Replace the two users of the ->data_unused_p dir ops method with a
direct calculation using ->data_entry_offset, and clean them up a bit.
xfs_dir2_sf_to_block already had an offset variable containing the
value of ->data_entry_offset, which we are now reusing to make it
clear that the initial freespace entry is at the same place that
we later fill in the 1 entry, and in xfs_dir3_data_init the function
is cleaned up a bit to keep the initialization of fields of a given
structure close to each other, and to avoid a local variable.
xfs: remove the ->data_dot_entry_p and ->data_dotdot_entry_p methods
The only user of the ->data_dot_entry_p and ->data_dotdot_entry_p
methods is the xfs_dir2_sf_to_block function that builds block format
directorys from a short form directory. It already uses pointer
arithmetics with a offset variable to do so for the real entries in
the directory, so switch the generation of the . and .. entries to
the same scheme, and clean up some of the later pointer arithmetics
to use bp->b_addr directly as well and avoid some casts.
xfs: remove the data_dotdot_offset field in struct xfs_dir_ops
The data_dotdot_offset value is always equal to data_entry_offset plus
the fixed size of the "." entry. Right now calculating that fixed size
requires an indirect call, but by the end of this series it will be
an inline function that can be constant folded.
xfs: devirtualize ->sf_get_ftype and ->sf_put_ftype
Replace the ->sf_get_ftype and ->sf_put_ftype dir ops methods with
directly called xfs_dir2_sf_get_ftype and xfs_dir2_sf_put_ftype helpers
that takes care of the differences between the directory format with and
without the file type field.
Replace the ->sf_get_ino and ->sf_put_ino dir ops methods with directly
called xfs_dir2_sf_get_ino and xfs_dir2_sf_put_ino helpers that take care
of the difference between the directory format with and without the file
type field. Also move xfs_dir2_sf_get_parent_ino and
xfs_dir2_sf_put_parent_ino to xfs_dir2_sf.c with the rest of the
low-level short form entry handling and use XFS_MAXINUMBER istead of
opencoded constants.
Now that the max bests value is in struct xfs_da_geometry both instances
of ->db_to_fdb and ->db_to_fdindex are identical. Replace them with
local xfs_dir2_db_to_fdb and xfs_dir2_db_to_fdindex functions in
xfs_dir2_node.c.
xfs: add a bests pointer to struct xfs_dir3_icfree_hdr
All but two callers of the ->free_bests_p dir operation already have a
struct xfs_dir3_icfree_hdr from a previous call to
xfs_dir2_free_hdr_from_disk at hand. Add a pointer to the bests to
struct xfs_dir3_icfree_hdr to clean up this pattern. To optimize this
pattern, pass the struct xfs_dir3_icfree_hdr to xfs_dir2_free_log_bests
instead of recalculating the pointer there.
xfs: make the xfs_dir3_icfree_hdr available to xfs_dir2_node_addname_int
Return the xfs_dir3_icfree_hdr used by the helpers called from
xfs_dir2_node_addname_int to the main function to prepare for the
next round of changes where we'll use the ichdr in xfs_dir3_icfree_hdr
to avoid extra operations to find the bests pointers.
Replace the ->free_hdr_to_disk dir ops method with a directly called
xfs_dir2_free_hdr_to_disk helper that takes care of the differences
between the v4 and v5 on-disk format.
Replace the ->free_hdr_from_disk dir ops method with a directly called
xfs_dir_free_hdr_from_disk helper that takes care of the differences
between the v4 and v5 on-disk format.
xfs: add an entries pointer to struct xfs_dir3_icleaf_hdr
All callers of the ->node_tree_p dir operation already have a struct
xfs_dir3_icleaf_hdr from a previous call to xfs_da_leaf_hdr_from_disk at
hand, or just need slight changes to the calling conventions to do so.
Add a pointer to the entries to struct xfs_dir3_icleaf_hdr to clean up
this pattern. To make this possible the xfs_dir3_leaf_log_ents function
grow a new argument to pass the xfs_dir3_icleaf_hdr that call callers
already have, and xfs_dir2_leaf_lookup_int returns the
xfs_dir3_icleaf_hdr to the callers so that they can later use it.
Replace the ->leaf_hdr_to_disk dir ops method with a directly called
xfs_dir_leaf_hdr_to_disk helper that takes care of the differences
between the v4 and v5 on-disk format.
Replace the ->leaf_hdr_from_disk dir ops method with a directly called
xfs_dir2_leaf_hdr_from_disk helper that takes care of the differences
between the v4 and v5 on-disk format.
xfs: add a btree entries pointer to struct xfs_da3_icnode_hdr
All but two callers of the ->node_tree_p dir operation already have a
xfs_da3_icnode_hdr from a previous call to xfs_da3_node_hdr_from_disk at
hand. Add a pointer to the btree entries to struct xfs_da3_icnode_hdr
to clean up this pattern. The two remaining callers now expand the
whole header as well, but that isn't very expensive and not in a super
hot path anyway.
Break up xchk_da_btree_entry and handle looking up leaf node entries
in the attr / dir callbacks, so that only the generic node handling
is left in the common core code. Note that the checks for the crc
enabled blocks are removed, as the scrubbing code already remaps the
magic numbers earlier.
Darrick J. Wong [Thu, 7 Nov 2019 23:05:21 +0000 (15:05 -0800)]
xfs: refactor "does this fork map blocks" predicate
Replace the open-coded checks for whether or not an inode fork maps
blocks with a macro that will implant the code for us. This helps us
declutter the bmap code a bit.
Note that I had to use a macro instead of a static inline function
because of C header dependency problems between xfs_inode.h and
xfs_inode_fork.h.
Conversion was performed with the following Coccinelle script:
Darrick J. Wong [Wed, 6 Nov 2019 17:11:23 +0000 (09:11 -0800)]
xfs: range check ri_cnt when recovering log items
Range check the region counter when we're reassembling regions from log
items during log recovery. In the old days ASSERT would halt the
kernel, but this isn't true any more so we have to make an explicit
error return.
Optimize the setting of full words of bits in xfs_buf_item_log_segment.
The optimization is purely within the bug triage process. No functional
changes.
Darrick J. Wong [Wed, 6 Nov 2019 16:53:54 +0000 (08:53 -0800)]
xfs: null out bma->prev if no previous extent
Coverity complains that we don't check the return value of
xfs_iext_peek_prev_extent like we do nearly all of the time. If there
is no previous extent then just null out bma->prev like we do elsewhere
in the bmap code.
Darrick J. Wong [Thu, 7 Nov 2019 01:19:33 +0000 (17:19 -0800)]
xfs: fix missing header includes
Some of the xfs source files are missing header includes, so add them
back. Sparse complains about non-static functions that don't have a
forward declaration anywhere.
Darrick J. Wong [Tue, 5 Nov 2019 23:33:57 +0000 (15:33 -0800)]
xfs: periodically yield scrub threads to the scheduler
Christoph Hellwig complained about the following soft lockup warning
when running scrub after generic/175 when preemption is disabled and
slub debugging is enabled:
If preemption is disabled, all metadata buffers needed to perform the
scrub are already in memory, and there are a lot of records to check,
it's possible that the scrub thread will run for an extended period of
time without sleeping for IO or any other reason. Then the watchdog
timer or the RCU stall timeout can trigger, producing the backtrace
above.
To fix this problem, call cond_resched() from the scrub thread so that
we back out to the scheduler whenever necessary.
Colin Ian King [Wed, 6 Nov 2019 16:07:46 +0000 (08:07 -0800)]
xfs: remove redundant assignment to variable error
Variable error is being initialized with a value that is never read
and is being re-assigned a couple of statements later on. The
assignment is redundant and hence can be removed.
Darrick J. Wong [Tue, 5 Nov 2019 23:33:56 +0000 (15:33 -0800)]
xfs: add missing early termination checks to record scrubbing functions
Scrubbing directories, quotas, and fs counters all involve iterating
some collection of metadata items. The per-item scrub functions for
these three are missing some of the components they need to be able to
check for a fatal signal and terminate early.
Per-item scrub functions need to call xchk_should_terminate to look for
fatal signals, and they need to check the scrub context's corruption
flag because there's no point in continuing a scan once we've decided
the data structure is bad. Add both of these where missing.
Darrick J. Wong [Sat, 2 Nov 2019 16:41:18 +0000 (09:41 -0700)]
xfs: decrease indenting problems in xfs_dabuf_map
Refactor the code that complains when a dir/attr mapping doesn't exist
but the caller requires a mapping. This small restructuring helps us to
reduce the indenting level.
Grouping the options parsing and mount handling functions above the
struct fs_context_operations but below the struct super_operations
should improve (some) the grouping of the super operations while also
improving the grouping of the options parsing and mount handling code.
Lastly move xfs_fc_parse_param() and related functions down to above
xfs_fc_get_tree() and it's related functions.
But leave the options enum, struct fs_parameter_spec and the struct
fs_parameter_description declarations at the top since that's the
logical place for them.
This is a straight code move, there aren't any functional changes.
Grouping the options parsing and mount handling functions above the
struct fs_context_operations but below the struct super_operations
should improve (some) the grouping of the super operations while also
improving the grouping of the options parsing and mount handling code.
Now move xfs_fc_get_tree() and friends, also take the oppertunity to
change STATIC to static for the xfs_fs_put_super() function.
This is a straight code move, there aren't any functional changes.
Grouping the options parsing and mount handling functions above the
struct fs_context_operations but below the struct super_operations
should improve (some) the grouping of the super operations while also
improving the grouping of the options parsing and mount handling code.
Start by moving xfs_fc_reconfigure() and friends.
This is a straight code move, there aren't any functional changes.
Ian Kent [Mon, 4 Nov 2019 21:58:46 +0000 (13:58 -0800)]
xfs: switch to use the new mount-api
Define the struct fs_parameter_spec table that's used by the new
mount-api for options parsing.
Create the various fs context operations methods and define the
fs_context_operations struct.
Create the fs context initialization method and update the struct
file_system_type to utilize it. The initialization function is
responsible for working storage initialization, allocation and
initialization of file system private information storage and for
setting the operations in the fs context.
Also set struct file_system_type .parameters to the newly defined
struct fs_parameter_spec options parsing table for use by the fs
context methods and remove unused code.
[darrick: add a comment pointing out the one place where mp->m_super is
null]
Ian Kent [Mon, 4 Nov 2019 21:58:46 +0000 (13:58 -0800)]
xfs: dont set sb in xfs_mount_alloc()
When changing to use the new mount api the super block won't be
available when the xfs_mount struct is allocated so move setting the
super block in xfs_mount to xfs_fs_fill_super().
Ian Kent [Mon, 4 Nov 2019 21:58:44 +0000 (13:58 -0800)]
xfs: refactor xfs_parseags()
Refactor xfs_parseags(), move the entire token case block to a separate
function in an attempt to highlight the code that actually changes in
converting to use the new mount api.
Also change the break in the switch to a return in the factored out
xfs_fc_parse_param() function.
Ian Kent [Mon, 4 Nov 2019 21:58:44 +0000 (13:58 -0800)]
xfs: avoid redundant checks when options is empty
When options passed to xfs_parseargs() is NULL the checks performed
after taking the branch are made with the initial values of dsunit,
dswidth and iosizelog. But all the checks do nothing in this case
so return immediately instead.
Ian Kent [Mon, 4 Nov 2019 21:58:43 +0000 (13:58 -0800)]
xfs: refactor suffix_kstrtoint()
The mount-api doesn't have a "human unit" parse type yet so the options
that have values like "10k" etc. still need to be converted by the fs.
But the value comes to the fs as a string (not a substring_t type) so
there's a need to change the conversion function to take a character
string instead.
When xfs is switched to use the new mount-api match_kstrtoint() will no
longer be used and will be removed.
Ian Kent [Mon, 4 Nov 2019 21:58:43 +0000 (13:58 -0800)]
xfs: add xfs_remount_ro() helper
Factor the remount read only code into a helper to simplify the
subsequent change from the super block method .remount_fs to the
mount-api fs_context_operations method .reconfigure.
Ian Kent [Mon, 4 Nov 2019 21:58:42 +0000 (13:58 -0800)]
xfs: add xfs_remount_rw() helper
Factor the remount read write code into a helper to simplify the
subsequent change from the super block method .remount_fs to the
mount-api fs_context_operations method .reconfigure.
Ian Kent [Mon, 4 Nov 2019 21:58:42 +0000 (13:58 -0800)]
xfs: merge freeing of mp names and mp
In all cases when struct xfs_mount (mp) fields m_rtname and m_logname
are freed mp is also freed, so merge these into a single function
xfs_mount_free()
Ian Kent [Mon, 4 Nov 2019 21:58:41 +0000 (13:58 -0800)]
xfs: use kmem functions for struct xfs_mount
The remount function uses the kmem functions for allocating and freeing
struct xfs_mount, for consistency use the kmem functions everwhere for
struct xfs_mount.
Ian Kent [Mon, 4 Nov 2019 21:58:41 +0000 (13:58 -0800)]
xfs: dont use XFS_IS_QUOTA_RUNNING() for option check
When CONFIG_XFS_QUOTA is not defined any quota option is invalid.
Using the macro XFS_IS_QUOTA_RUNNING() as a check if any quota option
has been given is a little misleading so use a simple m_qflags != 0
check to make the intended use more explicit.
Also change to use the IS_ENABLED() macro for the kernel config check.
Darrick J. Wong [Sat, 2 Nov 2019 16:40:36 +0000 (09:40 -0700)]
xfs: constify the buffer pointer arguments to error functions
Some of the xfs error message functions take a pointer to a buffer that
will be dumped to the system log. The logging functions don't change
the contents, so constify all the parameters. This enables the next
patch to ensure that we log bad metadata when we encounter it.
Darrick J. Wong [Sat, 2 Nov 2019 16:38:08 +0000 (09:38 -0700)]
xfs: relax shortform directory size checks
Each of the four functions that operate on shortform directories checks
that the directory's di_size is at least as large as the shortform
directory header. This is now checked by the inode fork verifiers
(di_size is used to allocate if_bytes, and if_bytes is checked against
the header structure size) so we can turn these checks into ASSERTions.
Always set XFS_ALLOC_USERDATA for data fork allocations, and check it
in xfs_alloc_is_userdata instead of the current obsfucated check.
Also remove the xfs_alloc_is_userdata and xfs_alloc_allow_busy_reuse
helpers to make the code a little easier to understand.
Move the extent zeroing case there for the XFS_BMAPI_ZERO flag outside
the low-level allocator and into xfs_bmapi_allocate, where is still
is in transaction context, but outside the very lowlevel code where
it doesn't belong.
Avoid duplicate userdata and data fork checks by restructuring the code
so we only have a helper for userdata allocations that combines these
checks in a straight foward way. That also helps to obsoletes the
comments explaining what the code does as it is now clearly obvious.
Move the EOF alignment and checking for the next allocated extent into
the callers to avoid the need to pass the byte based offset and count
as well as looking at the incoming imap. The added benefit is that
the caller can unlock the incoming ilock and the function doesn't have
funny unbalanced locking contexts.
By open coding xfs_bmap_last_extent instead of calling it through a
double indirection we don't need to handle an error return that
can't happen given that we are guaranteed to have the extent list in
memory already. Also simplify the calling conventions a little and
move the extent list assert from the only caller into the function.
Dave Chinner [Tue, 29 Oct 2019 20:04:32 +0000 (13:04 -0700)]
xfs: properly serialise fallocate against AIO+DIO
AIO+DIO can extend the file size on IO completion, and it holds
no inode locks while the IO is in flight. Therefore, a race
condition exists in file size updates if we do something like this:
aio-thread fallocate-thread
lock inode
submit IO beyond inode->i_size
unlock inode
.....
lock inode
break layouts
if (off + len > inode->i_size)
new_size = off + len
.....
inode_dio_wait()
<blocks>
.....
completes
inode->i_size updated
inode_dio_done()
....
<wakes>
<does stuff no long beyond EOF>
if (new_size)
xfs_vn_setattr(inode, new_size)
Yup, that attempt to extend the file size in the fallocate code
turns into a truncate - it removes the whatever the aio write
allocated and put to disk, and reduced the inode size back down to
where the fallocate operation ends.
Fundamentally, xfs_file_fallocate() not compatible with racing
AIO+DIO completions, so we need to move the inode_dio_wait() call
up to where the lock the inode and break the layouts.
Secondly, storing the inode size and then using it unchecked without
holding the ILOCK is not safe; we can only do such a thing if we've
locked out and drained all IO and other modification operations,
which we don't do initially in xfs_file_fallocate.
It should be noted that some of the fallocate operations are
compound operations - they are made up of multiple manipulations
that may zero data, and so we may need to flush and invalidate the
file multiple times during an operation. However, we only need to
lock out IO and other space manipulation operations once, as that
lockout is maintained until the entire fallocate operation has been
completed.
inode64 is the only value remaining in the unset array. Special case
the inode32/64 options with an explicit seq_printf that prints either
inode32 or inode64, and remove the unset array.
Rework xfs_parseargs to fill out the default value and then parse the
option directly into the mount structure, similar to what we do for
other updates, and open code the now trivial updates based on on the
on-disk superblock directly into xfs_mountfs.
Note that this change rejects the allocsize=0 mount option that has been
documented as invalid for a long time instead of just ignoring it.
xfs: remove the m_readio_* fields in struct xfs_mount
m_readio_blocks is entirely unused, and m_readio_blocks is only used in
xfs_stat_blksize in a max statements that is a no-op as it always has
the same value as m_writeio_log.
There is no real need for the local variables here - either they
are applied to the mount structure, or if the noalign mount option
is set the mount will fail entirely if either is set. Removing
them helps cleaning up the mount API conversion.
Darrick J. Wong [Mon, 28 Oct 2019 23:12:35 +0000 (16:12 -0700)]
xfs: refactor xfs_iread_extents to use xfs_btree_visit_blocks
xfs_iread_extents open-codes everything in xfs_btree_visit_blocks, so
refactor the btree helper to be able to iterate only the records on
level 0, then port iread_extents to use it.
Darrick J. Wong [Mon, 28 Oct 2019 23:12:35 +0000 (16:12 -0700)]
xfs: refactor xfs_bmap_count_blocks using newer btree helpers
Currently, this function open-codes walking a bmbt to count the extents
and blocks in use by a particular inode fork. Since we now have a
function to tally extent records from the incore extent tree and a btree
helper to count every block in a btree, replace all that with calls to
the helpers.
If we always have to write out of place preallocating blocks is
pointless. We already check for this in the normal falloc path, but
the check was missig in the legacy ALLOCSP path.
projects / linux.git / log