linux/fs/ext4/orphan.c

653 lines
18 KiB
C
Raw Normal View History

/*
* Ext4 orphan inode handling
*/
#include <linux/fs.h>
#include <linux/quotaops.h>
#include <linux/buffer_head.h>
#include "ext4.h"
#include "ext4_jbd2.h"
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
static int ext4_orphan_file_add(handle_t *handle, struct inode *inode)
{
ext4: Improve scalability of ext4 orphan file handling Even though the length of the critical section when adding / removing orphaned inodes was significantly reduced by using orphan file, the contention of lock protecting orphan file still appears high in profiles for truncate / unlink intensive workloads with high number of threads. This patch makes handling of orphan file completely lockless. Also to reduce conflicts between CPUs different CPUs start searching for empty slot in orphan file in different blocks. Performance comparison of locked orphan file handling, lockless orphan file handling, and completely disabled orphan inode handling from 80 CPU Xeon Server with 526 GB of RAM, filesystem located on SAS SSD disk, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Time Orphan locked Orphan lockless No orphan 1 0.945600 0.939400 0.891200 2 1.331800 1.246600 1.174400 4 1.995000 1.780600 1.713200 8 6.424200 4.900000 4.106000 16 14.937600 8.516400 8.138000 32 33.038200 24.565600 24.002200 64 60.823600 39.844600 38.440200 128 122.941400 70.950400 69.315000 So we can see that with lockless orphan file handling, addition / deletion of orphaned inodes got almost completely out of picture even for a microbenchmark stressing it. For reaim creat_clo workload on ramdisk there are also noticeable gains (average of 5 runs): Clients Vanilla (ops/s) Patched (ops/s) creat_clo-1 14705.88 ( 0.00%) 14354.07 * -2.39%* creat_clo-3 27108.43 ( 0.00%) 28301.89 ( 4.40%) creat_clo-5 37406.48 ( 0.00%) 45180.73 * 20.78%* creat_clo-7 41338.58 ( 0.00%) 54687.50 * 32.29%* creat_clo-9 45226.13 ( 0.00%) 62937.07 * 39.16%* creat_clo-11 44000.00 ( 0.00%) 65088.76 * 47.93%* creat_clo-13 36516.85 ( 0.00%) 68661.97 * 88.03%* creat_clo-15 30864.20 ( 0.00%) 69551.78 * 125.35%* creat_clo-17 27478.45 ( 0.00%) 67729.08 * 146.48%* creat_clo-19 25000.00 ( 0.00%) 61621.62 * 146.49%* creat_clo-21 18772.35 ( 0.00%) 63829.79 * 240.02%* creat_clo-23 16698.94 ( 0.00%) 61938.96 * 270.92%* creat_clo-25 14973.05 ( 0.00%) 56947.61 * 280.33%* creat_clo-27 16436.69 ( 0.00%) 65008.03 * 295.51%* creat_clo-29 13949.01 ( 0.00%) 69047.62 * 395.00%* creat_clo-31 14283.52 ( 0.00%) 67982.45 * 375.95%* Reviewed-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Lukas Czerner <lczerner@redhat.com> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-5-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:08 +08:00
int i, j, start;
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
struct ext4_orphan_info *oi = &EXT4_SB(inode->i_sb)->s_orphan_info;
int ret = 0;
ext4: Improve scalability of ext4 orphan file handling Even though the length of the critical section when adding / removing orphaned inodes was significantly reduced by using orphan file, the contention of lock protecting orphan file still appears high in profiles for truncate / unlink intensive workloads with high number of threads. This patch makes handling of orphan file completely lockless. Also to reduce conflicts between CPUs different CPUs start searching for empty slot in orphan file in different blocks. Performance comparison of locked orphan file handling, lockless orphan file handling, and completely disabled orphan inode handling from 80 CPU Xeon Server with 526 GB of RAM, filesystem located on SAS SSD disk, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Time Orphan locked Orphan lockless No orphan 1 0.945600 0.939400 0.891200 2 1.331800 1.246600 1.174400 4 1.995000 1.780600 1.713200 8 6.424200 4.900000 4.106000 16 14.937600 8.516400 8.138000 32 33.038200 24.565600 24.002200 64 60.823600 39.844600 38.440200 128 122.941400 70.950400 69.315000 So we can see that with lockless orphan file handling, addition / deletion of orphaned inodes got almost completely out of picture even for a microbenchmark stressing it. For reaim creat_clo workload on ramdisk there are also noticeable gains (average of 5 runs): Clients Vanilla (ops/s) Patched (ops/s) creat_clo-1 14705.88 ( 0.00%) 14354.07 * -2.39%* creat_clo-3 27108.43 ( 0.00%) 28301.89 ( 4.40%) creat_clo-5 37406.48 ( 0.00%) 45180.73 * 20.78%* creat_clo-7 41338.58 ( 0.00%) 54687.50 * 32.29%* creat_clo-9 45226.13 ( 0.00%) 62937.07 * 39.16%* creat_clo-11 44000.00 ( 0.00%) 65088.76 * 47.93%* creat_clo-13 36516.85 ( 0.00%) 68661.97 * 88.03%* creat_clo-15 30864.20 ( 0.00%) 69551.78 * 125.35%* creat_clo-17 27478.45 ( 0.00%) 67729.08 * 146.48%* creat_clo-19 25000.00 ( 0.00%) 61621.62 * 146.49%* creat_clo-21 18772.35 ( 0.00%) 63829.79 * 240.02%* creat_clo-23 16698.94 ( 0.00%) 61938.96 * 270.92%* creat_clo-25 14973.05 ( 0.00%) 56947.61 * 280.33%* creat_clo-27 16436.69 ( 0.00%) 65008.03 * 295.51%* creat_clo-29 13949.01 ( 0.00%) 69047.62 * 395.00%* creat_clo-31 14283.52 ( 0.00%) 67982.45 * 375.95%* Reviewed-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Lukas Czerner <lczerner@redhat.com> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-5-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:08 +08:00
bool found = false;
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
__le32 *bdata;
int inodes_per_ob = ext4_inodes_per_orphan_block(inode->i_sb);
ext4: Improve scalability of ext4 orphan file handling Even though the length of the critical section when adding / removing orphaned inodes was significantly reduced by using orphan file, the contention of lock protecting orphan file still appears high in profiles for truncate / unlink intensive workloads with high number of threads. This patch makes handling of orphan file completely lockless. Also to reduce conflicts between CPUs different CPUs start searching for empty slot in orphan file in different blocks. Performance comparison of locked orphan file handling, lockless orphan file handling, and completely disabled orphan inode handling from 80 CPU Xeon Server with 526 GB of RAM, filesystem located on SAS SSD disk, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Time Orphan locked Orphan lockless No orphan 1 0.945600 0.939400 0.891200 2 1.331800 1.246600 1.174400 4 1.995000 1.780600 1.713200 8 6.424200 4.900000 4.106000 16 14.937600 8.516400 8.138000 32 33.038200 24.565600 24.002200 64 60.823600 39.844600 38.440200 128 122.941400 70.950400 69.315000 So we can see that with lockless orphan file handling, addition / deletion of orphaned inodes got almost completely out of picture even for a microbenchmark stressing it. For reaim creat_clo workload on ramdisk there are also noticeable gains (average of 5 runs): Clients Vanilla (ops/s) Patched (ops/s) creat_clo-1 14705.88 ( 0.00%) 14354.07 * -2.39%* creat_clo-3 27108.43 ( 0.00%) 28301.89 ( 4.40%) creat_clo-5 37406.48 ( 0.00%) 45180.73 * 20.78%* creat_clo-7 41338.58 ( 0.00%) 54687.50 * 32.29%* creat_clo-9 45226.13 ( 0.00%) 62937.07 * 39.16%* creat_clo-11 44000.00 ( 0.00%) 65088.76 * 47.93%* creat_clo-13 36516.85 ( 0.00%) 68661.97 * 88.03%* creat_clo-15 30864.20 ( 0.00%) 69551.78 * 125.35%* creat_clo-17 27478.45 ( 0.00%) 67729.08 * 146.48%* creat_clo-19 25000.00 ( 0.00%) 61621.62 * 146.49%* creat_clo-21 18772.35 ( 0.00%) 63829.79 * 240.02%* creat_clo-23 16698.94 ( 0.00%) 61938.96 * 270.92%* creat_clo-25 14973.05 ( 0.00%) 56947.61 * 280.33%* creat_clo-27 16436.69 ( 0.00%) 65008.03 * 295.51%* creat_clo-29 13949.01 ( 0.00%) 69047.62 * 395.00%* creat_clo-31 14283.52 ( 0.00%) 67982.45 * 375.95%* Reviewed-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Lukas Czerner <lczerner@redhat.com> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-5-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:08 +08:00
int looped = 0;
/*
* Find block with free orphan entry. Use CPU number for a naive hash
* for a search start in the orphan file
*/
start = raw_smp_processor_id()*13 % oi->of_blocks;
i = start;
do {
if (atomic_dec_if_positive(&oi->of_binfo[i].ob_free_entries)
>= 0) {
found = true;
break;
}
if (++i >= oi->of_blocks)
i = 0;
} while (i != start);
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
ext4: Improve scalability of ext4 orphan file handling Even though the length of the critical section when adding / removing orphaned inodes was significantly reduced by using orphan file, the contention of lock protecting orphan file still appears high in profiles for truncate / unlink intensive workloads with high number of threads. This patch makes handling of orphan file completely lockless. Also to reduce conflicts between CPUs different CPUs start searching for empty slot in orphan file in different blocks. Performance comparison of locked orphan file handling, lockless orphan file handling, and completely disabled orphan inode handling from 80 CPU Xeon Server with 526 GB of RAM, filesystem located on SAS SSD disk, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Time Orphan locked Orphan lockless No orphan 1 0.945600 0.939400 0.891200 2 1.331800 1.246600 1.174400 4 1.995000 1.780600 1.713200 8 6.424200 4.900000 4.106000 16 14.937600 8.516400 8.138000 32 33.038200 24.565600 24.002200 64 60.823600 39.844600 38.440200 128 122.941400 70.950400 69.315000 So we can see that with lockless orphan file handling, addition / deletion of orphaned inodes got almost completely out of picture even for a microbenchmark stressing it. For reaim creat_clo workload on ramdisk there are also noticeable gains (average of 5 runs): Clients Vanilla (ops/s) Patched (ops/s) creat_clo-1 14705.88 ( 0.00%) 14354.07 * -2.39%* creat_clo-3 27108.43 ( 0.00%) 28301.89 ( 4.40%) creat_clo-5 37406.48 ( 0.00%) 45180.73 * 20.78%* creat_clo-7 41338.58 ( 0.00%) 54687.50 * 32.29%* creat_clo-9 45226.13 ( 0.00%) 62937.07 * 39.16%* creat_clo-11 44000.00 ( 0.00%) 65088.76 * 47.93%* creat_clo-13 36516.85 ( 0.00%) 68661.97 * 88.03%* creat_clo-15 30864.20 ( 0.00%) 69551.78 * 125.35%* creat_clo-17 27478.45 ( 0.00%) 67729.08 * 146.48%* creat_clo-19 25000.00 ( 0.00%) 61621.62 * 146.49%* creat_clo-21 18772.35 ( 0.00%) 63829.79 * 240.02%* creat_clo-23 16698.94 ( 0.00%) 61938.96 * 270.92%* creat_clo-25 14973.05 ( 0.00%) 56947.61 * 280.33%* creat_clo-27 16436.69 ( 0.00%) 65008.03 * 295.51%* creat_clo-29 13949.01 ( 0.00%) 69047.62 * 395.00%* creat_clo-31 14283.52 ( 0.00%) 67982.45 * 375.95%* Reviewed-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Lukas Czerner <lczerner@redhat.com> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-5-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:08 +08:00
if (!found) {
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
/*
* For now we don't grow or shrink orphan file. We just use
* whatever was allocated at mke2fs time. The additional
* credits we would have to reserve for each orphan inode
* operation just don't seem worth it.
*/
return -ENOSPC;
}
ret = ext4_journal_get_write_access(handle, inode->i_sb,
oi->of_binfo[i].ob_bh, EXT4_JTR_ORPHAN_FILE);
ext4: Improve scalability of ext4 orphan file handling Even though the length of the critical section when adding / removing orphaned inodes was significantly reduced by using orphan file, the contention of lock protecting orphan file still appears high in profiles for truncate / unlink intensive workloads with high number of threads. This patch makes handling of orphan file completely lockless. Also to reduce conflicts between CPUs different CPUs start searching for empty slot in orphan file in different blocks. Performance comparison of locked orphan file handling, lockless orphan file handling, and completely disabled orphan inode handling from 80 CPU Xeon Server with 526 GB of RAM, filesystem located on SAS SSD disk, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Time Orphan locked Orphan lockless No orphan 1 0.945600 0.939400 0.891200 2 1.331800 1.246600 1.174400 4 1.995000 1.780600 1.713200 8 6.424200 4.900000 4.106000 16 14.937600 8.516400 8.138000 32 33.038200 24.565600 24.002200 64 60.823600 39.844600 38.440200 128 122.941400 70.950400 69.315000 So we can see that with lockless orphan file handling, addition / deletion of orphaned inodes got almost completely out of picture even for a microbenchmark stressing it. For reaim creat_clo workload on ramdisk there are also noticeable gains (average of 5 runs): Clients Vanilla (ops/s) Patched (ops/s) creat_clo-1 14705.88 ( 0.00%) 14354.07 * -2.39%* creat_clo-3 27108.43 ( 0.00%) 28301.89 ( 4.40%) creat_clo-5 37406.48 ( 0.00%) 45180.73 * 20.78%* creat_clo-7 41338.58 ( 0.00%) 54687.50 * 32.29%* creat_clo-9 45226.13 ( 0.00%) 62937.07 * 39.16%* creat_clo-11 44000.00 ( 0.00%) 65088.76 * 47.93%* creat_clo-13 36516.85 ( 0.00%) 68661.97 * 88.03%* creat_clo-15 30864.20 ( 0.00%) 69551.78 * 125.35%* creat_clo-17 27478.45 ( 0.00%) 67729.08 * 146.48%* creat_clo-19 25000.00 ( 0.00%) 61621.62 * 146.49%* creat_clo-21 18772.35 ( 0.00%) 63829.79 * 240.02%* creat_clo-23 16698.94 ( 0.00%) 61938.96 * 270.92%* creat_clo-25 14973.05 ( 0.00%) 56947.61 * 280.33%* creat_clo-27 16436.69 ( 0.00%) 65008.03 * 295.51%* creat_clo-29 13949.01 ( 0.00%) 69047.62 * 395.00%* creat_clo-31 14283.52 ( 0.00%) 67982.45 * 375.95%* Reviewed-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Lukas Czerner <lczerner@redhat.com> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-5-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:08 +08:00
if (ret) {
atomic_inc(&oi->of_binfo[i].ob_free_entries);
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
return ret;
ext4: Improve scalability of ext4 orphan file handling Even though the length of the critical section when adding / removing orphaned inodes was significantly reduced by using orphan file, the contention of lock protecting orphan file still appears high in profiles for truncate / unlink intensive workloads with high number of threads. This patch makes handling of orphan file completely lockless. Also to reduce conflicts between CPUs different CPUs start searching for empty slot in orphan file in different blocks. Performance comparison of locked orphan file handling, lockless orphan file handling, and completely disabled orphan inode handling from 80 CPU Xeon Server with 526 GB of RAM, filesystem located on SAS SSD disk, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Time Orphan locked Orphan lockless No orphan 1 0.945600 0.939400 0.891200 2 1.331800 1.246600 1.174400 4 1.995000 1.780600 1.713200 8 6.424200 4.900000 4.106000 16 14.937600 8.516400 8.138000 32 33.038200 24.565600 24.002200 64 60.823600 39.844600 38.440200 128 122.941400 70.950400 69.315000 So we can see that with lockless orphan file handling, addition / deletion of orphaned inodes got almost completely out of picture even for a microbenchmark stressing it. For reaim creat_clo workload on ramdisk there are also noticeable gains (average of 5 runs): Clients Vanilla (ops/s) Patched (ops/s) creat_clo-1 14705.88 ( 0.00%) 14354.07 * -2.39%* creat_clo-3 27108.43 ( 0.00%) 28301.89 ( 4.40%) creat_clo-5 37406.48 ( 0.00%) 45180.73 * 20.78%* creat_clo-7 41338.58 ( 0.00%) 54687.50 * 32.29%* creat_clo-9 45226.13 ( 0.00%) 62937.07 * 39.16%* creat_clo-11 44000.00 ( 0.00%) 65088.76 * 47.93%* creat_clo-13 36516.85 ( 0.00%) 68661.97 * 88.03%* creat_clo-15 30864.20 ( 0.00%) 69551.78 * 125.35%* creat_clo-17 27478.45 ( 0.00%) 67729.08 * 146.48%* creat_clo-19 25000.00 ( 0.00%) 61621.62 * 146.49%* creat_clo-21 18772.35 ( 0.00%) 63829.79 * 240.02%* creat_clo-23 16698.94 ( 0.00%) 61938.96 * 270.92%* creat_clo-25 14973.05 ( 0.00%) 56947.61 * 280.33%* creat_clo-27 16436.69 ( 0.00%) 65008.03 * 295.51%* creat_clo-29 13949.01 ( 0.00%) 69047.62 * 395.00%* creat_clo-31 14283.52 ( 0.00%) 67982.45 * 375.95%* Reviewed-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Lukas Czerner <lczerner@redhat.com> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-5-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:08 +08:00
}
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
bdata = (__le32 *)(oi->of_binfo[i].ob_bh->b_data);
/* Find empty slot in a block */
ext4: Improve scalability of ext4 orphan file handling Even though the length of the critical section when adding / removing orphaned inodes was significantly reduced by using orphan file, the contention of lock protecting orphan file still appears high in profiles for truncate / unlink intensive workloads with high number of threads. This patch makes handling of orphan file completely lockless. Also to reduce conflicts between CPUs different CPUs start searching for empty slot in orphan file in different blocks. Performance comparison of locked orphan file handling, lockless orphan file handling, and completely disabled orphan inode handling from 80 CPU Xeon Server with 526 GB of RAM, filesystem located on SAS SSD disk, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Time Orphan locked Orphan lockless No orphan 1 0.945600 0.939400 0.891200 2 1.331800 1.246600 1.174400 4 1.995000 1.780600 1.713200 8 6.424200 4.900000 4.106000 16 14.937600 8.516400 8.138000 32 33.038200 24.565600 24.002200 64 60.823600 39.844600 38.440200 128 122.941400 70.950400 69.315000 So we can see that with lockless orphan file handling, addition / deletion of orphaned inodes got almost completely out of picture even for a microbenchmark stressing it. For reaim creat_clo workload on ramdisk there are also noticeable gains (average of 5 runs): Clients Vanilla (ops/s) Patched (ops/s) creat_clo-1 14705.88 ( 0.00%) 14354.07 * -2.39%* creat_clo-3 27108.43 ( 0.00%) 28301.89 ( 4.40%) creat_clo-5 37406.48 ( 0.00%) 45180.73 * 20.78%* creat_clo-7 41338.58 ( 0.00%) 54687.50 * 32.29%* creat_clo-9 45226.13 ( 0.00%) 62937.07 * 39.16%* creat_clo-11 44000.00 ( 0.00%) 65088.76 * 47.93%* creat_clo-13 36516.85 ( 0.00%) 68661.97 * 88.03%* creat_clo-15 30864.20 ( 0.00%) 69551.78 * 125.35%* creat_clo-17 27478.45 ( 0.00%) 67729.08 * 146.48%* creat_clo-19 25000.00 ( 0.00%) 61621.62 * 146.49%* creat_clo-21 18772.35 ( 0.00%) 63829.79 * 240.02%* creat_clo-23 16698.94 ( 0.00%) 61938.96 * 270.92%* creat_clo-25 14973.05 ( 0.00%) 56947.61 * 280.33%* creat_clo-27 16436.69 ( 0.00%) 65008.03 * 295.51%* creat_clo-29 13949.01 ( 0.00%) 69047.62 * 395.00%* creat_clo-31 14283.52 ( 0.00%) 67982.45 * 375.95%* Reviewed-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Lukas Czerner <lczerner@redhat.com> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-5-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:08 +08:00
j = 0;
do {
if (looped) {
/*
* Did we walk through the block several times without
* finding free entry? It is theoretically possible
* if entries get constantly allocated and freed or
* if the block is corrupted. Avoid indefinite looping
* and bail. We'll use orphan list instead.
*/
if (looped > 3) {
atomic_inc(&oi->of_binfo[i].ob_free_entries);
return -ENOSPC;
}
cond_resched();
}
while (bdata[j]) {
if (++j >= inodes_per_ob) {
j = 0;
looped++;
}
}
} while (cmpxchg(&bdata[j], (__le32)0, cpu_to_le32(inode->i_ino)) !=
(__le32)0);
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
EXT4_I(inode)->i_orphan_idx = i * inodes_per_ob + j;
ext4_set_inode_state(inode, EXT4_STATE_ORPHAN_FILE);
return ext4_handle_dirty_metadata(handle, NULL, oi->of_binfo[i].ob_bh);
}
/*
* ext4_orphan_add() links an unlinked or truncated inode into a list of
* such inodes, starting at the superblock, in case we crash before the
* file is closed/deleted, or in case the inode truncate spans multiple
* transactions and the last transaction is not recovered after a crash.
*
* At filesystem recovery time, we walk this list deleting unlinked
* inodes and truncating linked inodes in ext4_orphan_cleanup().
*
* Orphan list manipulation functions must be called under i_rwsem unless
* we are just creating the inode or deleting it.
*/
int ext4_orphan_add(handle_t *handle, struct inode *inode)
{
struct super_block *sb = inode->i_sb;
struct ext4_sb_info *sbi = EXT4_SB(sb);
struct ext4_iloc iloc;
int err = 0, rc;
bool dirty = false;
if (!sbi->s_journal || is_bad_inode(inode))
return 0;
WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
!inode_is_locked(inode));
/*
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
* Inode orphaned in orphan file or in orphan list?
*/
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
if (ext4_test_inode_state(inode, EXT4_STATE_ORPHAN_FILE) ||
!list_empty(&EXT4_I(inode)->i_orphan))
return 0;
/*
* Orphan handling is only valid for files with data blocks
* being truncated, or files being unlinked. Note that we either
* hold i_rwsem, or the inode can not be referenced from outside,
* so i_nlink should not be bumped due to race
*/
ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
if (sbi->s_orphan_info.of_blocks) {
err = ext4_orphan_file_add(handle, inode);
/*
* Fallback to normal orphan list of orphan file is
* out of space
*/
if (err != -ENOSPC)
return err;
}
BUFFER_TRACE(sbi->s_sbh, "get_write_access");
err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
EXT4_JTR_NONE);
if (err)
goto out;
err = ext4_reserve_inode_write(handle, inode, &iloc);
if (err)
goto out;
mutex_lock(&sbi->s_orphan_lock);
/*
* Due to previous errors inode may be already a part of on-disk
* orphan list. If so skip on-disk list modification.
*/
if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
(le32_to_cpu(sbi->s_es->s_inodes_count))) {
/* Insert this inode at the head of the on-disk orphan list */
NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
lock_buffer(sbi->s_sbh);
sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
ext4_superblock_csum_set(sb);
unlock_buffer(sbi->s_sbh);
dirty = true;
}
list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
mutex_unlock(&sbi->s_orphan_lock);
if (dirty) {
err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
if (!err)
err = rc;
if (err) {
/*
* We have to remove inode from in-memory list if
* addition to on disk orphan list failed. Stray orphan
* list entries can cause panics at unmount time.
*/
mutex_lock(&sbi->s_orphan_lock);
list_del_init(&EXT4_I(inode)->i_orphan);
mutex_unlock(&sbi->s_orphan_lock);
}
} else
brelse(iloc.bh);
ext4_debug("superblock will point to %lu\n", inode->i_ino);
ext4_debug("orphan inode %lu will point to %d\n",
inode->i_ino, NEXT_ORPHAN(inode));
out:
ext4_std_error(sb, err);
return err;
}
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
static int ext4_orphan_file_del(handle_t *handle, struct inode *inode)
{
struct ext4_orphan_info *oi = &EXT4_SB(inode->i_sb)->s_orphan_info;
__le32 *bdata;
int blk, off;
int inodes_per_ob = ext4_inodes_per_orphan_block(inode->i_sb);
int ret = 0;
if (!handle)
goto out;
blk = EXT4_I(inode)->i_orphan_idx / inodes_per_ob;
off = EXT4_I(inode)->i_orphan_idx % inodes_per_ob;
if (WARN_ON_ONCE(blk >= oi->of_blocks))
goto out;
ret = ext4_journal_get_write_access(handle, inode->i_sb,
oi->of_binfo[blk].ob_bh, EXT4_JTR_ORPHAN_FILE);
if (ret)
goto out;
bdata = (__le32 *)(oi->of_binfo[blk].ob_bh->b_data);
bdata[off] = 0;
ext4: Improve scalability of ext4 orphan file handling Even though the length of the critical section when adding / removing orphaned inodes was significantly reduced by using orphan file, the contention of lock protecting orphan file still appears high in profiles for truncate / unlink intensive workloads with high number of threads. This patch makes handling of orphan file completely lockless. Also to reduce conflicts between CPUs different CPUs start searching for empty slot in orphan file in different blocks. Performance comparison of locked orphan file handling, lockless orphan file handling, and completely disabled orphan inode handling from 80 CPU Xeon Server with 526 GB of RAM, filesystem located on SAS SSD disk, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Time Orphan locked Orphan lockless No orphan 1 0.945600 0.939400 0.891200 2 1.331800 1.246600 1.174400 4 1.995000 1.780600 1.713200 8 6.424200 4.900000 4.106000 16 14.937600 8.516400 8.138000 32 33.038200 24.565600 24.002200 64 60.823600 39.844600 38.440200 128 122.941400 70.950400 69.315000 So we can see that with lockless orphan file handling, addition / deletion of orphaned inodes got almost completely out of picture even for a microbenchmark stressing it. For reaim creat_clo workload on ramdisk there are also noticeable gains (average of 5 runs): Clients Vanilla (ops/s) Patched (ops/s) creat_clo-1 14705.88 ( 0.00%) 14354.07 * -2.39%* creat_clo-3 27108.43 ( 0.00%) 28301.89 ( 4.40%) creat_clo-5 37406.48 ( 0.00%) 45180.73 * 20.78%* creat_clo-7 41338.58 ( 0.00%) 54687.50 * 32.29%* creat_clo-9 45226.13 ( 0.00%) 62937.07 * 39.16%* creat_clo-11 44000.00 ( 0.00%) 65088.76 * 47.93%* creat_clo-13 36516.85 ( 0.00%) 68661.97 * 88.03%* creat_clo-15 30864.20 ( 0.00%) 69551.78 * 125.35%* creat_clo-17 27478.45 ( 0.00%) 67729.08 * 146.48%* creat_clo-19 25000.00 ( 0.00%) 61621.62 * 146.49%* creat_clo-21 18772.35 ( 0.00%) 63829.79 * 240.02%* creat_clo-23 16698.94 ( 0.00%) 61938.96 * 270.92%* creat_clo-25 14973.05 ( 0.00%) 56947.61 * 280.33%* creat_clo-27 16436.69 ( 0.00%) 65008.03 * 295.51%* creat_clo-29 13949.01 ( 0.00%) 69047.62 * 395.00%* creat_clo-31 14283.52 ( 0.00%) 67982.45 * 375.95%* Reviewed-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Lukas Czerner <lczerner@redhat.com> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-5-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:08 +08:00
atomic_inc(&oi->of_binfo[blk].ob_free_entries);
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
ret = ext4_handle_dirty_metadata(handle, NULL, oi->of_binfo[blk].ob_bh);
out:
ext4_clear_inode_state(inode, EXT4_STATE_ORPHAN_FILE);
INIT_LIST_HEAD(&EXT4_I(inode)->i_orphan);
return ret;
}
/*
* ext4_orphan_del() removes an unlinked or truncated inode from the list
* of such inodes stored on disk, because it is finally being cleaned up.
*/
int ext4_orphan_del(handle_t *handle, struct inode *inode)
{
struct list_head *prev;
struct ext4_inode_info *ei = EXT4_I(inode);
struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
__u32 ino_next;
struct ext4_iloc iloc;
int err = 0;
if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
return 0;
WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
!inode_is_locked(inode));
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
if (ext4_test_inode_state(inode, EXT4_STATE_ORPHAN_FILE))
return ext4_orphan_file_del(handle, inode);
/* Do this quick check before taking global s_orphan_lock. */
if (list_empty(&ei->i_orphan))
return 0;
if (handle) {
/* Grab inode buffer early before taking global s_orphan_lock */
err = ext4_reserve_inode_write(handle, inode, &iloc);
}
mutex_lock(&sbi->s_orphan_lock);
ext4_debug("remove inode %lu from orphan list\n", inode->i_ino);
prev = ei->i_orphan.prev;
list_del_init(&ei->i_orphan);
/* If we're on an error path, we may not have a valid
* transaction handle with which to update the orphan list on
* disk, but we still need to remove the inode from the linked
* list in memory. */
if (!handle || err) {
mutex_unlock(&sbi->s_orphan_lock);
goto out_err;
}
ino_next = NEXT_ORPHAN(inode);
if (prev == &sbi->s_orphan) {
ext4_debug("superblock will point to %u\n", ino_next);
BUFFER_TRACE(sbi->s_sbh, "get_write_access");
err = ext4_journal_get_write_access(handle, inode->i_sb,
sbi->s_sbh, EXT4_JTR_NONE);
if (err) {
mutex_unlock(&sbi->s_orphan_lock);
goto out_brelse;
}
lock_buffer(sbi->s_sbh);
sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
ext4_superblock_csum_set(inode->i_sb);
unlock_buffer(sbi->s_sbh);
mutex_unlock(&sbi->s_orphan_lock);
err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
} else {
struct ext4_iloc iloc2;
struct inode *i_prev =
&list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
ext4_debug("orphan inode %lu will point to %u\n",
i_prev->i_ino, ino_next);
err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
if (err) {
mutex_unlock(&sbi->s_orphan_lock);
goto out_brelse;
}
NEXT_ORPHAN(i_prev) = ino_next;
err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
mutex_unlock(&sbi->s_orphan_lock);
}
if (err)
goto out_brelse;
NEXT_ORPHAN(inode) = 0;
err = ext4_mark_iloc_dirty(handle, inode, &iloc);
out_err:
ext4_std_error(inode->i_sb, err);
return err;
out_brelse:
brelse(iloc.bh);
goto out_err;
}
#ifdef CONFIG_QUOTA
static int ext4_quota_on_mount(struct super_block *sb, int type)
{
return dquot_quota_on_mount(sb,
rcu_dereference_protected(EXT4_SB(sb)->s_qf_names[type],
lockdep_is_held(&sb->s_umount)),
EXT4_SB(sb)->s_jquota_fmt, type);
}
#endif
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
static void ext4_process_orphan(struct inode *inode,
int *nr_truncates, int *nr_orphans)
{
struct super_block *sb = inode->i_sb;
int ret;
dquot_initialize(inode);
if (inode->i_nlink) {
if (test_opt(sb, DEBUG))
ext4_msg(sb, KERN_DEBUG,
"%s: truncating inode %lu to %lld bytes",
__func__, inode->i_ino, inode->i_size);
ext4_debug("truncating inode %lu to %lld bytes\n",
inode->i_ino, inode->i_size);
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
inode_lock(inode);
truncate_inode_pages(inode->i_mapping, inode->i_size);
ret = ext4_truncate(inode);
if (ret) {
/*
* We need to clean up the in-core orphan list
* manually if ext4_truncate() failed to get a
* transaction handle.
*/
ext4_orphan_del(NULL, inode);
ext4_std_error(inode->i_sb, ret);
}
inode_unlock(inode);
(*nr_truncates)++;
} else {
if (test_opt(sb, DEBUG))
ext4_msg(sb, KERN_DEBUG,
"%s: deleting unreferenced inode %lu",
__func__, inode->i_ino);
ext4_debug("deleting unreferenced inode %lu\n",
inode->i_ino);
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
(*nr_orphans)++;
}
iput(inode); /* The delete magic happens here! */
}
/* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
* the superblock) which were deleted from all directories, but held open by
* a process at the time of a crash. We walk the list and try to delete these
* inodes at recovery time (only with a read-write filesystem).
*
* In order to keep the orphan inode chain consistent during traversal (in
* case of crash during recovery), we link each inode into the superblock
* orphan list_head and handle it the same way as an inode deletion during
* normal operation (which journals the operations for us).
*
* We only do an iget() and an iput() on each inode, which is very safe if we
* accidentally point at an in-use or already deleted inode. The worst that
* can happen in this case is that we get a "bit already cleared" message from
* ext4_free_inode(). The only reason we would point at a wrong inode is if
* e2fsck was run on this filesystem, and it must have already done the orphan
* inode cleanup for us, so we can safely abort without any further action.
*/
void ext4_orphan_cleanup(struct super_block *sb, struct ext4_super_block *es)
{
unsigned int s_flags = sb->s_flags;
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
int nr_orphans = 0, nr_truncates = 0;
struct inode *inode;
int i, j;
#ifdef CONFIG_QUOTA
int quota_update = 0;
#endif
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
__le32 *bdata;
struct ext4_orphan_info *oi = &EXT4_SB(sb)->s_orphan_info;
int inodes_per_ob = ext4_inodes_per_orphan_block(sb);
if (!es->s_last_orphan && !oi->of_blocks) {
ext4_debug("no orphan inodes to clean up\n");
return;
}
if (bdev_read_only(sb->s_bdev)) {
ext4_msg(sb, KERN_ERR, "write access "
"unavailable, skipping orphan cleanup");
return;
}
/* Check if feature set would not allow a r/w mount */
if (!ext4_feature_set_ok(sb, 0)) {
ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
"unknown ROCOMPAT features");
return;
}
if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
/* don't clear list on RO mount w/ errors */
if (es->s_last_orphan && !(s_flags & SB_RDONLY)) {
ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
"clearing orphan list.");
es->s_last_orphan = 0;
}
ext4_debug("Skipping orphan recovery on fs with errors.\n");
return;
}
if (s_flags & SB_RDONLY) {
ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
sb->s_flags &= ~SB_RDONLY;
}
#ifdef CONFIG_QUOTA
/*
* Turn on quotas which were not enabled for read-only mounts if
* filesystem has quota feature, so that they are updated correctly.
*/
if (ext4_has_feature_quota(sb) && (s_flags & SB_RDONLY)) {
int ret = ext4_enable_quotas(sb);
if (!ret)
quota_update = 1;
else
ext4_msg(sb, KERN_ERR,
"Cannot turn on quotas: error %d", ret);
}
/* Turn on journaled quotas used for old sytle */
for (i = 0; i < EXT4_MAXQUOTAS; i++) {
if (EXT4_SB(sb)->s_qf_names[i]) {
int ret = ext4_quota_on_mount(sb, i);
if (!ret)
quota_update = 1;
else
ext4_msg(sb, KERN_ERR,
"Cannot turn on journaled "
"quota: type %d: error %d", i, ret);
}
}
#endif
while (es->s_last_orphan) {
/*
* We may have encountered an error during cleanup; if
* so, skip the rest.
*/
if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
ext4_debug("Skipping orphan recovery on fs with errors.\n");
es->s_last_orphan = 0;
break;
}
inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
if (IS_ERR(inode)) {
es->s_last_orphan = 0;
break;
}
list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
ext4_process_orphan(inode, &nr_truncates, &nr_orphans);
}
for (i = 0; i < oi->of_blocks; i++) {
bdata = (__le32 *)(oi->of_binfo[i].ob_bh->b_data);
for (j = 0; j < inodes_per_ob; j++) {
if (!bdata[j])
continue;
inode = ext4_orphan_get(sb, le32_to_cpu(bdata[j]));
if (IS_ERR(inode))
continue;
ext4_set_inode_state(inode, EXT4_STATE_ORPHAN_FILE);
EXT4_I(inode)->i_orphan_idx = i * inodes_per_ob + j;
ext4_process_orphan(inode, &nr_truncates, &nr_orphans);
}
}
#define PLURAL(x) (x), ((x) == 1) ? "" : "s"
if (nr_orphans)
ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
PLURAL(nr_orphans));
if (nr_truncates)
ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
PLURAL(nr_truncates));
#ifdef CONFIG_QUOTA
/* Turn off quotas if they were enabled for orphan cleanup */
if (quota_update) {
for (i = 0; i < EXT4_MAXQUOTAS; i++) {
if (sb_dqopt(sb)->files[i])
dquot_quota_off(sb, i);
}
}
#endif
sb->s_flags = s_flags; /* Restore SB_RDONLY status */
}
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
void ext4_release_orphan_info(struct super_block *sb)
{
int i;
struct ext4_orphan_info *oi = &EXT4_SB(sb)->s_orphan_info;
if (!oi->of_blocks)
return;
for (i = 0; i < oi->of_blocks; i++)
brelse(oi->of_binfo[i].ob_bh);
kfree(oi->of_binfo);
}
static struct ext4_orphan_block_tail *ext4_orphan_block_tail(
struct super_block *sb,
struct buffer_head *bh)
{
return (struct ext4_orphan_block_tail *)(bh->b_data + sb->s_blocksize -
sizeof(struct ext4_orphan_block_tail));
}
static int ext4_orphan_file_block_csum_verify(struct super_block *sb,
struct buffer_head *bh)
{
__u32 calculated;
int inodes_per_ob = ext4_inodes_per_orphan_block(sb);
struct ext4_orphan_info *oi = &EXT4_SB(sb)->s_orphan_info;
struct ext4_orphan_block_tail *ot;
__le64 dsk_block_nr = cpu_to_le64(bh->b_blocknr);
if (!ext4_has_metadata_csum(sb))
return 1;
ot = ext4_orphan_block_tail(sb, bh);
calculated = ext4_chksum(EXT4_SB(sb), oi->of_csum_seed,
(__u8 *)&dsk_block_nr, sizeof(dsk_block_nr));
calculated = ext4_chksum(EXT4_SB(sb), calculated, (__u8 *)bh->b_data,
inodes_per_ob * sizeof(__u32));
return le32_to_cpu(ot->ob_checksum) == calculated;
}
/* This gets called only when checksumming is enabled */
void ext4_orphan_file_block_trigger(struct jbd2_buffer_trigger_type *triggers,
struct buffer_head *bh,
void *data, size_t size)
{
struct super_block *sb = EXT4_TRIGGER(triggers)->sb;
__u32 csum;
int inodes_per_ob = ext4_inodes_per_orphan_block(sb);
struct ext4_orphan_info *oi = &EXT4_SB(sb)->s_orphan_info;
struct ext4_orphan_block_tail *ot;
__le64 dsk_block_nr = cpu_to_le64(bh->b_blocknr);
csum = ext4_chksum(EXT4_SB(sb), oi->of_csum_seed,
(__u8 *)&dsk_block_nr, sizeof(dsk_block_nr));
csum = ext4_chksum(EXT4_SB(sb), csum, (__u8 *)data,
inodes_per_ob * sizeof(__u32));
ot = ext4_orphan_block_tail(sb, bh);
ot->ob_checksum = cpu_to_le32(csum);
}
int ext4_init_orphan_info(struct super_block *sb)
{
struct ext4_orphan_info *oi = &EXT4_SB(sb)->s_orphan_info;
struct inode *inode;
int i, j;
int ret;
int free;
__le32 *bdata;
int inodes_per_ob = ext4_inodes_per_orphan_block(sb);
struct ext4_orphan_block_tail *ot;
ino_t orphan_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_orphan_file_inum);
if (!ext4_has_feature_orphan_file(sb))
return 0;
inode = ext4_iget(sb, orphan_ino, EXT4_IGET_SPECIAL);
if (IS_ERR(inode)) {
ext4_msg(sb, KERN_ERR, "get orphan inode failed");
return PTR_ERR(inode);
}
oi->of_blocks = inode->i_size >> sb->s_blocksize_bits;
oi->of_csum_seed = EXT4_I(inode)->i_csum_seed;
oi->of_binfo = kmalloc(oi->of_blocks*sizeof(struct ext4_orphan_block),
GFP_KERNEL);
if (!oi->of_binfo) {
ret = -ENOMEM;
goto out_put;
}
for (i = 0; i < oi->of_blocks; i++) {
oi->of_binfo[i].ob_bh = ext4_bread(NULL, inode, i, 0);
if (IS_ERR(oi->of_binfo[i].ob_bh)) {
ret = PTR_ERR(oi->of_binfo[i].ob_bh);
goto out_free;
}
if (!oi->of_binfo[i].ob_bh) {
ret = -EIO;
goto out_free;
}
ot = ext4_orphan_block_tail(sb, oi->of_binfo[i].ob_bh);
if (le32_to_cpu(ot->ob_magic) != EXT4_ORPHAN_BLOCK_MAGIC) {
ext4_error(sb, "orphan file block %d: bad magic", i);
ret = -EIO;
goto out_free;
}
if (!ext4_orphan_file_block_csum_verify(sb,
oi->of_binfo[i].ob_bh)) {
ext4_error(sb, "orphan file block %d: bad checksum", i);
ret = -EIO;
goto out_free;
}
bdata = (__le32 *)(oi->of_binfo[i].ob_bh->b_data);
free = 0;
for (j = 0; j < inodes_per_ob; j++)
if (bdata[j] == 0)
free++;
ext4: Improve scalability of ext4 orphan file handling Even though the length of the critical section when adding / removing orphaned inodes was significantly reduced by using orphan file, the contention of lock protecting orphan file still appears high in profiles for truncate / unlink intensive workloads with high number of threads. This patch makes handling of orphan file completely lockless. Also to reduce conflicts between CPUs different CPUs start searching for empty slot in orphan file in different blocks. Performance comparison of locked orphan file handling, lockless orphan file handling, and completely disabled orphan inode handling from 80 CPU Xeon Server with 526 GB of RAM, filesystem located on SAS SSD disk, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Time Orphan locked Orphan lockless No orphan 1 0.945600 0.939400 0.891200 2 1.331800 1.246600 1.174400 4 1.995000 1.780600 1.713200 8 6.424200 4.900000 4.106000 16 14.937600 8.516400 8.138000 32 33.038200 24.565600 24.002200 64 60.823600 39.844600 38.440200 128 122.941400 70.950400 69.315000 So we can see that with lockless orphan file handling, addition / deletion of orphaned inodes got almost completely out of picture even for a microbenchmark stressing it. For reaim creat_clo workload on ramdisk there are also noticeable gains (average of 5 runs): Clients Vanilla (ops/s) Patched (ops/s) creat_clo-1 14705.88 ( 0.00%) 14354.07 * -2.39%* creat_clo-3 27108.43 ( 0.00%) 28301.89 ( 4.40%) creat_clo-5 37406.48 ( 0.00%) 45180.73 * 20.78%* creat_clo-7 41338.58 ( 0.00%) 54687.50 * 32.29%* creat_clo-9 45226.13 ( 0.00%) 62937.07 * 39.16%* creat_clo-11 44000.00 ( 0.00%) 65088.76 * 47.93%* creat_clo-13 36516.85 ( 0.00%) 68661.97 * 88.03%* creat_clo-15 30864.20 ( 0.00%) 69551.78 * 125.35%* creat_clo-17 27478.45 ( 0.00%) 67729.08 * 146.48%* creat_clo-19 25000.00 ( 0.00%) 61621.62 * 146.49%* creat_clo-21 18772.35 ( 0.00%) 63829.79 * 240.02%* creat_clo-23 16698.94 ( 0.00%) 61938.96 * 270.92%* creat_clo-25 14973.05 ( 0.00%) 56947.61 * 280.33%* creat_clo-27 16436.69 ( 0.00%) 65008.03 * 295.51%* creat_clo-29 13949.01 ( 0.00%) 69047.62 * 395.00%* creat_clo-31 14283.52 ( 0.00%) 67982.45 * 375.95%* Reviewed-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Lukas Czerner <lczerner@redhat.com> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-5-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:08 +08:00
atomic_set(&oi->of_binfo[i].ob_free_entries, free);
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
}
iput(inode);
return 0;
out_free:
for (i--; i >= 0; i--)
brelse(oi->of_binfo[i].ob_bh);
kfree(oi->of_binfo);
out_put:
iput(inode);
return ret;
}
int ext4_orphan_file_empty(struct super_block *sb)
{
struct ext4_orphan_info *oi = &EXT4_SB(sb)->s_orphan_info;
int i;
int inodes_per_ob = ext4_inodes_per_orphan_block(sb);
if (!ext4_has_feature_orphan_file(sb))
return 1;
for (i = 0; i < oi->of_blocks; i++)
ext4: Improve scalability of ext4 orphan file handling Even though the length of the critical section when adding / removing orphaned inodes was significantly reduced by using orphan file, the contention of lock protecting orphan file still appears high in profiles for truncate / unlink intensive workloads with high number of threads. This patch makes handling of orphan file completely lockless. Also to reduce conflicts between CPUs different CPUs start searching for empty slot in orphan file in different blocks. Performance comparison of locked orphan file handling, lockless orphan file handling, and completely disabled orphan inode handling from 80 CPU Xeon Server with 526 GB of RAM, filesystem located on SAS SSD disk, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Time Orphan locked Orphan lockless No orphan 1 0.945600 0.939400 0.891200 2 1.331800 1.246600 1.174400 4 1.995000 1.780600 1.713200 8 6.424200 4.900000 4.106000 16 14.937600 8.516400 8.138000 32 33.038200 24.565600 24.002200 64 60.823600 39.844600 38.440200 128 122.941400 70.950400 69.315000 So we can see that with lockless orphan file handling, addition / deletion of orphaned inodes got almost completely out of picture even for a microbenchmark stressing it. For reaim creat_clo workload on ramdisk there are also noticeable gains (average of 5 runs): Clients Vanilla (ops/s) Patched (ops/s) creat_clo-1 14705.88 ( 0.00%) 14354.07 * -2.39%* creat_clo-3 27108.43 ( 0.00%) 28301.89 ( 4.40%) creat_clo-5 37406.48 ( 0.00%) 45180.73 * 20.78%* creat_clo-7 41338.58 ( 0.00%) 54687.50 * 32.29%* creat_clo-9 45226.13 ( 0.00%) 62937.07 * 39.16%* creat_clo-11 44000.00 ( 0.00%) 65088.76 * 47.93%* creat_clo-13 36516.85 ( 0.00%) 68661.97 * 88.03%* creat_clo-15 30864.20 ( 0.00%) 69551.78 * 125.35%* creat_clo-17 27478.45 ( 0.00%) 67729.08 * 146.48%* creat_clo-19 25000.00 ( 0.00%) 61621.62 * 146.49%* creat_clo-21 18772.35 ( 0.00%) 63829.79 * 240.02%* creat_clo-23 16698.94 ( 0.00%) 61938.96 * 270.92%* creat_clo-25 14973.05 ( 0.00%) 56947.61 * 280.33%* creat_clo-27 16436.69 ( 0.00%) 65008.03 * 295.51%* creat_clo-29 13949.01 ( 0.00%) 69047.62 * 395.00%* creat_clo-31 14283.52 ( 0.00%) 67982.45 * 375.95%* Reviewed-by: Theodore Ts'o <tytso@mit.edu> Reviewed-by: Lukas Czerner <lczerner@redhat.com> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-5-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:08 +08:00
if (atomic_read(&oi->of_binfo[i].ob_free_entries) !=
inodes_per_ob)
ext4: Speedup ext4 orphan inode handling Ext4 orphan inode handling is a bottleneck for workloads which heavily truncate / unlink small files since it contends on the global s_orphan_mutex lock (and generally it's difficult to improve scalability of the ondisk linked list of orphaned inodes). This patch implements new way of handling orphan inodes. Instead of linking orphaned inode into a linked list, we store it's inode number in a new special file which we call "orphan file". Only if there's no more space in the orphan file (too many inodes are currently orphaned) we fall back to using old style linked list. Currently we protect operations in the orphan file with a spinlock for simplicity but even in this setting we can substantially reduce the length of the critical section and thus speedup some workloads. In the next patch we improve this by making orphan handling lockless. Note that the change is backwards compatible when the filesystem is clean - the existence of the orphan file is a compat feature, we set another ro-compat feature indicating orphan file needs scanning for orphaned inodes when mounting filesystem read-write. This ro-compat feature gets cleared on unmount / remount read-only. Some performance data from 80 CPU Xeon Server with 512 GB of RAM, filesystem located on SSD, average of 5 runs: stress-orphan (microbenchmark truncating files byte-by-byte from N processes in parallel) Threads Time Time Vanilla Patched 1 1.057200 0.945600 2 1.680400 1.331800 4 2.547000 1.995000 8 7.049400 6.424200 16 14.827800 14.937600 32 40.948200 33.038200 64 87.787400 60.823600 128 206.504000 122.941400 So we can see significant wins all over the board. Reviewed-by: Theodore Ts'o <tytso@mit.edu> Signed-off-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20210816095713.16537-3-jack@suse.cz Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2021-08-16 17:57:06 +08:00
return 0;
return 1;
}