[apue] Linux / Windows 系统上只能建立不超过 PATH_MAX / MAX_PATH 长度的路径吗?

问题的提出

在处理文件系统路径的时候,我们一般会先开辟一块内存区,用来接收路径、或者拼接好路径传递给系统调用。这是因为路径在各个系统上都有最大长度限制,在 Windows 上这个值是 MAX_PATH,一般不能超过 260;在 Linux 上这个值是 PATH_MAX,一般不能超过 4096 (或者通过 pathconf (_PC_PATH_MAX, ...) 来获取,但是一般也是 4096),就像下面这段典型的代码:

 1 int main ()
 2 {
 3 #ifdef WIN32
 4     char buf[MAX_PATH + 1] = { 0 };
 5     if (GetModuleFileNameA(NULL, buf, MAX_PATH) == 0)
 6     {
 7         printf("get current module path failed, errno %d", GetLastError());
 8         return -1;
 9     }
10 #else
11     char buf[PATH_MAX + 1] = { 0 };
12     if (readlink("/proc/self/exe", buf, PATH_MAX) < 0)
13     {
14         printf("read exe path failed, errno %d", errno);
15         return -1;
16     }
17 #endif
18 
19     printf("current executable file path: %s\n", buf);
20     return 0; 
21 }

 

它用来获取当前可执行文件的完整路径。在 Windows 与 Linux 上分别调用了 GetModuleFileName 与 readlink 系统调用,且事先在栈上分配了接收完整路径的内存区 buf。最后将成功获取的路径打印到控制台。在两种系统上,输出分别如下:

Windows

current executable  file path: E:\code\apue\02.chapter\Release\path_max.exe

 Linux

current executable  file path:  /home/yunhai/code/apue/02 .chapter /path_max

 

其它与路径相关的调用也是类似的。那么问题来了,这个最大路径长度是为了方便程序编写 (不然需要动态分配内存,且需要两次调用,其中一次用于获取最终的路径长度),还是说底层的文件系统就只能支持这么长的路径呢?

问题的验证

为了弄清楚这个问题,我专门写了一个测试程序:

path_max.c

  1 #ifdef WIN32
  2 #  ifndef _WIN32_WINNT            // 指定要求的最低平台是 Windows Vista。
  3 #  define _WIN32_WINNT 0x0600     // 将此值更改为相应的值,以适用于 Windows 的其他版本。
  4 #  endif
  5 #include <windows.h> 
  6 #include <direct.h> 
  7 #else 
  8 #include <limits.h>
  9 #include <unistd.h> 
 10 #include <fcntl.h> 
 11 #include <errno.h> 
 12 #include <sys/stat.h> 
 13 #endif 
 14 #include <string.h>
 15 #include <stdio.h> 
 16 #include <stdlib.h> 
 17 #include <time.h>
 18 
 19 void get_random_name (char *name, int len, int level)
 20 {
 21   int i, n; 
 22   sprintf (name, "%d", level); 
 23   n = strlen (name); 
 24   for (i=n; i<len; ++ i)
 25     name[i] = rand () % 26 + 'a'; 
 26 
 27   name [len] = 0; 
 28 }
 29 
 30 int main (int argc, char *argv[])
 31 {
 32   int dir_len = 0, file_len = 0; 
 33   char *name = 0; 
 34   int ret = 0, level = 0; 
 35 #ifdef WIN32
 36   int name_max = MAX_PATH; 
 37   int path_max = MAX_PATH; 
 38   HANDLE fd = 0; 
 39 #else 
 40   char const* path = "/"; 
 41   int name_max = pathconf (path, _PC_NAME_MAX); 
 42   int path_max = pathconf (path, _PC_PATH_MAX); 
 43 #endif 
 44 
 45   printf ("NAME_MAX = %d, PATH_MAX = %d\n", 
 46     name_max, path_max); 
 47 
 48   // add 1 (/) to 10 to be conveniently to compute how low the path is with level.
 49   dir_len = 9; 
 50   file_len = 100; 
 51 
 52   srand (time(0)); 
 53   name = (char *) calloc (1, (dir_len > file_len ? dir_len : file_len) + 1); 
 54   if (name == 0)
 55     return -1; 
 56 
 57 
 58   do
 59   {
 60     get_random_name (name, dir_len, ++level); 
 61 #ifdef WIN32
 62     //ret = _mkdir (name); 
 63     //if (ret == -1)
 64     ret = CreateDirectory(name, NULL); 
 65     if (!ret)
 66     {
 67       printf ("CreateDirectory %s failed, errno = %d\n", name, GetLastError ()); 
 68       break; 
 69     }
 70 #else 
 71     ret = mkdir (name, 0777); 
 72     if (ret == -1)
 73     {
 74       printf ("mkdir %s failed, errno = %d\n", name, errno); 
 75       break; 
 76     }
 77 #endif 
 78     
 79     printf ("mkdir %s\n", name); 
 80 #ifdef WIN32
 81     //ret = _chdir (name); 
 82     //if (ret == -1)
 83     ret = SetCurrentDirectory(name); 
 84     if (!ret)
 85     {
 86       printf ("SetCurrentDirectory %s failed, errno = %d\n", name, GetLastError ()); 
 87       break; 
 88     }
 89 #else 
 90     ret = chdir (name); 
 91     if (ret == -1)
 92     {
 93       printf ("chdir %s failed, errno = %d\n", name, errno); 
 94       break; 
 95     }
 96 #endif 
 97 
 98     //printf ("change to that dir\n"); 
 99     get_random_name (name, file_len, level); 
100 #ifdef WIN32
101     fd = CreateFile (name, GENERIC_READ | GENERIC_WRITE, FILE_SHARE_READ, NULL, OPEN_ALWAYS, 0, 0); 
102     if (fd == INVALID_HANDLE_VALUE)
103       printf ("open %s failed, errno = %d\n", name, GetLastError ()); 
104     else 
105     {
106       printf ("open %s OK.\n", name); 
107       CloseHandle (fd); 
108     }
109 #else 
110     int fd = open (name, O_RDWR | O_CREAT, 0644); 
111     if (fd == -1)
112       printf ("open %s failed, errno = %d\n", name, errno); 
113     else 
114     {
115       printf ("open %s OK.\n", name); 
116       close (fd); 
117     }
118 #endif 
119   } while (1); 
120 
121   free (name); 
122   return 0; 
123 }

 

这个小程序很简单,每次创建一个目录,切换进去,创建一个文件。这样周而复始,直到出错退出。目录名长度固定为 9,这样加上目录分隔符(/ 或 \)就正好凑成 10,方便之后根据目录深度计算路径总长度;文件名长度固定为 100,也是为了方便计算路径总长度。每个名字都是由 26 个字母随机组成的,同时将目录深度 (level)记录在名字开始。例如第 10 级目录可能长这样:“10jmvrfqv”,前面的数字 10 明确的标识了目录层级,同理,文件名前面也是层级数字,也是为了方便计算。好了,我们分别在两个系统上跑一下这个测试程序。

1. Windows 上的结果 ( Win10 x64)

NAME_MAX = 260, PATH_MAX = 260
mkdir 1tdobxciu
open 1gybhgvdtngozbkzvcuvpnitcpntxgtjhasljfycfowuuqmcvvitsthzdxpckttobaqtccxeabfmnvhoeknyjifabnynkrjwzhiq OK.
mkdir 2wafsehgp
open 2ssywxcafcvpuppgxjcpctjkleftvyhwpuzydlcaaxkbsgumljccmokznqpkvliklndweyegqxvmrmtcrfwzqyllezvneztvqwma OK.
mkdir 3vmkjwgrp
open 3bclyodmcwjgzmokchhkrmharodswgmpbusxozgqogaguvppohpthetfgqxrmihdjjmqsxvdzgondfsirxfmqbmgmexdnbdjyrqr OK.
mkdir 4xmvhqrfh
open 4hclzbqzoniuuolhumysibxonqutbpqgvroamdwdorhypsfkrkyskqykdukfrngtfipgjidvazgsvdfejofbjqqwpthkfxzgwubw OK.
mkdir 5rwoxiiwh
open 5sdeemddmvkhszsubkojdemrconicercdpdcmsmitbbuxowvsqbfghagpwkgmdrytzfnafoqvwsktiwdkjskfukxtxjeknctdjcm OK.
mkdir 6fnmlhdor
open 6vjrqxecbutxcospyzihwjabulwzbvbwimubvvchdxgyhqfjebjnpbbhvtwrkjlowmzbpoqkshwjbzoqysxfthztefvzwgarrdon OK.
mkdir 7sbvhcsxw
open 7ddmdylgkhghdeefydnjfsmiyxwutewbmujsppdezpoamtwodvqvkyaeswywbfnvjctofiaftmkiwapbjawuhjvuqsmefjpfgbew OK.
mkdir 8dwhpsgaj
open 8xepfibfufsaefmghnjukuddbqhyijunkazscfjxkznnylylpcaiauwyrzcfixbjfguvimzwrabcnvxomkaelrwtgnhsjxpiwmii OK.
mkdir 9wmpqcalh
open 9ldugmypucabxtjnpxqdvinxtevztnspuakrwwhebpshzsgbkedvmxbibkwyaoxctfelwjsglvdwjfjffuleijofuoinnrdcdymf OK.
mkdir 10vlwydie
open 10jihhaxrcpsviddxserikiiyrtbiwdumzdsyhkkodreimjxivkpnnxxtgpuathmqatlajmfueupwsdwjhyferopxiqggkywccvz OK.
mkdir 11awvhauw
open 11ypsctazqpuebetmzzgrzasmjmzppymtdabgcajahhwuwilzxlgvpupuksrjtmigjogjyqxhczuunlknxupfixnyxlyjsdhgwuf OK.
mkdir 12pdznsnw
open 12easlcjvcdudtktfkseedbtlbvnjhjtlchtjbbshmjbdsnpifidqjwsttalzwlyjezixtghwortsusrifigbxhpznmiigtvviym OK.
mkdir 13pvvjtcb
open 13lbpsmpbwyzzcnjlixarmvbxpqeketgycbduhzlhhakxykncmxwhmgqrwigcywiehemmhlinrhxnxbktentpvnlazrzgniwgcnf OK.
mkdir 14jrhanvs
open 14nlvfzgxyfzekthrzidzoeugldefwnxlelbntfetdnxvxkcupkpnzxhytidycrstkchdojzrsdjxclkvfgvmmsenysyyinkylsd OK.
mkdir 15yilwbim
open 15dhjbeoeybggfrjpstbyobosrmigtfhmpbajgwhjgczgklphxweefjoliwbuabyxyfsnzjzpiaxdagfbkujrqistslqkffqmilx OK.
mkdir 16owgedmp
open 16cqauvpdzsghgmzdbdxaqdclrvuyophifemvsoygwuhbtlkkmwmwxekwcxgmauagexxtyirthfgjbztmutljstjiwcbwmntkfri failed, errno = 3
mkdir 17xljxtht
open 17umkwtdbxiayekrljssfnxgvcmytthwevlmrxptivqmvwrkpqkvxoircuwadhkyidoaydlvnnafejpssqdewcymluzmrwrrkwmn failed, errno = 3
mkdir 18aywtihk
open 18ynydjmncmakzoezcupoeqfiahsunxhqvczbuapntehglrrhubpbydnnpnbpjdschmrkzkkpvgfajemsfsvofsjoclijforjkex failed, errno = 3
mkdir 19qsvtxwm
open 19qpfdrjgnxthljturhvuymuoqctsmixxrircoievqpgrnxpzufozwvpsjocdxfhjodpfzljbgegutfsquaehjoaswqhwtsqbgmh failed, errno = 3
mkdir 20vzvuovy
open 20zqqxsfnxrccsubxwevfffunrxfugkovltzidalrmppjjyrbnohxgmwnmwcseijuhzabngyfodgodnblmvhmvnmkvqfijucfqfz failed, errno = 3
mkdir 21idgwupz
open 21vlkqahypwulaaykizzvqviqztswdwikvloswluydfyqmgjcaixqqrilsrjireprfrvbgfvspulfmnuksbvsnfkdncybuhxddtw failed, errno = 3
mkdir 22plbyhbe
open 22rbekcixmxcohnvqlyhlppmgqtpsgsjwxhuwlmhnzspkbvdrnzechjbjsglzajiadiqguutgfjfkaqdtfqeevpcrsejiiotytse failed, errno = 3
mkdir 23vggwvra
open 23tlbobatwbdpaxouojluctxqyddllqedqbeotpkkxgddristpprablooxlomgvneixubhxhemtlvipqwzzfgqcxjtdltytpialw failed, errno = 3
mkdir 24dkpkfup
open 24fujnicyzpudgnsuzcqkkgihsousejefcuepnkkxdpdxbxxfryeglrmikgxxckazfokmdrgttnqxavphpoekagfbenjyuxpfntu failed, errno = 3
CreateDirectory 25aajjgcq failed, errno = 206

 

我是在 G:\ 根目录运行这个程序的,跑到第 16 级目录时,CreateFile 失败返回了,错误码为 3 (ERROR_PATH_NOT_FOUND:系统找不到指定的路径),对于文件创建失败,这里忽略,所以程序继续运行;跑到第 25 级目录时,CreateDirectory 也失败返回了,错误码为 206 (ERROR_FILENAME_EXCED_RANGE:文件名或扩展名太长),因为不能再继续了,所以整个程序终止。看了一下,到 16 级目录时整个路径是这样:

 

总长度达到 162,此时如果想再创建文件的话,总长度会升至 162 + 1 + 100 = 263  > MAX_PATH,所以接下来的文件创建失败了。然后再看下到 24 级目录时路径的样子:

 

总长度达到 243,如果再加上一个目录的长度的话,应该是253 < MAX_PATH,按理说应该还可以容纳一个目录的创建,但是我实地在这个目录下面尝试创建目录,得到了这样的错误:

 

如果是创建文件的话,会发现输入一定长度的文件名之后,就输入不了了:

 

这个长度目前是 16 (算上后缀 .txt 4个字符),加上之前目录的长度 243,总长度为 243 + 1 + 16 = 260,正好等于 MAX_PATH。可见,对于 Windows 而言,这个 MAX_PATH 就是底层文件系统的限制,再怎么玩也玩不出这个杠杠。尝试切换 CreateDirectory / SetCurrentDirectory  实现为 _mkdir / _chdir,看看有什么不同 :

line 62-65

ret = _chdir (name); 
if (ret == -1)
//ret = SetCurrentDirectory(name); 
//if (!ret)

line 81-84

ret = _mkdir (name); 
if (ret == -1)
//ret = CreateDirectory(name, NULL); 
//if (!ret)

 

重跑上面的例子,会得到完全一样的结果。

2. Linux 上的结果 (neokylin x64)

一开始我是在 CentOS i686 环境上跑的,没想到一阵跑之后再回来看,虚拟机桌面居然重置了,所有 shell 都退出了,没有截到任何输出,无奈只好采用另一个平台了。在这个中标麒麟的环境跑完了全程:

$ ./path_max
NAME_MAX = 255, PATH_MAX = 4096
mkdir 1rkxtumwh
open 1tovmschzmqujtonfqbqyfomyykrtwqcrezfxbmypesyygmfxpvxvjmvkwnfudhnhgsghgfwkzxkgjpfamcxwotimgniluvucpbl OK.
mkdir 2vijihgsn
open 2rkuswxpuniccqrkbliwpzzaxhkfqqzgklackzrgpclrscduqnsfnrikasrqkswuexwqwqwlshekkigaxyhkrpwrjniuieqodnea OK.
mkdir 3fanxiuju
open 3cqwzqdmivizeyjygopureytlajkkdtehlciegumddnhdxgjnxfedfaofjysmtytfbbjjyxmdmtgjbrwywdddfrkopcclbxqcabn OK.
mkdir 4yazbmsjy
open 4wawuybydgrowjqyutxmxynkypjcbelzaowwoxwsgqiczabtuaftyuezjpbmvomxekutjsnpixuhxxctxhnwdtvmiybemnbrzxmk OK.
mkdir 5sbcaywkx
open 5torqwgobzlqjlrnygrrhqdrigvletvbmjufhcvldjbnuscuavlhmqawwwhcpegeqalzfgkirnvmfaghvuojkpghlnlcsrgkurjz OK.
mkdir 6aujthffp
open 6howeinoseubrjntbgclavxzxrjraqwrzmqdudrphnsaxhtaovootlqqebheteysqoxntpccewcbevcstqimbacfbklwqjqixovs OK.
mkdir 7fzujwwnc
open 7upunhebiffpircfhmxsllpespacnxpptglgnrjxyoohgqopclhqzxuroutdtkumshufydexusgcjwtnhbdiaxaquvvnircaywgy OK.
mkdir 8cmxwffao
open 8btdlwhtygtqaplqxcszcofbsuyqzeqphmssizmjgfziwlyvpsvsjcvbwtsxzkmhwhzhinqovrztezpusmodqjepeyoejdnhkoos OK.
mkdir 9cgjxyiqe
open 9jhydvmgovlfaltguwvegjubndkmbufhdoghmupasahuoackzxohijiwmukpopwueedqbsstvzpjbrtcokkyvuujrfahwwbcdeue OK.
mkdir 10xmzunod
open 10rfytviftdcqmtxmatlewqkqujfvdsjgjqffnpkgtoyhhvwkrhqnxaftkmqphbysrdxeukmpylxigtuxckmbmuwyimopnmkgrjk OK.

……

mkdir 1341932sc
open 1341932rlcbgduqybifdacrgopzdnulnwxjtrlldpmltibrjlzolbhuqytvnpgbmgmhzzscpfniprbbearsezmuxhrnwaqkgcsfb OK.
mkdir 1341933kk
open 1341933qrzyhsckxcbpgcdcckupgufrchjjivvbnvbwndgmgjdomgroqndzjiqnrbwbwuclpejejrtrbwgndzdvoguxqmkhpikoc OK.
mkdir 1341934nz
open 1341934ttlyecrwdpcssdvqrdmovzaeqipeneggxrgbvxzzpbthgrzxwmnsnnwdylilpotohasczrdovzybqxzmjmewcdaaokngy OK.
mkdir 1341935gu
open 1341935hgmjgenwzmuaesbsbozaqccsqmfwmnttufcclrykfvnjnocqebswduovgvusjnlduqhghhspcfypuciafaxivlgdhcwqr OK.
mkdir 1341936jv
open 1341936lzerjlkaqryhndpoisltnwzqfdoxuysfzyzkklkccljqozgyrrrgpsyvvnupnpxonwbzjlblyldomjndcglszlpwyjony OK.
mkdir 1341937le
open 1341937ojfnsspftcjhouwtxderepgcqrqfsdjgmqvejmkcovlcpivonbfttnvjhmqzpahcqeibruffrrijbfxojfhcuelbqdcgd OK.
mkdir 1341938lk
open 1341938wquzhrfoiyytzerpnyyrscewvhadnmplejkncrblpaeqgxhtxhkqkoohxokkczvikgvnzxarxghderybyktkajrxybkca OK.
mkdir 1341939fl
open 1341939lnganfaecgnhmehqftjpvthvtkfvlkiyypynuarzggitnsjsluhhpqckbjgotootfogcoxduglptfylrvuaklevondeit OK.
mkdir 1341940sc
open 1341940bgiduigqotghzgssbmuozylpnqtvjlzktipnqxggrmnsshkwvekvevmtngqwtqjnaacsyiyruljpstlnzxkfuxyhfodyg OK.
mkdir 1341941ol
open 1341941gpqypywgvhpkbkxrlubqoaovfeaelprrghqvimbdwspzemqpgshwsvrabrgmixgohyjpmmuigjikxybfsibnfunjltvws OK.
mkdir 1341942dk failed, errno = 28

 

中间输出比较多,就省略了,只显示前后十几行内容。我是在 /home/vmware/code/apue/02.chapter 目录下运行这个程序的,跑到 1341942 级目录时,mkdir 失败返回了,错误码为 28 (ENOSPC:No space left on device)。总的目录路径长度达到 34 + 1341941 * 10 = 13419444,最深层的一个文件总路径长度达到 34 + 1341941 * 10 + 100 = 13419544,都远远大于 PATH_MAX (4096)了。此时我在运行程序的目录新建文件或目录,都会失败。

$  mkdir abc
mkdir : 无法创建目录 "abc" : 设备上没有空间

$  touch abc
touch : 无法创建 "abc" : 设备上没有空间

 

所以可以肯定的一点是,Linux 上的 PATH_MAX 并不是底层文件系统对路径最大长度的限制,只是一种便于程序编写的常量。那路径的最大长度到底由什么决定呢?看了上面 ENOSPC 的错误码,我第一反应是硬盘空间满了,然而在查看了磁盘剩余空间后,我否决了这个可能性。下面是我在执行测试程序前后分别记录的 df 输出。

 

可以看到所在的 /home 挂载点使用比例激增 (8%~22%),但是也没有达到 100%,所以磁盘空间还是比较充足的,很可能只是 inode 用光了(?)。不管怎样,本质上都是一种资源限制,这使得程序员可以在 Linux 上创建比较长的文件路径,比 Windows 提供了更大的灵活性。但是这样长的路径并不被所有程序所识别,例如,我在刚才创建测试目录的地方,递归列出所有文件,输出如下:

$ ls -R
.:
1rkxtumwh  Makefile  name_max.c  path_max    path_max.o    path_max.vcxproj
apue.o     name_max  name_max.o  path_max.c  path_max.sln

./1rkxtumwh:
1tovmschzmqujtonfqbqyfomyykrtwqcrezfxbmypesyygmfxpvxvjmvkwnfudhnhgsghgfwkzxkgjpfamcxwotimgniluvucpbl
2vijihgsn

./1rkxtumwh/2vijihgsn:
2rkuswxpuniccqrkbliwpzzaxhkfqqzgklackzrgpclrscduqnsfnrikasrqkswuexwqwqwlshekkigaxyhkrpwrjniuieqodnea
3fanxiuju

./1rkxtumwh/2vijihgsn/3fanxiuju:
3cqwzqdmivizeyjygopureytlajkkdtehlciegumddnhdxgjnxfedfaofjysmtytfbbjjyxmdmtgjbrwywdddfrkopcclbxqcabn
4yazbmsjy

./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy:
4wawuybydgrowjqyutxmxynkypjcbelzaowwoxwsgqiczabtuaftyuezjpbmvomxekutjsnpixuhxxctxhnwdtvmiybemnbrzxmk
5sbcaywkx

./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx:
5torqwgobzlqjlrnygrrhqdrigvletvbmjufhcvldjbnuscuavlhmqawwwhcpegeqalzfgkirnvmfaghvuojkpghlnlcsrgkurjz
6aujthffp

./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp:
6howeinoseubrjntbgclavxzxrjraqwrzmqdudrphnsaxhtaovootlqqebheteysqoxntpccewcbevcstqimbacfbklwqjqixovs
7fzujwwnc

./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp/7fzujwwnc:
7upunhebiffpircfhmxsllpespacnxpptglgnrjxyoohgqopclhqzxuroutdtkumshufydexusgcjwtnhbdiaxaquvvnircaywgy
8cmxwffao

./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp/7fzujwwnc/8cmxwffao:
8btdlwhtygtqaplqxcszcofbsuyqzeqphmssizmjgfziwlyvpsvsjcvbwtsxzkmhwhzhinqovrztezpusmodqjepeyoejdnhkoos
9cgjxyiqe

./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp/7fzujwwnc/8cmxwffao/9cgjxyiqe:
10xmzunod
9jhydvmgovlfaltguwvegjubndkmbufhdoghmupasahuoackzxohijiwmukpopwueedqbsstvzpjbrtcokkyvuujrfahwwbcdeue

./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp/7fzujwwnc/8cmxwffao/9cgjxyiqe/10xmzunod:
10rfytviftdcqmtxmatlewqkqujfvdsjgjqffnpkgtoyhhvwkrhqnxaftkmqphbysrdxeukmpylxigtuxckmbmuwyimopnmkgrjk

……

410dntuzt
ls: 无法打开目录./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp/7fzujwwnc/8cmxwffao/9cgjx
yiqe/10xmzunod/11ltybuja/12pddukks/13pgjmtme/14yxyuhvt/15qgddfps/16ipghtsx/17hrlvjqw/18mlwamyj/19ycqq
oum/20gdekplr/21mgkqtef/22uxjelyo/23yuogjxv/24gdyuetr/25ekyseyq/26mohxeho/27yhfufeo/28emftmwp/29detfm
nz/30ulhqorq/31lzvmcau/32vrkdiho/33pwrixvt/34ysjlzxm/35anbimen/36aazpcto/37dpzcvhb/38vbxsrwc/39urooly
a/40gabvtkg/41fswpnkc/42eyozjbb/43mbsphto/44szukcad/45lobskag/46njddwrk/47qmyaqip/48qywwpge/49jymbads
/50cevctwp/51dbrvvrg/52alaqbak/53wdxytcx/54iajcocc/55yulgnls/56xdvlohr/57rsyzzme/58avlsjct/59kzwpnmn/
60zuuyott/61cmyywqh/62fyuzvce/63djljats/64aopcwyi/65rxefdvy/66egqsclw/67jxyausw/68atlilhv/69xrxqsel/7
0isglkct/71fwkjdms/72mhygulk/73nlnwbrx/74sroynrd/75dimglds/76ulettre/77cvdbchx/78imrnssw/79uqfezsr/80
lkhipxs/81rbdpqqz/82lvyzxqt/83sgxgrxx/84tsgcvwa/85jaamdba/86zoneybo/87glbanpe/88tkzaefg/89rrwsack/90u
owxuoq/91rabkpyi/92dcnxiyz/93lacaahs/94frjwezd/95yhtpnte/96fblotmh/97mibhekj/98rzonuec/99topwrdn/100d
blgkn/101drisrx/102aqjjso/103uefcom/104yrflxq/105hyfxyh/106tlpbnl/107ndiswq/108uvgksb/109ozcwup/110qx
pxlo/111gttfdp/112nepxhw/113xwflah/114yibgjy/115dffvxj/116ukiosf/117fpvyvt/118kdreee/119ejgwiz/120lpo
qvc/121gihnlr/122umltry/123ajaaeq/124unorwj/125iuipco/126uwghec/127lzrzvo/128zytdwi/129xvkgeh/130mgfx
pr/131kekuik/132fjlbpy/133wogrgw/134eotsoo/135fquige/136smcyju/137obzwup/138fewkay/139dgizof/140gqrvp
w/141lwxiji/142gprlwz/143geyxsn/144otqwya/145ooyfvk/146ucigye/147xstuji/148ziprio/149ufrzci/150lmgpli
/151kczfla/152svbdit/153fzlyns/154yirhkv/155xbdixx/156mogbae/157goxflw/158fixfsg/159ncihug/160bnrvsj/
161ynpfeh/162cmanrm/163utjekr/164qcvzim/165ypxvkh/166ggnvdw/167nszvsr/168uyuayf/169fatffy/170znjsdj/1
71jqobhe/172syewyj/173fbdzgu/174bufnki/175qasxqn/176rigmyh/177zwoexf/178bdrxbg/179ggdmzx/180ljyrtm/18
1qbywid/182wsjhtx/183moknmr/184haaeaa/185jdpxza/186eerxfa/187kpfjiq/188wrdbbh/189uuhvct/190ofpwct/191
qnmtmy/192gqyowy/193eckugg/194oonhah/195dbjuxd/196qzbvuh/197kcqziu/198akbhoa/199hkszfc/200aqieum/201a
hkuqb/202hkikbi/203gywiex/204ibfmca/205vgczdp/206pchnch/207tfkxlo/208orztrq/209foperp/210umtatf/211rf
qqzo/212xuspzm/213bsqdet/214kbvvmg/215qsbepc/216qqqvea/217mtyasr/218eunbgv/219stppxz/220ncqzlb/221qgl
ips/222naqylv/223ymrfpk/224vavwhr/225dznual/226iuraac/227ztobic/228jynnfd/229iatquu/230vuokzi/231ykpc
bl/232wqldxe/233qeebyv/234nvspvv/235iqxfye/236nwnztb/237qjtgnn/238qyzyas/239lhcqon/240piorsv/241upvvm
j/242sjgemx/243ndojai/244whhuvw/245hmojut/246kwaiwh/247ohckmb/248ucmdrh/249eqzimi/250uymmqd/251cnglst
/252iysywl/253jridil/254nmoqzh/255ygguhk/256nwuiap/257zvtolc/258urycli/259kpiuab/260jempzc/261jyqcbe/
262sulagk/263mvljld/264ciiawa/265vjxora/266rxfgkg/267jwvpoy/268mpmgja/269xpxycf/270fjxobt/271zahncm/2
72soljdu/273jcdwrq/274wvrusy/275noyexu/276oksjfe/277wzfwvr/278tcsfue/279uhzgjs/280lfpypw/281cyjibv/28
2jwnlgu/283nnktxs/284udmuma/285xzbwnj/286jemqma/287xnedax/288dcvfle/289ddnbqe/290jheimg/291rscgbr/292
zlyhgw/293jbmjzx/294qocgsi/295mbyvyj/296ntwpkk/297jhofrm/298ibirpc/299kquens/300mbjnmo/301najyhm/302g
yerut/303blongi/304tafajx/305suvetf/306faeoiu/307dunnih/308ifvtsq/309vulqxe/310ilvtjv/311gzknfk/312ju
xeni/313bpaezt/314dcqgbb/315rqnuic/316ejhbrb/317wfzvzo/318hfndtr/319zgsznw/320dqahuh/321mdnyne/322xiu
lms/323lplrvo/324qtvodk/325tiigxk/326kljzob/327lxxopm/328qwjlms/329jwcbic/330nowwgr/331jqqbce/332qael
pn/333vyyvlr/334wvchpm/335afdbpb/336vsrvym/337ktyfii/338gasxaz/339gjjnco/340yqdxcc/341eysxtl/342ipsoc
o/343xrjucz/344aoetoq/345rtnyzv/346ptrthp/347gnxqkl/348vfpcrp/349jmoobg/350glqvzi/351vkznzy/352cmjseg
/353zgxeyh/354vfkgth/355fnhgjo/356cyouso/357hruehw/358jadbhv/359hklsla/360hwsife/361ragklc/362xydtpr/
363tjrhwj/364gpsojy/365sphazl/366fgwang/367tovqhr/368ybyzbe/369wasspn/370jjkehg/371fvmqie/372bdigtd/3
73afocqc/374agluqq/375jlrfsu/376gtbemx/377lqbrlg/378xmxxvu/379pwznpd/380gdryol/381cgzdfe/382egzxfd/38
3ciccll/384mpoifm/385ygdqiy/386ahilzb/387bhmyia/388cmfhah/389kuxsqg/390winlgk/391jnjlim/392vurqut/393
yehkub/394vkgknd/395pxnmgr/396bsnizx/397bzbbbs/398kdfjjs/399ugopik/400klxzic/401kaqhcd/402iyrruo/403d
tncvg/404feosyw/405cbohnn/406hfirse/407tqbohz/408udrjbo/409yhmpsz/410dntuzt: 文件名过长

 

中间输出比较多,就省略了。跑到 410 级目录时,ls 报错了,上一级目录路径的总长度达到 34 + 409 * 10 = 4124,其实已经比 PATH_MAX 大了,不过还是因为缓冲区不足而中断退出了。再尝试查找文件,输出如下:

$ find .
.
./name_max.o
./path_max.c
./Makefile
./name_max
./path_max
./path_max.o
./name_max.c
./1rkxtumwh
./1rkxtumwh/1tovmschzmqujtonfqbqyfomyykrtwqcrezfxbmypesyygmfxpvxvjmvkwnfudhnhgsghgfwkzxkgjpfamcxwotimgniluvucpbl
./1rkxtumwh/2vijihgsn
./1rkxtumwh/2vijihgsn/3fanxiuju
./1rkxtumwh/2vijihgsn/3fanxiuju/3cqwzqdmivizeyjygopureytlajkkdtehlciegumddnhdxgjnxfedfaofjysmtytfbbjjyxmdmtgjbrwywdddfrkopcclbxqcabn
./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy
./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx
./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp
./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp/6howeinoseubrjntbgclavxzxrjraqwrzmqdudrphnsaxhtaovootlqqebheteysqoxntpccewcbevcstqimbacfbklwqjqixovs
./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp/7fzujwwnc
./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp/7fzujwwnc/7upunhebiffpircfhmxsllpespacnxpptglgnrjxyoohgqopclhqzxuroutdtkumshufydexusgcjwtnhbdiaxaquvvnircaywgy
./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp/7fzujwwnc/8cmxwffao
./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp/7fzujwwnc/8cmxwffao/8btdlwhtygtqaplqxcszcofbsuyqzeqphmssizmjgfziwlyvpsvsjcvbwtsxzkmhwhzhinqovrztezpusmodqjepeyoejdnhkoos
./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp/7fzujwwnc/8cmxwffao/9cgjxyiqe
./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp/7fzujwwnc/8cmxwffao/9cgjxyiqe/10xmzunod
./1rkxtumwh/2vijihgsn/3fanxiuju/4yazbmsjy/5sbcaywkx/6aujthffp/7fzujwwnc/8cmxwffao/9cgjxyiqe/10xmzunod/10rfytviftdcqmtxmatlewqkqujfvdsjgjqffnpkgtoyhhvwkrhqnxaftkmqphbysrdxeukmpylxigtuxckmbmuwyimopnmkgrjk

……

 

最后有没有完成我没有得到结论,因为整个虚拟机历时 N 天这个命令还没跑完 (N > 10),运行中的 find 截图为证:

 

我是按内存占用从高到低排序的,可以看到经过 N 天的运行 find 命令的内存占用已经超过了整个图形界面(Xorg),另外与 find 命令关联的终端 (mate-terminal) 内存、CPU也在高位运行。不过至少可以说明 find 并没有使用 PATH_MAX 来简单限制路径长度,可能是通过动态分配内存来实现对长路径支持的。另外个人比较好奇这个命令的句柄开支,于是看了下 lsof 的输出:

$ lsof -p `pidof find`
COMMAND    PID   USER   FD      TYPE DEVICE  SIZE/OFF    NODE NAME
find    113339 vmware  cwd       DIR    8,5      4096 2095005 /home/vmware/code/apue/02.chapter
find    113339 vmware  rtd       DIR    8,3      4096       2 /
find    113339 vmware  txt       REG    8,3    203296 1180356 /usr/bin/find
find    113339 vmware  mem       REG    8,3 106374736 1187939 /usr/lib/locale/locale-archive
find    113339 vmware  mem       REG    8,3    141880 1191334 /usr/lib64/libpthread-2.20.so
find    113339 vmware  mem       REG    8,3     19512 1190784 /usr/lib64/libdl-2.20.so
find    113339 vmware  mem       REG    8,3    447240 1191294 /usr/lib64/libpcre.so.1.2.3
find    113339 vmware  mem       REG    8,3   2082456 1190695 /usr/lib64/libc-2.20.so
find    113339 vmware  mem       REG    8,3   1167000 1191118 /usr/lib64/libm-2.20.so
find    113339 vmware  mem       REG    8,3    154784 1191415 /usr/lib64/libselinux.so.1
find    113339 vmware  mem       REG    8,3    163184 1190512 /usr/lib64/ld-2.20.so
find    113339 vmware  mem       REG    8,3     26254 1448972 /usr/lib64/gconv/gconv-modules.cache
find    113339 vmware  mem       REG    8,3     30239 3015834 /usr/share/locale/zh_CN/LC_MESSAGES/findutils.mo
find    113339 vmware    0u      CHR  136,0       0t0       3 /dev/pts/0
find    113339 vmware    1u      CHR  136,0       0t0       3 /dev/pts/0
find    113339 vmware    2u      CHR  136,0       0t0       3 /dev/pts/0
find    113339 vmware    3r      DIR    8,5      4096 2095005 /home/vmware/code/apue/02.chapter
find    113339 vmware    4r  unknown                          /proc/113339/fd/4 (readlink: File name too long)
find    113339 vmware    6r  unknown                          /proc/113339/fd/6 (readlink: File name too long)
find    113339 vmware    7r  unknown                          /proc/113339/fd/7 (readlink: File name too long)
find    113339 vmware    9r  unknown                          /proc/113339/fd/9 (readlink: File name too long)
find    113339 vmware   10r  unknown                          /proc/113339/fd/10 (readlink: File name too long)
find    113339 vmware   11r  unknown                          /proc/113339/fd/11 (readlink: File name too long)

 

出人意料的是并没有很多打开中的文件句柄。按我的理解,每遍历一层目录应该有一个目录文件的句柄被打开,但是现在看没有,是因为我造的例子比较特殊,每个目录下只有一个子目录?还是 find 另辟蹊径不用占用许多句柄?暂时不得而知。总的来讲,find 的表现要好于 ls,顿时对这个神奇的命令产生了敬仰。最后插一句题外话,从上面的输出我们可以看到另一个现象,就是因为路径太长了,readlink 返回了错误,可见对长路径,很多系统 api 也是不支持的。最后你可以使用 rm -rf xxxx 来尝试清理这些目录与文件:

 

不出所料 rm 占用了很多的内存和 CPU,不过好歹可以工作 (而且耗时没有 find 那么夸张,也就十几分钟),所以我的虚拟机可以恢复到测试前的工况。下面是 lsof 针对 rm 的一些输出

$ lsof -p `pidof rm`
COMMAND   PID   USER   FD      TYPE DEVICE  SIZE/OFF    NODE NAME
rm      70867 vmware  cwd       DIR    8,5      4096 2095005 /home/vmware/code/apue/02.chapter
rm      70867 vmware  rtd       DIR    8,3      4096       2 /
rm      70867 vmware  txt       REG    8,3     62872 1181241 /usr/bin/rm
rm      70867 vmware  mem       REG    8,3 106374736 1187939 /usr/lib/locale/locale-archive
rm      70867 vmware  mem       REG    8,3   2082456 1190695 /usr/lib64/libc-2.20.so
rm      70867 vmware  mem       REG    8,3    163184 1190512 /usr/lib64/ld-2.20.so
rm      70867 vmware    0u      CHR  136,0       0t0       3 /dev/pts/0
rm      70867 vmware    1u      CHR  136,0       0t0       3 /dev/pts/0
rm      70867 vmware    2u      CHR  136,0       0t0       3 /dev/pts/0
rm      70867 vmware    3u  unknown                          /proc/70867/fd/3 (readlink: File name too long)
rm      70867 vmware    4u  unknown                          /proc/70867/fd/4 (readlink: No such file or directory)
rm      70867 vmware    5u  unknown                          /proc/70867/fd/5 (readlink: File name too long)
rm      70867 vmware    6u  unknown                          /proc/70867/fd/6 (readlink: File name too long)
rm      70867 vmware    7u  unknown                          /proc/70867/fd/7 (readlink: File name too long)
rm      70867 vmware    8u  unknown                          /proc/70867/fd/8 (readlink: File name too long)

 

和 find 一样,它没有多少打开中的文件句柄(否则打开文件句柄数早超限了)。另外从 lsof 的输出看,有一些节点已经被 rm 删除了,所以 readlink 报错。

结论

对超长的文件路径来说,不同的 Linux 命令的支持程度也是不同的,有的支持,有的不支持。而且考虑到传递给进程的命令行总长度也是有限制的,除了一些支持递归目录处理的命令外,普通的命令是不会支持这么长的路径的。另一方面,从程序员的角度考虑,对于大路径的支持也是比较麻烦的,首先是系统支不支持输入、输出超长路径;如果支持,输入的场景还比较好办,输出的场景就比较麻烦了。最简单的办法是自己定义一个大于 PATH_MAX 值的常量并使用它分配内存,但是这样也存在问题,一方面日常处理比较浪费内存;另一方面如果路径超过你自己定义的这个值,还是会出现接收截断的问题。而且程序员没有办法去预知要得到的路径长度,从而提前动态分配内存。

总结一下,这个最大路径限制并不是系统不能支持超长的路径,而且作为一种系统与应用之间的一个约定,有了这个约定之后,程序对路径的处理将得到简化。况且我们没有对超长路径的硬需求,绝大部分 Windows 应用在 260 个字符路径的限制下就跑得很好,Linux 4096 的限制就更不用说了。

这个测试程序的一个额外用处就是耗尽所在分区的 inode(?),并观察系统在这种极限情况下的一些表现,例如,当我试图截取屏幕时不能成功,得到下面的提示框:

 

 

当你需要验证程序在这种极限状态下的行为时,这个程序不失为一种有用工具。

posted @ 2020-07-14 10:44  goodcitizen  阅读(5831)  评论(0编辑  收藏  举报