}
} /* end of while loop */
/*
* Fill in resid count for original buffer.
* if any of the io's fail, the whole thing fails
*/
if (error) {
totxfer = 0;
}
bp->b_resid = totresid - totxfer;
return (error);
}
/*
* xfs_dio_write()
* This routine issues the calls to the disk device strategy routine
* for file system writes made using direct I/O from user space. The
* I/Os are issued one extent at a time.
*
* RETURNS:
* error
*/
STATIC int
xfs_dio_write(
xfs_dio_t *diop)
{
buf_t *bp;
struct dio_s *dp;
xfs_inode_t *ip;
xfs_trans_t *tp;
vnode_t *vp;
xfs_mount_t *mp;
xfs_bmbt_irec_t imaps[XFS_BMAP_MAX_NMAP], *imapp;
buf_t *nbp;
xfs_fileoff_t offset_fsb;
xfs_fsblock_t firstfsb;
xfs_filblks_t count_fsb, datablocks;
xfs_bmap_free_t free_list;
caddr_t base;
ssize_t resid, count, totxfer;
off_t offset, offset_this_req, bytes_this_req;
int error, reccount, bmapi_flag;
int end_of_file, totresid, exist;
int blk_algn, rt, numrtextents, sbrtextsize, iprtextsize;
int committed, unwritten, using_quotas;
xfs_fsize_t new_size;
int nres;
bp = diop->xd_bp;
dp = diop->xd_dp;
vp = diop->xd_vp;
ip = diop->xd_ip;
mp = diop->xd_mp;
blk_algn = diop->xd_blkalgn;
base = bp->b_un.b_addr;
error = resid = totxfer = end_of_file = 0;
offset = BBTOOFF((off_t)bp->b_blkno);
rt = using_quotas = 0;
numrtextents = iprtextsize = sbrtextsize = 0;
totresid = count = bp->b_bcount;
/*
* Determine if this file is using the realtime volume.
*/
if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) {
rt = 1;
iprtextsize = sbrtextsize = mp->m_sb.sb_rextsize;
if (ip->i_d.di_extsize) {
iprtextsize = ip->i_d.di_extsize;
}
}
if (XFS_IS_QUOTA_ON(mp)) {
using_quotas = 1;
if (XFS_NOT_DQATTACHED(mp, ip)) {
if (error = xfs_qm_dqattach(ip, 0))
goto error0;
}
}
/*
* Process the request until:
* 1) an I/O error occurs
* 2) end of file is reached.
* 3) end of device (driver error) occurs
* 4) request is completed.
*/
while (!error && !end_of_file && !resid && count) {
offset_fsb = XFS_B_TO_FSBT(mp, offset);
count_fsb = XFS_B_TO_FSB(mp, count);
tp = NULL;
exist = 1;
unwritten = 0;
retry:
XFS_BMAP_INIT(&free_list, &firstfsb);
/*
* We need to call bmapi() with the read flag set first to
* determine the existing
* extents. This is done so that the correct amount
* of space can be reserved in the transaction
* structure. Also, a check is needed to see if the
* extents are for valid blocks but also unwritten.
* If so, again a transaction needs to be reserved.
*/
reccount = 1;
xfs_ilock(ip, XFS_ILOCK_EXCL);
error = xfs_bmapi(NULL, ip, offset_fsb,
count_fsb, 0, &firstfsb, 0, imaps,
&reccount, 0);
if (error) {
xfs_iunlock(ip, XFS_ILOCK_EXCL);
break;
}
/*
* Get a pointer to the current extent map.
*/
imapp = &imaps[0];
/*
* Check if the file extents already exist
*/
if ((imapp->br_startblock == DELAYSTARTBLOCK) ||
(imapp->br_startblock == HOLESTARTBLOCK)) {
exist = 0;
} else if (imapp->br_state == XFS_EXT_UNWRITTEN) {
unwritten = 1;
}
reccount = 1;
count_fsb = imapp->br_blockcount;
/*
* If blocks are not yet allocated for this part of
* the file, allocate space for the transactions.
*/
if (!exist) {
bmapi_flag = XFS_BMAPI_WRITE;
if (rt) {
/*
* Round up to even number of extents.
*/
numrtextents =
(count_fsb + iprtextsize - 1)/
sbrtextsize;
datablocks = 0;
} else {
/*
* If this is a write to the data
* partition, reserve the space.
*/
datablocks = count_fsb;
}
/*
* Setup transaction.
*/
tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
xfs_iunlock(ip, XFS_ILOCK_EXCL);
nres = XFS_DIOSTRAT_SPACE_RES(mp, datablocks);
error = xfs_trans_reserve(tp, nres,
XFS_WRITE_LOG_RES(mp), numrtextents,
XFS_TRANS_PERM_LOG_RES,
XFS_WRITE_LOG_COUNT );
xfs_ilock(ip, XFS_ILOCK_EXCL);
if (error) {
/*
* Ran out of file system space.
* Free the transaction structure.
*/
ASSERT(error == ENOSPC ||
XFS_FORCED_SHUTDOWN(mp));
xfs_trans_cancel(tp, 0);
xfs_iunlock(ip, XFS_ILOCK_EXCL);
break;
}
/*
* quota reservations
*/
if (using_quotas) {
if (xfs_trans_reserve_blkquota(tp, ip,
nres)) {
error = XFS_ERROR(EDQUOT);
xfs_trans_cancel(tp,
XFS_TRANS_RELEASE_LOG_RES);
xfs_iunlock(ip, XFS_ILOCK_EXCL);
break;
}
}
xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
xfs_trans_ihold(tp, ip);
if (offset < ip->i_d.di_size)
bmapi_flag |= XFS_BMAPI_PREALLOC;
/*
* Issue the bmapi() call to do actual file
* space allocation.
*/
CHECK_GRIO_TIMESTAMP(bp, 40);
error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
bmapi_flag, &firstfsb, 0, imapp, &reccount,
&free_list);
CHECK_GRIO_TIMESTAMP(bp, 40);
if (error)
goto error_on_bmapi_transaction;
/*
* Complete the bmapi() allocations transactions.
* The bmapi() unwritten to written changes will
* be committed after the writes are completed.
*/
error = xfs_bmap_finish(&tp, &free_list,
firstfsb, &committed);
if (error)
goto error_on_bmapi_transaction;
xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES,
NULL);
}
xfs_iunlock(ip, XFS_ILOCK_EXCL);
/*
* xfs_bmapi() did not return an error but the
* reccount was zero. This means that a delayed write is
* in progress and it is necessary to call xfs_bmapi() again
* to map the correct portion of the file.
*/
if ((!error) && (reccount == 0)) {
goto retry;
}
imapp = &imaps[0];
unwritten = !(imapp->br_state == XFS_EXT_NORM);
bytes_this_req = XFS_FSB_TO_B(mp, imapp->br_blockcount) -
BBTOB(blk_algn);
ASSERT(bytes_this_req);
offset_this_req = XFS_FSB_TO_B(mp, imapp->br_startoff) +
BBTOB(blk_algn);
/*
* Reduce request size, if it
* is longer than user buffer.
*/
if (bytes_this_req > count) {
bytes_this_req = count;
}
/*
* Check if this is the end of the file.
*/
new_size = offset_this_req + bytes_this_req;
if (new_size >ip->i_d.di_size) {
/*
* File is being extended on a
* write, update the file size if
* someone else didn't make it even
* bigger.
*/
ASSERT((vp->v_flag & VISSWAP) == 0);
xfs_ilock(ip, XFS_ILOCK_EXCL);
if (new_size > ip->i_d.di_size) {
ip->i_d.di_size = offset_this_req +
bytes_this_req;
ip->i_update_core = 1;
ip->i_update_size = 1;
}
xfs_iunlock(ip, XFS_ILOCK_EXCL);
}
/*
* Setup I/O request for this extent.
*/
CHECK_GRIO_TIMESTAMP(bp, 40);
nbp = getphysbuf(bp->b_edev);
CHECK_GRIO_TIMESTAMP(bp, 40);
nbp->b_flags = bp->b_flags;
nbp->b_grio_private = bp->b_grio_private;
/* b_iopri */
nbp->b_error = 0;
nbp->b_target = bp->b_target;
if (rt) {
nbp->b_blkno = XFS_FSB_TO_BB(mp, imapp->br_startblock);
} else {
nbp->b_blkno =
XFS_FSB_TO_DADDR(mp, imapp->br_startblock) +
blk_algn;
}
ASSERT(bytes_this_req);
nbp->b_bcount = bytes_this_req;
nbp->b_un.b_addr = base;
/*
* Issue I/O request.
*/
CHECK_GRIO_TIMESTAMP(nbp, 40);
(void) xfsbdstrat(mp, nbp);
if ((error = geterror(nbp)) == 0) {
/*
* update pointers for next round.
*/
base += bytes_this_req;
offset += bytes_this_req;
count -= bytes_this_req;
blk_algn = 0;
}
/*
* Wait for I/O completion and recover buffer.
*/
biowait(nbp);
nbp->b_flags &= ~B_GR_BUF; /* Why? B_PRV_BUF? */
if (!error)
error = geterror(nbp);
if (!error && !resid) {
resid = nbp->b_resid;
/*
* prevent adding up partial xfers
*/
totxfer += (nbp->b_bcount - resid);
}
nbp->b_flags = 0;
nbp->b_bcount = 0;
nbp->b_un.b_addr = 0;
nbp->b_grio_private = 0; /* b_iopri */
putphysbuf( nbp );
if (error)
break;
if (unwritten) {
/*
* Set up the xfs_bmapi() call to change the
* extent from unwritten to written.
*/
tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
nres = XFS_DIOSTRAT_SPACE_RES(mp, 0);
error = xfs_trans_reserve(tp, nres,
XFS_WRITE_LOG_RES(mp), 0,
XFS_TRANS_PERM_LOG_RES,
XFS_WRITE_LOG_COUNT );
xfs_ilock(ip, XFS_ILOCK_EXCL);
if (error) {
/*
* Ran out of file system space.
* Free the transaction structure.
*/
ASSERT(error == ENOSPC ||
XFS_FORCED_SHUTDOWN(mp));
xfs_trans_cancel(tp, 0);
xfs_iunlock(ip, XFS_ILOCK_EXCL);
break;
}
xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
xfs_trans_ihold(tp, ip);
/*
* Issue the bmapi() call to change the extents
* to written.
*/
reccount = 1;
offset_fsb = XFS_B_TO_FSBT(mp, offset_this_req);
count_fsb = XFS_B_TO_FSB(mp, bytes_this_req);
CHECK_GRIO_TIMESTAMP(bp, 40);
error = xfs_bmapi(tp, ip, offset_fsb, count_fsb,
XFS_BMAPI_WRITE, &firstfsb, 0, imapp,
&reccount, &free_list);
CHECK_GRIO_TIMESTAMP(bp, 40);
if (error)
goto error_on_bmapi_transaction;
/*
* Complete the bmapi() allocations transactions.
* The bmapi() unwritten to written changes will
* be committed after the writes are completed.
*/
error = xfs_bmap_finish(&tp, &free_list,
firstfsb, &committed);
if (error)
goto error_on_bmapi_transaction;
xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES,
NULL);
xfs_iunlock(ip, XFS_ILOCK_EXCL);
}
} /* end of while loop */
/*
* Fill in resid count for original buffer.
* if any of the io's fail, the whole thing fails
*/
if (error) {
totxfer = 0;
}
bp->b_resid = totresid - totxfer;
/*
* Update the inode timestamp.
*/
xfs_ilock(ip, XFS_ILOCK_EXCL);
if ((ip->i_d.di_mode & (ISUID|ISGID)) &&
!cap_able_cred(dp->cr, CAP_FSETID)) {
ip->i_d.di_mode &= ~ISUID;
/*
* Note that we don't have to worry about mandatory
* file locking being disabled here because we only
* clear the ISGID bit if the Group execute bit is
* on, but if it was on then mandatory locking wouldn't
* have been enabled.
*/
if (ip->i_d.di_mode & (IEXEC >> 3))
ip->i_d.di_mode &= ~ISGID;
}
xfs_ichgtime(ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
xfs_iunlock(ip, XFS_ILOCK_EXCL);
error0:
return (error);
error_on_bmapi_transaction:
xfs_bmap_cancel(&free_list);
xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
xfs_iunlock(ip, XFS_ILOCK_EXCL);
goto error0;
}
/*
* xfs_diostrat()
* This routine issues the calls to the disk device strategy routine
* for file system reads and writes made using direct I/O from user
* space. In the case of a write request the I/Os are issued one
* extent at a time. In the case of a read request I/Os for each extent
* involved are issued at once.
*
* RETURNS:
* none
*/
int
xfs_diostrat(
buf_t *bp)
{
xfs_dio_t dio, *diop = &dio;
struct dio_s *dp;
xfs_inode_t *ip;
vnode_t *vp;
bhv_desc_t *bdp;
xfs_mount_t *mp;
off_t offset;
int error;
CHECK_GRIO_TIMESTAMP(bp, 40);
dio.xd_bp = bp;
dio.xd_dp = dp = (struct dio_s *)bp->b_private;
bdp = dp->bdp;
dio.xd_vp = vp = BHV_TO_VNODE(bdp);
dio.xd_ip = ip = XFS_BHVTOI(bdp);
dio.xd_mp = mp = ip->i_mount;
offset = BBTOOFF((off_t)bp->b_blkno);
ASSERT(!(bp->b_flags & B_DONE));
ASSERT(ismrlocked(&ip->i_iolock, MR_ACCESS| MR_UPDATE) != 0);
/*
* Check if the request is on a file system block boundary.
*/
dio.xd_blkalgn = ((offset & mp->m_blockmask) != 0) ?
OFFTOBB(offset & mp->m_blockmask) : 0;
/*
* We're going to access the disk directly.
* Blow anything in the range of the request out of the
* buffer cache first. This isn't perfect because we allow
* simultaneous direct I/O writers and buffered readers, but
* it should be good enough.
*/
if (!(dp->ioflag & IO_IGNCACHE) && VN_CACHED(vp)) {
xfs_inval_cached_pages(ip, offset, bp->b_bcount);
}
/*
* Alignment checks are done in xfs_diordwr().
* Determine if the operation is a read or a write.
*/
if (bp->b_flags & B_READ) {
error = xfs_dio_read(diop);
} else {
error = xfs_dio_write(diop);
}
/*
* Issue completion on the original buffer.
*/
bioerror(bp, error);
biodone(bp);
ASSERT(ismrlocked(&ip->i_iolock, MR_ACCESS| MR_UPDATE) != 0);
return (0);
}
/*
* xfs_diordwr()
* This routine sets up a buf structure to be used to perform
* direct I/O operations to user space. The user specified
* parameters are checked for alignment and size limitations. A buf
* structure is allocated an biophysio() is called.
*
* RETURNS:
* 0 on success
* errno on error
*/
int
xfs_diordwr(
bhv_desc_t *bdp,
uio_t *uiop,
int ioflag,
cred_t *credp,
uint64_t rw)
{
extern zone_t *grio_buf_data_zone;
xfs_inode_t *ip;
vnode_t *vp;
struct dio_s dp;
xfs_mount_t *mp;
uuid_t stream_id;
buf_t *bp;
int error, index;
__int64_t iosize;
extern int scache_linemask;
int guartype = -1;
vp = BHV_TO_VNODE(bdp);
ip = XFS_BHVTOI(bdp);
mp = ip->i_mount;
/*
* Check that the user buffer address is on a secondary cache
* line offset, while file offset and
* request size are both multiples of file system block size.
* This prevents the need for read/modify/write operations.
*
* This enforces the alignment restrictions indicated by
* the F_DIOINFO fcntl call.
*
* We make an exception for swap I/O and trusted clients like
* cachefs. Swap I/O will always be page aligned and all the
* blocks will already be allocated, so we don't need to worry
* about read/modify/write stuff. Cachefs ensures that it only
* reads back data which it has written, so we don't need to
* worry about block zeroing and such.
*/
if (!(vp->v_flag & VISSWAP) && !(ioflag & IO_TRUSTEDDIO) &&
((((long)(uiop->uio_iov->iov_base)) & scache_linemask) ||
(uiop->uio_offset & mp->m_blockmask) ||
(uiop->uio_resid & mp->m_blockmask))) {
/*
* if the user tries to start reading at the
* end of the file, just return 0.
*/
if ((rw & B_READ) &&
(uiop->uio_offset == ip->i_d.di_size)) {
return (0);
}
return XFS_ERROR(EINVAL);
}
/*
* This ASSERT should catch bad addresses being passed in by
* trusted callers.
*/
ASSERT(!(((long)(uiop->uio_iov->iov_base)) & scache_linemask));
/*
* Do maxio check.
*/
if (uiop->uio_resid > ctooff(v.v_maxdmasz - 1)) {
return XFS_ERROR(EINVAL);
}
/*
* Use dio_s structure to pass file/credential
* information to file system strategy routine.
*/
dp.bdp = bdp;
dp.cr = credp;
dp.ioflag = ioflag;
/*
* Allocate local buf structure.
*/
if (ip->i_d.di_flags & XFS_DIFLAG_REALTIME) {
bp = getphysbuf(mp->m_rtdev);
bp->b_target = &mp->m_rtdev_targ;
} else {
bp = getphysbuf(mp->m_dev);
bp->b_target = mp->m_ddev_targp;
}
bp->b_private = &dp;
bp->b_grio_private = NULL; /* b_iopri = 0 */
bp->b_flags &= ~(B_GR_BUF|B_PRV_BUF); /* lo pri queue */
/*
* Check if this is a guaranteed rate I/O
*/
if (ioflag & IO_PRIORITY) {
guartype = grio_io_is_guaranteed(uiop->uio_fp, &stream_id);
/*
* Get priority level if this is a multilevel request.
* The level is stored in b_iopri, except if the request
* is controlled by griostrategy.
*/
if (uiop->uio_fp->vf_flag & FPRIO) {
bp->b_flags |= B_PRV_BUF;
VFILE_GETPRI(uiop->uio_fp, bp->b_iopri);
/*
* Take care of some other thread racing
* and clearing FPRIO.
*/
if (bp->b_iopri == 0)
bp->b_flags &= ~B_PRV_BUF;
}
if (guartype == -1) {
/*
* grio is not configed into kernel, but FPRIO
* is set.
*/
} else if (guartype) {
short prval = bp->b_iopri;
bp->b_flags |= B_GR_BUF;
ASSERT((bp->b_grio_private == NULL) ||
(bp->b_flags & B_PRV_BUF));
bp->b_grio_private =
kmem_zone_alloc(grio_buf_data_zone, KM_SLEEP);
ASSERT(BUF_GRIO_PRIVATE(bp));
COPY_STREAM_ID(stream_id,BUF_GRIO_PRIVATE(bp)->grio_id);
SET_GRIO_IOPRI(bp, prval);
iosize = uiop->uio_iov[0].iov_len;
index = grio_monitor_io_start(&stream_id, iosize);
INIT_GRIO_TIMESTAMP(bp);
} else {
/*
* FPRIORITY|FPRIO was set when we looked,
* but FPRIORITY is not set anymore.
*/
}
}
/*
* Perform I/O operation.
*/
error = biophysio(xfs_diostrat, bp, bp->b_edev, rw,
(daddr_t)OFFTOBB(uiop->uio_offset), uiop);
/*
* Free local buf structure.
*/
if (ioflag & IO_PRIORITY) {
bp->b_flags &= ~(B_PRV_BUF|B_GR_BUF);
if (guartype > 0) {
grio_monitor_io_end(&stream_id, index);
#ifdef GRIO_DEBUG
CHECK_GRIO_TIMESTAMP(bp, 400);
#endif
ASSERT(BUF_GRIO_PRIVATE(bp));
kmem_zone_free(grio_buf_data_zone, BUF_GRIO_PRIVATE(bp));
}
bp->b_grio_private = NULL;
}
ASSERT((bp->b_flags & B_MAPPED) == 0);
bp->b_flags = 0;
bp->b_un.b_addr = 0;
putphysbuf(bp);
return (error);
}
lock_t xfs_refcache_lock;
xfs_inode_t **xfs_refcache;
int xfs_refcache_size;
int xfs_refcache_index;
int xfs_refcache_busy;
int xfs_refcache_count;
/*
* Insert the given inode into the reference cache.
*/
void
xfs_refcache_insert(
xfs_inode_t *ip)
{
int s;
vnode_t *vp;
xfs_inode_t *release_ip;
xfs_inode_t **refcache;
ASSERT(ismrlocked(&(ip->i_iolock), MR_UPDATE));
/*
* If an unmount is busy blowing entries out of the cache,
* then don't bother.
*/
if (xfs_refcache_busy) {
return;
}
/*
* The inode is already in the refcache, so don't bother
* with it.
*/
if (ip->i_refcache != NULL) {
return;
}
vp = XFS_ITOV(ip);
ASSERT(vp->v_count > 0);
VN_HOLD(vp);
/*
* We allocate the reference cache on use so that we don't
* waste the memory on systems not being used as NFS servers.
*/
if (xfs_refcache == NULL) {
refcache = (xfs_inode_t **)kmem_zalloc(xfs_refcache_size *
sizeof(xfs_inode_t *),
KM_SLEEP);
} else {
refcache = NULL;
}
s = mp_mutex_spinlock(&xfs_refcache_lock);
/*
* If we allocated memory for the refcache above and it still
* needs it, then use the memory we allocated. Otherwise we'll
* free the memory below.
*/
if (refcache != NULL) {
if (xfs_refcache == NULL) {
xfs_refcache = refcache;
refcache = NULL;
}
}
/*
* If an unmount is busy clearing out the cache, don't add new
* entries to it.
*/
if ((xfs_refcache_busy) || (vp->v_vfsp->vfs_flag & VFS_OFFLINE)) {
mp_mutex_spinunlock(&xfs_refcache_lock, s);
VN_RELE(vp);
/*
* If we allocated memory for the refcache above but someone
* else beat us to using it, then free the memory now.
*/
if (refcache != NULL) {
kmem_free(refcache,
xfs_refcache_size * sizeof(xfs_inode_t *));
}
return;
}
release_ip = xfs_refcache[xfs_refcache_index];
if (release_ip != NULL) {
release_ip->i_refcache = NULL;
xfs_refcache_count--;
ASSERT(xfs_refcache_count >= 0);
}
xfs_refcache[xfs_refcache_index] = ip;
ASSERT(ip->i_refcache == NULL);
ip->i_refcache = &(xfs_refcache[xfs_refcache_index]);
xfs_refcache_count++;
ASSERT(xfs_refcache_count <= xfs_refcache_size);
xfs_refcache_index++;
if (xfs_refcache_index == xfs_refcache_size) {
xfs_refcache_index = 0;
}
mp_mutex_spinunlock(&xfs_refcache_lock, s);
/*
* Save the pointer to the inode to be released so that we can
* VN_RELE it once we've dropped our inode locks in xfs_rwunlock().
* The pointer may be NULL, but that's OK.
*/
ip->i_release = release_ip;
/*
* If we allocated memory for the refcache above but someone
* else beat us to using it, then free the memory now.
*/
if (refcache != NULL) {
kmem_free(refcache,
xfs_refcache_size * sizeof(xfs_inode_t *));
}
return;
}
/*
* If the given inode is in the reference cache, purge its entry and
* release the reference on the vnode.
*/
void
xfs_refcache_purge_ip(
xfs_inode_t *ip)
{
int s;
vnode_t *vp;
/*
* If we're not pointing to our entry in the cache, then
* we must not be in the cache.
*/
if (ip->i_refcache == NULL) {
return;
}
s = mp_mutex_spinlock(&xfs_refcache_lock);
if (ip->i_refcache == NULL) {
mp_mutex_spinunlock(&xfs_refcache_lock, s);
return;
}
/*
* Clear both our pointer to the cache entry and its pointer
* back to us.
*/
ASSERT(*(ip->i_refcache) == ip);
*(ip->i_refcache) = NULL;
ip->i_refcache = NULL;
xfs_refcache_count--;
ASSERT(xfs_refcache_count >= 0);
mp_mutex_spinunlock(&xfs_refcache_lock, s);
vp = XFS_ITOV(ip);
ASSERT(vp->v_count > 1);
VN_RELE(vp);
return;
}
/*
* This is called from the XFS unmount code to purge all entries for the
* given mount from the cache. It uses the refcache busy counter to
* make sure that new entries are not added to the cache as we purge them.
*/
void
xfs_refcache_purge_mp(
xfs_mount_t *mp)
{
int s;
vnode_t *vp;
int i;
xfs_inode_t *ip;
if (xfs_refcache == NULL) {
return;
}
s = mp_mutex_spinlock(&xfs_refcache_lock);
/*
* Bumping the busy counter keeps new entries from being added
* to the cache. We use a counter since multiple unmounts could
* be in here simultaneously.
*/
xfs_refcache_busy++;
for (i = 0; i < xfs_refcache_size; i++) {
ip = xfs_refcache[i];
if ((ip != NULL) && (ip->i_mount == mp)) {
xfs_refcache[i] = NULL;
ip->i_refcache = NULL;
xfs_refcache_count--;
ASSERT(xfs_refcache_count >= 0);
mp_mutex_spinunlock(&xfs_refcache_lock, s);
vp = XFS_ITOV(ip);
VN_RELE(vp);
s = mp_mutex_spinlock(&xfs_refcache_lock);
} else {
/*
* Make sure the don't hold the lock for too long.
*/
if ((i & 15) == 0) {
mp_mutex_spinunlock(&xfs_refcache_lock, s);
s = mp_mutex_spinlock(&xfs_refcache_lock);
}
}
}
xfs_refcache_busy--;
ASSERT(xfs_refcache_busy >= 0);
mp_mutex_spinunlock(&xfs_refcache_lock, s);
}
/*
* This is called from the XFS sync code to ensure that the refcache
* is emptied out over time. We purge a small number of entries with
* each call.
*/
void
xfs_refcache_purge_some(void)
{
int s;
int i;
xfs_inode_t *ip;
int iplist_index;
#define XFS_REFCACHE_PURGE_COUNT 10
xfs_inode_t *iplist[XFS_REFCACHE_PURGE_COUNT];
if ((xfs_refcache == NULL) || (xfs_refcache_count == 0)) {
return;
}
iplist_index = 0;
s = mp_mutex_spinlock(&xfs_refcache_lock);
/*
* Store any inodes we find in the next several entries
* into the iplist array to be released after dropping
* the spinlock. We always start looking from the currently
* oldest place in the cache. We move the refcache index
* forward as we go so that we are sure to eventually clear
* out the entire cache when the system goes idle.
*/
for (i = 0; i < XFS_REFCACHE_PURGE_COUNT; i++) {
ip = xfs_refcache[xfs_refcache_index];
if (ip != NULL) {
xfs_refcache[xfs_refcache_index] = NULL;
ip->i_refcache = NULL;
xfs_refcache_count--;
ASSERT(xfs_refcache_count >= 0);
iplist[iplist_index] = ip;
iplist_index++;
}
xfs_refcache_index++;
if (xfs_refcache_index == xfs_refcache_size) {
xfs_refcache_index = 0;
}
}
mp_mutex_spinunlock(&xfs_refcache_lock, s);
/*
* Now drop the inodes we collected.
*/
for (i = 0; i < iplist_index; i++) {
VN_RELE(XFS_ITOV(iplist[i]));
}
}
![[BACK]](/icons/back.gif){kind=icon}
Return to xfs_rw.c CVS log ![[TXT]](/icons/text.gif){kind=icon}
![[DIR]](/icons/dir.gif){kind=icon}
Up to [Development] / xfs-linux