aha/fs/ubifs/budget.c
Artem Bityutskiy 547000da64 UBIFS: improve budgeting checks
Budgeting is a crucial UBIFS subsystem - add more assertions
to improve requests checking. This is not compiled in when
UBIFS debugging is disabled.

Signed-off-by: Artem Bityutskiy <Artem.Bityutskiy@nokia.com>
2008-08-13 11:34:27 +03:00

744 lines
23 KiB
C

/*
* This file is part of UBIFS.
*
* Copyright (C) 2006-2008 Nokia Corporation.
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 as published by
* the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* You should have received a copy of the GNU General Public License along with
* this program; if not, write to the Free Software Foundation, Inc., 51
* Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
*
* Authors: Adrian Hunter
* Artem Bityutskiy (Битюцкий Артём)
*/
/*
* This file implements the budgeting sub-system which is responsible for UBIFS
* space management.
*
* Factors such as compression, wasted space at the ends of LEBs, space in other
* journal heads, the effect of updates on the index, and so on, make it
* impossible to accurately predict the amount of space needed. Consequently
* approximations are used.
*/
#include "ubifs.h"
#include <linux/writeback.h>
#include <asm/div64.h>
/*
* When pessimistic budget calculations say that there is no enough space,
* UBIFS starts writing back dirty inodes and pages, doing garbage collection,
* or committing. The below constants define maximum number of times UBIFS
* repeats the operations.
*/
#define MAX_SHRINK_RETRIES 8
#define MAX_GC_RETRIES 4
#define MAX_CMT_RETRIES 2
#define MAX_NOSPC_RETRIES 1
/*
* The below constant defines amount of dirty pages which should be written
* back at when trying to shrink the liability.
*/
#define NR_TO_WRITE 16
/**
* struct retries_info - information about re-tries while making free space.
* @prev_liability: previous liability
* @shrink_cnt: how many times the liability was shrinked
* @shrink_retries: count of liability shrink re-tries (increased when
* liability does not shrink)
* @try_gc: GC should be tried first
* @gc_retries: how many times GC was run
* @cmt_retries: how many times commit has been done
* @nospc_retries: how many times GC returned %-ENOSPC
*
* Since we consider budgeting to be the fast-path, and this structure has to
* be allocated on stack and zeroed out, we make it smaller using bit-fields.
*/
struct retries_info {
long long prev_liability;
unsigned int shrink_cnt;
unsigned int shrink_retries:5;
unsigned int try_gc:1;
unsigned int gc_retries:4;
unsigned int cmt_retries:3;
unsigned int nospc_retries:1;
};
/**
* shrink_liability - write-back some dirty pages/inodes.
* @c: UBIFS file-system description object
* @nr_to_write: how many dirty pages to write-back
*
* This function shrinks UBIFS liability by means of writing back some amount
* of dirty inodes and their pages. Returns the amount of pages which were
* written back. The returned value does not include dirty inodes which were
* synchronized.
*
* Note, this function synchronizes even VFS inodes which are locked
* (@i_mutex) by the caller of the budgeting function, because write-back does
* not touch @i_mutex.
*/
static int shrink_liability(struct ubifs_info *c, int nr_to_write)
{
int nr_written;
struct writeback_control wbc = {
.sync_mode = WB_SYNC_NONE,
.range_end = LLONG_MAX,
.nr_to_write = nr_to_write,
};
generic_sync_sb_inodes(c->vfs_sb, &wbc);
nr_written = nr_to_write - wbc.nr_to_write;
if (!nr_written) {
/*
* Re-try again but wait on pages/inodes which are being
* written-back concurrently (e.g., by pdflush).
*/
memset(&wbc, 0, sizeof(struct writeback_control));
wbc.sync_mode = WB_SYNC_ALL;
wbc.range_end = LLONG_MAX;
wbc.nr_to_write = nr_to_write;
generic_sync_sb_inodes(c->vfs_sb, &wbc);
nr_written = nr_to_write - wbc.nr_to_write;
}
dbg_budg("%d pages were written back", nr_written);
return nr_written;
}
/**
* run_gc - run garbage collector.
* @c: UBIFS file-system description object
*
* This function runs garbage collector to make some more free space. Returns
* zero if a free LEB has been produced, %-EAGAIN if commit is required, and a
* negative error code in case of failure.
*/
static int run_gc(struct ubifs_info *c)
{
int err, lnum;
/* Make some free space by garbage-collecting dirty space */
down_read(&c->commit_sem);
lnum = ubifs_garbage_collect(c, 1);
up_read(&c->commit_sem);
if (lnum < 0)
return lnum;
/* GC freed one LEB, return it to lprops */
dbg_budg("GC freed LEB %d", lnum);
err = ubifs_return_leb(c, lnum);
if (err)
return err;
return 0;
}
/**
* make_free_space - make more free space on the file-system.
* @c: UBIFS file-system description object
* @ri: information about previous invocations of this function
*
* This function is called when an operation cannot be budgeted because there
* is supposedly no free space. But in most cases there is some free space:
* o budgeting is pessimistic, so it always budgets more then it is actually
* needed, so shrinking the liability is one way to make free space - the
* cached data will take less space then it was budgeted for;
* o GC may turn some dark space into free space (budgeting treats dark space
* as not available);
* o commit may free some LEB, i.e., turn freeable LEBs into free LEBs.
*
* So this function tries to do the above. Returns %-EAGAIN if some free space
* was presumably made and the caller has to re-try budgeting the operation.
* Returns %-ENOSPC if it couldn't do more free space, and other negative error
* codes on failures.
*/
static int make_free_space(struct ubifs_info *c, struct retries_info *ri)
{
int err;
/*
* If we have some dirty pages and inodes (liability), try to write
* them back unless this was tried too many times without effect
* already.
*/
if (ri->shrink_retries < MAX_SHRINK_RETRIES && !ri->try_gc) {
long long liability;
spin_lock(&c->space_lock);
liability = c->budg_idx_growth + c->budg_data_growth +
c->budg_dd_growth;
spin_unlock(&c->space_lock);
if (ri->prev_liability >= liability) {
/* Liability does not shrink, next time try GC then */
ri->shrink_retries += 1;
if (ri->gc_retries < MAX_GC_RETRIES)
ri->try_gc = 1;
dbg_budg("liability did not shrink: retries %d of %d",
ri->shrink_retries, MAX_SHRINK_RETRIES);
}
dbg_budg("force write-back (count %d)", ri->shrink_cnt);
shrink_liability(c, NR_TO_WRITE + ri->shrink_cnt);
ri->prev_liability = liability;
ri->shrink_cnt += 1;
return -EAGAIN;
}
/*
* Try to run garbage collector unless it was already tried too many
* times.
*/
if (ri->gc_retries < MAX_GC_RETRIES) {
ri->gc_retries += 1;
dbg_budg("run GC, retries %d of %d",
ri->gc_retries, MAX_GC_RETRIES);
ri->try_gc = 0;
err = run_gc(c);
if (!err)
return -EAGAIN;
if (err == -EAGAIN) {
dbg_budg("GC asked to commit");
err = ubifs_run_commit(c);
if (err)
return err;
return -EAGAIN;
}
if (err != -ENOSPC)
return err;
/*
* GC could not make any progress. If this is the first time,
* then it makes sense to try to commit, because it might make
* some dirty space.
*/
dbg_budg("GC returned -ENOSPC, retries %d",
ri->nospc_retries);
if (ri->nospc_retries >= MAX_NOSPC_RETRIES)
return err;
ri->nospc_retries += 1;
}
/* Neither GC nor write-back helped, try to commit */
if (ri->cmt_retries < MAX_CMT_RETRIES) {
ri->cmt_retries += 1;
dbg_budg("run commit, retries %d of %d",
ri->cmt_retries, MAX_CMT_RETRIES);
err = ubifs_run_commit(c);
if (err)
return err;
return -EAGAIN;
}
return -ENOSPC;
}
/**
* ubifs_calc_min_idx_lebs - calculate amount of eraseblocks for the index.
* @c: UBIFS file-system description object
*
* This function calculates and returns the number of eraseblocks which should
* be kept for index usage.
*/
int ubifs_calc_min_idx_lebs(struct ubifs_info *c)
{
int ret;
uint64_t idx_size;
idx_size = c->old_idx_sz + c->budg_idx_growth + c->budg_uncommitted_idx;
/* And make sure we have twice the index size of space reserved */
idx_size <<= 1;
/*
* We do not maintain 'old_idx_size' as 'old_idx_lebs'/'old_idx_bytes'
* pair, nor similarly the two variables for the new index size, so we
* have to do this costly 64-bit division on fast-path.
*/
if (do_div(idx_size, c->leb_size - c->max_idx_node_sz))
ret = idx_size + 1;
else
ret = idx_size;
/*
* The index head is not available for the in-the-gaps method, so add an
* extra LEB to compensate.
*/
ret += 1;
/*
* At present the index needs at least 2 LEBs: one for the index head
* and one for in-the-gaps method (which currently does not cater for
* the index head and so excludes it from consideration).
*/
if (ret < 2)
ret = 2;
return ret;
}
/**
* ubifs_calc_available - calculate available FS space.
* @c: UBIFS file-system description object
* @min_idx_lebs: minimum number of LEBs reserved for the index
*
* This function calculates and returns amount of FS space available for use.
*/
long long ubifs_calc_available(const struct ubifs_info *c, int min_idx_lebs)
{
int subtract_lebs;
long long available;
/*
* Force the amount available to the total size reported if the used
* space is zero.
*/
if (c->lst.total_used <= UBIFS_INO_NODE_SZ &&
c->budg_data_growth + c->budg_dd_growth == 0) {
/* Do the same calculation as for c->block_cnt */
available = c->main_lebs - 2;
available *= c->leb_size - c->dark_wm;
return available;
}
available = c->main_bytes - c->lst.total_used;
/*
* Now 'available' contains theoretically available flash space
* assuming there is no index, so we have to subtract the space which
* is reserved for the index.
*/
subtract_lebs = min_idx_lebs;
/* Take into account that GC reserves one LEB for its own needs */
subtract_lebs += 1;
/*
* The GC journal head LEB is not really accessible. And since
* different write types go to different heads, we may count only on
* one head's space.
*/
subtract_lebs += c->jhead_cnt - 1;
/* We also reserve one LEB for deletions, which bypass budgeting */
subtract_lebs += 1;
available -= (long long)subtract_lebs * c->leb_size;
/* Subtract the dead space which is not available for use */
available -= c->lst.total_dead;
/*
* Subtract dark space, which might or might not be usable - it depends
* on the data which we have on the media and which will be written. If
* this is a lot of uncompressed or not-compressible data, the dark
* space cannot be used.
*/
available -= c->lst.total_dark;
/*
* However, there is more dark space. The index may be bigger than
* @min_idx_lebs. Those extra LEBs are assumed to be available, but
* their dark space is not included in total_dark, so it is subtracted
* here.
*/
if (c->lst.idx_lebs > min_idx_lebs) {
subtract_lebs = c->lst.idx_lebs - min_idx_lebs;
available -= subtract_lebs * c->dark_wm;
}
/* The calculations are rough and may end up with a negative number */
return available > 0 ? available : 0;
}
/**
* can_use_rp - check whether the user is allowed to use reserved pool.
* @c: UBIFS file-system description object
*
* UBIFS has so-called "reserved pool" which is flash space reserved
* for the superuser and for uses whose UID/GID is recorded in UBIFS superblock.
* This function checks whether current user is allowed to use reserved pool.
* Returns %1 current user is allowed to use reserved pool and %0 otherwise.
*/
static int can_use_rp(struct ubifs_info *c)
{
if (current->fsuid == c->rp_uid || capable(CAP_SYS_RESOURCE) ||
(c->rp_gid != 0 && in_group_p(c->rp_gid)))
return 1;
return 0;
}
/**
* do_budget_space - reserve flash space for index and data growth.
* @c: UBIFS file-system description object
*
* This function makes sure UBIFS has enough free eraseblocks for index growth
* and data.
*
* When budgeting index space, UBIFS reserves twice as more LEBs as the index
* would take if it was consolidated and written to the flash. This guarantees
* that the "in-the-gaps" commit method always succeeds and UBIFS will always
* be able to commit dirty index. So this function basically adds amount of
* budgeted index space to the size of the current index, multiplies this by 2,
* and makes sure this does not exceed the amount of free eraseblocks.
*
* Notes about @c->min_idx_lebs and @c->lst.idx_lebs variables:
* o @c->lst.idx_lebs is the number of LEBs the index currently uses. It might
* be large, because UBIFS does not do any index consolidation as long as
* there is free space. IOW, the index may take a lot of LEBs, but the LEBs
* will contain a lot of dirt.
* o @c->min_idx_lebs is the the index presumably takes. IOW, the index may be
* consolidated to take up to @c->min_idx_lebs LEBs.
*
* This function returns zero in case of success, and %-ENOSPC in case of
* failure.
*/
static int do_budget_space(struct ubifs_info *c)
{
long long outstanding, available;
int lebs, rsvd_idx_lebs, min_idx_lebs;
/* First budget index space */
min_idx_lebs = ubifs_calc_min_idx_lebs(c);
/* Now 'min_idx_lebs' contains number of LEBs to reserve */
if (min_idx_lebs > c->lst.idx_lebs)
rsvd_idx_lebs = min_idx_lebs - c->lst.idx_lebs;
else
rsvd_idx_lebs = 0;
/*
* The number of LEBs that are available to be used by the index is:
*
* @c->lst.empty_lebs + @c->freeable_cnt + @c->idx_gc_cnt -
* @c->lst.taken_empty_lebs
*
* @empty_lebs are available because they are empty. @freeable_cnt are
* available because they contain only free and dirty space and the
* index allocation always occurs after wbufs are synch'ed.
* @idx_gc_cnt are available because they are index LEBs that have been
* garbage collected (including trivial GC) and are awaiting the commit
* before they can be unmapped - note that the in-the-gaps method will
* grab these if it needs them. @taken_empty_lebs are empty_lebs that
* have already been allocated for some purpose (also includes those
* LEBs on the @idx_gc list).
*
* Note, @taken_empty_lebs may temporarily be higher by one because of
* the way we serialize LEB allocations and budgeting. See a comment in
* 'ubifs_find_free_space()'.
*/
lebs = c->lst.empty_lebs + c->freeable_cnt + c->idx_gc_cnt -
c->lst.taken_empty_lebs;
if (unlikely(rsvd_idx_lebs > lebs)) {
dbg_budg("out of indexing space: min_idx_lebs %d (old %d), "
"rsvd_idx_lebs %d", min_idx_lebs, c->min_idx_lebs,
rsvd_idx_lebs);
return -ENOSPC;
}
available = ubifs_calc_available(c, min_idx_lebs);
outstanding = c->budg_data_growth + c->budg_dd_growth;
if (unlikely(available < outstanding)) {
dbg_budg("out of data space: available %lld, outstanding %lld",
available, outstanding);
return -ENOSPC;
}
if (available - outstanding <= c->rp_size && !can_use_rp(c))
return -ENOSPC;
c->min_idx_lebs = min_idx_lebs;
return 0;
}
/**
* calc_idx_growth - calculate approximate index growth from budgeting request.
* @c: UBIFS file-system description object
* @req: budgeting request
*
* For now we assume each new node adds one znode. But this is rather poor
* approximation, though.
*/
static int calc_idx_growth(const struct ubifs_info *c,
const struct ubifs_budget_req *req)
{
int znodes;
znodes = req->new_ino + (req->new_page << UBIFS_BLOCKS_PER_PAGE_SHIFT) +
req->new_dent;
return znodes * c->max_idx_node_sz;
}
/**
* calc_data_growth - calculate approximate amount of new data from budgeting
* request.
* @c: UBIFS file-system description object
* @req: budgeting request
*/
static int calc_data_growth(const struct ubifs_info *c,
const struct ubifs_budget_req *req)
{
int data_growth;
data_growth = req->new_ino ? c->inode_budget : 0;
if (req->new_page)
data_growth += c->page_budget;
if (req->new_dent)
data_growth += c->dent_budget;
data_growth += req->new_ino_d;
return data_growth;
}
/**
* calc_dd_growth - calculate approximate amount of data which makes other data
* dirty from budgeting request.
* @c: UBIFS file-system description object
* @req: budgeting request
*/
static int calc_dd_growth(const struct ubifs_info *c,
const struct ubifs_budget_req *req)
{
int dd_growth;
dd_growth = req->dirtied_page ? c->page_budget : 0;
if (req->dirtied_ino)
dd_growth += c->inode_budget << (req->dirtied_ino - 1);
if (req->mod_dent)
dd_growth += c->dent_budget;
dd_growth += req->dirtied_ino_d;
return dd_growth;
}
/**
* ubifs_budget_space - ensure there is enough space to complete an operation.
* @c: UBIFS file-system description object
* @req: budget request
*
* This function allocates budget for an operation. It uses pessimistic
* approximation of how much flash space the operation needs. The goal of this
* function is to make sure UBIFS always has flash space to flush all dirty
* pages, dirty inodes, and dirty znodes (liability). This function may force
* commit, garbage-collection or write-back. Returns zero in case of success,
* %-ENOSPC if there is no free space and other negative error codes in case of
* failures.
*/
int ubifs_budget_space(struct ubifs_info *c, struct ubifs_budget_req *req)
{
int uninitialized_var(cmt_retries), uninitialized_var(wb_retries);
int err, idx_growth, data_growth, dd_growth;
struct retries_info ri;
ubifs_assert(req->new_page <= 1);
ubifs_assert(req->dirtied_page <= 1);
ubifs_assert(req->new_dent <= 1);
ubifs_assert(req->mod_dent <= 1);
ubifs_assert(req->new_ino <= 1);
ubifs_assert(req->new_ino_d <= UBIFS_MAX_INO_DATA);
ubifs_assert(req->dirtied_ino <= 4);
ubifs_assert(req->dirtied_ino_d <= UBIFS_MAX_INO_DATA * 4);
data_growth = calc_data_growth(c, req);
dd_growth = calc_dd_growth(c, req);
if (!data_growth && !dd_growth)
return 0;
idx_growth = calc_idx_growth(c, req);
memset(&ri, 0, sizeof(struct retries_info));
again:
spin_lock(&c->space_lock);
ubifs_assert(c->budg_idx_growth >= 0);
ubifs_assert(c->budg_data_growth >= 0);
ubifs_assert(c->budg_dd_growth >= 0);
if (unlikely(c->nospace) && (c->nospace_rp || !can_use_rp(c))) {
dbg_budg("no space");
spin_unlock(&c->space_lock);
return -ENOSPC;
}
c->budg_idx_growth += idx_growth;
c->budg_data_growth += data_growth;
c->budg_dd_growth += dd_growth;
err = do_budget_space(c);
if (likely(!err)) {
req->idx_growth = idx_growth;
req->data_growth = data_growth;
req->dd_growth = dd_growth;
spin_unlock(&c->space_lock);
return 0;
}
/* Restore the old values */
c->budg_idx_growth -= idx_growth;
c->budg_data_growth -= data_growth;
c->budg_dd_growth -= dd_growth;
spin_unlock(&c->space_lock);
if (req->fast) {
dbg_budg("no space for fast budgeting");
return err;
}
err = make_free_space(c, &ri);
if (err == -EAGAIN) {
dbg_budg("try again");
cond_resched();
goto again;
} else if (err == -ENOSPC) {
dbg_budg("FS is full, -ENOSPC");
c->nospace = 1;
if (can_use_rp(c) || c->rp_size == 0)
c->nospace_rp = 1;
smp_wmb();
} else
ubifs_err("cannot budget space, error %d", err);
return err;
}
/**
* ubifs_release_budget - release budgeted free space.
* @c: UBIFS file-system description object
* @req: budget request
*
* This function releases the space budgeted by 'ubifs_budget_space()'. Note,
* since the index changes (which were budgeted for in @req->idx_growth) will
* only be written to the media on commit, this function moves the index budget
* from @c->budg_idx_growth to @c->budg_uncommitted_idx. The latter will be
* zeroed by the commit operation.
*/
void ubifs_release_budget(struct ubifs_info *c, struct ubifs_budget_req *req)
{
ubifs_assert(req->new_page <= 1);
ubifs_assert(req->dirtied_page <= 1);
ubifs_assert(req->new_dent <= 1);
ubifs_assert(req->mod_dent <= 1);
ubifs_assert(req->new_ino <= 1);
ubifs_assert(req->new_ino_d <= UBIFS_MAX_INO_DATA);
ubifs_assert(req->dirtied_ino <= 4);
ubifs_assert(req->dirtied_ino_d <= UBIFS_MAX_INO_DATA * 4);
if (!req->recalculate) {
ubifs_assert(req->idx_growth >= 0);
ubifs_assert(req->data_growth >= 0);
ubifs_assert(req->dd_growth >= 0);
}
if (req->recalculate) {
req->data_growth = calc_data_growth(c, req);
req->dd_growth = calc_dd_growth(c, req);
req->idx_growth = calc_idx_growth(c, req);
}
if (!req->data_growth && !req->dd_growth)
return;
c->nospace = c->nospace_rp = 0;
smp_wmb();
spin_lock(&c->space_lock);
c->budg_idx_growth -= req->idx_growth;
c->budg_uncommitted_idx += req->idx_growth;
c->budg_data_growth -= req->data_growth;
c->budg_dd_growth -= req->dd_growth;
c->min_idx_lebs = ubifs_calc_min_idx_lebs(c);
ubifs_assert(c->budg_idx_growth >= 0);
ubifs_assert(c->budg_data_growth >= 0);
ubifs_assert(c->min_idx_lebs < c->main_lebs);
spin_unlock(&c->space_lock);
}
/**
* ubifs_convert_page_budget - convert budget of a new page.
* @c: UBIFS file-system description object
*
* This function converts budget which was allocated for a new page of data to
* the budget of changing an existing page of data. The latter is smaller then
* the former, so this function only does simple re-calculation and does not
* involve any write-back.
*/
void ubifs_convert_page_budget(struct ubifs_info *c)
{
spin_lock(&c->space_lock);
/* Release the index growth reservation */
c->budg_idx_growth -= c->max_idx_node_sz << UBIFS_BLOCKS_PER_PAGE_SHIFT;
/* Release the data growth reservation */
c->budg_data_growth -= c->page_budget;
/* Increase the dirty data growth reservation instead */
c->budg_dd_growth += c->page_budget;
/* And re-calculate the indexing space reservation */
c->min_idx_lebs = ubifs_calc_min_idx_lebs(c);
spin_unlock(&c->space_lock);
}
/**
* ubifs_release_dirty_inode_budget - release dirty inode budget.
* @c: UBIFS file-system description object
* @ui: UBIFS inode to release the budget for
*
* This function releases budget corresponding to a dirty inode. It is usually
* called when after the inode has been written to the media and marked as
* clean.
*/
void ubifs_release_dirty_inode_budget(struct ubifs_info *c,
struct ubifs_inode *ui)
{
struct ubifs_budget_req req;
memset(&req, 0, sizeof(struct ubifs_budget_req));
req.dd_growth = c->inode_budget + ui->data_len;
ubifs_release_budget(c, &req);
}
/**
* ubifs_budg_get_free_space - return amount of free space.
* @c: UBIFS file-system description object
*
* This function returns amount of free space on the file-system.
*/
long long ubifs_budg_get_free_space(struct ubifs_info *c)
{
int min_idx_lebs, rsvd_idx_lebs;
long long available, outstanding, free;
/* Do exactly the same calculations as in 'do_budget_space()' */
spin_lock(&c->space_lock);
min_idx_lebs = ubifs_calc_min_idx_lebs(c);
if (min_idx_lebs > c->lst.idx_lebs)
rsvd_idx_lebs = min_idx_lebs - c->lst.idx_lebs;
else
rsvd_idx_lebs = 0;
if (rsvd_idx_lebs > c->lst.empty_lebs + c->freeable_cnt + c->idx_gc_cnt
- c->lst.taken_empty_lebs) {
spin_unlock(&c->space_lock);
return 0;
}
available = ubifs_calc_available(c, min_idx_lebs);
outstanding = c->budg_data_growth + c->budg_dd_growth;
c->min_idx_lebs = min_idx_lebs;
spin_unlock(&c->space_lock);
if (available > outstanding)
free = ubifs_reported_space(c, available - outstanding);
else
free = 0;
return free;
}